Mathematical Modeling of Water Quality PDF

INTERNATIONAL SERIES ON
APPLIED SYSTEMS ANALYSIS
MATHEMATICAL
MODELING OF
WATER QUALITY:
Streams, Lakes, and
.Reservoirs
Edited by
GERALD.T ORLOB
International Institute for

Applied Systems Analysis
MATHEMATICAL MODEUNG OF
WATER QUALITY:
Streams, Lakes, and Reservoirs
Edited by
Gerald T. Orlob
University of California, Davis
This book is the first to deal comprehensively with the

subject of mathematical modeling of water quality in
streams, lakes, and reservoirs. About one third of the
book is devoted to model development processes-iden-
tification, formulation, parameter estimation, calibra-
tion, sensitivity testing, and application-and a
thorough review of the mathematical principles and
techniques of modeling. Emphasis is placed on well
documented models, representative of the current state
of the art, to illustrate capabilities and limitations for the
simulation of water quality. About two thirds of the
book deals with specific applications of models for simu-
lation of water quality in natural water bodies. Topics
covered include modeling of temperature, dissolved
oxygen and phytoplankton growth in streams, develop-
ment and application of one-dimensional models of
stratified impoundments, two- and three-dimensional
modeling of circulation and water quality in large lakes,
thermally stratified plumes and cooling ponds, ecology
of lakes and reservoirs, modeling of toxic substances,
and the use of models in water quality management and
decision making.
Contents
Introduction
G. T. Orlob
2 A Procedure for Modeling
M. B. Beck
3 General Principles in Deterministic Water Quality
Modeling
P. Mauersberger
4 Modeling the Ecological Processes
S. E. Jf/Jrgensen
5 Simulation of the Thermal Regime of Rivers
J. Jacquet
(contents continued on back flap)
Mathematical Modeling of Water Quality:
Wiley IIASA International Series on Applied Systems Analysis
CONFLICTING OBJECTIVES IN DECISIONS

Edited by David E. Bell, University of Cambridge,
Ralph L. Keeney, Woodward-Clyde Consultants, San Francisco,
and Howard Raiffa, Harvard University.
2 MATERIAL ACCOUNTABILITY
Rudolf Avenhaus, Nuclear Research Center Karlsruhe,
and University of Mannheim.
3 ADAPTIVE ENVIRONMENTAL ASSESSMENT AND

MANAGEMENT
Edited by C. S. Holling, University of British Columbia.
4 ORGANIZATION FOR FORECASTING AND

PLANNING: EXPERIENCE IN THE SOVIET UNION AND
THE UNITED STATES
Edited by W.R. Dill, New York University,
and G. Kh. Popov, Moscow State University.
5 MANAGEMENT OF ENERGY/ENVIRONMENT SYSTEMS:

METHODS AND CASE STUDIES
Edited by Wesley K. Foell, University of Wisconsin, Madison.
6 SYSTEMS ANALYSIS BY MULTILEVEL METHODS

Yvo M. I. Dirickx, Twente University of Technology,
and L. Peter Jennergren, Odense University.
7 CONNECTIVITY, COMPLEXITY, AND

CATASTROPHE IN LARGE-SCALE SYSTEMS
John Casti, New York University.
8 PITFALLS OF ANALYSIS
Edited by Giandomenico Majone
and E. S. Quade, International Institute for Applied Systems Analysis.
9 CONTROL AND COORDINATION IN HIERARCHICAL
SYSTEMS
W. Findeisen, Institute of Automatic Control, Technical University,
Warsaw;
F. N. Bailey, University of Minnesota;
M. Brdys, K. Malinowski, P. Tatjewski, and A. Woiniak, Institute of
Automatic Control, Technical University, Warsaw.
10 COMPUTER CONTROLLED URBAN TRANSPORTATION

Edited by Horst Strobel, University for Transportation and
Communication" Friedrich List," Dresden.
11 MANAGEMENT OF ENERGY/ENVIRONMENT SYSTEMS:

NATIONAL AND REGIONAL PERSPECTIVES
Edited by Wesley K. Foell, University of Wisconsin, Madison,
and Loretta A. Hervey, International Institute for Applied Systems
Analysis.
12 MATHEMATICAL MODELING OF WATER QUALITY:

STREAMS, LAKES, AND RESERVOIRS
Edited by Gerald T. Orlob, University of California, Davis.
International Series on
12 Applied Systems Analysis
Mathematical Modeling of
Water Quality:
Edited by
Gerald T. Orlob
University of California, Davis
A Wiley-Interscience Publication
International Institute for Applied Systems Analysis
JOHN WILEY & SONS

Chichester-New York- Brisbane-Toranto-Singapore
Copyright © 1983 International Institute for Applied Systems Analysis
All rights reserved.
No part of this book may be reproduced by any means. or transmitted,

or translated into a machine language without the written permission
of the publisher.
Lihrary ofConRress CataloRing in Puhlication Data:
Mathematical modeling of water quality: streams,

lakes, and reservoirs
(International series on applied systems analysis;
12)
'A Wiley-Interscience publication.'
I ncludes index.
I. Water quality-Mathematical models. I. Orlob,
Gerald T. II. Series.
TD365.M27 1982 363.7'394'0724 82-17457
ISBN 0471 J00315(U.S.)
British Lih"ary Cataloguing in Puhlh'ation Data:
Mathematical modeling of water quality: streams,

lakes, and reservoirs-(Wiley international series in
applied systems analysis (IIASA)
I. Water quality-Mathematical models
I. Orlob, Gerald T.
ISBN 0471100315
Filmsel al Composition House Limited. Salisbury. Wiltshire.
Printed and hound in Greal Brilain al Ihe Pitman Press, Bath.

The Authors
GERALD T. ORLOB S. E. J0RGENSEN

Department of Civil Engineering The Royal Danish School of Pharmacy
University of California U niversitetsparken 2
Davis, CA 95616 2100 Copenhagen
USA Denmark
M. BRUCE BECK DANIEL P. LOUCKS

Department of Civil Engineering School of Civil and Environmental
Imperial College Engineering
London SW7 2BU Cornell University
England [thaca, NY 14853
USA
M. J. GROMIEC PETER MAUERSBERGER

Institute of Meteorology and Water Institute for Geography and Geoecology
Management Academy of Sciences of the GDR
ul. Podlesna 61 Miiggelseedamm 260
01-673 Warsaw 1162 Berlin
Poland German Democratic Republic
DONALD R. F. HARLEMAN OLEG F. VASILIEV

Ralph M. Parsons Laboratory for Institute of Hydrodynamics
Water Resources and Hydrodynamics Siberian Branch of USSR Academy of
Department of Civil Engineering Sciences
Massachusetts Institute of Technology 630090 Novosibirsk
Cambridge, MA 02139 Prospect Nauki 15
USA USSR
J. JACQUET MASATAKA WATANABE

Electricite de France National Institute for Environmental
64 Rue La Boetie Studies
75008 Paris Yatabe
France Tsukuba
Ibaraki 305
Japan
VlI
Preface
A capability to represent quantitatively the responses of the aquatic environ-

ment to actions of man is an essential part of resource management. It is
manifest in the mathematical model, which, if correctly designed, will simulate
real-world behavior under conditions yet to be experienced. Used with discretion
and appreciation of its limitations, the model can become a helpful tool in the
management process, enabling the user to explore new horizons of the imagina-
tion, to compare choices, and to identify pathways toward superior solutions
to practical problems.
Mathematical modeling of water quality presents a special challenge to the
systems analyst because it demands integration of so many disciplines. It is
dependent upon hydrology and hydromechanics for description of the move-
ment of water and the mechanisms of mixing. It calls upon climatology, meteor-
ology, and atmospheric physics to specify conditions at the air-water interface.
It draws on the chemistry of dilute solutions, chemical kinetics, and
biochemistry for determination of the fate of substances dissolved or suspended
in water. It requires knowledge of the interrelationships of aquatic life forms
and their environment-an understanding of aquatic ecology. Perhaps it is
this interdisciplinary aspect of water quality modeling that has attracted so
many competent scientists to become practitioners of the art. Whereas a
decade or so ago water quality models were novelties in the technical literature,
today they are acknowledged with increasing frequency as necessary elements
in environmental and resource management.
Yet, despite the enthusiasm with which modeling of aquatic systems has
apparently been embraced, there exists a gap between conception of the model
as an exercise of the mind and its use as a practical tool. One has only to examine
the literature to see that comparatively few water quality models have attained
a status that enables the technology they represent to be transferred to others.
IX
x
It was this deficiency in the art that prompted me to propose early in 1977 that
IIASA undertake to survey water quality modeling as a part of its State-of-the-
Art Series. Professor Zdzislaw Kaczmarek, then in charge of IIASA's water
program, invited me to visit Laxenburg for a two-week period and to outline
in detail a program for the survey. The proposal I presented to IIASA for a
monograph series on water quality modeling was accepted in principle and
work on this volume on Streams, Lakes, and Rservoirs was initiated in September
1978. We were successful, as you will see, in securing the collaboration of
an eminently qualified group ofcontributors to the modeling art, each individual
preparing chapters for which his experience best qualified him. This book is the
result of their combined efforts, stimulated from time to time by the editor, by
Professor Kaczmarek, and by Professor Oleg Vasiliev, who served both as
Deputy Director of IIASA and as a major contributor to the project.
The charge to our group was to capture as best we could the essence of
water quality modeling-the basic principles upon which it is based, the
practical problems in conceptualizing real-world phenomena in model form,
and the use of models as aids in decision making. I believe we have met this
charge and that you, the reader-likely a modeler yourself--will find that the
examples we have chosen to illustrate water quality modeling of streams, lakes,
and reservoirs are fairly representative of the state of our art.
G. T. Orlob
Benicia, California
Contents
Notation for Book xix
Introduction
G. T.Orlob
1.1 Water quality in perspective I

1.2 The beginnings of modeling 2
1.3 Growth of water quality modeling 3
1.4 Proliferation of models and the need for technology transfer 5
1.5 Scope of this volume 7
2 A Procedure for Modeling II

M. B. Beck
2.1 Introduction II
2.2 Definition of goals and objectives 14
2.3 Conceptualization 15
2.4 Selection of model type 17
2.4.1 A priori sensitivity analysis, 21
2.5 Computational representation 23
2.6 Calibration and verification 23
2.6.1 Observing the behavior or the system, 25
2.6.2 Experimental design, 28
2.6.3 Model structure identification, 30
2.6.4 Parameter estimation, 32
2.6.5 Verification, 35
2.7 Validation 36
2.7.1 A posteriori sensitivity analysis, 37
2.8 Summary and conclusions 38
XI
Xli
3 General Principles in Deterministic Water Quality Modeling 42

P. Mauersberqer
3.1 Introduction 42
3.1.1 The aim of this chapter, 42
3.1.2 Aims and types of deterministic water quality models, 43
3.1.3 Physical, chemical, and biological components and processes, 45
3.1.4 Balance equations, 46
3.1.5 Special types of balance equation, 47
3.1.6 The finite-difference method, 51
3.1.7 Thefinite-element method, 53
3.2 Conservation of mass-advection and diffusion 55
3.2.1 Continuity equation, mass fractions, 55
3.2.2 Chemical reactions, 56
3.2.3 Advection-diffusion equation; input-output models, 57
3.2.4 Biotic components, 59
3.2.5 Coupling of chemical and biotic components, 59
3.3 Conservation of momentum; types of flow systems 59
3.3.1 Navier-Stokes equations, 59
3.3.2 Special types offluidflow, 61
3.3.3 Waves, currents, and circulation: three-dimensional models, 63
3.3.4 Characteristic magnitudes, dimensionless equations, and
dynamic similarity, 63
3.4 Energy balance; thermal energy and heat exchange 66
3.4.1 Energy balance equations, 66
3.4.2 Model equations and types of solution, 68
3.4.3 Energy exchange, 69
3.5 Chemical, biochemical, and biological processes 70
3.5.1 Basic processes, constituents, interrelationships, and equations, 70
3.5.2 Energy balance for chemical reactions, 77
3.5.3 Reaction rates and reaction equations, 79
3.5.4 Chemical reactions in natural water bodies, 81
3.5.5 Photosynthesis and primary production, 83
3.5.6 Trophic interactions (grazing and feeding rates), 87
3.5.7 Egestion and excretion, 88
3.5.8 Respiration and nonpredatory mortality, 89
3.6 Further constraints on water quality models 90
3.6.1 Initial and boundary conditions, 90
3.6.2 Entropy, 91
3.6.3 The generalized Gibbs relation, 94
3.6.4 Thermodynamic relations, 95
3.6.5 The entropy balance equation, 96
3.66 The linear region of thermodynamic theory, 99
3.6.7 The nonlinear region of thermodynamic theory; stability
criteria, 99
3.6.8 Thermodynamics o.f primary production, 101
3.7 Problems and limits of deterministic water quality modeling 109
Xlll
4 Modeling the Ecological Processes 116

S. E. Jorgensen
4.1 Background 116
4.1.1 The ecosystem and chemical-biological processes, 116
4.1.2 Briefreview oj developmen t. 119
4.2 Modeling ecological processes in aquatic ecosystems 120
4.2.1 General considerations, 120
4.2.2 Photosynthesis, 120
4.2.3 Zooplankton grazing, 127
4.2.4 Respiration and nonpredatory mortality ofphytoplankton and
zooplankton, 130
4.2.5 Sinking, mineralization, and excretion, 131
4.2.6 Nitrogen .fixation, nitrification, and denitrification, 133
4.2.7 Release ofnutrient from sediment, 135
4.2.8 Effect of temperature on process rates. 136
4.2.9 Effect offish on water quality, 138
4.2.10 Other ecological processes, 138
4.2.11 Parameter values and stoichiometric ratio, 139
4.3 Concluding remarks 144
5 Simulation of the Thermal Regime of Rivers 150

J. Jacquet
5.1 Purpose of modeling 150
5.2 Thermal regime of a stream 151
5.3 Areas for modeling 152
5.4 Thermodynamics of a river 152
5.5 Simulation models 153
5.5.1 Deterministic approach, 154
5.5.2 Models for steady state conditions, 155
5.5.3 Exponential method, 155
5.5.4 Differential equation for steady state conditions, 156
5.5.5 Modelsfor nonsteady state conditions, 157
5.5.6 Simulation modelsfor nonsteady state conditions, 157
5.6 Data required for implementation 159
5.6.1 Simplified models for steady state conditions, 159
5.6.2 Simulation models for nonsteady state conditions, 159
5.6.3 Choosing a model, 162
5.7 Calculation techniques 162
5.8 Model adjustments and sensitivity analysis 163
5.9 Model application 164
Appendix. Surface heat exchange 170
XIV
6 Stream Quality Modeling 176

M. J. Gromiec, D. P. Loucks, and G. T. Orlob
6.1 Introduction 176
6.2 Types of stream quality model 177
6.3 Dissolved oxygen and biochemical oxygen demand 178
6.4 DOSAG I and QUAL I 188
6.5 Nitrogen cycle 196
6.6 Multiconstituent river ecosystem models 200
6.6.1 Bacteria and protozoa in pollutant degradation, 205
6.6.2 Michaelis-Menten models of aerobic nitrogen cycle, 209
6.7 Quality routing in unsteady flow 215
6.7.1 Hydrodynamic equations, 215
6.7.2 Convection-diffusion equation, 216
6.7.3 Solutions for unsteady flow, 216
6.8 Performance of receiving-water quality simulation models 217
6.9 Summary 218
Appendix 6.1 Rate coefficients of principal BOD decay processes 219
Appendix 6.2 Predictive models for reaeration coefficient 221
7 One-dimensional Models for Simulation of Water Quality in Lakes and

Reservoirs 227
G. T. Orlob
7.1 Need for mathematical models 227
7.2 One-dimensional approximation 228
7.3 Brief review of development 230
7.3.1 Early attempts to simulate lake temperatures. 230
7.3.2 Quantification of heat exchange, 231
7.3.3 First attempts at one-dimensional temperature modeling, 232
7.3.4 Extensions to segmented impoundments, 233
7.3.5 Dual-purpose one-dimensional models, 233
7.3.6 One-dimensional destratification simulation, 234
7.4 Conceptual representation of a one-dimensional lake or reservoir 234
7.5 Formulation for temperature prediction 236
7.5.1 Conservation ofmass, 236
7.5.2 Conservation ofheat energy. 237
7.5.3 Heat budget for an element, 237
7.5.4 Evaluation ofheat budget terms, 238
7.5.5 Effective diffusion. 240
7.5.6 Dimictic lakes: simulation of the freeze-thaw cycle, 242
7.6 Techniques for solution of one-dimensional temperature models 243
7.6.1 Implicit finite-difference method, 243
7.6.2 Explicit finite-difference method, 244
7.6.3 Finite-element method, 245
xv
7.7 Applications of one-dimensional temperature models of lakes and
reservoirs 245
7.7.1 Fontana Reservoir, North Carolina, 245
7.7.2 Hungry Horse Reservoir, Montana, 247
7.7.3 Lake Piiijiinne, Finland, 248
7.7.4 Ross Lake, Washington, 248
7.8 Comments on temperature simulation 250
7.9 Approach to one-dimensional modeling of water quality and ecology
in lakes 251
7.9.1 Brief review of development, 252
7.10 Conceptual representation of one-dimensional water-quality-
ecological models 253
7.11 Formulation of water-quality--ecological equations 253
7.11.1 Formulation for oxygen balance, 254
7.11.2 Formulation of the equation set, 255
7.12 Techniques for solution of one-dimensional water-quality--ecological
models 258
7.12.1 Implicit finite-difference method, 258
7.12.2 Explicitfinite-dif{erence method, 259
7.13 Other one-dimensional water-quality--ecological models 259
7.13.1 Finite-element model-Bam and Arnett, 259
7.13.2 Dynamic reserl'Oir water quality model-Imberger et aI., 261
7.13.3 Waterways Experiment Station model, 263
7.14 Applications of one-dimensional water-quality-ecological models 264
7.14.1 Lake Washington, Washington, 264
7.14.2 Wellington Reservoir, Australia, 264
7.14.3 Lake Hartwell, Georgia, 268
7.15 Concluding comments 268
8 Two- and Three-dimensional Mathematical Models for Lakes and

Reservoirs 274
M. Watanabe, D. R. F. Harleman, and O. F. Vasiliev

8.2 General characteristics of hydrothermal circulation models and
governing equations 276
8.2.1 Introduction, 276
8.2.2 Governing equations, 276
8.2.3 Turbulent shear and diffusion, 281
8.2.4 Eddy viscosity and eddy diffusil'ity, 283
8.2.5 Surface shear expressions, 285
8.2.6 Surface heat/lux. 286
8.3 Models of wind-driven circulation 288
8.3.1 Vertically averaged two-dimensional circulation models
(single-layer models), 289
8.3.2 Two- and multi-layer models, 294
XVI
8.3.3 Ekman-type models, 296

8.3.4 Laterally averaged two-dimensional/low models, 298
8.3.5 Three-dimensional models, 300
8.3.6 Concluding comments, 308
8.4 Two- and three-dimensional water quality and ecological models 309
8.4.1 Compartment models, 309
8.4.2 Network ofone-dimensional channel models, 312
8.4.3 Two- and three-dimensional models, 312
8.5 Mathematical models of cooling impoundments 316
8.5.1 Classification of cooling impoundments, 317
8.5.2 Distinction between stratified and vertically mixed
impoundments, 318
8.5.3 Vertically mixed cooling ponds, 321
8.5.4 Stratified cooling ponds, 325
9 Ecological Modeling of Lakes 337

S. E. JtJrgensen
9.1 Background 337
9.1.1 Water quality and ecological models of lakes, 337
9.1.2 Brief review of development, 338
9.2 Ecological lake models 338
9.2.1 Coupling ofsubmodels, 338
9.2.2 Nutrient budget models, 344
9.2.3 Multiconstituent models, 353
9.3 Some illustrative case studies 365
9.3.1 Application of case studies, 365
9.3.2 Dvre Heimdalsvatn, Norway, 366
9.3.3 Glums¢ S¢, Denmark, 367
9.4 Concluding comments 370
9.4.1 Discussion and conclusions, 370
9.4.2 Future research needs, 373
Appendix. Five lake models 373
10 Modeling the Distribution and Effect of Toxic Substances in Rivers and

Lakes 395
S. E. J¢rgensen
10.2 Processes and related parameters 397
10.2.1 General considerations, 397
10.2.2 Volatilization from water, 397
10.2.3 Sorption, 399
10.2.4 Chemical oxidation, 400
XVll
10.2.5 Photolysis, 400

10.2.6 Hydrolysis, 401
10.2.7 Ionization and complexation, 401
10.2.8 Biodegradation, 402
10.2.9 Interaction between biota and environment, 402
10.2.10 Migration of trace metals across water-sediment interface, 403
10.3 Models 411
10.3.1 Food chain/web models, 411
10.3.2 Simplifications, 414
10.3.3 Models considering a single trophic level, 417
10.4 Concluding discussion 420
II Sensitivity Analysis, Calibration, and Validation 425

M. B. Beck
11.2 Sensitivity analysis 426
11.2.1 A case study: Berkel River, Netherlands, 430
11.3 Calibration and verification 433
11.3.1 Parameter estimation, 434
11.3.2 Off-line estimation algorithms, 436
11.3.3 Recursive estimation algorithms: an example, 438
11.3.4 Some recent applications of parameter estimation, 447
11.3.5 Two case studies, 447
11.3.6 Verification, 458
II .4 Validation 459
11.5 Summary and conclusions 460
12 Models for Management Applications 468

D. P. Loucks
12.2 Management alternatives for water quality control 469
12.3 Management objectives and quality standards 470
12.4 Water quality control alternatives 475
12.4.1 Wastewater disposal on land, 475
12.4.2 Regional treatment and transport, 480
12.4.3 Multiple point source waste reduction to meet water quality
standards, 482
12.4.4 Flow augmentation, 486
12.4.5 Artificial in-stream aeration, 489
12.5 Lake quality management 491
12.5.1 Constant-volume well mixed lakes, 491
12.6 Predicting algal bloom potential 493
12.7 Concluding remarks 498
xviii
13 Future Directions 503
G. T.Orlob
13.1 Achievements in modeling water quality 503

13.2 Areas for improvement 505
13.2.1 Model reliability, 505
13.2.2 Data and data management, 506
13.2.3 Models and submodels, 506
13.2.4 First principles l'ersus empiricism, 507
13.2.5 Hydrodynamics in relation to water quality, 507
13.2.6 One, two, and three dimensions, 507
13.2.7 Kinetics and stoichiometry, 508
13.2.8 Documentation, 508
13.2.9 Register for software, 508
13.3 Conclusion 509
Index 511
Notation for Book
A cross-sectional area; algal biomass

A eddy viscosity (tensor)
A, affinity of rth chemical reaction
Bk Pk/P, mass fraction of biomass of kth biotic component
cp specific heat at constant pressure
C concentration
E exchange (bulk dispersion) coefficient
Ek rate of excretion by kth biocomponent
Eu Euler or Ruark number
f Coriolis parameter
F external force
Fr Froude number
9 gravitational acceleration
Ge Gibbs free enthalpy (per unit mass)
Gk grazing rate of kth biocomponent
h density of enthalpy
H height; depth
I light intensity
J diffusion flow
kB Boltzmann constant
K coefficient
L length; concentration of organic matter
vlt j molar mass ofjth chemical constituent
Mk nonpredatory mortality rate of kth biocomponent
nj number of moles of jth chemical constituent
N nitrogen concentration
Nj p)p, mass fraction ofjth chemical constituent
XIX
xx
p pressure
P phosphorus concentration
P dissipative component of pressure tensor (viscosity)
P k rate of production of kth primary producer
Pr Prandtl number
Q flow rate (downstream)
r space vector
Rk respiration rate of kth biocomponent
R o universal gas constant
Re Reynolds number
Ri Richardson number
Ro Rossby number
St Strouhal number
t time
T temperature
u density of internal energy
u, v, w velocity components
U QIA, downstream or wind velocity in one-dimensional flow
v velocity
V volume
w, ~" rate of rth chemical reaction
W vector of heat flow
x, y, z space coordinates
Z zooplankton concentration
IX lip, specific volume
,
!i.Laplace operator
free elevation of surface
'1 light extinction coefficient
() temperature coefficient
A wavelength of light
fl growth rate
flj chemical potential of jth chemical constituent (per unit mass)
v kinematic viscosity
Vj, stoichiometric coefficient ofjth chemical constituent in rth chemical
reaction
~, extent of rth chemical reaction
p(p) mass density (ofjth chemical constituent)
r shear stress
cP potential of external forces (e.g. gravity)
\¥ velocity potential (v = grad \¥)
w angular velocity.
Chapters 2-4 and 6-12 have individual lists of extra notation.

Introduction
1 G. T. Or/ob
1.1. WATER QUALITY IN PERSPECTIVE
Water as a natural resource has been a primary concern of man throughout

his existence. Only recently, however, has he learned that it is limited, not only in
quantity, but also in its capacity to assimilate the waste products of human
activity without endangering life itself.
Our earliest records acknowledge his requirements for a continuous source
of fresh, drinkable water and his dependence on the natural hydrological
cycle, which in its extremes could severely restrict, even preclude, life processes.
Agricultural societies developed and flourished where the natural cycle of
flooding brought new soil and nutrient to the land, but they also failed for want
of water of sustained high quality. Salinization of ground- and surface-waters
is believed to have been a primary cause of the demise of ancient agricultural
centers in the Fertile Crescent ofthe Middle East,just as it threatens agricultural
productivity in many parts of the world today.
While the gross differences in quality of natural waters-between sea- and
rain-waters, for example-were appreciated by early civilizations, the processes
resulting in changes of quality were not. It was not until the late nineteenth
century that a reasonably correct representation of the hydrological cycle was
advanced. A consciousness of quality was not developed until the connection
between enteric diseases in man and water pollution was recognized.
The identification of disease-producing organisms and of their transport
from source to receptor by water stimulated rapid development in the tech-
nology of water treatment and, as a matter of historical record, brought about
a sharp decline in the incidence of enteric diseases in those parts of the world
where this technology was practiced. At the same time, monitoring the quality of
natural waters for evidence of pollution became an essential part of disease
2
prevention. Analytical techniques were devised for measurement of specific
constituents in solution or suspension. With these tools in hand, scientists and
engineers began to inquire more deeply into the phenomenological behavior
of natural water bodies, especially freshwater sources-the streams, lakes, and
reservoirs with which this book is concerned.
1.2. THE BEGINNINGS OF MODELING
Stimulated by the need to control pollution of the major sources of freshwater

supply and to ensure protection of public health, sanitary engineers were
probably among the first to examine quantitatively the physical, chemical,
and biological responses of streams to loadings of nutrients, either natural or
man-induced. In the 1920s the Ohio River Commission in the United States began
an intensive study of sources of pollution and their impacts on domestic water
supply. From this investigation emerged one of the first mathematical models of
an aquatic environment, if not the first-the Streeter-Phelps equation de-
scribing the balance of dissolved oxygen in a stream (Streeter and Phelps, 1925).
This comparatively simple formulation recognized that the change in the
dissolved oxygen deficit below saturation in a stream with a steady discharge
could be represented by the sum of two processes: uptake of oxygen from
solution by the biochemical oxidation of dissolved and suspended organic
matter, and mass transfer of oxygen from the air to the turbulent stream
through the air-water interface. It was stated as a first-order ordinary dif-
ferential equation that, when integrated, provided an analytical solution
describing the dissolved oxygen deficit as a function of time (or days of flow)
from the point of discharge of a steady source of biodegradable pollutant. This
solution, called the "oxygen sag equation," was to command the attention
(even to the extent of preoccupation) of sanitary engineers for the next several
decades. Probably no single" model" in the history of our artt has been the
subject of more intensive study. Investigators sought to determine biodegrada-
tion and reaeration coefficients, both in the laboratory and in the field, and to
obtain even more elegant analytical solutions to the original differential
equation embellished with a variety of conceptual complications that arose
from an increased awareness of the behavior of the real system. A considerable
amount ofliterature grew out ofthese efforts, which are summarized in Chapter 6.
Yet, until the advent of modern computational hardware and techniques after
the Second World War, the Streeter-Phelps model was seriously restricted in
its usefulness. The necessary requirements for closed-form analytical solutions
precluded treating many cases that were considered essential to realistic
description of the water body.
t Or .. science:" if the reader prefers.

3
Development of the computer and the mathematical techniques that
accompanied it, especially numerical methods for solving previously intractable
problems, quickly effected, inter alia, a change in water resource technology.
In the relatively short span of five years or so in the late 1950s, methods were
developed for solving large sets of simultaneous algebraic equations and finite-
difference representations of more complex linear and nonlinear differential
equations. Members of the engineering professions quickly recognized the
potential ofthe methods to reduce the burden of work and to expedite solutions
to the more difficult problems posed by a society that was becoming ex-
ponentially more complex.
In this same period the rise in public concern for protection and enhance-
ment of the aquatic environment resulted in large expenditures for pollution
control through advanced wastewater treatment technology. Seeking to
evaluate the benefits of such measures in terms of the quantitative response of
the aquatic environment, engineers began to draw on the computer sciences
for tools to solve traditional problems, like that posed in the Streeter-Phelps
formulation.
Among the first models ofthis new era in water management was the Delaware
Estuary Comprehensive Study model, formulated by Thomann (1963) and his
associates for the Federal Water Pollution Control Administration in the
United States. The DECS model was an important extension of the classical
Streeter-Phelps equation to a case where multiple waste loads of varying
strengths were distributed along a narrow estuary of nonuniform cross section,
and where the rates of biodegradation and reaeration could be expected to
vary spatially and temporally with hydrological conditions. The uniqueness of
the model lay in its capability for rapid, quantitative assessment of alternative
strategies for pollution control in terms of water quality (of dissolved oxygen in
this example). This capability was only realizable because of advances in com-
puter hardware, which have continued unabated.
1.3. GROWTH OF WATER QUALITY MODELING
In the late 1960s the mounting public pressure in the United States for control
of pollution stimulated investment in a host of special studies to find the best
alternative ways of protecting the aquatic environment. The computer and the
computer sciences, especially mathematical modeling, were stressed by zealous
practitioners of the modeling art as aids to decision makers. Encouraged by
even the modest success of the DECS experience and a few others, governmental
agencies began to invest heavily in this new technology.
During this period the comparatively simple Streeter-Phelps model of
dissolved oxygen concentration and biochemical oxygen demand (DO-BOD)
appeared in a variety of computerized forms. The model DOSAG developed
4
by the Texas Water Development Board (1970) solved the steady state problem
for a multisegment river system, providing flexibility in dealing with variable
deoxygenation coefficients. QUAL I, also produced for the Texas agency
(Masch and Associates, 1970), simulated stream temperature as well as DO
and BOD, allowing temperature adjustments in rate coefficients to be made
internally during simulation. An extended version, QUAL II, produced for
the US Environmental Protection Agency (Water Resources Engineers, Inc.,
1973), included the capability to simulate more complex stream systems for
both steady and unsteady flow and to evaluate the impacts of nutrient loading
on the oxygen resources of the stream. Ecological interactions, including
photosynthesis and primary productivity, have been the subjects of the most
recent work on deterministic modeling of streams (Kelly, 1975; Stehfest, 1977).
Mathematical modeling of lakes was stimulated by two environmental
concerns: the impact of reservoir releases on downstream water quality, and the
nutrient enrichment of impoundments by tributaries. Applying the principle
of heat energy conservation, Water Resources Engineers, Inc. (WRE, 1968) and
a group at the Massachusetts Institute of Technology (MIT: Huber et aI., 1972)
developed models of the reservoir as a one-dimensional system of horizontal
slices to simulate the vertical distribution of heat in the impoundment over an
annual cycle. These models were extended to consider other quality constituents,
including DO, BOD (Markofsky and Harleman, 1973), nutrients, and even
biota (Chen and Orlob, 1975), and to predict the quality of downstream
releases from operating reservoirs. To obtain better agreement between model
and impoundment, various improvements were made: by representing inflow
and withdrawal more realistically, by better formulation of heat exchange at the
air-water interface (Tennessee Valley Authority, 1972), and by accounting
for detention time within volume elements (Fontane and Bohan, 1974). In
addition to the finite-difference explicit and implicit formulations used in the
WRE and MIT models, the one-dimensional lake model was expressed in
finite-element form by Baca and Arnett (1976).
The eutrophication of lakes such as Zurich See and Lake Erie drew the
attention in the late 1960s of biological researchers, who proposed fairly
simple models of nutrient enrichment oflakes (Vollenweider, 1969). They used
the models to focus on gross nutrient balances in two-layer (epilimnion-
hypolimnion) systems and to explore the cause-effect relationships between
carbon, nitrogen, and phosphorus inputs and primary productivity (O'Melia,
1974; Imboden, 1974; Vollenweider, 1975). In a somewhat parallel modeling
activity stimulated by the International Biological Program, more intricate
ecological models of lakes, like CLEANER (Park et aI., 1974), were developed.
Initially, these modeling efforts concentrated on ecological correctness and less
attention was given to the physical behavior of the lake system, e.g. circulation,
mixing, and stratification. More recently, this generation of models has been
adapted, as in the one-dimensional temperature models and LAKECO (Hydro-
5
logic Engineering Center, 1974), to multisegment systems, for example,
MS CLEANER (Leung et al., 1978). The ecological conceptualizations of such
models were the subject of continuing research and improvement throughout
the 1970s (Straskraba, 1973; J/2Jrgensen, 1976).
Modeling of large lakes, where transport processes are often governed by
wind stresses, was necessarily concentrated first on representing circulation and
mixing phenomena. The models evolved from the comparatively simple, single-
layer storm surge predictors (Platzman, 1963) to multilayer, multisegment
models based on the phenomenological equations of motion in two and three
dimensions (Simons, 1973). Applications of these models to large wind-driven
lakes like Lake Ontario and Lake Vanern (Simons et aI., 1977) have provided
essential input for modeling the water quality of such systems. t
Notwithstanding the limits to simulation of circulation processes in large
lakes, model developers have not hesitated to tackle problems of describing the
water quality and ecological behavior of such water bodies. Given a capability
to describe the principal components of wind-driven circulation in a com-
patible hydrodynamic model, Lam and Simons (1976) devised an advection-
diffusion water quality model of Lake Erie. Patterson et al. (1975) adapted a
link-node estuary model to Green Bay in northeastern Lake Michigan, driving
it with circulation derived from a two-dimensional, orthogonal-mesh, finite-
difference model patterned after that of Leendertse (1967). Di Toro et al. (1975)
developed a seven-segment-seven-constituent Lotka-Volterra type of primary
production model for western Lake Erie. A so-called "comprehensive" model
for Lake Ontario was developed by Chen et al. (1975), who used an approach
similar to that employed in LAKECO (Chen and Orlob, 1975). It was driven
by a three-dimensional circulation derived from a finite-difference hydro-
dynamic model and employed 15 state variables ranging from nutrients to
fish. The model conceptualized Lake Ontario as a system of seven layers over
the maximum depth and included more than 200 segments.
1.4. PROLIFERATION OF MODELS AND THE NEED FOR

TECHNOLOGY TRANSFER
In the late 1970s the technology of modeling, at least that of water quality,
reached a point where advancement seemed to depend more on the availability
of reliable data from the field than on the ingenuity of the modeler or on the
t Parallel development of models of the hydrodynamics of shallow estuaries and coastal seas has
contributed greatly to the technology of modeling large and complex water bodies.
6
computer. The focus shifted toward implementation of the model as a tool,
requiring that it be calibrated and validated against the water body. It was
necessary to determine the sensitivity of the model in terms of realistic measures
of the system response, e.g. dissolved oxygen and phytoplankton production,
if the model was to be accepted as an adjunct to the decision-making process.
It was clear, also, that there had been a proliferation of models, many of
which were merely instruments of research, often ends in themselves and of
little value in the management of aquatic resources. A redundancy was apparent.
While models were usually original, reflecting the "artistic" abilities of the
designers, there was a tendency to "reinvent the wheel" rather than to make
true advances in modeling. Moreover, there was a wide disparity in docu-
mentation, a step in the model development process that is absolutely essential
if the utility of the model is to be realized by other than the developer. Transfer of
modeling technology had been sadly neglected in the rush to develop.
It is in the nature of modeling, as it has been and is currently practiced,
that the documentation necessary for effective technology transfer rarely
appears in the literature. A survey of reference sources conducted by the
author in 1976 revealed that, of some 400 documents dealing with mathematical
modeling of surface water impoundments, about 90 percent were of limited
circulation, e.g. technical reports, conference proceedings, and technical
memoranda. In addition, the volume of information of relevance to a topic
like water quality modeling is too large for one to digest without proper
organization and dissemination.
In early 1976 the author proposed the development of a centralized clearing
house, or "register," for the coordination and transfer of technology related to
computer software for planning and managing water resources. The concept
was presented to IIASA through Professor Dr. Z. Kaczmarek and developed
further by the author while in Laxenburg as a visiting scientist in May 1976.
On this occasion a project within the scope of the IIASA state-of-the-art survey
was identified, outlines were prepared for a series of monograph on water quality
modeling, and a committee was formed to prepare this volume on Streams,
Lakes, and Reservoirs.
During late 1976, under the sponsorship of the Technology Transfer Program
of the Office of Water Research and Technology (OWRT), US Department of
the Interior, the author studied the feasibility of the proposed software register.
The work included a survey, through 1976, of the state of the art of mathe-
matical modeling of surface water impoundments, the development of a format
for documentation of software for a national register, an assessment of user
potentials, and a determination of the feasibility of an institutional structure for
technology transfer. A two-volume report comprising a narrative description
and review of capabilities of the principal useful models identified in the survey,
a bibliography of some 400 pertinent references, and specifications for about 90
working models was presented to OWRT (Orlob, 1977). The report served as a
7
source document for the present work, which was greatly expanded to provide
the necessary breadth of international coverage.
1.5. SCOPE OF THIS VOLUME
The original prospectus for consideration by IIASA discussed a series of

monographs dealing with water quality modeling, each encompassing a part,
or several parts, of the aquatic environment. This first volume is limited to
modeling water quality in streams, lakes, and reservoirs. However, simply
because it is the introductory volume and should set the tone for future works
in the same subject area, it gives special attention to certain topics of a general
nature, e.g., those concerned with the modeling process itself and with the basic
principles of modeling water quality. Overall, however, the emphasis is on
models, especially those that are of demonstrable utility in the management of
water quality and are well documented, i.e. in a suitable condition for effective
technology transfer.t
Having summarized our work, it is as well to indicate how we expect the
reader to use this book, before we proceed with the details. We have tried to
assemble and to describe carefully the best examples of the modeling art. Aside
from the chapters of general interest to modelers, i.e. the modeling process and
basic principles, each chapter presents at the outset a summary of historical
highlights. Each chapter tends to follow the general sequence by which models
are developed, i.e. conceptualization, formulation, computational representa-
tion, solution, calibration, validation, and application. Because our bias is
toward truly useful models, we have tried to give concrete examples of their
applications, normally showing comparisons between simulated and observed
measures of response. Each chapter concludes with a candid summary, by each
author, of the capabilities and limitations of models in the category discussed.
Chapter 2, prepared by Dr. Bruce Beck, describes the modeling process in
a stepwise fashion. It begins with some essential basic definitions of technical
terminology used by modelers. The procedure whereby the essence of an
aquatic environment is captured conceptually, described in mathematical
form, and translated into the language of computation is outlined. The nature
and importance of calibration, validation, and sensitivity testing are emphasized,
although a detailed exposure of these topics is deferred to Chapter 11.
Chapter 3, by Professor Peter Mauersberger, lays the foundation for de-
terministic water quality modeling with a comprehensive treatment of the
t This work does not really present the "state of the art" as such, but merely the best effort to
capture the essence of mathematical modeling of water quality as experienced by the individual
authors.
8
general physical, chemical, and biological principles. The coverage is mathe-
matically rigorous, yet the presentation emphasizes the general rather than the
special case.
Chapter 4, written by Professor S. E. Jorgensen, follows somewhat the
pattern of Chapter 3, in that basic ecological principles are identified, but
it endeavors as well to provide a broad coverage of the bases for translating
these principles into mathematical terms.
Chapter 5, by Dr. J. Jacquet, examines the modeling of the thermal regime in
streams. It includes the effects of multiple heat sources, diurnal heating and
cooling, longitudinal and lateral mixing, stratification, and the stochastic
nature of the heat exchange process. The appendix includes the most current
quantitative statement of the component heat fluxes at the air-water interface.
Chapter 6, prepared by Dr. M. J. Gromiec, Professor D. P. Loucks, and the
editor, deals with the classic problem of simulation of water quality in streams,
ranging from the historical foundation in the Streeter-Phelps equation to the
more elaborate predictive models, like QUAL II, that deal with a full range of
biological parameters. The appendices summarize the extensive research on
formulation of deoxygenation and reaeration rates.
Chapter 7, by the editor, reviews the development of one-dimensional models
of water quality in stratified impoundments. The chapter presents the basic
formulation for the one-dimensional temperature model and the implicit
solution technique for the resulting set of energy balance equations. Extension
of the model to treat multiple quality and biological state variables is described.
Illustrative examples are given of simulation of temperature and dissolved
oxygen in stratified reservoirs.
Chapter 8, by Dr. M. Watanabe, Professor D. R. F. Harleman, and Professor
O. Vasiliev, reviews the literature in modeling the hydrodynamic and water
quality processes of two- and three-dimensional lake systems. Formulations
based on the Navier-Stokes equations are given, solution techniques are de-
scribed, and the interrelationships between hydrodynamics and water quality
are emphasized. Illustrative examples of two- and three-dimensional modeling
of water quality in large lakes are presented. A special section deals with the
unique problem of thermally stratified plumes and cooling ponds.
Chapter 9, by Professor S. E. J0rgensen, provides a comprehensive examina-
tion of techniques for modeling the ecology of lakes and reservoirs. A discussion
is presented of the many variants in the formulation of interactions between
nutrients, biota, and environmental parameters that govern rates of growth,
respiration, mortality, etc. The performance of ecological models is demon-
strated by examples selected from actual case studies.
Chapter 10, also by Professor Jf2Jrgensen, presents the latest experience in
modeling the fate of toxic substances and evaluating their effects on aquatic
biota.
Chapter 11, prepared by Dr. Bruce Beck, is an extension of material presented
9
in Chapter 2 on the modeling process, dealing specifically with sensitivity
analysis, calibration, and validation. The theoretical bases for these essential
steps in modeling the aquatic environment are outlined. Illustrative examples are
presented.
Chapter 12, by Professor D. P. Loucks, describes the use of models in water
quality management. Methods are presented for the evaluation of alternative
wastewater treatment methods in relation to management objectives and
quality standards. The role of deterministic water quality models of streams and
lakes is discussed.
The final chapter, by the editor, summarizes the status of mathematical
modeling of water quality of streams, lakes, and reservoirs as this group of
authors has presented it. In particular, we try to highlight the major practical
accomplishments of the modelers and to bring the real limitations of models to
the attention of potential users. Some of the more notable deficiencies in
water quality modeling at present are identified, and we attempt, with some
trepidation, to identify the most promising avenues for future research and
development.
REFERENCES
Baca, R. G. and R. C. Arnett (1976). A finite-element water quality model for eutrophic lakes.
Finite Elements in Water Resources vol. 4, eds. W. G. Gray, G. Pinder, and C. Brebbia (Plymouth,
Devon: Pentech), pp. 125-147.
Chen, C. W., M. Lorenzen, and D. J. Smith (1975). A comprehensive water quality-ecologic model
for Lake Ontario. Report to Great Lakes Environmental Research Laboratory, National Oceanic
and Atmospheric Administration by Tetra Tech, Inc., 202 pp.
Chen, C. W., and G. T. Orlob (1975). Ecologic simulation of aquatic environments. Systems
Analysis and Simulation in Ecology vol. 3, ed. B. C. Patten (New York, NY: Academic Press),
ch. 12, pp. 475-528.
Di Taro, D. M., D. J. O'Connor, R. V. Thomann, and J. L. Mancini (1975). Phytoplankton-zoo-
plankton-nutrient interaction model for western Lake Erie. Systems Analysis and Simulation in
Ecology vol. 3, ed. B. C. Patten (New York, NY: Academic Press), ch. II, pp. 423-474.
Fontane, D. G., and J. P. Bohan (1974). Richard B. Russell Lake water quality investigation.
Waterways Experiment Station, Vicksburg, MS Technical Report H-74-14, 26 pp.
Huber, W. c., D. R. F. Harleman, and P. J. Ryan (1972). Temperature prediction in stratified
reservoirs. Proceedings of American Society of Civil Engineers, Journal of Hydraulics Division,
Paper 8839. pp. 645-666.
Hydrologic Engineering Center (1974) (revised 1978). WQRRS. generalized computer program for
river-reservoir systems. US Army Corps of Engineers. HEC, Dat'is, CA' User's Manual 401-100,
100A, 210 pp.
Imboden. D. M. (1974). Phosphorus model for lake eutrophication. Limnology and Oceanography
19:297-304.
Jorgensen, S. E. (1976). A eutrophication model for a lake. Journal ofEcological Modeling 2(2): 147-
165.
Kelly. R. A. (1975). The Delaware Estuary Model. Ecological Modeling in a Resource Management
Framework, ed. C. S. Russell. Resourcesfor the Future Working Paper QE-I (Baltimore. MD:
Johns Hopkins University Press), pp. 103-134.
10
Lam, D. C. L., and T. J. Simons (1976). Numerical computations of advective and diffusive trans-
ports of chloride in Lake Erie, 1970. Journal of Fisheries Research Board of Canada 33: 537-549.
Leendertse, J. (1967). Aspects of a computational model for well mixed estuaries and coastal seas.
Rand Corporation, Santa Monica, C A Research Memorandum 5294-PR.
Leung, K. K., R. A. Park, D. J. Desormeau, and J. Albanese (1978). MS CLEANER-An overview.
Proceedings of Conference on Modeling and Simulation. Pittsburg, PA (Troy. NY: Center for
Ecological Modeling, Rensselaer Polytechnic Institute), 8pp.
Markofsky, J., and D. R. F. Harleman (1973). Prediction of water quality in stratified reservoirs.
Proceedings of American Society of Civil Engineers, Journal of Hydraulics Division 99(HY5),
Paper 9730, pp. 729-745.
Masch, F. D., and Associates (1970). QUAL I, simulation of water quality in streams and canals,
program documentation and user's manual. Report to Texas Water Development Board, Austin,
TX.
O'Melia, C. R. (1974). Phosphorus cycling in lakes. North Carolina Water Resources Research
Institute, Raleigh Report 97, 45 pp.
Orlob, G. T. (1977). Mathematical modeling of surface water impoundments, vols. I and II. Report
to Office of Water Research and Technology by RMA, Inc., LaFayette, CA, 119 pp., appendix,
and references.
Park, R. A., D. Scavia, and N. L. Clesceri (1974). CLEANER, the Lake George model. US Inter-
national Biological Proqram, Eastern Deciduous Forest Biome Contribution J 84, 13 pp.
Patterson, D. J., E. Epstein, and J. McEvoy (1975). Water pollution investigations: Lower Green
Bay and Lower Fox River. Report to US Environmental Protection Agency, No. 68-01-1572,
371 pp.
Platzman, G. W. (1963). The dynamic prediction of wind tides on Lake Erie. Meteorology Mono-
graphs 4(26), 44 pp. .
Simons, T. J. (1973). Development of three-dimensional numerical models of the Great Lakes.
Canada Centerjor Inland Waters, Burlington, Ontario, Science Series No. 12.26 pp.
Simons, T. J., L. Funkquist, and J. Svensson (1977). Application of a numerical model to Lake
Viinern. Swedish Meteorology and Oceanography Institute Report RH09, 15 pp.
Stehfest, H. (1977). Mathematical modeling of self-purification of rivers. Projessional Paper
PP-77-11 (Laxenburg, Austria: International Institute for Applied Systems Analysis).
Straskraba, M. (1973). Limnological models of reservoir ecosystems. Proceedings of International
Symposium on Eutrophication and Water Pollution Control, Castle Reinhardsbrunn, DDR (Biology
Society, DDR).
Streeter, H. W., and E. B. Phelps (1925). A study of the pollution and natural purification of the
Ohio River. III. Factors concerned in the phenomena of oxidation and reaeration. US Public
Health Service, Washington, DC Bulletin 146,75 pp.
Tennessee Valley Authority (1972). Heat and mass transfer between a water surface and the atmo-
sphere. TVA Water Resources Research Engineering Laboratory, Norris, TN Report 14, 123 pp.
Texas Water Development Board (1970). DOSAG-I simulation of water quality in streams and
canals. Program documentation and user's manual.
Thomann, R. V. (1963). Mathematical model for dissolved oxygen. Proceedings of American
Society of Cil'il Engineers, Journal of Sanitary Engineering Dil'ision 89(SA5): 1-30.
Vollenweider, R. A. (1969). Moglichkeiten und Grenzen elementarer Modelle der Stoffbilanz von
Seen. Archil' fur Hydrobiologie 66: 1-36.
Vollenweider, R. A. (1975). Input-output models. Schweizerische Zeitschriji fur Hydrologie
37(1):53-82.
Water Resources Engineers, Inc. (1968). Prediction of thermal energy distribution in streams and
reservoirs. Report to California Department of Fish and Game, 90 pp.
Water Resources Engineers, Inc. (1973). Computer program documentation for the stream quality
model QUAL II. Report to US Environmental Protection Agency, 90 pp.
A Procedure for Modeling
2 M. B. Beck
2.1. INTRODUCTION
It is difficult to claim that one has available a universal systematic procedure

for model development. This chapter certainly makes no such claim. We are
concerned throughout this book with the state of the art, and art indeed is the
operative word in the matter of characterizing the nature of the modeling
procedure. Most analysts engaged in the development of models for the quality
of water bodies have assembled their own unique collections of techniques for,
and attitudes toward, model building. In this chapter we shall suggest a particular
sequence in the modeling procedure; it is not expected that such a sequence is
one to which rigid adherence is demanded or even possible. More important,
rather, we shall thereby cover the majority of problems associated with model
development. At this stage in the book our attention is merely to outline in
qualitative and nonmathematical terms the principal features of each step in the
modeling procedure. In Chapter 11, a more technical and detailed discussion
of the same topics will be given: in particular, case studies will be reported as
examples of the application of methods of sensitivity analysis, calibration, and
parameter estimation. Chapter 11 will also illustrate the problem of model
validation.
It is first necessary to define the overall problem that motivates the present
discussion of the modeling procedure. This chapter intends, therefore, to answer
the questions: how do we derive from the ensemble of general theory a set of
relationships, i.e. the model, for a specific water body; and how do we demon-
strate the good or poor approximation of the behavior of that model to the
observed behavior of reality? The two questions also define the way in which the
modeling procedure divides into two parts. Up to section 2.5 our procedure
describes the selection and combination of the various expressions required for
11
12
the specific model before any field experiment has been used to test the validity
of the model. From section 2.6 onward the discussion attends to the matter
of model modification after the collection of field data; in other words, the latter
half of the chapter discusses system identification, model calibration, and
validation.
The "before" and "after" character of the modeling procedure refers
additionally to the way in which we might view the a priori and a posteriori
knowledge of the behavior ofa system. Figure 2.1, adapted from Eykhoff(1974),
gives an example of the relationship between the two types of knowledge. The
plentiful sources of error in both our structural and measured knowledge of a
Exp erimental, - "Reality" I---- Physical, chemi cal,

field obs ervations and biological
theories
Partial
Measurement
--.
errors
Measured data
differential
equations
(nonlinear)
- Mo deling
err ors
r-------
I
I
Model I Partial
structure ---,I differential
equations
~ earization
erro rs
determination I
I (linear)
~-------i
I
I
Parameter I Ordinary
Estimation
--.
errors
estimates I differential ~ mping
erro rs
(state estimates) I equations
I
I
L _______ ~
--- Model I--
A posteriori A priori (theoretical)

measurement knowledge structural knowledge
FIGURE 2.1 Combining a priori and a posteriori knowledge in the modeling procedure
(adapted from Eykhoff, 1974).
13
process should emphasize the fact that models are at best working hypotheses
about the nature of reality. There is never a good reason for overconfidence in
the performance of a model. Again, we can use the dual nature of Figure 2.1
and the modeling procedure to illustrate two frequently polarized approaches
to modeling. The first, associated with a priori structural knowledge, is a de-
ductive reasoning approach: from an existing general theory we deduce the
model relationships for the specific case study. The second, associated with
a posteriori measurement knowledge, is an inductive reasoning approach:
assuming no a priori knowledge (theory) of process behavior, we attempt to
develop the specific information acquired from the particular sample set of
data into a more general model. The only valid reason for relying upon one of
these approaches alone may be the inevitable difficulty of obtaining experimental
data. We shall see aspects of this duality reflected throughout the chapter;
they have also been treated in a recent review by Somlyody (1981). Figure 2.2
shows the sequence of the modeling procedure.
Goals and objectives
1
Conceptualization
(section 2.3) I
Sensitivity analysis
1 (sections 2.4.1 and 2.7.1 )
Selection of model type t
(section 2.4)
1
Computational
representation
(section 2.5)
A priori
-----
A posteriori -------1--------- - - _.-
First sample
field data set --- Calibration and verification
(section 2.6)
I--
1
---
Second and
Validation I--
additional
field data sets (section 2.7)
!
Application
FIGURE 2.2 An outline of the modeling procedure and organization of the chapter.
14
There are five basic steps in the sequence:
(i) conceptualization (section 2.3)

(ii) selection of model type (section 2.4)
(iii) computational representation (section 2.5)
(iv) calibration and verification (section 2.6)
(v) validation (section 2.7).
Figure 2.2 indicates also some of the primary feedback, or iterative, loops in the
modeling procedure. It will thus be observed that sensitivity analysis has been
assigned a position of importance both at an early a priori stage and at a later
a posteriori step in two of the feedback loops. Like the rest of the diagram, this
location for sensitivity analysis does not have to be interpreted in any strict
sense. There are as many variants on the theme of Figure 2.2 as there are
authors on the subject of modeling. Much of what follows will be a distillate of
ideas from the works of Orlob (1975), J0fgensen (1978), Young (1978), Rinaldi
et al. (1979), and Beck (1979).
2.2. DEFINITION OF GOALS AND OBJECTIVES
Though perhaps not part of the modeling procedure itself, a common theme
among the majority of statements on the subject of this chapter is that the goals
and objectives for model application determine the nature of the model. As
pointed out in Chapter 1, our attitude is not one of seeking a universal model to
solve, in general, all manner of problems. For example, the definition of objec-
tives might state that the model is intended as a guide to long-term planning by
determining the year-to-year average response of a river system to patterns of
population and industrial growth and movement. It is clearly not sensible to
suggest that the same model would be required to fulfill these objectives as
would be needed to predict the probability of intermittent stream deoxygena-
tion resulting from the diurnal variations of a particular sewage discharge.
Two broad categories of goals and objectives can be distinguished. In
a research context the model has to provide indicators for further fruitful
directions of investigation. An awareness of the immediate use for the model is
not necessary before the study is undertaken. Rather, the concise representation
of a priori and measured information that the model offers and the possibility
for a gain in comprehension (of system behavior) are of primary importance.
An essentially research-oriented model may, nevertheless, be used to make
forecasts about the probable future behavior of the system. In a management
context the immediate application of the model must be known and carefully
specified. "Management" of water quality has traditionally been understood
15
to mean long-range planning, the design of treatment facilities, or the problem
of legislation for discharge consents and standard setting. Applications of
models to such problems will be discussed in Chapter 12. The term management,
however, may also include the design and operation of real-time control and
forecasting systems. Models used for this purpose in water quality management
are at present of lesser significance and are, therefore, not treated in detail in
this book. The reader will find more suitable accounts ofthese kinds of models by
Young and Beck (1974), Rinaldi et al. (1979), and Beck (1980).
2.3. CONCEPTUALIZAnON
Let us assume that we have chosen an objective and have a specific water body
in mind. The first step of the modeling procedure is conceptualization. At this
point the analyst is interested in how the physical system is represented in
three-dimensional space. It is of some importance, for instance, to know the
locations of control structures and tributaries along a river, or to know whether
various portions of a lake can be considered to be essent~ally deep or shallow.
Usually conceptualization will involve a choice regarding the possible (spatial)
segregation of the water body into a number of discrete segments and layers.
Besides a spatial separation of the water body it may be necessary to include a
grouping and differentiation of biotic species according to how one visualizes
their roles in the ecology of the water body. For example, a particular species
of phytoplankton with high nuisance value might best be considered in terms
of a component mass balance that is separate from a component mass balance
for all other phytoplankton species. In addition, though perhaps less obvious,
it might occasionally be appropriate to partition the temporal dimension into
ranges of quickly changing and slowly changing variations. A classic problem of
this kind is the matter of coupling, or decoupling, the hydrophysical and
ecological sectors of a lake system model (J0rgensen and Harleman, 1978;
J0rgensen, 1979). There is, of course, no guarantee that these choices of spatial,
temporal, and ecological aggregation/decomposition will turn out to be
correct at later stages of the modeling procedure. Moreover, as others have
argued eloquently elsewhere (e.g. Young, 1978), we should bear in mind that the
decomposition of a complex environmental system for individual analysis of
its component parts does not always imply that subsequent reassembly will
yield a characterization of the behavior of the whole.
With conceptualization ofthe modeling problem comes also model formula-
tion. The relevant variables for description of the desired water quality character-
istics and expressions for their interaction have to be chosen. The system
definition will have been completed so that in addition we can define in abstract
terms the groups of quantities shown by Figure 2.3. In physical terms we have
the following groups of variables.
16
Unmeasured
input Measurement
disturbances errors
((r,t) ,,1 r,t)
Process
+
Measured +
input c:====~ States c(r,t) 1-------"'-- Measured
disturbances d(r,t)
outputs
Parameters fJ(r,t) CO (r,t)
FIGURE 2.3 Definition of the system and variables.
(i) The group of variables denoted by the vector d, measured input dis-
turbances (or forcing functions), might comprise the recorded day-to-
day variations of total biochemical oxygen demand (BOD), suspended
solids, and ammonia nitrogen concentrations in a treated sewage
discharge to a river.
(ii) The group of variables denoted by I; represents unmeasured (unknown)
input disturbances. These might include items such as yearly rate of
organic phosphorus loading to a lake system from diffuse nonpoint
sources. A predominant characteristic of the unmeasured inputs is
that they will generally be expected to exhibit a random variability.
(iii) The process state variables c characterize the essential properties and
behavior of a process (or system) as functions of space and time.
Previously, in discussions of ecological models, the term compartment
variable has been used frequently to identify state variables. We have
adopted the notation c because for the most part in this book the state
of water quality is described by a vector of component concentration
variables.
(iv) The group of variables CO is defined as measured output variables. In
fact, by and large these variables are merely measurements of some of
the process state variables. However, it is not difficult to imagine a
situation in which CO is not so straightforwardly related to c: for
example, a particulate phosphorus measurement would include
phosphorus bound in algal cells, in zooplankton, and in detrital
material, with all three types of phosphorus binding possibly being
defined as model state variables.
(v) The last category of variables, TJ, represents the random and systematic
measurement errors that derive from process instrumentation and
laboratory analysis; such errors are inherent in all measurements co,
thus prohibiting the possibility of CO being an absolutely exact measure
of c.
17
One further group of variables in Figure 2.3 remaining to be discussed is the
model parameters p: for instance, the reaeration rate coefficient or chemical
kinetic rate constants that appear in the equations of the system model. A
highly desirable property ofthese parameters is that they should be invariant with
respect to time and space, i.e. truly constant. This desirable property will be
shown to be an extremely important feature of certain aspects of model de-
velopment and analysis. Yet we know that the value of a reaeration rate constant
varies with stream discharge and it is quite common that a BOD decay rate
coefficient will also be different for different reaches of a river. One is thus led
to question how constant a parameter ought to be. Indeed, the difference
between a quantity that is a state variable and a quantity that is a parameter
becomes almost negligible when one considers a state variable that does not
vary with time, i.e. when part of the system is at steady state, or a parameter that
displays seasonal, and therefore temporal, fluctuations. To attempt to preserve
the difference between the notions of state and parameter is actually not
particularly useful in the later discussion of the model calibration problem.
Perhaps an ambivalent attitude toward the distinction is desirable: sometimes
the difference between state and parameter is important, and sometimes it i~
not. For these kinds of reason the parameters are specified as functions of (three-
dimensional) space r and time t in Figure 2.3; all other groups of variables
discussed above are likewise defined. The parameters are, in fact, implicitly
functions of other variables, such as temperature and stream flow, and hence
they are functions of time and space.
The values of the parameters may not necessarily be known precisely at the
stage of conceptualization and formulation; it is more likely that the values can
only be said to lie within a certain range according to previously published
results. The principal objective of formulation is the specification of causal
relationships that describe the phenomena thought to govern the behavior of
the system. In other words, this is what we might call a priori model structure
determination. We therefore note the complementary activity of a posteriori
model structure identification, which will be discussed in section 2.6.3.
2.4. SELECTION OF MODEL TYPE
We have distinguished between categories of variables. It is also useful to

distinguish between various types of model and to discuss briefly their character-
istics. In section 2.2 we have actually made a division of models into the groups
of:
(i) "research" or "management" models.
18
Furthermore, Figure 2.1 allows us to classify the following pairs of model types:
(ii) distributed or lumped models;
(iii) nonlinear or linear models.
Some explanation is needed. A distributed model or, commonly, distributed-
parameter model, is one in which variations of all the quantities in Figure 2.3
are considered to be continuous functions of time and space. This form of model
arises rather naturally in the analysis of water bodies, the most familiar example
being the advection-diffusion model for the transport of a dissolved substance
along a stream. Several other examples of distributed-parameter models will
be found in Chapter 8.
A distributed-parameter model accounting for variations in the three
orthogonal directions (x, y, z) is probably the most accurate form of model that
one might use to describe the behavior of quality in a water body. It is also the
most difficult form of model to solve. Frequently, however, the analyst may
decide on the basis of prior experimental observations that horizontal gradients
of dissolved material in a lake, for example, are not sufficiently large to merit
inclusion in the model. Thus let uS suppose that we "lump" together parts of the
system description so that for certain finite volumes of the water body, or within
certain bounded spatial locations, water quality is assumed to be uniform and
independent of position within the defined volume. A distributed-parameter
model would then be reduced by that assumption to a lumped-parameter model.
A typical example of this kind of model is the continuously stirred tank reactor
(CSTR) idealization of lake water quality dynamics. Whereas the lumped-
parameter model is frequently expressed in ordinary differential equation
form, the distributed-parameter model is usually defined by partial differential
equations.
In general, the distributed or lumped models are nonlinear models. A special
case of the general class of nonlinear models is the linear model. Almost always
the analyst strives to obtain a linear system model because of the many powerful
techniques available for comprehensive analysis of such models. The great
advantage of the linear model is that it obeys the principle of superposition.
Hence it is possible to say that, for instance, if the model output response 0A is
related to the input forcing function I A' and likewise 0B is related to I B, then the
combination ofinputs (a'I A + aOI B ) will produce a model response (a'O A + aOO R ),
where a' and aO are proportionality constants. Where appropriate we may
exploit the linearity of a model in order to separate the free response of the
system due only to the initial state of the system, i.e. the general solution of the
differential equation, from the forced responses to the input disturbances, i.e.
the particular solution of the differential equation. The Streeter-Phelps model
of stream BOD-DO interaction (Streeter and Phelps, 1925) is a linear model;
the Lotka-Volterra model of predator-prey interaction is a typical nonlinear
model.
19
Another pair of complementary model types may be defined as:
(iv) stochastic or deterministic models.
A stochastic model of system behavior would reduce to a deterministic model

provided that the stochastic input disturbances ~ and the random measurement
errors 1') are assumed to be zero, i.e.
~(r, t) = 0 and 1')(r, t) = 0
for all r, t in Figure 2.3, and provided that the parameters are known exactly
(as opposed to being estimates stated in terms of statistical distributions). It is
worth remembering that the assumption of a deterministic model is tantamount
to the assumption that one has perfect knowledge of the behavior of a system.
In other words, the future response of the system is completely determined by a
knowledge of the present state and future measured input disturbances. Al-
ternatively, it has been said that (Papoulis, 1965) "probabilistic considerations
are necessary only because of our ignorance," a profound statement indeed if
we infer from it that all natural phenomena are deterministic. Stochastic models
are not treated in depth in this book-except for their application in stream
temperature prediction (Chapter 5)-and thus the reader is referred elsewhere
for the development of such models (e.g. Loucks and Lynn, 1966; Thayer and
Krutchkoff, 1967; Padget et al., 1977; Tiwari et al., 1978; Finney et al., 1982).
Our next classification of models, that between
(v) dynamic or steady state models,
makes one of the most significant distinctions that can be drawn, since it often
furnishes a dividing line in choosing the type of model best suited to particular
problems. Usually the analyst will be interested in approximating a distributed-
parameter description to a steady state model by assuming all variables and
parameters to be independent of time t. A classical example of the use of a
steady state model is that in which the average spatial variations of quality in a
river system are computed for an average time-invariant set of wastewater
discharge, temperature, and stream flow rate conditions. Strictly speaking, if
all variables but the stream discharge are held constant with time, then one is
dealing with a dynamic model; a time-varying discharge implies that water
quality at any fixed spatial location is not at a steady state. Like the lumped-
parameter model, the advantage of the steady state model is its potential for
simplifying subsequent computational effort through the elimination of one
of the independent variables in the model relationships. A typical dynamic
model application is that of examining over a period of years the response of a
lake ecological system after installation of nutrient removal treatment at an
adjacent wastewater treatment plant.
20
In the approximation of a lumped-parameter model, where behavior is
said to be uniform across a discrete portion of three-dimensional space, and in
the approximation of a steady state model, where behavior is effectively said
to be uniform over a discrete interval of time, the real questions ofjudgment are:
how great are the variations in time and space of the water quality characteristics
relevant to the problem; and how many of those individual variations can be
approximated by constants or neglected?
So far we have discussed models that are understood to be characterized by
a set of differential equations. Difference equation forms also arise readily in
water quality modeling. For instance, in practice it is rather atypical to have
available continuous-time (analog) measurements of process variables, just as
it is improbable to imagine a spatial continuum of measurements. Partly for
these reasons, namely that we shall often want to compare a model prediction
with discrete-time (digital, sample) or discrete-point measurements, partly
because numerical solutions to differential equations are obtained through
difference equation approximations, and partly because analytical solutions are
sometimes available that relate variables at discrete points of time and space
to variables at other discrete points, we shall thus refer frequently to difference
equation models.
Last, but not least, we introduce the classification of models into:
(vi) internally descriptive (mechanistic) or black box (input-output)

models.
It would be incorrect to suggest that the analyst has to make a choice between
one or the other of these model types; it is better to view them as defining the
two ends of the spectrum of models (Karplus, 1976). Here again we see the
duality of the modeling procedure. An internally descriptive model is closely
associated with a priori information and with a deductive reasoning process;
the black box model is much more naturally oriented toward a posteriori
information and inductive reasoning. One might say that the internally de-
scriptive model characterizes how the inputs are connected to the states and
how, in turn, the states are connected to each other and to the outputs of the
system. In contrast, the black box model reflects only what changes the input
disturbances will effect in the output responses. Of course, the major difference
between these two categories of models is that one of them, the internally
descriptive model, provides a description of the internal mechanisms, i.e. c and IJ,
of process behavior. The black box model makes no such explicit reference to
what is inside the process block of Figure 2.3; it deals only with what is measur-
able: the inputs and outputs.
The association of black box models with input-output relationships
allows us also to mention the subject of transfer function models and frequency
response analysis. Differential equations may be transformed into algebraic
21
equations by the application of Laplace transforms. It is customary to refer to
this as a transformation from time-domain analysis, in which time is the inde-
pendent variable (in the differential equation), to analysis in the frequency
domain. A transfer function model appears then as the ratio of the Laplace-
transformed output variable to the Laplace-transformed input variable, i.e. a
transfer function can be derived for each pairwise combination of an input and
an output variable. When the input forcing variable of the system exhibits
sinusoid-type variations (in time) it is possible to observe both input and system
output response oscillations and hence determine experimentally the form ofthe
transfer function model. Such a determination of the model is known as fre-
quency response analysis, since in effect one is analyzing the amplitude attenua-
tion and phase shift between input and output sinusoids at given frequencies.
Frequency response analysis is in fact a classical form of system identification
(section 2.6 and Chapter 11) typically applied in electrical engineering and
control engineering systems. Its use in water quality modeling has been limited
(e.g. Thomann, 1973), although as a component of systems analysis it offers a
perspective that conveniently complements the view provided by the time-
domain, state-space approach ofthis book. For example, it may be quite helpful
to decompose the nature ofthe behavior ofa system into responses resulting from
essentially low-frequency (slowly changing) input disturbances and responses
from essentially high-frequency (rapidly changing) disturbances (e.g. in the
problem of experimental design, section 2.6.2).
Two examples will illustrate the use of black box models. Let us suppose that
the system inputs and outputs are defined as time variations of quantities at two
spatial locations. A model that directly relates chlorophyll a variations (output)
at the center of a reservoir to variations in the phosphorus concentration
(input) ofthe major feed river can be said to be a black box model. Alternatively,
let us consider the case where inputs and outputs are defined as time variations
of several quantities at a single spatial location. If we wish to predict day-
to-day DO variations (output) from a multiple regression relationship using
conductivity and temperature variations as inputs, this too would come under
our definition of a black box model. We may thus conclude with the following
remark. Much of the eventual character of a black box model depends upon
how the analyst draws a line (conceptually) around the system-thus separating
the system from its environment-and how he chooses to categorize the quanti-
ties of interest into groups of variable types (according to Figure 2.3).
2.4.1. A Priori Sensitivity Analysis

Even without experimental field data having been collected for model evaluation,
certain important questions about the suitability of the model can now be
posed. The answers to these questions-questions of a priori sensitivity analysis-
may lead to a restructuring of the model at the conceptualization stage.
22
Figure 2.2 shows that sensitivity analysis is part of a feedback loop during both
the a priori and a posteriori phases of the modeling procedure.
The major point that we wish to establish from sensitivity analysis is the
relative sensitivity of the model predictions to changes in the values of the
model parameters p. From this analysis we might also be able to infer the
sensitivity of the model performance to various modes of behavior associated
with the various parameters and to those sectors of the model to which these
parameters are attached. A sensitivity coefficient sij for the change ~Ci in the
ith state variable of the model in response to a change ~{3j in the value of the jth
parameter may simply be defined as
~CJCi
(2.1)
sij = ~(3)/1j'
We have normalized the relationship by inclusion of a nominal reference value

/1j for the parameter, which would give a nominal reference value Cj for the
predicted state variable response. The notion of sensitivity analysis has been
well known in control engineering since the 1950s (with the work of Bode).
Only relatively recently has it been applied to water quality and ecological
systems, for example by Rinaldi and Soncini-Sessa (1978), J0rgensen et al. (1978),
and van Straten and de Boer (1979). In general, ~Cj and ~{3j are understood as
small changes in the neighborhoods of Cj and /1j' A definition of the type given
by eqn. 2.1 enables the analyst to investigate whether a certain percentage change
in a parameter has no real significance, sij ~ 0, whether {3j is a dominant param-
eter, or whether a small change in {3j induces instability in the model structure.
Two points can now be mentioned briefly in connection with sensitivity
analysis. First, if the output response of the model is found to be insensitive to
changes in the value of any parameter (i.e. sij = 0), that parameter is said to be
not identifiable. Such a problem of parameter identifiability, which will be dis-
cussed further in section 2.6.2, means that it is not possible to estimate certain
parameters unless the model relationships are specified in an alternative form.
Conversely, if a parameter has a strong influence on a particular output variable,
then measurements ofthat variable will influence the ability to estimate accurate
values of the parameter. In Chapter 11 this link between sensitivity coefficients
and parameter estimation will be made more explicit. Second, we may note that
stability analysis and sensitivity analysis are closely related. Both are concerned
with the nature of the behavior of the model in the region of some nominal or
equilibrium solution. Conditions can be derived for which linear models are
globally stable, but for nonlinear models-typical of those encountered in
modeling ecological systems-it is usually only possible to state the constraints
guaranteeing local, i.e. neighborhood, stability (e.g. Ikeda and Adachi, 1976;
Adachi and Ikeda, 1978).
Clearly, sensitivity analysis can yield much insight both into the nature ofthe
model and into the suitability of the model for calibration.
23
2.5. COMPUTATIONAL REPRESENTATION
It would be fortunate if all water quality models were sufficiently simple to yield
either an analytical solution or a solution requiring nothing more than pencil
and paper. Closed-form solutions to the model equations are partly becoming
the exception rather than the rule, and in part they are becoming increasingly
impractical. The discussion in section 2.4 reveals that our analyses are usually
strongly tied to the solution of differential equations; and if not differential
equations, then difference equations are employed that are readily amenable to
the recursive function of the modern digital computer. Hence numerical solu-
tion of ordinary differential equations has become so commonplace that the
techniques for such solution are a regular feature of standard mathematical
texts for scientists and engineers (e.g. Kreyszig, 1972). Because of this facility
seldom is little more demanded of the analyst than that he be able to write down
the functional forms of the differential equations for subsequent solution by
computer program library routines. The analyst would, naturally, be imprudent
if he did not check the degree of numerical error in his model solutions.
The case of partial differential equations and their solution is not so straight-
forward. For this reason a great deal in the character of a model may depend
upon how the differential equations are transformed into an approximate
set of difference equations. Clearly such considerations of numerical solution
are inherent in the earlier choices of lumping together certain groups of spatial
variations in the distributed-parameter model and in the choices of specifying
which parts of the dynamic behavior of the system can be said to be at a steady
state. Due reference will therefore be made from time to time in later chapters to
particular methods of distributed-parameter model solution.
The product from the modeling procedure at this point (Figure 2.2) is a
model potentially capable ofsimulating a portion, perhaps even a major portion,
or "reality." We have come thus to the boundaries of what is justified without
using the a posteriori measured information of field data.
2.6. CALIBRATION AND VERIFICATION
Let us briefly assess our situation. The raison d'etre of our book is a consequence
of the immense possibilities for complex system simulation created by the advent
of electronic computers. We may say that large computers have fostered large
water quality models-just as they have fostered large models in every other
field of technology and science. But the ability to conduct simulation exercises
with large, complex models has in no way necessarily increased our under-
standing of actual behavior or strengthened the validity of the models as
approximations of reality. In this section we shall discuss some procedures for
formally testing our models; in Chapter 11 it will become evident that only
24
rather low-order (i.e. small) models are amenable to rigorous calibration and
verification. One does not have to search far for an answer to why it is not
possible to verify many of the larger water quality models currently available,
e.g. those of Chen and Orlob (1975); we simply do not have the facilities to
gather all the field data that would ideally be required for model calibration
(Beck, 1978a; J0rgensen and Harleman, 1978; J0rgensen, 1979). Yet the
answer to the question is also partly bound to the fact that the behavior of water
quality in streams and lakes is rarely well defined and is often quite uncertain
(Young, 1978; Beck, 1981). We cannot therefore expect to place as much confi-
dence in large water quality models as we might have in a model of a petro-
chemical plant. Another question is: how, for instance, do errors of estimation
in the calibrated model parameter values affect the predictions obtained from
the model about future behavior? This topic, which might be termed a posteriori
sensitivity analysis, will be discussed briefly in section 2.7.1.
Figure 2.4 gives an outline of the calibration and verification phase of the
modeling procedure; calibration is here considered as somewhat more complex
than straightforward parameter estimation. For our subsequent discussion of
Field
1
data Experimental
r-- -
I design
I
I
I
I
I
1
I Model structure
r--- • identification
I
I
I
I
I
I
1
'---
I • Parameter estimation
I
I
I
I
I 1
I
L __
• Verification
1
FIGURE 2.4 An outline of the calibration and verification phase of the modeling pro-
cedure.
25
this topic (in Chapter 11) we shall draw heavily upon ideas and algorithms from
the fields of system identification (e.g. Eykhoff, 1974; Mehra and Lainiotis,
1976) and time-series analysis (e.g. Box and Jenkins, 1970). Before proceeding,
however, let us make a short digression and consider in greater depth some of
the implications of attempting to model poorly defined systems.
2.6.1. Observing the Behavior of the System

The central issue of model calibration and verification is that of obtaining
estimates, p, for the model parameter values, and of comparing some prediction,
co, of the outputs with the actually observed outputs CO. Predictions of the out-
puts, and not predictions, C, of the state variables, are required for the compara-
tive aspects of model verification. In other words, if we recall Figure 2.3 the
problem of calibration may be stated thus:
Given a set of experimental field data comprising the measured inputs d and
the measured outputs CO:
Determine values for the parameters ~ and the states c of the model chosen to
characterize the system behavior.
The theme of the a posteriori phase of the modeling procedure is centered upon
the retrieval, manipulation, and restructuring ofmeasured information: how can
we translate information about the" external" description of the system, d and
co, into information about the "internal" description of the system, pand c?
Since restricted measurement facilities and considerable complexity are the
dominant features of microbiological/ecologicalt systems, what is the likeli-
hood of success in the application of techniques of calibration and verification?
To answer these questions it is instructive to recast Figure 2.3 into the
representation of Figure 2.5. Block 1 includes the fundamental microbiology
and biochemistry of the system, such as phytoplankton production or micro-
organism-substrate interaction. At this level a high degree of literally micro-
scopic detail would be required to characterize, or model, all the phenomena
in the process under study. Yet the structure of these relationships, and the
changing patterns of dominant species in the ecological community, though
microscopic in detail, cannot necessarily be ignored, for they may have gross
macroscopic impacts on overall process conditions, such as algal blooms with
the consequences of severe oxygen depletion.
For block 2 the more macroscopic features of the state behavior of the system,
e.g. variations in pH and water temperature, will reciprocally influence what
happens at the microscopic, biochemical level. In general, however, most of
t This is not to ignore the complexities of hydrodynamics; we merely use microbiological aspects
of water quality as the illustrative example.
26
3
Process environment and instrumentation
/
"
// 2 Process state dynamics +
c"; +
dC;~F===::;>F=r~ Variations in (e.g.l: pH, temperature,
/; dissolved oxygen, chemical oxygen
~/ demand, chlorophyll a concentrations
/
cu
Biochemical kinetics and
microbiology
Microorganism - substrate
interaction
Predator- prey relationships
FIGURE 2.5 Observing the behavior of the system (to be compared with Figure 2.3).
the microscopic detail of block 1 belongs to the category of state variables that
are not easily measured, namely cu' Hence this fine detail is "lost," as it were,
to the process environment (block 3). That is to say, direct measurement of the
variables characteristic of block 1, for instance the concentrations of nitrifying
bacteria, is extremely difficult unless specialized experimental and analytical
facilities are available to the investigator. The relatively few variables in block 2
that are easily measured, Cm, amount only to the more macroscopic, some-
times crude, measurements of quantities like chemical oxygen demand, and
concentrations of chlorophyll a and suspended solids.
Block 3 of Figure 2.5 represents in part the system environment, from which
all manner of unobserved disturbances and unpredictable mechanisms of
behavior, 1;, will interact with the more deterministic features of the phenomena
accounted for in blocks 1 and 2. Block 3 also represents the instrumentation and
analytical procedures from which arise unavoidable components of measure-
ment error, 'I. Thus block 3 is intended to introduce elements of uncertainty
into the picture of system behavior.
Therefore, the following can finally be stated in answer to our earlier question
about the likelihood of success in the application of techniques for model
calibration and verification. Clearly, if measurements of only some of the
process inputs d and only some of the process outputs CO can be obtained, then
relatively little information is available with which to estimate the process
states c and parameters ~. In particular, it is unlikely that there will be much
information directly related to the microscopic detail of block 1 in Figure 2.5.
27
Measured
input c::=:::::::::---;:::::==~ "Reality"
disturbances d
)====:> Errors I
I
I
I
Model I
I
~
Predicted outputs I
I
" I
',~ ~~o~~~E~~~~~ __~ __ J
value adjustment
FIGURE 2.6 A rudimentary method of model calibration.
Moreover, the relationships between d, c, and CO are likely to be significantly

obscured by the uncertainty originating from the process environment and
instrumentation. One should therefore be modest in proclaiming the success of
currently published calibration studies.
A large proportion of the procedures already applied to water quality model
calibration might be classified as essentially procedures of "trial and error"
simulation. In other words, this is the type of informal calibration procedure
whereby, according to Figure 2.6, the analyst starts with some model structure
and set of associated parameter values, so that the simulated performance of the
model is compared with the observed behavior of the system under study.
Then, if the model is found to be inadequate in its characterization of reality,
the analyst may decide simply to adjust some of the parameter values on an
ad hoc basis until the desired (i.e. observed) performance is obtained. On the
other hand, the model may be so much in error that the analyst is required to
alter the structure of the relationships between the variables accounted for in
the model.
However, since reality is evidently subject to randomness in its observed
behavior, the techniques of interest here will be required to recognize the
presence of chance errors and disturbances (Figure 2.7). At the same time these
formal methods of estimation and calibration should be able to discriminate
effectively against such ever-present noise in the field data. It is not, in practice,
a matter of the analyst being unaware of the stochastic aspects of the modeling
problem, or of the informal trial-and-error simulation method being wholly
inadequate; for whether one calibrates model performance along the lines of
Figure 2.6 or 2.7, a major part of the calibration and verification exercise is
devoted precisely to filtering out the uncertainty in the observed patterns of
behavior. Moreover, these calibration methods should be able to assist one in
making inferences about the behavior of the inaccessible (i.e. not easily measure-
able), microscopic portion of the state variables from information on the more
28
Unmeasured
input Measurement
disturbances errors"
Measured "Reality" A
inpu t c::=:::=;-;::=c::=~ I
disturbances d L- ----l
9
o
\ r- - = = =c: =MO~dec: =1
l1_ Predicted outputs
= ==============:>i
t
h
m
Formal parameter
{J value adjustment
FIGURE 2.7 A more formal method of model calibration.
accessible, macroscopic sector of the state variables. If the algorithms of

calibration can fulfill these functions, even in some small measure, then we might
consider their application to have been beneficial.
2.6.2. Experimental Design

The success of any modeling procedure that sets itself the objective of demon-
strating how well, or how badly, the model simulates reality is strongly de-
pendent upon the quality of the field data available. The ideal would be the
ability to make certain specialized and deliberate in situ experiments. Seldom,
if ever, is the analyst permitted this luxury: it is difficult to imagine there being
sufficient control over the input forcing functions of the system for a specified
sequence of experimental events to be transmitted to the system. Thus the
analyst is not free to experiment with his system, and any "experiments" as
such reduce simply to the observation of behavior under "normal operating
conditions" (Eykhoff, 1974). Given these constraints, it is useful to examine
briefly how best to organize one's sampling effort in order to obtain pertinent
observations of the behavior to be modeled. Since we are concerned with in situ
experiments our remarks are not particularly relevant to those cases where the
analyst undertakes a laboratory experiment, for example a chemostat or more
complex hydrodynamic experiment, to examine hypotheses about the field
system.
There are common features shared by the problems of determining the data
requirements of parameter estimation algorithms and designing experimental
29
and routine monitoring programs. Indeed, algorithms that we shall use later for
illustrating parameter estimation in Chapter 11 are also algorithms that have
been applied for the specification of water quality sampling programs (Moore,
1973; Lettenmaier and Burges, 1977; Bras, 1978; Kitanidis et ai., 1978; Canale
et ai., 1980). The common theme of experimental design and parameter estima-
tion is the determination of the number and kind of measurements required for
distinguishing between "signal" and "noise" in the observed information.
In other words, we wish to be reasonably confident that our results are not
significantly impaired by measurement error and other sources of uncertainty
(as discussed in section 2.6.1).
If it is assumed that the objective is the calibration of a dynamic model, the
choice of sampling frequency, together with a decision regarding the duration
of the data collection program, is of crucial importance (Gustavsson, 1975).
Two very rough rules of thumb state that:
(i) The sampling interval should be at most as long as the minimum time
constant (or response time) of interest; alternatively, the sampling
interval should be one-sixth of the period of the fastest sinusoid-type
variation expected in the behavior of the system (Shannon and Weaver,
1949).
(ii) The length of the experiment should ideally cover a period with a
magnitude of at least ten times the magnitude ofthe largest time constant
of interest; to some extent this kind of determination is related to the
observation that the degree of subsequent parameter estimation error
is inversely proportional to the length, i.e. number of samples, of the
experiment.
Both rules are concerned with the speeds of response of the output variables to
changes in the input variables. Thus, if the idea of a time constant is approxi-
mately interpreted as, say, the detention time for water in a lake, and if this same
notion can also be crudely translated into the time scales for biological
growth of a species and rates at which nutrients are cycled, one has, according
to these rules, the basis of an experimental design.
Let us discuss the choice of sampling frequency in greater detail. For example,
it may well be that a fixed sampling frequency, as in a routine monitoring
program, is not always advisable. An "adaptive" strategy would be more
appropriate in collecting data for a phytoplankton model, in which an important
objective is the estimation of model parameters relating to the dynamics of a
bloom condition. Provided that the analyst has prior knowledge of the timing
of the algal bloom he may, for a short period of the year, allocate his sampling
effort to high-frequency measurements of chlorophyll a concentration. Another
view of this problem is that relatively rapid changes in water quality (i.e.
relatively small time constants in the system behavior) are effective only during
30
certain intervals and thus require only intermittent, intensive sampling effort.
Alternatively, sensitivity analysis of the a priori model might indicate that the
model responses are particularly sensitive to certain parameter values at
certain points in time (and space). In Chapter 9 an adaptive experimental
program will be reported for a lake modeling study.
For the calibration of a complex multivariable model having several inputs
and several outputs, the determination of sampling frequency alone does not
assist the analyst in making choices about which variables to measure. The
problem is that a wrong choice of measured variable combinations may eventu-
ally lead to difficulties of model structure and parameter identifiability, as already
noted in section 2.4.1. In general one has a problem of identifiability if the
structure of the proposed model is such that the information contained in the
field data cannot be translated into information about the values of certain
parameters. An analysis of the identifiability of the a priori model may therefore
reveal the following kinds of features (Cobelli et al., 1979):
(i) those parameters in the model that can be uniquely estimated from an
experiment with given input and output measurement combinations;
(ii) an appropriate combination of possible input and output measurements
that will allow the unique estimation of all the model parameters.
We would, for instance, have a typical problem of identifiability in the model

if it were possible to estimate only the sum of two rate constants, K 1 and K 2,
as a value p, i.e. K 1 + K 2 = p. There is no unique solution to this relationship
and many pairs of values of K 1 and K 2 will sum to the value p. Since this kind of
prior analysis requires a model, it will not have escaped notice that a good
experimental design almost implies the end point of the exercise, i.e. a reasonably
good model of reality.
2.6.3. Model Structure Identification

We come now to the problem of model structure identification. More specifically,
let us say that a posteriori model structure identification attempts, by reference
to the field data, to establish how the measured system input disturbances d
are related to the system state variables c, and how the state variables are in
turn related both to each other and to the measured system outputs CO. This is,
therefore, at least conceptually, a different problem from the straightforward
matter of parameter estimation usually associated exclusively with the activity
of model calibration. A posteriori model structure identification is technically
a much more difficult problem to solve than the estimation of parameter
values. In practice, as will be demonstrated later, the application of a parameter
estimation algorithm is frequently implicit in the solution of the model structure
identification problem (Beck, 1979).
31
It may be helpful to visualize model structure identification as analogous to
the choice of whether to fit a straight line or a curve to a set of experimental data.
Or again, within the above-mentioned broad definition of this problem, model
structure identification is also concerned with identifying the correct form of
the mathematical expressions that are contained in the model equations. Let us
suppose that, as an example, we are examining the uptake or removal of a
nutrient/substrate in a batch chemostat reaction, and that our first hypothesis is
a linear kinetic model,
Model I
(2.2)
in which the dot notation refers to differentiation with respect to time t; c 1, the
concentration of substrate, is the state variable, and /31 is a parameter repre-
senting a first-order kinetic decay rate constant. For our second hypothesis
about the observed system behavior we might propose a Monod-type kinetic
expression together with the presence of a mediating microorganism in the
reaction,
Model II
(2.3a)
(2.3b)
where the additional state variable C2 is the microorganism concentration and

we have a vector [P~, /32' /33' /3~] of associated model parameters. Now we
recall that there are presumably some noise-corrupted measurements available
from this experiment, but that we do not know which, if either, of Models I and II
better characterizes the nature of the observed behavior. Model structure identifi-
cation is then the problem of choosing, by reference to the in situ data, the
number of state variables to be accounted for in the model, the problem of
defining how these state variables depend upon each other, and the problem
of identifying the correct form of the expression to go inside the square brackets
of eqns. 2.2 and 2.3. If both models are thought, a priori, to be good approxima-
tions of reality, we might also call this a problem of model discrimination in the
sense used by Shastry et at. (1973). However, if neither hypothesis is adequate and
if a more complex pattern of behavior is suggested by the analysis of the data,
the definition given above will be the most useful interpretation of model
structure identification to be borne in mind here.
32
The matter of model structure identification will be discussed again in the
next section and elsewhere. Its importance in the modeling procedure is partly
related to the fact that water quality and ecological systems are not well defined,
and therefore the investigator is frequently ignorant of the "true" observed
relationships between the system variables. Thus let us return briefly to the
discussion of internally descriptive and black box modeling approaches.
Probably the best view of the two approaches is that internally descriptive and
black box models represent complementary conceptual frameworks for system
identification; some attempts have been made to illustrate this view (Beck,
1978b). Since the two approaches are complementary, more is to be gained from
their joint application than from the exclusive use of either model type. For
much of the time system identification, and model structure identification in
particular, is confronted with the need to offer plausible hypotheses about
"unexplained" relationships in a set of field data. It seems only prudent,
therefore, to approach each such problem from a variety of angles, and to gather
all the available evidence for synthesis of the next hypothesis.
2.6.4. Parameter Estimation

Parameter estimation deals with the computation of values for the parameters
that appear in the model equations, once the structure of these relationships
has been properly identified. A basic principle of parameter estimation, as
illustrated by Figure 2.7, is that the estimates p, say, of the model parameters
JJ are obtained by minimizing some function of the errors,
(2.4)
between the output response observations CO and model predictions of those
output variables, CO, One of the simplest and most well known of parameter
estimation schemes is that of least-squares estimation, where the loss (error)
function to be minimized is
(2.5)
In (2.5) the summation is carried out over all available observations; if £ is
assumed to be a column vector with p elements, then J is a scalar function,
since
T
££=£1+£2
2 2 + 2
"'+£p, (2.6)
where superscript T denotes the transpose of a vector or matrix.
At this point in the book it is not necessary to develop further the subject of
parameter estimation; ample illustration of the application of techniques of
parameter estimation will be provided in Chapter 11. It is useful, however, to
draw a distinction between off-line schemes (block data processing) and on-line
schemes of parameter estimation, since this leads to another view of the prior
33
pO • ""'"
•
to
•
t,
• - - --- - - - - - - --- - . Parameter estimate
t2 tN update
---p'..)
(a)
(b)
FIGU RE 2.8 Methods of parameter estimation: (a) olf-line; (b) recursive. The notation tk
in this example represents the kth discrete sampling instant in a time series with N samples.
problem of model structure identification. We shall deal first with a summary of

the off-line scheme. As shown in Figure 2.8(a), an off-line procedure keeps the
parameter estimates constant at their a priori values po., while the complete
set of field data (the sampled measurements for the period from t I to t N ) is
processed by the algorithm. A value for the loss function, such as that given by
(2.5), is calculated at the end of each iteration. The algorithm attempts then to
minimize the value of the loss function with respect to the model parameters
and computes an updated set of parameter values, pI, for substitution into the
next iteration through the data.
In contrast to the off-line algorithms, a recursive estimation algorithm
computes revised parameter estimates P°(tk), sayJ at each sampling instant t k
of the field data. Here, t k denotes the time of the kth observation of system
behavior. The data are therefore processed in a serial fashion, as shown by
Figure 2.8(b). The minimization of the value of the error loss function is in-
cluded implicitly, rather than explicitly, in the algorithms. At the end of the
block of data the estimates PO(t N ) are substituted for the a priori parameter
values Pl(t O) of the next iteration through the data. Further iterations through
the set of field data are required since any initially incorrect estimates PO(t o)
contribute errors to the calibration loss function that are larger than the errors
t We use the argument of discrete time merely for illustration; otherwise. it is easy to imagine a
recursive algorithm being applied in a stepwise manner to data obtained from a series of fixed
spatial locations along a river.
34
c,
/1,
(a) C2 /1 3 C4
j
C, C3
/1 2
/1 5
/1, C5
/16
C2 P3 C4
(b)
F1GURE 2.9 A conceptual picture of the problem of model structure identification.
contributed by initially correct estimates. By implication, therefore, the mini-

mum of the loss function is unlikely to have been located after the first itera-
tion.
Now let us reconsider model structure identification in the light of our
knowledge of recursive parameter estimation. We suppose that the state
variables c in a water quality model may be represented conceptually by the
nodes of Figure 2.9(a), and that the parameter values are visualized as the
"elastic" connections between the state variables. If the assumption has been
made that all the parameter values are constant with time and yet a recursive
algorithm yields an estimate of one or more of the parameters (/34' say) that is
significantly time-varying, one may question the correctness of the chosen
model structure. The reason for this is as follows. The general nature of an
estimation procedure is to fit the model (i.e. state variable) predictions to the
field observations. Hence when any persistent structural discrepancy is detected
between the model and observed patterns of behavior, this will manifest itself
as an attempt by the estimation procedure to adapt the model, i.e. the parameter
values, toward "reality." Such time variations of the parameter values can, of
course, occur for different reasons; for instance, the parameter may be truly
35
time-varying in accordance with some seasonal fluctuation. However, for the
purposes of our example in Figure 2.9(a) we might suppose that the actual struc-
ture of the relationships underlying the observed system behavior would be
better represented by the introduction of a new state variable and two new
parameters, as shown in Figure 2.9(b). If this were indeed the correct model
structure, recomputation of the parameter values should give recursive estimates
that are essentially constant.
Our example has two objectives. First, it should emphasize the earlier
statement that model structure identification and parameter estimation are
closely related and that the former problem can sometimes be solved by
recourse to a parameter estimation routine. Second, it should be apparent that
an exercise in accurate parameter estimation is of dubious significance if the
prior problem of model structure identification has not been satisfactorily
resolved.
2.6.5. Verification
"Verification" and "validation" are easily confused and both have come to be
interpreted in several different ways. However, since we have placed "verifica-
tion" together with "calibration," thus implying association with the a posteriori
phase of the modeling procedure, the following will serve as our working
definition: verification is the determination of whether the" correct" model has
been obtained from a given single set of experimental data. Findeisen et al.
(1978), in their sample glossary of systems analysis, give a different definition:
"a model is said to be verified if it behaves in the way the model builder wanted
it to behave." Such a definition, in the present discussion, would be more relevant
to the a priori stages of the modeling procedure, for instance at the stage of
a priori sensitivity analysis (section 2.4.1). On reflection it must seem that both
definitions suffer from being vague; and in fact the arguments for satisfying
oneselfthat the model is verified, in an a posteriori sense, are also rather circular.
Usually these arguments are as follows. We assume that a model structure has
been identified, that the parameters have been estimated, and thus that a
sequence of final model response errors can be computed, as illustrated by
Figure 2.10. Almost inevitably it will have been necessary at some stage during
the analysis of the field data to have made assumptions about the statistical
properties of the noise sequences idealized in the model of reality, for example
the processes ~ and" in Figure 2.3. If these assumptions are valid, the model
response errors should also conform to certain statistical properties, and in
particular to those of white noise sequences. To conform with the properties of
white noise any error sequence should broadly satisfy the following constraints:
that its mean value is zero; that it is not correlated with any other error sequence;
and that it is not correlated with the sequences of measured input forcing
functions. Evaluation of the error sequences in this fashion can, therefore,
36
Unmeasured
input Measurement
disturbances { errors "
n
"Reality"
Measured Errors
input
disturbances d
Model
Predicted outputs
Measured input
sequence Error sequence
.. Correlated
variations?
.
Time Time
FIGURE 2.10 Model verification: computing the residual error sequences and checking
their statistical properties.
provide a check essentially on whether the final model invalidates some of the
assumptions inherent in its development.
Should the error sequences not conform to their desired properties, this
suggests that the model does not characterize adequately all of the (relatively)
more deterministic features of the observed dynamic behavior. A strong cor-
relation between variations in a given input and the variations in the model
response errors of a given output, for example, would indicate that the model
structure should be modified to accommodate additional significant relation-
ships between those two variables. Analysis of the model performance along
these lines, therefore, directs attention once again to the problem of model
structure identification.
2.7. VALIDATION
It is clear that calibration and verification represent the bulk of the procedure
for model development and testing, once an experimental data set has been
obtained. There is, however, no guarantee that the validity of the model extends
37
beyond the sample data set against which it has been calibrated. Validation
is, then, the testing of the adequacy of the model against a second, independent
set of field data. Because validation thus entails the design and implementation
of new experiments, it is unfortunately a step in the analysis that is all too rarely
attempted. We may further observe that Findeisen et al. (1978) give the following
as part of their definition of validation: "a model can never be completely
validated; we can never prove that its results conform to reality in all respects;
it can only be invalidated." Similar notions are evident in an article by Schweppe
(1978), who discusses the use of statistical hypothesis testing for the determina-
tion of model validity. Statistical hypothesis testing, as he says, does not prove
that a hypothesis (model) is true; it may merely indicate whether it is possible
to reject the hypothesis on the basis of the available observations and the chosen
set of test statistics.
Perhaps too much time should not be spent in lamenting any lack of interest
shown by analysts in exhaustively validating their models. For example,
Mankin et al. (1977) offer the following advice:
Let us dismiss the question: Have you proven that your model is valid? with a quick
NO. Then let us take up the more rewarding and far more challenging question:
Have you proven that your model is useful for learning more about the ecosystem?
2.7.1. A Posteriori Sensitivity Analysis

Much of what has been said above implies that validation is not a "once and
for all" step. Indeed, like the other topics of this chapter, validation defines only
a location in the iterative procedure of model development. Figure 2.2 shows
two feedback loops returning from the validation stage: one to the calibration
and verification phase; and one to the block labeled "sensitivity analysis."
A priori sensitivity analysis (section 2.4.1) establishes the relative magnitudes
of changes in the simulated model output responses to changes in the model
parameter values. A posteriori sensitivity analysis, if we may call it such, ex-
amines the distribution of model responses that are possible, given the distri-
butions of estimated parameter values. This requires some further explanation.
Calibration of a model does not imply that all parameter values will be known
exactly once the model is calibrated; there remain irreducible errors of estima-
tion associated with the calibrated parameter values. In fact these estimation
errors reflect a mixture of uncertainty and errors in the field data used for
calibration, and uncertainty in the values initially specified for the parameters
prior to calibration. If the calibrated model is now applied to the prediction of
future events, the uncertainty in the parameter estimates (together with the
uncertainty in the present state of water quality and uncertainty in the future
input forcing functions, e.g. solar radiation) would be propagated forward
as a degree of uncertainty, or error, in the predicted response of the water body.
A simple example may illustrate the point. Let us suppose that, during model
38
calibration, it turns out that one particular parameter has a large estimation
error. Among several reasons, such as the problem of identifiability discussed
in sections 2.4.1 and 2.6.2, this large error might result from the fact that the
type of system behavior associated with that parameter is not exhibited in the
available field data. In other words, there is no information in the observations
with which to estimate the given parameter accurately. Thus, when the model
is used for prediction, and if the assumed range of future forcing functions
drives the model into a pattern of behavior that is sensitive to this particular
parameter, then the following should occur. The distribution of future output
responses should become wide, indicating both little confidence in model
performance and great uncertainty about the pattern of responses. In effect,
such analysis reveals that certain types of behavior cannot be predicted ac-
curately because little previous empirical evidence is available for calibration
of the model of that behavior.
The attention given so far to this kind of a posteriori sensitivity analysis has
been modest, although current developments indicate a rapidly increasing
interest in related topics. Typically, one might approach the problem under the
heading of Monte Carlo simulation, as did Mankin et al. (1976) and Tiwari et al.
(1978), although both of these studies lack the direct connection of such analysis
with prior model calibration results. Some analytical results have also been
obtained for a number of basic simple forms of ecological models by O'Neill
and Gardner (1979) who, with co-workers, have undertaken a comprehensive
examination of the effects of error and uncertainty on model predictions. Other
approaches are suggested by Argentesi and Olivi (1976), Di Toro and van Straten
(1979), Beck et al. (1979), Fedra et al. (1981), and Beck and van Straten (1983).
2.8. SUMMARY AND CONCLUSIONS
In this chapter a procedure for modeling has been introduced. Several salient
problems within that procedure have been discussed in qualitative terms.
The major objective of the chapter has been to show that the modeling procedure
divides into two parts: that which can be accomplished on the basis of existing
knowledge and theory; and that which ought to be accomplished when ex-
perimental field data are available. Much of the former part, the a priori phase
of the modeling procedure, is largely intuitive and not methodological. The
a posteriori stage of modeling covers more technical topics that may be less
familiar to the reader. For that reason Chapter 11 has been designed as a
complement to this chapter. Sensitivity analysis, system identification, time-
series analysis, and parameter estimation will be the subjects of Chapter 11.
An important point emerging from the present chapter is that (a posteriori)
model structure identification can be a particularly difficult problem to solve
during calibration. Indeed. calibration as a whole is not a trivial exercise, not
the least because good-quality field data are rarely available. Thus it may not
39
always be possible to adhere to the calibration, verification, and validation
phases of the modeling procedure, though such adherence would be highly
desirable. Subsequent chapters of this book emphasize in general the develop-
ment of specific models from basic theoretical principles (i.e. the a priori aspect
of modeling). The reader should keep in mind the problems that each particular
model might pose if it were to be calibrated and verified against field data,
REFERENCES
Adachi, N., and S. Ikeda (1978). Stability analysis of eutrophication models. Research Memorandum
RM-78-53 (Laxenburg, Austria: International Institute for Applied Systems Analysis).
Argentesi, F., and L. Olivi (1976). Statistical sensitivity analysis of a simulation model for the
biomass-nutrient dynamics in aquatic ecosystems. Proceedings of 4th Summe.r Computer Simu-
lation Conference, Simulation Councils, La Jolla, CA, pp. 389-393.
Beck, M. B. (1978a). Mathematical modeling of water quality. Collaborative Proceedings CP-78-10
(Laxenburg, Austria: International Institute for Applied Systems Analysis).
Beck, M. B. (l978b). Random signal analysis in an environmental sciences problem. Applied Mathe-
matical Modelling 2( I): 23-29.
Beck, M. B. (1979). Model structure identification from experimental data. Theoretical Systems
Ecology, ed. E. Halfon (New York, NY: Academic Press), pp. 259-289.
Beck, M. B. (1980). Forecasting and control of water quality. Real-Time Forecasting/Control of
Water Resource Systems, ed. E. F. Wood (Oxford: Pergamon), pp. 179-202.
Beck, M. B. (1981). Hard or soft environmental systems" Ecological Modelling II: 233-251.
Beck. M. B.. E. Halfon. and G. van Straten (1979). The propagation of errors and uncertainty
in forecasting water quality-Part I: Method. Working Paper WP-79-100 (Laxenburg, Austria:
International Institute for Applied Systems Analysis).
Beck, M. B., and G. van Straten (eds.) (1983). Uncertainty and Forecasting of Water Quality (Berlin:
Springer) (in the press).
Box, G. E. P., and G. M. Jenkins (1970). Time-series Analysis, Forecasting and Control (San Fran-
cisco, CA: Holden-Day).
Bras, R. L. (1978). Sampling network design in hydrology and water quality sampling: A review of
linear estimation theory applications. Applications of Kalman Filter to Hydrology, Hydraulics
and Water Resources, ed. C-L. Chiu (Pittsburgh, PA: University of Pittsburgh, Stochastic
Hydraulics Program), pp. 155-200.
Canale, R. P., L. M. de Palma, and W. F. Powers (1980). Sampling strategies for water quality in the
Great Lakes. US Environmental Protection Agency, Washington, DC Report EPA 600/3-80-055.
Chen, C W., and G. T. Orlob (1975). Ecologic simulation of aquatic environments. Systems
Analysis and Simulation in Ecology vol. 3, ed. B. C Patten (New York, NY: Academic Press),
pp.476-588.
Cobelli, C, A. Lepschy, and G. Romanin-Jacur (1979). Structural identifiability of linear com-
partmental models of ecosystems. Theoretical Systems Ecology, ed. E. Halfon (New York, NY:
Academic Press), pp. 237-258.
Di Toro, D. M., and G. van Straten (1979). Uncertainty in the parameters and predictions of
phytoplankton models. Working Paper WP-79-27 (Laxenburg, Austria: International Institute
for Applied Systems Analysis).
Eykhoff, P. (1974). System Identification- Parameter and State Estimation (Chichester: Wiley).
Fedra, K., G. van Straten, and M. B. Beck (1981). Uncertainty and arbitrariness in ecosystems
modelling: a lake modelling example. Ecological Modelling 13: 87-110.
Findeisen, W., A. lastrebov, R. Lande, J. Lindsay, M. Pearson, and E. S. Quade (1978). A sample
glossary of systems analysis. Working Paper WP-78-12 (Laxenburg, Austria: International
Institute for Applied Systems Analysis).
40
Finney, B. A., D. S. Bowles, and M. P. Windham (l9g2). Random differential equations in water
quality modeling. Water Resources Research 18(1): 122-134.
Gustavsson. I. (1975). Survey of applications of identification in chemical and physical processes.
Automatica 11 :3-24.
Ikeda, S., and N. Adachi (1976). Dynamics of the nitrogen cycle in a lake and its stability. Ecological
Modelling 2:213-234.
),irgensen. S. E. (1978). Review of eutrophication models. Modelling, Identification and Conlrol in
Environmental Systems, ed. G. C. Vansteenkiste (Amsterdam: North-Holland), pp. 473-500.
J,;rgensen, S. E. (ed.) (1979). Hydrophysical and ecological models of shallow lakes and reservoirs.
Collaboralil'e Proceedings CP-78-14 (Laxenburg, Austria: International Institute for Applied
Systems Analysis).
Jergensen, S. E., and D. R. F. Harleman (eds.) (1978). Hydrophysical and ecological modelling of
deep lakes and reservoirs. Collaborative Proceedings CP-78-7 (Laxenburg, Austria: International
Jergensen, S. E., H. Mejer, and M. Friis (1978). Examination of a lake model. Ecological Modelling
4 :253-278.
Karplus, W. J. (1976). The future of mathematical models in water resources systems. System
Simulation in Water Resources, ed. G. C. Vansteenkiste (Amsterdam: North-Holland), pp. 11-18.
Kitanidis, P., C. S. Queiroz, and D. Veneziano (1978). Sampling networks for violation of water
quality standards. Applications of Kalman Filter to Hydrology, Hydraulics and Water Resources,
ed. c.-L. Chiu (Pittsburgh, PA: University of Pittsburgh, Stochastic Hydraulics Program),
pp.213-229.
Kreyszig, E. (1972). Advanced Engineering Mathematics (New York, NY: Wiley).
Lettenmaier, D. P., and S. J. Burges (1977). Design of trend monitoring networks. Proceedings of
American Society of Civil Engineers, Journal of Environmental Engineering Division 103(EE5):
785-802.
Loucks, D. P., and W. R. Lynn (1966). Probabilistic models for predicting stream quality. Water
Resources Research 2: 593-605.
Mankin, J. B., R. H. Gardner, and H. H. Shugart (1976). The COMEX computer code: Monte
Carlo analysis of ecosystem attributes. Proceedings of 1976 Summer Computer Simulation Con-
ference, pp. 433-436.
Mankin, J. 8., R. V. O'Neill, H. H. Shugart, and B. W. Rust (1977). The importance of validation
in ecosystem analysis. New Directions in the Analysis of Ecological Systems, ed. G. S. Innis.
Simulation Council Proceedings Series 5(1): 63-72.
Mehra, R. K., and D. G. Lainiotis (eds.) (1976). System Identification.' Advances and Case Studies
(New York, NY: Academic Press).
Moore, S. F. (1973). Estimation theory applications to design of water quality monitoring systems.
Proceedings of American Society of Civil Engineers, Journal of Hydraulics Division 99: 815-831.
O'Neill, R. V., and R. H. Gardner (1979). Sources of uncertainty in ecological models. Methodology
in Modelling and Simulation, eds. 8. P. Zeigler et al. (Amsterdam: North-Holland), pp. 447-463.
Orlob, G. T. (1975). Present problems and future prospects of ecological modeling. Ecological
Modeling in a Resource Management Framework, ed. C. S. Russell. Resources for the Future
Working Paper QE-I (Baltimore, MD: Johns Hopkins University Press), pp. 283-312.
Padgett, W. J., G. Schultz, and C. P. Tsokos (1977). A stochastic model for BOD and DO in
streams when pollutants are discharged over a continuous stretch. International Journal of
Enl'ironmental Studies II: 45-55.
Papoulis, A. (1965). Probability, Random Variables and Stochastic Processes (New York, NY:
McGraw-Hill).
Rinaldi, S., and R. Soncini-Sessa (1978). Sensitivity analysis of generalized Streeter-Phelps models.
Advances in Water Resources I: 141-146.
Rinaldi, S., R. Soncini-Sessa, H. Stehfest, and H. Tamura (1979). Modeling and Control of River
Quality (New York, NY: McGraw-Hili).
41
Schweppe, F. C. (1978). Model identification problems. Applicalions of Kalman Filter to Hydro-
logy, Hydraulics and Water Resources, ed. c.-L. Chiu (Pittsburgh, PA: University of Pittsburgh,
Stochastic Hydraulics Program), pp. 115-133.
Shannon, C. E., and W. Weaver (1949). The Malhematical Theory of Communication (Urbana, IL:
University of l1Iinois Press).
Shastry, J. S., L. T. Fan, and L. E. Erickson (1973). Nonlinear parameter estimation in water
quality modeling. Proceedings of American Society of Cil'il Engineers, Journal o{ Em'ironmental
Engineering Division 99(EE3): 315-331.
Somlyody, L. (1981). Water quality modelling: A comparison of transport-oriented and bio-
chemistry-oriented approaches. Working Paper WP-81-117 (Laxenburg, Austria: International
van Straten, G., and B. de Boer (1979). Sensitivity to uncertainty in a phytoplankton-oxygen model
for lowland streams. Working Paper WP-79-28 (Laxenburg, Austria: International Institute
Streeter, H. W., and E. B. Phelps (1925). A study of the pollution and natural purification of the
Ohio River. US Public Health Service, I'Vashington, DC Bulletin 146.
Thayer, R. P., and R. G. Krutchkoff (1967). Stochastic model for BOD and DO in streams. Pro-
ceedings o{American Society oIG!·il Engineers, Journal o/Sanitary Engineering Division 93(SA3):
59-72.
Thomann, R. V. (1973). Effect of longitudinal dispersion on dynamic water quality response of
streams and rivers. Water Resources Research 9:355-366.
Tiwari, J. L., J. E. Hobbie, J. P. Reed, and M. C. Miller (1978). Some stochastic differential equation
models of an aquatic ecosystem. Ecological Modelling 4: 3-27.
Young, P. C. (1978). General theory of modelling for badly defined systems. Modelling, Identificalion
and Control in Environmelllal Systems, ed. G. C. Vansteenkiste (Amsterdam: North-Holland),
pp.103-135.
Young, P. c., and M. B. Beck (1974). The modeling and control of water quality in a river system.
Automatica 10: 455-468.
CHAPTER 2: NOTATION
C, e vector of state variables, state estimates

Cm' Cu vector of measurable, not easily measurable state variables
cO, CO vector of measured, model-predicted output response variables
c nominal deterministic reference trajectory for the state vector
d vector of measured input (forcing) variables
J (squared error) loss function
r vector representing three-dimensional space (x, y, z directions)
Sij sensitivity coefficient
t, tk independent variable of time, kth discrete instant of time
~, P vector of parameters, parameter estimates
Pj nominal reference value of parameter j
€ predicted model response errors (or residual errors)
lJ vector of (stochastic) measurement errors
~ vector of (stochastic) unmeasured input disturbances of system,
General Principles in Deterministic
3 Water Quality Modeling
P. Mauersberger
3.1. INTRODUCTION
3.1.1. The Aim of this Chapter
The aim of this chapter is to present the theoretical background of deterministic

water quality modeling. For any scientific discipline, as well as for management
activities, it is of great importance to ensure against defects in the fundamental
premises. Therefore, in this chapter the equations used in water quality modeling
will be derived from fundamentals of macroscopic physics, chemistry, and
biology. The resulting system of equations represents a complicated mathe-
matical problem. For this reason, simplified versions of these equations and the
assumptions under which they are valid will also be discussed. Anyone who is
especially interested in the application of a model for a river reach or for a
special lake may pass quickly through parts of this chapter to one of the later
chapters.
Every deterministic water quality model must be based on the following
general principles:
conservation of mass and elements

narrow bands of biomass composition
conservation of momentum
conservation of energy
boundary conditions and initial conditions
laws governing chemical, biochemical, and biological processes
the second law of thermodynamics.
42
43
Therefore, this chapter will discuss:
the main "components" of deterministic water quality models and numerical

methods for solving the typical" balance equations" (section 3.1);
basic and specialized model equations derived from the laws of conservation
of mass, momentum, and energy (sections 3.2-3.4);
mathematical descriptions of chemical reactions and biological processes
(section 3.5);
the role of entropy in water quality modeling, and some relationships between
deterministic, stochastic, and cybernetic modeling methods (section 3.6).
Figure 3.1 surveys the topics covered in the chapter. The formulation and
application of special models will be described in later chapters.
It is necessary that a deterministic ecosystem model obey the second law of
thermodynamics, according to which dissipation (or entropy production) must
be positive. While the balance equations for mass, momentum, and energy
play the roles of "book-keepers," the entropy principle, like the "director,"
determines the development of the ecosystem. The structure, state, and evolution
of the aquatic ecosystem are regulated by the mutual effects of entropy-producing
and entropy-reducing processes inside the water body and across its boundaries.
Furthermore, entropy and entropy production depend on practically all
hydrophysical, hydrochemical, and hydro biological variables and processes
of the aquatic ecosystem. Therefore, by the entropy principle the many phe-
nomenological relationships are reduced to a few important equations and basic
principles. This is of significance for the unified treatment of physical, chemical,
and (to a first approximation) biological processes in the ecosystem. It is also of
importance if we are looking for" new" types of variables (especially" general-
ized" forces and potentials) suited for the description of complex ecosystems
with the aid of a reduced number of variables.
Before we discuss these aspects in section 3.6.2, we must assemble the
basic physical, chemical, and biological concepts of water quality modeling.
3.1.2. Aims and Types of Deterministic Water Quality Models

Macroscopic, deterministic models are widely used in studying the water
quality of lakes, rivers, estuaries, and other types of water resources. These
models, taking into account the physical, chemical, and biological processes
inside the system as well as the fluxes of matter and energy across its boundaries,
are intended to serve both as research tools and as bases for water resources
management. Further improvement in our capability to model water quality
is necessary, but may be seriously limited without better knowledge of the
important processes taking place in aquatic systems and without the applica-
tion of the most advanced modeling techniques.
Conservation Advection- Special types
c-- of mass I--- diffusion equation of balance
(3.2) (3.2.3) equation
(3.1.4, 3.1.5)
~I ~
- Conservation
of momentum
(3.3.1)
Special types of
flow (3.3.2, 3.3.3)
Methods for the
numerical solution
I Dimensionless I----- of nonlinear
equations, ~ (partial)
dynamic similarity differential
(3.3.4) equations
(3.1.6, 3.1.7)
~
Energy balance Special cases
1--1 (3.4.1 ) \3.4.2, 3.4.3)
Energy balance
Chemical, bio- Chemical for chemical
I-----
chemical, and reactions reactions
1-
biological inter-
relationships in
the ecosystem
- in natural
water bodies
(3.5.4)
I---
(3.5.2)
!
(3.5.1 ) Reaction rates
and reaction
equations of
Biological phenomena I-----
chemical and
(primary production,
I-- biological
trophic interactions, etc.)
processes
(3.5.5 - 3.5.8)
(3.5.3)
Initial and boundary

conditions I Water quality models and their
I--
(3.6.1) ~ applications (following chapters)
y Entropy balance
(3.6.2 ~ 3.6.8) ~
I Types of water resource, data base,
goals of modeling, etc.
FIGURE 3.1 Survey of Chapter 3 (section numbers are in parentheses).
Since hydrophysical and ecological equations are invariably coupled, water

quality models are usually characterized by high degrees of complexity and very
difficult mathematical problems. Selecting the appropriate degree of complexity
is a challenging task. Usually, models are generated in accordance with preset
objectives of the modeling effort and with predetermined amounts and quality
45
of field observations. For instance, the long-term ecological response of an
entire lake, including the" hydrodynamics," can be modeled only by introducing
the fluxes necessary to satisfy gross mass and energy balances, or by introducing
fluxes related to spatial differences in concentration. However, if the highly
transient response of a lake (e.g. in an extreme situation) is sought, it may be
essential for dynamic models to be coupled to the ecological equations.
There is no "general" water quality model that can deal with the myriad
problems of different types of streams, lakes, and reservoirs. The structure of a
model is influenced by its particular goal. Scientific investigations may be
structured by combining detailed measurements with elaborate submodels of
the processes under consideration. Planning and management of water re-
sources frequently require "engineer-developed models," derived through
gross simplification of both the hydrodynamic behavior and the aquatic
ecosystem. In such cases, scientific rigor is relaxed in favor of practicability.
Each model represents a compromise between the complexity of the real water
resources system and the simplicity of a well posed mathematical problem
suited for the description and prediction of selected features of the real system.
Thus, in order to satisfy best the requirements of both science and practical
water management, it is necessary to establish a methodology of water quality
modeling. Basic concepts of this methodology will be stated in this chapter.
We shall start with the fairly general" basic laws" of hydrothermodynamics and
ecology and then proceed to simplified mathematical statements by which we
can more easily make the transition to special models.
3.1.3. Physical, Chemical, and Biological Components and Processes
Important processes within the water body include chemical reactions, primary
production, grazing, egestion, excretion, respiration, and nonpredatory
mortality. Many of these processes are deeply influenced by variables describing
the physical state of the system, e.g. temperature and turbidity, or by hydro-
dynamic characteristics, e.g. velocities, velocity gradients, shears, and turbu-
lence.
The set of equations for a particular deterministic water quality model can
be chosen to simulate the properties and interrelationships of any or all of the
following components and processes:
Physical Solar radiation, temperature, pressure, density, external and frictional

forces, flow velocity, diffusion flows, heat flow, kinetic and internal energy,
entropy, sedimentation, etc.
Chemical Dissolved inorganic and organic chemicals, particulate organic
material (detritus), inaccessible nutrients, heavy metals, complex synthetic
and toxic compounds. etc.
46
Biological Primary producers (diatoms, green algae, blue-green algae,
phytobenthos), zooplankton (raptors, selective filterers, nonselective filterers,
etc.), zoobenthos, fishes, biota in different life stages, etc.
In the following sections we shall identify the basic concepts upon which
models must be founded. However. we shall not deal with questions of selecting
the variables and the equations best suited for a special type of water body. The
development, calibration, validation, and application of specialized water
quality models will be treated in subsequent chapters.
3.1.4. Balance Equations

In view of both the hydrodynamic and ecological aspects of natural water
bodies, the system of equations for a water quality model must consist of at
least the balances of mass, momentum, energy, and entropy and the equations
describing chemical, biochemical, and biological processes. Appropriate
initial values and boundary conditions must be included. From the mathe-
matical point of view, a nonlinear initial-boundary-value problem of differential
equations must be solved. A question arises if there is one and only one stable
solution uniquely dependent on the initial and boundary values. Furthermore,
stable, converging numerical methods are required for the solution of these
equations. Here, we confine ourselves to some preliminary remarks on the
general types of basic equations that may result and on some of the more
popular numerical methods used for solution (sections 3.1.6 and 3.1.7).
The general type of balance equation for a scalar quantity F (which may be a
function of space, time, and other parameters) is
d
dt
f.
V(t)
F(r, t)d V = f. V(l)
G(F, ... ; r, t)d V - 1
rA(t)
S" dA. (3,1)
G denotes the sources and sinks of the field F, and may depend on F in a non-
linear manner. S" is the flux across the boundary A(t) of the volume Vet),
directed outward and normal to A(t). Table 3.1 shows some examples of sources
and fluxes.
A deterministic water quality model generally consists of a set of inter-
connected balance equations. However, for simplicity we shall confine ourselves
to only one equation of this type and assume that there is an independent
determination of the flow field vCr, t), a fairly common approach. We shall leave
discussion of the larger, more complex problems to later sections and chapters.
By use of the relationship
d
dt
f.
V(l)
F dV = f.
V(tl
(dd
F
t
+ F div V)d V = f.
V(t)
(~F
(t
+ diV(FV»)d V,
47
TABLE 3.1 Examples of Sources G and Fluxes S of the Field F.
F G s
p, mass density o o
Pi' mass density of thejth chemical
component
lI" electromagnetic energy density -j'E S" flux of electromagnetic energy

lit = lI, + (li + [,2/2 + ¢)p, total LPiVi,Fi w + S, + v· (P - Ip)
energy density
E, electric field; F i , external force; I. unit tensor ;j, electric current density; J, diffusion; Ai' molar
mass of the jth constituent; P, dissipative part of the pressure tensor (viscosity); p, pressure; vi'
velocity of the ;th constituent; v, velocity of center of mass; "',. reaction rate of the rth chemical
reaction; W, heat flow density; Yi , rate of biochemical reactions (cf. eqns. 3.85-3.86); vi'" stoichio-
metric coefficient ofthe;th constituent in the rth chemical reaction.
the following differential equation can be derived from the "integral" balance
equation 3.1 :
of .
-;:;- + dlV(vF + S) = G(F, ... ; r, t). (3.2)
ot
The local time variation of/at results from convection vF, from nonconvective
transport S, and from local sources and sinks G. Initial and boundary conditions
must be added; for instance,
F(r,O) = F o(r), rEV, (3.3)

vnF + sn = g(F, r, t), rEA, t ~ O. (3.4)
3.1.5. Special Types of Balance Equation
Reduction to an Ordinary Differential Equation

In water quality modeling, very often the field variable F is considered to be
constant within a finite time-independent volume V, or the mean values are
used:
F=~f.
V VUl
FdV, G* = r G dV,
w Vir)
S* =
r1.
AU )
Sn dA. (3.5)
The balance equation 3.1 then reduces to an ordinary differential equation,
d(V:{t» = G*(F, t) - S*(t), (3.6)

48
which, in general, is also nonlinear. G* describes changes due to processes
inside V, while S* results from the exchange of matter and energy with the
surroundings through the surface A (outflow, S* > 0; inflow, S* < 0). Appro-
priate initial conditions must be prescribed but, as a rule, in water quality
modeling the exchange processes S* govern the state and development of the
system. The initial values are most often of minor consequence.
Averaging Along the Vertical

We derive a balance equation for the mean value F of the field F, taken along the
vertical by integrating (3.2) from the bottom of the basin or river to the free
surface. Cartesian coordinates (x, y, z) are used, with z increasing toward the
zenith. The flow velocity v, flux S, and V operator are divided between the
horizontal components V h' Sh, V h and the vertical components VZ , S" ajaz.
Furthermore, we introduce: z = (o(x, y) for the bottom of the water body;
z = (x, y, t), the equation of the free water surface; H(x, y, t) = (x, y, t) -
(o(x, y), the depth of the water body; and F(x, y, t) = (ljH) no
F(x,y,z, t)dz, the
mean value of F. Analogous definitions are valid for G, Vh , and Sb' Then, from
(3.2) the following balance equation for F can easily be derived:
a(~F) + Vh ' (HvhF + S*) = HG*, (3.7)
where
and
-
S* = Sh + -I r'v~F' dz. (3.9)
Ht
Here, v~ and F' denote the deviations of the local values Vh and F from the
averages:
F' = F - F.
The source factor G* in (3.7) consists of the mean value G of the source in (3.2)
and of contributions from fluxes through the air-water interface and across
the bottom of the basin or river, which are determined by the boundary con-
ditions. The horizontal flux S* is determined not only by the average value Sh
of the nonconvective flux Sh' but also by the dispersion due to the deviations
v~ and F', which is very often the dominating process.
Averaging Over the Cross-Sectional Area of a River

In order to derive from (3.2) or (3.7) the type of balance equation used in one-
dimensional models of river flow (section 3.2.3 and Chapter 7), we define:
49
(a) The equations describing the positions of the river banks:
y = Yz(x, t).
The water flows in the x direction. The positions of the banks can
be functions of time, e.g. if the flow rate varies with time.
(b) The cross-sectional area A(x, t) of the river:
y,
I Y2(X,
Y2(X.l)f~(X' I)
A(x, t) = dz dy
Y ,(x, r) ~o(x, y)
I)
=
f r ,(x.1)
H(x, y, t)dy. (3.10)
(c) The mean value F of F across A(x, t) (and analogously the mean values
G and Sx):
F(x, t) = A1 IY2f~ F(x, y, z, t)dz dy

y I '0
I
=A I
Y,
Y
2 -
H(x, y, t)F(x, y, t)dy. (3.11 )
(d) The net downstream velocity:
f
Y2f~
u(x, t) = vxCx, u, z, t)dz dy
YI '0
= -I
A
I Yt
Y
2
H(x, y, t)uxCx, y, t)dy. (3.12)
(e) The mean flow rate:

Q(x, t) = u(x, t)A(x, t). (3.13)
Integrating (3.2) over A(x, t) or (3.7) over the river width, Y l ::;; y ::;; Yz ,
yields the following" one-dimensional" river flow equation:
a(AF) + ~ (QF + AS**) = AG**, (3.14)

at ax
where
ay ay
G** = G* + [ H (F -a + vxF -a + vyF +
t X
s; )]Y=Y2
y=Y
(3.15)
1
and
S:- + -AI I H(v x -

Y
2 __
S** = u)(F - F)dy. (3.16)
Y1
50
In (3.14), the quantities A, F, Q, S**, and G** may be functions of x and c.
According to (3.15), the sources G** include additional terms resulting from
boundary effects. To discuss (3.16) we assume that the nonconvective flux S in
(3.2) results from molecular diffusion and turbulent mixing. Then, S may be
approximated by
Sex, y, z, c) = -(D M + ()grad F(x, y, z, c), (3.17)
where the molecular diffusion coefficient D M and the coefficient ( of turbulent

mixing (eddy diffusivity) are functions of time and space. If, as usual, it is assumed
that
aF(x, c)
S *(x,
*) c = c
(,)
- Ex ax' (3.18)
it follows from (3.9) and (3.16) that the effective diffusion coefficient E,
E(x, c) = DM + "( + D', (3.19)
includes effects of molecular and turbulent diffusion as well as the dispersion

coefficient D' due to the deviations of the actual fields from their averages. The
relative effect of the dispersion term depends on how well the net downstream
velocity u is determined. If it is well done, dispersion can be minimized or even
reduced to the level of computation errors.
When Q, A, E, G**, and F are independent of time, (3.14) reduces to
d (_ dF(X»)
dx Q(x)F(x) - A(x)E(x) ~ = A(x)G**(x), (3.20)
the solution of which is
-
F(x) =
-
F(xo)exp
(IX
Xo
U(O
E(O d~
)
_ IX
Xo
Q(xo)F(xo) - A(xo)E(xo)F*(xo) ex
A(OE(O P
(rJ~
X
U(O d V)d v
E(O (, S
_ IX It G**(OA(Od( ex
Xo A(OE(~) P
(rJ~
X
U(O d v)d
E(O (,
v
S, (3.21)
where F(x o) is the value of F at an arbitrarily chosen point Xo,
-* )_
F (x o -
I' dF(x)
1m -d-'
x-+xo x
51
Q(xo)F(x o) is the transport by downstream flow at x = X o ,
- A(xo)E(xo)F*(x o) is the downstream transport caused by the effective
diffusion at x o , and
U(x) = Q(x)
A(x)
is the mean flow rate.
The one-dimensional river flow model will be used in the following chapters
of this book.
3.1.6. The Finite-Difference Method

We cannot present the details of mathematical methods for solving the non-
linear balance equations 3.1, 3.2, or 3.6. We only wish to give a short outline
that will serve to illustrate the application of the finite-difference method.
Applying (3.1) to a time-independent finite volume element V;, the center
(node) of which is situated at rj, and integrating (3.1) over the interval from III
to t ll + I' we obtain
LiF(r, til + I)d V - LiF(r, tll)d V
=
f '~·jf. G(F, ... ; r, t)d V dt -
rIA Vi
f'~+j
tlA
L
m
f
Ai,m
(v"F + S")dA dt. (3.22)
Ai. m is one of the surface elements of V;, and L m is the sum over all the elements
(Figure 3.2).
Let us introduce the following quantities, abbreviations, and approxima-
tions:
(3.23a)
(3.23b)
(3.23c)
E = v"F + S" (3.23d)
i L (E m. lI + 1 + E m. 1I + Ej,lI+ 1 + Ej,II)Ai,mhw (3.23e)

m
52
FIGURE 3.2 Example of a simple system of finite volume elements 11" Vj' Vk suitable
for the application of finite-difference methods for solving the balance equation.
Using (3.23) in (3.22), we arrive at
1 1
h (F i ,l1+ 1 - F i ,l1) = r(G i. l1 + 1 + Gi ,l1)
11
I'\' Ai.m E E E ) (3.24)

-4:L..T(Em'I1+1+ m,l1+ i,I1+1+ i.I1'
m ,
where
If G = G(F, ... ; r, t), then Gk • v depends on F k • v. Thus, eqns. 3.24 are, in general,
also nonlinear. These equations are the basis for the solution of (3.1) and (3.2)
by the finite-difference method. The volume V of the water body must be divided
into elements Vi (i = 1,2,3, ...) and the resulting system of equations 3.24 must
be solved step by step (/1 = 1, 2, 3, ...). One possibility for linearization consists
in using Fk . 11 instead of Fk 'l1 + 1 on the right-hand side of (3.24), especially if it is
assumed that Gi ,I1+1 ~ Gi ,l1'
Equations 3.24 may easily be transformed into the following system of
equations:
BiiFi.I1+1 +I BimFm.I1+1 = Pi,11 (i =f. m), (3.25)

m
53
where
(3.26)
(i "# m)
and
(3.27)
Here, i, j, m = 1, 2, 3, ... are the numbers of the space elements. Written in

matrix form, (3.25) becomes
[B]{F} = {P}. (3.28)
I t should be pointed out that {P} comprises {F} if G depends on F. If this is so,
an iteration method must be used for solving (3.28). An example of the applica-
tion of the finite-difference method is given, for instance, in Chapter 7. Finite-
difference methods are widely treated in the literature, e.g. by Collatz (1959),
Richtmyer and Morton (1967), and Marchuk (1975).
3.1.7. The Finite-Element Method

The initial-boundary-value problem represented by (3.2)-(3.4) can be trans-
formed into an initial-value problem of an ordinary differential equation by the
Galerkin method. It is supposed that the volume V of the water body is divided
into space elements J.'i (k = 1, 2, 3, ...) and that a set of weighting functions
wk(r) (k = 1,2,3, ...) is given that fulfills the conditions
wk(r k) = 1, wk(rJ = 0 if i "# k

wk(r) "# 0 inside J.'i (3.29)
wk(r) = 0 outside J.'i.
At least the first derivatives of wk(r) are square-integrable. Figure 3.3 shows a
special type of two-dimensional finite element V;. The two-dimensional weight-
ing function wk(x, y) identically vanishes outside J.'i. We are seeking the solution
F(r, t) of (3.2)-(3.4) in the form of a spline interpolation:
N
F(r, t) = L a/t)wj(r) (3.30)
j; 1
N
F(r, 0) = L aj(O)w/r), (3.31)
j; 1
54
y
I/.~.
V/_'
x
FIGURE 3.3 Three two-dimensional finite elements ~, ~, v,. , each consisting of six
triangles. ~j is common to ~ and ~. ~j # 0 only if Pi and Pj are neighboring points.
Multiplying (3.2) by wk(r), integrating over V, and using Green's theorem, the
boundary conditions (3.4), and the expansion (3.30), we produce the following
system of equations (k = 1, 2, 3, ...), from which the unknown functions a/t)
can be derived:
(3.32)
where
A jk = Iv",'j(r)wk(r)dV = A kj (3.33)
B jk = J. w/')v . grad ",-.(r)d V #- B kj

V
(3.34)
Ck(t) = Iv(G(F, . .. ; r, t)wk(r) + S' grad wk(r»d V
- fA wk(r)g(F, . .. ; r, t)dA. (3.35)
On account of the conditions (3.29), the integrals over V in (3.33)-(3.35) are

reduced to integrals over the small volume V]k common to V] and Vk. The matrix
elements Ajk and B jk are equal to zero if the finite elements V] and Vk do not
55
intersect. Thus, the matrices are sparse (with diagonal predominance and
banding). A jk and B jk can be evaluated for very complicated geometries of V
if triangular, rectangular, or curvilinear volume elements v,. and suitable
functions wk(r) are defined. Elements can be varied in size, shape, and orienta-
tion to give the modeler the maximum opportunity to describe correctly the
geometrical properties of the water body.
With the help of a difference method (e.g. the Crank-Nicolson method),
we can integrate (3.32) with respect to time, starting from the values
j = 1 (1) N, (3.36)
which result from (3.31). In general, the right-hand side, C k , of (3.32) depends
on the unknowns ak(t); therefore, iteration procedures are to be applied.
Finite-difference and finite-element methods are powerful tools for the
numerical solution of the much more complicated system of differential equa-
tions characterizing a deterministic water quality model. The reader should
find the following literature relevant: Collatz (1959), Richtmyer and Morton
(1967), Zienkiewicz and Cheung (1967), Strang and Fix (1973), Marchuk (1975),
Mitchel and Wait (1977), and Pinder and Gray (1977). We now turn to the basic
equations of water quality models.
3.2. CONSERVATION OF MASS~ADVECTION AND DIFFUSION
3.2.1. Continuity Equation, Mass Fractions

In water quality modeling we use a description in terms of a limited number of
macroscopic observables, such as the density of water, Po [ML - 3J, the mass
density Pj ofthejth chemical component (j = 1,2,3, ...), and the mass densities
Pk of the biotic components of the aquatic ecosystem (k = 1,2,3, ...). We also
introduce the mass fractions
N·} = Pj , Bk -- Pk (3.37)
P P
and the total mass density p:
(3.38)
which fulfills the continuity equation
~:
et
+ div(pv) = 0 or dp + P d'IV V =
dt 0. (3.39)
56
3.2.2. Chemical Reactions
Chemical reactions (with reaction rate wr [mol L - 3 T- 1 ] for the rth reaction)
and biochemical processes lj [ML - 3 T- 1] are to be taken into account in the
mass balances of the chemical components:
aaPj + div(pjv) + div J(Pj)

t
= I r
Vjr"HjW r + Yj (3.40)
convection diffusion chemical biochemical
reactions reactions
ort
(3.41)
Equations 3.41 follow from (3.40), by substitution of (3.37) and (3.39). The
essential parts of (3.40) and (3.41) are:
con vective displacement:
velocity of center of mass:

"Pi Vi. (3.42)
v=~-,
i P
and effective diffusion consisting of molecular diffusion, turbulent "eddy
diffusivity," and other types of random mixing processes:
(3.43)
The scales of all the mixing phenomena and diffusion processes are of particular
importance for the validity of assumptions like
J(p) = -D grad Pj or J(N) = -D* grad N j , (3.44)
where D [L 2 T - 1] and D* are scalar diffusion coefficients or diffusion tensors

of rank 2. Furthermore, in (3.40) and (3.41) we use:
Vjr the stoichiometric coefficient of the chemical reaction (v jr > 0 for products,
but Vjr < 0 for reactants),
A j the molar mass of the jth chemical component.
Even in simple cases the reaction rates W r are nonlinear functions of the densities
Pj of the reacting chemicals. Consequently, the mass balances (3.40) and (3.41)
t -a (pN) + div(pNjv) = dN
p ---" + N j (op
- + div(pv) ) = dN
p_'.
at dt 01 dt
57
are nonlinear partial differential equations. These" diffusion equations" are a
common basis for the development of the majority of water quality models in
use today. For a conservative substance (of concentration C) from (3.40) and
(3.44) we obtain
~~ + div(Cv) = div(D grad C). (3.45)
The values of the diffusion coefficient D must be derived from field experience.
Dependence on some empiricism cannot be avoided.
3.2.3. Advection-Diffusion Equations; Input-Output Models

Applying the methods described in section 3.1.5 (especially eqns. 3.7 and 3.14)
to eqn. 3.40 or 3.45, we obtain two- and one-dimensional mass transport
equations, called" advection-diffusion equations."
One-dimensional river flow models, which can be employed, for instance, to
predict the biochemical oxygen demand (BOD) and/or the dissolved oxygen
concentration (DO) in the river, are based on the differential equation 3.14, i.e.
a(AC)
- - + -a (QC - AE -
ac) = AG. (3.46)
at ax ax
The cross-sectional area A, the concentration C of the particular constituent
[M L - 3], the net downstream flow Q [L 3 T - 1], and the effective diffusion
coefficient E may depend not only on time t but also on the downstream distance
x. The source-sink term G is a function of C, x, and t. Equation 3.19 defines E.
Table 3.2 gives an overview of mass balance equations of different types.
If we assume that a well mixed lake has a constant volume Vo , constant
discharge Qo [L 3 T- 1 ], constant input Go [M T- 1 ], and constant net decay
coefficient K o [T- 1] of a nonconservative substance, we obtain the following
input-output model of type (3.6):
dC(t) Go
-- = - - KC(t) (3.47)
dt Vo '
Integration of (3.47) from the initial concentration C(O) yields
C(t) = %~o + ( C(O) - %~JexP( - Kt). (3.48)
Models of this kind have been used, for instance by Vollenweider (1969), to
predict the phosphorus concentrations in large lakes.
Vl
00
TABLE 3.2 Mass Balances and Advection-Diffusion Equations of Different Types.

vC
Continuity equation for a single substance (Pj = C): - + div(vC + J(C)) = G(C, Y, I), derived from (3.40)
VI
vC
Use of J(C) = -D grad C: - + div(vC - D grad C) = G(C, Y, I) (3.50)
VI
vAC V ( VC)
One-dimensional case, v = {u(x, I), 0, O}, etc.: -,- + -:;- QC - AE ~ = AG(C, x, I)
I' ex IX
(3.46)
(
One-dimensional case without sources and sinks: -vAC + -V QC - AE -VC) = 0
VI I'x VX
Integration of (3.50) over a finite time-independent volume Vo ~ f C dV + ~ (vC - D grad C)dA = f G(C, Y, l)dV
and application of Green's theorem: dt Vo A Vo
V
dC(I)
Introduction of the time-dependent mean values Crt), 5(1), G(I): Vo ----;Jt = SU) + v;,G(I)
input-output internal
prod uction/destruction
In the special case of G = - K o C and 5 = Go - Qo C, we obtain (3.47).
59
3.2.4. Biotic Components
The balance equations of the biotic components are of the type
a~k + div(Pk v) + div J(Pk) = Qk - PkRk - PkMb (3.49)
P dB + d'IV J(B k) = Qk - PkRk - Pk Mk·

k
Cit (3.50)
where. for k = 1,2,3....•
Qk is the production and consumption term [M L - 3 T- 1 J.

Rk is the rate of respiration of the kth component [T- 1 J.
Mk is the nonpredatory mortality rate of the kth biocomponent [T- 1].
Again, these equations are of the" advection -diffusion" type. The sources and
sinks Qb R b M k will be discussed in section 3.5.
3.2.5. Coupling of Chemical and Biotic Components

The chemical constituents and the biotic components are coupled in a very
complicated manner by biochemical processes. represented by 1] in (3.40).
and by terms on the right-hand sides of (3.49) and (3.50). To illustrate these
effects we use the balance of dissolved oxygen as an example. The local concentra-
tion of dissolved oxygen is influenced by chemical processes, such as oxidation
and reduction, and by physical processes like advection, diffusion, and surface
aeration. The coupling to biotic components is effected through photosynthetic
oxygenation, respiration. and biochemical oxidation processes. The coupling
of chemical and biotic components of the aquatic ecosystem is termed "bio-
chemical" and is illustrated in Figure 3.4 (section 3.5).
3.3. CONSERVATION OF MOMENTUM; TYPES OF

FLOW SYSTEMS
3.3.1. Navier-Stokes Equations

The hydrodynamic processes in streams, lakes, and reservoirs are integral
components of the complex aquatic ecosystem. Water movements at different
scales and of different types significantly influence not only the aggregation and
distribution of microorganisms and plankton, but also the distribution of
nutrients, dissolved gases, and temperature. Hence, circulation, convection,
wave phenomena, and turbulent mixing are major influences on the distribution
of biota and the productivity of natural water bodies.
60
A realistic hydrodynamic description of the behavior of the water body
should be founded as far as possible on the basic hydrodynamic equations.
Conservation of momentum is ensured by
dv
P- + 2poo x v= L: PiFj - P grad ¢ - grad p + V' P, (3.51)
dt .. i ..
COflolis force external forces pressure fflctlOn
where
00is the angular velocity of the Earth's rotation [T- 1 J,

Fjrepresents the nonconservative external forces (per unit mass) acting on
the component of density Pi'
¢ is the time-independent potential of external forces (e.g. gravity),
p is the pressure [M L - 1 T - 2J,
P is the dissipative part of the pressure tensor, i.e. the friction tensor including
turbulent friction [M L - 1 T- 2].
Viscous and turbulent stresses can be combined linearly and represented by

V' P = V' A· Vv (Boussinesq/Prandtl), (3.52)
where A is the "eddy diffusivity tensor" (of rank 2). In isotropic fluid flows A
reduces to a scalar quantity and (3.52) becomes
V. P = div(A grad v) = AVv + grad A • Vv. (3.53)
The coefficient A or the tensor A must be determined, respectively, by measure-
ments or by statistical investigations.
The Coriolis acceleration 200 x v can be important in large embayments that
are nearly circular and have diameters of 30 km or more. Using the transforma-
tion
dv av av
dt = at + v . Vv = at + V(~V2) - V X rot v, (3.54)
we obtain from (3.51):

av F.p.
-a -
t
v x (200 + rot v) = L: -'-'
P
j
- grad(~v2 + ¢)
1
- - grad p + V' P. (3.55)
P
Ifnonconservative external forces are absent (F j == 0) and if pressure and density
are directly related, i.e. P = p(p), we introduce
dp'
f
p
p v(p) = -(----,-) (3.56)
Po PP
61
and arrive at
8v
at - v x (2m + rot v) = V· P - grad(-tv 2 + ¢ + P v )' (3.57)
Modern mathematics offers a variety of tools for solving hydrodynamic

problems (e.g. Lions, 1969). "Practical" solutions are normally computed
using simplified basic equations. Some examples of those equations are sum-
marized in the next section. Initial and boundary conditions must be added
(cf. section 3.6.1).
3.3.2. Special Types of Fluid Flow

From (3.57) the following special cases can easily be derived.
(a) Stationary, rotational flow of a non viscous fluid:
-tv 2 + ¢ + Pv = constant along streamlines. (3.58)
(b) Irrotational motion of a non viscous fluid (Coriolis acceleration

omitted):
rot v = 0, v = grad lJ'(r, t),
8lJ' 1 2
at + zV + ¢ + P v = f(t). (3.59)
The kinetic energy per unit mass, -tv 2 , depends on lJ'. The quantities lJ'
and P v are unknown. A second equation results from the continuity
equation (3.39):
8p
at + grad lJ' . grad p + p!1lJ' = O. (3.60)
Equations 3.56, 3.59, and 3.60 form a mathematically closed system.

(c) Stationary irrotational motion of a non viscous incompressible fluid
(dp/dt = 0) without Coriolis acceleration: from (3.60) and (3.59),
!1lJ'(r) = 0, (3.61)
where !1 is the Laplace operator, and
-tv 2 + !!. + ¢ = constant (Bernoulli). (3.62)

p
The solution of (3.61) provides the velocity field, v = grad lJ', from
which, after substitution in (3.62), the pressure can be calculated.
62
(d) "Quasistatic approximation" (where vertical accelerations are neglec-
ted): the Coriolis force is approximated by its horizontal component,
(f == 2w cos 0), (3.63)
where e3 is the unit vector toward the zenith, f is the "Coriolis param-
eter," ro is the angular velocity of the Earth's rotation, 0 is the geo-
graphical polar distance, and Vh = e!v! + e2v2 is the horizontal
component of the velocity. It follows that v = Vh + e3 v 3 . If we assume
that F i == 0 and that horizontal and vertical eddy viscosity coefficients
Ah and A 3 differ from each other, then from (3.51) and (3.52):
ap
-=pg. (3.64)
aX 3
In (3.64) we have used the gravitational acceleration 9 and the operator
d a a 3 a
-
dt
= -
at
+ v •V = -
at
+ L Vi - .
i~! aX i
Equations 3.64 are characteristic of "two-dimensional" circulation

models of lakes (single-layer and multilayer models). Boundary
conditions comprise surface wind stresses, bottom drag, and heat
fluxes through the surface. Vertical velocities are computed from hori-
zontal flow divergences if, additionally, the fluid is supposed to be
incompressible:
av! aV 2
div v = 0, ---- (3.65)
ax! aX 2
(e) Equations describing the gross circulation patterns and the average
mass transport in well mixed water bodies are obtained by integration
over depth from the bottom of the basin to the free surface. However,
from an ecological point of view even small parameter variations along
the vertical axis may be very important. Biological cycles are related
to solar influx at the air~water interface as well as to benthic processes
and to vertical exchange in the water column.
63
(f) Ekman-type models of wind-generated currents in the ocean and in
large lakes result from (3.64) and (3.65) by omission of the accelerations
dujdt. Vertical shearing stresses A] D2U)Dx~ (j = 1,2) are balanced by
the Coriolis force (and by a pressure gradient, if a slope current exists).
Table 3.3 provides simplified balance equations.
3.3.3. Waves, Currents, and Circulation; Three-Dimensional Models

Wind-generated surface waves (progressive waves) in shallow water can prevent
aquatic plants from growing and may transport recently deposited material
to deeper areas. Currents and circulation systems, but also internal waves and
seiches in lakes and reservoirs, are of more or less importance in controlling
the extent of heat intrusion into the hypolimnion. They may also influence
nutrient return from the hypolimnion to the epilimnion, for instance in late
summer and fall. For large water bodies with significant spatial gradients in
nutrients and in biomass, three-dimensional hydrodynamic models are
necessary, since some of the most important physical influences on the ecosystem
originate from such relatively short-term phenomena as upwelling and from
eddy diffusivity in localized areas. Turbulent eddies are able to circulate plankton
and to influence sinking rates. Concentrated nutrient loadings may be re-
distributed and transported into nutrient-poor zones, etc.
Classical wave theory (Lamb, 1932) is based on (3.59) and (3.60). Modern
theory involves additionally the concepts of random processes. Since the aim
of this chapter is to summarize basic concepts and equations, we have chosen
to eliminate details of the theories of waves, currents, circulation systems,
upwelling, etc. from the discussion. The interested reader is referred to a number
of excellent texts on these subjects, cited at the end of section 3.3.
3.3.4. Characteristic Magnitudes, Dimensionless Equations, and

Dynamic Similarity
The characteristic magnitudes (maximum values) of the terms of the Navier-
Stokes equation,
av
at + v· V'v + 20> X V = -g - p1 grad p + P1 V' . P, (3.66)
may be denoted by IDv/atl m. Iv' V'vl m, 120> x vim' etc. Here, 9 is the gravity
vector. Introducing
av = I Dv I (~V)', v· V'v = Iv· V'vlm(v' V'v)',

at at m ot
0'1
.j:>.
TABLE 3.3 Simplified Versions of the Balance of Momentum.

Using eqn. 3.52 in eqn. 3.51 and assuming: . . . we arrive at the following balance of momentum:
dv
(i) No nonconservative forces F i , p- + 2pro x v + p grad r/J = - grad p + V' A • Vv
dt
dV h
(ii) No vertical component of v or of the Coriolis force P -dl + pe 3 .r x Vh = - grad h P + V' A· V"-n
(cf. eqn. 3.63),
(3.64)
ar/J _ ap
p aX -
3 aX3
(iii) No acceleration, A becomes a constant scalar A. pe 3 f x Vh = -grad h p + A~Vh

dZvh(z) Ekman
(iv) Vh = vh(z), P = p(z), peJi' x vh(z) = A - - z -
dz
(v) One-dimensional flow in the x direction: v = {u(x, t), 0, O},
p ~ + pu ~ + p ~ = - ~)
without nonconservative forces and Corio1is acceleration, at ax ax ax
~ + ax
~ (A ax
65
etc. into (3.66) and dividing, for instance, by I v • V'v 1m' we obtain the dimensionless
form:
1
St
(au)'
at + (v • n)' + (2m x v)'
"'----R-o~
vV
g'
---
[{1/p)grad p]'
Fr Eu
I
+ Re [{1/p)V' . P]', (3.67)
where
Eu = Iv' V'vl m Euler or Ruark number
I(l/P)V'Plm
Fr = Iv' V'vl m Froude number

Iglm
Re= IV'V'vl m Reynolds number (3.68)
l(l/p)V" Plm
Ro = IV'V'vl
- m-
Rossby number
12m x v 1m
St = Iv,V'vl m Strouhal number.
I av/at
1m
If all these flow numbers are very small compared with unity, the nonlinearity
v • V'v may be neglected in (3.66). It is possible to estimate these numbers with the
help of the characteristic magnitudes of the variables v, p, P, r, t or the" similarity
constants" U, R, P, L, T. Two fluid motions are said to be dynamically similar
if they are related by
v = Uv, p = Rp, p = Pp, r = Lr, t = (L/U)l.
It can then be shown that, inter alia,
UT U U2
St=- Ro=----- Fr=--
L' L· 2w cos E>' gm L '
(3.69)
where Va is the characteristic magnitude of the kinematic viscosity.

The dimensionless forms of differential equations and the characteristic
flow numbers are useful for:
investigating the magnitudes of different terms in these equations (lineariza-
tion and other simplifications of the mathematical model);
solving these equations numerically;
comparing laboratory experiments with field observations; etc.
66
The turbulent exchange coefficient can be expressed as a function of the
Richardson number Ri. Strong vertical density gradients sometimes exist in a
lake, e.g. between epilimnion and hypolimnion. As the difference between the
flow velocities on both sides of this interface approaches some critical value,
disturbances in the flow field grow steadily in amplitude and finally break into
vortices. When the critical value is exceeded, a transition layer is generated
across which there is a velocity gradient av/az and a density gradient ap/az.
Turbulence is maintained or increased if
. g(ap/az)
RI = p(av/az)2 ::; 0.25. (3.70)
The characteristic flow numbers allow the estimation of the type and stability
of flow systems.
Out of the very large number of texts on hydrodynamics we mention:
Lamb (1932), Proudman (1953), Lin (1955), Stoker (1957), Hinze (1959),
Serrin (1959), Eckart (1960), Milne-Thomson (1960), Alder et al. (1964),
Kinsman (1965), Monin and Jaglom (1965-67), Krauss (1966), Vasiliev et al.
(1975), and Kamenkovich and Monin (1978).
3.4. ENERGY BALANCE; THERMAL ENERGY AND

HEAT EXCHANGE
3.4.1. Energy Balance Equations

The following way of deducing the energy balance equations stems from basic
principles. We start with the total energy of the system, which consists of
internal, mechanical, and electromagnetic energy. The electromagnetic field
is included because the propagation and absorption of electromagnetic radia-
tion (as light energy) are often prominent in water quality and ecological
models (e.g. eqns. 3.126 and 3.127).
In the basic equations we have to use field theory. Internal energy comprises
the energy content of the chemical and biological constituents of the aquatic
ecosystem. By subtracting the mechanical and electromagnetic energy (eqns. 3.72
and 3.73) from the total energy (eqn. 3.71), we shall derive the balance of the
internal energy of the ecosystem (eqn. 3.75). Introduction of enthalpy and
specific heat will yield the heat conduction equation 3.78. Use of (3.80) and of
notation q for the right-hand side of(3.78), and specialization to one-dimensional
problems will lead to (3.81). These steps will now be developed in more detail.
The balance of the total energy of the system is given by
d
dt
f.
V(t)
CUe + p(u + 1V2 + ¢)]dV
= f. V(t)
LPiVj,FjdV
i
+ t
~A(t)
[v·(P -Ip) - W - Se] ·dA, (3.71)
67
which states that the time variation of the total energy results from the work
of nonconservative external forces F i (per unit mass) within the volume Vet)
occupied by the system, and from the actions of stress, pressure, heat flow W,
and flux Se of electromagnetic energy on the surface A(t) of Vet). On the left-
hand side of(3.71),
U is the internal energy density (per unit mass),

Ue is the electromagnetic energy density (per unit volume),
tv 2 + ¢ is the sum of kinetic and potential energy (per unit mass).
Multiplying (3.51) by v, we obtain the balance of the mechanical energy:
d
P d- (tv 2
t
+ ¢) = L Pi V • F
i
i - v· Vp + (V' P)· v. (3. 72)
work or pressure rriction

external rorces
The balance of the electromagnetic energy density U e (cr. textbooks on electro-

magnetism, e.g. Stratton, 1941; Landau and Lifshitz, 1960; Phillips, 1962) is
given by
-dive + U e v) - j ' E - i(I), (3.73)

Rux or convection Joule absorption
electromagnetic heat
energy
where the part i(I) of radiation of intensity I taken up by biota (k = 1,2, 3, ...)
can be approximated by
i(I) = ~ Pk J "k(A.)I(r, t, A)dA. (3.74)
Subtracting (3.72) and (3.73) from (3.71), we derive the balance of the
internal energy density u:
du da
P- =
dt
L J. .
i I
F. - div W
I
+j . E - pp -
dt
- P" Vv + i(I). (3.75)
work or heat supply compression dissipation absorption

external rorces
The increase of internal energy results from the work of the nonconservative
forces on diffusive motions J i , from external and internal heat supplies, from
the transformation of convective energy by compression and friction (dis-
sipation), and from the transformation i(I) of radiative energy into biochemical
energy, e.g. by photosynthesis. It is assumed that the specific absorption "k(J..)
of the species k, e.g. of primary producers, depends on the wavelength J.. of the
68
light, the intensity of which is 1. In water quality modeling the Joule heat may
be neglected. We introduce the density of enthalpy, h = u + prx, where rx = lip,
and use the continuity equation (3.39), so that the balance (3.75) becomes
dh dp . .
p- - -
dt dt
+ dlV W = LJ i
i • F; + J' E + p .. Vv + i(1). (3.76)
Assuming that h = h(T, p, N j , B k ) and taking into account that
~I = cp , (3.77)
aT p.Nj,B k
where c p is the specific heat at constant pressure, we convert (3.76) into the
"heat conduction equation" or "temperature equation":
dT .
pCp dt + dlV W* = P"'Vv + '1*, (3.78)
where (for m #- j and n #- k)
'1* = (1 - p :h [ ) ~ + LJ i • F; +j . E + i(1)
P T,Nj,B k I
_"~I p~
j
aN.j p. T,Nm.B k
dN
dt
j
_
p "~I
L.. aB
k k p, T,Nj,B
dB
dt .
n
k
(3.79)
Very often in water quality modeling, '1* is omitted from (3.78), while
W* = W + W** = -K grad T + W** (3.80)
is taken to be the sum of heat conduction Wand the non-Fourier heat supply
W**, including radiation. Equations 3.41 and 3.49 may be introduced into (3.79)
so that we can examine the role of chemical and biochemical processes in the
heat balance (3.78). In general, the energy fluxes between the chemical and
biological components of an aquatic ecosystem and the energy stored within
these components are small compared with the hydrothermodynamic energy
transformations.
3.4.2. Model Equations and Types of Solution

Since (3.75), (3.76), and (3.78) augmented by the continuity equation 3.39 may
be transformed into balance equations of the type (3.2),
p du + div W = ~ pu + div(pu + W) = ...

dt at '
the methods discussed in sections 3.1.4-3,1.7 apply to the treatment of these
equations. Finite-difference and finite-element methods are well suited to
numerical solution. By these means models with quasicontinuous temperature
69
distributions, variable exchange coefficients, and time-dependent thermocline
depths can be generated. Heat exchange with the atmosphere can be represented
as a function of surface temperature. In simple cases (low inflow and outflow,
small variation in thermocline depth, etc.) two- or even multi-layer models have
been successfully developed. Ordinary differential equations like (3.6) can be
used only for spatial averages within a finite volume V.
A very simple one-dimensional river flow model for the transport of heat is
given by
8T(x, t) aT(x, t) 8 2 T(x, t) q(x, t)

----'------'---+u =K +--. (3.81)
at ax ax 2 pCp
Similar equations are used also in reservoir temperature models, taking into
account horizontal advection, as well as vertical eddy diffusivity. The flow
velocity u and the coefficient K of turbulent heat transfer must be measured (or
determined otherwise). Comparing (3.81) with (3.46), we note that mass dif-
fusivity and heat diffusivity often differ even in the order of magnitude. Precise
and concentrated measurements of temperature and other parameters (oxygen,
phosphates, etc.) offer data for estimating mixing processes.
3.4.3. Energy Exchange

In (3.81) the heat supply factor q(x, t) includes the absorption of shortwave
solar radiation (after attenuation in the atmosphere and reflection at the water
surface) and of longwave, indirect atmospheric radiation, as well as heat losses
by the longwave radiation flux from the water, evaporation, etc. The flux of
"sensible heat" (conduction across the interface and convection in the atmos-
phere) can be directed away from or toward the water surface. Heat exchange
processes determine the boundary conditions of thermal models for lakes or
reservoirs. The transmission and absorption of light within the water body are
described approximately by the Lambert-Beer law:
1(z, t) = 1o(t)exp( - tlz), (3.82)
where 1(z, t) is the light intensity or irradiance at depth z. The bulk extinction
coefficient tI depends on the wavelength and is a composite measure of light
extinction by water and absorption by suspended particles and biotic com-
ponents, as well as by dissolved compounds.
As a rule, absorption of solar radiation constitutes the main heat supply to
lakes and reservoirs. However, indirect heating or cooling can become significant
in lakes and reservoirs that have high-volume input from surface runoff or
from groundwater sources. Energy input, temperature, and density stratifica-
tions are important regulators of nearly all physicochemical cycling processes
70
and of biological productivity and metabolism. Therefore, the correct de-
termination and simulation of energy exchange play dominant roles in water
quality modeling.
The problems of this section have been treated by, for example, Meixner and
Reik (1959), Serrin (1959), Truesdell and Toupin (1960), J0rgensen and
Harleman (1978), Kamenkovich and Monin (1978), and J0rgensen (1979a).
3.5. CHEMICAL, BIOCHEMICAL, AND BIOLOGICAL PROCESSES
3.5.1. Basic Processes, Constituents, Interrelationships, and Equations

Potentially significant chemical processes in natural water bodies can be
classified broadly into:
oxidation-reduction reactions
acid-base reactions
gas-solution processes and outgassing
coordination reactions of metal ions and ligands
precipitation and dissolution of solid phases
adsorption-desorption processes at interfaces.
We shall not go into the details of hydrochemistry, but shall confine the dis-
cussion to processes described by (3.41) and to the reaction equations in sections
3.5.2 and 3.5.3.
Aquatic organisms influence the concentrations of many substances by
metabolic uptake, transformation, storage, and release. Particulate and dis-
solved organic matter serve as substrates for decomposer organisms, which,
as a byproduct of their metabolism, generate inorganic substances. Primary
production is the main process by which dissolved inorganic substances
(C0 2 , POl-, HPOl-, N0 3, N0 2, etc.) are consumed. Excretion, egestion,
and non predatory mortality produce dissolved and particulate organic
substances; matter and energy are transported and stored in the food chain;
and chemical elements are cycled through the ecosystem. To a certain extent,
the element cycles are independent, e.g. luxury uptake of phosphorus by phyto-
plankton, and different rates of excretion of nitrogen and phosphorus by
zooplankton and fish. Therefore, "stoichiometric models," coupling the intake
and release of chemicals by the biotic components stoichiometrically to the
growth and remineralization of biomass, are valid only within bounded changes
of the amount and composition of the nutrient loadings to the aquatic eco-
system. Otherwise, "element cycle models" must be used, allowing for an
independent element cycle description. Generally, for the improvement of the
71
.. causal" description of biological processes it is necessary to study certain
details in the laboratory and others in the actual ecosystem in order to obtain the
basic knowledge essential for a good working model.
To formulate the basic concepts of deterministic water quality models suf-
ficiently generally, we consider an arbitrary number of chemical constituents:
dissolved inorganic matter DIM j , j = 1 (l)n 1 ;

dissolved organic matter DOM j , j = n 1 + 1 (I) n z ;
particulate organic matter paM j ' j = n z + 1 (1) n 3 ·
We also consider biotic components BC.:
several species of primary producers (phytoplankton and phytobenthos),

k= 1(1)L 1 ;
several species of herbivorous zooplankton and zoobenthos.
k = L 1 + l(l)L z ;
several species of decomposers, k = L z + 1 (I) L 3 ;
several species at every trophic level of the food chain.
The interrelationships of the chemical and biological components of the aquatic

ecosystem are illustrated by Figure 3.4. Chemical reactions between the dis-
solved components DIM j and DOM j are described with the help of stoichio-
metric coefficients "jro molar masses At j , and reaction rates wr :
The mathematical description of biochemical and biological processes may

be demonstrated by an example: the grazing of one species BC. on several
members BC l that are lower in the food chain, and the effects of egestion and
excretion by BC•.
G. is the grazing rate [T - 1] of BC•. It depends on "internal properties" of

the predator population (physiological conditions, age structure,
carrying capacity, etc.) and on physical and chemical state parameters
of the environment (temperature, oxygen concentration, etc.);
g.IG. is the partial feeding rate of the predator BC. on the prey BC I , where
gkl (dimensionless) describes the portions of different types of prey
populations BC I in the food of the predator BCk and depends, inter alia,
on the preference coefficient t kl and on the relative densities of the
populations (cf. eqn. 3.139);
-..l
tv
Dissolved ( Respiration )-
inorganic
( Primary production) Phytoplankton
constituents ~
and
DIM - - - { Excretion by decomposers )-
phytobenthos
species
I
( Chemical
reactions
1
Grazing
r---<
f 1
Dissolved - - - { Egestion and excretion of DOM) Zooplankton
organic and
,---.
constituents zoobenthos
DOM
~( Uptake by decomposers
I- species
c ~
o 5l
'"
1 0)0
I c c. 1
Decomposer species .- E Grazing
CHydrolysis 1;j 0 H
~ u
I C)Ql
-0
~
I l
I Higher-level
Particulate ' - - - - - ( Grazing by decomposers I-------
food
organic
chain
constituents
I--{ Egestion and excretion of POM "\ species
POM
t Nonpredatory mortality J
FIGURE 3.4 Interrelationships of the chemical and biological components of the aquatic ecosystem.
73
dB,
dt
is the resulting decrease of one prey population BC l ;
= ~*
k B G
-dB L, gkl k k
dt I
is the increase of the predator population BC k caused by grazing on several

preys BC I ;
PklgklBk Gk is the partial rate [T- I ] of egestion and excretion of POM by BC k ;

Ski gkl B kGk is the partial rate [T - I] of egestion and excretion of DO M by BC k
(there is a more detailed description in section 3.5.7);
dB k = L* (1 - Pkl - Skl)gkl Bk Gk
dt I
is the increase of the biomass of the predator population BC k if egestion and

excretion are taken into account.
Other processes in the aquatic ecosystem, e.g. uptake of DOM by decom-
posers, respiration, mortality, and hydrolysis of POM, are treated mathe-
matically in a similar way. Figure 3.5 provides a general overview of the
succeeding development of the necessary equations to describe these processes.
We shall adopt the following summation symbols.
Li chemical and biotic components of the ecosystem including water

(i = 0),
Lj'Lm chemical components of the ecosystem,
Lk'LI biotic components of the ecosystem,
DI] primary producers (k = 1,2, ... , L I ),
D 2
] herbivorous zooplankton and zoobenthos species (k = L I + 1,
L I + 2, ... , L 2 ),
D31 food chain members,
L1 ]
4
decomposers (k = L 2 + 1, L 2 + 2, ... , L 3 ),
Lt members of the food chain before BC k (I < k),
Lt* members of the food chain after BC k (I > k), excluding decomposers,
D5] DI M (j = I, 2, ... , 11 d,
D6 J DOM(j=11 1 + 1,11 1 +2, ,11 2 ),
L~7J POM (j = 11 2 + 1,11 2 + 2, ,11 3 ),
LT chemical reactions (excluding hydrolysis of POM).

-.J
.j::o.
Chemically Bjk Pk R k : input to DIMj by R k : rate of
interd~ting respiration of BC k respiration of BC k
dissolved Phytoplankton and
inorganic DIMj participating
b jl PI ~ : ~ : rate of primary phytobenthos
constituents DIM I in primary production production of BCI species of density p/
i= 1 (11 n, I = 1 (11 L,
DIM/ excreted
elk Pk E k : Ek : rate of excretion
by decomposer BCk by decomposer BC k
1
Sjkl gkl Pk Gk : input to

DOMj by egestion and
excretion of BC k
fk /' Pk" Fk ,,: uptake of DOM I

by decomposer BC k "
~~ him Pm H m : input Fk" : uptake rate

o to DIMj by hydrolysis of DOM/ by BCk~
a
B of POM m
:;
c.
.~
':1 dk"m Pk" Gic,,: grazing of decomposers BC k on POM m

<0:'
~~ Particulate
I organic
constituents POM m
m=n2 + 1 (1) n J
Cmk Pk M k : input to POM m
~v mortality of BC k
M k : mortality rate of BCk J. ,
FIGURE 3,5 Illustration of the mathematical expressions describing the interrelationships between chemical and biological components
of the aquatic ecosystem.
75
In the mass balances (3.41) and (3.50) of the chemical and biotic components
of the ecosystem,
P dN j + div leN) =
dt ,
I l'j,ultjlV, + lj, (3.83)
diffusion chemical biochemical
reactions reactions
dB k
Pdt + d'IV l(B k) = Qk - PkRk - PkMko (3.84)
respiration mortality
the sources and sinks lj and Qk are of the following types.
1. The jth Dissolved Inorganic Constituent DIM j
[4] [11
lj = Ik ajkPkRk + Ik ejkPkEk - IbjkPkPk,

k
(3.85)
respi ra tion excretion by primary

decomposers production
where
Rk is the rate of respiration of BC k [T- 1],

Ek is the rate of excretion by BC k [T- 1],
P k is the rate of primary production of BC k by photosynthesis [T- 1],
a jk is the fraction of DIM j produced by the respiration of BC k ,
I! jk E k is the fraction of the excreted DIM j in E k ,
bjk is the fraction of DIM j used or generated by the primary production
of BC k (b jk < 0 if DIM j , such as oxygen, is produced).
In the aquatic ecosystem, DIM j may be produced or consumed by biotic

components through respiration and excretion, or by primary production.
2. Thejth Dissolved Organic Constituent DOM j
[3]
lj = Ik
I*Sjklgk/PkGk
I
+ I(l - Cjk)PkMk
k
cgestioniexcretion mortality
[4] (7)
-I k
fkj Pk F k +I
m
hjm Pm H m' (3.86)
uptake by hydrolysis
decomposers
76
where
Gk is the grazing rate of the biocomponent BC k [T- 1 ],

gkl are coefficients describing trophic interactions,
gkl G k is the fraction of the lth prey BC I consumed by the kth predator
BC k [T-I],
Sjkl is the input fraction to DOM j by egestion and excretion,
Mk is the rate of nonpredatory mortality of BC k [T- I],
c jk is the fraction ofPOM j generated by nonpredatory mortality ofBCk ,
Fk is the rate of uptake of DO M by the kth decomposer [T - I],
hjF k is the fraction of DOM j in F k [T- 1 ],
Hm is the rate of hydrolysis of POM m [T- 1 ],
hjmH m is the input to DOM j by the hydrolysis of POM m •
DOM j is produced by egestion, excretion (including defecation), and non-

predatory mortality of the members of the food chain, and by hydrolysis of
detritus. DOM j is diminished through uptake by decomposers.
3. The jth Particulate Organic Constituent POM j
[3]
}j = I I* PjklgklPkGk + I
k I k
CjkPkMk
egest ion/excretion mortality
[41 [6]
-I dkjPk G : - IhmjpjH j , (3.87)
k m
grazing by hydrolysis
decomposers
where
Pjkl is the input fraction to POM j by excretion and egestion,

G: is the rate of grazing on POM by decomposers,
d kj is the fraction of POM j consumed by the kth decomposer.
Particulate organic components are generated by excretion (including de-

fecation), egestion, and nonpredatory mortality, but are reduced by grazing
and hydrolysis.
4. The Primary Producers (k = 1,2,3, ... , L 1 )

[2]
Qk = PkPk - I g,kP,G 1 , (3.88)

I
where P k is the rate of primary production of BC k by photosynthesis [T- 1 ].

In the mass balance, given by (3.83) and (3.88), for phytoplankton and/or
77
phytobenthos, apart from primary production, respiration, and mortality,
only the losses by zooplankton/zoo benthos are taken into account (BC , is the
predator; g,k G1 is the partial grazing rate).
5. The Food Chain Groups BC k

[4]
Qk = L* (1 - Pkl - Skl)gkl Pk Gk + L gkl Pk G~

I I
grazing. egestioniexcretion grazing on
decomposers
- L** g,kPIG" (3.89)

I
grazing by higher levels
of the food chain
where
[7] [61
Pkl = L Pmkl and Ski = L Sjkl (3.90)
m j
are measures of the egestion and excretion of POM and DOM by BC k (section
3.5.7). G~ is the rate of grazing on decomposers by BC k •
6. Decomposers (k = L 2 + 1, L 2 + 2, ... , L 3 )
[7] [6]
Qk = LdkmPk G: + L.fklPkFk
m j
grazing uptake
on POM ofDOM
[3] [5]
- L glkP,G; - LejkPkEk' (3.91)

I j
grazing by food excretion bv
chain members decomposers
The uptake of POM and DOM is compensated by grazing and excretion.

The mathematical relationships used in (3.83--3.91) are shown by Figure 3.5.
3.5.2. Energy Balance for Chemical Reactions

In this section biotic components are neglected. The rth chemical reaction at
constant pressure and temperature can be represented by
vAA + vBB + ... veC + vDD + ...
reactants products
or
(3.92)
78
where the dimensionless stoichiometric coefficients are Vjr > for products °
°
but Vjr < for reactants. As a result of this reaction the Gibbs free enthalpy Ge
is changed with respect to the extent ¢r (mol) of the reaction:
drG e = L.ujdn j = L.ujvjrd~r = -Ard¢r. (3.93)
j j
The summation is taken over all components and all phases:
.uj is the molar chemical potential;

nj is the number of moles of the jth constituent.
Ar = - L .ujV jr (3.94)
j
denotes the affinity, and - A r the free energy change of the rth reaction. The
internal entropy production d i S of the total system is then given by
(3.95)
At equilibrium,
diS = 0, (3.96)
The Gibbs free enthalpy is a minimum. However, for a spontaneous reaction,
diS> 0, (3.97)
For the jth component in an aqueous solution, generally
.uj = .uY + R o T In()'jX), (3.98)
where
ji y is the standard state chemical potential,

Ro is the gas constant,
Yj are the activity coefficients,
xj are the molar fractions.
As a convention, the ideal behavior of the solution is specified by Yj ....... I as

°
x j ....... for solutes (and as Xj ....... 1 for the component(s) whose mole fraction(s)
can approach unity in solution). For solutions of electrolytes, "mean activity
coefficients," etc. must be introduced. The values of the standard state chemical
potentials, concentrations, and activity coefficients depend on the scale used.
The molal and molar concentration scales are commonly used for aqueous
solutions. The mole fraction scale is appropriate to satisfy the Gibbs-Duhem
relationship (3.170).
From (3.94) and (3.98),
(3.99)
79
where
Qr = n (~!jX)\'J'
j
(3.100)
is the reaction quotient, and
Kr(p, T) = exp (- L~~~jr) (3.101)
is the equilibrium constant. If Qr < K r , an irreversible reaction tends toward

equilibrium, which is characterized by Ar = 0, Qr = K r. The quantities Ar ,
K" Q" 'Yj, Pj , etc. depend on temperature and pressure. In deep lakes and
reservoirs the hydrostatic pressure may become large enough to require
quantitative consideration of its effects on the chemical parameters and
processes.
For the calculation of chemical equilibrium states and/or the direction of
spontaneous reactions we need the following basic data (at standard pressure
and temperature):
-0
Ilj is the standard state chemical potential, which is equal to the partial
molar free enthalpy gJ,
are the standard state enthalpy and entropy, since gJ = liJ - TSJ,
are the partial molar volume and heat capacity.
Thermodynamic data have been obtained by calorimetry, chemical equilib-

rium measurements, electromotive force measurements of galvanic cells,
calculations of entropy based on the third law of thermodynamics, etc. Chemical
potentials and other useful materials have been published by, for instance,
Latimer (1952), Rossini et al. (1952), Kaye and Laby (1959), Lewis and Randall
(1961), Klotz (1964), Wagman et al. (1968-69), Martell and Smith (1974-77),
and Naumov et al. (1974).
3.5.3. Reaction Rates and Reaction Equations

In the kinetic approach, an elementary reversible chemical reaction at constant
temperature and pressure is described by a mass balance (3.92) and by the
relationship between the forward and backward reaction rates WI' wb :
Kr
W, = - Wb . (3.102)
Qr
As a first approximation we assume that
Wf = n
k f l()'.x.)!Y;rl
· I , Wb = kb n( )'YkXk
Yk •.
• (3.103)
k
reactants products
The rate coefficients k, and kb must be determined experimentally. The reaction

quotient Qr characterizes the direction of the net reaction:
(3.104)
80
when the reaction (3.92) proceeds from left to right;
k
Q, = K, = kr (3.105)
b
at equilibrium. The resulting reaction rate W, = Wr - Wb (mol s - I) is, according

to (3.99) and (3.102),
(3.1 06)
The conclusions (eqns. 3.104-3.106) remain valid even if there are rate-limiting
steps and various pathways in a complex chemical reaction (for example, when
the expressions (3.103) do not contain powers identical with the stoichiometric
coefficients). Most realistic descriptions of the chemical reactions lie within the
nonlinear range A, 2 R o T. Thus, methods are required to study nonlinear
thermodynamics of irreversible processes (section 3.6).
In simplified water quality modeling semiempirical expressions of the
following types are widely used to describe chemical, biochemical, and bio-
logical processes:
(1) dC = KC n (3.107)
dt .
The concentration C becomes irreversibly diminished by a reaction of order

n= 0, by a first-order reaction (n = 1), or in a nonlinear manner (n #- 0, n #- 1,
where n is determined empirically).
(2) (Michaelis-Menten). (3.108)
For instance, the dependence of the uptake rate on the external nutrient
concentration is approximated by this nonlinear expression.
(3) dC I = KC" or (3.109)

dt 2
These simple laws of the nth order for the relationship between the concentra-
tions C 1 and C 2 of two components (with or without saturation concentration
C 2.0) are used, for instance, to describe reaeration or adsorption.
(4) /1 #- O. (3.110)
The Streeter-Phelps equation (eqn. 3.110 with n = 1) and its generalizations

(/1 #- 1) are used in theories of adsorption-desorption processes, of interrelation-
ships between biotic and chemical constituents of the aquatic ecosystem, of the
81
oxygen concentration in rivers, etc. For example, the source-sink terms for BOD
and DO in (3.46) are:
G(BOD, x, t) = -(K I + K 3 )BOD + Q, (3.111)
G(DO, x, t) = Kz(DO max - DO) - K 1 • BOD + P, (3.112)
where
KI is the rate of deoxygenation [T- I],
Kz is the reaeration rate,
K 3 is the rate of sedimentation,
Q is the BOD addition due to runoff and local sources [M L - 3 T- I],
P is the net photosynthetic oxygen generated by primary producers.
By integration of these differential equations the location(s) of the critical

point(s) of maximum oxygen deficit can be determined (oxygen sag curve).
(5) dCI = KI(C 1 0 - CI)q - K z C I. (3.113)

dt .
By (3.113) it is possible to describe adsorption bounded by the saturation value
C I. 0 of the concentration of the adsorbed phase.
(6) dCI - K C Cz (Monod). (3.114)

dt - 1 I K
z
+ Cz
Equation 3.114 is used, for instance, for the change of the biomass C I of a
primary producer if the growth rate is restricted by the concentration C z
of a limiting nutrient. The half-saturation value K z cannot always be assumed
to be a constant. In the case of nutrient inhibition, a Monod expression for
competitive inhibition by a second nutrient C 3 is
dC z _ KC Cz (3.115)
dt - 11 Kz + C z + aI C3 '
where a l is the inhibition constant. For allosteric inhibition we may assume that
dC z _ K C Cz (3.116)
dt - I l ( K z +C z)(l+a Z C )'
3
Very rapid reactions can be described by

dCI dC z
(7) dt = KI dt - KZC 1 (3.117)
and by similar expressions. (References to the literature are on pp. 89-90.)
3.5.4. Chemical Reactions in Natural Water Bodies

In the hydrochemistry of natural water bodies we are seldom interested in only
one chemical reaction. For a system of one or several phases (:x = 1, 2, 3, ...)
82
and several chemical components (j = 1,2,3, ...) the Gibbs potential is
(neglecting biotic components)
Gp(p, T, 11) = I. I. Ilj iii(p, T, xj), (3.118)
a j
where the summation is carried out for all species and phases of the system.
Hence, the change of Gp is
aGp
dG p = - I dp + -aGp I dT+ I.-P
aG I dll j (k i' j)
ap T.n} aT p.n} j all j T.p.nk
= ~ (va dpa - sa d P + ~ iii dll j ). (3.119)
At constant temperature and pressure we locally encounter

(a) equilibrium if
I.a I. iij dnj = 0,
j
Gp = minimum; (3.120)
(b) spontaneous change if

I. I. iii dni < O.
a j
(3.121)
If there are R chemical reactions or other physicochemical processes in an iso-

baric~isothermal system, the complete equilibrium state is defined by
d,G p = -A, d~, = 0 for r = 1,2,3, ... , R. (3.122)

For a system of simultaneous reactions not at complete equilibrium the total
internal entropy production is positive:
(3.123)
If reactants and/or products are common to several reactions, so that the re-
actions cannot be treated independently, the direction of a particular reaction
may not be deduced from its own affinity alone. In natural waters, for instance,
hydrogen ions, electrons, hydroxide ions, oxygen, bicarbonate ions, and several
metal ions are interlocking constituents of a large number of reactions.
The interpretation of multicomponent and multi phase chemical processes
in aquatic ecosystems requires the solution of nonlinear systems of (differential)
equations and, therefore, the utilization of modern numerical techniques and
electronic computers. Most of the computerized models being used to calculate
multicomponent-multiphase chemical equilibria can be classified either as an
"equilibrium constant approach" or as a "Gibbs free energy approach"
(Jenne et al., 1979). In the former approach equilibrium constants are needed
and the mass action expressions are substituted into the mass balance con-
ditions, yielding a set of nonlinear equations that must be solved simultaneously
or by an iterative procedure (Newton-Raphson method, linearized matrix
83
inversion, etc.). This approach is often preferred, especially for the calculation
of large complex systems.
In the Gibbs free energy approach, the total Gibbs potential is minimized
for a given set of chemical species and their mole numbers subject to the mass
balance requirement. Optimization techniques such as pattern search, linear
programming, steepest descent, and gradient methods are appropriate to the
solution of the mathematical problems. For relatively simple systems for which
Gibbs free energy values are available and reliable, this approach is convenient.
The method will probably find greater use as soon as accurate and consistent
sets of thermodynamic data become available.
Computerized chemical models suited for calculating equilibria in aqueous
systems have been described and compared by, e.g., Jenne et al. (1979), and
van Zeggeren and Storey (1970).
Using one of the above-mentioned approaches, we assume that a unique
solution exists. Natural aqueous systems (or aquatic ecosystems) may have
several thermodynamically metastable states corresponding to local minima
in the Gibbs function. Therefore, nonunique solutions of the equilibrium
problem are possible, since the solution may occur not only at the most stable
equilibrium point but also at a local minimum. For the application of the above-
mentioned computerized models in aquatic chemistry, this problem of non-
uniqueness is very important. For instance, the problem arises in the interpre-
tation of solid precipitation and dissolution processes.
Furthermore, a number of processes in rivers, lakes, reservoirs, and other
natural water systems cannot be treated by the "equilibrium theory" of chemical
reactions. The flow of matter and energy, gradients ofconcentration and tempera-
ture, and exchanges of matter and energy with the surroundings characterize
an open, continuous thermodynamic system that is poorly approximated by
the equilibrium state of a closed homogeneous system. Knowledge of the
temperature dependence of some of the standard quantities (section 3.5.2)
is sometimes inadequate. Moreover, the characteristic time scales of river
flows and reservoir management often do not allow for chemical equilibrium
if the reactions are very slow. Local equilibrium conditions might be expected
if the reaction rate is sufficiently high, even in a slowly changing environment
(e.g. at the sediment-water interface). Local equilibrium within each small mass
element can be realistically assumed to exist (section 3.6.3), but over larger
regions such as the biologically active photic zone of a eutrophic lake, thermo-
dynamic equilibrium is never realized. In such cases it is necessary to use the
method of thermodynamics of irreversible processes (section 3.6).
3.5.5. Photosynthesis and Primary Production

Through the medium of photoactivated plant pigments (chlorophylls), light
energy in the wavelength range 390-710 nm is transformed into chemical energy.
84
Carbohydrates, proteins, nucleic acids, and other metabolites are formed from
CO Z, HzO, and constituents containing N, P, S, Fe, etc. The photosynthetic
production of algal protoplasm and the reverse destructive processes are
summarized by
106CO z + 16NO; + HPO~- + 122 HzO + 18 H+

+ trace elements + energy ~ C106Hz630110N16P + 1380 z .
(3.124)
The net primary productivity is defined by gross photosynthesis (light and dark
bottle) minus loss by respiration during the period of light minus losses due to
respiration and mineralization over the daily cycle.
The primary production rate P k of the kth biotic component BC k of the
ecosystem (phytoplankton or phytobenthos) is a complicated function of bio-
mass, local light intensity, temperature, pH, vitality of BCk> availability of
intracellular and extracellular nutrient supply, and intracellular processes. In
water quality modeling, P k is approximated by expressions of the type
(3.125)
where J(...) represents functions of different types. Different assumptions about

P k, 0 and the functions of light intensity I, temperature T, and nutrient con-
centrations Pi are used. Some examples are given below.
(I) The maximum rate Pk,o for optimum conditions of light, temperature,
and nutrients is supposed to depend mainly on cell properties.
(2) For the description of the relationship between photosynthesis and
irradiance, functions of the type
1/1*
J(I) = [I + (I/I*)Z]IIZ' I(z, t) = lo(t)exp( - '1z), (3.126)
or
rJ. > 0, (3.127)
or a Michaelis- Menten expression, or other assumptions are used.

lo(t) includes the annual and diurnal variations.
(3) The effects of temperature on biological rates are approximated, for
instance, by
T
J(T) = a + b- or J(T) = KT-To, (3.128)
To
85
or
(3.129)
(3.130)
(4) The limiting expression f(p) can be, for instance,
f(p) = n
[5)
j
K,
j
p"
~ P., j
(3,131)
or
(3,132)
where w j are weights, or

pj
f(P"P2,P3'''') K* (3.133)
=
j + Pj*'
where pj is the mass density of the limiting nutrient.
For very low nutrient concentrations,
Pj ~ K
Kj + Pj ~ Pj
Sometimes, the Monod kinetics of nutrient uptake are replaced by other

assumptions. An example is
_ 1_ (_ Lj ckj(Pj - Pj.min))
.{( P"P2"" ) - exp . (3.134)
KkPk
The total photosynthesis per unit surface area of a lake can be determined by
integrating over the depth z. If the extinction coefficient 1] is constant, then
from (3.126):
f°oo
f(/)dz =
1{I [ + (I1~
~ In 1~ + 1
)2]'/2} (3.135)
::::: ~I In (10)
2 1* if10~1*. (3.136)
Logarithmic light expressions can be found in water quality modeling, but it

should be recognized that the whole of (3.126) must be integrated over z and
that, as a rule, 1] depends upon z.
86
Observations of algae kinetics show that intracellular concentrations Pkj
of the jth inorganic nutrient in BC k must be introduced; thus (3.85) is generalized
to
[4J [11
Y; = I
k
ajkPkRk +I
k
f'jkPkEk - Ik
Pk U kj , (3.137)
where Ukj is the rate of DIM j uptake by the kth phytoplankton component.
A simple assumption is that the uptake of nutrients is controlled by the vitality
(or temperature) and by both the extracellular concentration Pj and intra-
cellular concentration Pkj of the nutrient:
(3.138)
If we suppose that photosynthesis and primary production are controlled by

intracellular, but not by extracellular concentrations, the function f(p) in
(3.131)-(3.134) must be replaced by f(Pk). Figure 3.6 shows this variant of
primary production.
Nj : mass fraction of the Changes of concentrations

nutrient DIMj
Pk Ukj : uptake of DIM j by the Ukj :cf. (3.138)

primary producer BC k
Pkj : intracellular concentration

of the jth nutrient in the
biocomponent BC k
Pk P k : primary production by
the biocomponent BCk
where
Pk = Pk,o f ( I ) f (T) f (Pkl'

Bk = Pk / P : mass fraction of the
biocomponent BCk Pk2' Pk3' ... )
FIGURE 3.6 Intracellular storage of nutrients and primary production by the bio-
component Be,.
87
3.5.6. Trophic Interactions (Grazing and Feeding Rates)
The feeding relationship forms an important nonlinear linkage between the
biotic components of the ecosystem. At intermediate biomass densities the
feeding rate depends on the biomass of both interacting populations. A prefer-
ence term differentiates between various types e1f food. The feeding rate function
exhibits a maximum at a specific temperature and is generally higher for
juveniles than for adults. It decreases when the population reaches its maximum
density.
Formulating the partial feeding rate gklGk of the predator (consumer) BC k
on the prey (food supply) BC I , we use the Monod kinetic:
tkl(PI - PI. min)! Pk

(3.139)
(3.140)
where
t kl is the preference coefficient (for BC I as food for BC k ),

PI,min is the threshold concentration of the prey population below which the
grazing on this population is zero; the assumption PI, min ---> 0 is im-
portant for the simplification of the model,
Pk, max is the carrying capacity (maximum possible density) of the predator
population,
F(Pk/Pk, max) ---> 0 if Pk ---> Pk, max'

Ck is a correction factor (not specified here) that takes into account
physiological effects, the age structure of the predator population,
oxygen concentration, toxic impacts, etc.
Similar expressions are used for the rate of grazing, gkl G~, on decomposers by
the BC k and for the partial rate ofuptakeJkJk, ofDOM j by the kth decomposer:
r _ tkj(Pj - Pj,min)/Pk
(3.141)
.Ikj - K k(T) + '\'l61(
L,m Pm - Pm,min )/ Pk .
Equation 3.139 involves two asymptotic cases.
(a) For optimum feeding conditions the total feeding rate of BC k is
(3.142)
88
(b) For very low food concentrations,
Gk t kl
gkl GkPk ::::: ~ (P, - P" mil,). (3.143)
Sometimes an exponential law is used instead of (3.139):
'\'*
L gk/
G - G
k - k
[1 _exp (_ '\'*
L
tk,(p, -
K
PI,min»)] ' (3.144)
I I Pk k
leading also to the asymptotic expressions (3.142) and (3.143).
3.5.7. Egestion and Excretion

The fraction of food consumed but not assimilated by the kth component BC k
of the food chain (zooplankton, zoo benthos, and the higher trophic levels)
consists of POM and DOM. The partial egestion rates are
PjklgklGk for POM and Sjklgk,G k for DOM j .
Hence, the partial egestion rate connected with the trophic interaction between
the predator BC k and the prey BC, is
[7] [6])
( ~ Pjkl + ~ Sjkl + Skl)gkl Gk·

gkl Gk = (Pkl (3.145)
The fraction of food assimilated and subsequently excreted (including

defecation) by BC k also consists of POM j and DOM j . The partial excretion
rates are
Pjk I* gklGk for POM j and Sjk I* gklGk for DOM j .
I I
Using the notations

Pjkl = Pjkl + Pjb (3.146)
[7] [6]
Pkl = Ij
Pjk" Ski = Ij
Sjkl' (3.147)
we may combine the processes of egestion and excretion by BC k :
E(BC k) = I* (Pkl + Skl)gkl Gk· (3.148)

I
The relations (3.86), (3.87), and (3.89) are based on (3.148).

The rate of excretion of dissolved inorganic matter by the kth decomposer is
approximated by
[6]
E k = Ek,o I ikJk (3.149)

j
89
or by
[6]
E k = E k. O I (f~Jk - R k) (3.150)
j
if uptake I~6] !kjF k is greater than respiration. E k. O depends on physiological

conditions, which are in turn related to the state of the environment.
3.5.8. Respiration and Nonpredatory Mortality

Respiration and mortality losses are considered proportional to biomass and to
ingestion, respectively. Modifications result from the physiological responses
to the structure and changes of environmental parameters (e.g. temperature,
dissolved oxygen). Nongrazing mortality may be caused by overcrowding, if
the population approaches the maximum that can occupy the ecosystem (the
system carrying capacity).
One of the following types of expressions may be used for the respiration
rate R k of the BC k :
R k = ak!(T), (3.151)
Rk = bk!(T)F(~),
Pk. max
(3.152)
R k = ck!(T) I* gk/G/, (3.153)

/
R k = dk!(T) I* (1 - Pk/ - SkagklGk, (3.154)

/
or a linear combination of (3.151) or (3.152) with (3.153) or (3.154).

For the rate of nonpredatory mortality M k a number of different approxima-
tions are in use; for instance,
M k = qk!(T)F(dT)(1 - Ct. k I** g/kG/Pl) (3.155)

dt I Pk
or laws of the types (3.151 )-(3.154) as well as linear combinations of these.

According to (3.155), in addition to the temperature effect!(T), a second func-
tion F, depending on the time variation of the temperature, influences the
mortality rate. Furthermore, the non predatory mortality is diminished by
grazing losses (0 < ct.k < 1; Ct. k ~ 0.5).
The reader interested in a more detailed exposition of hydrochemistry and
hydrobiology is referred to the literature, e.g. Stumm and Morgan (1970),
Rheinheimer (1971), Ciaccio (1971-73), Daubner (1972), Lehninger (1972),
Romanenko and Kuznecov (1974), Golterman (1975), Sa1<inki and Ponyi
90
(1975), Uhlmann (1975), Wetzel (1975), and Goldberg et al. (1977). Methods of
mathematical modeling are dealt with in the following chapters and also by,
e.g., Volterra (1932), Vinberg and Anisimov (1966), Menshutkin (1971), Biswas
(1972), Romanenko and Kuznecov (1974), Bagotskiy and Bazykin (1975),
Romanovskiy et al. (1975), Antomonov (1977), Mitropolskiy (1977), JlJrgensen
and Harleman (1978), and J0rgensen (I 979a). Data are available in a handbook
edited by JlJrgensen (I 979b).
3.6. FURTHER CONSTRAINTS ON WATER QUALITY MODELS
3.6.1 Initial and Boundary Conditions

To represent the water quality model by a well posed mathematical problem,
it is necessary to supplement appropriate initial and boundary conditions of the
system of differential equations. These equations describe the physical, chemical,
and biological processes within the water body, while the boundary conditions
reflect the constraints of the surrounding world acting upon the fluid motions
and upon the components of the aquatic ecosystem.
The number and type of initial and boundary conditions depend strongly
on the nature of the particular water body, on the specific problem of interest,
and on the type ofmodelthat weintend to use. Therefore, the initial and boundary
values in water quality modeling are of considerable variety. We confine our-
selves to some general remarks, leaving the formulation of particular initial-
boundary-value problems to later chapters that discuss special models.
An aquatic ecosystem exchanging energy and matter with the external
world is called an "open system." In this case, at the surface separating the water
body from its surroundings, we must prescribe the boundary values of
the fluxes of energy and matter across the surface, or

the temperature and the concentrations (e.g. densities Pi) of the constituents
of the ecosystem, or
linear combinations of these quantities at the surface.
For instance, if thermal effects are included, a boundary condition on the

temperature is required, specifying either the temperature or the heat flux, or a
linear combination (boundary conditions of the first, second, or third type).
At the free surface of a water body or at an interface between two different
water bodies (one of which we are investigating), a kinematic boundary con-
dition must be fulfilled and the stress vector should be continuous. As a rule,
this guarantees the continuity of the pressure and velocity distributions across
the free surface or interface. The energy fluxes across this surface must be
specified. There is no exchange of water through this type of surface.
91
At solid boundaries the fluid must adhere to the solid, at least at low tan-
gential stresses and high pressures, if the viscosity is significant. If the viscosity
is not large, however, we may safely assume that the fluid slips along the surface
without appreciable drag. The energy fluxes across this boundary must be
specified. If the solid boundary is moving a kinematic condition must be imposed.
Of very great importance in water quality modeling is the exchange of matter
(nutrients, biota, detritus, etc.)
through free surfaces by precipitation, nitrogen fixation, outgassing, etc.,

across interfaces of water bodies by diffusion,
at solid boundaries by sedimentation, adsorption, desorption, chemical
reactions, transport of interstitial water into the free water, etc.
Sometimes it is very difficult to prescribe the boundary values since they must
be measured, but it should be carefully done, especially if the values are time-
dependent. Changed environmental influences may lead to the activation of
"possibilities" of the ecosystem not realized formerly.
3.6.2. Entropy
The concept of entropy was first introduced in physics with reference to thermo-
dynamic phenomena. In recent years this idea has been so generalized that it
has found fruitful application in fields far from thermal physics. For instance,
in chemistry, cosmology, information theory, and biology, entropy has revealed
itself as a powerful tool, as a comprehensive measure of irreversibility if we look
at the system from outside, and as a measure of stability seen from inside the
system.
The second law of thermodynamics postulates the existence of a state func-
tion S, the entropy, which is linked to other state functions by the fundamental
Gibbs relation (eqn. 3.160). The change of entropy with time can be split into
two parts, the entropy production inside the system and the exchange of entropy
with the outside world (eqn. 3.186). The internal production, caused by dissipa-
tive processes, is never negative. The entropy flow across the boundary of the
system is a necessary condition for its maintenance far from thermostatic
equilibrium. Many of the physical and chemical processes in natural water
bodies, as well as practically all biochemical processes in living systems, occur
far from equilibrium. Therefore, the entropy principle offers the possibility of
making deductions about the existence of stable, stationary states of an eco-
system.
Furthermore, without integrating the nonlinear model equations, we can
derive from the entropy balance some conclusions concerning the evolution
and further development of the aquatic ecosystem. Under unstable conditions,
92
relatively small fluctuations (in space and/or time) of physical, chemical, and/or
biological components or parameters of the water body (e.g. tides, seiches,
day-night dependence of biological processes) can suddenly increase and give
rise to a transition of the ecosystem to a "new" stationary state. This is especially
likely when the environmental conditions or some internal variables/parameters
of the ecosystem have undergone such variations that the inherent stability,
characteristic of the stationary state of the ecosystem, has been lost. From this
point of view, we may approximate the anthropogenically influenced history
and further development of lakes, reservoirs, and other water resources as a
succession of transitions from one stationary state to a new one caused by
fluctuations of ecosystem parameters or components under unstable conditions.
The entropy principle also connects the phenomenological, stochastic, and
cybernetic approaches to water quality modeling, as may be indicated by the
following remarks and formulas.
(a) Entropy can be defined as "potential information" (Shannon):
(3.156)
where Pk is the probability of each microstate. Entropy measures how

much information we would gain if we knew not only the" macrostate"
of the system but also the "microstate" within it (von Weizsacker,
1972). In information theory and ecology the function (3.156) is called
the Shannon index and is used as a measure of the information content
of a code. It is a special case (<5 -> 1) of a more general class of" diversity
indices." For a code of n kinds of symbols and proportions Pk of the
kth kind (k = 1,2,3, ... , n), the function
(3.157)
is known as the entropy of order <5 of the code or, equivalently, of the
set (Pk) (Pielou, 1975).
As the number of microstates or "elementary arrangements"
contained in a macrostate of the system increases, the entropy increases.
If we take into account the binding energies and if the total energy of
the system (or the temperature) is sufficiently small, macrostates of
high order, like crystals, can have high entropy. Entropy is not the
"general measure of disorder." Diffusion processes and heat conduction
in gases, by which it is usually demonstrated that disorder increases
together with entropy, are insufficient examples for the discussion of the
meaning of increasing entropy, since gases generally are without
structure.
93
(b) Microscale and macroscale fluctuations of physical, chemical, and/or
biological components of the ecosystem are typical phenomena in
water bodies. In spite of these fluctuations, some macroscale character-
istics of the system are constant. Only after a fluctuation has occurred
and been extended to a sufficiently large macroscopic range may it
give rise to a change in the state of the ecosystem. Therefore, besides
the stability of the stationary state we must specify the a priori prob-
ability of there being fluctuations and the probability that a fluctua-
tion spreads and attains a large amplitude and range, allowing for
changes in the state and structure of the ecosystem.
The determination of these probabilities belongs to the stochastic
theory of aquatic ecosystems. The probability W of fluctuations around
a stationary state is connected with the second variation (52 S of entropy
by the Einstein formula:
(3.158)
where k B is the Boltzmann constant. This equation offers a statistical

foundation for the thermodynamic stability criteria.
(c) The processes in living systems are controlled by special regulating
factors. Therefore, the theory of aquatic ecosystems should be based
on a combination of thermodynamics and cybernetics. Cybernetic
water quality models contribute valuable knowledge about
the feedback mechanisms in systems,

the isomorphic structures of complex systems,
the evolution of self-organizing systems, and
the hierarchy of systems and subsystems, etc.
The value of this knowledge for the investigation of water bodies has
been pointed out, for instance, by Lyapunov (1972) and Straskraba
(1977). On the other hand, the formation and stability of structures of
aquatic ecosystems, the feedback mechanisms within them, etc. depend
strongly on physical, chemical, and biological constraints. Therefore,
only the combination of thermodynamic methods, taking into account
the basic physical, chemical, and biological laws, with the stochastic
and cybernetic investigations of the behavior of water bodies can
ensure the further successful development of water quality modeling.
The progress of stochastic methods for the investigation of non-
equilibrium systems has been extraordinary in recent years (cf. Nicolis
and Prigogine, 1977.) Since this chapter is concerned primarily with
94
deterministic water quality modeling, we confine ourselves to the
thermodynamic aspects, especially to:
formulation of the generalized Gibbs relation (section 3.6.3);

derivation of the entropy balance equation for aquatic systems
(section 3.6.5);
stability criteria (section 3.6.7);
ansatz equations for primary production (section 3.6.8).
3.6.3. The Generalized Gibbs Relation
An extension of the concept of entropy is needed for application to open systems

and to situations far from thermostatic equilibrium since, from the thermo-
dynamic point of view, the aquatic ecosystems of lakes, reservoirs, rivers, and
other types of water resources are
"open systems" in nature, exchanging energy and matter with their sur-
roundings;
far from thermostatic equilibrium because matter and energy flow through
them;
nonlinear systems because of the nonlinearity of the internal processes: the
Reynolds stress terms, the temperature dependence of chemical reaction rates,
etc.
Important exchange processes between the aquatic ecosystem and the external
world include the transfer of energy, mass, and biomass across the water-water,
air-water, water-plant, and water-soil interfaces, insolation, nitrogen fixation,
input of chemicals by precipitation, and phosphate release from sediments.
The exchange of energy and matter is an essential element of aquatic plant
and animal populations.
The Gibbs relation may be generalized and applied to water bodies if
the concentration gradients are limited, in that the mass densities of the
constituents do not vary appreciably within distances of the order of the
mean free path;
chemical reactions are not too fast, in that the rate of elastic collisions is
larger than the rate of reactive collisions, which in general is true also for all
biological processes in dense media.
95
Interfaces inside the ecosystem call for special "jump conditions" (Glansdorff
and Prigogine, 1971; Nicolis and Prigogine, 1977).
Thus, our basic postulate is that not only the physical and chemical processes
but also the biological phenomena can be studied with reference to the laws of
generalized thermodynamics, at least to a good approximation on the macro-
scopic scale. The system as a whole is out of thermostatic equilibrium. However,
it is assumed that the macroscopic evaluation of internal energy, entropy, and
entropy production, although very difficult for biological components, still
remains possible, since" local equilibrium" exists within each mass element of
the medium. Between the thermodynamic quantities u, s, T, etc. and the local
macroscopic variables the same relations are supposed to be valid as between
the state variables of the whole system at equilibrium. Thus, we assume that the
internal energy density u (per unit mass) of the water body is determined by the
local physical, chemical, and biological state parameters and one further,
dimensionally independent scalar parameter, the entropy density s:
u = u(s, IX, Nj , Bd. (3.159)
We also assume the Gibbs relation to be valid:
T ds = du + P dlX - I. Pj dNj - I. 13k dB k· (3.160)

j k
For the description of systems containing electrolytes the chemical potentials

Pj must be replaced by the electrochemical potentials. We introduce the
potentials 13k of the biotic components. In a first approximation 13k is equal to
the free enthalpy of BC k • In this idealized description the trophic relationships
are determined only by differences in the energy contents of BC k and BC/.
The driving force is proportional to the difference, 13k - 13/ = 9k - 9/. However,
in reality the trophic relationship results also from regulatory processes that
probably cannot be described adequately by chemical kinetics. We shall not
go into detail here.
Differentiating (3.160) in the direction of center-of-mass motion produces
T ds = du + dlX _ ,,9!'j ._ I. dB k 13 . (3.161)

dt dt P dt 'J dt Pj k dt k
This is the generalized Gibbs relation, which may be applied to water quality
modeling (Mauersberger, 1978, 1979).
3.6.4. Thermodynamic Relations

In this section we bring together some of the relations used in the thermodynamic
theory of aquatic ecosystems. Adopting the "proper" independent variables of
96
the thermodynamic quantities u, g, s, etc., we may easily derive from (3.159)

and (3.160):
u(s, ~, N j , Bk) = Ts - p~ + L J.1 j N j + L PkBk, (3.162)
j k
ou ou ou ou
os = T, o~ = -p, oN
j
= J.1j' OBk = Pb (3.163)
g(T, p, N j , Bk) = u - Ts + p~ = L J.1 j N j + L PkBb (3.164)

j k
og ag og
oT = -s, op = ~, oN. = J.1j' (3.165)
)
Using further the independent variables T, p, N j ' Bk , we find from equation

3.165 :
(os)
op T.NjoB. = -
C~)
aT p.Nj,B.
(3.166)
os CJ.1j -
os a/Jk OJ.1j o~
(3.167)
oN j - ci' aBk oT' ap oN·'
)
apk o~ 0J.1i CJ.1j OPk OJ.1j

- -
(3.168)
cp oBk ' cN j ONi' oN j aBk
The thermodynamic relations
(opo~)
Cp
- = -~x (3.169)
T.Nj.B. ' T
and the Gibbs~Duhem relationship
-~ dp + s dT + L N j dJ.1j + L Bk dPk = 0 (3.170)

j k
are valid too (X is the isothermal compressibility).
3.6.5. The Entropy Balance Equation

The entropy balance equation follows from the Gibbs relation and from the
basic equations 3.75, 3.83, and 3.84:
c(ps) .
----at + dlV S = (1. (3.171)
The entropy flow density S is
(3.172)
97
For the dissipation Ta,
Ta = TWgrad(~) +j'E + i(l) + P"VV

heat flow Joule absorption friction
heat
diffusion of chemical components
"'diffusion"' of biotic components

[tl
+ L w,A, + L PkPkA(Pk)
k
chemical primary
reactions production
+ L Pk(RkA(R k) + MkA(M k))

k
respiration mortality
[3]
+ L Pk( Gk A(Gk) + G~ A( G~))
k
grazing in the food chain and on decomposers
[41
+ L Pk(GtA(Gt) + FkA(F k) + EkA(E k))
k
grazing on paM, uptake of DaM. and excretion by decomposers
[7]
+ L PjHjA(H) (3.173)
j
hydrolysis
Dissipation is produced by heat transfer, Joule heating, absorption of light

(especially by primary producers), friction, diffusion of chemical and bio-
logical components, chemical reactions, primary production, respiration,
mortality, grazing, excretion, and hydrolysis of particulate organic matter.
The entropy production a is a bilinear function of "flows" and "forces";
(3.174)
Flows W j
Forces grad(l/T) E
98
The" affinities" are defined by
A, = - L vj,.Jt j J1j (chemical reactions) (3.175)
j
[5]
A(P k) = L bjk J1j - 13k (primary production) (3.176)

j
[5]
A(R k) = 13k - L ajkJ1j (respiration) (3.177)

j
[6] [7]
A(M k) = 13k - L (l - C jk)J1j - L Cmk J1m (mortality) (3.178)

j m
(grazing)
(3.179)
[4]
A( G~) = L gkl(131 - 13k) (grazing on decomposers) (3.180)

I
[7]
A(G*) = L dkm(flm - 13k) (grazing of decomposers on POM) (3.181)
m
[6]
A(F k) = L j~iflj - 13k) (uptake of DOM by decomposers) (3.182)

j
[5]
A(E k) = L ejk(13k - fl) (excretion by decomposers) (3.183)
j
[6]
A(H m) = L hjm (J1m - J1) (hydrolysis). (3.184)
(The summation symbols were defined in section 3.5.1, p. 73.)

Integrating the entropy balance equation 3.171 over the whole ecosystem,
we obtain the total entropy production Ps :
P = f.adV = i LJnXndV ~ 0 (3.185)

v Jv n
and the balance equation
o.S = _ ~ S' df+ P =

(3.186) oe S + DjS
(It:t at ' s - at
arriving at the decomposition of the total change, as/at, of the entropy of the
system into an external part 0eS/ot related to the flux from or into the surround-
ings and an internal part OJ S/ot, which, by the second law of thermodynamics,
can never be negative.
99
3.6.6. The Linear Region of Thermodynamic Theory
In the linear range of irreversible processes it is assumed that the Onsager
relations are valid:
L nn > 0, (3.187)
Open linear systems subject to time-independent external constraints reach,

after a sufficiently long time, a steady nonequilibrium state characterized by
constant entropy production, (J = (Jo or P, = Po. Stable steady states exist.
To maintain such a state, a continuous negative flow of entropy into the system
is necessary.
The stationary states and the evolution of linear open systems are character-
ized by the theorem of Prigogine:
dP,
_. <0
P, ~ 0, (evolution) (3.188)
dt
P, = Po = minimum, dP.,
dt
= ° (steady state). (3.189)
The total entropy production P., becomes a minimum compatible with the
constant constraints applied to the system. According to Lyapunov's theorem,
(3.188) guarantees the stability of the steady nonequilibrium state because the
Eulerian derivative dPjdt is semidefinite of sign opposite to the Lyapunov
functional P.,. Perturbations of this state (either internal fluctuations generated
by the system itself or external excitations) give rise to an entropy production
larger than (Jo or Po, respectively. According to (3.188) the system is driven back
to the reference state (Jo or Po. In linear open systems entropy production plays
the same role as thermodynamic potentials in closed systems.
The linear Onsager theory is valid, as far as transport phenomena are con-
cerned, but in order to set up a satisfactory description of chemical and bio-
chemical reactions or biological processes, it is necessary to extend the theory to
the nonlinear range.
3.6.7. The Nonlinear Region of Thermodynamic Theory; Stability Criteria

In the nonlinear range of irreversible processes, neither eqns. 3.187 nor 3.188 and
3.189 are valid. Under the condition of fixed concentrations·or zero fluxes at the
boundaries, the evolution criterion of Glansdorff and Prigogine (1971),
(3.190)
100
states that the change of the forces X n always proceeds in such a way as to lower
the entropy production p.,. However, ax p,jat = 0 does not necessarily mean that
P s goes to a minimum. No state variable or potential is known that tends to a
minimum or maximum if a nonlinear system is far from thermodynamic
equilibrium. While no .. global" equilibrium criterion exists, a "local" stability
criterion may be derived. Starting from the Gibbs relation (3.160), we can
calculate the second-order differential of the entropy s(u, IX, N j , Bk ):
2
b s = b(~)bU + b(:)blX - tb($)bNj
- ~ b(~ )bBk. (3.191)
Again, using (3.160) we arrive at
T b2s = -bT bs + bp blX -I bpj bN j - I bPk bB k· (3.192)

j k
If now p, T, N j , Bk are supposed to be independent variables, from (3.192) and

(3.166)-(3.170) it follows that
Tb 2s = - ( IX X(b P)2 + cp (bT)2
+ '\ '\ apj '\ '\ apk )

L.. L.. ~a~ bN m bN j + L.. L.. aB bB I bBk . (3.193)
j m Nm k I I
Thus, the function
(3.194)
may serve as a Lyapunov function if the necessary and sufficient conditions
X> 0,
and the sufficient condition
dL
->0 (3.196)
dt -
are fulfilled. Since L is a homogeneous function of second degree in the per-

101
turbations bN j , bB k , bv, bp, bT (or bu, blX), we obtain from (3.196) by means of
Euler's theorem
~ dL = b(~) dbu + b(~) d blX _ ~bv' d bv

2 dt T dt T dt T dt
" -(f3k) -
- L. 0 -
k
d bB
T
--
k
dt
,,-(flT
L. 0 -
j
j) d bN j
--:;:::
dt
O. (3.197)
The time derivatives on the right-hand side of (3.197) are given by the energy
and mass balance equations of the perturbed state. The inequalities (3.195)
and (3.197) represent a general criterion of the local stability of nonequilibrium
states or processes (including convective as well as dissipative effects) and can
be used to obtain explicit criteria for specific situations.
Let us consider a steady state solution that corresponds to thermodynamic
equilibrium, e.g. to the minimum of the specific free enthalpy ge for a closed
system at given temperature and pressure. Far from equilibrium this steady state
is stable if the excess entropy production bxPs is positive definite:
bxPs = rI
Jv n
blnbX n dV > 0 (3.198)
The vanishing of bxPs determines the thermodynamic threshold (or "bifurca-

tion point") that separates the stable regime (bxP s > 0) from the unstable
regime (bxPs < 0). The relation
(3.199)
enables us to calculate the constraints acting on the ecosystem, which are
capable of inducing an instability of the state under consideration. Beyond
this instability another solution of the nonlinear system of basic equations may
exist, radically different from the former type of solution, describing a new type
of structure of the ecological system. The new structure is maintained in con-
ditions far from equilibrium by flows of energy and matter into, through, and
out of the system. The variety of steady states accessible to an open system may
become much larger in such conditions. Chemical and biological processes
involve situations beyond the instability. Therefore, investigations of this type
are of great importance in water quality modeling. Stability criteria derived from
the entropy principle must play an important role in these investigations.
(Of course, we have to differentiate exactly between the "biological stability"
of the ecosystem, the stability of the mathematical model, and the stability of
the numerical method for the solution of the model equations.)
3.6.8. Thermodynamics of Primary Production

Thermodynamic principles, acting as "constraints" in the theory of water
quality modeling, may be used to improve the modeling of biological processes
102
and, furthermore, their interrelationships with physical phenomena like dif-
fusion. At least the types of mathematical functions suited for describing these
processes can be inferred, for instance, from the local entropy production.
Logistic growth laws are shown to govern the processes of net primary pro-
duction and of grazing. As soon as accurate thermodynamic data are available,
the method will probably find greater use in water quality modeling.
In this section, only the thermodynamics of primary production can be
outlined. A more complete presentation is given by Mauersberger (1982).
Basic Assumptions
The thermodynamic theory of primary production is based on the following
assumptions.
(a) Autotrophic species transform light energy into chemical energy

via photoactivated pigments, mainly chlorophylls, and by means of
a complex series of reactions. As a result, an amount [k of energy per
unit mass of the kth species is stored, e.g. incorporated in adenosine
triphosphate (ATP) molecules and other "internal components"
(f = I, 2, 3, ... , h). For every species the balance of this internally
stored energy is formulated as
dtk _ '"' dmkf

dt - L.J1kf & (3.200)
f
dtk
-d = <Dk(T, 1)K k - Ap kPk
t . '
+ AR kRk - \f'k(T)qk> (3.201)
where:
mkf' J1kJ are the mass fraction and chemical potential ofthefth
storing component" inside" the kth species,
K k is the specific absorption by the kth species [L 2 M - 1 ],
J is the mean light intensity in the activating interval of
wavelength [energy L - 2 T - 1],
Ap , k is the energy demand for primary production [energy
M- 1 ],
AR.k is the increase in tk caused by respiration [energy
M- 1 ],
qk is the energy dissipated or used for reproduction
[energy M -1 T- 1 ],
<D k( T, 1), \f' k( T) are the dimensionless functions describing the temper-
ature dependence of biological processes in which
enzymatically catalyzed reactions play an important
role (cf. Figure 3.7).
103
(b) The internal storage of energy has to be taken into account in the
generalized Gibbs relation by the addition of a term to the right-hand
side of (3.161):
kf
- f" 7' crt .
Bk " Jikf dm
(c) The entropy flux connected with the" absorption" of solar radiation
by the primary producers counteracts the local entropy production a.
Hence, the total production per unit volume is
a tot = W grad(~) + ... + ~ ~ PkPkA/',k

1 [II
+ ... - - L Pk Kk<l\(T, I). (3.202)
To k
The terms omitted from (3.202) can be found in (3.173). To is the tempera-
ture of the light source, for instance the radiation temperature of the
solar photosphere. A p • k denotes the affinity of primary production:
[5]
Ap,k = L bjkJij - 13k + Ap,k' (3.203)
j
(d) The specific entropy s and specific free enthalpy ge per unit mass of an
aquatic ecosystem can be determined approximately by applying the
theory of dilute solutions (cf. Planck, 1913; Stumm and Morgan,
1970; etc.) not only to the dissolved inorganic and organic chemical
components and dead particulate organic materials in the water body,
but also to the biota (Mauersberger, 1980a, 1981a). This is only a very
rough approximation for higher trophic levels. Let us denote by .if0
and X o = no/n : : : : 1 the molar mass and molar fraction of pure water.
The mean molar mass JII of all the components of the ecosystem (water
included) is defined by
M
JII = - = xoJll o + LXjJll j + LBkJII. (3.204)
n j k
sand ge are found to be

o
S = soNo +L Nj(S/P, T) - R In Xj)
j JII j
o
+ L Bk[Sk(P, T, mkj, ... , mkf.) - R In(JII Bk)J
k JII k jfk
(3.205)
and
ge = goN o + L NjJij(p, T, N) +L Bk f3k(P, T, Bk), (3.206)

j k
104
where x j are the molar fractions of the chemical components and No,
N j , B k are mass fractions, while R o denotes the universal gas constant.
Furthermore,
(3.207)
(3.208)
(3.209)
(3.210)
From (3.205) it follows that the entropy increases with the number of
species and with their specific entropies Sk' From (3.203), (3.207), and
(3.208) it follows that
(3.211)
where
[5]
AI~)~ = I Ojkg/P, T) - gk(P, T, mk1 ,···, mkJ .) + XP• k' (3.212)
j
The bar denotes molar quantities:
(3.213)
gj = ultjg j ,
From AI'.k ~ 0 it follows that
B.k <
-
It (lSI
B km • x = ~H.
n _) exp
X bjk
J
(A~
lOl
R T .
) (3.214)
./I't J 0
The biomass of every species in a given region is bounded by an upper

limit Blt a" which depends on water temperature and pressure as well
as on the concentrations of the dissolved chemicals that take part in
primary production.
The Rate of Primary Production

From the local entropy production (3.202), by the rules of thermodynamics
of irreversible processes we conclude that the rate of primary production, P k ,
105
depends on the affinity Ap • k , the light intensity I, the temperature T, and other
quantities in the following manner:
A • ' et>k(T, 1),... ) .

Pk = Pk(T
p k
(3.215)
Only for small values of A p . k and et>k(T, 1) is this relation linear (the Onsager
relation). Furthermore, we know that P k vanishes outside of the T and I
intervals. In the nonlinear region of the theory we use an ansatz equation
determined by an optimization principle (Mauersberger, 1982), from which the
relations between rates and affinities of biological processes like primary
production, respiration, grazing, etc. are inferred to be of exponential character:
Pk = p:;"a{ 1 - ex p ( -Cket>k(T, 1) : : ; ) l (3.216)
where Ck is a coefficient. Introducing (3.211) and (3.214) into (3.216), we obtain

the following expression for the primary production rate:
P. = p rnax [1 _ (~)Ck<l>klT.I)]. (3.217)

k k B:;"ax
In a somewhat simplified version (3.217) can be written as
Pk ( Bk [02]b O )Ck<l>klT.11
p:;"ax = 1- C . [N0 ]b [P0 ]b 2 [ •• • ]b J • (3.218)
P k 3 ' 4
The coefficient C".k depends slightly on the temperature; [N0 3 ] denotes the
concentration of nitrate in water. By using (3.217) or (3.218) we can quantify
the influences of light intensity, temperature, and pressure on Pk , and also the
effects of the biomass and of the concentrations of nutrients, oxygen, CO 2 ,
and so on. It should be stressed that these relations are inferred from the
entropy balance equation. The type of the function et>k(T, 1) is shown by Figures
3.7 and 3.8, while the dependences of Pk on the light intensity, temperature,
T~ T r~
FIGURE 3.7 The function <lJk(T, I) describes the temperature and light dependence of
biological processes in which enzymatically catalyzed reactions play an important role.
The function vanishes outside the interval from T k to T~. It is possible that this function
has two peaks.
106
....
1'1< (TJ
FIGU RE 3.8 The type of dependence on light intensity I of the function <I\(T, J). While
the lower boundary l~ is approximately constant, the upper limit IZ depends on temperature
T.
biomass, and concentrations of inorganic chemicals in the water body are shown
qualitatively by Figures 3.9-3.14.
Growth Laws
Introducing (3.217) into (3.84), we can deduce the growth law for phytoplankton
or the law governing gross primary production:
d:k = BkPk_... = praxB{1 _ (;~xr<l>k(T.l)] _.... (3.219)
Investigating respiration from the thermodynamic point of view, we find

that the rate R k is
(3.220)
(3.221 )
FIGURE 3.9 The rate of gross primary production P, as a function of the light intensity I
for different values of temperature. The arrow indicates the change of this function with
increasing temperature.
107
x
'E""
"
0...
.......
0...""
FIGU RE 3.10 Pk as a function of temperature at constant values of nutrient concentra-

tions and biomass. The arrow indicates the influence of increasing light intensity on this
function.
ct:""
I
et
ci°f---...J-I'.-=------+-L---t---..
FIGURE 3.11 Temperature dependence of gross primary production rate P k , respira-

tion rate R k , and net primary production rate Pk - Rk for constant light intensity, nutrient
concentrations, and biomass.
x
'"
E
0...""
'-
0...""
--
FIGU RE 3.12 The rate of primary production, Pk , as a function of the mass fraction Bk
of the species k. In reality, B;:'ax is very large. Arrow 1 indicates the change with decreasing
nutrient and increasing oxygen concentrations in the water. Arrow 2 shows the change with
increasing temperature and/or light intensity available for photosynthesis.
108
Xj
FIGURE 3.13 The rate of primary production as a function of the molar fraction Xj
of one of the inorganic nutrients. The arrow indicates the change of this function with
increasing biomass fraction Bk of the primary producer species k.
FIGURE 3.14 Dependence of P k on the oxygen concentration [02] in water. Positive

curvature (broken line) is possible for small values of I and T.
The very small quantity Erin is determined by characteristics of the kth species
and by environmental factors. Thus, the result is that the net production of the
phytoplankton species k follows a "logistic" growth equation:
The coefficients C l' C 2' C 3' not written explicitly here, depend on T, I, and Xj.
Furthermore, the thermodynamic analysis of grazing shows that not only
phytoplankton species but also zooplankton species obey logistic growth laws.
A simplified version of these logistic growth equations has been discussed by
Mauersberger (1980b, 1981b). The type of this growth law is shown by Figure
109
r:r::.'"
Light Darkness
t
FIGURE 3.15 Characteristic growth of a phytoplankton population: mass fraction Bk
as a function of time t. Broken line: very high initial value BkCO).
3.15. It agrees with the laws proposed and used by von Bertalanffy (1941),
Straskraba (1978), and many others (cf. Majkowski and Uchmanski, 1980).
Conclusion
The use of the entropy principle in ecosystems research is not only necessary
(because the second law of thermodynamics cannot be ignored) but also of
great advantage, supporting a powerful method for the analysis and synthesis
of complex ecosystems, as has been demonstrated in this section.
3.7. PROBLEMS AND LIMITS OF DETERMINISTIC WATER

QUALITY MODELING
Aquatic ecosystems are open, nonlinear systems far from thermostatic equilib-
rium, exchanging energy and matter with the surroundings and maintaining
their more or less stationary state by fluxes of energy and matter through the
systems. That is, they are able to use part of the energy and/or matter exchanged
with the outside world to establish a macroscopic internal structure that is
characterized by dissipative processes. There is a definite relation between the
structure and the function of the ecosystem. These dissipative structures of a
"self-organizing system" are not necessarily stable. The succession of structures
and instabilities forms the anthropogenically influenced history of streams,
lakes, reservoirs, and other water bodies.
The formation and maintenance of self-organizing systems are the result of
nonequilibrium constraints, of appropriate nonlinear couplings, and of the
competition between the entities constituting the ecosystem. This competition
ItO
becomes significant whenever the resources necessary for the synthesis, growth,
and/or survival of biological components are limited or become scarce. The
result may be the elimination of some of the entities or a "dynamic equilibrium"
of widely different, coexisting species.
Mathematically, the interactions between the dissipative system and the
external world are specified either by boundary conditions on the separating
surface or through constraints intervening explicitly within the differential
equations. These interactions deeply influence the entropy-producing and
entropy-reducing processes inside the water body and across its boundary,
which are coupled to the structure, state, and further development of the aquatic
ecosystem.
The nonlinear basic equations of water quality modeling (including boundary
and initial conditions) may have more than one solution (bifurcation): more than
one structure or state of the ecosystem may be possible. This fact should be
taken into account in water quality modeling.
If the nonlinear basic equations describing the aquatic ecosystem allow for
more than one solution without justifying preference for anyone of them,
internal fluctuations of components of the ecosystem (generated by the system
itself or excited by the external world) have an important influence on the transi-
tion of the system to a new type of ordered configuration. This transition often
occurs near a bifurcation point. Since fluctuations of physical, chemical, and/or
biological parameters and/or components of aquatic ecosystems (e.g. day-night
variations, seiches) are typical phenomena in water bodies, the in vestigation of
fluctuations is a main task of water quality forecasting.
The responses of an ecosystem to radically changing conditions must be
simulated by nonlinear relationships. The solution of the nonlinear initial-
boundary-value problem representing the water quality model is not necessarily
unique; stationary states are not necessarily stable. The multiplicity of solutions
and the possibility of changing to another stationary state correspond to a
gradual acquisition of autonomy from the environment. The existence of more
than one solution to the model equations and the stability or instability of
stationary states can depend upon:
the value of one or more parameters,

the numbers and types of variables/components taken into account, and
the types of connections between the components of the system.
If the stationary solution of the complex set of model equations is stable, this
situation may change, for instance, if only one further component is taken into
account. This corresponds to the change in behavior of an ecosystem caused by
one new species, e.g. a virus.
The more complex the system of equations, the higher the degree of the
characteristic equation determining the stability of the system and the greater
111
the probability of an instability of a stationary state (May, 1973). This indicates a
limit of the application of multicomponent deterministic water quality models.
Of course, besides this "theoretical limit," which calls for the introduction of
"new" variables allowing for the description of the ecosystem by a limited
number of equations, there are practical demands for choosing the appropriate
model complexity, e.g. to balance the modeling efforts with the data base
available and/or to ensure that the model can be sustained economically for a
sufficiently long time.
The scientific background, as well as other material, has been covered in
works by La Salle and Lefshetz (1961), Morowitz (1970), Glansdorff and
Prigogine (1971), Lyapunov (1972), Zotin (1972), Beserskiy and Popov (1975),
Bautin and Leontovich (1976), Ebeling (1976), Rubin (1976), J0rgensen and
Mejer (1977), Kogan (1977), Nicolis and Prigogine (1977), Dubois (1979), and
Mejer and J0rgensen (1979).
REFERENCES
Alder, B., S. Fernbach. and M. Rotenberg (eds.) (1964). Fundamental Methods in Hydrodynamics
(New York, NY: Academic Press) (Russian translation 1967, Moscow: Mir).
Antomonov, J. G. (1977). Modeling of Biological Systems. (Kiev: Naukova Dumka) (in Russian).
Bagotskiy. S. V., and A. D. Bazykin (1975). Mathematical Ecology. A Bibliographic Index of Soviet
Publications 1953-1974 (Moscow: Scientific Computing Centre, Academy of Sciences. USSR)
(in Russian).
Bautin, N. N., and E. A. Leontovich (\976). Methods and Procedures ofthe Qualitatir'e Inl'estigation
of Dynamic Systems in Two Dimensions (Moscow: Nauka) (in Russian).
von Bertalanffy, L. (1941). Stoffwechseltypen und Wachstumstypen. Bioloqisches Zentralblatt
61: 510-532.
Beserskiy, V. A., and E. P. Popov (1975). Theory ofSystems ofAutomatic Control3rd edn. (Moscow:
Nauka) (in Russian).
Biswas, A. K. (ed.) (1972). Modelling of Water Resources Systems (Montreal: Harvest House).
Ciaccio, L. C. (1971-73). Water and Water Pollution Handbook (New York, NY: Dekker).
Collatz, L. (1959). The Numerical Treatment of Differential Equations, 3rd edn. (Berlin: Springer).
Daubner, 1. (1972). Mikrobiologie des Wassers (Berlin: Akademie-Verlag).
Dubois, D. M. (1979). Catastrophe theory applied to water quality regulation of rivers. State of the
Art in Ecological Modelling vol. 7, ed. S. E. Jl'lrgensen (Oxford: Pergamon), pp. 751-758.
Ebeling, W. (1976). Strukturbildung bei Irreversibelen Prozessen (Leipzig: Teubner).
Eckart, C. (1960). Hydrodynamics of Oceans and Atmospheres (Oxford: Pergamon).
Glansdorff, P., and 1. Prigogine (1971). Thermodynamic Theory of Structure, Stability and Fluctua-
tions (London: Wiley).
Goldberg, E. D., I. N. McCave, J. J. O'Brien, and J. H. Steele (eds.) (1977). The Sea vol. 6 Marine
modeling (New York. NY: Wiley).
Golterman, H. L. (1975). Physiological Limnology (Amsterdam: Elsevier).
Hinze. J. O. (1959). Turhulence (New York, NY: McGraw-Hill) (Russian translation 1963, Moscow:
Fizmatgiz).
Jenne, E. A.. et al. (1979). Chemical modeling in aqueous systems. Speciation, Sorption, Solubility,
and Kinetics. Symposium Series of American Chemical Society.
112
Jorgensen, S. E. (ed.) (l979a). Hydrophysical and ecological models of shallow lakes and reservoirs.
Summary Report of an IIASA Workshop, April 11-14, 1978. Collaborative Proceedings CP-78-14
(Laxenburg, Austria: International Institute for Applied Systems Analysis).
Jorgensen, S. E. (ed.) (l979b). Handbook o/Environmental Data and Ecological Parameters (Copen-
hagen: International Society for Environmental Modelling).
Jorgensen, S. E., and D. R. F. Harleman (eds.) (1978). Hydrophysical and Ecological modelling of
deep lakes and reservoirs. Summary Report of a IIASA Workshop, December 12-15, 1977.
Collaborative Proceedings CP-78-7 (Laxenburg, Austria: International Institute for Applied
Systems Analysis).
Jorgensen, S. E., and H. Mejer (1979). A holistic approach to ecological modelling. Ecological
Modelling 7: 169-189.
Kamenkovich, V. M., and A. S. Monin (eds.) (1978). Oceanology and Phvsics of the Ocean vol. I
Hydrodynamics of the ocean; vol. 2 Hydrodynamics (Moscow: Nauka) (in Russian).
Kaye, G. W., and T. H. Laby (1959). Tables o/Physical and Chemical Constants, 12th edn. (London:
Longman) (Russian translation 1962, Moscow: Fizmatgiz).
Kinsman, B. (1965). Wind Waves (Englewood Cliffs, NJ: Prentice-Hall).
Klotz, I. M. (1964). Chemical Thermodynamics: Basic Theory and Methods, revised edn. (New York,
NY: Benjamin).
Kogan, A. B. (1977). Biolo.qical Cybernetics (Moscow: Vishaya Shkola) (in Russian).
Krauss, W. (1966). Interne Wellen. Methoden und Ergebnisse der Theoretischen Ozeanographie vol. 2
(Berlin: Gebriider Borntraeger) (Russian translation 1968, Leningrad: Gidrometizdat).
Lamb, H. (1932). Hydrodynamics (London: Cambridge University Press) (1945 New York, NY:
Dover) (Russian translation 1947, Moscow: Gostekhizdat).
Landau, L., and E. Lifshitz (1960). Electrodynamics oj" Continuous Media (London: Pergamon)
(Russian version 1959, Moscow: Fizmatgiz).
La Salle, J., and S. Lefshetz (1961). Stability by Lyapunov's Direct Method (New York, NY:
Academic Press) (Russian translation 1964, Moscow: Mir).
Latimer, W. M. (1952). Oxidation Potentials, 2nd edn. (Englewood Cliffs, NJ: Prentice-Hall).
Lehninger, A. L. (1972). Biochemistry (New York, NY: Worth) (Russian translation 1976, Moscow:
Mir).
Lewis, G. N., and M. Randall (1961). Thermodynamics, 2nd edn. (New York, NY: McGraw-Hili).
Lin, C. C. (1955). The Theory of Hydrodynamic Stability (London: Cambridge University Press)
(Russian translation 1958, Moscow: Inostr. LiL).
Lions, J. L. (1969). Quelques Mtithodes de Resolution des Problemes aux Limites non Lineaires
(Paris: Dunod; Gauthier-Villars) (Russian translation 1972, Moscow: Mir).
Lyapunov, A. A. (1972). Cybernetic Problems in Biology (Moscow: Nauka) (in Russian).
Majkowski, J., and J. Uchmanski (1980). Theoretical foundations of individual growth equations in
animals. Polish Ecological Studies 6: 7-31.
Marchuk, G. I. (1975). Methods of Numerical Mathematics (New York, NY: Springer) (Russian
version 1973, Novosibirsk: Nauka).
Martell, A. E., and R. M. Smith (1974-77). Critical Stability Constants vol. I Amino acids (1974);
vol. 2 Amines (1975); vol. 3 Other organic ligands (1977); vol. 4 Inorganic complexes (1976)
(New York, NY: Plenum).
Mauersberger, P. (1978). On the theoretical basis of modeling the quality of surface and sub-
surface waters. Modeling the Water Quality of the Hydrological Cycle. Proceedings of Baden
Symposium, September. International Association of Hydrological Sciences Publication 125, pp.
14-23.
Mauersberger, P. (1979). On the role of entropy in water quality modelling. Ecological Modelling
7: 191-199.
Mauersberger, P. (I 980a). Ergebnisse der Thermodynamik aquatischer Okosystems. Gerlands
Beitrage zur Geophysik 89: 339-341.
Mauersberger, P. (l980b). Growth equations of phyto- and zooplankton deduced from thermo-
113
dynamic principles. Proceedings ol"lnternational Symposium on Simulation ol"Systems in Biology
and Medicine: Si-Sy '80, Prague, 17-20 NOl'ember, vol. 3, pp. 55-65.
Mauersberger, P. (l98Ia). Entropie und freie Enthalpie im aquatischen Okosystem. Acta Hydro-
physica 26: 67-90.
Mauersberger. P. (l98Ib). Laws for population and community dynamics in aquatic ecosystems
based on the balances of mass. energy, and entropy. Proceedings of International Workshop on
Ecosystem Dynamics in Freshwater Wetlands and Shallow Water Bodies, Minsk-Tskhaltoubo,
USSR, 12-26. July.
Mauersberger, P. (1982). Zur Bestimmung der nichtlinearen Beziehungen zwischen Raten und
Affinitiiten bei Produktions- und Abbauprozessen im aquatischen Okosystem. Acta Hydro-
physica 27: 125-130.
May, R. (1973). Model Ecosystems (Princeton, NJ: Princeton University Press).
Meixner, J., and H. R. Reik (1959). Thermodynamik der irreversibelen Prozesse. Handbuch der
Physik vol. 3/2, ed. S. Fliigge (Berlin: Springer), pp. 413-523.
Mejer, H., and S. E. J0rgensen (1979). Energy and ecological buffer capacity. State of the Art in
Ecological Modelling vol. 7, ed. S. E. J0rgensen (Oxford: Pergamon), pp. 829-846.
Menshutkin, V. V. (1971). Mathematical Modeling of Populations and Faunal Communities in
Aquatic Ecosystems (Leningrad: Nauka) (in Russian).
Milne-Thomson, L. M. (1960). Theoretical Hydrodynamics, 4th edn. (New York, NY: Macmillan)
(Russian translation 1964. Moscow: Mir).
Mitchel, A. R., and R. Wait (1977). The Finite Element Method in Partial Differential Equations
(New York. NY: Wiley).
Mitropolskiy,1. A. (ed.) (1977). Mathematical Methods in Biology (Kiev: Naukova Dumka) (in
Russian).
Monin, A. S., and A. M. Jaglom (1965-67). Statistical Hydrodynamicsvols. 1, 2 (Moscow: Nauka) (in
Russian).
Morowitz, H. J. (1970). Entropyfor Biologists (New York, NY: Academic Press).
Naumov, G. B., B. N. Ryzhenko, and I. L. Khodakovsky (1974). Handbook of Thermodynamic Dara.
US Geological Survey WRD-74-001. National Technical Information Service PB 226 722/AS.
Nicolis, G .. and I. Prigogine (1977). Self-Organization in Nonequilibrium Systems (London: Wiley).
Phillips, M. (1962). Classical electrodynamics. Handbuch der Physik vol. 4, ed. S. Fliigge (Berlin:
Springer), pp. 1-108.
Pielou, E. C. (1975). Ecological Diversity (New York, NY: Wiley).
Pinder, G. F., and W. G. Gray (1977). Finite Element Simulation in Surface and Subsurface Hydrology
Planck, M. (1913). Vorlesun.qen uber Thermodynamik, 4th edn. (Berlin: Veit).
Proud man. J. (1953). Dynamical Oceanography (London: Academic Press) (Russian translation
1957, Moscow: Inostr. Lit.).
Rheinheimer. G. (1971). Mikrobiologie der Gewiisser (lena: VEB Gustav Fischer).
Richtmyer. E. D., and K. W. Morton (1967). Difference Methodsfor Initial Value Problems, 2nd
edn. (New York, NY: Interscience).
Romanenko, V. I., and S. I. Kuznecov (1974). Ecology 01" the Microorganisms of Freshwater Bodies
(Leningrad: Nauka) (in Russian).
Romanovskiy, 1. M., N. V. Stepanova, and D. S. Chernavskiy (1975). Mathematical Modeling
in Biophysics (Moscow: Nauka) (in Russian).
Rossini. F. D., D. D. Wagman. W. H. Evans, S. Levine, and I. Jaffe (1952). Selected values of
chemical thermodynamic properties. National Bureau of Srandards, us Department of Com-
merce, Washington, DC Circular 500.
Rubin, A. B. (1976). Thermodynamics of Biological Processes (Moscow: University Press) (in
Russian).
Sal;lnki. J .. and J. E. Ponyi (eds.) (1975). Limnology of Shallow Waters: Proceedings of Symposium.
Tihany, Hungary, 1973 (Budapest: Akademiai Kiad6).
114
Serrin, J. (1959). Mathematical principles of classical fluid flow. Handbuch der Physik vol. 8/1,
ed. S. Fliigge (Berlin: Springer), pp. 125-263.
Stoker, J. J. (1957). Water Wal'es(New York, NY: Interscience)(Russian translation 1959, Moscow:
Inostr. LiL).
Strang, G., and G. J. Fix (1973). An Analysis of the Finite Element Method (Englewood Cliffs, NJ:
Prentice-Hall) (Russian translation 1977, Moscow: Mir).
Straskraba, M. (1977). Natural control mechanisms in models of aquatic ecosystems. Proceedings
o(/FAC Symposium on Control Mechanisms in Bio- and Ecos.l'stems. Leipzi,q, September vol. I.
pp.74-108.
Straskraba, M. (1978). Theoretical considerations on eutrophication. Verhandlung /nternationaler
Vereini,qung Limnoloqie 20: 2714-2720.
Stratton, J. A. (1941). Electromagnetic Theory (New York, NY: McGraw-Hili).
Stumm, W., and J. J. Morgan (1970). Aquatic Chemistry (New York. NY: Wiley).
Truesdell, C, and R. Toupin (1960). The classical field theories. Handbuch der Physik vol. 3/1, ed.
S. Fliigge (Berlin: Springer), p. 692.
Uhlmann. D. (1975). Hydrobiologie (Jena: VEB Gustav Fischer).
Vasiliev, O. F., V. I. Kvon, Ju. M. Litkin, and I. L. Rozovskiy (1975). Stratified flows. Hydro-
mechanika 8: 74-131 (in Russian).
Vinberg, G. G., and S. I. Anisimov (1966). Investigation of a mathematical model for an aquatic
ecosystem. Publications olAIl-Union Research Institute for Fishery Economics and Oceanography,
MoscoII' 67:49-75.
Vollenweider, R. A. (1969). Moglichkeiten und Grenzen elementarer Modelle der Stoffbilanz
von Seen. Archil"fur Hydrobiologie 66: 1-36.
Volterra, V. (1932). Le~'ons sur la Theorie Mathematique de la Lulie pour la Vie. (Paris: Gauthier-
Villars) (Russian translation 1976, Moscow: Nauka).
Wagman, D. D., W. H. Evans, V. B. Parker, I. Halow, S. M. Bailey, and R. H. Schumm (1968-69).
Selected values of chemical thermodynamic properties. National Bureau olStandards Technical
Note 270-3 (1968), 270-4 (1969).
von Weizsiicker, C F. (1972). Evolution und Entropiewachstum. Nova Acta Leopoldina, Neue Folge
206: 515-534.
Wetzel, R. G. (1975). Limnology (Philadelphia, PA: Saunders).
van Zeggeren, F., and S. H. Storey (1970). The Computation of Chemical Equilibria (London:
Cambridge University Press).
Zienkiewicz, O. C, and Y. K. Cheung (1967). The Finite Element Method in Structural and Con-
tinuum Mechanics (London: McGraw-Hili).
Zotin, A. I. (1972). Thermodynamic Aspects of Developmental Biology (Basel: Karger).
CHAPTER 3: NOTATION
a jk' bjk' Cjk' )

dkm , ejk,hj, fractions of ecosystem components produced or consumed
gkt, hjm , Pjkl' by internal biochemical or biological processes
Pjk' Sjkl> Ski'
Bk mass fraction of biota
BC k biotic component k of aquatic ecosystem
D,D* scalar diffusion coefficients
DIM(DIM) dissolved inorganic matter (component j)
115
DOM(DOM j ) dissolved organic matter (component j)
Fk rate of uptake of DOM by kth decomposer
ge specific free enthalpy
G sources, sinks
Gp Gibbs potential
Hj rate of hydrolysis of jth type of particulate organic matter
(POM)
j electric current density
In generalized flows
Nj mass fraction of chemicals
Ps total entropy function
Pv pressure function
POM(POM) particulate organic matter (component j)
Qk production and consumption term
Q, reaction quotient
entropy density
total entropy of system
entropy flow density
preference coefficient (for biocomponent / as food for k)
rate of uptake of DIM j by kth primary producer
generalized forces
rates of biochemical reactions
specific volume
potentials of biocomponents
activity coefficients
coefficient of turbulent mixing (eddy diffusivity); internal
energy per unit mass of species k
thermal conductivity
energy supply or demand
JII j !1j, molar chemical potential of jth chemical constituent
entropy production rate
x isothermal compressibility.
Modeling the Ecological Processes
4 S. E. J(j)rgensen
4.1. BACKGROUND
4.1.1. The Ecosystem and Chemical-Biological Processes

The basic chemical, biochemical, and biological processes mentioned in section
3.5 have been quantitatively described by application of a great variety of
equations, which will be surveyed and represented in this chapter in forms that
can easily be translated into computer language (e.g. CSMP).
An ecosystem is defined by several biotic and abiotic variables, which will
change in space and time owing to the above-mentioned processes. This is illus-
trated in Figure 4.1, which shows the flow of energy in various forms through a
succession of trophic levels within an aquatic environment characterized by
quality parameters such as the concentrations of dissolved oxygen, carbon
dioxide, phosphorus, and nitrogen, and amounts of suspended matter, detritus,
and bottom sediment. The mass and energy transformations are regulated by
processes such as growth, respiration, mortality, and decomposition; these in
turn are governed by environmental quality parameters such as temperature,
toxicity, and nutrient concentrations. The system is highly coupled, as energy
and mass balances for individual constituents are invariably linked to several
others.
Ecological models, like other models, represent a compromise between the
reality of the system and the simplicity of the mathematical description. Fitted
regression equations might compare favorably in predictive capacity with
causally based functions, but they provide little or no insight into the underlying
processes and they are most often specific and cannot be used in several case
studies without modification. Causal models require biological, chemical, and
physical details in construction to enhance reliability over a broad environ-
116
117
Aquatic plants
Light Sediment Carbon,
Oxygen and
nitrogen.
carbon dioxide { phosphorus, etc.
1 (
Thermocline
Rough fish
Hypolimnion
~
Epilimnion
Thermocline
Hypolimnion
Benthos
FIGURE 4.1 Ecological relationships in a lake environment (redrawn from GrIob, 1975).
A, advection; B, benthic animals; Bac, bacteria; D, diffusion; Det, detritus; F I' pelagic
fish; F 2 , rough fish; N" N 2 , soluble nitrogen (NHt and NO;); DOl' D0 2 , dissolved
oxygen; OS, soluble carbonate; PI' aquatic plants; P 2 , algae; Tox, toxic substances;
Z, zooplankton.
118
DENIT
SN
FIGURE 4.2 Lake model: only nitrogen is considered. I, Uptake of nitrogen by phyto-
plankton; 2, growth of phytoplankton (photosynthesis); 3, nitrogen fixation; 4, grazing;
5, fecal (by grazing): 6, fish preying on zooplankton; 7, fecal (by process); 8, mortality and
autolysis of phytoplankton; 9, mineralization of detrital nitrogen; 10, sinking of phyto-
plankton; II, sinking of detritus; 12, excretion of ammonia by phytoplankton and fish:
13, release of ammonia from sediment; 14, nitrification; 15 and 16, inflow and outflow of
ammonia and nitrate; 17, denitrification; 18, inflow and outflow of phytoplankton. DENIT,
denitrification; NA, concentration of nitrogen in phytoplankton; NFIX, nitrogen fixation;
PHOTO, photosynthetic rate; SN, nitrogen in upper layer of sediment.
mental spectrum. The most frequently applied causal dynamic models have a
number of constituents: water, bacteria, detritus, phytoplankton, zooplankton,
fish, sediment, etc. The relation to the underlying processes is illustrated by
Figure 4.2, where only nitrogen is considered, for simplicity. Similar diagrams
could be added to illustrate the constituents and processes for other biologically
important elements such as carbon, phosphorus, oxygen, hydrogen, sulfur,
and silicon. The conceptual diagram, Figure 4.2, contains seven constituents,
which are the state variables (Chapter 2) in the mathematical model, and 18
processes that describe the flow of nitrogen from one constituent to another.
The changes of the state variables are expressed by differential equations
based upon the mass conservation principle (section 3.2). For example, the
differential equation for the ammonium ion concentration (NHt ) is:
dNH+
-d 4 = process 9 + process 13 + process 12 + process 15
t -
- process 14 - process 1. (4.1)

The models of the individual processes, such as processes 9, 13, 12, 15, 14, and
1, and their related parameters, which can be termed submodels, will be surveyed
in this chapter. The submodels are components of water quality models, which
119
include ecological processes, such as the river model QUAL II (section 6.6) and
the lake model CLEANER (section 9.2.1). The coupling of the submodels will
be treated in relation to river and lake models in Chapters 6 and 9.
4.1.2. Brief Review of Development

Among the earliest ecological submodels were those that addressed the problem
of predator-prey relationships, e.g. the set of equations referred to as the
Lotka-Volterra predator-prey model (Lotka, 1924). Another field, which was
approached at an early stage by the use of models, was the interaction between
organisms and their feed or substrate. The so-called Monod kinetic has been
widely used for many years:
s (4.2)
)1 = )1max K
s
+ S'
where)1 is the growth rate, )1max the maximum rate, S the substrate concentration,
and K s the half-saturation constant. The equation is represented by Figure 4.3,
which demonstrates that K s is equal to S when )1 = t)1max' This equation was
derived from the Michaelis-Menten formulation of biochemical enzyme
kinetics.
The trend in ecological modeling over the past several years has been to
include more biological realism, so that more and more detailed submodels
attempt to take into consideration feedback mechanisms and couplings to
other constituents and to environmental factors. The experience gained by
working with river and lake ecological models has provoked new biological
experiments devised to aid formulation of model construction by representing
------------~-
Jlmax
...
Q)
~
~
~ O.5/ 1max
o...
<.9
Substrate concentration S
FIGURE 4.3 Michaelis-Menten kinetic expression for microbial substrate utilization
and growth.
120
environmental responses for specific processes. As a result of this effort a wide
range of mathematical formulations of the ecological processes in bodies of
water is available today.
4.2. MODELING ECOLOGICAL PROCESSES IN AQUATIC

ECOSYSTEMS
4.2.1. General Considerations

The plentiful sources of error in both our structural and measured knowledge
of an ecological process make it hardly possible to select the only correct
formulation for a specific process in an aquatic ecosystem. The reason for this
is the difficulty in obtaining harmonized, regular, high-frequency spatial and
temporal field data with reasonable accuracy. Such difficulties can partly be
overcome by examination of the actual processes in the laboratory, but this is
only, as indicated, partly a solution, since:
(i) the processes in the ecosystem can never be isolated from their environ-
ments, and
(ii) it is impossible to take account of all natural coupling mechanisms.
Consequently, each process has several alternative mathematical formulations

that, at this stage of ecological modeling, are equally valid to a certain extent.
As for the total model, the submodels should be selected in accordance with
the problem, the data available, and the case study. The processes of interest
for water quality modeling and their mathematical formulations will be surveyed
in the following sections, and the realism and causality of the equations will be
discussed. The notation used in Tables 4.1-4.12 is explained in Table 4.13.
4.2.2. Photosynthesis
The generally used mathematical formulations of photosynthesis are surveyed
in Table 4.1. Only eqns. 5, 8, and 14 include phosphorus in addition to nitrogen
and carbon as the controlling nutrients, but the other equations can easily be
expanded to cover all three nutrients and even silicon, which must be included
when diatoms are of importance for the eutrophication process.
Equations 10 and 12-14 consider a two-step description of the photo-
synthesis: uptake of nutrient, and growth controlled by the intracellular con-
centration of nutrient. If only the first process is indicated the growth is, or can
be, formulated by use of the other growth equations. This two-step formulation
121
TABLE 4.1 Photosynthesis.
(1) Broqvist (1971)
/ PS NS)
PHOTO = MY(T)· PHYT· min ( - -,-
/(0) 'PS o NS o
(2) Chen (1970), Chen and Orlob (1975)

/ NS PS
PHOTO = MY(T). ~-. . . PHYT
/ + IK KN + NS KP + PS
MY(T) = MYMAX· f(T)
(3) Parker (1972)

PHOTO = MY(T)' / . PS· NS· PHYT
(4) Anderson (1973)

PHOTO = MYMAX . PHYT . (PS + NS)
(5) Dahl-Madsen and Strange Nielsen (1974)

PS NS CS
PHOTO = MY(T). f(I) . . .---
KP + PS KN + NS KC + CS
MY(T) = MYMAX· g(T)
(6) Jansson (1972)

PHOTO = MY· PHYT· PS· /T
(7) Lassen and Nielsen (1972)

PHOTO = MY· PHYT . f(PS) . FD . f(I)
(8) Patten et al. (1975)
PHOTO = MY· PHYT . f(I)(f(NS) + f(PS) + f(CS»/3
(9) Larsen et al. (1974)

PHOTO = MY· PHYT . min(f(I), f(PS), f(NS))
F
R(t)
lU)=~
. IK + R(t)
There are corresponding equations for f(PS) and f(NS).
(10) Bierman et al. (1974)
UP = UPMAX ( I - ---- 1)
I + PKI . PA I + PKI . PS
(11) Gargas (1976)

MY = MYMAX . f(I)f(PS)f(NS)f(T) . FD . F AC
(continued over)
122
TABLE 4.1 (continued)
(12) Cloern (1978)
PHOTO = MY(T)._I_ eXP(1 - _1_){(PA){(NA)

I op,( T) lop,( T) . .
MY(T) = 0.02 exp(0.17 T)
I op,( T) = 0.06 exp(0.22 T)
PAP
.f(PA)- -PA-P---PA-M-I-N
NAP
f(NA) = NAP _ NAMIN
PS
UP = UPMAX . - - -
KP + PS
NS
UN = UNMAX . - - -
KN + NS
(13) Nyholm (1978)
PHOTO = MY· f(l)f(NA, PA)
2
f(NA, PA) = - - - - - -
l/f(PA) + l!f(NA)
(sec also Bloomfield et aI., 1974)
KPA + PAMAX - PAMIN PA/PHYT - PAMIN

f(PA) = PAMAX _ PAMIN KPA + PA/PHYT - PAMIN
NA/PHYT - NAMIN
f(PN) = NAMAX _ NAMIN
Excess supply of nutrients:
dPS dPHYT
-=PAMAX--
dt dT
dNS dPHYT
-=NAMAX--
dt dT
UP = MY . PAMAX
UN = MY· NAMAX
Limiting conditions: UP and UN arc equal to supply.

123
(14) Jorgensen (1976)
PHOTO = MYMAX· f(T)f(PA)f(NA)f(CA)

PA - PAMIN . PHYT
f(PA) = PA
NA - NAMIN . PHYT
f(NA) = NA
CA - CAMIN . PHYT
{(CAl = - - - - . - - -
. CA
CAMAX· PHYT - CA CS
UC = f(l)· UCMAX . CAMAX. PHYT _ CAMIN . PHYT . PHYT . KC + CS
PAMAX· PHYT - PA PS
UP = UPMAX· . PHYT . - - - -
PAMAX· PHYT - PAMIN . PHYT KP + PS
NAMAX· PHYT - NA NS
UN = UNMAX· . PHYT . - - -
NAMAX· PHYT - NAMIN . PHYT KN + NS
has the following advantages:
(i) It is in accordance with algal kinetic observations (Fuhs, 1969; Droop,

1974; Nyholm, 1975; Rhee, 1978).
(ii) Some case studies show better accordance between model and observa-
tions (Serruya and Berman, 1975; Jorgensen, 1976; Park et aI., 1979).
However, the disadvantage of a model consisting of independent element

cycles is that more parameters are introduced into the model. This drawback
can be partly eliminated, as the minimum and maximum concentrations of
the elements in phytoplankton are well known. In lake studies where the ratios
of element concentrations are kept fairly constant, the stoichiometric model
might be used as a less complex alternative to description by independent
element cycles (Jorgensen and Harleman, 1978).
The growth of phytoplankton is controlled by light and by nutrients. In
Table 4.1 the light limitation is indicated by !(1) and Table 4.2 gives some
generally applied formulations for f(1). Many of the equations include the
extinction of light with increasing depth and use a Michaelis-Menten ex-
pression to describe the relationship between photosynthesis and irradiance.
However, integration over depth is of course possible for all the formulations
shown. Figures 4.4(a-i) present the relationships between photosynthesis and
irradiance, excluding the influence of the depth Z on I.
Equation 3 in Table 4.2 includes the diurnal variation and eqns. 2 and 6 the
length of day. Equations 2, 5, and 7 take the attenuation coefficients of water
124
TABLE 4.2 Light.
(I) Steele (1974)
I(O)exp( -£Z) ( I(O)exp( -£Z))

((I, Z) = exp I - - - - -
. IK IK
(2) Ahlgren (I 973a, b)
. Z) = -
j(l, 2.72 -
FD [ exp (IDA. exp(£z») - exp (IDA)]
-
£Z IK IK
FD = 0.5 + 0.26 sin(t - 82) (latitude 60°)
IDA = 1(0)/24 FD
1(0) = 350 + 300 sin(l - 82) - 75[1 + sin 2(1 - 37)](m Wh cm - 2 day - 1)
£ = 0.5 + 14 PA(I)
£ = 0.5 + 29 PA(2)
(3) Schofield and Krutchoff (1974)
~: (t -
2
1(0, t) = IDA· K(sin 6)r
=0 t<6,t>18
K = 2.6225
since
IB
J 6
1(0, t)dt = IDA
211: 11:)
REF = 0.08 + 0.02 sin ( - . Week - -
52 2
(5) Gargas (1976)

1
1 S; IK
IK
f(l)= IK < 1 < IH
max
o
{I - KLYS(I- IH) IH S 1
1 = I(O)exp[( -ex - j1PHYT - yZOO - bDET)Z]
(6) Nyholm (1978)

1(0) = IDA . CFGI/FD
IH = 0.71(0) + 3.10 5 (cal m -2 day-I)
and as (5)
125
Growth rate of phytoplankton as
A 1(0) + lK
fU) - In
- «(1 + pPHYT) V I(O)exp[( - (1 - pPHYT) V/A] + lK
(8) Thomann et al. (1974)

Growth rate as
fU) = MYMAX [ 2.718 . exp ( - 1(0)

- exp[ -«(1 + /JPHYT)Z] ) - exp (/(0»)]
--
(1 + pPHYT IOPI IOPI
(9) Shelef et at. (1969)

fU) = KI I ~ lK
fU) = maximum I ;:0: lK
(10) Shelef et al. (1972)

fU) = I - expU/lK)
(II) lkusima (1966)
PHOTO) I
fU) = KI ( I - . K=-
PHOTO max lK
(12) Tailing (1957)

l/lK
lK corresponds here to I at PHOTO = 0.7 PHOTOmax-
(13) Takahashi et al. (1973)

f(1) = KI exp(l - KI/K J + K 2 T)
(14) Vollenweider (1965)

l/lK I
fU) = . ------:-:=
[I + U/IK)2]1 12 [I + «(11)2]K/2
lK corresponds here to I at PHOTO = 0.6 PHOTO m...
(15) Steel (1973)

1/2lK
fU) = 2 I + U/2lK/
126
PHOTO max
a a
f- f-
a a
I I
a.. a..
(a) (b)
~~TOm ..
(c) I
PHOTO max
----- PHOTO max
a a
f- f-
a a
I I
a.. a..
(e) ( f)
~c
a
f-
a
I
a..
(g) / (hI
Photoinhibition
above this level
PHOTO max
a
f-
a
I
a..
( il
FIGURE 4.4 PHOTO = f(1) according to different equations in Table 4.2: (a) eqns. 9;
(b) eqns. 5; (c) eqn. 10; (d) eqn. 11; (e) eqn. 12; (f) eqn. 13; (g) eqn. 14; (h) eqn. 15; (i) eqn. 7,
when the integration over depth is omitted. A Michaelis-Menten function is applied.
127
and of phytoplankton into consideration, and (5) even the attenuation by
zooplankton and detritus. Equations 5, 7, 13, 14, and 15 also include the in-
hibition of irradiance above a certain level. However, for eutrophic lakes the
inhibition will be significant only in the upper few millimeters of the body of
water. As seen from Table 4.1, the existing mathematical models combine the
effects of light, temperature, and nutrient concentration in a variety of ways.
Some of the methods of combining the effects are described below (Park et al.,
1979).
(1) Multiplication of the effects (eqns. 5 of Table 4.1) has been used by
Oi Toro et al. (1971,1975), Chen and Orlob (1975), Lehman et al. (1975),
Youngberg (1977), and Kremer and Nixon (1978). Equations 14 of
Table 4.1 use multiplication of temperature and nutrient-limiting
effects, but the light-limiting effect only influences the uptake of carbon.
This implies that the factors are independent of each other and are of
equal importance.
(2) Some of the limitations are combined by using a mean expression and
then multiplying it by the third limiting effect, often the temperature
limitation (eqn. 8 ofTable 4.1) (Bloomfield et al., 1974; Park et aI., 1974;
Patten et al., 1975). This implies interaction between nutrients and light
and results in a partial compensation for one of the other factors, while
temperature is treated independently. Equation 8 uses an arithmetic
average, while eqns. 13 use the harmonic mean.
(3) The minimum of two or more limitations is used (eqns. 1 and 9 of
Table 4.1; Scavia and Park, 1976). This assumes that light and nutrients
cannot be limiting at the same time.
(4) A complex interaction of the limitations, implying independence of the
nutrient and light limitations, both being temperature-dependent (eqns.
12 of Table 4.1; Figure 9.11 on page 348 represents the complex
combination used in MS CLEANER (Park et aI., 1979».
4.2.3. Zooplankton Grazing
Zooplankton grazing plays an important role in lakes and reservoirs by limiting

phytoplankton blooms. Table 4.3 summarizes the zooplankton grazing equa-
tions. Equation 5 describes classic logistic growth, which is modified in eqn. 6
by multiplication with a variable that is related to the availability of food and is
dependent upon temperature and the concentrations of phytoplankton,
oxygen, and toxic substances. Other authors use a Michaelis- Menten expression.
Steele (1974) has modified this equation by using a threshold concentration
below which the grazing ceases, while Frost (1975) uses a minimum feeding
rate below the threshold concentration. This expression has the advantage of
128
TABLE 4.3 Zooplankton Grazing Rate.
(I) Dodson (1975)
dPHYT
- - = -K·PHYT·ZOO
de
(2) Steele (1974)

PHYT - KTR
GRZ = MYZ . . ZOO
KZ + PHYT
MYZ = MYZMAX . f(T)
(3) Walsh and Dugdale (1971)

GRZ = g(ZOOj(PHYT - KTR)
For example,
g(ZOO) = MYZ . ZOO
(4) O'Brien and Wroblewski (1972)

GRZ = ZOO· MYZ{I - exp[Dp(PHYT - KTR)]}
MYZ = MYZMAX· f(T)
(5) Lotka (1924)
ZOO)
GRZ = MYZ· ( I - CK . ZOO
(6) Odum (1972)
GRZ = MYZ· AV· (I - Z~KO). ZOO

AV = f(PHYT)f(OX)f(T)f(TOX)
(7) Gargas (1976)
PHYT - KTR
MYZ· ·ZOO PHYT > KTR
KZ + PHYT
GRZ =
PHYT
{ MYZ·-- PHYT ~ KTR
ZOO
MYZ = MYZMAX . f(T)
(8) Chen and Orlob (1975), Chrisholm ee al. (1975), Canale ee at. (1976)
PHYT
GRZ = MYZ . . ZOO
KZ + PHYT
129
(9) Canale et al. (1976)

KMFM . PHYT + KFLM
GRZ = MYZ . - - - - - - - - -
PHYT + KFLM
IX~ PHYT
PREF = =--=----
L
IX~ PHYT
(10) lost et al. (1973)

PHYT 2
GRZ = MYZ . . ZOO
(KZI + PHYT)(KZ2 + PHYT)
(II) Gause (1934)
MYZ . PHYT'/2 . ZOO

GRZ = -------;-;-:--
KZ + PHYT'12
(12) Dugdale (1975)
GRZ = MYZ· ZOO[l - exp( -KZ· PHYT - KTR)]
(13) Richey (1977)

GRZ = AK . (BL)2 . T
being in accordance with several observations, e.g. the grazing rate of filter
feeders.
Canale et al. (1976) distinguish between raptors (eqn. 8), selective filter
feeders (egns. 9), and nonselective filters (eqn. 1). For the first two classes a con-
stant assimilation efficiency is used, and for the latter Canale et al. use a simple
but unconfirmed formulation:
. .. ffi . . ffi . HMFL
assImIlatIon e clency = maXImum e clency' PHYT + HMFL'
where HMFL is the half-maximum-efficiency food level.

Jost et al. (1973) have developed a multiple kinetic model, which is in better
agreement with their observations than was the Monod kinetic (eqn. 10). Gause
(1934) used another model (eqn. 11), which predicts that the predator becomes
more effective in capturing prey at a low prey concentration.
A persistent difficulty of the predicted zooplankton biomass is that it does
not decrease sufficiently in the fall and winter. The model CLEANER contains
a very complex construct for predator-prey interaction (section 9.2) (Park
et al., 1974; Scavia et al., 1976), yet the model cannot predict very well the
zooplankton biomass in the fall and winter. It is probably necessary to include
starvation and the formation of resting stages, but unfortunately very few
quantitative studies have been conducted on these processes.
130
4.2.4. Respiration and Nonpredatory Mortality of Phytoplankton and
Zooplankton
Respiration and nonpredatory mortality are often described as first-order
reactions (Table 4.4, eqns. 1) but a more causal expression is shown in eqn. 2
of Table 4.4, where the respiration is dependent upon the intracellular concen-
tration of phytoplankton.
Some models do not include phytoplankton mortality as a separate factor,
but combine it with settling, since each can be approximated as a first-order
process.
The physiological state of algae is important in determining the point at
which growth ceases and death (and settling) begins. When growth conditions
become unfavorable (e.g. nutrient-deficient), algae become susceptible to
decomposition. DePinto (1979) presents a comprehensive discussion of non-
predatorial mortality.
In the model of Lehman el al. (1975) the death rate is a function of the
number of days of algal growth in suboptimal conditions. Bierman (1976) uses
a second-order algal death rate, as shown in (3) of Table 4.4. Thomann el al.
(1975) and Canale el al. (1976) use a temperature-dependent mortality rate, as
indicated in (4). In the model of Scavia and Park (1976) and in MS CLEANER
(Park el aI., 1979) mortality rate is a function of temperature as well as of
physiological state (Figure 9.11).
TABLE 4.4 Respiration and Mortality of Phytoplankton
(I) Gargas (1976)
RESP = RESPK . f(T)· PHYT

MORT = MORTK . PHYT
(2) Lehman et al. (1975)
RESP = RESPK . (~) . f(T)· PHYT

CAMAX .
(as carbon)
(3) Bierman (1976)
MORT = MORTK . PHYT 2
(4) Thomann et al. (1975), Canale et al. (1976)

MORT = MORTK· f(T)
(5) Stra~kraba (1976)

RESP = 3.1 exp(O.09T)[mg O 2 (mg chlorophyll)-' h '] 0 ~ T ~ 25°C
131
TABLE 4.5 Zooplankton Respiration, Excretion, and Mortality.
(I) Oi Toro et al. (1975)
MORT = MORTK· ZOO

RESP = RESPK· !(T)' ZOO

19 RESP = 0.0364T - 0.34181g W + 0.6182
A first-order kinetic is also often used for zooplankton mortality and respira-
tion (Table 4.5). As discussed in section 4.2.3, it is, however, sometimes necessary
to use more complex models, if a more realistic description of zooplankton
dynamics is required.
4.2.5. Sinking, Mineralization, and Excretion

Sinking of phytoplankton is a function of shape, specific gravity, size, and the
viscosity and turbulence of the surrounding water. Some algae are vacuolated
and their sinking rate is controlled differently. Several factors controlling the
sink ing rate have been suggested:
(i) physiological state: cells in the stationary phase sink two to four times
faster than cells in the exponential phase of growth (Eppley et al., 1967;
Smayda, 1970);
(ii) nutrient depletion (Smayda, 1974; Titman and Kilham, 1976);
(iii) light (Eppley et ai., 1967; Steele, 1974);
(iv) viscosity of water: is a function of temperature in fresh water (eqn. 2
of Table 4.6);
(v) turbulence (Stefan et ai., 1976; Titman and Kilham, 1976).
Most submodels formulate sinking rate as a first-order reaction (eqn. I of

Table 4.6), which is an oversimplification. Others include a reduction in
sinking rate due to turbulence, applying the Reynolds number as a site variable.
At present only the model of Scavia and Park (1976) considers the physiological
state as a controlling factor.
The rate with which detritus, including dead algal cells, decomposes is
important for understanding nutrient cycling. In most submodels the decompo-
sition is formulated by a first-order reaction scheme as shown in Table 4.7.
However, as discussed by J0rgensen (1979a), the detritus consists of a wide range
of organic compounds with different decomposition rates. This implies that a
more correct description should consider two or more fractions with different
rate coefficients. It seems that especially allochthonous nitrogen compounds
have a lower decomposition rate than other nitrogen compounds (discussed
by J0rgensen, 1979; J0rgensen and Mejer, 1979).
132
TABLE 4.6 Settling.
(I) Chen and Orlob (1975)
SEV(P)
SEV=--
SEV(O)
(2) Jorgensen (1976), Jorgensen and Harleman (1978)

T' 2
SETTL = (SEPH . PHYT + SEOT . OET). - " -
ZT:!2
(3) Oi Giarno et al. (1978)

BPHYT
SETTL = SEPH . - -
BZ
(4) Nyholm (1978)
0.20 m day-l
SEPH = min {
0.05Z m day- 1
(5) Scavia and Park (1976)

SETTL = SEPH . PHYT(I + constant· T)
TABLE 4.7 Detritus.
Jorgensen (1976)
OERV = MYOET· OET

MYOET = MYOETMAX . f(T)
Both phytoplankton and zooplankton (process 12 of Figure 4.2) excrete

phosphorus, nitrogen, and organic carbon (discussed by ]0rgensen, 1979).
These processes are not included in most models, but phytoplankton excretion
is accounted for in the two-step phytoplankton model (Table 4.1, eqns. 12-14)
by consideration of the nutrient uptake as a net uptake. Zooplankton excretion
is often omitted, as it is of minor importance for the nutrient budget. However,
Patten et al. (1975) have used the formulation shown in Table 4.8.
TABLE 4.8 Excretion

Patten et al. (1975)
EXCR = EXCRK . ZOO-N (ammonia excretion)
EXCR = 1.35Z00- 0 . 6 ' (phosphorus excretion)
133
4.2.6. Nitrogen Fixation, Nitrification, and Denitrification
In many lakes, where nitrogen is a limiting factor, nitrogen-fixing algae may
appear during the summer (J0rgensen, 1979b). In such case studies the
nitrogen fixation must be accounted for to obtain a realistic nitrogen budget
(J0rgensen and Mejer, 1979). As shown in Table 4.9, the rate of nitrogen fixation
is related to the concentration of nitrogen-fixing algae. Equations 2 show how
the ratio ofthe nitrogen-fixing species can be connected to the "lack of nitrogen"
relative to phosphorus. Since the algae need approximately five times as much
nitrogen as phosphorus, 5PS - NS expresses the surplus of phosphorus.
TABLE 4.9 Nitrogen Fixation and Nitrification.
(I) Gargas (1976)
NFIX = KNFIX . (I _ NS ) . PHYT(N)

KN + NS
(2) Jorgensen and Harleman (1978)

NFIX = KNFIX . PHYT(N)· (5PS - NS)
or
NFIX = KNFIX . PHYTN)

PHYT(N) 5PS - NS
---- 5PS ~ NS
PHYT 5PS
PHYT(N) = 0 NS> 5PS

NITRI = NITRIK . NH4
In most ecological water quality models, ammonia and nitrate are con-
sidered as one constituent: dissolved inorganic nitrogen. If the two compounds
are separated, however, the nitrification rate will account for the change from
ammonia to nitrate. The most generally applied mathematical formulation for
this process is the first-order kinetic, eqn. 3.
In many eutrophic lakes the sediment becomes anaerobic during the summer,
which implies that denitrification at the sediment-water interface may occur.
Nitrogen balances of hypereutrophic lakes demonstrate that this process is of
great significance for nitrogen balance. As much as 30 % or more of the total
nitrogen input may be removed by denitrification (case study by J0fgensen
et aI., 1973).
Denitrification is a microbiological process and a more complete description
must account for the dynamics of the denitrifying bacteria, as demonstrated by
134
TABLE 4.10 Denitrification.
(I) O'Connor et al. (1976)
dM
- = MYM·M
dt
MYMMAX· NIT
MYM=------
KNIT + NIT
M = M(O) + STO . (NIT(O) - NIT)
Substitution of the last two equations into the first gives upon integration:
KNIT NIT(O)
MYMMAX . t = - - - - - - - - In - - - - -
M(O) + STO . NIT(O) NIT(O) - I/STO(M - M(O»
STO . KNIT + M(O) + STO . NIT(O) M

+ In--
M(O) + STO ' NIT(O) M(O)
(2) lergensen (1976)
DENIT = DENITK . NS
Nitrite nitrogen,
NOrN
Dissolved
organic
nitrogen
Nitrobacter
bacteria
Particulate
organic
nitrogen
Zooplankton Phytoplankton
nitrogen nitrogen
FIGURE 4.5 Aerobic nitrogen cycle.

135
O'Connor et al. (1976) (eqns. 1 of Table 4.10). However, a simple first-order
formulation gives a satisfactory description in some cases (eqn. 2).
Najarian and Harleman (1975) and Harleman (1978) have suggested an
even more detailed nitrogen cycle (Figure 4.5). Seven nitrogen constituents are
considered: ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, phytoplankton
nitrogen, zooplankton nitrogen, particulate organic nitrogen, and dissolved
organic nitrogen. In addition, three species of bacteria are considered:
Nitrosomonas, Nitrobacter, and heterotrophic bacteria. However, normally in a
lake the number of bacteria will generally be constant compared with the
situation in a chemostat, where the bacterial concentration is low initially and
increases exponentially in the first phase (J0rgensen and Harleman, 1978).
Furthermore, the biomass of the bacteria is generally not known or, at best, is
known with great uncertainty. Consequently, a simple quantitative model
description of the nitrogen cycle is not only satisfactory, but must be recom-
mended in most case studies, unless the data allow the calibration and validation
of a more complex model.
4.2.7. Release of Nutrient from Sediment

A detailed description of release of nutrients from sediment is often significant
for predictive models, as discussed by J0rgensen et al. (1973) and J0rgensen
(1976). A completely different prognosis is produced, depending on whether a
simple first-order kinetic or the more comprehensive submodel proposed by
Kamp-Nielsen (1974, 1975) and J0rgensen et al. (1975) is applied. The prin-
ciples of this submodel are demonstrated in Figure 4.6(a). The model
distinguishes between exchangeable and nonexchangeable phosphorus, which
is of great importance when the phosphorus loading is diminished. The phos-
phorus is released in a two-step process:
(i) decomposition of phosphorus compounds, raising the concentration

of interstitial phosphorus, and
(ii) diffusion of interstitial phosphorus to the lake or river water.
Process (i) is different under aerobic and anaerobic conditions. The profile of
the phosphorus indicates which part of the settled phosphorus is exchangeable
and which part is not (Figure 4.6(b». This model does not take all problems into
consideration, e.g. groundwater infiltration.
Another complex submodel for bottom nutrient release has been suggested
(Kozerski, 1977) and applied, with good results, to a lake by Schellenberger
et al. (1978). The sediment model contains nine balance equations and represents
the sedimentation and decay of organic material, the diffusion of phosphorus
and iron from the sediment into water, the diffusion of oxygen into the sediment,
the adsorption of phosphates on to iron compounds, the reduction and oxidation
136
Sinking
Mineralization IPl [g (kg dry maner)-']
at interface Release of P
by diffusion
E
u
.s:: A
Nonexchangeable Exchangeable Interstitial E.
CIl
P P P C
(a) (b)
FIGURE 4.6 Submodel of the exchange of phosphorus between water and sediment.
In (b), A represents the nonexchangeable phosphorus, and B the exchangeable phosphorus.
processes of iron and the compounds of calcium, and the transfer of material
from the thin top sediment layer (5-10 mm) to the deeper, undisturbed layers.
Langmuir's adsorption isotherm was used for the adsorption, and the order
processes were simulated as first-order reactions. Most modelers, however,
have used the first-order formulation (eqn. I of Table 4.11; Chen and Orlob,
1975). Since an essential part of the nutrient flow passes through the sediment,
it is strongly recommended that a more detailed formulation of this submodel
is considered, which implies that the process must be examined in the laboratory
to obtain geochemical site constants.
TABLE 4.11 Release of Nutrient from Sediment.
(I) Nyholm (1978)
REL = RELK . SETTL
4.0 SN + 0.08 A
REL (N) = 1000 . II J(T)
REL (P): submodel (Kamp-Nielsen. 1974, 1975; Jorgensen et al., 1975)
4.2.8. Effect of Temperature on Process Rates

All chemical and biological process rates are strongly dependent upon tempera-
ture. The temperature response of phytoplankton and zooplankton should be
modeled realistically, otherwise large discrepancies can arise between model
and observations. Table 4.12 contains several formulations of the relationship
between process rate and temperature; most of them consider an optimum
temperature.
137
TABLE 4.12 Effects of Temperature upon Biological Rates.
(I) Chen and Orlob (1975)

K(T) = K 2D • KOT T - 2D
(2) Lassiter and Kearns (1974)

T. - T )O( T~ .. -rapd
K(T) = K
oP' exp[a(T - Top,rl ( ----'-m__"_ _
Tmax. - T opt
(3) Lehman et al. (1975)
K(T) = Kopt exp[ -2.3(T - TOP,)2/(Tmox - ToPI )2]
K(T) = K OPI exp[ -2.3(Top, - T)2/(Top, - Tm;o)2]
(4) J0rgensen (1976)
K(T) = K opt exp ( -2.31 T ~5ToPII)

K(T) = KoP' exp(KT)
(5) Lamanna and Malette (1965)
K(T) = KOPI(~)"
T
eXP[1 - (~)"]
T
0 < T < ToP'
opt oP'
K(T) = Kop,[1 _ ( T -_ T
opt
)m] Topl < T < Tmox
Tmu ~Pt
(6) Park et al. (1979)

PHOTO(T) = exp[K(K 1 T 2 - K 2 T K, - I)]
K = -In PHOTO (ocC)
K
I + (In PHOTO (Topt K
- -----,,------:=-------;cc--~-c_
»/
2 - (Tm.. )K,-2(Top, l' - (Top,l'
21 /(K, - 2) T
oP'
K3 Tmox
(7) Stra~kraba (1976)

ToP' = T + 28 exp( -O.115T)
PHOTO(T) = PHOTO(ToPI )' exp[ - K(Topt - T)2]
138
The interactions between the responses of phytoplankton to temperature,
nutrient, and light have been discussed in section 4.2.2. Seasonal adaptation to
water temperature has been considered by Straskraba (1976) (eqns. 7 of Table
4.12).
4.2.9. Effect of Fish on Water Quality

A wide range of fish population models has been developed to optimize fishery,
but the models are far too detailed to be included in water quality models and
focus on a completely different problem. It might be tempting to model year
classes of dominant species, as demonstrated by E. Ursin (1976, private com-
munication), but the computational load is probably prohibitive. Fish is
included as a constituent of a water quality model when there are trophic inter-
actions with phytoplankton (herbivorous fish) and with zooplankton (car-
nivorous fish).
Chen and Orlob (1975) have suggested a compromise, whereby major growth
stages are differentiated. This suggestion is also applied in MS CLEANER
(Park et al., 1979). The weakest point in this modeling approach is the popula-
tion dynamics of the larval and immature stages; the model is highly sensitive
to parameters used for these stages and it is hardly possible to obtain them by
calibration. However, it seems satisfactory in most cases to simulate fish as one
constituent using the same mathematical formulation as for zooplankton, or
even to consider the fish population as a constant or as a driving variable based
on fishery statistics.
MS CLEANER contains a function to determine the migration of fish (and
of zooplankton). The function includes responses to dissolved oxygen levels,
food availability, and temperature. It has also been suggested that the response
to light intensity and swimming rates should be added.
4.2.10. Other Ecological Processes

Many toxic chemicals discharged into rivers and lakes may affect the water
quality. The problem of modeling the distribution and effect of toxic substances
is treated in Chapter 10. The relation between the concentrations of a toxic
compound and of a biological constituent is often expressed as
MORTALITY = MORT + pC"
where MORT is the natural mortality, p the toxicity coefficient (m 2 g-l day-I),
and C x the concentration of the toxic compound.
Toxic substances can also inhibit growth. There are two possibilities, com-
petitive or allosteric inhibition, for which Michaelis-Menten expressions can
be used:
139
Competitive
inhibition
Allosteric
inhibition
Concentration of inhibitor
FIGURE 4.7 Process rate as a function of inhibitor concentration for both competitive
and allosteric inhibition.
Competitive Inhibition
dN l max N1 B
-aNjBJ.lBN, K + N 1 + a2 N 2 ' (4.3)
dt N,
where N 1 is the concentration of inhibited nutrient, N 2 the concentration of

inhibiting nutrient, a 2 a dimensionless inhibition constant for N 2' and B the
concentration of bacteria.
Allosteric Inhibition
dN l _ max NIB
dt aN1BJ.lBN'(KN,+Nl)(I+a2N2)· (4.4)
It is assumed that the affinity of the bacteria for the inhibitor is as strong as
for the inhibited nutrient. If the concentration N 2 is insignificant, then (4.3) and
(4.4) are identical. In competitive inhibition the rate of N 1 degradation is less
dependent on the concentration N 2 than in allosteric inhibition. An allosteric
inhibitor will tend to block the degradation of a nutrient or constituent much
faster than if it is only a competitive inhibitor. These distinctions and effects
are illustrated in Figure 4.7.
4.2.11. Parameter Values and Stoichiometric Ratio

Table 4.14 summarizes some generally adopted values of parameters. Table
4.15 surveys the most used stoichiometric ratios, which might be employed to
compute, for example, phytoplankton concentration from chlorophyll con-
centration, etc. These values should be considered only as means, since no
140
TABLE 4.13 Notation for Tables 4.1 to 4.12.
a constant (OC - I)
Aarea (m 2 )
AK constant
ASSEassimilation efficiency
AVavailability of food
BL body length (mm)
CA concentration of carbon in phytoplankton (g C m - 3)
CAMAX. CAMlN maximum. minimum concentrations of carbon in phytoplankton
[g C (g dry matter) - ']
CFGl conversion factor from global irradiance to light active in photosynthesis
CK carrying capacity (g m - 3)
CS concentration of inorganic soluble carbon (g m -3)
Dp constant (m 3 g - ')
DENIT rate of denitrification (g m - 3 day - 1)
DENITK rate coefficient of denitrification (day - I)
DERV detritus mineralization rate (g m - 3 day - I)
DET detritus concentration (g m - 3)
EXCR excretion rate (g m - 3 day - ')
EXCRK excretion rate coefficient (day - ')
((x) function of x
Fmax maximum fractional reduction in daily specific growth rate over eutrophic
depth
FAC correction factor for biochemical growth activities during dark periods
FD relative day length
g(y) function of y
GRZ grazing rate (g m- 3 day-I)
I irradiancet
1(0) irradiance, surface intensityt
lopt optimum irradiancet
IDA daily average irradiancet
IH light saturation parametert
lK light saturation parametert
K unspecified constant
K 20 constant or coefficient at 20°C
K(T) constant or coefficient at T
K opt constant at optimum temperature
KC half-saturation constant of uptake of inorganic soluble carbon (g m - 3)
KFLM food level, where multiplier is 1(1 + KMFM)
KLYS constant for light inhibition
KMFM minimum filtering rate multiplier
KN half-saturation constant of uptake of soluble inorganic nitrogen (g m - 3)
KNFIX nitrogen fixation coefficient (day-I)
KNIT nitrogen fixation coefficient for denitrification (g m - 3)
KOT temperature coefficient
KP half-saturation constant of uptake of soluble inorganic phosphorus
(g m - 3)
KPA saturation constant for intracellular phosphorus
KTR threshold concentration for grazing (g m - 3)
KZ half-saturation concentration for grazing (g m - 3)
L o constant
m constant (a number)
M concentration of denitrification microorganism (g m - 3)
M(O) concentration of denitrification microorganism at t = 0 (g m - 3)
t A number of different units are used.

141
MORT mortality (g m - j day-I)

MORTK mortality coefficient (day- 1)
MY growth rate of phytoplankton (day- ')
MYDET detritus mineralization rate coefficient (day-I)
MYDETMAX maximum detritus mineralization rate coefficient(day-I)
MYM growth rate coefficient of denitrification microorganism (day- ')
MYMAX maximum growth rate of phytoplankton (day- ')
MYMMAX maximum growth rate coefficient of denitrification microorganism (day- ')
MYZ growth rate of zooplankton (day - I)
MYZMAX maximum growth rate of zooplankton (day-I)
n constant (a number)
NA concentration of nitrogen in phytoplankton (g m - 3)
NAMAX. NAMIN maximum. minimum concentrations of nitrogen in phytoplankton
[g N (g dry matter) - ']
NAP NA/PHYT
NFIX nitrogen fixation (g m - 3 day - ')
NH4 concentration of ammonia nitrogen (g m - 3)
NH4L NH4 beyond which nitrate nitrogen uptake is minimal
NIT concentration of nitrate nitrogen (g m - 3)
NIT(O) concentration of nitrate nitrogen at t = 0 (g m - 3)
NITR I nitrification rate (g m - 3 day - 1)
NITRIK nitrification rate coefficient (day-I)
NS concentration of soluble inorganic nitrogen(g m - 3)
NS o constant
OX oxygen concentrations (g m- 3)
PA concentration of phosphorus in phytoplankton (g m - 3)
PAM AX. PAMlN maximum, minimum conccntrations of phosphorus in phytoplankton
[g P (g dry matter)- 1]
PAP PA/PHYT
PHOTO photosynthetic rate (g m - 3 day- I)
PHYT concentration of phytoplankton (g m - 3)
PHYT(N) concentration of phytoplankton with nitrogen fixation (g m - 3)
PKI equilibrium constant for reaction between phosphorus and carrier (1 mol-I)
PREF preference ratio
PS concentration of soluble inorganic phosphorus (g m - 3)
PSo constant
REF reflected light (units as 1)
R(t) total daily radiation (units as 1)
REL rate of release of nutrients (g m - 3 day- ')
RELK release coefficient
RESP respiration (g m- 3 day-I)
RESPK respiration coefficient (duy- I)
SEDT settling rate of detritus (m day-I)
SEPH settling rate of phytoplankton (m day - I)
SETTL rate of removal of nutrients by settling (g m - 3 day- ')
SEV settling rate (m day - ')
SEV(O) surface overflow rate (m day-I)
SEV(P) settling rate of particles (m day-I)
SN nitrogen in upper layer of sediment (g 1- I)
STO stoichiometric constant [mg bacterial mass (mg substrate) - 1]
t time (days)
T temperature CC): T; = T for j in element. T; and 7;, are T in inflow and
outflow
(continued over)
142
T. absolute temperature (K)

T mu ' Tm;n maximum and minimum temperatures, at which rate coefficient is zero
TOPI optimum temperature
T, reference temperature
TL ." TL • O temperature at depth z and time t or 0
TOX concentration of toxic material (g m - 3)
V, V" V 2 mass flow (kgday-l)
UC uptake rate of inorganic carbon (g m -3 day-I)
UCMAX maximum uptake rate of inorganic carbon(day - I)
UN uptake rate of nitrogen (g m -3 day-I)
UNMAX maximum uptake rate of nitrogen (day-I)
UNMIN(NIT) minimum uptake rate of nitrate nitrogen (day-I)
UP uptake rate of phosphorus (g m -3 day- I)
UPMAX maximum uptake rate of phosphorus (day-I)
V volume (m 3 )
~V change of volume (m 3 )
W weight (kg)
Z depth (m)
ZOO concentration of zooplankton (g m - 3)
ct attenuation of water (m - J)
ctk preference coefficient
Ii spccific attenuation coefficient of phytoplankton (g -I m 2)
y specific attenuation coefficient of zooplankton (g-I m 2)
c\ specific attenuation coefficient of detritus (g-I m 2)
( attenuation coefficient (m - 1)
P density (g m - 3)
TABLE 4.14 Parameter Values.
ASSE 0.65 3 0.60-0.65 5 0.6 10 0.6" 0.63 16

CAMAX (g g-I) 0.6 16
CAMIN (g g-l) 0.18 16
CFGI 0.41 7
DENITK (day - I) 0.002 16
eddy diffusion
cocfficient (m 2 day - I) 0.1-0.5 9
Fmax 0.27 6
FAC 1.33 7
IH 20. 10 5 (cal m - 2 day- 1)1 350 (Iy day- 1)40.03-0.2 (ly min -1)13
IK 0.006 (kcal m - 2 S - 1)4300- 350(ly day- 1)5 150(lyday - 1)6 300(ly day- ')10
350 (lY day-I)" 300-350 12
KC (g m- 3 ) 0.5-0.6 4
KFLM (g food C m - 3) 0.2'7
KMFM (g food C m- 3 ) 0.1 17
KN (g m- 3 ) 0.05' 0.02530.3-0.440.02550.01460.058
0.025 10 0.025" 0.005-0.008 12 0.0014-0.QI" 0.2 16
K NFIX (g m - 3 day - I) 0.02 1
KOT 1.024 1.14 7 1.07 7 1.066" 1.08" 1.02 16
143
KP(gm-') 0.005' 0.01' 0.03-0.05 4 0.001 6 0.005-0.0"

0.01 -0.1" 0.002 11 0.005-0.015" 0.0 I 5-0.15 1' 0.03 16
KTR(gm-') 0.5"
KZ (g m - ') 0.5 4 0.7' 4 7 0.51 8 I" 0.5 16
MORTK (day- ') 0.015 1 0.075 10 0.001" 0.125 16
(zooplankton)
MYDETMAX (day- ') 0.002T' 0.001 4 0.035-0.14' 0.01 7 0.007T '0
(organic N) 0.I'oO.00175T II 0.1 16
MYDETMAX (day- ') O.02T.I 0.001 4 0.14-0.40' 0.1 7 0.007T '00.1 10
(organic P) 0.007T II
MYMAX (day-I) 2.5' 0.1 + 0.06T' 1-24 1.3-2.5' 2.4'
0.8-2.471.8-3.9100.58111.1-1.6" 1--2'42.0 16
MYZMAX (day- I) 1.3' 0.012 + 0.02IT' 0.15 4 0.13-1.2'
0.1-0.6 7 0.2-2.0 10 0.06T '10.6-0.85" 0.18'6
NAMAX 0.\7 0.08 16
NAMIN (g g-') 0.04 7 0.02 16
NITRIK (day- I) 0.002 + 0.0025T' 0.03 4 0.04-0.2' 0.0\ T'o 0.002T 1'1
PAMAX 0.02 7 0.03 16
PAMIN 0.00146 7 0.003 16
RELK (day-I) 0.95-1.8 7
RESPK (day-I) 0.06 1 0.01" 0.DI5 16
(phytoplankton)
RESPK (day- ') 0.0007( T - S)'" 0.01 4 0.02-0.16' 0.001 T II
(zooplankton)
SEDT (m day-I) 0.05' 0.2 4 0.01-0.6 7
SEPH (m day-I) 0.2 1 0.05-0.2 4 0.\' 0.05 -0.3 7 0.1 - 5" 0.1 1, 0.2" 0.1 _0.4 14 0.08 16
Tm " (OC) (blue-greens) 45 11
Top, (0C) (blue-greens) 33"
Tm " (OC) 35"
(nanophytoplankton)
Top, (0C)
(nanophytop1ankton)
Tm" eC) 35"
(net phytoplankton)
'J.,p, eC)
(net phytoplankton)
UCMAX (day- I) 0.45'6
UNMAX (day -I) 0.0096'6
UP MAX (day-I) 0.0035 16
C1. (m- I
) 0.16 6 0.3 4 0.27' 6
/1 (m' g-I) 0.24 0.19 6 0.1" 0.18 16
)'(m'g-I) 0.2 1
(j(m ' g- 1 ) 0.01'
1 Gargas (1976) 10 O'Connor et al. (1973)

1 Dahl-Madsen and Strange Nielsen (1974) [ , Thomann et al. (1975)
, Di Toro et al. (1975) 11 Scavia and Park (1976)
4 Chen and Orlob (1975) 13 Lehman et al. (1975)
, O'Connor et al. (1976) 14 Imboden (1974)
6 Larsen et al. (1974) I' Stumm and Morgan (1970)
7 Nyholm (1978) 16 Jorgensen (1976)
8 Ikeda and Adachi (1978) 17 Canale et al. (1976)
9 Lewis and Nir (1978)
144
TABLE 4.15 Stoichiometric Ratios.
C/chlorophyll 50 5 50' 50"

N/chlorophyll 7-10 5 7' 7.2 6 10 '0 lO"
P/chlorophyll 15 13 0.63 6 1'0 1'1
o ,/detritus 24 4
O,/phytoplankton 1.6
CO,/BOD 0.2 4
N zooplankton 0.14 3
P zooplankton 0.02'
C/dry weight of phytoplankton 0.670.33-0.43'00.33'5
N/dry weight of phytoplankton 0.044-0.084 '0 0.045 15
P/dry weight of phytoplankton 0.011-0.029 '0 0.07 15
1 Gargas (1976) 10 O'Connor et al. (1973)
, Dahl-Madsen and Strange Nielsen (1974) 1I Thomann et al. (1975)

'Di Toro et al. (1975) 12 Scavia and Park (1976)
4 Chen and Orlob (1975) 13 Lehman et al. (1975)
5 O'Connor et al. (1976) 14 Imboden (1974)
6 Larsen et al. (1974) 15 Stumm and Morgan (1970)
7 Nyholm (1978) 16 Jorgensen (1976)
B Ikeda and Adachi (1978) 17 Canale et al. (1976)
9 Lewis and Nir (1978)
biota have constant stoichiometry; the element cycles are independent although
some realistic upper and lower limits can be set. A more comprehensive collec-
tion of parameters and stoichiometric ratios has been assembled by J0rgensen
(l979b).
4.3. CONCLUDING REMARKS
As demonstrated in this chapter, the mathematical expressions describing

ecological processes vary from model to model. It is generally possible in
models or in the biological literature to find different mathematical descriptions
of the same biological process. A more detailed examination usually reveals
that the differences are caused by environmental factors being included or ex-
cluded. Theoretically, as many details as possible may be included in the de-
scription of biological processes in rivers and lakes. The question is whether
these will provide additional advantages for the total model, in view of the
objectives of the model. Generally, it must be recommended that the complexity
of the submodels is selected in accordance with the objectives and the data
available for the calibration and validation of the model. However, a causal
submodel that has been used with good results in a number of case studies
might be used where the data do not allow the calibration and validation of the
submodel, but where previously experienced parameters can be applied. In
145
most cases a causal submodel should be preferred to an empirical one, since
it is easier to transfer from one study to another.
Very little has been published, up to the present, about the effect on eco-
system models of replacing one submodel with another. It seems absolutely
necessary to intensify such studies so that modelers can gain more experience
of where and how a certain submodel can be used. Such research should go
hand in hand with laboratory or microscale investigation of important sub-
models.
It is a great advantage to know the process parameters within certain limits.
The first attempt at a comprehensive collection of useful parameters was made
by Jorgensen (I 979b). If, however, such knowledge of limits to parameters is
not available, it might be necessary to conduct experiments in the laboratory
or in situ to obtain the basic data required for good parameter estimations (see
also section 9.4 and Chapter 11). Generally, more complex submodels are used
for processes directly influencing the state variables that are the model object-
ives. For instance, it is of more interest to have a good, detailed submodel for
phytoplankton growth and nutrient uptake than it is for fish feeding on zoo-
plankton in a eutrophication model. Consequently, rather complex phytoplank-
ton growth models have been developed (Table 4.1) for inclusion in
eutrophication models. The same is true for the sediment-water exchange of
nutrients, which should be modeled accurately, as an essential part of the
nutrients in an aquatic ecosystem goes through the sediment (discussed in
section 4.2.7). However, more complex models contain more parameters to be
calibrated, which again requires more data for calibration and validation.
Nevertheless, if the state variable in focus (for eutrophication models it is the
phytoplankton concentration) is sensitive to simplification, the more complex
and causal model should be implemented and the parameters estimated by
separate laboratory or in situ experiments (discussed in section 9.4), if the
measurements are not sufficient to allow for a direct calibration.
REFERENCES
Ahlgren, I. (l973a). Limnologiska studier av Sjan Norrviken. III. Avlastningens effekter. Scripra
Limnolo,qica Upsaliencia No. 333,
Ahlgren, I. (1973b). Limnological studies of Lake Norrviken. a eutrophicated Swedish lake.
Nutrients and phytoplankton before and after sewage diversion. Scripra Limnoloqica Upsaliencia
No. 334.
Anderson. J. M. (1973). The eutrophication of lakes. Towards Glohal Equilihrium, ed. D. Meadows
(Cambridge. MA: MIT Press). pp. 117-140.
Bierman, Y. J. (1976). Mathematical model of the selective enhancement of blue-green algae by
nutrient enrichment. Modeling Biochemical Processes in Aquaric Ecosysrems. ed. R. P. Canale
(Ann Arbor. MI: Ann Arbor Science), pp. 1-32.
146
Bierman, V. J., F. H. Verhoff, T. C. Poulson, and M. W. Tenney (1974). Multinutrient dynamic
models of algal growth and species competition in eutrophic lakes. Modeling the Eutrophication
Process, eds. E. Middlebrooks, D. H. Falkenberg, and T. E. Maloney (Ann Arbor, MI: Ann
Arbor Science), pp. 89-109.
Bloomfield, J. A., R. A. Park, D. Scavia, and C. S. Zahorcak (1974). Aquatic modeling in the eastern
deciduous forest biome, US International Biological Program. Modeling the Eutrophication
Process, eds. E. Middlebrooks, D. H. Falkenberg, and T. E. Maloney (Ann Arbor, MI: Ann
Arbor Science), pp. 139-158.
Broqvist, S., (1971). Matematisk modell for ekosystemet i en sjo. Forskningsgruppen fOr Planerings-
teori, Matematiske Institution, Tekniska Hogskolan, Stockholm.
Canale, R. P., L. M. DePalma, and A. H. Vogel (1976). A plankton-based food web model for Lake
Michigan. Modeling Biochemical Processes in Aquatic Ecosystems, ed. R. P. Canale (Ann Arbor,
MI: Ann Arbor Science), pp. 33-74.
Chen, C. W. (1970). Concepts and utilities of ecological models. Proceedings oj'the American Society
oj' Civil Engineers, Journal oj'the Sanitary Engineering Division 96(SA5): 1085-1097.
Analysis and Simulation in Ecology vol. 3, ed. B. C. Pallen (New York, NY: Academic Press), pp.
476-588.
Chris holm, S. W., R. G. Stross, and P. A. Nobbs (1975). Environmental and intrinsic control
of filtering and feeding rates in Arctic Daphnia. Journal oj' Fisheries Research Board oj' Canada
32:219-226.
Cloern, J. E. (1978). Simulation model of Cryptomonas O/'ata population dynamics in Southern
Kootenay Lake, British Columbia. Ecological Modelling 4: 133-150.
Dahl-Madsen, K.I., and K. Strange Nielsen (1974). Eutrophication models for ponds. Vand5: 24-31.
DePinto, J. V. (1979). Water column death and decomposition of phytoplankton: an experimental
and modeling review. Perspectives on Lake Ecosystem Modeling, eds. D. Scavia and A. Robertson
(Ann Arbor, MJ: Ann Arbor Science), pp. 25-52.
Di Giarno, F. A., L. Lijklema, and G. van Straten (1978). Wind-induced dispersion and algal
growth in shallow lakes. Ecological Modellin.l/ 4: 237-252.
Di Toro, D. M., D. J. O'Connor, and R. V. Thomann (1971). Nonequilibrium systems in natural
water chemistry. American Chemical Society, Adl'ances in Chemistry Series 106: 131-180.
Di Toro, D. M .. D. J. O'Connor, R. V. Thomann, and J. L. Mancini (1975). Phytoplankton-
zooplankton-nutrient interaction model for western Lake Erie. Systems Analysis and Simulation
Ecology vol. 3, ed. B. C. Pallen (New York, NY: Academic Press), pp. 423-474.
Dodson, I. S. (1975). Predation rates of zooplankton in arctic ponds. Limnology and Oceano-
graphy 20(3): 426-433.
Droop, M. R. (1974). The nutrient status of algal cells in continuous culture. Memoranda oj'Marine
Biological Association, UK 54:825-855.
Dugdale, R. C. (1975). Biological modeling. Modeling oj'Marine Systems, ed. J. C. Nihoul (Amster-
dam: Elsevier), pp. 187-206.
Eppley, R. W., R. W. Holmes, and J. D. H. Strickland (1967). Sinking rates of marine phyto-
plankton measured with a t1uorometer. Journal of Experimental Marine Biology and Ecology
I: 191-208.
Frost, B. W. (1975). A threshold feeding behavior in Calanus paci!icus. Limnology and Oceano-
graphy 20: 263-266.
Fuhs, G. W. (1969). Phosphorus content and rate of growth in the diatoms Cyclotella nana and
Thalossiosira fluL'iatilis. Journal 0/ Phycology 5: 312.
Gargas, E. (1976). A three-box eutrophication model of a mesotropic Danish Lake. Water Quality
Institute, Horsholm, Denmark.
Gause, G. F. (1934). The Struggle o/Existence (New York, NY: Hafner), p. 133.
Harleman, D. R. F. (1978). A comparison of water quality models for the aerobic nitrogen cycle.
Research Memorandum RM-78-34 (Laxenburg, Austria: International Institute for Applied
Systems Analysis).
147
Ikeda, S., and N. Adachi (1978). A dynamic water quality model of Lake Biwa. Ecolo,qical Modellin,q
4: 151-172.
Ikusima, I. (1966). Ecological studies on the productivity of aquatic plant communities. II. Botanical
Magazine, Tokyo 79:7-19.
Imboden, D. M. (1974). Phosphorus model for lake eutrophication. Limnology and Oceanography
19:297-304.
Jansson, B. O. (1972). Ecosystem approach to Baltic problem. Swedish Natural Science Research
Council Bulletins/rom Ecolo,qical Research Committee No. 16.
.Jorgensen. S. E. (1976). A eutrophication model for a lake. Ecological Modelling 2: 147-165.
Jorgensen, S. E. (I 979a). Lake Mana,qement(Oxford: Pergamon).
Jorgensen, S. E. (ed.) (l979b). Handbook ofEnvironmental Data and Ecological Parameters (Oxford:
Pergamon; Copenhagen: International Society for Ecological Modelling).
Jorgensen, S. E., and D. R. F. Harleman (eds.) (1978). Hydrophysical and ecological modelling of
deep lakes and reservoirs. Summary Report of a IlASA workshop, December 12-15, 1977.
Collaborative Proceedings CP-78-7 (Laxenburg, Austria: International Institute for Applied
Systems Analysis).
Jorgensen, S. E., O. S. Jacobson, and I. Hoi (1973). Prognosis for a lake. Vatten 29: 383-404.
Jorgensen, S. E., L. Kamp-Nielsen, and O. S. Jacobsen (1975). A submodel for anaerobic mud-
water exchange of phosphate. Ecological Modelling I : 133-146.
Jorgensen, S. E., and H. F. Mejer (1979). A holistic approach to ecological modelling. Ecolo,qical
Modellin,q 7: 169-190.
Jost, J. L., I. F. Drake, A. G. Frederickson. and H. M. Tsandriya (1973). Interactions of Tetra-
hymena pyr!lormis, Escherichia coli, Az tobacter rinelandii and glucose in minimal medium.
Journal ol Bacteriolo,qy 113: 834-840.
Kamp-Nielsen, L. (1974). Mud-water exchange of phosphate and other ions in undisturbed
sediment cores and factors affecting the exchange rate. Archil' .fur Hydrobiologie 2: 218-
237.
Kamp-Nielsen, L. (1975). A kinetic approach to the aerobic sediment-water exchange of phos-
phorus in Lake Esrom. Ecological Modelling I: 153-160.
Kozerski, H.-P. (1977). Ein einfaches mathematisches Modell fiir Phosphataustausch zwischen
Sediment und Freiwasser. Acta Hydrochimica et Hydrobiologia 5: 53-65.
Kremer, J. N., and S. W. Nixon (1978). A Coastal Marine Ecosystem Simulation and Analysis
(Berlin: Springer), 217 pp.
Lamanna, c., and M. F. Malette (1965). Basic Bacteriology (Baltimore, MD: Williams and
Wilkins).
Larsen, D. P., H. T Mercier. and K. W. Malveg (1974). Modeling algal growth dynamics in
Shagawa Lake, Minnesota. Modeling the Eutrophication Process, eds. E. J. Middlebrooks,
D. H. Falkenberg, and T E. Maloney (Ann Arbor, M I: Ann Arbor Science), pp. 15-33.
Lassen. H., and P. B. Nielsen (1972). Simple mathematical model for the primary production as a
function of the phosphate concentration and the incoming solar energy applied to the North
Sea. Danmarks Fiskeri- og Havundersogelser, International Council for the Exploration of the
Sea, Plankton Committee 1972.
Lassiter, R. R .. and D. K. Kearns (1974). Phytoplankton population changes and nutrient fluctua-
tions in a simple aquatic ecosystem model. Modeling the Eutrophication Process, eds. E. J.
Middlebrooks, D. H. Falkenberg. and T E. Maloney (Ann Arbor, MI: Ann Arbor Science),
pp.131-138.
Lehman, J. T, D. B. Botkin, and G. E. Likens (1975). The assumptions and rationales of a computer
model of phytoplankton population dynamics. Limnology and Oceanography 3: 343-364.
Lewis. S., and A. Nir (1978). A study of parameter estimation procedures of a model for lake
phosphorus dynamics. Ecological Model/ing 4: 99-118.
Lotka, A. J. (1924). Elements ol Mathematical Biolo,qy (New York, NY: Dover) (reprinted 1956).
Najarian, TO., and D. R. F. Harleman (1975). A real-time model of nitrogen-cycle dynamics in an
estuarine system. MIT Department ol Cil'il En,qineering, R. M. Parsons Laboratory Report 204.
148
Nyholm, N. (1975). Kinetics studies of phosphate-limited algae growth. Thesis, Technical Uni-
tersity of Copenhagen.
Nyholm, N. (1978). A simulation model for phytoplankton growth and nutrient cycling in eutro-
phic, shallow lakes. Ecological Modelling 4: 279-310.
O'Brien, J. J., and J. S. Wroblewski (1972). An ecological model of the lower marine trophic levels
on the continental shelf of West Florida. Geophysical Fluid Dynamics Institute, Florida State
Unil'ersity. Tallahassee, FL Technical Report, 170 pp.
O'Connor, D. J., R. V. Thomann, and D. M. Di Toro (1973). Dynamic water quality, forecasting
and management. US Enl'ironmental Protection Agency, Washington, DC Report EPA 660/3-73-
009.
O'Connor, D. I., R. V. Thomann, and D. M. Di Toro (1976). Ecologic models. Systems Approach
to Water Management, ed. V. Biswas (New York, NY: McGraw-Hill), pp. 299-333.
Odum, H. T. (1972). An energy circuit language. Systems Analysis and Simulation in Ecology vol. 2.
ed. B. C. Pallen (New York, NY: Academic Press), pp. 139-211.
Orlob, G. T. (1975). Modeling of aquatic ecosystems. Integrity 01' Water. US Enl'ironmental Pro-
tection Agency Report (Washington, DC: US Government Printing Office), pp. 189-201.
Park, R. A., T. W. Groden, and C. J. Desormeau (1979). Modification to model CLEANER,
requiring further research. Perspecth'es on Lake Ecosystem Modeling, eds. D. Scavia and A.
Robertson (Ann Arbor, MI: Ann Arbor Science), pp. 87-108.
Park, R. A., et al. (1974). A generalized model for simulating lake ecosystems. Simulation 23: 33-50.
Parker, R. A. (1972). Estimation of ecosystem parameters. Verhandlung Internationaler Vereinigung
Limnologie 18: 257-263.
Pallen, B. c., D. A. Egloff, and T. H. Richardson (1975). Total ecosystem model for a cove in Lake
Texoma. Systems Analysis and Simulation in Ecology vol. 3, ed. B. C. Pallen (New York, NY:
Rhee, G. Yu. (1978). Effects ofN: P atomic ratios and nitrate limitation on algal growth, cell com-
position, and nitrate uptake. Limnology and Oceanography 23: 10-25.
Richey, J. E. (1977). An empirical and mathematical approach toward the development of a
phosphorus model of Castle Lake. Ecosystem Modeling in Theory and Practice, eds. C. A. S. Hall
and I. W. Day, Jr. (New York, NY: Wiley-Interscience), pp. 267-287.
Scavia, D., B. J. Eadie, and A. Robertson (1976). An ecological model for the Great Lakes. Great
Lakes Em'ironmental Research Laboratory, National Oceanic and Atmospheric Association,
Ann Arbor, MI Contribution 64.
Scavia, D., and R. A. Park (1976). Documentation of selected constructs and parameter values in
the aquatic model CLEANER. Ecological Modelling 2: 33-53.
Schellenberger, G., H.-P. Kozerski, H. Behrendt, and S. Hoeg (1978). A mathematical ecosystem
model applicable to shallow water bodies. Modeling the Water Quality ofthe Hydrological Cycle.
Proceedings of Baden Symposium, September. International Association of Hydrological Sciences
Publication 125, pp. 128-136.
Schofield, W. R., and R. G. Krutchoff (1974). Stochastic model for a dynamic ecosystem. Virginia
Water Resources Research Center Bulletin VPI-WRRC 60.
Serruya, c., and T. Berman (1975). Phosphorus, nitrogen and the growth of algae in Lake Kinneret.
Journal 01' Phycology 11(2): 155-162.
Shelef, G., W. J. Oswald, and C. G. Golueke (1969). The continuous culture of algal biomass on
wastes. Proceedings 01' 4th Symposium on Continuous Culth'ation 01' Microorganisms, Prague
(Oxford: Pergamon), pp. 601-629.
Shelef, G., M. Schwarz, and M. Schechter (1972). Prediction of photosynthetic production in
biomass in accelerated algal-bacterial wastewater treatment systems. Proceedings of 6th Inter-
national Symposium on Water Pollution Research (Oxford: Pergamon), Section 5, Paper a, pp.
1-10.
Smayda, T. J. (1970). The suspension and sinking of phytoplankton in the sea. Oceanography and
Marine Biology Annual Rel'iew 8: 353-414.
149
Smayda, T. J. (1974). Some experiments on the sinking rates of two freshwater diatoms. Limnology
and Oceanography 19: 628.
Steel, J. A. (1973). Reservoir algal productivity. The use of mathematical models in water pollution
control. Unh-ersity o{Ne.....castle upon Tl'ne Symposium. ed. A. James. pp. 107~135.
Steele. J. H. (1974). The Structure oj Marine Ecosystems (Oxford: Blackwell), pp. 74-91.
Stefan, H., T. Skoglund, and R. Megard (1976). Wind control of algae growth in eutrophic lakes.
Proceedings 0/ American Societyo/Ch·i/ Engineers, Journal o{Environmental Engineering Dieision
102(EE6): 1201-1213.
Straskraba. M. (1976). Development of an analytical phytoplankton model with parameters
empirically related to dominant controlling variables. Um .....eltbiophysik, eds. R. Glaser, K. Unger,
and M. Koch (Berlin: Akademie-Verlag), pp. 33-65.
Stumm, W., and J. J. Morgan (1970). Aquatic Chemistry (New York, NY: Wiley), 432 pp.
Takahashi, M., K. Fujii, and T. R. Parsons (1973). Simulation study of phytoplankton photo-
synthesis and growth in the Fraser River Estuary. Marine Biology 19: 102-116.
Tailing, J. F. (1957). The phytoplankton populations as a compound photosynthetic system. Ne .....
Phytology 56: 133-149.
Thomann, R. V., D. M. Di Toro, and D. J. O'Connor (1974). Preliminary model of the Potomac
Estuary phytoplankton. Proceedings o{ American Society o{ Civil Engineers, Journal o{ Enl'iron-
mental Engineering Division I 00(EE3): 699-715.
Thomann, R. V., D. M. Di Toro, R. P. Winfield, and D. J. O'Connor (1975). Mathematical modeling
of phytoplankton in Lake Ontario. US Environmental Protection Agency National Environ-
mental Research Center, Corvallis, OR Report EPA 660/3-75-005.
Titman, D., and P. Kilham (1976). Sinking in freshwater phytoplankton: some ecological implica-
tions of cell nutrient status and physical mixing processes. Limnology and Oceanography
21(3) :409-417.
Vollenweider, R. A. (1965). Calculation models of photosynthesis-depth curves and some implica-
tions regarding day rate estimates in primary production. Memorie dell Istituto Italiano di
Idrobiologia 18 Suppl.: 425-457.
Walsh, J. J., and R. C. Dugdale (1971). Simulation model of the nitrogen flow in the Peruvian
upwelling system. Im·estigacion Pesquera 35: 309-330.
Youngberg, B. A. (1977). Application of the aquatic model CLEANER to a stratified reservoir
system. Center jor Ecological Modeling, Rensselaer Polytechnic Institute. Troy, NY Report 1.
CHAPTER 4: NOTATION
aN,B constant
a2 inhibition constant
B concentration of bacteria
ex concentration of toxic compound
Ks half-saturation constant
N1 concentration of inhibited nutrient
N2 concentration of inhibiting nutrient
S concentration of substrate
f3 toxicity coefficient.
Table 4.13 lists the symbols used in Tables 4.1-4.12.
Simulation of the Thermal Regime of Rivers
5 J. Jacquet
5.1. PURPOSE OF MODELING
Water temperature has an important influence on the water quality, as well as

the aquatic life, of a river. It varies according to a seasonal rhythm, upon which
are superimposed random fluctuations that are closely linked to prevailing
meteorological conditions. Knowledge of river temperature is more or less
good, depending on the interest that investigators have attached to it; long
time series of temperatures are available for several river sections, but for others
temperature measurements are completely missing.
The increasing importance given to solving environmental problems has
made it necessary to study the thermal regime of streams in considerable detail.
The following points are important with regard to the impacts of technology.
(1) It is necessary to understand in detail the thermal regime of a river reach
upon which a thermal power plant is to be located, in order to design
the cooling system, which includes internal piping, pumping equipment,
and the condenser.
(2) The design of the inlet and outlet structures of a thermal power plant
must take into account the risks of recycling rejected hot water.
(3) One must understand that the thermal regime of a reach where a power
plant is located is disturbed by all other installations upstream.
(4) The use of potable water from a stream requires maintenance of a mini-
mum quality of water; specifically, the temperature should not exceed
a certain threshold, beyond which problems might arise in water
treatment and in delivery to the consumer.
From the environmental point of view, knowledge of the thermal regime of
a stream is fundamental. A forecast of the impacts of any development on water
150
151
quality and on aquatic life is often made by estimating the modification of this
regime. Modification of water temperature can have direct, even lethal effects
on the aquatic fauna or flora. It can also have an indirect effect through the
disequilibrium imposed on the population by the competition between species
reacting differently to the modification. Statistical methods, for example, permit
one to indicate the domain of variations to which the distribution of populations
is sensitive.
5.2. THERMAL REGIME OF A STREAM
The thermal regime of a stream depends essentially on the climatic conditions of

the regions through which it passes, on its hydraulic characteristics, and on the
temperatures of its successive water inputs. In practice, the water temperature
becomes more closely linked to local climatic conditions as the distance from
the source increases. Thus, the temperature of a river flowing slowly through a
plain, where the climate is often uniform, depends almost entirely on local
meteorological conditions, and it can be easily calculated from these. On the
other hand, a mountain stream or a river with a steep slope goes through very
different thermal regimes; the water temperature does not depend strongly on
meteorological conditions, but reflects rather those of its tributaries.
Modification of the hydraulic characteristics of a stream or a change of
inflows will disturb its thermal regime. One of the most common causes of
modification is the development of a stream for navigation. By creation of
successive flow levels, the flow is usually slowed and the depths are very often
considerably increased. This type of development generally does not change
the average temperature, but changes appreciably the thermal inertia and
eliminates the daily fluctuations of temperature. In a region of natural cold
waters an average temperature increase may be observed. Similarly, the con-
struction of a reservoir on a stream or its tributaries can cause a serious change
in the thermal regime downstream, depending on the scale of the development.
Normally, the temperature will be lower in summer and higher in winter. A
third cause is industrial or municipal discharge, which supplies significant
amounts of water at a different temperature from that of the stream. This can
modify the thermal regime of the receiving medium. Among such discharges
are those from treatment plants, discharges from various industries (e.g. metal-
lurgical industries), and, especially, discharges from thermal power plants,
which are often the most important ones.
Other sources of modification of the thermal regime are the reduction of
friction losses through the construction of hydroelectric power plants, resulting
in cooling; heat input from navigation; and heat inputs or losses through bio-
chemical reactions of self-purification in streams.
152
Each of these modifications to the thermal regime is of a quite different
nature and each requires a different approach for it to be properly understood.
The two most important causes of perturbation to the thermal energy balance
of streams are (in order) heat discharge by industry, and the construction of
large reservoirs. The study of the latter type of perturbation requires particular
knowledge of stratified impoundments, to be discussed in Chapter 7. The
problem posed for discussion in this chapter is that of the heat discharge to
streams by large power plants.
5.3. AREAS FOR MODELING
The study of perturbations to the thermal regime of a stream, caused by the

discharge of hot water or by modifications of the hydraulic characteristics,
reveals two different problems. The" near field" problem is concerned with
a very limited zone of a stream near the point of maximum perturbation. How
does the thermal perturbation act locally? In other words, how, for example,
do the hot waters mix with the receiving medium? What is the local modification
in the daily thermal cycle due to hydraulic changes? This type of problem
requires, most often, the use of physical hydraulic models or special mathe-
matical models (Chapter 8, section 8.5: Mathematical models of cooling
impoundments).
In "far field" problems, the main emphasis is placed on global effects. What
is the effect of a perturbation on the thermal regime of a river a considerable
distance downstream of the disturbance, after the initial mixing with hot water?
Up to what distance will the average temperature increase of the stream re-
main significant? How might a series of power plants within the same river basin
modify the thermal regime of the basin and its principal drainage courses? One
considers the stream to be homogeneous in a given section and studies the
perturbations in the average temperature, considered significant only at the basin
level.
5.4. THERMODYNAMICS OF A RIVER
The thermal regime of a stream can be defined by a great number of parameters,

depending on the area of interest:
temperature time series corresponding to various sections of the stream;

periodic characteristics, including the periods, phases, and amplitudes of
daily and annual cycles;
variations in temperature with space or time, including deviations observed
simultaneously at a section of the river; deviations of temperature on the
same day, deviations from one day to the next, from one year to the next, etc.;
153
extreme changes in temperatures of very short duration, or daily or monthly
averages; and
stochastic structure of temperature, considered as a continuous process.
For example, to optimize the cooling circuit of a power plant, one would
have to know the temperature time series; but to judge the development of a
basin as a whole, one requires statistical parameters. The biologist who would
like to estimate the possible impact of a development would be interested in
still other parameters: extreme values, for example, or the thermal constraints
applied to the aquatic life, estimated after the modification of average temper-
atures in a given critical period.
The thermal regime of rivers is governed generally by the following
phenomena: exchange of energy with the atmosphere; exchange of energy with
the banks; heat input by inflowing water, a tributary, or a waste discharge; heat
exchange at a weir, waterfall, or rapids; artificial heat injection; evaporation
and condensation; internal heat dissipation by friction; heat liberated by bio-
chemical reaction; and convection and diffusion in the water mass. The study of
the local modification of the temperature field does not usually require a
completely reliable description of all these phenomena. Usually this is only
necessary for a particular development, like the installation of a once-through
cooled thermal power plant or a planned wastewater discharge. In practice,
such a study is made either with a physical model or with a mathematical model
that emphasizes the hydraulic mechanisms of heat transfer.
In contrast, problems related to water temperature changes over long
distances or periods can only be treated with mathematical models. The more
important processes considered in such models include energy exchange with
the atmosphere or the ground.
5.5. SIMULAnON MODELS
The use of either scaled physical or mathematical simulation models is essential

for studying the effects of various developments on the thermal regime or local
thermal structure of a stream. These models have to describe the evolution of
the temperature regime (or of the field of temperatures in a given zone, in
physical models) starting from limited data. The mechanisms governing this
evolution must be represented with sufficient realism that the effects of a modifi-
cation of the system can be studied. As suggested in section 5.3, two types of
model have to be considered. There are those describing local phenomena,
such as the field of temperatures in the proximity of a discharge, in which the
phenomena are essentially governed by mechanisms internal to the water
mass. Then there are models that describe the phenomena in an integrated
154
fashion, where the exchanges of energy between water and the atmosphere play
a most important role and hydraulic phenomena can be strongly schematized.
5.5.1. Deterministic Approach
Simulation of the evolution of the temperature regime along a river by con-

sidering the river in its entirety, or even as a hydrogeographic basin, is only
done at the cost of certain simplifications. The first simplification is that the
river is homogeneous in temperature in all cross sections. This hypothesis is
not always verified, but it results in appreciable errors only in exceptional
cases. In effect, heterogeneity of temperature in a cross section is only caused
by local phenomena: zones of dead water, and mixture zones with tributary
waters or with industrial discharges. These phenomena normally have only
little influence on the thermal regime downstream. Kaisers6t and Mitschel
(1977) made a theoretical study of these extreme cases.
Hydraulic phenomena must also be schematized, sometimes even to an
extreme. Practice shows, however, that these simplifications are justifiable.
Turbulence and longitudinal dispersion, under special circumstances (sudden
variations of stream discharge or sporadic industrial discharges), can some-
times cause appreciable errors if they are not taken into account.
With these simplifying assumptions, the evolution of water temperature in
space and time can be described according to the conservation of energy
principle by
aT
peH (at + U
aT)
ax = R s + Ra - Re - E- C +P +
as
ax' (5.1)
where
p is the density of water (kg m - 3),
e is the specific heat of water (J kg - I),
H is the average water depth (m),
T is the water temperature (OC),
x is the distance downstream (m),
Rs is the solar radiation flux (ly day - I),
Ra is the atmospheric radiation flux (ly day - I),
Re is the radiation flux emitted from the water surface (ly day-I),
E is the energy flux exchanged by evaporation, normalized per unit area of
water surface (ly day - I),
C is the energy flux exchanged by convection at the air-water interface,
normalized per unit area of water surface (ly day- I),
155
P is the heat flux introduced artificially into a given section (ly day-l),
as/ax is the heat flux exchanged with the ground and the banks of a stream,
normalized per unit area of water surface (ly day-l),
1 ly is one langley (l calorie cm - 2).
Equation 5.1 can be written simply as
pcH (
aT +
at U
aT)
ax = F(T, x, t). (5.2)
5.5.2. Models for Steady State Conditions

There are various methods for solving (5.2). The first studies attempted merely
to simplify the solution. Accordingly, we note that for a mass of water moving
downstream,
aT
pcH at = F(T, t). (5.3)
The assumption of steady state conditions permits direct integration of this

equation. Very often, for the equation in this form, the idea of an "equilibrium
temperature" 1;, is introduced. This may be taken to be the temperature of a
mass of water without thermal inertia, i.e. for which aT/at would be zero, or
F(1;" t) = O.
Equation 5.3 can be written simply as

dT .
pcH Cit = F(T, t) - F(1;" t)
aF
= aT (T - 1'.) = K(T - 1;,). (5.4)
5.5.3. Exponential Method

Under steady state conditions 1;, is constant. If T is close to 1;" or varies only
slightly along the stream, K can also be considered constant, which leads to
~ ~ i = exp(~c~). (5.5)
where T is the temperature at time t = O. Therefore, under steady state con-

ditions the water temperature tends toward the equilibrium temperature
exponentially with distance (or flow time) from the source. At this point con-
fusion often arises between the natural water temperature Tn and the equilibrium
temperature 1;" confusion that can lead to a mistake when the progressive
156
disappearance of a temperature perturbation is studied. In fact, it would lead
to writing (5.3) as
d(T - Tn) o f ,
pcH dt = F(T, t) = F(Tn, t) = at (T - Tn) = K (T - Tn)' (5.6)
K' has a different value from K in (5.5) because it corresponds to the derivative
of the function F for another value of water temperature (Kahlig (1977) dis-
cusses this problem).
This type of approach permits simple assessment of questions of the evolution
of a temperature perturbation generated, for example, by a discharge from a
thermal power plant or by a release from a reservoir. Two essential weaknesses
limit its usefulness.
(i) In practice, all stream parameters vary along the stream or in time. In
particular, hydraulic parameters (e.g. average depth, stream discharge,
roughness) vary considerably from upstream to downstream. Therefore,
the simplified model can only be applied to reaches that are relatively
short and along which the flow and geometric parameters vary only
slightly.
(ii) The model, under steady state conditions, cannot be used to calculate
the natural evolution of the water temperature along a stream, but only
the evolution of the average difference in temperature with respect to
either the equilibrium temperature or the natural temperature.
It is, therefore, impossible to compare calculated with measured values and

to test the validity of the model. Nevertheless, provided that the coefficient K
of energy exchange is correct, the calculations based on this method allow one
to determine the approximate size of the residual temperature increase down-
stream of a section where heat was injected.
On the other hand, this method is completely inadequate for estimating the
effect of a modification in flow on the thermal regime of a stream (e.g. the effect
of channelization).
5.5.4. Differential Equation for Steady State Conditions

In order to carry out more realistic calculations it is possible to integrate (5.1)
stepwise, still keeping the hydrometeorological conditions steady, but allowing
geometric data to change from one step to the next. In this way the masses of
water are followed downstream. This method was used by the Uinderarbeits-
gemeinschaft Wasser (1971) working group to study thermal energy changes in
the Rhine, by the Hydraulic Institute of Karlsruhe for the Neckar River (Fleig
and Flinsbach, 1973), and by the Hydrologisches Institut Miinchen (1976) for
the Danube. The method allows one to simulate typical cases (winter, spring,
157
summer, fall) by choosing fixed conditions for the water discharge of the stream,
its natural temperature, and coefficients, representative of the season, for the
exchanges of energy (average and extreme values).
5.5.5. Models for Nonsteady State Conditions

The approach outlined above can be modified still further by varying the set of
parameters for the whole length of the stream. The calculation (for a given mass
of water) is the same as given above, except that for each time step all para-
meters are allowed to vary, especially the hydrometeorological parameters.
Various approaches have been proposed to determine values of these
parameters.
Some case studies use real data (chosen more or less arbitrarily to represent
the periods or conditions given, such as seasons, or periods of low water level).
An example of this approach is a model developed by Raphael (1962) and
applied by Motor-Columbus to the Rhine. Each season was represented by a
period of ten days. The model introduces a longitudinal diffusion term by the
following artifice. For each section the calculation is made for two masses of
water flowing at different speeds; the residence times are different and the
temperatures of the masses evolve differently; at each end of a reach the masses
of water are mixed, and the average temperature calculated corresponds to the
temperature of the mixture. Then the model is calibrated according to the
evaporation coefficient. This calibration consists of the determination of the
value of the coefficient that gives the best agreement between the calculated and
measured temperatures for the four ten-day periods.
5.5.6. Simulation Models for Nonsteady State Conditions

Instead of considering some typical periods for this type of simulation, one can
carry out the calculation over very long periods. Smitz (1975) proposed this
solution for the Meuse River, for which he simulated the thermal regime for
the whole year, with 24-hour time steps. He took into consideration two
phenomena that have been neglected in the models previously described: heat
exchange with the river bed; and longitudinal dispersion.
Smitz showed that in the calculation of the natural temperature of a stream,
the first of these terms plays an important role, whereas longitudinal dispersion
can almost always be neglected. This does not apply, however, to the calculation
of the temperature of water influenced by a time-varying heat discharge. Heat
exchanges with the ground were calculated by a far-field method based on
simple conduction in the soil (Smitz, 1975; Jobson, 1977).
During the one-year application of a nonsteady temperature model to the
Meuse, Smitz simulated a great number of hydrometeorological conditions.
Thus, it was possible to make a statistical estimation of temperature increases.
158
Smitz did not proceed further with calibration of his model, which gave a good
agreement between calculations and measurements for the year studied.
A similar approach was used by Gras and Jacquet (1971), Gras and Martin
(1976), and a working group of the Commission Internationale pour la
Protection du Rhin Contre la Pollution (1973- 74). The calculation technique
is very similar to that of the Smitz model but has the following differences.
(i) It permits simulation over a great number of years (between ten and
twenty).
(ii) It is possible to adjust the exchange coefficients through comparison of
calculated and measured temperatures for one year. The adjustment
criteria are based on the annual average deviation and the typical
deviation of the daily differences between calculated and measured
values. Then, this adjustment is verified for another year.
(iii) A "loop" term is introduced to take into account globally the
phenomena that cannot be estimated separately:
exchange of heat with the ground;
water inputs at different temperatures from that of the stream (from
underground waters, small tributaries, etc.);
heat inputs from turbulence, navigation, internal chemical reactions,
and small industrial or urban discharges; and
meteorological data that are poorly representative of the larger data
set, especially the more or less systematic (seasonal) deviations
between the air temperature and humidity over the water surface
and the corresponding observations at the meteorological station.
The loop term is also estimated by adjustment between the calculated
and measured temperatures.
(iv) Since the hydrometeorological sample is very important, the results
obtained can be examined statistically.
(v) Since the number of cases studied is especially large, it is possible to
introduce other considerations into the simulation. For example, one
can take into account the random character of heat discharges from
power plants.
On the other hand, this approach does not deal with longitudinal dispersion.
Neglecting this phenomenon can introduce appreciable errors in the calculation
of instantaneous values or of temperature increases in sections close to the heat
injection point. In practice, however, it does not influence results that are
treated statistically.
Gras (1970) used this method to generate time series of natural temperatures
in those sections of the stream that were sufficiently far from the source. It is
sufficient for this purpose to integrate (5.3) stepwise, considering the water mass
to be stationary. He demonstrated that, independently of the initial temperature
159
taken, T converges rapidly toward the same temperature T(t), which is the
natural temperature of a water mass of average depth H influenced by local
meteorological conditions.
This method is very convenient for completing an insufficient set of data.
For example, it may be used to calculate the temperatures of tributaries or dis-
charges for which no measurements are available or to construct the temper-
ature time series that serves as an input to a model.
5.6. DATA REQUIRED FOR IMPLEMENTATION
The collection of necessary data for the application of a model constitutes an

essential step in its implementation. Very often difficulties arise that require
large amounts of data to resolve them, especially for simulation models of
nonsteady state conditions.
5.6.1. Simplified Models for Steady State Conditions

These models are used most often in specific case studies where the choice of
data is more or less arbitrary. The stream to be studied is divided into reaches
delimited by special sections, such as the point of heat injection, the confluence
with another stream, or the point where the temperature increase or the water
temperature itself is stipulated (e.g. a potable water intake).
The equation of temperature evolution can be written in either of two forms:
(-Kt)
T - 7:e = exp _ _ (-KIt)
T - 7:n = exp _ _ (5.7)
To - 7;, pcH ' To - Tn pcH
or
(-KA)
T - 7:e _ e x p - - T - 7:
To - i= exp
(-KIA)
pcQ . (5.8)
To - 7;, - pcQ'
In practice, the second expression, using the water surface area A (defined
reach by reach) and the flow Q, is much more convenient. The exchange co-
efficients K and K ' can be taken from the literature. Gras and Jacquet (1971)
have provided a convenient diagram for estimation. Sweers (1976) formulated
K as a function of T. The equilibrium temperature 7;, or the water temperature
Tn, as well as the stream discharge Q, is arbitrarily chosen depending on the case
studied.
5.6.2. Simulation Models for Nonsteady State Conditions

Whichever specific model is chosen, nonsteady state simulation requires the
collection of a considerable amount of data. Calculations are invariably made
with very small time and space steps. The geometry of the stream as a function of
160
the stream discharge has to be known with good precision. In addition to the
length of the reach, which is specified before constructing the model, at least two
of the following parameters have to be known:
average depth H = H(Q)

average velocity v = v(Q)
average width of reach / = /(Q).
The stream discharge is given by
Q = v/H.
In practice, there are no great difficulties in collecting data for these param-
eters, but results are not particularly sensitive to errors in the parameters.
If necessary, missing data can be supplied by interpolation, correlation, or
regression techniques.
To simulate the thermal regime it is necessary to define time series of water
temperatures, stream discharges, and representative meteorological data.
Temperatures of the stream at the head of the first reach studied and the
temperatures of effluents, especially the main ones, must be specified. Often
these time series are not completely available. Gras (1970) proposed a method
for generating the series from only meteorological data, by considering only
the natural temperatures of the stream in sections that were relatively distant
from the source.
Stream discharges must include discharges for each of the reaches and the
principal tributaries.
Representative meteorological data include, as a minimum, wind speed, air
temperature, humidity, atmospheric pressure, and solar and atmospheric
radiation. The time series collected through the established meteorological
network are generally sufficient. In practice, it is advisable to limit the number
of meteorological stations used, even for very long reaches. For example, for a
study of a 700 km reach of the Rhine, initially seven meteorological stations
were taken into consideration; but it was found that this number could be
reduced to three without affecting appreciably the quality of the results. There-
fore, it is essential to verify whether the meteorological stations selected are
indeed representative.
For models that do not proceed to a calibration stage, the choice of meteorolo-
gical stations has to be made very carefully (Smitz, 1975). The quality of the
results depends completely on selection of the most appropriate stations.
For models that are to be calibrated, the adjustment of evaporation co-
efficients can compensate for differences between meteorological parameters,
especially between the wind velocity at the water level and the wind velocity
at the meteorological station, where these values are sufficiently well correlated.
161
If tI, and tim are the average wind velocities at the water level and as measured
at the meteorological station, respectively, and e(tI,) is the evaporation co-
efficient as a function of the wind velocity, then the necessary adjustment
equations are
(5.9)
and
e(tI,) = a + btl"
where a, b, and f are constants to be adjusted. It follows that
e(tI,) = a + bextl m + b(, (5.10)
where a, b, ex, and f are constants to be adjusted. The adjustments of a, b, and ex

will take into account explicitly the relationship between tim and tI,.
Direct measurements of solar and atmospheric radiation are very often
deficient or lacking altogether. Calculating methods allow estimation of these
fluxes from other meteorological data, measured or estimated at the ground
station. These usually include dry and wet bulb air temperatures, humidity,
cloud cover or duration of radiation, wind velocity, atmospheric pressure, and
sometimes pan evaporation rates. This subject has been discussed by Gras
(1970), Wunderlich and Hsieh (1971), and Klein and Momal (1979). The
appendix summarizes a set of heat flux equations of this type developed by the
Tennessee Valley Authority (1972).
At this point, attention is focused on the quality of the data assembled, which
is often questionable. The data must be analyzed before they are introduced into
a model. Experience shows, for example, that it is rare for temperature time
series not to be affected by important errors. Certain errors can be detected
easily, such as those originating from poor measurement techniques or scale
errors that occur in carrying forward the values read. Other errors can be found
only by careful study of the time series, especially systematic errors caused by
a change of instruments or use of a nonrepresentative measurement station.
The direct comparison of temperature time series can help to detect these
errors, but often the model itself has to be used to complete this data analysis.
On these occasions it is often possible to reveal the influence of particular
phenomena. An example might be a temperature measurement station under
the influence of a heated water discharge, even though it is a significant distance
downstream from the injection point. Another might be a measurement station
upstream of a thermal power plant where hot water is recycled sporadically.
Measurement stations under the influence of a single tributary or heat source
may reveal systematic differences that can be discovered only by careful probing
of the time series.
162
5.6.3. Choosing a Model
The choice of a model depends, of course, on the problem to be solved. Local
problems, e.g the temperature field in the locale of a water discharge or the
risk of recirculating water between intake and discharge, will be handled by
models adapted to mixing zones. When problems of the thermal regime of a
stream have to be studied, one-dimensional models, which consider the water
to be homogeneous in temperature throughout the cross section, have to be
used.
It appears that, of the methods available, nonsteady state modeling permits
solution of the greatest variety of problems. Nevertheless, steady state solutions
are often useful for preliminary studies and for checking more costly nonsteady
state model results.
5.7. CALCULAnON TECHNIQUES
Two general methods of computation, describing the thermal regime in a river,

are indicated. The exponential method gives a broad, generalized treatment of
the heat balance; and the simulation method provides considerable detail in
space and time, even to the point of facilitating statistical treatment of results.
Implementation of models on the macrolevel with the exponential method
does not present any particular difficulty. It can be applied with a simple pocket
calculator. The simulation method, on the other hand, requires access to a
computer. Calculations are generally independent of the model used, although
in simulations using several years of data the results of calculation may be so
voluminous as to require storage on tape or disk for later statistical analysis.
When the simulation technique is employed for large and complex river
systems, like the Meuse, it is necessary to continually update the time series of
temperatures at intervals corresponding to the selected time step. The technique
for this calculation will now be described briefly.
The temperature profile along a stream is T = f(x) at time t, as shown in
Figure 5.1. Each section of water moves downstream at a velocity u 1(x) and,
warming in the process, creates a new temperature profile at t + M. Two time
series, for t and t + /:"t, are thus defined at intervals of u, (x)M along the x axis.
Intermediate values of the temperature at particular stations (x s ) are inter-
polated between adjacent members of the same series with respect to either
distance or time, depending on which suits better the requirements of the
problem.
If the flow changes from Q, to Q2 in the interval from t + /:,.t to t + 2/:"t, one
can interpolate on the temperature profile for t + /:,.t at intervals of U 2 (x)/:"t to
obtain a new time series as shown in Figure 5.2. This can, in turn, be compared
with the series representing the state of the system at t + 2/:"t, and so on, follow-
ing the procedure outlined in Figure 5.1. The method allows calculation of
163
B
.f:r==--------r--
---:;"'/~
------ .....
/ ..... -- ....
U T (1) .----------- r - /~ !'.T.
~' ------------ - ~ ~ i
" .... T, (0) - - - - - - - - - ....,~"----------+_---'----
X, x2 xi
FIGURE 5.1 Temperature profile change along stream, under steady state conditions.
A and B are the profiles at ( = 0 and !'.t, respectively. /iTj is the temperature increase
during flow Q(I) from Xl to X 2 in time /it. ll(t) = Q(t)/i(A,(O) + A 2 (O».
Pro fie
I at
_-9- t
-- --
= 2M
Interpolated
temperature
~
-;::;
T,(2)
T,( 1)
- - - - - - -
\--
-~
~
.----
------
---- - - -
!'.7j
- ---
T,(O) - - --- -
!'.x=u(2)M
x, x, +!'.x x2 xi
FIGURE 5.2 Adjusted temperature profile after change in rate of flow. /iTj is the tem-
perature increase during flow Q(2) from Xl to Xl + /ix.
temperature time series for the entire system with a constant time interval At
(taken as 3 h in the Meuse study) between values.
5.8. MODEL ADJUSTMENTS AND SENSITIVITY ANALYSIS
After critical analysis of the water temperature and hydrometeorological data,

it is the usual practice to adjust the empirical coefficients used, i.e. to make a
calibration. For macroscale models, this adjustment is most often made to the
164
evaporation and convection coefficients, e.g. in the Motor-Columbus model.
Also, it may be performed on the loop coefficient, which takes globally into
account the simplifications made in the model, such as neglecting heat exchanges
with the ground or heat inputs from minor sources. The loop coefficient may
also account for the fact that the meteorological data are not accurately repre-
sentative of the conditions prevailing at the stream, that is, the meteorological
station may be some distance from the stream and at a different altitude.
The justification for such an adjustment lies simply in improved simulation
results. Moreover, empirical evaporation formulas, abundant in the literature,
have generally been established by an analog adjustment between the observa-
tion station and the water surface. Therefore, evaporation coefficients used in
these formulas are not universal. It appears prudent, therefore, to carry out an
independent calibration and adjust the more important coefficients for each
new case (or model) studied.
The problem that remains is one of uniqueness. Will an adjustment made
for a given year remain valid for another? The answer can be found by testing
a year other than the one initially studied, as long as the system remains the
same and the meteorological station and measurement devices remain un-
changed. Experience shows that a model that is adjusted carefully for one or
two given years of records usually produces excellent results for all other years.
An anecdote illustrates the utility of the method. In a study by the
International Commission for the Protection of the Moselle River Against
Pollution, a thermal regime model had been adjusted. Applied for a time series
of 13 years, the model gave good results for the first 12 years but very poor
results for the last year. It was discovered that in order to produce reasonable
results for that year the wind speed would have to be multiplied by a factor of
about 2. A careful review of the data revealed an error of this order: the wind
speed had been expressed in meters per second rather than knots. Use of the
proper units produced a consistent simulation.
Calibration is most often carried out by trial and error when the number of
adjustable coefficients is small. Comparisons are made between observed and
simulated values of temperature time series with the objective of minimizing
the absolute value of the annual average difference between measured and
calculated temperatures, and of minimizing the sum of squared errors between
measured and calculated temperatures.
Chapter 11 provides additional discussion of the topics of sensitivity testing,
calibration, and validation.
5.9. MODEL APPLICATION

Implementation of a temperature simulation model will be illustrated by an
application to the Moselle River by Gras and Martin (1976). Their objective
was to evaluate the impact of construction of a nuclear power plant a few kilo-
165
meters upstream of the France-Luxemburg border, i.e. to determine whether
the resultant water temperature and temperature increase would conform with
the European standards for international streams. An additional consideration
was the plan of Luxemburg to construct a second power plant just downstream
of the border. The investigation was conducted in five phases.
Phase 1. Selection of Model

European standards for international streams include limiting the fraction
of the time for which the temperature standards may be exceeded. On the other
hand, the operator of the power plant delivering rejected heat to the stream can
limit the load of the plant for short periods. In order to make his operation
decisions, he must know accurately the risk he takes. The risk must, therefore,
be expressed by the mathematical probability of load reductions. The model
chosen for simulation of temperature in the stream must be sufficiently dis-
aggregated to permit analysis of extreme cases and to evaluate the probabilities
of their occurrence while at the same time calculating relatively modest changes
in temperature. Since one of the criteria for decision making is the value of the
extreme temperature, the model must not be biased toward the highest
temperatures.
Having considered all of these factors, the investigators studying the Moselle
chose a simulation model for non steady state conditions, which could be cali-
brated for the river and would provide statistical results.
Phase 2. Data Collection and Analysis

Two stations of the meteorological network near the Moselle, at Metz and at
Nancy, could have been used for the study. Calculation of the natural tempera-
ture using data from both stations revealed that the data from Nancy gave
somewhat better results. Two distinct periods seem to be evident from the data
of the Metz station. This anomaly was attributed to a change of meteorological
equipment at the station. Consequently, data from only the Nancy station were
retained for the study.
Daily records of flow were available for several locations along the reach
studied. Various time series of water temperature data, unfortunately quite
dissimilar in the locations and periods covered, were available from stations
at Millery (1973- 74), Metz (1970-73), La Maxe (1972- 73), Palzem (1959-62,
1971-75), Trier (1959-64), Cochem (1960-75), Miiden (1971-75), and Giils
(1972- 74). Since 1964 two relatively large power plants with once-through
cooling have been established on the Moselle (KKWI and KKW2, shown in
Figure 5.3). Their discharges disturb the water temperatures downstream, so it
was necessary to take into account these disturbances in order to study the
coherence of the temperature time series. It had to be established whether
there existed a systematic difference that could be attributed either to a real
166
se'
{) ~O GERMANY
Trier(J •
. ~
j
.I
(, (> Meteorological station
,. / • Temperature measurement
...... _.,
"' * Flow measurement
<> Central power plant
H Limits of reach
FIGURE 5.3 Sketch of the Moselle River basin, showing the division of the river into
reaches.
167
phenomenon, e.g. climatological modifications along the stream, the effect of a
tributary, etc., or to systematically incorrect measurements, and whether the
precision of measurements was acceptable in view of the natural dispersion of
temperature observations.
To simplify the calculations the Palzem station was chosen as a reference by
virtue of its favorable location near the border and about midway along the reach.
Observations at the other stations were correlated to those of Palzem over
their common periods. Analysis of these correlations revealed that the differ-
ences observed between the stations could, in large part, be explained by the
effects of the existing power plants. The temperature time series were considered
as coherent.
Phase 3. Development a/the Model

The Moselle was divided into reaches by taking as endpoints for the calcu-
lation sections: sites of power plants; confluences with important tributaries
(e.g. the Saar); stream discharge gauging stations; and temperature measure-
ment stations. Figure 5.3 shows how the division was made. For each reach,
H = H(Q) and v = v(Q), the average depth and average velocity as functions
of discharge, were determined.
Phase 4. Model Calibration

Minimum residuals between calculated and observed values of temperature
were obtained by successive adjustment of the evaporation coefficient e and
the loop coefficient a. The evaporation coefficient was defined a priori to be of
the form
e(u) = a(l + u)
for the whole set of reaches to which meteorological data from the Nancy
station were applied. Figure 5.4 shows the evaluation of these two parameters
considered in the adjustment. The minimum of (J/x = y corresponds closely to
the value of a for which dTm is zero, independently of the measurement station
considered.
Experience shows that the loop coefficient can be represented by a function
of the air temperature. In other words, it is a term of seasonal character that
makes adequate recognition of phenomena like heat exchange with the ground
or temperature inputs from small tributaries that are not otherwise explicitly
included in the model. For ease of calculation, the authors combined this co-
efficient with that of atmospheric radiation, which can also be expressed as a
function of the air temperature.
Figures 5.5 and 5.6 show comparisons between the calculated and measured
temperatures of the Moselle at the France-Luxemburg border for 1972 and
1973, respectively.
168
2.0
1.6
1.2
0.8
u
0
:::.. 0.4
II
:><
........
\::> 0
0.0030 a 0.0035
-0.4
-0.8
-1.2
FIGURE 5.4 Determination of the optimum of the evaporation coefficient

e(u)= a(l + u). The data are from stations at Millery (e), Metz (+), La Maxe (x), and
Palzem CA.).
Phase 5. Results of Simulation

Simulations of the Moselle to study the influence of power plants were
performed using data for the period 1961-73. The primary results of this
investigation comprised:
(i) time series of natural stream temperatures at different key stations;

(ii) time series of perturbed stream temperatures downstream of power
plants after mixing of cooling water discharges;
(iii) time series of temperatures of effluent discharges from power plants
situated along the Moselle;
(iv) time series of temperature increases above the natural state down-
stream of power plants and at the France-Luxemburg border; and
(v) distribution curves of these parameters, indicating the frequency of
occurrence of stipulated levels of temperature and temperature
increase.
169
28
x
•
~
24
rx Ix ~•
"xl ~ ;~
&~ ~!~
W"~ \J~ ~
x
20
x t.'~
~ ~l!
f V'1~
u "'l '1/ ~
~~f ,
0
U
0; 16
h
U
1 x * \
Ul
1:>
xx~
x II' ~~
~x
f:,
hO II I
X
~ r
12
"
x 'I~"
x >x'
~
l i
~~
~ I,
x .i1,w *~
x.~ ~ ~h
x\' ~ x ~ x
x~ *~
~ x
4 ~ ~
o 100 200 300 400
Days
FIGURE 5.5 Observed (broken line) and calculated (x) temperatures of the Moselle
River at the France-Luxemburg border in 1972.
It was possible to use this information to evaluate the effects of different

regulation hypotheses based on maximum temperatures and allowable in-
creases above the natural state. In the specific case of the power plant at
Cattenom (Figure 5.3) it was possible to make an important change in the
cooling system to relieve the thermal load on the Moselle. This plant will now
be equipped with cooling towers, the blow-down from which will be diverted
through a cooling pond of several hundred hectares before discharge to the
river. In calculation of the operation of the cooling towers, the meteorological
conditions have been simulated, as well as the thermal behavior of the cooling
pond.
170
28
20
u
0
(.)
m
1--.(.)
.... 16
..c
0
1--.
o 100 200 300 400

Days
FIGURE 5.6 Observed (broken line) and calculated (x) temperatures of the Moselle
River at the France-Luxemburg border in 1973.
APPENDIX. SURFACE HEAT EXCHANGE
Heat exchanges through the air-water interface depend upon both the internal
hydromechanical behavior of the water body and the physics of its interaction
with the overlying air mass. Meteorological factors such as solar radiation,
wind, humidity, pressure, and cloudiness figure prominently in the many
physical processes involved. The aggregate effect of the most important of
171
these processes is represented in the general heat flux equation:
(AS. I)
where
qn is the net energy flux across the air-water interface [J L - 2 T- I],
qsn is the net solar radiation flux,
qat is the net atmospheric radiation flux,
qws is the water surface radiation flux,
q. is the evaporation heat flux,
qh is the sensible heat flux.
Among the early investigations to formulate explicitly the independent
energy flux terms, one of the most comprehensive was undertaken in the late
1960s by the Tennessee Valley Authority Engineering Laboratory. This led to
publication of Report 14 (TVA, 1972), from which the following formulations
have been derived.
Net Solar Radiation Flux qsn

The net solar radiation flux is the residual flux through the water surface of the
solar radiation that is received at the top of the atmosphere and reduced through
attenuation by the atmospheric column, interception by clouds, and reflection
at the water surface:
(AS.2)
where
qo is the solar radiation flux on a horizontal plane at the top of the atmos-
phere [J L -2 T- I],
f(A) is the atmospheric attenuation factor, a function of the optical properties
of the air column, dust content, moisture, and terrestrial surface re-
flectance (Klein, 1948),
Rt is the reflectivity of the water surface (R t ~ 0.03),
C is the degree of cloudiness (0 < C < 1.0).
The solar radiation flux at the top of the atmosphere at a particular location
and time is given by
qo = (lo/r 2 ) sin IX, (AS.3)

where
lois the solar constant, equal to the solar radiation incident normal to the top
of the atmosphere when the earth is at its mean distance from the sun
[J L -2 T- 1 ] (10 ~ 21y min-I),
172
r is the radius vector, i.e. the ratio of the actual earth-sun distance to the
mean distance (0.9833 < r < 1.0167 and 0.9669 < r 2 < 1.0337),
LX is the solar altitude (radians), a function of latitude, and of the declination
and local hour angle of the sun (0 < LX < n/2).
Net Atmospheric Radiation Flux qat

The atmospheric radiation flux is the net longwave flux from the atmospheric
air mass, including clouds, after reflection from the water surface. It is a function
of absolute air temperature, cloudiness, and water surface reflectivity:
(A5.4)
where
qat is the longwave atmospheric radiation flux from a cloudy sky [J L - 2 T- 1 ],
Cat is an empirical coefficient: 0.906· 10- 5 < Cat < 0.999· 10- 5; Swinbank's
coefficient is 0.938 . 10 - 5,
(J is the Stefan-Boltzmann constant, 2.041 . 10 - 3 J m - 2 h - 1 K - 4,
T2 = 273.2 + Ba2 is in kelvins; Ba2 is the dry bulb air temperature 2 m above
the ground (0C).
Water Surface Radiation Flux, qws

The water surface radiation flux is the longwave flux [J L- 2 T- 1 ] from the
water mass emitting as a black body at a specific surface temperature. It is a
function of water surface emissivity and surface temperature:
(A5.5)
where
(w is the emissivity of the water surface, i.e. the ratio of the radiation emission
of the water surface to that of a perfect black body; (w ~ 0.96 (dimension-
less),
Ta =273.2 + Ba is in kelvins; Ba is the average water surface temperature (Oe).
Evaporation Heat Flux qe

Heat is lost from the water surface in the form of the latent heat of evapora-
tion:
qe = - Pw L E, (A5.6)
where
Pw is the density of water [force· T 2 L - 4 or M L - 3]: Pw = 998.2 kg m - 3 at
20°C,
L is the latent heat of evaporation [cal M - I] : L = 597.1 - 0.57 Bo (cal g - I),
E is the evaporation rate [L T- I].
173
The TVA Engineering Laboratory study provides an extensive review of
various evaporation formulas, concluding that the Marciano- Harbeck formula,
derived from detailed investigation of evaporation on Lake Hefner in Oklahoma
(Marciano and Harbeck, 1954), probably gives the most consistent practical
results from standard weather observations. This formula (converted to metric
units) is
E = 4.33.10- 5 u(e o - ea)' (A5.7)
where
E is the evaporation rate (mm day - 1),
U is the wind velocity (km h - 1),
eo is the saturation vapor pressure at the water surface temperature (mm Hg),
ea is the vapor pressure of air (mm Hg).
Adjusting for observations made at specific elevations above the water surface
yields
(A5.8)
where the subscripts 2 and 4 refer to observations made at 2 and 4 m above the
surface.
An alternative form of the Lake Hefner expression that takes account of
stability of the overlying air mass is
E = N(PIPo)u(C o - Ca)f(Ri), (A5.9)
where
E is the evaporation rate (m h - 1),
N is the evaporation coefficient (dimensionless),
P is the atmospheric pressure (mbar),
Po is the standard atmospheric pressure at sea level, 1013 mbar; PIP 0 may
be approximated by PIP o = (l - 0.0226z)5.256 with z in kilometers
above mean sea level,
u is the wind speed (km h - 1),
Co = 0.622 eolP is the water vapor concentration of the saturated air at
the water surface (kg vapor/kg air),
eo is the vapor pressure of saturated air (mbar),
ea is the vapor pressure of air (saturation vapor pressure at dewpoint)
(mbar),
f(Ri) is a function of the Richardson number:
(l - 22Ri)o.80 for 0 ~ Ri ~ -I
f(Ri) = (I + 34Ri)-o.8o for 0 ~ Ri ~ 2
{
1 for Ri = 0 (neutral case)
· __ g(Pa - Po)z
R1 - 2'
Pa U
174
where
g is the acceleration due to gravity [L T- 2],
Pa is the density of air (usually at z = 2 m) [M L - 3],
Po is the density or saturated air at the water surface temperature [M L - 3],
U is the wind speed (usually at z = 2 m) [L T- 1],
Z is the height above the water surface [l].
Values of N in (A5.9) depend on particular conditions of meteorological
observations:
for airport data,
N = 1.17.10- 6 , z = 5.5m;
for land/lake data,
N = 1040.10- 6 , z = 2m;
P and ea observed at an upwind land station are unaffected by lake evaporation;
for lake data,
N = 1.54.10- 6 , z = 2m.
Sensible Heat Flux qh

Sensible heat is transferred between air and water by conduction and trans-
ported from or toward the air-water interface by convection associated with
the moving air mass:
(A5.1O)
where
qn is the forced convective sensible heat flux [J L - 2 T- 1],
eo is the water surface temperature caC),
ea is the dry bulb air temperature caq,
Cp is the specific heat of air at constant pressure [J M - 1 °C- 1].
Other Heat Losses

In addition to the five principal heat fluxes cited in (A5.1), heat may be
transferred to the water by condensation from an overlying supersaturated air
mass or into or out of the system by advective transfers associated with precipi-
tation, evaporation, and/or condensation. The condensation flux is of the same
general form as the combination resulting from (A5.6) and (A5.9) for evapora-
tion, except that C a and Co are interchanged and C a is required to be less than
the saturation vapor concentration of the air. The heat gain by condensation
is usually relatively small, so it is often neglected in energy calculations or con-
175
sidered as a part of the net evaporative heat loss. Heat exchanges associated
with advective transport of water through the interface are usually neglected
except in the most detailed studies.
REFERENCES
Commission Internationale pour 'Ia Protection du Rhin Contre la Pollution (1973- 74). Plan de
charge thermique du Rhin. Rapports Internes tlu Sous-groupe .. Eaux de Refroitlissemenl dli (Jroupe
de Tramil T.
Fleig, H., and D. Flinsbach (1973). Wiirmelastplan Neckar. Ministerium fiir Erniihrung. Land-
wirtschaft. und Umwelt, Baden- Wiirttemberg.
Gras. R. (1970). Estimation des elements du bilan radiatifpour Ie calcul des temperatures naturelles
d'un secteur d'un cours d'eau. Electricite de France Direclion des Etudes el Recherches, Centre de
Chatou Rapport HF 41-70. No. 26.
Gras. R., and J. Jacquet (1971). Simulation du comportement d 'un reseau hydrographique equipe
d'un ensemble de centrales thermiques, application au bassin de la Seine. Xles Journees de
I'Hydraulique Paris, 1970, Hydrotechnil'Ol Society of France, Question IV Rapport II, p. 10.
Gras, R., and P. Martin (1976). Modele de simulation du comportement thermique de la Moselle.
Electricitede France Direction des Ellides el Recherches, Cenlrede Chalou Rapport E 31-76, No. 18.
Hydrologisches Institut Miinchen (1976). Wiirmelastplan Donau. Bayrisches Landesamt fiir
Wasserversorgung und Gewiisserschutz.
Jobson, H. (1977). Bed conduction computation for thermal models. Proceedings of American
Sociely or Ciril Enqineers, Journal of Hydraulics Dil'ision I03(HY 10): 1213-1237.
Kahlig, P. (1977). Zur Exponentialmethode del' Flul3temperaturberechnung. Archirfur Meleoro-
logie, Geophysik !Jnd Biologie. Serie A 25: 357-365.
Kaisers6t. K., and H. Mitschel (1977). Ein Beitrag fiir die Berechnung des Wiirmeaustausch-
koeffizienten bei del' Abkiihlung eines kiinstlich erwiirmten Flu13es. BrennstofJ- Wiirme-Krafi
29(6): 250-255.
Klein, P., and P. Morna! (1979). Presentation et mise en oeuvre numerique d'une methode d'esti-
mation du rayonnement solaire. Ele£'lricile de France Direclion des Eludes el Recherches, Cenlre
de ChalOu Rapporl E 31-79. No. 15.
Klein, W. H. (1948). Calculation of solar radiation and the solar heat. Journal or Meleorology 5,
NO.4.
Liinderarbeitsgemeinschaft Wasser (1971). Grundlage fur die Beurleilling del' Wiirmebelaslling !'On
Gewiissern (Wiesbaden: Kochler und Hermann).
Marciano, T. T .. and G. E. Harbeck. Jr. (1954). Mass transfer studies. WaleI' Loss Im·esl(qation.
Lake Herner Sludies, Technical Reporl, Uniled Stales Geological Sun'ey Professional Paper 269.
Raphael, J. M. (1962). Prediction of temperature in rivers and reservoirs. Proceedings of American
Societv or Cil'il Engineers. Journal or POIIW Dil'ision 88(P02): 157-181.
Smitz. J. S. (1975). Modele mathematique de la temperature naturelle des rivieres. Bulletin of Royal
Socielyfor Science, Be(qium.
Sweers. H. E. (1976). A nomogram to estimate the heat exchange coefficient at the air-water interface
as a function of wind speed and temperature, a critical survey of some literature. Journal or
Hydrology 30: 375-40 I.
Tennessee Valley Authority (1972). Heat and mass transfer between a water surface and the atmos-
phere. TVA. WaleI' Resources Research Engineering Laboralory, Norris, TN Reporl 14.
Wunderlich, W.O., and L. W. Hsieh (1971). The computation of daily mean stream temperatures.
TV A WaleI' Resources Research Enqineering Laboratory, Norris. TN Report 25.
6 Stream Quality Modeling
6.1. INTRODUCTION
The achievement of regional water quality goals, especially in the more de-
veloped areas of the world, often involves substantial capital investments and
changes in public attitudes concerning resource management. Economic
impacts may include not only the cost of facilities designed to reduce the dis-
charge of contaminants into natural waters or to improve the quality of waters
receiving waste, but also any limitations on economic development in a par-
ticular region or river basin. Those responsible for the formulation and approval
of water quality plans or management policies must have a means of estimating
and evaluating the temporal and spatial economic, environmental, or ecological
impacts of these plans and policies. This need has stimulated the development
and application of a wide range of mathematical modeling techniques for pre-
dicting various impacts of alternative pollution control plans and policies.
There are many different types of stream quality model. The appropriate
model and the required data depend on the purpose of the specific study.
Long-range regional water quality planning does not require the detail that is
appropriate, for example, when evaluating a single proposed industrial waste
outfall or discharge site. There is no best single water quality model for all
streams and for all planning situations. An important decision that must be
made early in the planning process is the selection of the modeling method or
methods appropriate for planning and capable of development, calibration,
validation, and execution within the limits of available time and money.
Most stream quality predictive models in use today apply to streams re-
ceiving wastewater from point sources. As the quantities of wastes discharged
from point sources are reduced, nonpoint or distributed sources of wastewater
from agricultural and urban runoff become increasingly important. Models are
176
177
needed to help predict nonpoint source waste inputs to surface waters. The
outputs of these non point source wastewater generation models provide the
inputs to water quality models of the receiving stream. For example, urban
stormwater management models are often integrated from various models for
runoff, sewer routing, and prediction of the quality of the waste-receiving water.
This chapter reviews a number of the more typical stream quality predictive
models developed for and applied to waste-receiving streams. The models
range from the simple to the complex, yet each has proven effective in certain
planning situations. The inclusion of various water quality management al-
ternatives, and their costs, within these predictive models will be reviewed in
Chapter 12.
6.2. TYPES OF STREAM QUALITY MODEL
It is useful to distinguish between certain types of model and to discuss their

characteristics. This will provide an opportunity to define a few of the terms and
to illustrate some of the concepts used by those who develop and apply water
quality models. Many ofthe models in use are extensions of two simple equations
proposed by Streeter and Phelps in 1925 for predicting the biochemical oxygen
demand (BOD) of various biodegradable constituents, and the resulting dis-
solved oxygen concentration (DO) in rivers (Thomann, 1972). Often used with
these BOD-DO models are other fairly simple first-order exponential decay,
dilution, and sedimentation models for additional nonconservative and
conservative substances.
More complex multiconstituent water quality models have also been
proposed and have been applied to predict the physical, chemical, and biological
interactions of many constituents and organisms found in natural water bodies.
These multiconstituent simulation models generally require more data and
computer time, but they also can provide much more detailed and compre-
hensive information on the quantity and quality of water resulting from various
water and land management policies.
Water quality models can be used to evaluate steady state conditions, fdr
which the values of the water quality and quantity variables do not change with
time. They can also be used to evaluate dynamic or time-varying conditions.
The latter type of model permits an evaluation of transient phenomena such as
nonpoint stormwater runoff and ~pills of pollutants. Steady state models are
usually simpler and require less computational effort than dynamic or transient
models, and are more relevant to long-term planning than to short-term
management and control.
Assumptions pertaining to the mixing of pollutants in water bodies dictate
the spatial dimensions of the model. Sufficient accuracy may be obtained in
many river systems by modeling only one or two dimensions. One-dimensional
178
models of river systems assume complete vertical and lateral mlxmg. One-
dimensional lake models usually assume complete mixing in all but the vertical
direction. Two-dimensional models may assume either lateral mixing, as in
stratified estuaries or lakes, or vertical mixing, as in relatively shallow and wide
rivers.
Undoubtedly, the most data-demanding model type is the stochastic or
probabilistic model as compared with its deterministic counterpart. Most
deterministic models yield estimates of mean values of various quality con-
stituents, whereas probabilistic models explicitly take into account the random-
ness or uncertainty of various physical, biological, or chemical processes.
Validation of stochastic models is especially difficult because of the quantity of
data on the water body that is necessary for comparing probability distributions
of variables rather than just their expected or mean values. This introduction
will be confined to a review of one-dimensional deterministic models of steady
and unsteady state flow. These relatively simple models are used for long- and
short-term planning, management, and control of water quality during periods
when complete mixing exists in the other two dimensions and flows are constant.
6.3. DISSOLVED OXYGEN AND BIOCHEMICAL

OXYGEN DEMAND
In connection with a study of the Ohio River, Streeter and Phelps (1925) de-
veloped the first important water quality model describing the BOD-DO
relationship in a stream. In their pioneering work the simplest system was
considered, in which biodegradable waste is discharged to the stream and
consumes oxygen, atmospheric reaeration being the only source of oxygen.
The rate of change in the dissolved oxygen deficit, dD/dt, was assumed to be
directly proportional to the unsatisfied oxygen demand and to the oxygen
deficit in the stream. Therefore, the differential equation for the process can be
written as
(6.1)
where
D is the oxygen saturation deficit [M L - 3] or the difference between the DO

saturation concentration and the concentration at time t [T],
L is the carbonaceous biochemical oxygen demand BODe [M L - 3],
Kd is the deoxygenation rate coefficient [T- 1],
K. is the reaeration rate coefficient [T- 1 ].
179
The solution to (6.1) is
KL
Dr = d 0 [exp( -Kdt) - exp( -K.t)] + Do exp( -K.t) (6.2)
K. - K d
or
(6.3)
where
Dr is the dissolved oxygen saturation deficit after time t,

L o is the oxygen demand at an initial reference time (t = 0),
Do is the dissolved oxygen deficit at t = 0,
kd = 0.434K d , k. = 0.434K•.
Equation 6.2 or 6.3 is the Streeter-Phelps "oxygen sag formula," and a
profile of DO along the stream is referred to as a "dissolved oxygen sag curve"
(Figure 6.1). With Do, L o , K d , K. known the deficit Dr may be computed at
any time t, and when Do, Dr' K d , L o are given (6.2) or (6.3) may be solved for K a •
In many cases, the main interest is directed to the critical point, which is
determined by the critical deficit Dc and critical time t c' At this point there is no
Waste discharge point.

BOD of stream = L o DO saturation
concentration Cs
;;-----r------i-----------
Dc Dr
dD
OJ
E
o
o
o t Time (days)
FIGURE 6.1 Dissolved oxygen sag curve.
180
change in the deficit (reaeration balances deoxygenation):
dD
dt = KdL - K.D = O. (6.4)
Therefore, the critical deficit is

Kd K
Dc = - L = - d L o exp( -KdtC> (6.5)
K. Ka
or
kd
Dc = k LoIO-kdfc, (6.6)
a
and the critical time can be expressed as
(6.7)
Fair (1939) introduced a new constant f = k./k d , called the self-purification

factor, and defined the critical time by
tc = ki/- I) Ig[f(1 - (f - 1) ~:) J (6.8)
and the critical deficit by
(6.9)
These relationships were used for studies of the self-purification of natural

streams. t was taken to be the time of flow. The dissolved oxygen sag equations
are also useful in analyzing the influence of various parameters.
The Streeter-Phelps model is based on a constant pollution load discharged
at a single given point along a stream having a constant flow rate and a uniform
cross section. The lateral and vertical concentrations of oxygen and BOD are
assumed uniform throughout any cross section. Other basic assumptions are
that the deoxygenation and reaeration are first-order reactions, that the reaction
rates are constant, and that the net change of oxygen deficit is a function only of
deoxygenation and atmospheric reaeration by gas absorption through a liquid-
air interface. While in many cases the kinetics of the BOD reaction can be better
described by a second-order model, the assumption of first-order kinetics is
reasonably accurate for many streams. It is also convenient because it describes
a linear system that may be readily analyzed by many techniques.
In addition to the bacterial oxidation of the organic matter and the atmos-
pheric reaeration considered in the Streeter-Phelps model, numerous processes
181
are taking place in natural streams. Dobbins (1964) summarized some of these
processes:
(I) the reduction of BOD by sedimentation or adsorption;

(2) the increase of BOD from scoured bottom deposits or from the dif-
fusion of partly decomposed organic products from the benthic layer
into the water above;
(3) the increase of BOD along the stream from runoff;
(4) the removal of oxygen from the water by diffusion into the benthic
layer to satisfy the oxygen demand in the aerobic zone of this layer;
(5) the removal of oxygen from the water by the purging action of gases
rising from the benthic layer;
(6) the addition of oxygen by the photosynthetic action of plankton and
fixed plants;
(7) the removal of oxygen by respiration of plankton and fixed plants;
(8) the continuous redistribution of both BOD and DO by longitudinal
dispersion.
Frankel and Hansen (1968) stated that the following factors should also be
considered:
(9) the variation of K d with time, particularly at the onset of nitrification,

which precludes assuming K d to be constant for larger values of t (time
of travel);
(10) changes in channel configuration that alter the characteristics of
surface turbulence and consequently the rate of transfer of oxygen
from the atmosphere;
(II) the effects of suspended and dissolved substances upon the rate of
diffusion of oxygen from the surface into the main body of the stream;
(12) diurnal variation in oxygen content, BOD, temperature, and flow
rate of influent discharges.
Various modifications of the Streeter-Phelps model have been proposed to

take into account some of these processes. Such models, used to describe the
BOD profiles along a river reach, have been based on various assumptions.
A brief summary of some of these models is presented below.
Thomas (1948) pointed out that part of the BOD can be removed by sedimen-
tation without consumption of the dissolved oxygen, and that the removal rate
is directly proportional to the remaining BOD. Therefore,
dL
CIt = -(K d + Ks)L, (6.10)
182
where K s is the rate constant for BOD removal by sedimentation [T- 1 l Then
(6.11 )
and
KdL a
Dt = K _ (K + K ) {exp[ -(K d + Ks)t] - exp( -Kat)} + Daexp( -Kat).
a d s
(6.12)
In this case, the amount of bottom sediment is assumed to be small and the
reaction rate so low that no measurable oxygen is drawn from the sediment;
photosynthesis is not considered.
Camp (1963) proposed that the BOD and DO profiles can be described by
the following differential equations:
BOD profile,
dL
-(K d + Ks)L + B; (6.13)
dt
DO profile,
dD
(6.14)
dt
where
B is the rate of addition of BOD to the overlying water from the bottom
deposits [M L -3 T- 1],
P is the rate of oxygen production in the euphotic zone by photosynthesis
[M L -3 T-1l
The solutions of these equations are
L, = (La - K B K )ex P [ -(K d + Ks)t] + K B K (6.15)

d+ s d+ s
and
Dr = K d + K) ( La - K
K _ (K B)
+ K {exp[ -(Kd + Ks)t] - exp( -Kat)}
ads d s
+ K-Kd (B K P)
- -K [1 - exp(-Kat)] + Daexp(-Kat). (6.16)
a Kd + s d
If the BOD is added to the overlying water from the bottom sediments, and the
reduction of BOD by settling is negligible, these equations can be presented as
(6.17)
183
and
Dr = Kd
Ka - Kd
(L O - ~)[(exP(
K
-Kdt) - exp( -Kat)]
d
B-P
+ - - [1 - exp( - Kat)] + Doexp( -Kat). (6.18)
Ka
If BOD is reduced by settling, but benthic demand has no measurable effect
on the oxygen deficit of the stream, then (6.16) becomes
KL
D, = K _ (K: ; K ) {exp[ -(K d + Ks)t] - exp( - Kat)}
a s
P
-j([l-exp(-Kat)]. (6.19)
a
If K s ' B, and P are zero, (6.18) becomes the original Streeter-Phelps oxygen
sag equation.
In addition, Camp (1963) considered a stream with significant longitudinal
mixing and in which BOD and DO do not vary with the temporal mean depth.
An element of unit width, length, and depth was considered, and the equations
for the BOD and DO profiles in the x direction at steady state were written as:
BOD profile,
(6.20)
DO profile,
d2C dC
E-
dx2
- U-
dx
+ Ka
(C - C) - K d L + P = 0
s '
(6.21)
where
E is the turbulent transport (longitudinal mixing) coefficient [Ll T- 1],
U is the average stream velocity [L T- 1 ],
C is DO [M L - 3],
C, is DO at saturation [M L ~ 3].
O'Connor (1962) assumed that in natural streams mass transport by turbu-

lent diffusion (i.e. longitudinal mixing) is insignificant. Assuming only de-
oxygenation by organic matter oxidation and atmospheric reaeration, he
defined the nonsteady state distribution of dissolved oxygen by
(6.22)
184
Under the steady state condition 8C/8t = 0, (6.22) reduces to
dC
U-
dx - KdL - K •(Cs - C) = ° .
(6.23)
Dividing through by U gives

dC Kd K.
dx - U L +U (C s - C) = 0. (6.24)
If the variable L is expressed as a function of the downstream distance x, where

L = L o exp( - Krx/U), and K r is the coefficient of BOD removal in the stream,
which may be different from that of oxidation, K d , (6.24) may be integrated to
·L
C = Cs - ~ [exp( -jrx) - exp( -j.x)J - (Cs - Co)exp( -j.x). (6.25)
J. - Jr
The boundary condition for (6.24) is C = Co at x = 0, and the coefficients are
defined as ja = Ka/U, jd = Kd/U, jr = Kr/U. The units of the coefficients are
in terms of distance, i.e. L ~ 1.
The following assumptions were made by Dobbins (1964):
(1) The stream flow is steady and uniform.

(2) The process for the stretch is in a steady state; the conditions at every
cross section do not change with time.
(3) BOD removal by bacterial oxidation and by sedimentation and/or
adsorption are first-order reactions; the rates of removal at any cross
section are proportional to the BOD present.
(4) The removal of oxygen by benthic processes and by plant respiration, the
addition of oxygen by photosynthesis, and the increase of BOD from
the benthic layer or local runoff are all uniform along the stretch.
(5) The BOD, L, and DO, C, are uniformly distributed over each cross
section, thus permitting the equations to be written in the usual one-
dimensional form.
Using these assumptions, Dobbins derived a set of two differential equations

describing the BOD and DO profiles as functions of distance downstream:
BO 0 profile,
(6.26)
DO profile,
(6.27)
185
where
L add is the rate of increase of BOD along the stretch [M L - 3 T - 1],
Db is the rate of removal of oxygen caused by benthic demand and plants
[ML- 3 T- 1}
The Camp-Dobbins models take no account of the nitrification stage, since it

was assumed that all deoxygenation occurs as first-stage deoxygenation.
Hansen and Frankel (1965) suggested an important modification of Camp-
Dobbins models. They assumed that the classical pattern of diurnal DO
profiles in a stream can be represented by a periodic function. Therefore, if
factors involving BOD are neglected, (6.14) may be written as
(6.28)
in which Pm is the maximum rate of oxygen production/consumption by

photosynthesis/respiration [M L - 3 T- I} Equation 6.28 can be integrated to
Dr = [Do - a(K a cos ¢ + w sin ¢)]exp( -Kat)

+ a[K a cos(wt + ¢) + sin(wt + ¢)], (6.29)
in which a = Pm/(w 2 + K;), w is the frequency, and ¢ the phase. The maximum
or minimum deficit (depending on the sign of the function) is given by D m =
(Pm/w)sinex = (Pm/Ka)cosex,whereex = wt + ¢ = tan-1(w/K a ).
It was proposed that the maximum rate [M L - 2] of oxygen production/
consumption should be related to the solar energy input, the mean depth of
the stream, and turbidity. The rate of addition of BOD to overlying water
from the benthic layer, B [M L - 3 T - 1], was assumed to represent the rate of
addition through only physical mechanisms; therefore,
(6.30)
where
L d is the total areal BOD of the benthos [M L - 2],

H is the depth of the stream [L],
K b is the areal demand rate constant [T- I}
Integration of (6.30) results in
(6.31)
186
The rate of change of BOD from suspended and dissolved matter, L s , was
given by Frankel and Hansen (1968) as
dLs
Cit = -KL s + B, (6.32)
which can be integrated to
L (L - KB)exP(-Kt) + ~K'
s,
= So
(6.33)
where
K = K d + K r + K s [T-I],
Kd is the laboratory rate of BOD increase by deoxygenation,
Kr is the river rate of BOD increase by deoxygenation due to attached aquatic
growths and slimes,
Ks is the rate of BOD reduction by sedimentation and/or adsorption.
Also, when sedimentation occurs rather than scour, the fractional change in
suspended BOD that is due to sedimentation was expressed by the ratio K.IK,
and
1 dLd
(6.34)
H dt
The total oxygen deficit resulting from biological oxidation of the suspended
and dissolved BOD load, oxygen consumption by anaerobic decomposition
products of the benthos, "decay" of the initial oxygen deficit, and production/
consumption of oxygen by photosynthesis/respiration is expressed as
+ ~:)~ ;
K
Dr = (K d K.l ) (L so - ;)[exp( -Kt) - exp( -Kat)]
+ Kb dO
(L + ~)[exP( -Kbt) - exp( -Kat)]
Ka - Kb H Kb
+ [Do - a(K a cos ¢ + w sin ¢)]exp( - Kat)
+ a[K a cos(wt + ¢) + w sin(wt + ¢)]. (6.35)
O'Connor and Di Toro (1968, 1970) assumed that the temporal form of the
photosynthetic oxygen source in streams may be represented by a half-cycle
sine wave. The daily rate of photosynthetic oxygen production as a function of
time, pet), was expressed as
pet) = Pm sin[(n/p)(t - ts)] when ts s t S ts +P
(6.36)
pet) = 0 when Is + pst s ts + 1,
187
where
t, is the time, expressed as a fraction of the day, at which the source becomes
active,
p is the fraction of the day during which the source is active (period of sunlight).
A periodic extension of the temporal and spatial variations of photosynthetic
oxygen production was expressed as a Fourier series:
P(x, t) = P m ( ~
2P
+ Jl
00
b. cos[2nn(t - t s - pI2)] ,
)
(6.37)
in which
4nlp
b. = 2 2 cos nnp.
(nip) - (2nn)
The following equation for the DO deficit, D(x, t) = C s - C(x, t), was set
forth:
oD(x, t) Q oD(x, t)
---
at
- A ax - K.D(x, t) + KdL(x) + KnN(x)
+ Sex) + R(x) - P(x, t), (6.38)

where
A is the cross-sectional area [U],
Q is the flow rate [L 3 T- 1 ],
U = QIA is the stream velocity [L T- 1],
L(x) is the concentration of carbonaceous BOD at x [M L - 3],
N(x) is the concentration of nitrogenous BOD at x [M L - 3],
Kd is the deoxygenation rate coefficient of carbonaceous BOD [T- 1 ],
Kn is the deoxygenation rate coefficient of nitrogenous BOD, which
reflects both the removal of ammonia and the utilization of oxygen
[T- 1 ],
K. is the reaeration rate coefficient [T - 1],
Sex) is the benthic respiration rate (sink) [M L -3 T- 1 ],
R(x) is the algal respiration rate (sink) [M L -3 T- 1 ],
P(x, t) is the algal photosynthetic oxygen production rate (source)
[M L -3 T-1l
The respiration sinks in (6.38) were assumed to be constant. The solution of
(6.38) may be expressed as
D(x, t) = DoCt - x/U)exp( -j.x)+ Fd,r[exp( -jrx) - exp( -j.x)]
+ Fn.r[exp( -jnx) - exp( -j.x)]
+ (SIK) [1 - exp( -j.x)] + (RIK)[l - exp( -j.x)]
- P m {{2plnK.)[1 - exp(-j.x)] + j(x, t)}, (6.39)
188
where
" Kr
Jr=U'
F = _K--,,-"_N--=-o_(I_-_X/_U_)
".r Ka - Kr
~ [K;
j "( x, t ) = n~l b n
+ (2rrn?]1/2 [(
cos 2rrn t - ts -"2p) - tan
_ K:
12rrn]
00 b
- exp( -jaX)n~l [K; + (;rrn)2]1/2
. cos 2rrn t - t - -P - -x ) - tan _ 1 2rrn]

.-
[ ( s 2 U Ka •
Equation 6.39 was considered to be applicable to streams in which the primary

cause of the diurnal variation in DO is the algal oxygen production. However,
it was recognized that larger, rooted plants may also be important in some
cases.
Appendix 6.1 presents several rate coefficients for the BOD decay processes
that are frequently of major importance in river water quality modeling. Many
excellent reviews of stream reaeration have been written (Bennett and Rathbun,
1972; Lau, 1972a, b; Wilson and MacLeod, 1974; Rathbun, 1977). A detailed
survey offormulas for the reaeration coefficient is summarized in Appendix 6.2.
6.4. DOSAG I AND QUAL I
DOSAG I is one of two stream quality models developed by the Texas Water
Development Board (l970a) to simulate the spatial and temporal variations in
BOD and DO under various conditions of flow and temperature. The other
model, QUAL I, was designed to simulate the spatial and temporal variations
in water temperature and conservative mineral concentration in addition to
BOD and DO. These programs may be used as complements to each other.
DOSAG I analyzes a multiple-reach, branching river system.
A junction is the confluence between two streams within the river basin being
modeled.
A stretch is the length of river between two junctions.
A headwater stretch is the length of river from a headwater to its first junction
with another stream.
A reach is the subunit of length within any stretch.
189
When a significant change occurs in the hydraulic, biological, or physical
characteristics of the stream (including the addition of a waste load or with-
drawal of water from the stream), then a new reach is added to any point in the
stretch.
The purpose of DOSAG I is to predict the BOD and the minimum DO
in a stream, as well as to estimate the required flow augmentation to bring the
DO up to the target level. The removal of carbonaceous biochemical oxygen
demand BO DC and nitrogenous biochemical oxygen demand BOon is expressed
by first-order equations:
dBOD C
_K C. BOD c (6.40)
dt
and
dBOD n
-Kn·BOD n. (6.41 )
dt
The bio-oxidation rates for BOD c and Boon, K C and K n , are considered to be
constant for a given reach. The standard equation for atmospheric reaeration
is used in the form
dDO
- = K (DO' - DO) (6.42)
dt a
and the DO equation is given as

KC
DOt = DO' + [exp( -Kat) - exp( -KCt)]BOD o
K a + KC
Kn
+ [exp( -Kat) - exp( -Knt)]BOD~
Ka + Kn
- (DO' - DOo)exp( - K a t). (6.43)
A Lagrangian solution technique, in which a coordinate system moves' with a
particle of water downstream, was applied to solve the DO equation. At each
change in the reach and at every junction a mass balance is computed to
determine the BOD and DO in the next reach. The model has provisions for
waste discharges and for withdrawal of water at any location within the stream
system. As an example, this model was applied to the upper portion of the San
Antonio River basin (Figure 6.2).
QUAL I was developed during 1969-70 by F. D. Masch and Associates in
collaboration with the Texas Water Development Board (I970b). This inte-
grated system of interrelated mathematical models of water quality is capable
of routing through a one-dimensional completely mixed branching stream
system: (a) conservative minerals, (b) BOD/DO, and (c) temperature. QUAL I
190
~
0)50
E40
g 30
CD
~ 20
III
.~ 10
§ L.::d---.-r1r----------=====::::::::::::::::===
10
'I 8
0) 6
E
o 4
I:) 2
O'---_ _--L- ----'- ----'-_
384 288 192

Distance from river mouth (km)
FIGURE 6.2 DOSAG I model applied to San Antonio River basin (Texas Water
Development Board, 197Gb).
possesses the following capabilities and characteristics (Texas Water Develop-

ment Board, 1971):
(1) The stream may be discretized into elements of suitable length and
variable cross section to obtain any degree of resolution that is warranted.
(2) It can account for heat exchange across the air-water interface and is
capable of handling waste inputs and withdrawals at selected points
along the channel axis.
(3) It allows for transport by advective and dispersive mechanisms along
the principal axis of flow (longitudinal axis).
(4) Solutions provide for a temporal and spatial description of conservative
material, BOD and DO, and temperature variation throughout a stream
system.
(5) It provides for determining flow augmentation requirements based on
selected minimum allowable concentrations of dissolved oxygen.
191
(6) It has the capability of an integrated system so that the results of any
submodel can be used as the input to another model provided that such
feedback is required; each submodel also has a stand-alone capability.
(7) It is so structured as to be completely general, and can be applied to any
stream system by choosing the appropriate parameters and providing
the necessary data that relate to a specific case.
QU AL I is a series of mass balances of a given water quality constituent over a

time interval dt on a stream segment of cross-sectional area A and of length
dx along the principal axis of flow. In steady state nonuniform flow, conserva-
tive and nonconservative constituents in a stream may be described by
ac A = a(AE ac;ax) _ a(AUC) + AS
(6.44)
at ax ax - ,
where
C is the concentration of the constituent [M L - 3],
E is the longitudinal dispersion coefficient [U T- 1 ],
S represents sources or sinks of a nonconservative constituent [M L - 3 T- 1].
There is a similar equation to (6.44) in which C represents temperature and S

represents sources of heat. When (6.44) is written for a control volume ~ of
element i in the stream system (as illustrated in Figure 6.3 for steady state non-
uniform hydraulics), the resulting expression is
aC i (AE ac;aX)i+ 1/2 - (AE aClax)i-I/2
at ~
(6.45)
where
~ is the volume ofa control element [L 3 ], ~ = Aidx,
Ai is the mean cross-sectional area of the control element [U],
Ai = -!(A i - 1 / 2 + A i + 1/2),
(AE aClax)i_ 1/2 is the total longitudinal dispersion of the constituent
[L 3 T- 11M L - 3] or the temperature [U T- 1;oC] on the
inflow side of the control volume,
(AE ac;aX)i+ 1/2 is the total longitudinal dispersion of the constituent
[L 3 T- 11M L - 3] or the temperature [L 3 T - l;oCJ on the out-
flow side of the control volume,
Qi _ 1/2 is the rate of flow into the control volume [L 3 T- 1],
Qi + 1/2 is the rate of flow out of the control volume [L 3 T- 1],
QXi represents local inflows or withdrawals [L 3 T- 1],
C i - 1 is the concentration of the constituent [M L - 3], or tempera-
ture (DC), in the in flowing water,
192
Control volume
V,
0i +)1,
Hydrological balance
E ilC)
IA t>x ax i-)I,
0i + )I,Ci
E ilC)
IA t>x ax i+)I,
Material balance
FIGURE 6.3 Discretized stream system (Water Resources Engineers. Inc., 1967).
Cj is the concentration of the constituent [M L - 3], or tempera-

ture ("C), in the control volume,
C x, is the concentration of the constituent [M L - 3], or tempera-
ture (OC), in Qx •
Sj represents sour~es or sinks of a nonconservative constituent
[M L - 3], or temperature ("C).
193
Under the assumption of steady state conditions a hydraulic balance for the
control volume has been represented by the following continuity equation:
(6.46)
where
Pr is the precipitation rate,
E, is the evaporation rate.
In many cases E v and P r are neglected. Equation 6.44 without a source or sink
term S may be used to describe the behavior of a conservative mineral within a
stream:
aM o(AE aM/ax) o(AUM)
-A= ---- (6.47)
at ax ax'
where M is the concentration of the conservative mineral [M L - 3]. Also, if there
is complete mixing, (6.44) written with a source term for temperature is
~~ A = o(AE aT/ax) _ o(AUT) +~ ST

(6.48)
at ax ax - YC p ,
where
y is the density of water [M L - 3J,
Cp is the specific heat of water [cal M - 1 °C - 1 J,
ST is the heat source term (the net rate of heat gain or loss per unit volume of
water) [cal L -3 T- 1 ].
For a stream oflength ~x and mean surface area W the total rate of heat transfer
across the air-water interface is TN W, where TN is the net energy flux passing
through the interface. This heat is distributed uniformly throughout the under-
lying volume A~x. The heat term can be defined as
TNW -I
ST = A~x = TNH , (6.49)
where H is the mean hydraulic depth of the stream. Therefore, the basic one-
dimensional heat transport equation is
aT A = a(AE aT/ax) _ a(AUT) + AT".

(6.50)
at ax ax - YCpH
F or practical applications the parameters y and C p are considered constant.
The rate of oxygen utilization due to BOD is expressed by the first-order
reaction (the ultimate BOD is considered), and the reaeration process is
expressed in the same way as in DOSAG I. If the transverse section of the stream
194
is completely mixed, (6.44) can be written for DO and for BOD, respectively, as
aDO A = a(AE aDo/ax) _ a(AUOO) + AS (6.51 )

DO
~ ~ ~-
and
aBO 0 a(AE aBoo/ax) a(AUBOO) AS
----at A = ax - ax ± I.'
(6.52)
where
SDO = K.(OO' - DO) - (K d + K.)BOO,
S/. = -(K d + K.)BOO,
Kd is the deoxygenation rate constant [T- 1],
K. is the rate constant for BOD removal by sedimentation [T- 1].
OKLAHOMA ARKANSAS/
20 NEW ~- ............. ......,.-~~
iV1EXI(O
_J ~u" WU<l~'" D~lliI~
15
1-
OJ
E
10
0
0
OJ
5
6
1
OJ
E 4
0
0
2
O'--_---'---_ _--'-----_-----'-_ _--'-_----I_ _--l...

354 322 290 257 225 193 161
FIGURE 6.4 BOD and DO simulation results obtained by application of QUAL I to
the San Antonio River basin (Texas Water Development Board, 1971). The graphs show
the observed points and the computed curves. Kd was estimated from field data.
195
200 --,-------------------------,
o
100 --o;...----Q-~
0+---,.-----,------,----,-------,.----1
354 322 290 257 225 193 161
.... 900,-----------------------,
0-o.=-
e:: Ql I
o >-
'';:; 0 OJ o
~ .~
e::
E
-0 (/)
700-
~ro~
§oo
u +" (/) 500 +------,--------,---,-------,-------r------1
354 322 290 257 225 193 161
FIGURE 6.5 Results of simulations for conservative minerals obtained by application
of QUAL I to the San Antonio River basin (Texas Water Development Board, 1971).
The graphs show the observed points and the computed curves.
A finite-difference method is used to obtain an analytical solution. QUAL I

uses a longitudinal dispersion coefficient E [U T- 1 ], defined as E = C U* H,
where U* is the bed shear velocity [L T- 1]. Since for steady state flow the shear
velocity was described as U* = R2/3nS~·5, whereR is the hydraulic radius, Se is the
slope of the energy gradient, and n is the Manning coefficient for the reach, then
the resulting dispersion equation was expressed as E = C'nUHO. 833 . The mean
velocity U and the mean depth of flow in a reach, H, are calculated in a similar
way as in DOSAG I, by equations of the form U = rt.QP and H = yQ/;, where
constants rt., p, y, , are usually determined from stage-discharging rating
curves and Q is the mean discharge in a reach. QUAL I was applied to part of
the San Antonio River basin. The results from the BOD and DO simulations are
shown in Figure 6.4. The simulations for conservative minerals are shown in
Figure 6.5.
196
6.5. NITROGEN CYCLE
Organic
nitrogen Ammonia Nitrate
waste waste waste
sources sources sources
j j I
Organic Hydrolysis Ammonia Nitrosomonas Nitrite Nitrobacter Nitrate

nitrogen nitrogen nitrogen nitrogen
i I i I
Nitrate reduction
Death Plant and

• I
animal nitrogen
FIGURE 6.6(a) Major features of the nitrogen cycle, showing nitrification (oxidation)
and nitrate reduction.
In river systems where nitrogen is a major constituent, the assumption of a

constant reaction rate K n (as in eqn. 6.41) may not be appropriate. The rate may
vary along the system simply because of the changing concentrations of the
various forms of nitrogen, each form having its own rate constant. A modifica-
tion of (6.44), together with data on the input of each form of nitrogen and its
corresponding reaction rate, will permit the prediction of the separate com-
ponents of the nitrification process as illustrated in Figures 6.6(a) and (b).
c
o
.+=:
...c~ Ammonia
/'
/" "Nitrate
Q)
u /
c /
o /
u /
c /
Q) /
Ol
e
.<.=
z
Distance downstream
FIGURE 6.6(b) Sequential reactions in nitrification with increasing distance down-
stream from point source of ammonia waste.
197
O'Connor and Di Toro (1970) proposed some sequential reaction models to
account for each nitrogen component.
Let us consider the following concentration components [M L - 3]: N l'
organic nitrogen; N z, ammonia nitrogen (NH 3 - N); N 3, nitrite nitrogen
(NOz-N); and N 4 , nitrate nitrogen (N0 3 -N). We shall let Ki now represent
the first-order decay (including settling) of nitrogen form i [T- 1], Ki. i + 1 the
forward reaction coefficient [T- 1 ], and J.t;(X) the discharge of form i at site
X [M L - 3 T - 1]. The solution of the following four equations will permit a
prediction of the concentration of each form of nitrogen in steady state river
systems:
dZN dN
o = E~-z1 - U - 1 - K~Nl + W1(X)
dX dX
Z
0= E d ~ _ U dN i _ K"N· + K~-l ·N·- l + W,'(X) i = 2,3,4, (6.53)
dXz dX '1 ',I 1
where
E is the dispersion coefficient (applicable to estuaries) [U T - 1],
U is the net downstream velocity [L T- 1].
The decrease in the dissolved oxygen concentration in such river systems due
to the individual nitrogen constituents is caused by oxidation of ammonia,
3.43K z3 N z, and of nitrite, 1.14K 34 N 3' The dissolved oxygen deficit concen-
tration D" resulting from the oxidation of these two nitrogen forms can be
predicted by
dZD" dD"
o= E dX z - U dX - KaD" + 3.43 K Z3 N z(X) + 1.14 K 34 N 3 (X), (6.54)
in which K a is the reaeration rate constant [T- 1 ] and N z (X) and N 3 (X) are
obtained from the solution of eqns. 6.53.
This model can be further expanded to include denitrification and algal
utilization. Denitrification may involve the reduction of N0 3 to NO z and the
conversion of NO z to N z gas. These reactions occur in low concentrations of
dissolved oxygen. Assimilatory nitrate reduction can convert N0 3 to NH 3 .
The utilization of ammonia nitrogen and nitrate nitrogen by phytoplankton
may also take place, and hence produce organic nitrogen, thereby completing a
very simplified version of the nitrogen cycle. This process is generally defined
by reaction rate constants Kij (i, j = 1, 2, 3,4). Feed-forward reactions involve
reaction rates Kij (i < j) where j is usually i + 1, and feedback reactions involve
reaction rates Kij (i > j) where.i is usually i - 1. All these reactions (KijN i )
are then included in (6.53):
dZN· dN·
0= E dX z' - U dX' - K?N j + 2: (Ki;N) + J.t;(X) i = 1,2,3,4. (6.55)
j*i
198
By finite-difference approximations four algebraic equations, one for each
form of nitrogen, can be written for each finite section k of the river system as
appropriate.
0= L [-Qkj(rtkjN ik + 13kj N ij ) + E~j(Nij -
j
N ik )]
~(Krk N ik - L K~ik N hk) + ~k i = 1,2, 3,4, Vk, (6.56)

h
where
Qkj is the net flow from an adjacent section k to section j (positive if from k
toj, negative iffromj to k) [L 3 T- 1 ],
~ is the volume of segment k [L 3],
rt kj is a finite-difference weight (max (1, 1 - E~)Q)),
13kj is 1 - rtkj,
E~j is the exchange (bulk dispersion) coefficient [L 3 T - 1J, defined by
, 2E kj A kj
(6.57)
Ekj = L k + L/
N ik is the concentration [M L - 3J of nitrogen form i in section k,

K~ik are the reaction rate constants [T-1J for the conversion of nitrogen
form h to form i in section k,
~k is the direct discharge [M T-1J of nitrogen form i into section k.
In equation 6.57,
E~j is the dispersion coefficient between sections k and j,
A kj is the interfacial cross-sectional area,
L k , Lj are the lengths of sections k and j.
The spatial distribution of the dissolved oxygen deficit concentration Df

caused by ammonia oxidation (from NH 3 to N0 2 , reaction rate 3.43K z3 )
and by nitrite oxidation (from N0 2 to N0 3 , reaction rate 1.14K 34 ) can be
predicted from the simultaneous solution of similar finite-difference equations:
0= L [-Qkj(rtkjDf + fikjD'J) + E~j(D'J - DflJ
j
- KakDf ~ + 3.43K z3 N 2k ~ + 1.14K 34 N 3k ~, (6.58)

in which N 2k and N 3k represent, respectively, ammonia and nitrite nitrogen
concentrations [M L - 3J in section k and K. k is the reaeration rate constant
[T-1J for section k.
Equations 6.56 and 6.58 represent five equations for each section of the
river. They can be combined and put into a convenient matrix form for solution
using any of a number of computer programs available for solving large sets of
linear equations. Since some of the feedback reaction coefficients are functions
199
of the dissolved oxygen deficit, parameter adjustments will be necessary to
ensure an adequate prediction of the concentrations of dissolved oxygen and of
each form of nitrogen. Just how these adjustments are made is indeed part of
the art of water quality modeling. The results of these nitrogen models, such as
those applied to the Delaware Estuary, are illustrated in Figure 6.7 (O'Connor
etal., 1976).
2.0 '""'"Ba-c'k-g-ro-u-nd-,,------=-.---r.--------------,
z_ assumed due •• • • •
.~~
c: - 1.0
to plankton iII"'--':::: • •
ell OJ • _._1./ . ~.-t----!
EJ E • • I I. • •
0-
0'------'---'------'-------'------'---'-=---'---------"-----'
4.0,---------------------,
I 3.0f-
OJ
E a....... •
2.0e- ~
~
~ 1.0f-
. (' 1\~,
• •
I-!.-. •
,.-i,-I".
iJ· I ,
t I
,
•,,·.. . .
., "--;-.
O~--'------.AL.C:=--=-----'------'---'-------'---------'----.J
193 161 129 97
~ 2.0,---------------------------,
OJ
E
z I
1.0
N
o
Z 0 ........-=--'------===-____'__--'-_ _ ~~_ _'___'_"_"""" -.J
5.0.-------------------,
I
~ I/j\,
1* .\.
4.0-
E
z
OJ
3.0
ill... -
o 2.0
'"
• :/III\i
• t •
: t~
•
z ~ '-1/
1.0 ,Ii., , '"
193 161 129 97
Distance from estuary mouth (km)
FIGURE 6.7 Observed (points) and computed (curves) nitrogen profiles for the Delaware
Estuary (O'Connor el al., 1976). Observations were made in 1964 on 30 July CA.), 10
August (.), and 31 August (e).
200
The models discussed above are all first-order kinetic steady state models.
While they vary in complexity and in their requirements for data, they remain
relatively simple. In the models, flows and temperatures do not vary with time
and the complex nonlinear kinetic interactions between the microorganisms
and the constituents are approximated by linear or first-order reactions.
6.6. MULTICONSTITUENT RIVER ECOSYSTEM MODELS
Over the past decade there has been an increasing emphasis on the effects of
various constituents, especially nutrients, on the aquatic ecosystem, i.e. on the
production of bacteria, protozoa, phytoplankton, zooplankton, fish, and
other trophic levels within natural water bodies (Chen and Orlob, 1975).
QUAL II, developed for the US Environmental Protection Agency (Water
Resources Engineers, Inc., 1973), is a good example of many of the operational
aquatic ecosystem simulation models. It predicts a variety of water quality
constituents, including conservative substances; algal biomass and chlorophyll
a; nitrogen from ammonia, nitrite, and nitrate; phosphorus, carbonaceous
biochemical oxygen demand, and benthic oxygen demand; dissolved oxygen;
coliforms and radionuclides. The following paragraphs define the mathematical
relationships that describe the individual reactions and interactions among
these constituents. The dispersion and advection terms are not shown in the
following equations, but are included in the actual model.
The chlorophyll a concentration Co is considered to be proportional to the
concentration A of phytoplanktonic algal biomass [M L - 3] :
(6.59)
The time-varying growth and production of algal biomass, dA/dt, depends on

the specific growth rate of algae, J.lA [T- 1 ], the respiration rate (or specific loss
rate) p [T - 1], the algal settling rate (J 1[L T - 1], and the average stream depth
H,,[L], all at a particular location X in the river system. Although not in the
QUAL II model, the concentration of algae is also a function of mortality M
and of grazing Gz by higher trophic levels such as zooplankton (of concentration
Z). Hence,
(6.60)
The growth rate of algal biomass is dependent upon the temperature and
availability of nutrients (nitrogen, carbon, phosphorus) and upon light. The
standard Michaelis--Menten formulation, illustrated in Figure 6.8, defines the
specific growth rate J.l at a given site in a river system.
201
Jlmax --------------
....
Q)
~
..r::.
~ O.5J1max
e
t:J
Substrate concentration S
FIGURE 6.8 Michaelis-Menten kinetic expression for microbial substrate utilization
and growth. Growth rate 11 = Ilmax[S/(K s + S)], where K s is the half-saturation concentra-
tion.
This Michaelis- Menten formulation assumes that Il is a parabolic function

of the essential substrate concentration S:
S
Il = Ilmax K + S· (6.61)
s
The parameter K, has the value of S when Il is half of the maximum growth
rate Ilmax' and is called the half-saturation concentration. At low substrate
concentrations, II is proportional to S. For high values of S, Il approaches the
limiting saturation value Ilmax' In QUAL II the algal growth rate IlA may be
limited by nitrate nitrogen N0 3 , phosphorus P, carbon e, light intensity, and
temperature. Hence,
(6.62)
in which
Il~~~o is the maximum growth rate of algal biomass at 20 0 e [T- I],

o is a temperature coefficient ranging from 1.02 to 1.06,
T is the water temperature cae),
N0 3 , P, e are the nitrate nitrogen, orthophosphate phosphorus, and
carbon concentrations [M L - 3],
K NOJ ' K p , K c are the temperature-dependent half-saturation concentrations
for algal growth [M L - 3],
202
I is the light intensity [ly T- 1 ],
K J is the half-saturation coefficient for light [ly T- 1 ],
A is the light extinction coefficient in the river [L -I].
Equation 6.62 couples algal biomass production to the available supply of

nutrients, light, and heat; hence algae and chlorophyll a will vary in time and
space in response to the concentrations of elements needed for growth. By
eqn. 6.62, if any of the critical growth element concentrations are zero, then flA
is zero. This expression is based on a multiplicative growth hypothesis, in
contrast to a threshold growth hypothesis. The latter approach includes only
the Michaelis-Menten term of the most limiting nutrient, or of light, that con-
strains algal growth. This threshold hypothesis is expressed in the following
equations:
- (flmax ()T- ZO) ( N0 3 ) if N0 3 is limiting

fl
A - A.ZO NO
3
+K NO,
fl = (flmax ()T - ZO) ( P ) if P is limiting

A A.ZO P+K p
(6.63)
fl
A
= (flmax ()T -
A,ZO
ZO) (
C +C K c ) if C is limiting
fl - (flmax ()T-zo)[_l_ln( K[ +I )] if I is limiting.

A - A,ZO AHa K J + Iexp(-AHa)
Respiration is also temperature-dependent:
Pr = PZO {)T- ZO, (6.64)
where PZO is the respiration rate [T- 1 ] at 20°e. All temperature-dependent rate
constants are defined by equations having the form of eqn. 6.64, the constant ()
being defined for each rate constant.
The mortality rate M [T- 1 ] of algal biomass can be expressed as a linear
function of the natural mortality rate M N and the mortality caused by the
toxicity of the water, PC x:
(6.65)
where
P is a toxIcIty coefficient (rate of mortality per unit concentration)
[T- 1 M- 1 L 3 ],
Cx is the concentration of the toxic constituents [M L - 3].
The growth rate flz of zooplankton can be defined by
Ilz = (fl,;.a~o{)T-ZO) (K A A+ A)' (6.66)

203
where K A is the temperature-dependent half-saturation concentration for algal
biomass [M L - 3]. The product of zooplankton growth rate Ilz [T- I], concen-
tration Z [M L - 3], and a conversion coefficient FA. z [M M - I], which indicates
the mass of algal biomass required per unit mass of zooplankton, is an estimate
of the loss in algal biomass due to zooplankton grazing Gz . The reliability of
current models for predicting trophic levels higher than phytoplankton (such
as zooplankton and fish) is relatively poor (Russell, 1975), hence the omission
of these higher trophic levels from operational models such as QUAL II.
The nitrogen cycle in QUAL II is described by differential equations
governing the transformation of nitrogen from one form to another. For the
ammonia nitrogen concentration NH 3 [M L - 3],
dNH 3 0"3
-----crt = C(lpA - PI NH 3 + Ax' (6.67)
where
C(I is the fraction of respired algal biomass resolubilized as ammonia nitrogen
by bacteria,
PI is the temperature-dependent rate of biological oxidation of NH 3 [T -I],
0" 3 is the benthos source rate for NH 3 [M T - I L - I],
Ax is the average stream cross-sectional area at location X [U].
For the nitrite nitrogen concentration N0 2 [M L - 3],

dN0 2
~ = PI N H3 - P2 N0 2 , (6.68)
in which
PI is the rate of oxidation of NH 3 to N0 2 [T - I ],
P2 is the rate of oxidation of N0 2 to N0 3 [T- I ].
For the nitrate nitrogen concentration N0 3 [M L - 3],

dN0 3
~ = P2 N0 2 - C(IIlA A. (6.69)
Equations 6.67-6.69, together with eqn. 6.64, close the loop of the nitrogen
cycle. Equations 6.67 and 6.69 demonstrate that the fraction of respired algal
biomass resolubilized as NH 3 -N is assumed to equal the fraction of biomass
that is N0 3 -N.
The phosphorus cycle is modeled in less detail than the nitrogen cycle. Only
the interaction of phosphorus and algae and a sink term are considered. Thus
the differential equation describing the time-varying concentration P [M L - 3]
of orthophosphate phosphorus is written as
dP 0"2
Cit=C(2 A (P-IlA)+ Ax' (6.70)
204
in which
(xz is the fraction of algal biomass that is phosphorus,
(J z is the benthos source rate for phosphorus [M T - 1 L - 1l
Carbonaceous biochemical oxygen demand BOOc [M L -3] is formulated

as a first -order reaction:
dBOO C
- KCBOO c - KsBOOC, (6.71 )
dt
where
KC is the temperature-dependent deoxygenation or decay rate of carbonaceous
BOD [ r 1],
Ks is the rate constant for loss of carbonaceous BOD by settling [T-1l
The benthic oxygen demand BOOb [M L - 3] is assumed to be fixed, depen-

dent on the cross-sectional area Ax at location X:
dBOO b
(6.72)
dt
where K b is a constant benthic source rate [M T - 1 L - 1l
The differential equation that describes the rate of change in dissolved
oxygen is
(6.73)
where
OO~ is the temperature-dependent dissolved oxygen saturation at location
X [M L -3],
is the dissolved oxygen concentration at X [M L - 3],
is the temperature-dependent reaeration rate constant [T- 1J,
is the rate of oxygen production through photosynthesis per unit of
algal biomass [M M - 1],
Cl 4 is the rate of oxygen uptake from respiration per unit of algal biomass
[M M-1J,
is the rate of oxygen uptake per unit of oxidation of ammonia nitrogen
[M M-1J,
is the rate of oxygen uptake per unit of nitrite nitrogen oxidation
[M M-1l
205
The complete set of equations that is numerically solved by QUAL II
includes the effects of dispersion, advection, constituent reactions, and inter-
actions up to the phytoplankton trophic level. The set also includes a source
term S [M T- 1 J that is assumed uniform over the length ~x of the river section
at location X.
6.6.1. Bacteria and Protozoa in Pollutant Degradation

The QUAL II model structure, a part of which has just been outlined, is typical
of many multiconstituent water quality models. In their application some
assumptions have been changed and modifications made to suit particular
river conditions. However, an alternative modeling approach that explicitly
includes the role of bacteria and protozoa in the degradation of biodegradable
pollutants and in the reduction of dissolved oxygen in aerobic river systems
has been developed by Stehfest (1977). Michaelis- Menten kinetics for bacterial
and protozoan growth are assumed, including the possibility of differing
nutrient uptake rates (caused by the presence of more than one type of enzyme)
or inhibition, i.e. the reduction in the rate of degradation of one nutrient due to
the presence of another nutrient, toxic substance, or degradation byproduct.
The overall model structure can be introduced by first considering a simple
self-purification process consisting of a single nutrient concentration N, such
as that of a particular form of nitrogen, a bacterial biomass concentration B,
and a dissolved oxygen concentration DO, in an aquatic environment suitable
for bacterial growth. In this situation the rate of change dN /dt in concentration
will be a function of the consumption of N by bacteria:
dN
(ll =
max(K N+ N ) B,
- a NB J1BN (6.74)
N
where
aNB is the mass of nutrient N consumed per unit increase of bacterial biomass
[M M-1J,
J1'};~x is the maximum growth (increase) of bacterial biomass resulting from a
unit mass of nutrient N per unit time [M M - 1 T- 1 J,
KN is the nutrient half-saturation concentration for bacterial growth
[M L -3].
The rate of change dB/dt in the bacterial biomass concentration will depend
on bacterial growth and endogenous respiration:
(6.75)
206
where PB is the endogenous respiration rate of bacteria, i.e. the mass decrease
of a unit mass of bacteria per unit time [M M - 1 T - 1].
The rate of change in the concentration of dissolved oxygen, DO, will be
governed by physical reaeration and by the consumption of oxygen from nutrient
degradation and endogenous respiration:
d~tO = Ka(DO
S
- DO) - a~B Jl'Jl~x (K N: N)B - aOB PBB, (6.76)
where
a~B is the mass of oxygen required per unit increase in bacterial biomass
resulting from nutrient N [M M - 1],
aOB is the mass of oxygen required per unit decrease in bacterial biomass due
to endogenous respiration [M M - 1].
The Michaelis-Menten expression on the right-hand side of (6.74) represents

a two-parameter approximation of the kinetics of nutrient degradation. For
low nutrient concentrations the rate of change of N will be approximately
proportional to the product of the nutrient and bacterial biomass concentra-
tions, N B. For high nutrient concentrations dN /dt will be independent of N;
it will depend only on B.
The first term on the right-hand side of (6.75) is identical to - aBN dN /dt,
where aBN is the mass of bacteria increase resulting from a unit mass of nutrient
(l/aNB)' Hence the change in bacterial biomass per unit change in nutrient
concentration, dB/dN, is assumed constant and equal to -a BN . The second
term on the right-hand side assumes that the decrease of bacterial biomass
through endogenous respiration is proportional to the biomass concentration.
In (6.76) oxygen consumption in nutrient degradation is assumed propor-
tional to dN /dt. Oxygen consumption in endogenous respiration is assumed
constant per unit of bacterial biomass.
If a second nutrient is added, it could be one that is degraded independently
of the first constituent. The resulting model would include an equation similar
to (6.74) for the second nutrient, and appropriate terms in (6.75) and (6.76).
The procedure can be extended to include any number of independently
degraded constituents.
Nutrient growth can be hindered by either competitive or allosteric inhibi-
tion. A Michaelis-Menten expression for competitive inhibition is
dN 1 max ( N1 )
(6.77)
----crt = -aN,BJlBN, K N, + Nt + a2 N l B,
where
N1 is the concentration of the inhibited (or more slowly utilized) nutrient,
N2 is the concentration of the inhibiting nutrient,
a2 is the inhibition constant for N 2' a dimensionless parameter.
207
An expression for allosteric inhibition is
dN 1 _ _ max ( N1 ) B (6.78)
crt - aN \BJ1BN, (K N \ + N 1 )(1 + azN z )
as long as the affinity of the bacteria to the inhibitor is as strong as it is to the
inhibited nutrient.
If N z is insignificant, (6.77) and (6.78) are identical. In competitive inhibition
the rate of N 1 degradation is less dependent on the concentration of N z than
it is in allosteric inhibition. An allosteric inhibitor will tend to block the degrada-
tion of a nutrient or constituent much faster than will a competitive inhibitor.
These distinctions and effects are illustrated in Figure 6.9.
The consideration of protozoa feeding on bacteria requires another term
in (6.75) that reflects the consequent decrease in bacterial biomass, a separate
equation to reflect the rate of change in protozoa mass P, and finally terms in
(6.76) defining the oxygen requirements for protozoa feeding and respiration.
Hence, in addition to (6.74), which defines dN /dt, the single-nutrient model
will include the following three equations:
(6.81)
--- -----
Allosteric inhibition
Inhibitor concentration I
FIGURE 6.9 Effect of inhibitor concentration Jupon rate of degradation ofa constituent
having concentration C. Full curves, high C; broken curves, low C.
208
40
-
COD of
pollutants
(mg 1- 1)
••
Bacterial
concentration
4
(mg 1- 1) 62t·-· ••
o LI_---'--_----'-_---L_ _____L_ _____L_-----'-_-----'-_
Protozoan
o.a~ x~
0.4~ ~
concentration
(mg 1-')
o t_-LI _~_____L
_ _____L_-----'-I_-----'I_----'\_ ____'I_ ____'IL__LI-
6
4
2
o
40''-=O-L.--=-50-:-:0,------'------=6-':-OO,,------'---7-Lc:------'--a--lo-o----..J-
OO
River distance (km)
E N N C en
C C ;0 Q)
Q) Q) Q)
..c co ::£
:0 a:
c ~ 0
c ::£
co
~
FIGURE 6.10 Solution (full curve) of model of the River Rhine, and some measured
values (Stehfest, 1977).11',,11'2' and »'3 are the concentrations of easily degradable, slowly
degradable, and nondegradable pollutants, respectively.
where
aBP is the mass of bacteria consumed per unit Illcrease III protozoa mass
[MM- 1],
J.1 pax is the maximum growth (increase) of protozoa mass resulting from unit
mass feeding on bacteria per unit time [M M - I T - 1],
KJi is the bacteria half-saturation concentration for protozoa growth
[M L -3],
209
pp is the endogenous respiration rate of protozoa mass [M M - \ T - 1],
a~p is the mass of oxygen required per unit increase in protozoa mass resulting
from feeding on bacteria [M M - \],
aop is the mass of oxygen required per unit decrease in protozoa mass due to
endogenous respiration [M M - \].
The modeling approach just defined has been applied to the prediction of
organic pollution, and dissolved oxygen, in the Rhine in Germany, a relatively
benthos-free, fast-flowing, and deep river (Stehfest, 1977). In this application
the parameter values were identified on the basis of measurements of the con-
centrations of chemical oxygen demand COD, bacterial biomass B, protozoa
mass P, and dissolved oxygen DO. COD was divided into easily and slowly
degradable nutrients, N \ and N 2' respectively (COD = N 1 + N 2)' Equation
6.74 describes the degradation kinetics of N \' whereas (6.77) or (6.78) was
considered appropriate for the degradation kinetics of N 2' in which it was
assumed that N \ acted as an inhibitor to reduce the degradation rate of N 2'
Equation 6.77 defining competitive inhibition was considered more appropriate.
For a constant mass discharge rate of COD (where etCOD = N \ and
(l - et)COD = N 2 [M T- 1]) into a river reach having a flow rate Q [L 3 T -I]
and velocity U [L T - 1], the predictive model for the river reach, excluding
advection and dispersion terms, is defined in Tables 6.1 and 6.2. The result,
applied to the Rhine, is illustrated in Figure 6.10.
6.6.2. Michaelis-Menten Models of Aerobic Nitrogen Cycle

Figure 6.11 summarizes an extension of this Michaelis- Menten kinetic modeling
approach using the aerobic nitrogen cycle as an example (Harleman, 1978).
The components of the aerobic nitrogen cycle include nitrogen in the chemical
forms of ammonia, nitrite, nitrate, and the organic nitrogen content of phyto-
plankton and zooplankton. The cycle also includes particulate and dissolved
organic nitrogen. Not included are free nitrogen and the exchange of nitrogen
between the atmosphere and bottom sediments. The transformation of phyto-
plankton directly to dissolved organic nitrogen is assumed to be negligible
relative to other nitrogen transformation processes.
The model assumes a bacterially mediated transformation from dissolved
organic nitrogen to ammonia, to nitrite, and finally to nitrate nitrogen.
Michaelis- Menten kinetics are assumed for the uptake of nitrate and ammonia
by phytoplankton and the grazing of zooplankton on phytoplankton.
The concentration of heterotrophic bacteria affects the rate of conversion
of dissolved organic nitrogen to ammonia nitrogen. The concentration of
Nitrosomonas bacteria affects the rate of conversion of ammonia nitrogen to
nitrite nitrogen and the N itrobacter bacteria concentration affects the conversion
rate of nitrite to nitrate nitrogen.
N
..-
o
TABLE 6.1 Water Quality Interactions for a Benthos-Free River (Stehfest, 1977).
Easily degradable inhibiting nutrient dN! m" ( N, ) VIXCOO
concentration (N!) dt -aN,B/lBN, K , + N, B + --Q-
N
Slowly degradable inhibited nutrient dN 2 m" ( N2 ) V(I - IX)COO
concentration (N 2) dt -aN,B/lBN, K + N + N B + Q
. N2 2 al 1
j
Bacterial biomass concentration (B) dB _ N
m" ( )B + m"( N2 )B _ m,,(_B_)p _ B
dt - /lBN, K N, + N j /lBN, K N, + N 2 + ajN, aBP/lp KB + B PB
Concentration of protozoa mass (P) dP = /lJ)" (~_)p _ PpP

dl KB + B
Dissolved oxygen concentration (DO) dOO = K (DO' - DO) - a~iJ/lB~~ ( K N

+N! ) B - aO's/lBN,
N m" ( K + N2
N + a,N, ) B
df a Nt 1 N1 2
B "m" \K
- aOp,.p L_~)p+ B - aOBPBB - aOpppP
B
211
TABLE 6.2 Parameter Values for Rhine River Model (Table 6.1) (20 0 e,
1.25 . Mean Flow) (Stehfest, 1977).
Parameter Definition Value
Mass of N, consumed per unit increase of bacterial biomass 2.6 mgNdmgB

Mass of N 2 consumed per unit increase of bacterial biomass 3.4 mgN 2 jmgB
Mass of bacterial biomass consumed per unit increase in
protozoa mass 3.0 mgBjmgP
Mass of oxygen required per unit increase in bacterial biomass
from nutrient N I 1.6 mgOjmgB
Mass of oxygen required per unit increase in bacterial biomass
from nutrient N 2 2.4 mgOjmgB
Mass of oxygen required per unit increase in protozoa mass
from feeding on bacteria 2.0 mgOjmgP
Mass of oxygen required per unit decrease in bacterial biomass
from endogenous respiration 1.0 mgOjmgB
Mass of oxygen required per unit decrease in protozoa mass
from endogenous respiration 1.0 mgOjmgP
Inhibition constant for nutrient N, 3.0
Reaeration rate constant varies, near
0.25 h -,
Nutrient N [ half-saturation concentration for bacterial growth 20.0 mg 1-'
Nutrient N 2 half-saturation concentration for bacterial growth 20.0 mg \-,
Bacterial half-saturation concentration for protozoa growth 12.0 mg \-1
Volume flow in river reach [L 3 T- '] varies
Flow velocity in river reach [L T- I] varies
Proportion of COD that is nutrient N I varies
Maximum growth rate of bacterial biomass resulting from
nutrient N,
"max
rBN 2 Maximum growth rate of bacterial biomass resulting from
nutrient N 2 0.10 h- I
Maximum growth rate of protozoa mass resulting from
feeding on bacteria 0.36 h-[
PB Endogenous respiration rate for bacterial biomass 0.06 h- '
Pp Endogenous respiration rate for protozoa mass 0.07 h- I
In addition to the endogenous respiration rates, the first-order coefficients

include K Z2 ' the fraction of zooplankton nitrogen N z converted to ammonia
nitrogen N 2 per unit time; K Pp' the fraction of phytoplankton nitrogen N p con-
verted to particulate organic nitrogen Nf. per unit time; and K pd , the fraction of
particulate organic nitrogen Nf. that is transformed to dissolved organic
nitrogen N~ per unit time. Model constants also include the maximum
212
Nitrite nitrogen,
N0 2 -N
Dissolved
organic
nitrogen
Nitrobacter
bacteria
Particulate
organic
nitrogen
Zooplankton Phytoplankton
nitrogen nitrogen
FIGURE 6.11 Aerobic nitrogen cycle.
phytoplankton nitrogen uptake rates and half-saturation constants for ammonia

nitrogen (J1Np N2' K~,) and nitrate nitrogen (J1NpN 4 , K~.), and the zooplankton
grazing rates (J1 Nz Np' K~p)' Inhibition by ammonia of nitrate uptake by phyto-
plankton has been observed and could be included in the model (eqns. 6.77 and
6.78), as could ammonia regeneration by phytoplankton and the temperature
and light dependence of phytoplankton uptake rates and zooplankton grazing
rates. The constants Q~ [M M - 1] are the previously defined coefficients indica-
ting, for this model, the mass of nitrogen of form i required per unit increase in
the mass of material j resulting from the uptake of nutrient N (if other than 0.
The model, as illustrated at the end of this section, has 22 constants and
requires ten initial concentrations [M L - 3] for the seven nitrogen compounds
(N~, Nf, N 2' N 3, N 4' N p , N z) and the three species of bacteria (B Ns ' B Nb , B H ).
All first-order constants have dimension T-l, as have all maximum growth
rates J1. Hence each term in each equation has dimensions M L - 3 T - 1. The
values of most of these constants are available in the literature.
This combination of first-order Michaelis-Menten models has been com-
pared with the strictly first-order kinetic models (section 6.5) and has been
found to fit observed data more accurately. Whether or not the increased
accuracy, and expense, is warranted will depend on the management problem
being solved as well as on the availability of data (Harleman, 1978).
213
I. Ammonia nitrogen, N 2
decrease in N 1 from growth

of NirrosomOllCls, BN ,
increase in N 1 from increase in N 1

growth of hetero- from regeneration
trophic bacteria, B H by zooplankton, N z
- J1NPN2(K~2N: N JN p
decrease in N 1 from
growth of phytoplankton, N p
2. Nitrite nitrogen, N 3
increase in N 3 from growth of decrease in N 3 from growth of

Nirrosomonus. B N , N irrobucrer. B Nb
3. Nitrate nitrogen, N 4
dN 4
----cit = aN3BNbJ1BNb K
(
N3
N3
+ N 3 ) B Nb J1NPN4(K~4N: NJN p
increase in N 4 from growth of decrease in N 4 due to growth

N irrobucler. BNb of phytoplankton. N p
4. Phytoplankton nitrogen, N p
dN
----cit
p
= J1Np N2(KN
P + N
2) N p + J1N pN4(KN 4 )N p
P + N
N2 2 N 44
increase in N p from increase in N p from

phytoplankton uptake phytoplankton uptake
of ammonia nitrogen. N 1 of nitrate nitrogen, N 4
- J1N2NP(K~pN: NJN z
decrease in N p from decrease in N p from
zooplankton grazing endogenous respiration
214
5. Zooplankton nitrogen, N z
d~z = IlNzNP(K~:~ NJN z

increase in N z from decrease in N z due decrease in N z from
zoopJank ton grazing of to regeneration of endogenous respiration
phytoplankton, N p ammonia nitrogen. N 1
6. Particulate organic nitrogen, N~

dN~
dt
increase in N from r concentration of Nr
endogenous respiration of transformed to
phytoplankton. N p • and death dissolved organic
and defecation of zooplankton, N z nitrogen, N1
7. Dissolved organic nitrogen, N1

dN1
- aN~BHIlBH
N" (KN1 +1N1) B H
dt N
increase in N1 from decrease in N1 from growth
solution of particulate of heterotrophic bacteria,
organic nitrogen, N'; B H , and production of
ammonia nitrogen. N 1
8. N itrosomonas, B Ns
dBNs
-cit =
(N+ N
IlBNs K
2 )
B Ns
N, 2
growth of B N , from uptake reduction of BN , from

of ammonia nitrogen, N 2 endogenous respiration
9. Nitrobacter, B Nb
dB Nb
~ =
(N+ N
IlBNb K N
3 )
B Nb
3 3
growth of B Nb from uptake reduction of BNb from
of nitrite nitrogen, N 3 endogenous respiration
10. Heterotrophic bacteria, B H

N
d:tH IlBH (K N1 ; N1 )B H
growth of B H from uptake of reduction of B H from
dissolved organic nitrogen, N I endogenous respiration
215
6.7. QUALITY ROUTING IN UNSTEADY FLOW
All of the models described in this chapter are based on an assumption of

hydraulic steady state, i.e. dQ/dt = O. Whenever the flow changes rapidly with
time, as may happen when a hydraulic transient is admitted to the stream, it
becomes necessary to give greater attention to temporal description of the flow
regime and its effect on the transport and mixing of quality constituents identi-
fied with the flow. Such cases of interest include the routing of load patterns
associated with diurnal fluctuations in wastewater discharges, "pollutographs"
of storm runoff to stream systems, accidental spills of pollutants, and periodic
flow oscillations in the tidal reaches of stream systems.
6.7.1. Hydrodynamic Equations

To provide the necessary spatial and temporal description of the quality
routing problem it is required that we include the equations for unsteady open-
channel flow (following the one-dimensional approach of Vasiliev (1976»:
Equation of continuity
oA oQ
at + ox = q, (6.82)
where
A is the area of flow cross section [L 2 ],
Q is the discharge [L 3 T- I ], Q = UA,
U is the mean velocity [L T- I],
q is the lateral discharge along the axis of flow (sources or sinks) [U T - I l
Momentum equation
o 0
ot (AU) + oX (QU) = -gA
(0:ox + UIUI HeOP)
C 2R + P oX ' (6.83)
where
is the acceleration due to gravity [L T - 2],
g
is the water surface elevation [L],
Z
C is the Chezy resistance coefficient [L 1/2 T- I ],
R is the hydraulic radius [L],
P is the density [M L - 3],
He is the mean channel depth [Ll
216
6.7.2. Convection- Diffusion Equation
Following the traditional approach in water quality modeling, we may couple
the hydrodynamic equations to water quality through the convection-diffusion
equation, which for conservation of temperature may be written as
a a a ( a0)
at (A0) + ax (Q0) = ax AD ax - K T A(0 - 0 E) + q0), (6.84)
where
o is the temperature (QC),
D is the diffusion coefficient [U T- 1 ],
KT is the heat transfer coefficient [T- 1 ],
0E is the equilibrium temperature (QC),
01 is the temperature of lateral inflow (sources or sinks){QC).
For a conservative substance carried with the flow the mass balance equation is
a a a ( as)
at (AS) + ax (QS) = ax AD ax + qSI, (6.85)
where S is the concentration of a conservative substance [M L - 3]. These

equations must be supplemented with an equation of state relating water
density p to temperature and salinity:
p = j(0,s). (6.86)
To describe the hydrochemical and hydrobiological processes in the stream,

(6.85) may be modified by adding terms representing the appropriate chemical-
biological linkages and interactions. This step follows closely the structure of
QUAL II.
6.7.3. Solutions for Unsteady Flow

Numerical solutions of (6.82)-(6.86) have been obtained by Vasiliev and
Voyevodin (1975) and Vasiliev et al. (1976) for unsteady flows in open-channel
networks.
Depending on the particular case examined, it may prove practical to
simplify the equations before seeking a solution. For example, in stream systems
or well mixed estuaries it may be reasonable to neglect the interdependence
of the hydrodynamic and advection-diffusion equations through the equation
of state (6.86), thus uncoupling the equation set and allowing independent
solution of the hydrodynamic problem. This approach has been successful in
the modeling of channel networks and shallow estuaries with braided channels
(Orlob, 1972).
217
Additionally, longitudinal dispersion may prove troublesome because it is
empirical in nature and some numerical techniques introduce" mixing" effects
(Bella and Dobbins, 1968). These difficulties have sometimes been successfully
overcome by considering that longitudinal dispersion is negligible near the
downstream boundary (Daily and Harleman, 1972; Vasiliev, 1976). Problems
with numerical mixing, especially troublesome in explicit solution techniques,
have sometimes been partially resolved by relying on a combination of the
numerical and physical processes to describe what is actually observed in the
water body. This approach has not been wholly successful since it does not
address the fundamental question of a rigorous description of the internal
mixing processes.
6.8. PERFORMANCE OF RECEIVING-WATER QUALITY

SIMULAnON MODELS
The stream quality prediction models outlined in this chapter provide only an
introduction to the variety of types of steady state model used to simulate or
predict water quality in water bodies. The relative reliabilities of various con-
stituent concentration predictions provided by current water quality models
are listed in Table 6.3. The table also summarizes the major impacts of various
constituents in natural river systems.
TABLE 6.3 Quality Impacts and Current State of Modeling.
Component Quality Model Reliability

Impacts Streams Estuaries
Transport. steady state Good Fair

Transport. dynamic Fair Poor
Conservative substances Water supply Fair Fair
Suspended solids Water supply Poor Poor
Recreation
Bacteria. protozoa Water supply Fair Poor
Recreation
BOD, DO Aquatic ecosystem Good Good
Simple chemicals Water supply Fair Fair
and metals Ecosystem
Synthetic chemicals Water supply Poor Poor
and complex metals Ecosystem
Nutrients Aquatic ecosystem Fair Fair
Recreation
Eutrophication (algae) Recreation Fair Poor
Zooplankton and fish Recreation Unsatisfactory Unsatisfactory
Temperature Aquatic ecosystem Fair Fair
Virus Watcr supply Unsatisfactory Unsatisfactory
Recreation
Floating substances Recreation Unsatisfactory Unsatisfactory
Color and turbidity Recreation Unsatisfactory Unsatisfactory
Water supply
218
6.9. SUMMARY
This chapter has been an introduction to water quality simulation modeling

of rivers. The modeling procedures or approaches have ranged from fairly
simple one-dimensional steady state approximations to procedures that are
considerably more involved.
These simulation models, or their more complex extensions, are relatively
crude approximations of the interactions among various constituents that occur
in water bodies. Yet in spite of their current limitations, they are the only
reasonable means available for predicting surface water quality. The state of
the art in water quality modeling and an understanding of the physical, chemical,
and biological processes that affect water quality are improving rapidly.
Readers interested in pursuing this area of modeling activity are encouraged
to study in greater detail alternative water quality modeling approaches and
solution procedures, many of which are cited in the reference list.
APPENDIX 6.1. RATE COEFFICIENTS OF PRINCIPAL BOD DECAY PROCESSES.
Processes Equations Coefficients Illustrations Temperature

Coefficients
Definitions Comments
Biochemical Oxidation of carbonaceous K. carbonaceous Depends on type and concentration of Second stage (BOOn) 1.047-1.135
oxidation of organic matter: BOD decay carbonaceous organic maller, time,
organic matter coefficient temperature. pH, inorganic and organic --' =J
dL
in laboratory - ~ - KL nutrients, and toxic materials. ~
dt Second-order and higher-order models
tv are also used.
o
...... L, = L o exp( - Kt) o
'00 en v ..
Y ~ Lo(l - 10-")
Time ITI
Y is the carbonaceous BOD
exerted in time t [M L - 3].
k 1 = kI20nlf-1O.
r is the temperature (cC).
Biochemical Oxidation of carbonaceous K d , deoxygenation Depends mainly on type and 1.047-1.075

oxidation of organic matter, L: rate coefficient concentration of organic matter
M
organic matter (for soluble remaining. and temperature. May be
L = L o exp( - KdxlU) ....J
in streams carbonaceous obtained by determining the BOD of
organic material) filtered samples. ~
crl] o
o
Cll
L = L o exp( - K,x/U) Kr , total removal Depelld~ on concentration of lotal
rate coefficient for organic matter and temperature, and is
c
carbonaceous affected by sedimentation, scour. and Distance downstream I U.
•
organic matter volatilization of certain organic 1.00-1.075
[rl] compounds.
K $ =K r - K d K~. coeffieient for Depends on concentration of organic c

rate of removal by matter thal can settle. and stream
o
c '';:; _
mechanisms other velocity. Has a small positive value in a Q;l ~M
cn ...... .:....J
than oxidation rapidly moving stream (U > 0.3 m s - I). 1.106 for
[ r I] and a large value in a more sluggish
e u~ ~
.~ Nirrosomonas
Z c 1.072 for
slrea m (U < 0.18 m s - I). o
u I\'irrohacler
(continued over)
APPENDIX 6.1. (continued)
Oxidation of nitrogenous organic K n , rate coefficient Depends on type and concentration of
matter, N: for nitrification nitrogenous organic matter. temperature,
[r ' ] dissolved oxygen level, pH.
N = No exp( - K"xjU)
or
N = No! 1+ exp[K"No(t,/, -i)]}
I", is the time for half completion

or reaction.
N
Benthic dL b Kb K b , benthic Depends on the diffusion of soluble
~ decomposition - - = - - Lb demand rate oxidizable products of anaerobic
dr 1+" coefficient [T - I] decomposition into a stream. aerobic
L b is the areal BOD of the stabilization of the top layers of the
bottom deposits [M L -, T- I], benthos, the transport of degradable
" is the coefficient of retardation. particles back into the flowing body of
water, etc.
Photosynthetic 2P 00 )
Pm' maximum rate Depends mainly on solar energy;
oxygen P(X.I) = Pm - I hn cos 2nnt of photosynthetic therefore it is affected by the availability
( 7[ + 11= 1
production and oxygen production and intensity of sunlight, the availability
respiration (see [M L- J r ' ] of a carbon source and nutrients, and
temperature. The respiration of plants
r' = t - I s P2
also Chapter 4)·
is a reverse reaction, with oxygen being
ib:~~
Q.C\ C\.
b _ 4rr/p consumed in the oxidation of organic
" - (rrjp)' _ (2rrn)' cos nrrp matter. Time [T]
Reaeration dC _ KLA (C, _ C) KL , mass transfer

<5 Q) Interface
d/- v coefficient [L T - I]
5~ I
';: ~ Watefl Air
III Q)
l=
c.
K a , reaeration Depends mainly on turbulence and c_
dD __ IS D = _ K,D ~.~
1.0241
d/- H coefficien t [T - I] temperature of air and water, and the c t:
presence of surface-active agents.
8 ~I~~-=-:----- -.
Distance downstream [L1
APPENDIX 6.2. PREDICTIVE MODELS FOR REAERATION COEFFICIENT.
Investigators Formulas for k, (base 10) (day - I) Comments and Units of Variables
Streeter and Phelps (1925) k, CU"H,;2 U is the mean velocity of flow (m s - I)

H o is the mean depth of water above extreme low water (m)
C, n are empirical constants
n ranged from 0.57 to 5.40
r varied between 23.96 and 13.06.
O'Connor and Dobbins (1956) Anisotropic turbulence D M is the coefficient of molecular diffusivity (m 2 d-')
IV
IV k, = 356.7D~·5S0.25H- 1.25 D M(20T) = 0.000181 m 2 d- I
k'(20 C) = 4.85"'·25 H- 125 S is the slope of the river channel (m m - 1)
Isotropic turbulence H is the mean depth of flow (m)
k, = 127D fi/U o.5H-1.5 U is the mean velocity of flow (m s - 1).
o
k,f2o n = 1.71 U 05 H-1.5
Krenkel and Orlob (1962) k,t20T) = 2.4·10-2E1·32IH-2.32 £ 1 is the longitudinal dispersion coefficient (m 2 min - I)
k,t200C) = 1.98· 10 - 'Ed· 40B H- 0.66 £2 is the energy dissipation per unit mass of flow (m 2 min - 3)
H is in meters.
A multiple correlation analysis was applied to the data from experiments
in a laboratory flume.
Krenkel and Orlob (1963) k, = 2.6E~·237 H- 2.087 E,. is the average vertical eddy diffusivity (m 2 s -I)
H is in meters.
A regression analysis was applied to data from a laboratory flume.
Churchill et al. (1962) k'(20'C) = 2.18U 09b9 H -1.673 H is in meters, U in m s - I.
k,(20"c) = 2.26UH - 1.67 A dimensional analysis and multiple regression techniques were applied
to the data from several tributary streams in the upper Tennessee Basin.
Owens 1'1 al. (1964) H is in meters, U in m s - I.
k'f20Cl = 2.316UO."7H -1.85 Regression analysis was performed on the data from Gameson el al.
(1955), Churchill et al. (1962), and Owens et al. (1964).
k'(20-C) 3.0UO. 73 H -1.75 Regression analysis was performed on the restricted data set from the
Water Pollution Research Laboratory.
(cO/lli/lIIed over)
APPENDIX 6.2. (continued)
Investigators Formulas for k. (base 10) (day-I) Comments and Units of Variables
Langbein and Durum (1967) k. llOT ) = 2.23 UH - 1.33 H is in meters, U in m s - 1.

Regression analysis was performed on the field data of Churchill el al.
(1962), together with the data O'Connor and Dobbins (1956), the
laboratory data of Streeter el al. (1936), and the data of Krenkel and
Orlob (1962).
Isaacs and Gaudy (1968) H is in meters, U in m s - I.
IV k. 120 "CI = 2.22UH -1.5 A least-squares analysis was applied to the field data of Churchill el al.
IV
IV ( 1962).
k. 120 "C) = 2.064U H- '.5 Obtained from 29 of the 30 data points of Churchill el at. (1962).
k.120oq = 1.347U H- 1.5 A least-squares analysis was used on the data of Krenkel (1960) and
Krenkel and Orlob (1962).
k. llO -C ) = 1.685U H- 1.5 Obtained from the laboratory data of Isaacs and Gaudy (1968).
Isaacs and Maag (1969) k. = 1.645(XI(X2UH-15 (x, is a nondimensional variable that changes with the channel geometry.
k. = 2.057UH-1.5 (X2 is a nondimensional variable that is a measure of surface velocity.
H is in meters, U in m s - I.
For the data of Churchill el al. (1962) the average value of (x, was 1.078,
and the average of (X2 was 1.16.
Thackston and Krenkel (1969) k. = 10.8(1 + Fro. 5 )U. H- I H is in meters, U in m s- '.

Fr = U(gH)-U.5 is the Froude number
U. is the shear velocity (m s - ')
g is the acceleration of gravity (m l S-I).
Cadwallader and McDonnell (1969) k. = 25.7Eo. 5H- 1 ED is the energy dissipation per unit mass of flow (m l s - 3).
H is in meters
A multivariate analysis was performed using a restricted data set, and the
data of Churchill el al. (1962), Owens el al. (1964), and the Water
Pollution Research Laboratory channel data.
Negulescu and Rojanski (1969) ka = 4. 74U 0.85 H - 0.85 Developed from a flume with depth less than 0.5 m and velocity not
greater than 1.2 m s - '.
Padden and Gloyna (1971) k. 120 ,c) = 1.963U O . 703 H-1.0545 H is in meters. U in m S-l.
During this experimental study performed in a research flume, the
velocities ranged from 0.03 to 0.14 m s - I and depths from 0.035 to 0.19 m.
Bennett and Rathbun (1972) ka = 2.33U 0.074 H -1.865 H is in meters, U in m s - I.
The equation was developed using the data of Churchill et al. (1962) and
Owens et al. (1964).
Tsivoglou (1967) kaI20'C) = 1.63t!.Ht- 1 t!.H is the water surface elevation change (m)
N Tsivoglou and Wallace (1972) t is time of flow (days)
N
W Corrected to 20°C using I), = 1.0241.
A gas tracer technique was applied to non tidal streams.
Lau (1972a.b) kaI20'C) = lO88.64U~U-2H-l H is in meters, U in m s - 1.
U * is the friction velocity (m s - 1)
A dimensional analysis was performed using the data of Krenkel (1960),
Churchill et al. (1962), and Thackston and Krenkel (1969).
Foree (1976) kaI20'C) = 0.116 + 2147.8S1. 2 S is the channel slope (m m -1)

Corrected to 20°C using 0, = 1.0241.
A gas tracer technique was used on streams in Kentucky.
224
REFERENCES
Bella, D. A., and W. E. Dobbins (1968). Difference modeling of stream pollution. Proceedings of
American Society of Or-if Engineers, Journal of Sanitary Engineering Dicision 94(SA5): 995-1 0 16.
Bennett, J. P" and E. R. Rathbun (1972). Reaeration in open-channel flow. US Geological Surver
Professional Paper 737.
Cadwallader, T. E., and A. J. McDonnell (1969). A multivariate analysis of reaeration data. Water
Research 3: 731-742.
Camp, T. R. (1963). Water and Its Impurities (London: Chapman and Hall).
Chen, C. W., and G. T. Orlob (1975). Ecological simulation of aquatic environments. Systems
Analvsis and Simulation in Ecology vol. 3. ed. B. C. Patten (New York, NY: Academic Press),
ch. 12.
Churchill, M. A .. H. L. Elmore, and R. A. Buckingham (1962). The prediction of stream reaeration
rates. Tennessee Valley Authority, Challanooga, TN Report.
Daily, J. E.. and D. R. F. Harleman (1972). Numerical model for the prediction of transient water
quality in estuary networks. MIT Department of Ch-il Engineering, R. M. Parsons Laboratory
Report 158, p. 226.
Dobbins, W. E. (1964). BOD and oxygen relationships in streams. Proceedings ofAmerican Society
of Or-if Engineers, Journal (jfSanitary Engineering Dil'ision 90(SA3): 53-78.
Fair, G. M. (1939). The dissolved oxygen sag-An analysis. Sewage Works Journal 11(3) : 445.
Foree, E. G. (1976). Reaeration and velocity prediction for small streams. Proceedings of American
Society of Ci/-il Engineers, Journal of Em-ironmental En.qineering Dil-ision 102(EE5):937-952.
Frankel, R. J., and W. W. Hansen (1968). Biological and physical responses in a freshwater dissolved
oxygen model. Adrances in Water Quality Improvement, eds. E. F. Gloyna and W. W. Ecken-
felder (Austin, TX and London: University of Texas Press), pp. 126-140.
Gameson, A. L. H., G. A. Truesdale, and A. L. Downing (1955). Reaeration studies in lakeland
beck. Journal of Institution (jf Water Engineers 9: 571 ·594.
Hansen, W. W., and R. J. Frankel (1965). Economic evaluation of water quality, a mathematical
model of dissolved oxygen concentration in freshwater streams. Sanitary Engineering Laboraton'.
Unil-ersity of Cali/t)rnia. Berkeley Report 65-11.
Harleman, D. R. F. (1978). A comparison of water quality models of the aerobic nitrogen cycle.
Research Memorandum RM-78-34 (Laxenburg, Austria: International Institute for Applied
Systems Analysis).
Isaacs. W. P" and A. F. Gaudy (1968). Atmospheric oxidation in a simulated stream. Proceedings
(j/ American Society 0/ Civil Engineers, JOl/rnal 0/ Sanitary Engineering Division 94 (SA2): 319-
344.
Isaacs, W. P., and J. A. Maag (1969). Investigation of the effects of channel geometry and surface
velocity on the reaeration rate coefficient. Purdue Unil-ersity, Lafayelle. IN Engineering Bulletin
vol. 53.
KrenkeL P. A. (1960). Turbulent diffusion and the kinetics of oxygen adsorption. Ph.D. Dissertation,
Unil'ersity 0/ Cali/ornia, Berkele.v.
Krcnkcl, P. A" and G. T. Orlob (1962). Turbulent diffusion and the reaeration coefficient. Pro-
feedill.'!,\' ofAmeril'lln Soeietl' o/Ocif EII.t/ineers. Journal ofSanitary En.'!ineerinq Dil'ision 88(SA2):
53-83.
Krenkel, P. A" and G. T. Orlob (1963). Discussion of turbulent diffusion and the reaeration co-
efficient. Transactions o/American Society o/Cicil Enqineers 128(3):293-334.
Langbein, W. B., and W. H. Durum (1967). The reaeration capacity of streams. US Geoloqical
Sun'ey Circl/lar 542.
Lau, L. Y. (1972a). A review of conceptual models and prediction equations for reaeration in open-
channel flow. Department ofEnl'ironmenl. Ollawa Technical Bulletin 61.
Lau. L. Y (1972h). Prediction equation for reaeration in open-channel flow. Proceedin.'!s ofAmeri-
can Society of CiI·if Enqineer.\, Journal of Sanitary Engineering Divi.'1·on 96(SA6): 1063-1068.
225
Negulescu. M., and Y. Rojanski (1969). Recent research to determine reaeration coefficient. WaleI'
Research 3: 189-202.
O'Connor. 0.1. (1962). The effect of stream flow on waste assimilation capacity. Proceedinqs or I7lh
Industrial Waste Conference, Purdue Uni1'ersil.\' LalavelIe, IN, 1-3 May. Engineering Extension
Series No. 112, pp. 608-629.
O'Connor. D. 1.. and D. M. Di Toro (1968). The distribution of dissolved oxygen in a stream with
time-varying velocity. WaleI' Resources Research 4(3): 639-646.
O'Connor, D. J., and D. M. Di Toro (1970). Photosynthesis and oxygen balance in streams. Pro-
ceedinqs 0.1 American Sociely of01'il Enqineers, Journal ()l'Sanilary Enqineering Dirision96(SA2):
547-571.
O'Connor, D. J.• and W. E. Dobbins (1956). The mechanism of reaeration in natural streams. Pro-
ceedings ofAmerican Sociely orCh'il Engineers, Journal orSanilary Enqineerinq Dirision 82: 1-30.
O'Connor, D. L R. Y. Thomann, and D. M. Di Toro ([976). Eco[ogic models. Syslems Approach
/0 Waler MOIl(/{temenl. ed. A. K. Riswas (New York. NY: McGraw-Hili). ch. 8.
Orlob, G. T. (1972). Mathematical models of estuarial systems. Proceedings or International Con-
jerence on Wall.'l' Resource Syslems, o lIa I\'a , May vol. I, pp. 78--128.
Owens, M., R. W. Edwards, and 1. W. Gibbs (1964). Some reaeration studies in streams. Inler-
nalional Journal o{ Air and WaleI' Pollulion 8:469-486.
Padden, T. J., and E. F. Gloyna (1971). Simulation of stream process in a model river. Cenler/or
Research in WaleI' Resources, UnirersilY of Texas al Auslin Report EHE-70-23, CR WR-72.
Rathbun, R. E. (1977). Reaeration coefficient of streams-State of the art. Proceedings o{American
Sociely or Ol'il Engineers, Journal or Hydraulics Dit'ision 103(HY4): 409-424.
Russell, C. S. (ed.) ( 1975). Ecolo.qical Modeling in a Resource Managemenl Framework. Resources/or
Ihe FUlure Working Paper QE-I (Baltimore. MD:Johns Hopkins University Press).
Stehfest, H. (1977). Mathematical modeling of self-purification of rivers. Projessional Paper
Streeter, H. W.. and E. B. Phelps (1925). A study of the pollution and natural purification of the
Ohio River. US Public Heallh Sen'ice. Washinqlon, DC Bullelin 146.
Streeter, H. W., C. T. Wright. and R. W. Kehr (1936). Measures of natural oxidation in polluted
streams, Part III, An experimental study of atmospheric reaeration under stream flow conditions.
Sewage Works Journal 8(2) :282- 316.
Texas Water Development Board (1970a). DOSAG I simulation of water quality in streams and
canals. Program Documenlalion and User's Manual.
Texas Water Development Board (l970b). QUAL I simulation of water quality in streams and
canals. Proqram Documenlalion and User's Manual.
Texas Water Deve[opment Board (1971). Simulation of water quality in streams and canals.
Reporl 128, PB 202-975.
Thackston. E. L.. and P. A. Krenkel ([ 969). Reaeration prediction in natural streams. Proceedings
0/ American Sociely or Cit'll Engineers, Journal or Sanilary Engineering Dil'ision 95(SA I): 65-94.
Thomann, R. Y. (1972). Syslems Analvsis and WaleI' QualilY Managemenl (New York, NY:
Environmental Research and Applications, Inc.).
Thomas, H. A. (1948). Pollution load capacity of streams. WaleI' and Sewage Works 95 :409.
Tsivoglou, E. C. (1967). Tracer measurement in stream reaeration. Federal Waler Pollution Conlrol
Adminislralion, US Departmenl ol'Inlerior, Washingloll, DC Reporl.
Tsivoglou, E. C .. and J. R. Wallace (1972). Characterization of stream reaeration capacity. Office
or Research and Monilorinc;. US Enrironmelllal Proleclioll Aqency. Washinqlon, DC Reporl
EPA R3-72-0 12.
Yasiliev, O. F. (1976). Mathematical modeling of water quality in rivers and water bodies. Pro-
ceedings 01'41h All-Union Hydrological Congress vol. 9 (Leningrad: Gidrometizdat), pp. 161-168.
Yasiliev, O. F .. and A. F. Yoyevodin (1975). Mathematical modeling of water quality in systems of
open channels. Dinamica Spoloshnoi Sredl'. Inslilul Gidrodinamiki SO AN USSR, NOl'osibirsk
No. 22, pp. 73-88.
226
Vasiliev, O. F., A. F. Voyevodin, and A. A. Atavin (1976). Numerical methods for the calculation
of unsteady flow in systems of open channels and canals. /nlernalivnal Symposium on Unsleady
Flo ..... in Open Channels,/AHR, Ne .....caslle upon Tyne, April. Paper E2, pp. J 5-22.
Water Resources Engineers, Inc. (1967). Prediction of thermal energy distribution in streams and
reservoirs. Reporr 10 California Deparlmenl of Fish and Game. June (rel'ised Augusl /968),90 pp.
Water Resources Engineers, Inc. (1973). Computer program documentation for the stream quality
model QUAL II. Reporl 10 US Em'ironmental Prolection Agency, WashingfOn, DC.
Wilson, G. T., and N. MacLeod (1974). A critical appraisal of empirical equations and models for
the prediction of the coefficient of reaeration of deoxygenated water. Waler Research 8: 341-366.
CHAPTER 6: NOTATION
B rate of addition of BOD to overlying water from bottom deposits

D dissolved oxygen saturation deficit
Db rate of removal of oxygen caused by benthic demand and plants
E turbulent transport coefficient; longitudinal dispersion coefficient
Ev evaporation rate
L carbonaceous BOD
Ld total areal BOD of benthos
n Manning coefficient
NI, N2 concentration of inhibited nutrient, of inhibiting nutrient
P period of sunlight
P rate of oxygen production; protozoa mass
Pr precipitation rate
q lateral discharge along axis of flow
R algal respiration rate; hydraulic radius
S benthic respiration rate
S. slope of energy gradient
U,U· average stream velocity, bed shear velocity
w concentration of pollutant
Z water surface elevation
!XI fraction of respired algal biomass resolubilized as ammonia nitrogen
!X2 fraction of algal biomass that is phosphorus
!X 3 to !X6 rates of oxygen uptake
f3 toxicity coefficient
f31' f32 rates of oxidation
y density of water
e temperature
A. light extinction coefficient
p respiration rate
0'1 algal settling rate
0'2,0'3 benthos source rate for phosphorus, for ammonia.
One-dimensional Models for Simulation
7 of Water Quality in Lakes and Reservoirs
G. T. Orlob
7.1. NEED FOR MATHEMATICAL MODELS
Mathematical modeling of surface water impoundments received its greatest

impetus in the early 1960s with an awakened interest worldwide in environmen-
tal conservation and pollution control. Among the more obvious problems in
need of solution were those identified with accelerating nutrient enrichment of
both natural and artificial impoundments. Such widely publicized examples as
Lake Erie and the Zurich See, which were often characterized as "dead '.'-
beyond hope of recovery-drew the attention of environmentalists, pollution
control regulatory authorities, and scientists to this special class of environmen-
tal problems. "Eutrophication" became synonymous with extremes in quality
degradation, rather than merely a change in state of the biodynamic cycle.
Efforts were intensified to find practical solutions for eliminating or significantly
reducing nutrient accretions to lakes and reservoirs, often without an under-
standing of the consequences to the ecological and water quality balances. Even
though technology then existed for removal of phosphorus from wastewaters,
denitrification, and regulation of flows, there was uncertainty concerning the
effectiveness of such measures in achieving desired responses in the water body.
This lack of general understanding of the fundamentals of hydrodynamic,
water quality, and ecological behavior of natural water bodies, particularly
lakes, and of their responses to external stimuli encouraged additional field
work. This work produced data from water bodies and stimulated development
of analytical techniques suited to prediction of changes resulting from alterna-
tive strategies for management of these important water resources. One such
technique is the one-dimensional mathematical model for prediction of thermal
energy changes in a stratified impoundment, which, as will be shown later, is
an essential first stage leading toward development of a general capability to
describe the responses of water quality and ecological systems of natural and
artificial impoundments.
227
228
7.2. ONE-DIMENSIONAL APPROXIMAnON
The earliest limnological studies of the thermal structure of lakes (Hutchinson,

1957) recognized that during the period of greatest stratification, especially
with the formation of a distinct thermocline, there was comparatively little
variation in temperature over a horizontal plane parallel to the water surface.
Although in large lakes longitudinal and lateral gradients were easily identified
with transient phenomena, e.g. extreme hydrological episodes and wind dis-
turbances, these were often rapidly dissipated by gravitational forces and the
lake was restored to a condition in which the dominant variations in tempera-
ture over the greater part of the annual cycle were in the vertical direction.
(Strictly speaking, we should refer to density rather than temperature although
observations were seldom made of density.) For lakes of small to medium size,
say less than 50 km along the major axis, the representation of the temperature
structure as one-dimensional has been found by experience to be reasonable.
Exceptions are lakes that are relatively narrow and deep and of relatively
small volume compared with the peak rate of inflow.
A criterion that gives some guidance to the applicability of the one-
dimensional approximation was suggested by Water Resources Engineers, Inc.
(1969). The lake is characterized by a densimetric Froude number that compares
the inertial force, represented by an average flow-through velocity, with the
gravitational force tending to maintain densimetric stability:
U 1
Fr = [(~p/PO)gdJl/2 <~' (7.1)
where
U = Q/bd is the average flow-through velocity [L T- 1J,

Q is the volumetric discharge [L 3 T-1J,
d is the average depth [LJ,
~p is the density difference over depth d [M L - 3 J,
Po is the reference density [M L - 3 J,
9 is the acceleration due to gravity [L T- 2].
If the length and volume of the impoundment are introduced as characteristic

parameters, (7.1) becomes
Fr = ~Q (pO)I/2 (7.2)
d V gfJ '
where
I is the length [LJ,

V = lbd is the volume [L 3 J,
fJ = ~p/d is the density gradient [M L -4].
TABLE 7.1 Stratification Characteristics of Selected Impoundments.
Impoundmenl Location Length Mean Depth Discharge: Densimetric Comments

I (km) d(m) volume Froude Number
Q/V (S-I) Fr
Lake Roosevelt Washington 200 70 5.0.10- 7 0.46 Weakly stratified

Hungry Horse Reservoir Montana 47 70 \.2.10- 8 0.0026 Strongly stratified
Wells Reservoir Washington 46 26 6.7.10- 6 3.8 Fully mixed
Cayuga Lake New York 60 55 4.9.10- 9 0.015 Strongly stratified
Lake Piiijiinne Finland 120 17 3.3.10- 10 0.0004 Strongly stratified
Ross Lake Washington 32 122 \.2.10- 7 0.0042 Strongly stratified
N
N
'0
230
Deep, well stratified impoundments, for which one-dimensional models are
best suited, are those for which Fr ~ l/n. Weakly stratified impoundments,
for which a two-dimensional representation is sure to be necessary to describe
the temperature (density)-velocity relationship, are generally those for which
0.1 < Fr < 1.0. Fully mixed systems are defined by Fr > 1.0.
Illustrations of this classification system are summarized in Table 7.1.
Typically, Hungry Horse Reservoir in Montana (l = 47 km, d = 70 m,
Q/V = 1.2· 10- 8 s -I) is a strongly stratified system, i.e. Fr = 0.0026 ~ lin.
In contrast, Wells Reservoir on the Columbia River (I = 46 km, d = 26 m,
Q/V = 6.7· 10- 6 s - I), studied by Raphael (1962a) in one of the early attempts
to model the temperature regime in reservoirs, is classified as fully mixed:
Fr = 3.8. Between these extremes, Lake Roosevelt behind Grand Coulee Dam
(l = 200 km, d = 70 m, Q/V = 5.0.10- 7 S-I) is a weakly stratified reservoir:
Fr = 0.46. Indeed, observations of this impoundment reveal isothermal
.. planes" inclined downstream toward the outlet, a physical circumstance that
clearly indicates strong coupling between hydrodynamic behavior and density
changes brought about by heat influx through the air-water interface along
the major axis of the reservoir (Water Resources Engineers, Inc., 1969).
This chapter will deal exclusively with strongly stratified impoundments
(Fr ~ lin), for which the assumption of horizontal isothermal planes through
the thermally stratified water body is reasonably consistent with reality.
This case will be regarded, therefore, as one-dimensional in the mathematical
sense and it will be implicit that mixing of heat introduced in the horizontal
plane is instantaneous and complete. The only gradients treated will be those
along the vertical axis. We begin with consideration only of thermal energy
changes induced by advected flows, diffusional transport, and heat exchange
across the air-water interface. The models that have evolved from these as-
sumptions, restrictive though they may seem at first, are an important class,
covering many situations of practical interest and providing basic foundations
for further advances in water quality and ecological modeling.
7.3. BRIEF REVIEW OF DEVELOPMENT
7.3.1. Early Attempts to Simulate Lake Temperatures

The first important attempts to describe mathematically the annual thermal
cycle in lakes appear to have been those of the pioneering limnologists, who,
having observed the development of thermal stratification in small ponds and
lakes, sought to quantify the vertical transfer of heat. McEwen (1929) was
among the first to estimate heat transfer coefficients using assumed temperature
profiles typical of thermally stratified lakes when a thermocline was evident.
Ertel (1954) designed a diffusional model of the thermocline formation, assuming
231
a constant coefficient of diffusion, independent of time or depth. Hutchinson
(1957), in his monumental work, A Treatise on Limnology, reported these efforts
and his own using data derived from field studies to explain quantitatively the
formation of the thermocline, that feature of the annual thermal cycle that is
so pronounced in its influence on physical, chemical, and biological responses
of the lake. However, these efforts were largely unsuccessful in producing
credible representations of the entire annual cycle of thermal energy changes
that had been so well documented in the annals of Iimnological research. The
most apparent deficiency at this point in the development of a quantitative
description of the thermal stratification process was an adequate account of
heat exchange through the air-water interface.
In the period after the Second World War, the booming construction of
large dams and impoundments worldwide focused some (if not sufficient)
attention on the impacts of such developments on the quality of waters released
downstream. Particular concerns developed where anadromous fish migrations
were likely to be affected by temperature changes and flow regulation, as in the
unusually intense development of the Columbia River in the northwestern
United States, a major salmon-producing river with an enormous hydro-
electric power potential. Under pressure of fisheries interests, power companies
sought to predict the modification that could be expected in the downstream
temperature as a result of storage and flow regulation. Raphael (1961, 1962a)
was among the first to devise a method for quantifying the thermal energy
budget ofan operating reservoir. His technique, which was carried out by tedious
manual calculation, gave specific attention to heat energy gains and losses by
advection, insolation, evaporation, and conduction. Results of his method,
applied first to the well mixed reservoirs of the Middle Columbia River
(Fr > 1.0), compared favorably with field observations. Sub~equently, he
adapted the method to predict temperatures of releases from deep reservoirs
like Oroville Reservoir in California (Raphael, 1962b). The method, while
giving reasonable estimates of downstream temperatures for specific projects,
was not capable of describing the distribution of heat energy within the
impoundment. Moreover, it was not developed in model form or computerized
for general use. Similar methods were developed and applied by Burt (1958,
1960, 1963) to several reservoirs in the Pacific Northwest, but like the Raphael
procedures, these were essentially manual and were not adopted by others.
7.3.2. Quantification of Heat Exchange

Among the first concerted efforts to quantify the heat exchange process, the
landmark work in the late 1960s of the Tennessee Valley Authority (TV A)
Engineering Laboratory under the direction of Rex A. Elder stands out as
unique. Elder's co-workers, W. O. Wunderlich and R. Gras, are credited with
the thorough and careful research leading to the authoritative report, Heat and
232
Mass Transfer Between a Water Surface and the Atmosphere (Tennessee Valley
Authority, 1972). This document served as an important building block for
parallel development of the first working mathematical models of the thermal
stratification process in deep reservoirs. It is the primary source for the quan-
titative description of the heat exchange process to be described in this chapter
(see also the appendix to Chapter 5).
7.3.3. First Attempts at One-Dimensional Temperature Modeling

Formal mathematical modeling of temperature changes in deep, stratified
impoundments appears to have been stimulated by the efforts of both the TVA
and the California Department of Fish and Game, the latter agency being
concerned with the effects of large impoundments, e.g. Oroville Reservoir, on
salmon migration. In 1965 the Department contracted with Water Resources
Engineers, Inc. to develop a predictive model (Orlob, 1965). In the following
year the TVA and WRE collaborated in developing the model. The TVA
quantified heat exchange phenomena and conducted field studies on several of
its reservoirs to provide data for model calibration and validation (Elder and
Wunderlich, 1968). This combined effort culminated in a working model that
was first applied to Fontana Reservoir in the TVA system; it was later revised
(WRE, Inc., 1968) as a result of experience with several reservoirs in the north-
western United States (WRE, Inc., 1969). The characteristics of the model and
preliminary test results were first reported by Orlob and Selna (1967, 1970).
Subsequently, it was thoroughly documented for the Environmental Protection
Agency (Gaume and Duke, 1975). It is currently being used in various forms by
many United States governmental agencies.
In a parallel research and development effort, spanning the same period in
the late 1960s and also in collaboration with the TVA Engineering Laboratory,
D. R. F. Harleman and his co-workers at the Massachusetts Institute of Tech-
nology developed a comparable one-dimensional temperature simulation model
(Huber et aI., 1972). The development effort at MIT focused more strongly at
first on fundamental heat transfer mechanisms, utilizing laboratory models as
prototypes for mathematical development (Dake and Harleman, 1966).
Subsequently, however, the MIT model was extended to simulation of actual
reservoirs, e.g. Fontana Reservoir.
The MIT model, which has also been well documented and tested, is presently
used by the TVA and other United States governmental agencies. Apart from
some refinements in treating inflow and withdrawal processes, which will be
discussed later, the MIT model is substantially equivalent in performance to
the WRE model.
A one-dimensional temperature model designed for deep, stratified lakes
was developed at the Cornell Aeronautical Laboratory (Sundaram et aI., 1969)
and applied to Cayuga Lake in upper New York State in a study of power plant
233
cooling water discharges. The model is based on the one-dimensional diffusion
equation and is limited by assumptions of a constant cross-sectional area
(horizontal plane), absorption of all incoming heat energy in the surface layer,
and neglect of heat advected either laterally or vertically in the water column,
except that associated with power plant withdrawals or discharges. Wind
effects are included. Apparently, because of its case-specific nature, the model
has been applied only to Cayuga Lake.
7.3.4. Extensions to Segmented Impoundments

Other significant developments in mathematical modeling of temperature in
deep, stratified impoundments include several attempts to extend the one-
dimensional concept to segmented, weakly stratified reservoirs. Water
Resources Engineers, Inc. (1968) modeled Lake Roosevelt as a six-segment
system and Baca et al. (1974), using a modification of the WRE model, simulated
American Falls Reservoir with a three-segment system. In each instance, model
results compared favorably with observations of the impoundment, but diffi-
culties were experienced in interfacing of segments according to the densimetric
criteria used for introducing advective flows into the water column.
7.3.5. Dual-Purpose One-Dimensional Models

Several one-dimensional models, capable of simulating temperature in addition
to other quality constituents, are worthy of note at this point for reason of
completeness, although they will be discussed more fully in section 7.9, which
deals with ecological and water quality models.
The model LAKECO, also developed by WRE, includes the temperature
simulation routines of the earlier Deep Reservoir Model in addition to the
capability for simulation of some 20-25 abiotic and biotic state variables. This
model, in turn, is embedded in the package WQRRS (Water Quality River-
Reservoir Simulation) developed for the Hydrologic Engineering Center of the
United States Army Corps of Engineers by WRE and others (HEC, 1974). The
solution technique used in the original temperature model (WRE, Inc., 1968,
1969) has become an integral part of each of these packages.
A one-dimensional water-quality-ecological model developed by Baca and
Arnett (1976) utilizes a finite-element technique for solution of the governing
equations, including those for thermal energy balance. The solution technique
is apparently superior in some respects to those of the WRE and MIT models,
coping more reliably in the one case with steep gradients and in the other with
avoiding instabilities. The technique will be presented in some detail later in the
chapter.
A unique model based on principles of energy conservation, including
kinetic energy induced by wind shear, has been developed by Imberger et at.
234
(1978). The model, which is considered suitable for small to medium-sized
reservoirs (less than 10 km long), simulates temperature and salinity variations
in one-dimensional systems that are stratified in density. The model will be
described more fully in section 7.13.2. A somewhat similar approach has been
applied by Stefan and Ford (1975) to simulate temperature dynamics of small
lakes in the north-central United States.
7.3.6. One-Dimensional Destratification Simulation

Henderson-Sellers (1978) described use of a one-dimensional model to simulate
the result ofjet-ind uced destratification. Details of the model were not provided;
however, it appears that inflows and outflows (associated with destratification)
are treated in much the same manner as with the WRE and MIT models, by
observing the requirements for neutral buoyancy of reinjected flows and densi-
metric stability in the water column. Similar applications of the WRE tempera-
ture and LAKECO models have been made in studies of the effects ofrecircula-
tion in pumped storage schemes (Chen and Orlob, 1972).
7.4. CONCEPTUAL REPRESENTATION OF A ONE-DIMENSIONAL

LAKE OR RESERVOIR
In most of the models described above (an exception being that of Imberger
et al., 1978), the one-dimensional impoundment is conceptualized from an
Eulerian viewpointasacontinuum of horizontal slices, usually of equal thickness,
as illustrated in Figure 7.1. The slices, or volume elements, are of fixed volume
and constant thickness, except for that at the surface, which is allowed to vary
in some models in accordance with changes in impoundment volume. (In the
WRE model, for example, slices are added or subtracted as the water surface
fluctuates.) Each volume element is capable of receiving laterally advected
flows, discharging from the impoundment, and transferring advected flows
along the vertical axis while preserving mass continuity. Heat or mass passes
through the bounding horizontal planes by advection and diffusion, and heat
energy by direct insolation depending on the location of the slice. Of course,
heat may be transferred into or out of a slice laterally by advection. Inflows
occur in accordance with densimetric criteria, that is at elevations where there
is a correspondence between inflow density and the density of the water within
the slice. Outflows occur at specified withdrawal points, including the surface
in the case of natural lakes and losses due to evaporation. The general notation
applied to successive slices is illustrated in Figure 7.2.
The notation adopted in Figure 7.1 corresponds to that of the WRE model,
but is virtually the same as for the MIT model. It will be used in development of
the general formulation for the one-dimensional temperature model as follows.
235
Element j
FIGURE 7.1 Conceptual representation of a stratified reservoir (after WRE, Inc., 1968).
Qin' inflow to reservoir; Qx' local drainage; Qj, advected flow to elementj; Qo, withdrawal
from element j; Qj' Q;+ l' vertically advected flows to element j; Q04' withdrawal from
reservoir at levelj = 4; E, evaporation rate; P, precipitation rate.
236
Zj+2!'J.z-c
Zj t z
+ !'J.z ----'.---'\,..,..,7"'''--------:--+1.....
Zj+!'J.Z - I
L
Zj L -.,..,s,..-.::..-----f1
Element j - 1
FIGURE 7.2 Mass continuity between adjacent reservoir elements (after WRE, Inc.,
1968).
7.5. FORMULATION FOR TEMPERATURE PREDICTION
7.5.1. Conservation of Mass

Mass conservation for a volume element ~ within the impoundment is expressed
by
(7.3)
where
~ = i (a z + a z + dZ )~z is the volume of the jth element [L 3 ],
a z , a z + dZare the areas of horizontal planes bounding ~ [L 2],
~z is the thickness of the element [L],
z is the depth, measured from the bottom [L],
j is the element index;j = I at the bottom,
237
Qz is the vertical flow rate [L 3 T - 1],
Qi is the flow advected into the jth element in the horizontal plane
[L 3 T- 1],
Qo is the flow advected out of the control volume in the horizontal plane
[L 3 T- 1].
a~/at = 0, except for the surface element.
7.5.2. Conservation of Heat Energy

Conservation of heat energy in volume element Vj is expressed by
aH·
a/ = (hi - ho + hsz)j - (h wj - hw • j + I) - (h dj - hd • j + 1)' (7.4)
where
Hj = cp l-j8 j is the heat content ofthejth element (1),
c is the specific heat [1 M- 1 DC-I],
P is the density [M L - 3],
8 j is the temperature (0C),
h j = cp Qi 8 j is the heat advected by inflow [1 T- 1 ],
ho = cp Qo 8 0 is the heat withdrawn by outflow [1 T- 1],
hsz = gHZ qszQ z dz is the direct insolation [1 T- 1 ],
qsz = (1 - f3)qsne - ~z is the solar radiation flux at depth z [1 L - 2 T- 1],
qsn is the net solar radiation flux penetrating the surface [1 L - 2 T- I],
f3 is the ratio of radiation absorbed at the surface to the net incoming
radiation,
Y/ is the bulk extinction coefficient [L - 1],
hwj , hw • j + 1 = cpQ(z)8 z represent the heat advected along the vertical axis
[1 T- I ],
hdj , hd . j + 1 = cp EzQz a8/az represent the heat diffused along the vertical axis
[1T- 1 ],
Ez is the coefficient of vertical diffusion [L 2 T- 1 ].
7.5.3. Heat Budget for an Element

The heat energy budget for element Vj at temperature 8 j is given by
local ad vection solar radiation
(7.5)
vertical advection vertical diffusion

238
Solution of (7.5) for 0 j ,j = 1,2,3, ... , n, requires knowledge of tj, a, Qj, Qo'
qS2' and E. ~{z) and a(z) are physical characteristics that can be obtained from
area-volume-elevation curves if L\z is fixed. The other terms may be evaluated
as follows.
7.5.4. Evaluation of Heat Budget Terms
Local Advection
In the earliest versions of the one-dimensional temperature model the local
inflow at temperature 0 j was simply introduced into an element tj, where the
condition 0 j > 0 j > 0 j _ 1 was satisfied. Outflow was considered to occur from
an element spanning the vertical dimension of the outlet(s). In a later version
for the US Environmental Protection Agency (EPA) (WRE, Inc., 1969) outflows
were distributed uniformly over a layer estimated by Debler's (1959) criterion,
which is related to the densimetric Froude number in the zone of the outlet.
For withdrawal well below the thermocline sensitivity analysis indicated that
a uniform distribution produced virtually equivalent results to those obtained
when the outflow was given a pattern of Gaussian form. Since velocity patterns
in the region of the outlet were actually poorly defined the added refinement of
a nonuniform distribution was not considered justified.
Huber et al. (1972) have employed the theoretical relationship of Kao (1965)
to estimate the thickness of the withdrawal layer:
_
f> - 4.8
(q2) 1/4, (7.6)
g(
where
f>is the thickness of a withdrawal layer of uniform velocity approaching a
line sink [LJ,
q is the outflow per unit width of reservoir [L 2 T - 1 J,
( = (lip) dpldz [L -I].
They distribute the withdrawal in a Gaussian pattern over a layer assumed to
be twice as thick as estimated by (7.6).
In the MIT model, inflows are also distributed in a Gaussian pattern defined
by a standard deviation in the inflow velocity field.
Solar Radiation Flux

The net flux of solar radiation to the water column at the air-water interface,
qsn' is either measured directly or estimated from meteorological measurements.
Procedures for estimating qsn that have been incorporated into the WRE and
MIT reservoir models, as well as the stream models, QUAL I and QUAL II
239
(Chapter 6), were developed by the Engineering Laboratory of the Tennessee
Valley Authority (1972). Details of the estimation equations for the principal
heat flux components are given in the appendix to Chapter 5.
The incremental supply of solar radiation to an element Vj in the water
column is obtained by integrating qsza z between z and z + Llz. The heat energy
supplied to the uppermost element U = n) is the total heat flux across the air-
water interface less that passing through the bottom of the element (j = n - 1),
I.e.
(7.7)
where
is
Vn the volume of the surface element [L 3 ],
en is the temperature of the surface element (DC),
an is the area of the reservoir surface [L 2 ],
an _ 1 is the area at the bottom of the surface element [L 2 ],
zn is the thickness of the surface element [L],
c is the specific heat [cal force - I],
P is the density [force· T 2 L - 4 or M L - 3],
qsn is the net solar radiation heat flux [J L - 2 T- 1],
qat is the net atmospheric radiation heat flux [J L - 2 T- I],
qws is the water surface radiation heat flux [J L - 2 T- 1],
q. is the evaporation heat flux [J L -2 T- 1],
qh is the sensible heat flux [J L - 2 T- I],
f3 is the ratio of absorbed to net incoming radiation,
'1 is the bulk light extinction coefficient.
For each element below the uppermost element in the water column the
incremental solar radiation flux is given by
a(V.J e·)
at J = 1
-(1
cp
- f3)q
sn
I
j
j_ 1
e-~Zja.dz
J'
(7.8)
in which 1 < j < n - 1. The extinction coefficient '1 may itself vary with depth,
particularly within the epilimnion of eutrophic impoundments. For most
practical purposes there is no net solar radiation flux to elements below the
thermocline.
The appendix to Chapter 5 provides details of the estimation of the heat flux
terms in (7.7) and (7.8).
240
7.5.5. Effective Diffusion
In addition to vertical advection due to flow imbalances at various levels in the
impoundment, heat is transferred by diffusion, by the random motions of
ambient turbulence, and by secondary currents not otherwise represented in
the one-dimensional approximation. Collectively, these diffusion and dispersion
mechanisms have been described as "effective diffusion" (HEC, 1974) by con-
sidering their combined net effects in heat transfer as analogous to the truly
random process of molecular diffusion. The magnitude of the effective diffusion
coefficient E z varies widely in both time and space, depending on the physical cir-
cumstances of the water body being represented by the one-dimensional model. In
laboratory experiments in the total absence of fluid turbulence, E z may approach
minimal levels in the range of 1 . 10- 6 to 1 . 10- 8 cm z s - 1, while in modeling
of large lakes and reservoirs E z may be in the range of 1 . 10- 4 to 1 . 10- z cm z s - 1
(WRE, Inc., 1968). In these cases the coefficient is regarded as essentially em-
pirical and must be derived from observations of the impoundment with due
regard for the unknown mechanisms, including even characteristics of the
model itself, e.g. "numerical mixing."
E z is evaluated from field data by integrating the heat budget equation
between the limits of the reservoir bottom (z = 0) and a specific elevation z,
accounting for all explicitly defined heat transport or flux terms, i.e.
Z
-z =
E 1 f"(f 0 z Qz dz - Q.0. - Qz 0 z
Q z o01oz /, 0
- cp~ fZ
0
qsz Q z . 1 dZ) dt, (7.9)
where
Ez is the average effective diffusion coefficient at level z over the time
intervalt z - t 1 [UT- 1],
is the horizontal area at level z [L Z],
is the average temperature gradient at level z over the interval t z - t 1
[OC L -1],
oz is the temperature at level z eC),
Q. is the lateral advected flow, Q.(z, t) [L 3 T- 1],
o• is the temperature of the lateral advected flow, 0.(z, t)(°C),
Qz.l is the vertical advected flow at level Z l' Qz. 1(z, t)[L 3 T- 1],
qsz is the shortwave insolation, qsz(z)[J L - z T- 1].
Analyses of temperature profiles in actual lakes and reservoirs have demon-

strated that E z varies widely with depth, with minima occurring in the region
of the thermocline and near the bottom (Orlob and Selna, 1967, 1970; WRE,
Inc., 1968). Highest values occur, of course, in the epilimnion and are clearly
241
related to mixing induced by wind shear at the surface. At intermediate depths
in the hypolimnion Ez may be several orders of magnitude greater than at the
thermocline.
Various empirical representations of E z have been used in one-dimensional
temperature models, ranging from those derived by the method of (7.9) to
merely applying a constant, independent of depth. Water Resources Engineers,
Inc. (l969) derived a functional relationship between Ez and densimetric
stability (lIp) dpldz that indicated a maximum of about 2 cm 2 s- 1 for densi-
metric stabilities above about I· 10- 6 m - l. The Hydrologic Engineering
Center, in its WQRRS model (HEC, 1974), recommends using a constant value
of Ez for density gradients less than the critical, and lower values depending on
the gradient within the zone of the thermocline. Experience has shown that in
most instances minimum values two or three orders of magnitude greater than
the molecular diffusion coefficient are appropriate, although several investi-
gators have noted that simulation of heat transport deep in the reservoir is
especially sensitive to the density gradient and to the numerical methods of
calculation (Orlob and Selna, 1967; Harleman and Hurley, 1976). Explicit
numerical techniques, in particular, often introduce numerical mixing effects
that may be greater than those of the physical processes being approximated
in the model (Bella and Do bbins, 1968).
Investigations of the development and erosion of the thermocline (Sundaram
and Rehm, 1973; Spaulding and Svensson, 1976; Svensson, 1978) have led to
improved descriptions of the mixing induced by wind shear at the water surface.
From the classical mixing length concepts of Prandtl, Spaulding and Svensson
obtained a description of E z that takes the form:
(7.10)
where
C is an empirical constant,
K is the turbulent kinetic energy,
is the rate of turbulent kinetic energy dissipation per unit mass.
Values of K and ( are calculated from two transport equations derived from
the Navier-Stokes equations. Comparison of results from laboratory flume
experiments (under conditions of steady wind) with the model predictions
indicates development and erosion of the vertical structure comparable with
observations of lakes and reservoirs (Svensson, 1978). However, there may be
some practical limitations in the use of this approach in long-term simulation
at lake temperatures because of computational requirements (Bloss and
Harleman, 1979).
242
The combined effect of mixing processes in small to medium-sized lakes on
heat energy distribution has been successfully simulated using a total energy
integration approach (Stefan and Ford, 1975; Imberger et al., 1978). This
modeling approach accounts for both potential and kinetic energy exchanges
through a continuum of one-dimensional segments. Depending on their
relative importance the energy budget can include terms for heat exchange
through the air-water interface, convective mixing due to diurnal cooling,
wind-induced turbulence, internal waves and seiches, and viscous damping.
The Stefan-Ford model, known also as the Minnesota Lake Temperature
Model (MLTM), has been applied with excellent results to a number of small
dimictic lakes in the northern United States. DYRESM, the Imberger et al.
model, has been applied to several medium-sized reservoirs, notably Wellington
Reservoir in Australia. It is described more fully in section 7.13 and some
representative results are illustrated in Figure 7.10.
An intermediate approach to representation of mixing processes in stratified
reservoirs has been taken by Hurley-Octavio et al. (1977) and Bloss and
Harleman (1979). These investigators have accounted for wind-induced
entrainment across the thermocline by equating turbulent kinetic energy input
from wind to potential energy increases due to mixing against the density
gradient. Transient and dissipative effects of the entrainment process are in-
cluded in an updated algorithm for the MIT Lake and Reservoir Model.
7.5.6. Dimictic Lakes: Simulation of the Freeze-Thaw Cycle

Lakes in the north or south temperate and polar regions experience two cycles
of stratification-destratification annually. Inasmuch as the winter cycle,
associated with the formation of an ice cover over the impoundment during
extended periods of subzero temperature, is of special consequence from both
water quality and ecological viewpoints it is desirable to include in temperature
simulation models a capability to represent the freeze-thaw cycle.
The original WRE one-dimensional temperature model was modified to
include the freeze-thaw cycle for applications to two impoundments on the
border between the United States and Canada, Lake Koocanusa (Chen and
Orlob, 1973) and Lake Ross (Norton and King, 1975). An additional term,
accounting for the heat exchange accompanying the change of state of water
from liquid to solid, was added to the heat budget equation. When the tempera-
ture in the water column is dropped to 4°C, a continuing net loss of heat at the
surface reduces surface temperatures toward O°e. When freezing temperatures
are reached, further loss of heat, proportional to the latent heat of fusion,
results in ice formation. Other heat transfer processes are modified in accordance
with the restriction imposed by the ice sheet. In the spring, net warming results
in melting of the ice until open water develops, whereupon the water column
returns to isothermal conditions at the maximum density point, 4°e.
243
7.6. TECHNIQUES FOR SOLUTION OF ONE-DIMENSIONAL
TEMPERATURE MODELS
7.6.1. Implicit Finite-Difference Method

The following solution technique, devised by Water Resources Engineers, Inc.
(1968, 1969), is employed in the WRE and WQRRS versions of the one-
dimensional temperature model. Equation 7.5 is formed into a finite-difference
set for the discretized impoundment according to the notation of Figures 7.1
and 7.2. The rate of change of temperature 0 j in the jth element is given by
- (Q+
. = [1
0· -Ea) ] 0· 1- [1
- (Ea)
- + -1 (Ea)
-
} J.j ~z j }- J.j ~z j J.j ~z j +I
+ ~ (Qj+l + Qo + ~)] 0 j j + [~(~:}+J 0 j +1

1
+ -(Q 0 j j )j + -1 (qsza)
- . (7.11 )
J.j Vj cp j
Written in matrix form, where the bracketed { } terms (vectors) are coefficients
of 0 z , (7.11) becomes
[I] {0} = [5] {0} + {P}, (7.12)
where [I] is an identity matrix, [5] is a tridiagonal matrix of coefficients, and

{P} is a vector of the known parameters and external heat sources and sinks.
The finite-difference equations are integrated by a linear acceleration
method, wherein the incremental change in temperature is approximated by
0 tHt = 0/ + t~t (8 t + 8 tHt ) (7.13)
or
(7.14)
in which 0 and its derivative 8 are known at time t. If the time subscripts are
removed to simplify notation, the temperature of the jth element becomes
I .
0 j = 2M 0 j + bj , (7.15)
which in matrix form for the full equation set is
{0} = t~t {0} + {b}, (7.16)
where {b} is a matrix of 0 and its time derivative for time t.
Equation 7.16 is then substituted in eqn. 7.12, giving
[I] {0} = [5] {b} + t~t [5] {0} + {P}, (7.17)
244
which may be simplified to
[5*J {0} = {P*}, (7.18)
where
[5*J = [IJ -1l1t [5J
and
{P*} = [5J {b} + {P}.
The computational procedure is as follows.
(1) Form {b} from initial conditions or the most recently computed values
of {0} and {0}.
(2) From known values of coefficients and boundary conditions, determine
for the end of the time step the values of [5J, {P}, [5*], and {P*}.
(3) Solve for {0} from (7.18).
(4) Solve for {0} from (7.16).
(5) Repeat for all succeeding time steps.
The steady state solution, i.e. e= 0, is readily obtained by solving

[5J {0} = - {P}, (7.19)
a special case of(7.12).
The solution technique outlined above has been demonstrated as efficient
in many practical applications. Its particular advantages are stability of solution
and flexibility in adapting to temporal variations in boundary conditions.
7.6.2. Explicit Finite-Difference Method

An explicit solution technique has been used in the MIT model (Ryan and
Harleman, 1971), wherein the equation set is solved in spatial order (j = I, 2,
3, ... , n) for successive times (t = 1,2,3, ... , k). The method requires adherence
to the more restrictive of two stability criteria,
1 l1z 2
l1t < - - (7.20)
2 £(z)
or
l11 < l1z a(z) (7.21)

Q(z)'
where £(z) is the diffusion coefficient and Q(z)/a(z) the vertical advection
velocity. Selection of an appropriate l1t depends, therefore, on the desired
245
spatial detail as well as on hydrological and operational conditions. Elements
at reservoir levels near large-capacity bottom outlets often represent limiting
conditions for choosing ilt and ilz.
The explicit method demands less computer storage capacity than the
implicit technique, but this advantage may be offset partially by increased
computational effort when the ilt required is smaller than that needed to
describe the requisite temporal variations in temperature.
7.6.3. Finite-Element Method

Baca and Arnett (1976) have employed the finite-element method in a general
water quality model that includes temperature simulation. Advantages of the
method, which is also implicit in its treatment of the governing equations, are
avoidance of numerical mixing and instability (inherent difficulties of explicit
methods), flexibility in time-step selection, and adaptability to steep gradients
that cause problems when linear approximations are made. The solution
technique will be outlined in section 7.13.1.
7.7. APPLICATIONS OF ONE-DIMENSIONAL TEMPERATURE

MODELS OF LAKES AND RESERVOIRS
There have been many practical applications of one-dimensional temperature

models, far too numerous to describe in detail here. A few typical examples have
been selected to indicate the general capabilities, as well as limitations, of these
models.
7.7.1. Fontana Reservoir, North Carolina

One of the first attempts to simulate the annual cycle of a thermally stratified
impoundment was an application of an early version of the WRE model to
Fontana Reservoir in the TVA system. Data for the simulation were derived
from a carefully planned and executed field survey by the TVA Engineering
Laboratory.
Figure 7.3, derived from a report to the California Department of Fish and
Game (WRE, Inc., 1968), compares simulated temperatures with observations
from March to December 1966. Broken lines are isotherms predicted by the
model, while full lines represent in situ observations. The model gives a credible
representation of the distribution of temperature throughout most of the annual
cycle, particularly during the period of strongest stratification. The pattern of
thermocline formation is followed reasonably well by the model, although
warming of strata near the surface in the spring was predicted to occur more
246
1700
-
....
c::
.Q
1600
1500
~
>OJ
W
1400
1300
60 120 180 240 300 360
Time (daysl
FIGURE 7.3 Computed (broken lines) and observed (full Jines) thermal energy distri-
bution in Fontana Reservoir using different diffusion coefficients (after WRE, Inc., 1968).
The temperature CF) is shown for each isotherm.
slowly than actually happened, possibly due to attenuation of shortwave

insolation by turbid spring runoff not properly accounted for in the model. The
uncertain ability of the model to represent convective mixing as a consequence
of cooling during fall overturn is evident in the disparity between model and
impoundment during late fall. It is exemplary, as well, of a general deficiency
of this class of models in which the instability brought about by surface cooling
is only crudely represented by numerical mixing driven by temperature anom-
alies in the upper strata. Overall, the simulation is a credible depiction of the
annual thermal cycle for such a reservoir, rather typical of results expected with
one-dimensional temperature models.
Comparable results for Fontana Reservoir employing the MIT model were
reported by Harleman and Hurley (1976), as illustrated in Figure 7.4. In this
example, the investigators examined the sensitivity of vertical heat transfer
during thermocline development to the magnitude of the "diffusion coefficient."
In Figure 7.4(a), corresponding to 27 April 1966, the assumption of molecular
diffusion as the only mixing process results in a considerable disparity between
model and impoundment. This difference is rectified somewhat by increasing
the diffusion coefficient by two orders of magnitude, but it appears that in this
period, at least, an even greater mixing effect might be necessary.
Later, in the early summer with a well established profile the model tends to
produce more heat transfer in the region below the thermocline than was
observed in the water body. Since these simulations were made largely for
investigating the mechanics of internal heat transfer, no attempt was made to
calibrate the model. Results, however, do show that the model is capable of
reasonable representation of the gross processes of heat transfer in a thermally
stratified deep reservoir.
247
500 ~-
(a)
.......-:;
E 480
c
a
.;:; 460
ctl
>
Q) ~tCf
w 440
420
0 4 8 12 16 20 24 28
520
:, ...
500 (b)
......
E 480 .'
c
a 460
';:;
ctl
> Outlet Cf.
~
w 440 ;1
i
420 !
0 4 8 12 16 20 24 28
Temperature (oG)
FIGURE 7.4 Comparison of predicted temperature profiles in Fontana Reservoir using
different diffusion coefficients (after Harleman and Hurley, 1976). (a) 27 April 1966; (b)
22 June 1976. Full line, measured; dotted line, predicted (molecular diffusion) ; broken line,
predicted (diffusion coefficient = 100· molecular value).
7.7.2. Hungry Horse Reservoir, Montana

Results of a simulation of Hungry Horse Reservoir in Montana (Table 7.1) are
presented in Figure 7.5, showing the capability of the one-dimensional tempera-
ture model to represent a strongly stratified reservoir (Fr = 0.(026) with a very
shallow epilimnion. The thermocline developed late in the spring as the reservoir
water level was rising, owing to flow of snowmelt into the bottom of the reservoir
(4°e isotherm near the bottom). Apparently, because of higher turbidity in
the waters of the impoundment, shortwave energy influx during the summer
was confined to surface strata, the body of water below about 20 m remaining
at less than goe throughout the year, even though surface temperatures ex-
ceeded 20 0 e in August. The agreement between model and impoundment is
considered excellent for the most part, except that the model seems to predict
fall cooling and associated convective mixing earlier than they actually oc-
curred, illustrating once again the difficulty in simulating this phenomenon.
248
1090 Water surface

16 18 10
1060 ---y-----_
E
c
a
' co::; 1030 \
\
>
OJ
\
\
jjj \
\
\
\
\ I
\ I
J
1000
970 +--,-----,--------,---------'~-----",-------'---------,------,--------!
June July August September October
FIGURE 7.5 Simulated and observed thermal regimes for Hungry Horse Reservoir,
1965 (after WRE, Inc., 1969). Full lines, computed; broken lines, measured. The tem-
perature (0C) is shown for each line.
7.7.3. Lake Piiijiinne, Finland

The model EPAECO (Gaume and Duke, 1975) was used to simulate the
principal thermal cycle of Lake Piiijiinne in Central Finland (Kinnunen et aI.,
1978) from late May, when temperatures were isothermal at 5,SOC, through the
summer when surface temperatures reached about 20°C. Typical temperature
profiles for June and August in three successive years are illustrated in Figure
7.6. Results are regarded as generally satisfactory, although there is evidence
that surface heating occurs more rapidly in late spring in the model than in
reality.
7.7.4. Ross Lake, Washington

The earliest versions of one-dimensional temperature models neglected dimictic
lakes, which go through two distinctive periods of thermal stratification, one
occurring during the period of ice cover. This capability was required in order
to study Ross Lake on the Skagit River in the Cascade Mountains of Washington
State. As a result of prolonged subzero air temperatures and the formation of
ice cover, temperatures of surface strata during winter months drop to near
249
80 80 ,
1974 x~
14 June x x 19 August
E 60 60
9,'
J1
c
0
I
.~ 40 40
>
~
w 20 20
x
9
¥
0 10 20 0 10 20
80 80
1975
E 60
1;~June 60 *26 August
x
c 9x /
0
.~ 40 40 x
x
I
x
>
~
w 20
I
20 I
I
x
I
I I
0 10 20 0 10 20
80 80
1976 19 August
12 June x
-60 60 x/ 18 August
E /
x
c /
0
.;:; 40 x 40 x
co
>
9x I
x
~ I
OJ
w 20 20 x
? Ix
¢ I
0 10 20 0 20
Temperature (DC)
FIGURE 7.6 Observed (0) and simulated (x) temperatures in Lake Paijanne, Finland
in 1974. 1975. and 1976.
zero. Isothermal conditions are reached twice each year, once with the spring
thaw in March or April and again with the fall overturn in October and
November.
The Deep Reservoir Model as modified by Norton and King (1975) was
used to simulate the entire annual cycle of the lake and the predicted tempera-
tures were compared with those recorded in monthly surveys during 1971, as
illustrated in Figure 7.7. A random sampling of temperatures from model and
lake indicates agreement to within ± O.soC throughout most of the year. The
double cycle of stratification and subsequent mixing is faithfully simulated by
the model.
250
0
20
E
~
40
C.
al
0
60
80 Jan Feb April June July Aug Sept

Mar May Oct Nov Dec
(al
0
I I
I I
I I
I I
20 I
I ,I
\ I
E \ I
\ I
~
40 \ I
C.
al \ I
0 \ I
\ I
'j
60
80
Jan Feb Mar April May June July Aug Sept Oct Nov Dec
(bl
FIGURE 7.7 (a) Observed and (b) simulated temperatures (wC) in a dimictic lake. Lake
Ross. Washington. 1971 (after Norton and King. 1975).
7.8. COMMENTS ON TEMPERATURE SIMULAnON
For the class of lakes or reservoirs that can be considered to undergo strong
stratification (Fr ~ 0.1), one-dimensional models are capable of representing
the principal features of the annual thermal cycle, including even winter strati-
fication under ice. Some difficulties are still experienced with such models in
simulation of convective mixing because of cooling in the fall, although this
deficiency does not appear to be serious, at least insofar as temperature simu-
lation is concerned. When such models are employed to drive simulation of
other state variables, water quality and biota for example, this weakness may
become more of a limitation.
251
Another deficiency appears to lie in the characterization of mixing processes
in the epilimnion after the onset of stratification, but this may be more of
academic interest than of consequence from a practical water quality viewpoint.
As noted in section 7.5.5, the effects of wind-induced mixing in the upper strata
are currently under investigation, which holds promise of a more rigorous
description of these processes. As regards water quality changes, it may be of
greater importance to examine more critically what takes place in the region of
the thermocline and below, inasmuch as these areas are likely to be the more
sensitive ecologically.
7.9. APPROACH TO ONE-DIMENSIONAL MODELING OF WATER

QUALITY AND ECOLOGY IN LAKES
Mathematical models for impoundments have evolved along two different

lines that have gradually merged in recent years to provide a fairly comprehen-
sive capability for simulation of water-quality-ecological relationships in deep,
stratified water bodies. Those that may be treated by the one-dimensional
conceptualization were presented earlier in this chapter. First, there was the
rather logical extension of the one-dimensional stratified reservoir model of the
WRE-MIT type to include vertical transport of quality constituents, both
abiotic and biotic. These developments are well represented by the work of
Chen and Orlob (1968), Markofsky and Harleman (1973), and Chen and Orlob
(1975). Then there were the modeling activities that focused primarily on the
ecosystem and nutrient balance of the lake, like the models of Vollenweider
(1965, 1969) and the International Biological Program team of researchers at
Rensselaer Polytechnic Institute (Park et al., 1974a, b) who initiated the model
series beginning with CLEAN, which was subsequently modified to deal with
multisegment systems, e.g. MS CLEANER (Leung et al., 1978). In both in-
stances, models emerged that were essentially one-dimensional (although
attempts have been made to extend them to systems that have higher dimen-
sionality and deal with both water quality and ecology). Examples of such
extensions of the basic concepts of modeling the water quality and ecology of
lakes will be presented in Chapter 8. For the present, however, we will confine
our discussion to the somewhat simpler one-dimensional case, using the
general principles outlined in Chapters 3 and 4.
Because of the disciplinary predilections of the development teams, one
engineering-oriented and the other concentrating more on biology, the two
lines ofone-dimensional water-quality-ecological model development tended to
place different emphasis on two important and essential considerations in realistic
presentation of the stratified water body. The WRE-MIT approach, founded as
it was on consideration& of thermal energy balance mechanisms and advective-
diffusive transport along the vertical axis of a strongly stratified impoundment,
252
concentrated more on vertical movement and distribution of the water quality
state variables of traditional pollution control concern, e.g. DO, BOD, con-
servative substances, nutrients, coliforms, e!c., and tended to treat the ecosystem
in a more aggregated state. For example, algae, zooplankton, and fish were
divided conceptually into simple groupings, rather than by species, age classes,
life stages, etc. (Figure 4.1). The line of development represented by CLEAN,
CLEANER, and MS CLEANER tended toward a more rigorous representation
of biological phenomena at the sacrifice of some realism from the hydrological
and hydrodynamic viewpoints (Figures 9.10 and 9.11). Both schools of model
development have gradually combined, with each drawing from the other to
provide more correct, yet practical, capability to model the one-dimensional
class of impoundments.
Zero-Dimensional and Two-Layer Models

Virtually all mathematical water-quality-ecological models have evolved from
application of the law of mass conservation, supplemented by the principles of
kinetics applicable to chemical and biological systems. Among the earliest such
models of lake systems were those addressing the problem of nutrient balance
in lakes. They were generally gross nutrient budget models that described the
entire lake as a continuously stirred tank reactor (CSTR), fully mixed over the
period of interest, often the full annual cycle (Vollenweider, 1965, 1969). At
first, no particular attempt was made to discretize spatially and the models
often treated only a single water quality constituent, e.g. nutrients considered
to be important regulators of eutrophication, like carbon, nitrogen, and phos-
phorus. The Vollenweider approach was seized upon by other investigators
(O'Melia, 1974; Snodgrass and O'Melia, 1975; Larsen and Mercier, 1975; Bella,
1976) in dealing more explicitly with nutrient and algal budgets in well mixed
systems. Imboden (1974) and O'Melia (1974) adapted the approach to two-
segment (epilimnion-hypolimnion) systems.
Transition to Multisegment Models

Among the early models for simulation of eutrophication of lakes, the one
proposed by Chen and Orlob (1968) adopted the CSTR analogy to describe a
set of interactive (coupled) mathematical relationships for production of algae
from nutrient input to natural systems. This development took place, however,
with the full expectation that the generalized set of mass balance equations that
might evolve for the multiplicity of water quality and ecological compartments
would be embedded directly in one-dimensional transport models, such as
those already being operated to simulate the thermal structure of deep reser-
253
voirs. In fact, since the basic one-dimensional segmented model had already
been constructed and tested for the temperature simulation problem (WRE,
Inc., 1968, 1969), the transition to an operational water-quality-ecological
model for this case was more easily accomplished. A similar transition was
effected between the MIT temperature model (Huber et aI., 1972) and extensions
to deal with DO-BOD balance in deep reservoirs (Markofsky and Harleman,
1973).
An important step in the evolution of water-quality-ecological models was
the development of CLEAN, a model conceptualized by some 25 investigators
of the Eastern Deciduous Forest Biome of the US International Biological
Program (Park et al., 1974a). At first, the model considered only a single spatial
segment but as applications were made to real systems it was modified to
accommodate at least two-segment systems, such as stratified lakes in which the
epilimnion and hypolimnion could be treated as unique biological entities. In
the version CLEANER, 20 state variables (mostly biological) were represented
(Park et al., 1974b; Scavia and Park, 1976). As Chapter 9 will show, it evolved
further into a version (MSCLEANER) capable ofrepresenting the impoundment
in multiple segments with up to 40 state variables (Leung et aI., 1978).
7.10. CONCEPTUAL REPRESENTATION OF ONE-DIMENSIONAL

WATER-QUALITY-ECOLOGICAL MODELS
For this class of models our conceptual view of the water system is virtually
identical to that of the vertically stratified, deep reservoir, depicted in Figure
7.1, wherein each control volume, a segment of the water column, is treated as
a CSTR, i.e. as a zero-dimensional submodel. Thus, the physical, chemical, and
biological interactions are all assumed to occur within the confines of a discrete
volume element. Exchanges occur across the boundaries of such an element as
a result of exogenous forces, e.g. advection, diffusion, insolation, sedimentation,
etc., and the" states" of variables from time to time are governed by kinetic
principles that must be defined as functions of space and time. Finally, all
discrete elements are linked in space to provide a continuum in model form
that represents the water body.
7.11. FORMULATION OF WATER-QUALITY-ECOLOGICAL

EQUATIONS
The general form and notation for the water-quality-ecological equations

resemble those set forth earlier in this chapter for one-dimensional temperature
models. The general one-dimensional mass conservation equation for a volume
254
element ~ is given by
o(VC.)
at J = (Q;C i - QoCo)j - (QzC)j + (QzC)j+l + (Ea OC)
OZ j
_ (Ea OC)
OZ j+ 1
local advection vertical advection
vertical diffusion
oV dC
+ c._
at
J
J
+ V-
J dt
± s, (7.22)
volume change process change sources and sinks
where C j is the concentration of any constituent, abiotic or biotic, that moves

with the fluid mass or may be transferred by a diffusive process in proportion
to the gradient oC/oz. The total derivative dC/dt represents all processes, other
than advection, diffusion, and volume change, that act to modify Cj • Equations
of this form can be written for each quality parameter of interest, either inde-
pendent or coupled, conservative or nonconservative, through an appropriate
formulation of dC/dt. Customarily, the process reactions exemplified by dC/dt
comprise first- or zero-order terms, the coefficients of which may be determined
by either preset exogenous conditions or the values of other state variables.
7.11.1. Formulation for Oxygen Balance

In modeling the water quality of a stratified impoundment, perhaps the best
illustration of the explicit formulation of an equation of type (7.22) is the mass
balance equation for dissolved oxygen. For simplicity, the subscript j has been
dropped:
O(VO)
-A +D _ Oov
at at
advection diffusion volume change
+ as K 2 (0* - 0) -K 1 LV- K 4 a b (OS)*

reaeration BOD benthic demand
ammonia oxidation nitrite oxidation detritus oxidation
(7.23)
biotic respiration. photosynthesis external sources
e.g. of algae and sinks
where
o is the dissolved oxygen concentration [M L - 3],
0* is the dissolved oxygen concentration at saturation [M L -3],
A is the advection rate [M T- 1],
D is the diffusion rate [M T- 1 ],
255
as is the surface area of the uppermost element [L 2 ],
K2 is the reaeration coefficient [L T - I],
K1 is the biochemical oxygen demand coefficient [T- I],
L is the carbonaceous biochemical oxygen demand [M L - 3],
K4 is the benthic sediment oxygen demand coefficient [T-I J,
ab = aj - aj_ 1 is the bottom area associated with element j [V],
(OS)* is the organic sediment accumulation per unit of bottom area
[M L -2],
is the ammonia nitrogen concentration [M L - 3],
is the nitrite nitrogen concentration [M L - 3],
is the detritus concentration [M L- 3],
are the stoichiometric equivalences with oxygen for ammonia,
nitrite, and detritus [M M-1J,
are the decay coefficients for ammonia, nitrite, and detritus [T- 1 ],
is the phytoplankton concentration [M L - 3],
is the phytoplankton respiration rate [T- 1 ],
is the phytoplankton growth rate [T- I],
is the biota activity coefficient (dimensionless),
is the stoichiometric oxygenation factor for algal growth [M M -I],
is the source or sink [M T- 1].
In eqn. 7.23 we perceive the complexity of interactions between water

quality and the ecosystem. The first three terms represent the primary processes
of the impoundment that affect oxygen transport, i.e. advection, diffusion, and
change in water surface elevation. The next three terms represent the classical
formulation of the Streeter-Phelps equation (Chapter 6) for oxygen sag.
Nitrogenous demand for oxygen is represented in the oxidation of ammonia and
nitrite. Oxidation of organic detritus and respiration of living forms, e.g. algae,
account for additional decrements in the available resource, while photo-
synthesis adds oxygen to the pool in proportion to the production of new algal
cell material. Sources and sinks of oxygen, exogenous to the element, are
represented for completeness.
7.11.2. Formulation of the Equation Set

A complete formulation for an impoundment requires writing similar equations
of the form of (7.22) for each state variable represented. Table 7.2, which has
been adapted from the LAKECO subroutine of the WQRRS model (HEC,
1974; Chen et al., 1975), illustrates the approach applied to a system involving
15 state variables. Figure 7.8 shows the relationships between the various
constituents of the aquatic system.
TABLE 7.2 Mass Balance Equations for a One-Dimensional Water-Quality-Ecological Model of Stratified Impoundments.
Constituent Concentration [M L - J] Mass Balance Equation
Conservative substances. C CV = A + D + CV
Biochemical oxygen demand. L LV = A + D - LV - K1LV
Detritus, DET (DET)V = A + D - (DET)V - fJJ(DET)V - absD(DET)V
decay sedimentatl0n
IV
Vl + Kb/lZZV(~ - -
I) + Fz(~ I) + FB(~ 1) -
0\ ~u ~~ ~"
zooplankton excreta fish excrela
Organic sediment. OS (OS)V = -K,ab(OS)·V + abspPV + absD(DET)V

decay phytoplankton detritus
Carbon. C CV = A + D - CV + a,Kc(CO! - CO,W + P,cD(DET)V + K1cLLV + Kb(cprpP+ C7r7Z + cFrFF + cBrBB)V

reaeration detritus release BOD decay respiration of biota
Ammonia nitrogen, NH, (NHJ)V = A + D - (NHJ}V - til(NH,)V + 'PDnD(DET)V + 'Psns(OS)V

decay detritus release organic sediment release
Nitrite nitrogen. NO , (NO,)V = A +D - (NO,)~' - P,(NO,)V

decay
Nitrate nitrogen. NO J (NOJ)V = A + D - (NO,) V + P,(NO, ) V - Khnp/lpPV

nitrite oxidation phytoplankton
uptake
Phosphate phosphorus, P0 4 (P0 4 )V = A +D -- (P0 4 )V + 'PDPD(DET)V + 'PsPs(OS)V - KbPP/lpPV

detritus release organic sediment phytoplankton
release uptake
Coliform bacteria. CB (CB)V = A + D - (CB)V - K,(CB)V

decay
Algae. P PV = A + D - PV + Kb(/lp - rp)PV - abspPV - Kb/lzZV/~u

net growth sedimentation zooplankton
grazing
Zooplankton, Z ZV = A + D - ZV + Kb(/lz - r z - "'z)ZV - FZ/~F
net growl h grazi ng by
tish
Benthic animals, B B V = K b(/lB - r B - mB)B*u b - F B/~B

net groWl h grazi ng by
fish
Nekton, F FV = Kb(/lF - r F - "'F)FV - hFV

net growth harvest
Dissolved oxygen. 0 OV = A + D - OV + a,K 2 (0* - 0) K,LV - Kpb(OS)V -{llO(,(NH 3)V - fJ20(2(N0 2) V - /i 30(3(DET)V
tv reaeratlOn BOD benthic ammonia nitrIte delritus demand
Vl demand oxidation oxidation
-.J
- Kb(rpP + rzZ + rFF)V + Kb(/lpP)yV

biotic respiration photosynthesis
Terms not previously defined:

B* is the saturation concentration of benthic animals per unit of bottom area [M L - 2].
/lp, /lz, /lB' /IF are the growth rates of phytoplankton. zooplankton, benthic animals, and nekton [T-'].
r p , rz , r B , r F are the respiration rates of phytoplankton, zooplankton. benthic animals. and nekton [T- 1 ],
"'/' "'0' "'F are the mortality rates of zooplankton. benthic animals, and nekton [T-'].
Sp, So are the sedimentation rates of phytoplankton and detritus [L -2 T- 1],
'I' is the release rate of ammonia nitrogen [T - I],
~,," is the efficiency of predator a in digesting prey b,
c is the proportion of carbon,
n is the proportion of nitrogen,
r is the proportion of phosphate phosphorus,
h is the harvest rate constant [T - t],
K, is the rate of organic sediment decay per unit of bottom area [T - , L - 2].
ab is the area of bottom sediment [L 2].
Kc is the mass transfer rate of carbon as CO 2 per unit of surface area [T- I L - 2],
a, is the area of the water surface [L 2].
258
r--------'-''-i
A A
Atmosphere 1---'--'----------,
Carbonaceous
B BOD
E,M
R R
M E
R R
M
B B
B B
FIGURE 7.8 Water quality and ecological relationships in LAKECO (after HEC,
1978). A, aeration; B, bacterial decay; C, chemical equilibrium; E, excreta; G, growth; H,
harvest; M, mortality; P, photosynthesis; R, respiration; S, settling.
7.12. TECHNIQUES FOR SOLUTION OF ONE-DIMENSIONAL

WATER-QUALITY-ECOLOGICAL MODELS
7.12.1. Implicit Finite-Difference Method

The set of equations resulting from formulation of j = 1, 2, ... , n equations
for each of i = 1,2, ... , k state variables may be solved by an implicit technique
like that outlined in section 7.6 for solution of the heat energy equations. This
procedure has been used in LAKECO (Chen and Orlob, 1975) and WQRRS
259
(HEC, 1974) and in other versions of the original model, like EPAECO (Gaume
and Duke, 1975).
7.12.2. Explicit Finite-Difference Method

Extensions of the MIT model to include water quality parameters like DO and
BOD (Markofsky and Harleman, 1973) have followed the original solution
procedure, a step-forward explicit method that requires observance of certain
stability criteria (eqns. 7.20 and 7.21).
7.13. OTHER ONE-DIMENSIONAL WATER-QUALITY-

ECOLOGICAL MODELS
Three additional one-dimensional water quality models are worthy of special

mention in this chapter. They are the Finite-Element Model of Baca and Arnett
(1976), a "dynamic" reservoir model developed by Imberger et al. (1978), and
an adaptation of the Deep Reservoir Model by the US Army Waterways
Experiment Station (Fontane and Bohan, 1974).
7.13.1. Finite-Element Model-Baca and Arnett

Baca and Arnett (1976) developed a water quality model for eutrophic lakes
and reservoirs that, in addition to employing the finite-element method for
forming and solving the basic differential equations, provides an improved
representation of the nitrogen and phosphorus cycles in lakes.
Formulation of the basic mass balance equations in their model follows the
general pattern outlined in Table 7.2, with a few important distinctions. These
relate primarily to the nitrogen and phosphorus cycles and the recycling of
these nutrients between the water column and the sediment.
Formulation of the Baca and Arnett model involves a set of j = 1,2, ... , n
equations, corresponding to n reservoir slices, for each of i = 1, 2, ... , k state
variables, just as in LAKECO. These are solved by transforming the equations
and applying the finite-element method of implicit solution. The method is
summarized below.
The advection-diffusion equation can be written as a complex function of C:
L(C) = - ac + w-
ac - -a (Ez -ac) + AC - S, (7.24)
at az az az
where
C is the nutrient concentration [M L - 3 J,
w is the vertical advection velocity [L T- 1 J,
Ez is the diffusion coefficient [L 2 T- 1J,
A is the decay (or growth) coefficient [T- 1J,
260
and in vector notation as
{a}b = ~
8z
[z-1j 1j]{ C
+1 - Z
Cj + 1
j }. (7.25)
By employing the Galerkin method of weighted residuals, we can recast the

governing model eqn. 7.24 in integral form:
x= LL(C)a-;dz, (7.26)
where a-; is an arbitrary set of weighting functions. If the approximating function

~ for C is taken to be the same as ~ according to Galerkin's method, then
k
c= L j= 1
Cj~ (7.27)
and
-
C(z, t) = [WI' W2 ] {Cj~
c. } I ' (7.28)
where the weighting functions are given by
(7.29)
and
1
W2 = 8Z(z - Zj). (7.30)
Expanding (7.26) and integrating by parts results in
X=
(
R
ac
f. at - a-; + w -ac
az
a-; - E 2 -
ac aa-; + ACa-; -
az az
--
)
Sa-; dz, (7.31)
which can be partitioned so that

n
X= L xj, (7.32)
j= I
where n is the number of subdomains. This may be expanded to give

22 2
X= dz + I23 dz + .. , + I '" dz. (7.33)
I 2, [] 22 [] 2", _, [ ]
Individual terms in the Galerkin functional eqn. 7.26 and the expanded form,
eqn. 7.33, are approximated by
ac { C. }
at = [WI' W2 ] Cj~I ' (7.34)
261
oC [ ~z'I ~zI]{CC
7h = -
j
+I
}
'
(7.35)
j
and
oW; _ [_ ~ ~] (7.36)
oz - ~z' ~z .
Substituting these quantities into the Galerkin functional for a particular

subdomain results in a set of integrals that can be evaluated by
rZj
+1 Y fJ _ y!f3!~z
I + f3 + r
WI W 2 dz - (7.37)
~Zj y
From these result specific expressions for the time derivative, advection,
diffusion, decay, and the sources. Evaluation, summing, and gathering terms
leads to a general matrix equation:
[I]{C} = [S]{C} + {R}, (7.38)
in which [I] is an identity matrix, [S] is a tridiagonal matrix of the coefficients
of C, and {R} is a vector of all known parameters and external sources of C.
Solution of the equation for each quality constituent is accomplished sequen-
tially according to the method outlined for (7.11), i.e. the procedure adopted
for the WRE Deep Reservoir Model and for LAKECO (WRE, Inc., 1968, 1969;
HEC, 1974).
The advantages of the finite-element technique cited by Baca and Arnett lie
in avoidance of numerical mixing effects, stability of solution, flexibility in
length of time step, and adaptability to steep gradients. Experience in use of
the method in even more complex problems, involving hydrodynamic processes,
seems to confirm these attributes (Gallagher et al., 1973; King et al., 1974).
7.13.2. Dynamic Reservoir Water Quality Model-Imberger et al.

Additional rigor in modeling the hydromechanics of small to medium-sized
impoundments is provided in a new approach by Imberger et al. (1978; Spigel,
1978; Fischer et al., 1979). A Lagrangian one-dimensional model is constructed
by forming a set of turbulent kinetic energy budget equations. In Imberger's
notation, the rate of change of mean potential energy in a control" slab" is
given by
(7.39)
where
V is the potential energy,
AK is the kinematic flux of turbulent kinetic energy imposed at the top surface,
262
As is the rate of production of turbulent kinetic energy by the shear across
the bottom,
AT is the rate of increase of turbulent kinetic energy in the slab (or column),
AL is the rate of leakage of energy by radiation of internal waves into the
quiescent fluid,
AD is the total rate of energy dissipation by viscosity.
Drawing on research in oceanography, primarily that of Niiler (1975),

Imberger transforms (7.39) into a simple entrainment law:
1 dh CK
(7.40)
q dt Ri'
where
q is the characteristic turbulent velocity scale,
h is the depth of the mixed layer,
CK is a constant that determines the entrainment rate,
Ri = IXdegh/q2 is the Richardson number, in which IX is the expansion
coefficient for water and de is the temperature change.
The entrainment rate is estimated from observations of the impoundment in

terms of two characteristic damping depths, ho and hI' that determine the
effectiveness of mixing by wind energy and convective mixing by mechanical
buoyancy, respectively.
Effective turbulent mixing, which accounts for vertical transport of turbulent
kinetic energy, is determined empirically by using the depth of the reservoir,
the stability due to density differences from top to bottom, and a characteristic
mixing time. A constant-flux model is proposed:
KH 2
{(z) = T. S' (7.41)
m
where
£(z)is the effective turbulent mixing coefficient along the vertical (z) axis,
H is the depth of the reservoir,
Tm = V/(P I + P 2) is the mixing time scale, in which PI and P 2 are rates of
work of inflowing streams and wind, respectively,
S = - (dp/dx)H/(p(O) - p(H» is the stability,
K is the reservoir constant.
Application of the model requires estimation, from impoundment measure-

ments, of four so-called "universal" constants: CD, a drag coefficient of the
entering streams; K, the constant of proportionality of the diffusion coefficient
(eqn. 7.41); and ho and hI' In demonstrating the model on Wellington Reservoir
263
in Australia, Imberger et al. (1978) derived best values of the constants by
simulation during representative periods when only one of the four constants
was dominant. For the case examined CD was found to be about 8.5· 10- 3,
K was 4.8.10- 4 , hI = 5 m, ho = 20 m, and ( varied from 0 to 10- 4 m 2 S-I.
The procedure for setting up the Imberger model is as follows. The reservoir
is divided into slabs of arbitrary thickness for which the initial values of tempera-
ture, salinity, density, volume, etc. are known. Surface thermal energy, deter-
mined by a meteorological subroutine, is added according to an exponential
penetration law at quarter-hour intervals up to one day, or when the surface
slab has cooled by 3°e. A mixing routine then adjusts for instabilities to produce
a new density profile, which is subsequently relaxed by the diffusion coefficient
calculated according to (7.41). This is followed by adjustments for energy input
by inflow and energy extraction by outflow. Volume continuity is maintained
within each Lagrangian slab and the bounding planes are tracked as they move
up and down. No convection is allowed between slabs.
In simulation of Wellington Reservoir with the model DYRESM (Dynamic
Reservoir Model) a typical run of 100 days required 5.5 s per day of CPU time
on a Cyber 73 computer. Preliminary results of the calibrated model give an
excellent account of its capability to simulate changes in temperature and
salinity in a medium-sized reservoir under rather complex operational
conditions (lmberger, 1981).
7.13.3. Waterways Experiment Station Model

The Waterways Experiment Station (WES) ofthe US Army Corps of Engineers
developed a one-dimensional model for the simulation of temperature and
oxygen balance in deep reservoirs (Fontane and Bohan, 1974), known as
WESTEX. It was patterned after an early version of the WRE Deep Reservoir
Model, extended by Fogg and Fruh (1973) for use on Lake Lyndon B. Johnson
in Texas, and designed to overcome an inherent difficulty of one-dimensional
models: failure to represent correctly the actual detention time for flows
entering when the reservoir is strongly stratified. Paradoxically, this problem
is especially acute when the reservoir has a large volume relative to the rate of
inflow, just the conditions that favor the one-dimensional temperature model.
Under such conditions, organic loads entering the reservoir may distribute
their demands for oxygen variably over space and time depending on BOD
concentrations, temperature and relative density of inflow, and the time of flow
through the reservoir. Clearly, the problem requires more than a simple one-
dimensional model, wherein the loads are instantaneously distributed through-
out the horizontal" slice" into which the load is introduced. The WES has
devised a method whereby the times of flow through the reservoir can be
estimated with the aid of a simple physical model of the stratified system. These
times are then utilized in the simulation to give a more plausible description
264
of the dissolved oxygen distribution along the reservoir and over the season.
Comparison of simulated DO profiles in operating reservoirs with actual
observations indicate that WESTEX can produce more realistic results than
the unmodified one-dimensional model.
7.14. APPLICATIONS OF ONE-DIMENSIONAL WATER-QUALITY-

ECOLOGICAL MODELS
While there have been many applications of one-dimensional water-quality-

ecological models, since their first appearance in the mid-1970s, to deep lakes
and reservoirs, relatively few cases have yielded sufficient data for comparative
purposes. The best results seem to have been for simulation of dissolved oxygen
balances, although some success has been demonstrated with nutrient budgets
and primary productivity. Three examples have been selected to indicate the
general capabilities of such models, as well as to point out certain inherent
limitations.
7.14.1. Lake Washington, Washington

Figure 7.9 illustrates an early test of the model LAKECO on Lake Washington.
The variation of dissolved oxygen concentration in time and space appears to
be fairly well simulated, although certain details, notably the steeper gradients
revealed in field observations, are missed. The model can characterize the
moderate depletion of dissolved oxygen in the upper part of the hypolimnion
during the late summer and early fall, as well as the progressive decline in DO
levels at the bottom from early spring until the fall overturn. The initial and
peak stages of phytoplankton growth, also simulated in the model, were repre-
sented satisfactorily, although during the fall observed algal biomass (as
indicated by chlorophyll a concentrations) tended to be much higher than
predicted (Chen and Orlob, 1975).
7.14.2. Wellington Reservoir, Australia

Figure 7.10 shows results of a simulation using DYRESM, the model of
Imberger et at. (1978), in which both temperature and salinity (NaCl) distribu-
tions are compared with detailed observations in the field. The simulation
begins with day 133, at which time the reservoir was slightly stratified thermally
with a thermocline about 10 m above the bottom. Salinity at this time was
almost uniform from top to bottom. Subsequently, the reservoir cooled (day
190) and there was a large influx of saline water. This inflow was apparently
cooler and denser (owing to temperature and salinity) with the result that
there was a large alteration in quality of the impounded water below about 13 m
265
20
E
r.
....c.
(ll 40
Cl
60
10 9 8 7 4
Jan July Aug Sept Oct Nov Dec
(al
O--,----'T-----------'--.---------,-----'-i---~-___=;_--___T_-__r--_rn_rl
20
E
r.
C. 40
(ll
Cl
60
10 9 8 7 6 5
Jan Feb Mar April May June July Aug Sept Oct Nov Dec
(bl
FIGURE 7.9 (a) Observed and (b) simulated dissolved oxygen concentrations in Lake
Washington, 1963 (after Chen and Orlob, 1975).
from the bottom. These changes were fairly well simulated by the model, as
indicated by the broken and dotted lines.
The subsequent warming period, illustrated by days 348 and 405, is closely
followed by the model, especially in the region of the thermocline. Upward
transport of salinity from lower strata is evident in reservoir data and seems to
be followed well by the model. Simulation of the dynamics of the wind-mixed
layer and internal mixing phenomena are still under investigation (Spigel and
1mberger, 1980; 1mberger, 1981).
266
-
..c.
o
c.
Q)
Temperature (OC)
(a)
30
20
E
-
..c.
o
c.
Q)
10
° °
°°
0'--_ _----'-- --'- ---'-- -'- --'--- --'
200 400 600 800 1000 1200 1400
Salinity (l\JaCI) (ppm)
(b)
FIGURE 7.10 Simulation of (a) temperature and (b) salinity in Wellington Reservoir
using DYRESM, and comparison with observations, 1975-76 (after Imberger et aI.,
1978).
267
Temperature (OCl
o.---_~---=-10~:>--y--~2:;-0_~_..::;30
I
0/)
I
01
/
10 DO ( 0
/
I
I
r. I
6
r>
6
E 20 I
\ 0 6
6
I 6
..r::.
.... I 66 Temperature
c- I
Ol 01
o 30 \
\
40 ~
I
I June 1967
1
0
5O'----- -----L -'----- ---.J
o 15
(a)
o.-----_.,---':..;=-_--.&_=-;=-_~_~30
I
C?
10 00/~1--;
I
0/
o
/
~/
E 20 t'~ 0
I
I 0
..r::.
.... I
c-
Ol <>
o 30 I
I
I
I
I
\
40 \
,
" August 1967
50 L - L- L -_ _----l
o 15
lb)
FIGURE 7.11 Observed (points) and simulated (lines) temperatures and dissolved
oxygen concentrations in Lake Hartwell, Georgia, 1967 (after Fontane and Bohan. 1974).
268
7.14.3. Lake Hartwell, Georgia
This reservoir, located on the Savannah River, experiences an annual thermal
cycle with the formation of a strong density gradient in the region of the thermo-
cline that, in turn, restricts transport of oxygen into the hypolimnion. In
addition, because the reservoir is large and summer flows are regulated by
upstream projects, the demand imposed on the hypolimnion often depletes
dissolved oxygen to the point of anaerobiosis by late summer or early fall.
This system has been simulated successfully using WESTEX, the specialized
one-dimensional water quality model of the Waterways Experiment Station
(Fontane a Bohan, 1974).
Figure 7.11 shows some representative" verification" test results for simula-
tion of temperature and dissolved oxygen concentration in Lake Hartwell for
1967. The model appears to give a fair account of the gradual depletion of
dissolved oxygen in the hypolimnion. Especially noteworthy is the simulation
of a steep DO gradient in the region of the thermocline, a characteristic of deep,
stratified impoundments with a high nutrient supply.
7.15. CONCLUDING COMMENTS
One-dimensional models of lakes and reservoirs have proven to be useful tools

for the assessment of water quality changes in impoundments that are small or
medium-sized, say less than about 50 km in major horizontal dimension, and
that experience a high degree of stratification. Attempts to simulate thermal
energy changes in such water bodies have been generally successful and a
variety of useful models are available for this purpose. Also, some results of
oxygen balance prediction with one-dimensional models indicate that they
can be used as tools to evaluate the impacts of operational or structural changes
in deep reservoirs. Considerable care is required, however, in interpreting
responses generated by these models, since they are designed to represent only
average conditions in a fully mixed, horizontal segment. They cannot account
for the spatial and temporal variability in organic loadings and biological
activities that affect the distribution of dissolved oxygen in the real system.
Consequently, they cannot predict the occurrence of extremes, such as anaerobic
conditions, that may be of greatest interest to decision makers.
The same caution should be exercised in extensions of the one-dimensional
conceptualization to simulate the aquatic ecosystem. While some one-dimen-
sional models appear to provide reasonable accounts of gross nutrient budgets
and primary production, they are increasingly unreliable in representing
biological responses at the higher trophic levels. Such phenomena as patchiness,
localized scavenging, and preferential grazing by motile predators cannot be
described adequately in a one-dimensional conceptualization ofthe water body.
269
These limitations are highlighted here to emphasize the need for balance
between the capabilities of the one-dimensional model and the complexities of
the real world. For larger water bodies, for those that are not strongly stratified,
and for cases where higher trophic levels are of major concern, we shall require
more realistic models. These requirements suggest more complexity and rigor,
multidimensional conceptualization, improved hydrodynamic representation,
and more information from the field. They impose a challenge to modelers that
the modelers seem to have been eager to accept. Evidence of this will be clearly
demonstrated in subsequent chapters of this book.
REFERENCES
Baca, R. G .. and R. C. Arnell (1976). A finite-element water quality model for eutrophic lakes.
Finite Elements in Water Resources vol. 4. eds. W. G. Gray, G. Pinder and C. Brebbia (Plymouth,
Devon: Pentech). pp. 125-147.
Baca, R. G., M. W. Lorenzen. R. D .. Mudd. and L. V. Kimmel (1974). A generalized water quality
model for eutrophic lakes and reservoirs. Bal/elle Pacific Northl1'est Lahoratories. Report
PB-256925.
Bella, D. A. (1976). Simulating the effect of sinking and vertical mixing in algal population dynamics.
Journal 0/ Water Pollution Control Federation 4(5), Part 2: R 140-R 152.
Bella. D. A .. and W. E. Dobbins (1968). Difference modeling of stream pollution. Proceedinqs 0/
Amerimn Society o/Cil'il En,qilleers. Journal o/SanitwT Ellyineerillq Dirisioll94 (SA5), Paper 619,
pp.995-1016.
Bloss, S.. and D. R. F. Harleman (1979). Effect of wind mixing on the thermocline formation in
lakes and reservoirs. MIT Department o( Cil'il Ellgineerinq, R. M. Parsons Lahoratory Technical
Report 249. 68 pp.
Burt. W. V. (1958). Heat budget terms for Middle Snake River Reservoir. US Fish and Wildlife
Serrice Report 58-7. 23 pp.
Burt. W. V. (\960). A forecast of temperature conditions for the research behind the proposed
Bruces Eddy Dam. H/ater Research Associates. Inc., Corrallis. OR Report. 16 pp.
Burt. W. V. (1963). Preliminary study on the predicted water changes in the Lower Snake River.
Waler Research Associates. Inc .. Con-allis. OR Report, 16 pp.
Chen. C. W .. M. Lorenzen, and D. 1. Smith (1975). A comprehensive water-quality ecologic model
for Lake Ontario. Report to Great Lakes Enrironmellt Research Lahoratory, NationalOceani,'
al/(I Atmospheric Administration hr Tetra Tech. Inc .. 202 pp.
Chen. C. W.. and G. T. Orlob (1968). A proposed ecologic model for a eutrophic environment.
Report 10 Federal Wafer Qualitr A.lfeI1lT, Southwest Regioll hy Water Resources Engineers, Inc.
Chen, C. W .. and G. T. Orlob (1972). Predicting quantity effects of pumped storage. Proceedin.l/s
o/American Society o/Cilil Ellqineers. Journal 0/ Power Dirisioll 98(POI), Paper 8984.
Chen. C W .. and G. T. Orlob (1973). Ecologic study of Lake Koocanusa. Report to US Army
Corps o(ElIlfineers. District o/Seal/le hr Water Resources En.lfineers. Inc .. 63 pp.
Chell. C. W .. and G. T Orlob (197~). Ecologic simulation of aquatic environments. Systems
Analysis and Simulation in Ecoloqy vol. 3. ed. B. C. Patten (New York. NY: Academic Press).
ch. 12. pp. 475-528.
Dake.1. M. K .. and D. R. F. Harleman (1966). An analytical and experimental investigation of
thermal stratification in lakes and ponds. MIT Departmellt o(Ciril E!ujincerilllj. R. M. Parsolls
Laboratory Tee/mical Report 99.271 pp.
270
Debler. W. E. (1959). Stratified flow into a line sink. Proceedin.4s oj" American Society oj" Ciril
Engineers, Journal of Engineerin.4 Mechanics Dil'ision 85(EM3): 51-65.
Elder, R. A .• and W. O. Wunderlich (1968). Evaluation of Fontana Reservoir field measurements.
Paper presented af ASCE Specialty Conference on Current Research into Effects of Reserroirs
on Water Quality. Porfland. OR.
Ertel, H. E. (1954). Theorie derthermischen Sprungschicht in Seen. Acta Hydrophysica I: 151-171.
Fischer, H. B., E. J. List. R. C Koh. J. Imberger, and N. H. Brooks (1979). Mixing in Inland and
Coastal Wafer (New York, NY: Academic Press). 483 pp.
Fogg, J. B., and E. G. Fruh (1973). A dissolved oxygen model for the hypolimnion of Lake Lyndon
B. Johnson. Center for Research in Water Resources, Unit'Crsity of Texas, Ausfin, CRWR Report
103,90 pp.
Fontane. D. G .. and J. P. Bohan (1974). Richard B. Russell Lake water quality investigation.
Waterways Experiment Sfal/on, Vicksburg, MS Technical Report H-74-14, 26 pp.
Gallagher, R. H .. J. A. Liggett, and S. T. K. Chan (1973). Finite-element shallow lake circulation
analysis. Proceedinys oj" American Societv of Ciril Engineers, Journal oj" Hydraulics Dil'ision
99(HY7), Paper 9855, pp. 1083-1096.
Gaume, A. M., and J. H. Duke. Jr. (1975). Computer program documentation for the reservoir
ecologic model EPAECO. Report to US Enrironmell/al Protection Agency, WashinguJn, DC by
Water Resources Engineers, Inc.
Harleman. D. R. F., and K. A. Hurley (1976). Simulation of the vertical thermal structure of lakes
under transient meteorological conditions. Dynamics oj" Stratification and oj" Stratified Flows i/1
Large Lakes: Proceedin.4s of Workshop of the Comll1illee on Lake Dynamics (Windsor, Ontario:
International Joint Commission, Regional Office), pp. 79 96.
Henderson-Sellers, B. H. (1978). Forced plumes in a stratified reservoir. Proceedings of American
Societv of Cieil Engineers. Journal of Hl'draulics Dieision I 04(HY4): 487-50 I.
Huber, W. C, D. R. F. Harleman. and P. J. Ryan (1972). Temperature prediction in stratified
reservoirs. Proceedings (jj" American Society of Cil'il Enyineers, Journal oj" Hydraulics Dit'ision.
Paper 8839, pp. 645-666.
Hurley-Octavio, K. E., G. H. Jirka, and D. R. F. Harleman (1977). Vertical heat transport mech-
anisms in lakes and reservoirs. MIT Department of Civil Engineering. R. M. Parsons Laboratorv
Technical Report 227. 70 pp.
Hutchinson, G. E. (1957). A Treatise on Limnology vol. I (New York, NY: Wiley).
Hydrologic Engineering Center (1974) (revised 1978). WQRRS. generalized computer program
for river-reservoir systems. US Army Corps of Engineers, H EC, Davis, CAUser's Manual
40\-100. 100A, 210 pp.
Imberger, J. (1981). Influence of stream salinity on reservoir water quality. Agricultural Water
Managemen/4, 9 pp.
Imberger, J., J. Patterson, B. Hebbert. and I. Loh (1978). Dynamics of reservoirs of medium size.
Proceedinqs oj" American Society of Cil'il Engineers. Journal oj" Hydraulics Dil'ision 104:
725-743.
19:297-304.
Kao, T. W. (1965). The phenomenon of blocking in stratified flow. Journal oj"Geophysical Research
70(4): 815-822.
King. r. P., W. R. Norton, and K. R. Iceman (l974). A finite element model for two-dimensional
flow. Finite Elements in Flow Problems. eds. J. T. Oden et al. (Huntsville, AL: University of
Alabama. Huntsville Press). pp. 133-137.
Kinnunen. K., J. S. Neimi, and J. Eloranta (1978). Adaptation of the EPAECO model to a lake in
Central Finland. Modelin.4 the Water Qualitl' oj" the Hydrological Cycle. Proceedings oj" Baden
Symposium, Septemher. International Association of Hydrological Sciences Publication 125.
pp.115-127.
Larsen, D. P.. and H. T. Mercier (1975). Shagawa Lake recovery characteristics as depicted by
271
predictive modeling. Paper prcsentcd at American Institute of Biological Scil'llces Meeting.
Corrallis, OR.
Leung. K. K., R. A. Park. D. J. Desormeau. and J. Albanese (1978). MS CLEANER-An overview,
Proceedings of Conference on Modeling and Simulation, Pittsburgh, PA (Troy, NY: Center for
Ecological Modeling, Rensselaer Polytechnic Institute), 8 pp.
Markofsky, J., and D. R. F. Harleman (1973). Prediction of water quality in stratified reservoirs.
Proceedings of American Society of CiL'i/ Engineers, Journal of Hydraulics Division 99(HY5),
Paper 9730, pp. 729-745.
McEwen. G. F. (1929). A mathematical theory of the vertical distribution of temperature and
salinity in water under the action of radiation, conduction, evaporation and mixing due to the
resultant convection, Scripps Institute ofOceanoyraphy Technical Bulletin 2: 197-306.
Niiler, P, O. (1975). Deepening of the wind-mixed layer. Journal of Marine Research 33 :405-422,
Norton. W. R., and I. P. King (1975), Mathematical simulation of water temperature to determine
the impact of raising an existing dam. Progress in Astronautics and Aeronautics 36: 53-70.
O·Melia. C R. (1974), Phosphorus cycling in lakes. North Carolina Water Resources Research
Institute. Rale(qh Report 97, 45 pp.
Orlob, G. T. (1965), A mathematical model of thermal stratification in deep reservoirs. Conference
of American Fisheries Society, Portland, OR, 14 pp,
Orlob, G. T, and L. Selna (1967), Prediction of thermal energy distributions in deep reservoirs,
Proceedings of 6th Annual Sanitation and Water Resources Engineering Conference, Vanderhilt
UnicersilY, Nash1,ille. TN, pp. 64-77.
Orlob, G. T, and L. Selna (1970). Temperature variations in deep reservoirs. Proceedings of Ameri-
can Society ofCiril Engineers, Journal of Hydraulics Dicision 96(HY2), Paper 7063.
Park, R, A., et al. (1974a). A generalized model for simulation of lake ecosystems. US International
Biological Program Technical Report.
Park, R. W .. D. Scavia, and N. L Clesceri (1974b). CLEANER, the Lake George model. US
International Biological Program, Eastern Deciduous Forest Biome Contrihution 184, 13 pp.
Raphael. J. M. (1961), The effect of Wanapum and Priest Rapids Dams on the temperature of the
Columbia River. Public Utilily District No. 2 (J( Grant County, Ephrata, W A Report,
22 pp.
Raphael, J, M. (1962a). The effect of Wells and Rocky Reach Dams on the temperature of the
Columbia River. Puhlic Utility Districi No, 2 of Grant County, Ephrata, WA Report, 9 pp.
Raphael, J, M. (1962b), Prediction of temperature in rivers and reservoirs. Proceedings of American
Society ofCil'i1 Engineers, Journal of Power Di1,ision 88(P02): 157-181.
Ryan, P. J., and D. R. F. Harleman (1971). Prediction of the annual cycle of temperature changes in a
stratified lake or reservoir: Mathematical model and user's manual. MIT Department of Ciril
Engineering, R. M. Parsons Laboratory Technical Report 137,
Scavia, D .. and R. A. Park (1976). A user's manual for the aquatic model CLEANER. Rensselaer
Fresh Water Institute at Lake George. Rensselaer Polytechnic In.Hitute, Troy, NY Report 76-16,
81 pp.
Snodgrass, W. J., and C R. O'Melia (1975). A predictive phosphorus model for lakes-Sensitivity
analysis and applications. Encironmental Science and Technology 9: 937-945.
Spaulding, D. B., and V. Svensson (1976). The development and erosion of the thermocline. Heat
Transfer alld Turhulent Buoyant Conncction (Washington, DC: Hemisphere), pp. 113-122.
Spigel, R. H. (1978). Wind mixing in lakes. Ph.D. Dissertation. Unil'ersity of CalIfornia. Berkeley.
Spigel, R. H., and J. I mberger (1980). The classification of mixed-layer dynamics in lakes of small
to medium size. Journal of Pln'sical Oceanography 10(7): 1104-1121.
Stefan. H., and D. E. Ford (1975). Temperature dynamics in dimictic lakes. Proceedings o/Amen'-
can Society 0/ Ciril Engineers, Journal of Hydraulics Dil'ision, Paper 11058, pp. 97-114.
Sundaram. T R., C C, Easterbrook. K. R. Piech, and G. Rudinger (1969). An investigation of
the physical effects of thermal discharge into Cayuga Lake. Cornell Aeronautical Lahoratory.
BufJalo, NY Report VT-2616-0-2.
272
Sundaram, T. R., and R. G. Rehm (1973). The seasonal thermal structure of deep temperate lakes.
Telbes 25: 157-167.
Svensson, J. (1978). A mathematical model or the seasonal thermocline. Department or Water
Resources Enqineerinq, Lund Institute or Technoloqy', Lund, Sweden Thesis Report 1002.
Tennessee Valley Authority (1972). Heat and mass transfer between a water surface and the atmo-
sphere. TVA Water Resources Research Enqineerinq Laboratory, Norris. TN Report 14, 123 pp.
Vollenweider. R. A. (1965). Calculation models or photosynthesis-depth curves and some implica-
tions regarding day rate estimates in primary production. Memorie dell IstitufO Italiano di
Idrobioloqia 18 Suppl: 425-457.
Vollenweider, R. A. (1969). Miiglichkeiten und Grenzen e1ementarer Madelle der StoRbilanz von
Seen. Archil' Hydrobioloqie 66: 1- 36.
Water Resources Engineers, Inc. (1968). Prediction or thermal energy distribution in streams and
reservoirs. Report to California Department or Fish and Game, 90 pp.
Water Resources Engineers, Inc. (1969). Mathematical models ror prediction or thermal energy
changes in impoundments. US Em'ironmental Protection Aqency Water Pollution Control
Research Series 16130 EXT, Contract 14-22-422. US Government Printing Office, Washington,
DC, 157 pp.
CHAPTER 7: NOTATION
a area of horizontal plane through a reservoir

A advection rate
b average width of a reservoir
B benthic animal biomass concentration
d average depth of a reservoir
D diffusion rate
E effective diffusion rate
F nekton (fish) biomass concentration
h heat flow
H heat content of a volume element
K turbulent kinetic energy
K 1 BOD decay coefficient
K 2 reaeration coefficient
K 4 benthic demand coefficient
I length of a reservoir
L biochemical oxygen demand
m mortality rate
P phytoplankton biomass concentration
q discharge per unit width or depth; heat flux per unit area
Qa lateral advected flow
r respiration rate coefficient
s sedimentation rate
S source or sink of constituent
U average flow-through velocity of a reservoir
273
w vertical advection velocity
z depth (above bottom of reservoir)
f3 density gradient; ratio of absorbed to net incoming radiation
b thickness of withdrawal layer
( rate of energy dissipation per unit mass
y/ bulk (light) extinction coefficient; digestion efficiency of predator
e temperature
(J stoichiometric oxygenation factor
lJ1 release rate.
8 Two- and Three-Dimensional Mathematical
Models for Lakes and Reservoirs
M. Watanabe, D. R. F. Harleman,
and O. F. Vasiliel'
8.1. INTRODUCTION
A knowledge ofthe hydrophysical processes in lakes and reservoirs is important

for the understanding of ecological processes, particularly (i) the transport of
various materials such as nutrients needed by aquatic organisms, or pollutants;
(ii) temperature distribution; (iii) population dynamics and transport of
plankton; and (iv) distribution of water quality parameters such as DO, pH,
BOD, and COD.
There is no universal model detailed enough to describe the entire physical
and biochemical behavior of lakes. Simplifications must be introduced in order
to formulate models and to make solutions possible. The degree of simplifica-
tion, including averaging or the dropping of higher-order terms, necessarily
reduces the generality of a mathematical model.
One-dimensional models can simulate the vertical (or longitudinal) behavior
of water bodies. While the vertical structure is largely important for the analysis
of lake behavior, from an ecological point of view vertical, one-dimensional
models cannot predict the horizontal divergences and convergences of the
wind-driven surface current, the return current in lake basins, the effects of the
earth's rotation in large basins, or upwelling and downwelling. Patterns of
distribution or circulation that extend over large space and time scales are
often determined by the shape of the lake or by short-term strong disturbances
(e.g. localized shear instabilities) that are characterized by much shorter
space and time scales. This suggests the necessity for three-dimensional circula-
tion models capable of simulating the important mechanisms over a wide
range of scales (Mortimer, 1974).
274
275
Although the equations of fluid dynamics can describe motions that include
all of these scales, practical difficulties prohibit (or make unnecessary) the use
of the full equations of motion for real problems involving a wide variety of
phenomena. Considerable effort and ingenuity have been expended to approxi-
mate these equations in order to obtain simpler equations and methods of
solution.
The predictive power and, therefore, the usefulness of such models will be
greatly dependent on the depth of understanding and recognition of those
mechanisms that are physically important (excellent reviews have been written
by Mortimer (1974), Boyce (1974), and Csanady (1975» and also on well
planned model verifications in actual lakes. In the last few years, hydrodynamic
models for the calculation of lake circulation have become quite detailed and
realistic and have contributed greatly to our understanding of lake behavior.
This chapter is concerned with the state of the art of two- and three-di-
mensional mathematical models for describing the hydrophysical behavior
of lakes and reservoirs under natural influences (such as wind). In addition,
mathematical models of cooling ponds or reservoirs under artificial heat
loading from power plants will be discussed. The latter topic is relatively new,
but significantly important for the understanding of eutrophication processes
and for predicting other possible effects of excess heat loading on lake eco-
systems. Artificial heat loads may change the entire flow and temperature
fields and affect the rates of biochemical reactions.
In section 8.2, a brief discussion of some general characteristics of currents in
lakes is followed by the three-dimensional, time-dependent hydrodynamic and
heat exchange equations, boundary conditions, and parameters basic to all the
models. These general equations are simplified, based on length and time scales
significant for particular situations.
Mathematical models of wind-driven circulation can be classified broadly as
horizontal two-dimensional models, two- and multi-layer models, Ekman-type
models, vertical two-dimensional stratified flow models, and three-dimensional
models.
Various models are compared in terms of their basic structure and assump-
tions, physical validity, and range of applicability. Brief reviews of numerical
techniques, such as finite-difference and finite-element methods, and stability
criteria are presented. The next section briefly reviews two- and three-
dimensional water quality models in which hydrodynamic and water quality
models are coupled. Because computer capacity is limited, integration of com-
plex hydrodynamic circulation models and ecological models causes problems
in terms of computation time, core memory, and cost. Simplifications based on
physical reality are essential.
Following this, a critical survey is presented of two- and three-dimensional
mathematical models for lakes, reservoirs, and cooling ponds, which require
different simplifications and solution techniques.
276
8.2. GENERAL CHARACTERISTICS OF HYDROTHERMAL
CIRCULATION MODELS AND GOVERNING EQUATIONS
8.2.1. Introduction
Circulation in lakes is primarily caused by wind shear acting on the water
surface. Temperature or density differences produce some additional hydro-
dynamic effects that result in density currents. At the same time, density gradients
affect currents and play an important role in stratified water bodies.
The geometry of a lake (or reservoir) has a significant influence on its
circulation. Water circulation influenced by wind, density gradients, and many
other factors is therefore complex and includes various motions associated
with different length and time scales. Boyce (1974) summarized motions in
lakes and their associated length and time scales, as shown in Table 8.1.
The relative importance of each force can be evaluated by selecting the basic
scale of interest. Therefore, the comparison of various forces as they affect
motions at these scales allows the simplification of the governing equations of
motion by neglecting the less important terms. The differences between existing
mathematical models are therefore related to the different scales that are sig-
nificant for the specific conditions. In this section, the basic equations describing
fluid motion and heat transfer in lakes or reservoirs are presented before each
model is discussed in detail.
8.2.2. Governing Equations

The foundation for the development of models of fluid motion lies in the Navier-
Stokes equations for the conservation of momentum. In addition, conservation
laws for mass and energy are required. The complete set of equations is without
formal, general solutions, yet methods have been devised to obtain solutions
for particular cases (including numerical methods).
The three components of velocity, u, v, w, and the pressure p are decomposed
according to the Reynolds conceptualization into mean and fluctuating com-
ponents:
u = u+ u', v = l' + v', w= w + w', p = p + p'.
The following approximations are introduced in the Navier-Stokes equations.
(i) The Boussinesq approximation: the density variation in a water body

is much smaller than the density itself; therefore, constant density can
be used in the equation of motion, except in the term for the buoyancy
force.
(ii) The hydrostatic approximation: in lakes and reservoirs, the vertical
velocity component is usually small. Therefore, the vertical component
of acceleration is much smaller than the gravitational acceleration.
277
If the expressions above are substituted into the Navier-Stokes equations
and the time averages are taken, the Reynolds equations for a fluid body on a
rotating earth are as follows:
aii a a a I ap 2-
- + - iiii + - iiI' + - iiw - fv = - - - + vV u
at ax ay az . pax
- -a-,-,
u u - -a-,-,
u v - -0 - -
u'w' (1)
8.
ox ay az'
ou 0 __ 0 __ 0 __
- + - vu + - vv + - vw + j u =
._
-
1 op 2-
- - + vV v
at ox oy oz poy
-
a -,-, a -,-, a -,-,
- vu - - vv - - vw (8.2)
ax oy oz'
lop
o= - g - - - (8.3)
paz'
and the continuity equation can be expressed as
(8.4)
where
t is time,
x, y, z are the coordinate axes ( + z is vertically upward),
u, v, ware the mean velocities in the x, y, z directions,
u;uj is the time-averaged turbulent eddy transport of momentum,
g is the gravitational acceleration,
p is the density.
p is the pressure.
.f is the Coriolis parameter,
v is the (molecular) kinematic viscosity,
V2 = a21ax 2 + a21ay2 + a2laz 2.
Similarly, the equation of heat balance for a turbulent flow is
-af + -a T--u + -a T--v + -a T--w = Xvn2T - -a T" u - -a T" v - -a T-----'-W-----' + QH,

at ax ay az ax ay oz
(8.5)
where X is the (molecular) thermal diffusivity, T'u; is the time-averaged turbulent

eddy transport of heat in direction Xi and QH is the heat source. Advective
transport fUi and turbulent eddy transport T'u; depend on the state of flow.
tv
--.l
00
TABLE 8.1 Classification of Motions in Lakes (according to Boyce, 1974). Letters in parentheses refer to the type of scale
used. M, amplitude of motion; S, distance over which phenomenon varies significantly; P, period; T, time interval over which
phenomenon varies significantly; C, wave speed; V, vertical particle velocity; H, horizontal particle velocity. The governing
terms in the equations of motion and continuity are listed in column 6 according to the code: I, time-dependent horizontal
accelerations; 2, time-dependent vertical accelerations; 3, advective component of horizontal acceleration; 4, advective
component of vertical acceleration; 5, Coriolis force; 6, pressure gradient force due to slope of free surface; 7, pressure gradient
force due to slope of the thermocline; 8, pressure gradient force due to atmospheric pressure field; 9, variations in bottom
topography; 10, wind energy/stress; 11, internal stresses arising from horizontal current shear; 12, internal stresses arising
from vertical current shear; 13, friction against lake boundaries; 14, potential energy changes due to surface heating and cooling.
Phenomenon Length Scale Timc Velocity Scale Dynamics of Notcs

Scale Major Components
Horizontal Vertical
Wind-driven surface 10 m (S) 1m (M) IS (P) 10 m s -I (C) 1.2,6,10

gravitational waves 1m s-' (V, H)
Surface gravitational
100 km (S) 10 cm (M) 2cms- I (H)
waves-seiches 2-10 h (P) 1,6,9,10
Short, freely propagating
internal waves 100 m (S) 2m (M) 5 min (P) 2 cm s - I (H) 1.2,7,10
Long, propagating internal
waves steered by
topography 10 km (S) 2m(M) I day (T) 50 em s - I (C) 1,5,7,9,10
Internal gravitational
standing waves or seiches 10 km (S) 2m(M) 16 h (P) 10 em s - 1 (H) 1,5,7,10
Surface wind drift 10 cm (S) 2 cm s - 1 (H) 10,12
Coastal currents 10 km (S) I day (T) 10 cm S-I (H) All Horizontal scale measured
perpendicular to coast
Upwelling and
downwelling 10 km (S) 10m (M) I day (T) < I cm s - I (V) All Horizontal scale measured
N
--J perpendicular to coast
\0
Wind-driven horizontal
circulation 100 km (S) 100 m (S) I day (T) 10 cm s- I (H) All
Basin-wide horizontal
diffusion 100 km I month(T) All Integrated effect of all
motions
Langmuir'circulations,
vertical mixing of
epilimnion 10 m (S) I h (T) I cm s -I (V) I, 2. 3, 4, 10. 12, Dynamics not yet
14. and others understood
Formation and decay

of thermocline 10-100 m (S) I month (T) 10, 12. 14
280
In addition, an equation of state defines the relationship between temperature
and density:
p = peT)· (8.6)
Initial conditions and boundary conditions have to be specified for the six
unknowns ui , p, p, and r. Also, the eddy transport terms u; uj and T' u; need to be
parameterized through the flow state variables.
8.2.2.1. Boundary Conditions

The specification of boundary conditions depends on the nature of the problem.
At the free surface

(1) A kinematic free-surface condition:
a/'] _a/'] _a/'] _)

(-at + u ax
- + u-
ay
- w
z=ry
= 0, (8.7)
where/'] is the elevation of the free surface;

(2) a dynamic free-surface condition for the pressure on the air-water
interface:
p(x, y, /'], t) = Pa(x, y, t). (8.8)
where Pa is the atmospheric pressure;

(3) wind stress on the air-water interface:
t(x, y, /'], t) = ts(x, y, t), (8.9)
where t s is the specified surface wind stress;

(4) surface heat flux:
q(x, y, /']. t) = rPn(x, y, t), (8.10)

where rPn is the net heat flux through the surface.
At the bottom and side boundaries

(5) A kinematic condition:
ah + uah _
(u ax w) = 0, (8.11)
ay ==-11
where h is the height from the datum to the bottom;

(6) no fluid motion or a specification of surface stresses in terms of near-
surface velocities;
(7) a specification of either temperature or heat flux.
281
At the inflow or outflow boundaries
(8) Inflow and outflow velocities or volume flux are specified;
(9) inflow temperature is given by
T = fT(X, y, z, t); (8.12)
(10) outflow temperature satisfies the following relation (Vasiliev, 1978a,b):
of of of of
-01 + if -ax + ii -ay + W-az = QH' (8.13)
The rigid lid approximation can be considered as a special free-surface

boundary condition that assumes no change in surface elevation. This assump-
tion eliminates all surface waves; therefore, currents resulting directly from wind
shear can be calculated.
The rigid lid approximation simplifies the kinematic condition at the free
surface:
w = 0 at z = 0 (8.14)
and the pressure can be set to an arbitrary value at any point. However, the
rigid lid does not act like a solid boundary with respect to friction and shear.
Therefore, in the rigid lid approximation pressure differences are transmitted
instantaneously everywhere in the water body instead of propagating with
the celerity of gravity waves.
Since the propagation of gravity waves is very rapid compared with the
response of the wind-driven current, this approximation has little effect on the
response of currents. The advantage of the rigid lid approximation is that it
results in a simpler computational process and the ability to use longer time
steps. Besides the boundary conditions, the initial conditions should be given:
the distributions of velocity and temperature in the water body and the free-
surface position at an initial time.
8.2.3. Turbulent Shear and Diffusion
8.2.3.1. Turbulence Models

The covariances pu;uj in (8.1)-(8.3), which result from time averaging over the
scale of the turbulent fluctuations, represent an effective shear acting in direction
Xi on a plane perpendicular to Xj' In order to solve the governing equations,
these covariances need to be specified. This specification has been the subject
of different theories of turbulence, starting with Reynolds and Boussinesq in
the later part of the last century. A useful summary of these theories and their
practical limitations has been given by Vreugdenhil (1973).
A complete and accurate description of turbulence is complicated by the fact
that different mechanisms govern the turbulent fluctuations (measured by
282
its kinetic energy e):
e = !(U'l + V'l + W'l).
These are (Schlichting, 1968): (a) the generation of turbulence as a result of
gradients in the mean motion, (b) the transfer of turbulence from more turbulent
to less turbulent zones, and (c) the dissipation of turbulence by internal friction
between fluid elements. Thus, successful descriptions are usually limited to
cases in which a local equilibrium persists between these three mechanisms,
such as at a cross section of uniform pipe or open-channel flow.
Attempts to quantify the turbulent shear stresses range from the eddy
viscosity concept to direct transport equations for the shear stresses. All
hypotheses involve empirical coefficients and need to be verified with experi-
ments. According to Boussinesq, the turbulent shear stress rL can be formulated
as
t
Tij =
-,-,
- pUjUj = Pf.j aX
(au + aXj
au
j j)
, (8.
15
)
j
completely in analogy to the laminar (molecular) shear stress. The coefficient

f.j in (8.15) is termed the kinematic eddy viscosity and is represented as a vector
signifying the directional dependence. (This is a simplification because the eddy
viscosity should be a higher-order tensor, as shown by Hinze (1959; also Monin
and Yaglom, 1965).)
8.2.3.2. Diffusion Models

The covariances ui T' represent an effective transport of heat by the random,
turbulent displacement of lumps of fluid and the subsequent exchange with
adjacent fluid masses. The specification of ui T' presents some difficulty because
temperature, by virtue of the equation of state (8.6), also affects the density of
the fluid. On the other hand, most research on turbulent diffusive transport
uie' has been focused on passive (inert) scalar quantities of small concentrations
e, which therefore do not change the density of the ambient fluid.
Similarly to the turbulent shear models, two approaches have been followed
for diffusion models, namely the eddy diffusivity concept and the direct specifica-
tion of ui T' by means of the equations for higher-order statistical moments.
Discussion will concentrate on the eddy diffusivity method since there are
computational restrictions with the other method. The turbulent diffusive
transport of heat is expressed by
ql = -uiT' = K Ti aar, (8.16)

Xi
where K Ti is the eddy diffusivity, in analogy to the Fourier law of heat con-
duction by molecular diffusion.
283
8.2.4. Eddy Viscosity and Eddy Diffusivity
8.2.4.1. Eddy Viscosity

Unlike the molecular viscosity, the eddy viscosity £j is not simply a fluid property,
but rather a function of the flow conditions. An approach to describe this
dependence is the mixing length concept, first proposed by Prandtl, and used
by Taylor, von Karman, and others. The mixing length concept is, in principle,
an empirical relation:
(8.17)
where I is the characteristic (mixing) length, du/dz is the mean velocity gradient
perpendicular to the flow, and z is the distance from the bottom. This concept
has been useful for describing certain equilibrium shear flows, in particular for
uniform open-channel flow in which the vertical eddy viscosity can be written as
1= KZ(1 - z/H), (8.18)
where
u. = (T o/p)1/2 is the shear velocity,
To is the bottom shear,
K = 0.4 is the von Karman constant,
H is the water depth.
Another approach can be made by using turbulence energy e and a length

scale of turbulence:
(8.19)
This requires an additional transport equation for e (derived from the Navier-
Stokes equations), involving higher-order covariances (statistical moments)
and requiring more empirical coefficients. The equation of the turbulent energy
balance is discussed in detail by Monin and Yaglom (1965).
A more complex turbulence closure model is based on two transport equa-
tions: one for e and the other for the turbulent energy dissipation £ (Jones and
Launder, 1973; Launder, 1976). These more elaborate turbulence models
require substantial computational effort. Individual models have usually been
verified with a limited class of experiments (mostly boundary layer types).
Examples of application of turbulent models of the latter types will be given in
section 8.3.5.
Effects of density stratification

Turbulence in the presence of a density stratification is discussed by Phillips
(1969). A stable stratification hampers vertical displacement of fluid. Thus,
284
for turbulence to be maintained there must be a sufficiently great destabilizing
supply of turbulent kinetic energy to overcome the stabilizing effect of the
potential energy of stable density stratification. The parameter that character-
izes these relative effects is the Richardson number:
Ri = (g/p)op/oz (8.20)
(OU/OZ)2 '
where z is the coordinate in the direction of g and u is the horizontal velocity.
The Richardson number is a local parameter and varies with depth as the density
gradient op/oz and velocity gradient ou/oz change. Large values of Ri reflect a
high degree of stability and reduced vertical transport of momentum (shear).
A theoretical condition for turbulence to be maintained is
Ri < i. (8.21)
An elementary derivation of this criterion, including critical remarks about its
significance, is given by Long (1970; also Monin and Yaglom, 1965).
Little is known in exact, quantitative terms about the dependence of the
vertical eddy viscosity coefficient £z on the Richardson number. Karelse (1974)
compared available data with a variety of empirical formulas proposed by
different investigators and discovered considerable scatter. A characteristic
empirical formula is one by Munk and Anderson (1948):
£z = £zo(1 + 10 Ri) - 1 12, (8.22)
where £zo is the eddy viscosity under nonstratified conditions.
In many instances, when the stratification is significant and a distinct
interface exists, it is advantageous to make use of the discretely layered structure
and apply one- or two-dimensional stratified flow theory with more or less well
defined interfacial friction factors.
8.2.4.2. Eddy Diffusivity

Eddy diffusivities K j are again strongly dependent on flow conditions. The ratio
between the eddy viscosity and the eddy diffusivity is termed the turbulent
Prandtl number:
£.
Pr =~. (8.23)
Kj
The Prandtl number is about unity for wall turbulence such as the shear flow
considered in (8.17) and (8.18), so that these equations are directly applicable
for the vertical diffusivity K z •
Diffusion in open water, such as in lakes or coastal zones, is in contrast to
open-channel shear flow in which the diffusive transport can be effectively
related to the mean flow characteristics. Oceanic or lake turbulence represents a
285
spectrum of different eddies resulting from the breakdown of large-scale circu-
lations in shore zones and perturbation by wind and waves. Attempts to analyze
this situation have shown that the horizontal diffusive transport K h depends on
the length scale I of the phenomenon. The most widely used formula is the four-
thirds power law:
(8.24)
where AD is the dissipation parameter. Useful summaries of lake and ocean
diffusion data have been provided by Yudelson (1967), Okubo (1968), and
Ozmidov (1968).
Effects of density stratification

Density stratifications have a pronounced effect on the vertical turbulent
diffusivity K z • Furthermore, K z is more influenced by stratification than is the
eddy viscosity Ez . Available laboratory and field data on the relation of Kz/E z (the
inverse of the Prandtl number) to the local Richardson number have been
summarized by several investigators. For example, the results of Kullenburg
(1974) are shown in Figure 8.1. Since eddy diffusion coefficients are more readily
determined from tracer diffusion experiments, qualitative relations such as those
in Figure 8.1 are of great practical significance, as they allow one to estimate the
eddy viscosity coefficients, which are difficult to determine directly.
There have also been many attempts to relate K z to the local Richardson
number. A typical formula is that of Munk and Anderson (1948):
Kz = K zo (1 + 13° Ri)-3/2, (8.25)
where K zo is the eddy diffusivity without stratification, indicating a reduced
value of K z for large Ri. Further compilations of data and comparisons of
different formulas have been made by Okubo (1962), Koh and Fan (1970), and
Karelse (1974).
8.2.5. Surface Shear Expressions

The turbulent shear stress t s exerted at the water surface is usually expressed as
(8.26)
where W is the wind velocity measured at a certain height, usually 10 m, Pa is the
air density, and C w is a "wind stress coefficient." The coefficient depends on the
surface roughness, which in turn is dependent on wind velocity. Cw is also
dependent on the fetch, the stability of the air mass, the relative temperatures
of air and water, and the topography and roughness upwind of the water body.
In general, the roughness of the air-water interface is created by a great variety
of waves differing in height, shape, and phase velocity. Moreover, the waves are
subjected to continuous and irregular changes.
286
•."
... ,...
'\.
• v~
••
10
Ri
FIGU RE 8.\ Ratio of the vertical diffusion coefficient and the vertical momentum
transfer coefficient plotted against the Richardson number. The straight line represents
(K)(J Ri = 0.05. The curve represents Ellison theory with the critical flux Richardson
number equal to 0.005. Experimental data are from: 0 Bowden. 0 and D, Ellison and
Turner. • Lofquist. and x Swinbank (after Kullenburg. 1974).
Depending on wind speed in general, two regions can be distinguished from

available data (Wu, 1969): (i) the region of roughness establishment, in which
surface roughness increases with wind velocity, and (ii) the region of established
roughness, in which surface roughness reaches an equilibrium value. Table 8.2,
taken from Wu (1969), summarizes different expressions for the surface rough-
ness Cw = C 10 (wind velocities in meters per second measured at 10 m) as a
function of wind speed, sea state, or Beaufort number.
8.2.6. Surface Heat Flux

Several thermophysical processes, such as radiation, conduction, and evapora-
tion, determine the heat transfer through the water surface. Detailed descrip-
tions of these processes were presented in Chapter 5.
TABLE 8.2 Wind Stress Coefficients C to as Functions of Wind Conditions (from Wu, 1969).
Type of Breeze Light Wind Strong Wind

Wind
I I I I I I I I I I I I I I I I I I I I I
Wind speed I I I I I I I
I 3 5 7 9 II 13 15 18 21 24 27 30
W,o(ms-')
Sea state 0 1 2 3 4 5 6 7 8 9
tv
00
-..J
Beaufort
0 I 2 3 4 5 6 7 8 9 10 II 12
number
Boundary Aerodynamically Transition Aerodynamically rough flow

layer flow smooth flow
conditions
Region of
surface Region of roughness establisment Region of established roughness
roughness
Surface
roughness 0.03 < k < 0.3 0.3 < k < 1.5 8
k (em)
Wind stress
J
formula ~ 2.6·IO-
In(9.l90jv)
( r Cn(~'SjOOI~6C'OUTO)
for CIO
r
Approximate 0.00125
wind stress 0.5W: 02 .1O- J 2.6. 10- J
U '' ,
1O
formula
288
In the analysis of the heat balance for water bodies the net surface heat flux
cPo is usually introduced as a boundary condition directly at the free surface.
This is only approximately correct inasmuch as radiation absorption does not
occur simply at the air-water interface but rather in a water layer close to the
surface, the thickness of the layer being dependent on the absorptive character-
istics, such as turbidity. This approximation is sufficiently accurate for many
applications. In specific applications, such as the thermal stratification of lakes,
the actual radiation absorption should be accounted for in the model formula-
tion.
The general form of cPo is nonlinearly dependent on the water surface
temperature. The equation may be linearized over a certain range of surface
temperatures upon introduction of an equilibrium temperature TE • This is the
water surface temperature at which, under given meteorological conditions, the
net surface heat flux is zero. A linear approximation can be defined as follows:
(8.27)
where K H denotes the heat transfer coefficient valid for a certain range of
surface temperatures Ts. This linearization of surface heat transfer has an
advantage only with analytical solutions or simple steady state models. Since
both K H and TE are functions of time, direct computation of cPo is recommended
for transient numerical models (Ryan and Harleman, 1973).
Plate and Wengefeld (1979) made a survey of the transport processes for
momentum, heat, and mass at a lake surface. The survey includes, in particular,
the description of the formation of boundary layers near the free surface and
the analysis of wave generation.
8.3. MODELS OF WIND-DRIVEN CIRCULAnON

Although circulation and heat exchange in lakes or reservoirs can be expressed
in general by the equations of motion, the continuity equation, and the thermo-
dynamic energy equation, modelers wish to isolate the particular phenomena
that they want to simulate and to retain only the corresponding terms in the
governing equations.
Existing circulation models can be categorized into the following groups,
based on the relevant assumptions and simplifications:
(1) horizontal two-dimensional circulation (single-layer) models,

(2) two- and multi-layer models,
(3) Ekman-type models,
(4) vertical two-dimensional stratified flow models, and
(5) three-dimensional models.
Each type of model has its own advantages and applicability. The derivation of
the governing equations and limitations of the models will now be discussed.
289
8.3.1. Vertically Averaged Two-Dimensional Circulation Models
(Single-Layer Models)
Single-layer models are formed by averaging the three-dimensional equations
over the depth, with bottom and surface boundary conditions. The result of
these manipulations is a set of reduced two-dimensional equations, which are
comparatively easy to analyze and require relatively little computer time. These
models do not give details of the vertical variation of the flow and calculate
only the total mass transport.
~The vertically averaged formulation was originally developed in the theory of
long waves and in open-channel hydraulics. The first numerical model of storm
surges in the North Sea based on such a formulation was developed by Hansen
(1956). Platzman (1963) applied a similar model to several of the Great Lakes,
Reid and Bodine (1968) to the Gulf of Mexico, and Heaps (1969) to the North
Sea. Later, this approach was used to simulate circulation in shallow, vertically
well mixed systems by Leenderste (1967), Cheng and Tung (1970), Simons (1971),
Cheng (1972), Abbott et al. (1973), Gallagher et al. (1973), and Wang and
Connor (1975).
The principal limitation of vertically averaged models is that they do not
consider the effects of velocity and density variations in the vertical direction,
though these details are often necessary for the complete understanding of the
flow characteristics and for an accurate description of dependent problems in
ecological applications (for example, when there are bottom return currents in
wind-induced lake circulation).
However, the two-dimensional model can be adequate for the consideration
of pronounced unsteady flows in shallow water bodies and might be useful for
preliminary investigations of flows. This type of model is also the basis for
multilayer models. The basic equations for this model are obtained by integrat-
ing the equations of motion and continuity from the bottom, Z = - hex, Y),
to the surface, Z = r,(x, y). (For convenience, hereinafter the bar indicating
mean velocity will be omitted.)
The integrated continuity equation is
f
~ au f~ av
- dz + -a dz + w I~ - w 1- h = O. (8.28)
-h aX -h Y
According to the Leibnitz rule,
-
a f~ u dz - a1'
- / + u 1_ h -
u I~
a( --h)
+ -a f~ v dz
ax -h ax ax aY-h
a1'/ a( -h)
- vl~ - + Vl-h -a- + wl~ - wl- h = O. (8.29)
ay y
If the kinematic conditions (8.7) and (8.11) at the surface and bottom, re-
spectively, are used, the following integrated continuity equations can be
290
obtained:
0rt 0 0
-+-VH+-VH=O (8.30)
ot ox oy ,
where the averaged velocities are defined as:
1 f~
V = H _h U dz,
and H = h + rt. The following assumptions are introduced for momentum

equations.
(i) The hydrostatic approximation is imposed in the vertical direction

because the horizontal scale L ~ H:
op
oz = -pg.
(ii) The Boussinesq approximation is introduced, whereby the density in

all terms is replaced by the constant mean density Po.
(iii) Horizontal exchange of momentum is neglected, on the assumption
that it is much less significant than other processes such as bottom and
surface friction, etc. t
(iv) The distribution of horizontal velocities is almost uniform along the
vertical.
With these assumptions, the momentum equations are integrated from z = - h

to z = rt, resulting in
oV
---;- + oV
V~ + oV
V --;- - fV = -
0 (
~ grt + -Pa) + -H
1 (rsx - 'bx), (8.31 )
ut uX uy uX Po Po
-oV + V oV
- + oV
V- + fV = - -
0 (
grt + -Pa) + - 1 ('s - 'b ). (8.32)
ot ox oy oy Po PoH Y Y
The bottom shear stresses depend on the character of the flow and on the
vertical velocity distribution, particularly in the zone near the bottom, which is in
turn related to the horizontal pressure gradients. As an example, for steady
wind-driven flows in lakes with well developed bottom return currents, vertically
averaged velocities V and V can be negligible in comparison with the actual
velocities. Therefore, the evaluation of bottom shear stresses is not easy in
the vertically integrated model. Usually, these stresses are parameterized
t Consideration of the problem taking into account horizontal momentum exchange can be found.
e.g.. in Simons (1980).
291
through U and V, which can be more or less substantiated only for the so-called
translation wave motion (in which the particle velocity is the same, or nearly
so, at all points along the vertical):
2 2
'bx = c r Po(U + V )1/2U, , by = cr Pa(U 2 + V 2)1/2V, (8.33)
where Cr is a resistance coefficient. Several empirical expressions, the Darcy-
Weisbach, Chezy, and Manning formulas, were originally derived from measure-
ments of steady flow in channels or pipes:
Cr = fo/8 Darcy-Weisbach
Cr = g/C 2 Chezy (8.34)
2 1 3 1 6
Cr = n g/H / Manning (C = H / /n).
C is a Chezy coefficient and fa (or A) is a friction coefficient in the Darcy-

Weisbach formula for head losses in pipes. Values of Manning's n are known for
fully developed, rough turbulent flow, which normally prevails in lakes.
The nonlinear bottom friction can produce strong eflects at large velocities
and small depths, since it has a quadratic dependence on velocity and an inverse
quadratic dependence on depth.
There are many computational methods for solving the vertically integrated
momentum and continuity equations numerically with appropriate initial and
boundary conditions. These methods differ according to: (a) whether they solve
governing equations directly for the "primitive variables" U, V, and pt or
instead solve for the vorticity and stream function; (b) the type of time integra-
tion scheme, such as implicit or explicit; and (c) the kind of Spatial differencing,
such as finite-difference or finite-element methods. An effective numerical
method to solve the problems being considered is a splitting method. Different
aspects of the usage of this method for problems of atmospheric and ocean
dynamics have been considered by Marchuk (1974, 1975) and Marchuk and
Zalesny (1974). A complete review of computational techniques applicable to
the equations describing the circulation in water basins is far beyond the scope
of this chapter. For the interested reader there are reviews by Lick (1976),
Cheng et al. (1976), Siindermann (1979), and Simons (1980).
Types of Variable
Most circulation models use primitive variables U, V, and p. Primitive-variable
models tend to be more accurate for similar computational labor probably
because the vorticity-stream-function approach requires finite-difference
approximations to more critical derivatives than does the primitive-variable
approach.
t The pressure p is represented in the momentum equations 8.31 and 8.32 by the surface elevation Yf.
292
Time-Differencing Procedures
In the numerical calculation it is desirable to use space and time steps that are as
large as possible and yet consistent with physical meaning. However, other
restrictions on the allowable steps are related to the numerical methods used
and to their accuracy and stability. The basic equations 8.31 and 8.32 describe
the propagation of long waves in a shallow water body, and the effects of
various time integration schemes on the accuracy of solution are important for
numerical stability.
Detailed discussions of computational (or numerical) stability in connection
with time-variable circulation problems have been presented by Fischer
(1959, 1965), Platzman (1963), Kasahara (1965), Gates (1968), and Simons
(1973). For computing unsteady flows in shallow water bodies, one of the most
simple and familiar methods is based on the scheme of centered time differences.
For centered differences in the space and time domains, the stability criterion of
Courant-Friedrichs and Levy is imposed:
Ax
At < (2gH)1/2 . (8.35)
This condition indicates that the time step is related to the time for a surface
gravity wave (of speed c = (gH)1/2) to travel a distance Ax.
The nonlinear acceleration terms and Coriolis term may cause some numerical
instability; Lilly (1961), Simons (1973), and others have discussed improved
methods for the evaluation of these terms.
Space-Differencing Procedures
It is impossible to discuss space-resolving techniques completely independently
from the time-resolving procedures described above. Usually, each of them is
designed in close relation to a particular technique for another dimension.
Thus, the scheme of centered time differences is commonly used together with
the staggered grid in space, in which U, V, and p are computed at different grid
points.
One advantage of the staggered grid is that it allows the use of central
differences in space and time, which is desirable for accuracy and numerical
stability while keeping the number of variables low and the variables themselves
partly uncoupled. However, there are problems in representing the physical
boundaries properly and special treatment is necessary to avoid errors and
instability. These problems have remained in later modeling efforts.
For marine applications (circulation in shallow seas) Hansen (1956) solved
the vertically integrated equations by the finite-difference method, using the
variables '7, U, and V on a staggered grid in space and time.
293
Leendertse (1967) discussed the numerical stability and accuracy of the
alternative-direction implicit (ADI) method in its application to the single-layer
circulation model. The treatment of the nonlinear terms causes severe problems
and time-centered differences cannot be used for the convective terms in the
momentum equation.
Heaps (1969) used a staggered grid in space with velocities at the same points
for the linearized and vertically integrated dynamic equations. Care was taken
to center the differences in space and an explicit time integration scheme was
used.
For the Great Lakes, Simons (1971) developed a finite-difference model
based on the vertically integrated equations, including horizontal eddy viscosity,
using space- and time-staggered grids to avoid problems with the convective
terms. Several variations on the treatment of bottom friction and convective
terms were tried.
Finite-difference methods normally make use of orthogonal grids. These
may cause some difficulties in the representation of the geometry of natural
water bodies. To suit complex boundaries better, other types of mesh are applied
that can be used even with finite-difference schemes (Bauer, 1979). An unsteady,
two-dimensional, depth-averaged model was developed for long, shallow
reservoirs by Tatom and Waldrop (1978). The model uses orthogonal curvilinear
coordinates and considers reservoir inflows and outflows, cooling water
circulation, and surface wind stress. Irregular meshes, e.g. of triangular shape,
are more often used with another effective numerical method, the finite-
element method.
Gallagher et al. (1973) analyzed steady wind-driven circulation for shallow
lakes using the rigid lid approximation (eqn. 8.14). Full advantage of the freedom
of varying the grid was not taken in the examples given, but the possibility of
using existing general-purpose finite-element programs was emphasized.
Taylor and Davis (1972), Cheng (1972), Grotkop (1973), Norton et al. (1973),
and King et al. (1973) developed finite-element models for application in a
number of different cases. Wang and Connor (1975) devised a finite-element
model for near-coastal circulation and discussed the fundamental transforma-
tion of the original vertically integrated equations to the so-called weighted
residual form. This method was applied to simulate wind- and tidally driven
circulation in Massachusetts Bay.
Cheng (1972) developed a finite-element method for wind circulation in
Lake Erie. Linearized momentum equations, which were obtained by neglecting
inertia terms and introducing linear bottom shear, were written in the form of a
stream function. The finite-element grid consisted of 516 three-node triangular
elements and 308 nodal points. This gave a reasonably accurate spatial resolu-
tion in the domain of Lake Erie. Figure 8.2 shows the finite-element grid and the
mean circulation streamlines driven by a linear wind distribution (along the lake
surface).
294
CANADA
Buffalo
Region of solution n ....,~,', ,,17170
\
/.'[/V VVIVIVIV,VIV.-V
r-.. (i',/ I"V IVV IVIV 'IV
'// VI/1/ r/I/ /1/ I/V
A typical element / VV VV / / I/IX1/1/1/1/ IVI/ VIV 1/1/
/ --- VV VV / / IVIV I/IV IVI/iVV VA '/
~t!'\)( /V VV VV IVIV I/IV IVIVIVIVY I
Y Detroit New York
! r?IVVIV V VV VV VV IVIV 1/1/VILh- I
.,..., / /I/VVI/ / / / / / V VI/ LV V I
1/ 1// / ~ Boundary
l/V VY:: l%l%t% 1/1// / / / / /,.,.;; ......
11VV ./ v: /1/// V V / I
X
""W..Iv;, /I?; 1/1/1/ / /I~ I
v: //IV 1/ .6 Pennsylvania
~'I;:~~ /1/ I
I..-''';:~
" Ohio
Xl
FIGURE 8.2 Finite-element grid and two-celled mean circulation driven by a quadratic
wind stress distribution: application to Lake Erie, l/J is the stream function.
8.3.2. Two- and Multi-Layer Models

In this type of model, the water body is represented by layers in each of which the
density is assumed to be constant. The thickness of each layer is variable and the
layers move in response to free-surface and internal waves. This type of model,
discussed by Simons (1973, 1980), Wang and Connor (1975), and Cheng et al.
(1976), is useful for strong stratification with little interfacial mixing. There are
relatively few applications of this type of model to lakes or reservoirs. A four-
layer model was used by McNider and O'Brien (1973); however, the most
common version of moving-interface models is a two-layer scheme representing
the epilimnion and hypolimnion.
The vertically integrated equations for continuity and momentum are derived
by integrating three-dimensional equations over each constant-density layer.
295
Momentum is transmitted between layers by interfacial stresses and by mass
transfer (entrainment) through the interfaces. Wang and Connor (1975)
treated the mass transfer across the interface in the continuity equation as
follows:
(8.36)
where
is the thickness of layer k,
f
~k
1 f~k
Uk = H Uk dz, qk = S dz,
k ~k - I '1k- I
and S is the source (or sink) strength; Wk' Wk _ 1 are the relative normal velocities
at the interface between layers k and k - 1, that is, W k is a net entrainment or
mixing velocity between these layers. Vertically integrated momentum equations
for each layer are obtained similarly to those in the single-layer formulation,
with hydrostatic and Boussinesq approximations.
Bottom shear stress and surface wind stress are determined with the same
relationships that are used in single-layer formulations. The idealization of
constant-density layers cannot represent explicitly the mass and momentum
exchanges in the transition region between layers. Wang and Connor (1975)
included a shear stress and a velocity of entrainment at the interface to simulate
these processes; therefore, their functional dependence on the mean flow
variables must be specified.
Interfacial shear stress is related to the square ofthe velocity differences of the
layers:
(8.37)
where C is an interfacial shear stress coefficient that depends on the Reynolds

and densimetric Froude numbers. Derivations of governing equations for
multilayer models are also discussed by Simons (1973). Wang and Connor
(1975) have applied this model to a two-layer flow that was investigated ex-
perimentally by Hyden (1974). The multilayer models present difficulties in
practical application when upwelling or downwelling effects are significant;
use of this type of model is limited to strongly stratified conditions with little
interfacial mixing.
296
8.3.3. Ekman-Type Models
The approach initiated by Welander (1957) is based on the earlier work of
Ekman, who investigated the rotational effects of the earth on oceanic circula-
tion. The approach is specifically designed for wind-driven currents. If the Rossby
number, Ro = (inertial terms)/(rotational terms) = ulfL, is small and the
hydrostatic approximation is valid, the nonlinear convective terms can be
neglected. It is also assumed that the role of horizontal turbulent exchange is
small in comparison with the vertical exchange; that is, horizontal shearing
motions (such as in zones near the shore) are not considered. The governing
equations are, therefore:
ou 1 op 02 U
at - fv = - pox + Ez OZ2
(8.38)
ov 1 op 02 1'
ot + fu = - poy + Ez OZ2 '
(8.39)
OU ov ow
ox + oy + oz = o. (8.40)
This model can be categorized as an Ekman-type model. The class of models

allows one to solve the governing equations through a combination of analytical
and numerical solutions. At least for the steady state case, the vertical velocity
distribution is determined analytically in terms of the prescribed wind stresses
and the unknown pressure gradients. The pressure gradients must be obtained
numerically through the depth-integrated equations.
Liggett and Hadjitheodorou (1969) applied the Ekman-type model to
simulate a steady state wind-driven circulation in a rectangular lake with variable
depth. Applications of the Ekman-type models have been demonstrated by
Gedney and Lick (1970, 1972) and by Bonham-Carter and Thomas (1973) in
their studies of wind-driven circulations in Lake Erie and Lake Ontario.
Liggett and Hadjitheodorou (1969) considered steady state on the basis of
the following assumptions. (i) The vertical eddy viscosity is constant through-
out. (ii) The no-slip condition is imposed at the bottom boundary:
u = v = W = 0 at z = -hex, y). (8.41 )
(iii) The rigid lid approximation is applied at the surface and the wind stress is
prescribed as follows:
(8.42)
From the momentum equations with boundary conditions, the general solution
for velocity is obtained analytically as a function of the imposed wind stress,
297
the bottom topography, and the pressure distribution (which is unknown):
. sinh A(h + z)Q . (COSh AZ 1)

W =u+w= -I - P (8.43)
A cosh Ah cosh Ah '
where
ap . ap
P=-+I-
ax ay'
and Ev = tzllh l is the vertical Ekman number. If the general solution is inte-
grated vertically and a stream function is introduced that satisfies the vertically
integrated continuity equation, a Poisson-type equation for a stream function is
derived:
v l '¥ = a a'll + b a'll + c, (8.44)

ax ay
where a, b, and c are functions of the wind stress, Ekman number, and local
depth. This equation, with proper boundary conditions, is solved using finite-
difference methods. The numerical results for the stream function give the
pressure distribution to complete the general solution for the velocity, eqn. 8.43.
The steady state model was applied to an artificial basin of rectangular
planform, shown in Figure 8.3. A vector diagram of velocities in the test lake
at the surface and at two depths is also shown for when a south wind is blowing.
The highest velocities on the surface occur near the boundaries and are in the
downwind direction, and the return flow occurs in the deeper, central portion of
the lake in the upwind direction. Rao and Murty (1970) presented a similar
Ekman-type model and applied it to Lake Ontario.
Gallagher et al. (1973) applied a finite-element method to the steady state
Ekman-type model described by eqns. 8.43 and 8.44. The same method was
applied for an unsteady state in shallow, homogeneous lakes by Liggett (1969).
Lee and Liggett (1970; Liggett and Lee, 1971) also applied the Ekman-type
model for two-layer stratified flow conditions. The solutions obtained from the
Ekman-type model are applied to each layer separately. Additional boundary
conditions between the layers, consisting of equating velocities at the interface,
and continuity of pressure and shears across the interface, are imposed as follows:
(8.45)
PI = Pl' (8.46)
where the subscripts I and 2 refer to the upper and lower layers, respectively.
These boundary conditions, together with a kinematic condition, lead to the
interface specifications.
This type of model has certain limitations in its application to real
water bodies: (i) simplifications of governing equations are necessary for
298
~cross Cross ~ /
section section
A--A B-B
, - - - - - - - " .
~~:;;~-\-~, B B
· · . · .. . .
l>'I\\, 1 1
f//".r;'11
L·.I 1 ··
\
• f
.... \ . · 11
.. , .
.......
·
..-
\
... rI
• I 1'/ •
~ .-' ..--
.. · ··
4- ~ ~ rI •
11,/·,1,11 1
!. 1 • I t 1 ~ l .. •
·,, ·. ····,,
N 1 " I
I j J •
f/,/~/h' 1 \ 1 I j , •
1/,/~/l.ll 1 \ 1
r ·· ··, ·····,,
I ' J •
T'I/~/~1 I \ 1
f !'~/%ll
%/~/ll
I
I
\
\
I
1
I 1' :
! : .
T!/~/tl1 1 /
,· ·
\ 1 j J
Ih/~/I,TI 1
, ·•· .\ 1
, ·· ·
\ I •
I 1, I . 1 7, 1 1 I
·
/ I , •
17,/~//,lj
, · .·
\ I I , .
I /
,
A
~h"~/tl
7,/~//,1
I
A
1
I
· ·,,
\
\
I
I
I , .
l , •
, ·· ·
\
tl'l,'~'hll I t t I lJ •
II/"~/tll
, · ··,
d/,'~/hl1
1
I
t \
\, 1
I
j,. J •
, • ·· 1
\
l'l,'~'hl1 I j , •
1 f ,'~/l,ll
'1,' ~/hl
; '7.'~/1t11
I
I , \\
, \ ·· •t
. , 1I
1
j J
j J •
, '/"~/7,1
I
1 , \
t , \ · ··· , II
, J
j ,
L.L..LJ
,/"
, '7,1
'1./~//,tl
1
.• //,111
1 ,, • \
· · ,
, 1t
I J
j ,
·, ··
0 15 \
cm S-1 '!./~/!.ti
'7.'~/!.11
t
1 \ ·
, t
\ •.. , 1
, I
J ,
''I,/~/!.ll
I
t
t
, · \ I , I , •
I , •
'7,/~/11
1/rl/~/
1 .... /1[1
1
t
1
1
t
\ " t
\ ..
\ .
...· .· ,
!
I
I
j I
1
11\ \ 1 ' 1 / 1
1\\\\'/1/
II/Ill I
'-~
·. ·, ·..." ·.
t \
•
.,- r'
,
~ I
,/
j I •
Ii
I •
•
lit. lit. lit. lit. , ., jl

• ~o'
· ···
Surface Depth: 0.2 Depth: 0.4
FIGU RE 8.3 Horizontal velocities in the test lake under a south wind: at the surface and
at 0.2 and 0.4 of the total depth (source: Liggett and Hadjitheodorou, 1969),
analytical solutions, such as neglecting the convective terms and horizontal

diffusion terms; (ii) the assumption of constant vertical eddy viscosity may not
be realistic for stratified flows; and (iii) the analytical solutions are quite complex
for a multilayer system.
8.3.4. Laterally Averaged Two-Dimensional Flow Models

If the lateral dimension of the water body is sufficiently small compared with
the longitudinal dimension, the system can be approximated as two-dimensional
(in the longitudinal-vertical plane). This type of model may be applicable in
299
reservoirs with strong through flows, but is less important for natural lakes
(another field of application is in estuarine hydrodynamics).
Withdrawal patterns, i.e. withdrawal "layers," are primarily affected by the
degree of stratification in the immediate vicinity of the outlet. This selective
withdrawal influences the distributions of flow and temperature within the
entire impoundment. The resulting patterns of flow and temperature are
essentially interdependent and they, in turn, govern the distributions in space
and time of other water quality characteristics of the impoundment.
A notable example is Lake Roosevelt behind Grand Coulee Dam in
Washington. In this reservoir the period of maximum runoff corresponds
roughly with the period of greatest insolation. The result is a pronounced
longitudinal temperature gradient as well as vertical stratification. So-called
"tilted isotherms," planes of equal temperature inclined downstream, are
evident.
The two-dimensional formulation of inhomogeneous fluid motion, including
vertical and longitudinal distributions of density and velocity, can be obtained
by averaging the momentum and heat balance equations in the lateral direction
(Vasiliev et al., 1973, 1974; VasiIiev, 1978a,b). After averaging, the governing
equations are as follows (the hydrostatic approximation is used for the vertical
momentum equation):
au au
-+u-+w-= - g a
- (r,
au + f~ -
P-- d
Po z )
at ax oz AX z Po
a(b(v + f ) -a;;au).
+ -b1 -oz Z
-
u r
-
b'
(8.47)
~(bu) o(bw) _ 0
ax + oz - , (8.48)
(8.49)
u-I
_ 2 [
1+ (Ob
- i )2 + (Ob
- i )2]1/2 , (8.50)
;=1 ax oz
r = (~) lu lu,
where
PoCo is the product of average density and specific heat,
b = b l + b 2 is the channel width,
b I (x, z), bix, z) characterize the shape of the lateral surface of the channel,
r, is the free-surface elevation,
v is the coefficient of molecular viscosity
fz is the coefficient of turbulent viscosity,
300
X is the coefficient of molecular heat diffusivity,
K _ is the coefficient of turbulent heat diffusivity,
T is the frictional resistance of the lateral surface,
,\, is the frictional resistance coefficient,
qn is the heat flux through the lateral surface (per unit surface
area),
¢ is the heat source or the volumetric heat influx (section 8.2.2).
King et al. (1973) applied a finite-element method to two-dimensional

stratified flow. The vertically two-dimensional model was applied to a laboratory
experiment at the Waterways Experiment Station. Numerical results were
compared with measurements from the vicinity of a broad-crested weir. This
model was also applied to Lower Granite Reservoir by Norton et al. (1973).
A similar model was developed by Edinger and Buchak (1978) and applied
to Sutton Reservoir, West Virginia. Results of the test simulations indicated
four possible circulation regimes. These depend on inflow and outflow rates,
inflow density, and surface heat exchange.
8.3.5. Three-Dimensional Models

The most advanced models can be found in this group, in which a three-
dimensional discretization of the water body is constructed by using a spatial
Eulerian grid and the unknown variables are computed at each node for succes-
sive time steps. The development of high-speed computers has facilitated
remarkable progress in this type of modeling (Simons, 1972, 1973, 1974, 1975,
1980; Bennett, 1974; Vasiliev and Kvon, 1977; Wittmiss, 1979; Tjomsland,
1979; Bauer, 1979; Raney et al., 1979; Tsvetova, 1979).
The governing equations for three-dimensional models of a thermally
stratified turbulent flow were given in section 8.2.2. The equations can be solved
by the straightforward application of conventional numerical techniques,
whereby the continuous derivatives are replaced by finite differences.
A number of models of lake circulation have been developed by Simons, as
previously noted. Adapting the concepts of modeling utilized in the three-
dimensional representation of atmospheric circulation, Simons has modeled
several of the Great Lakes, as well as Lake Vanern in Sweden. Calibration and
verification tests using field data have shown that these models simulate
reasonably well the patterns of circulation induced by wind and affected by
temperature changes and Coriolis forces. The approach employed is illustrated
by Simons' model VANERN (Simons et al., 1977), which was an adaptation of
his earlier multilevel model for the Great Lakes.
A three-dimensional finite-difference model can be visualized as a sequence
of fixed but permeable levels. Simons (1980) discussed the advantages of this
interpretation of a finite-difference model as follows: (i) the general circulation
problem can be reduced to a quasi-two-dimensional problem; (ii) the model
301
equations are obtained by vertical integration over each layer instead of applying
the equations at given levels, and this procedure ensures that the conservative
character of the original differential equations and the relevant volume integrals
is preserved.
The equations of motion, heat conservation, and continuity developed by
Simons are:
Motion
ou + V· (Vu
-;- - €h Vu) + -0 ( wu - ou)
€z - = 0 (P
fv - ;- + Q) (8.51)
ut oz oz uX
ov 0 ( ov) o + Q)
ot + v· (Vv - €h Vv) + oz wv - €z oz = -fu - - (P
oy
(8.52)
Heal conservation
(8.53)
Continuity
OW ou ov
---- (8.54)
oz ox oy'
where
u, v, ware velocity components along the x, y, z axes,

V is the horizontal component of the velocity vector,
T is the temperature,
€h' €z are the horizontal and vertical eddy viscosities,
K h , K z are the horizontal and vertical heat diffusivities,
V is the horizontal gradient operator,
f is the Coriolis parameter.
Motion is induced by Coriolis forces fv and fu and by barotropic and
baroclinic pressure components P and Q:
P = 91'/ + Pa (8.55)
Po
r~ tlp
Q = 9 ,_ -dz, (8.56)
·z Po
where
'1 is the free-surface elevation,

9 is the gravitational acceleration,
302
Pa is the atmospheric pressure at the air-water interface,
Po is the mean water density,
tip is the density difference.
Vertical velocities are computed from the horizontal flow divergence through
application of the continuity equation, with the lower boundary condition that
there can be no flow normal to the bottom. Integration of the same equation
along the vertical axis results in a determination of the free-surface elevation,
which, in turn, determines the barotropic pressure function (8.55). Surface wind
stresses, bottom friction, and heat fluxes at the surface and bottom comprise
boundary conditions for fluxes of momentum and heat. At lateral boundaries,
where normal components of velocity must vanish, tangential velocities are
stipulated as required by "slip" or "no slip" assumptions that fit a particular
model.
Numerical solutions of the equations presented above are obtained by
finite differencing on a staggered grid. The vertical structure of the model,
which fixes the computational scheme, comprises a series of arbitrarily fixed
levels at which vertical velocities, stresses, and vertical fluxes are calculated.
Temperatures and currents are defined as averages for the intermediate layers.
The effects of free-surface oscillations on the computation of internal flows are
minimized by reducing the layered system of equations to a single equation for
vertically integrated flow and a set of equations for the shears between adjacent
moving layers. Integration over time is essentially explicit, except that the
Coriolis term is treated implicitly. Simons (1973) provided other details of the
solution technique and numerical approximations. He applied the above-
mentioned multilevel model to three cases.
(i) The model was applied by Simons (1974) to Lake Ontario in spring
and early summer. The predicted water levels, currents, and tempera-
tures were compared with observations made shortly after the passage
of tropical storm Agnes during the latter part of June 1972. Figure 8.4
indicates that the currents follow the wind on the shallow, northern
side of the lake, whereas the resulting pressure gradient returns the
water masses in the deeper parts of the lake. Figure 8.5 compares
observed and predicted water levels during the storm. The model
simulation of the observed circulation pattern is substantially correct.
For the period of weak stratification the model is found to be most
sensitive to parameters related to the vertical flux of momentum.
However, satisfactory simulations of observed water levels and
currents require a wind stress coefficient considerably larger than those
obtained from direct measurements.
303
Cobourg
N
(a)
L------.J
(b) 5cm s-l
FIGURE 8.4 (a) Observed and (b) computed currents in Lake Ontario averaged over the
three-day period, 22-24 June 1972. The contours are in meters. Numbers I to 4 indicate
depths of 10,15,30, and 50 m, respectively (from Simons, 1974).
(ii) The model was applied to the strong stratification that prevailed in
Lake Ontario during and after a storm on 9 August 1972 (Simons,
1975). Good agreement between predicted and observed data was
obtained for water levels, as shown in Figure 8.6. The model results
shown in Figure 8.7 neglect heat diffusion and include only heat
transport by advection, There is good agreement between model and
observations with regard to the wave-like pattern of temperature in
the deeper parts of the lake,
(iii) The model was applied by Simons et al. (1977) to Lake Vanern in
Sweden. Verification checks were made for two distinct conditions,
one when the lake was essentially homogeneous and another when the
lake was thermally stratified. Sensitivity tests, involving alternative
FIGURE 8.5 Observed (full curves) and computed (broken curves) water levels at four
stations on the shore of Lake Ontario during storm Agnes in June 1972 (from Simons, 1974).
All time series were filtered to remove periodic components having periods of less than 5 h.
levels of spatial resolution, two different computational grids, and

various values of empirical coefficients, were performed to determine
the best model structure. Results of the study demonstrated the capa-
bilities of the model to describe correctly the major characteristics of
circulation and water level fluctuation induced by wind. However, the
tests indicated that attempts to improve the results of verification by
increasing the resolution (using a smaller grid size, more layers,
shorter time steps, a refined computational scheme) tended toward
a level of diminishing returns against the added time and cost of com-
putation. The most sensitive parameters were the empirical mixing
coefficients and the eddy viscosities. These, in turn, through their
influence on current structure, controlled the ability of the model to
simulate temperature changes. Heat diffusion, per se, was of little
consequence. A general conclusion derived by Simons and his
colleagues from these experiments was that "very expensive, non-
linear, high-resolution models" are not justified for dealing with
slowly varying lake-wide circulations.
305
50
30
10
-10
-30
-50
30
10
E -10
E -30
Qi -50
> 30
~
10
...coQ;
~
6 14
August
FIGURE 8.6 Observed (full curves) and computed (broken curves) water levels at four
stations on the shore of Lake Ontario during and after a storm on 9 August 1972 (from
Simons. 1975).
A similar model was also developed by Bennett (1974). This model employed
a rigid lid approximation to eliminate the short-term dynamic response of the
water surface and used a stream function to ensure continuity. As a result, this
model allows a computational time step of the order of an hour, whereas the
Simons model required a time step of a few minutes. However, this advantage
is offset by the requirement of iterative procedures to solve the pressure equa-
tion. Thus, instead of more time steps, one needs more iterations and so requires
about the same computer time.
Most circulation models have used constant eddy diffusivities for computation
of the momentum equations. Vasiliev and Kvon (1977), Vasiliev (1978a,b), and
Kvon (1979a,b) developed more comprehensive circulation models, including
a one- or two-equation turbulence closure model with two parameters.
Using the Boussinesq approximation and the hydrostatic condition and
v.>
o
0\
(a)
(b)
FIGURE 8.7 (a) Observed and (b) computed temperature changes CC) in Lake Ontario during 2-5 August
1972, corresponding to the second layer of the model (between 10 and 20 m) (from Simons. 1975).
307
eliminating horizontal turbulent exchange, they obtained the following govern-
ing equations for a temperature-stratified turbulent flow:
au
- +u- +
au au au fv = - 9 -a [1] + IZO (p--=-----.2.
p) dz] + -a ((; -au) ,
at ax v-
ay + w az
- -
ax Z Po az az
(8.57)
av av
- + u- + v - +
av av - 9 -a [ 1] + IZO (p-- -
po) dz] + -a ((; -av) ,
at ax ay w-
az + fu =
ay Z Po az az
(8.58)
au av aw
ax + ay + az = 0, (8.59)
aT + uaT + vaT + w aT = ~ (K aT), (8.60)

at ax ay az az az T
P - Po
-f3(T - To)·
Po
On the lateral boundaries, either the velocity component normal to the boundary
is zero, or it must be specified. At the free surface, z = zoo
az au t
at = w, [-=-
az Po'
(8.61)
and at the bottom, z = h(x, y):
az o az o aT
w =u ax + v ay' az = 0, (8.62)
where u = (u, v), t = ('x, 'y), K l is the coefficient of the bottom tangential
stress, TE is the equilibrium temperature, and Zo is the average free-surface
elevation. Vasiliev et al. (1973) determined the turbulent exchange coefficients,
(; and K T' from the turbulent kinetic energy e and a length scale of turbulence,
L, as follows:
where k e is the coefficient of total exchange for turbulent energy transfer and c
is a constant. The Richardson number Ri and the ratio aT of the turbulent
308
exchange coefficients for heat and momentum transfer are given by:
(g/Po)(Jp,'oz
Ri= (8.64)
(OU/OZ)2 + (OV/OZ)2
(1 + 10 Ri)1/2
(8.65)
aT = a o [1 + (10/3) RiJ3/2'
where ao is a constant. The model was solved numerically by means of an implicit
finite-difference scheme.
Later, Kvon applied the two-equation turbulence model founded on two
transport equations for the turbulent energy and the turbulent energy dissipa-
tion (Jones and Launder, 1973; Launder, 1976). On this basis he developed the
flow model with a slip condition at the bottom. The model was used by Kvon
(1979) in the consideration of a three-dimensional temperature-stratified flow
in a water body.
8.3.6. Concluding Comments

A well developed capability exists for simulating vertically averaged two-
dimensional circulation and water level fluctuations in shallow, well mixed
lakes and impoundments. Single-layer circulation models solved by finite-
difference methods have been successfully applied to shallow lakes, such as
Lake Erie, producing sufficient information in the form of depth-averaged
velocities to drive compatible water quality models. Added flexibility in pro-
viding local detail and in fitting irregular boundaries and topography is avail-
able in the form of finite-element and finite-difference models, both of which have
been demonstrated in a number of practical cases. Documentation of vertically
averaged two-dimensional circulation models is generally adequate for trans-
ferring this technology to new users. The state of the art of modeling circulation
in the horizontal two-dimensional formulation is fairly well advanced.
Another type of simulation technique, which is also of two-dimensional
character, has been developed relatively recently for deep, oblong reservoirs
and lakes when density stratification takes place. Though even vertical one-
dimensional models can be helpful in many of these cases, the vertical two-
dimensional models allow much better resolution of the problem and can
include the simulation of withdrawal patterns, "tilted isotherms," flow patterns
in the upstream reach of a reservoir, etc. However, experience with the applica-
tion of these models is less than for the horizontal two-dimensional models.
The state of the art of mathematical modeling of circulation in large lakes is
exemplified by the three-dimensional models of Simons. This type of model can
be visualized by a sequence of fixed but permeable levels, except for the free
surface and the bottom. Therefore, a generalized system of model equations is
reduced to a quasi-two-dimensional form. Such models have been applied to
many lakes and, therefore, are well documented for other users.
309
Conceptually, all of the most important physical mechanisms that govern
water movement are included in the models, although it is generally recognized
that model performance is most sensitive to empirically determined coefficients
of mixing. A more comprehensive approach to determine the coefficient of
turbulent exchange, as discussed by Vasiliev (1978) and by Kvon (l979a,b), is
desirable. Modeling efforts are likely to be extended in the direction of better
characterization of mechanisms of turbulent transport along both the vertical
and horizontal axes.
8.4. TWO- AND THREE-DIMENSIONAL WATER QUALITY

AND ECOLOGICAL MODELS
In the one-dimensional models for stratified impoundments, hydrodynamic

considerations are much simplified. A continuity equation for the vertical
direction is used to determine the advection caused by inflows and outflows
at different elevations. The vertical transport of heat is governed by advection
and diffusion and by internal heating due to shortwave solar radiation. For
deep impoundments that are dominated by inflows and outflows, as opposed to
diffusion and internal heating, these models are generally reliable (Orlob and
Selna, 1967; Huber and Harleman, 1968; Ryan and Harleman, 1971; Hurley-
Octavio et aI., 1977).
Simulation of the annual temperature cycle in such water bodies has been
accomplished with reasonably good agreement between mathematical model
and impoundment. However, extension of the advection-diffusion approach
to include other nonconservative parameters (for example, dissolved oxygen,
BOD, nutrients) that are not distributed in the impoundment by the same
mechanisms as heat energy raises questions as to the appropriateness of the one-
dimensional approximation (Markofsky and Harleman, 1971, 1973; Chen and
Orlob, 1972). This concern is further accentuated for long, narrow or broad,
shallow impoundments, in which temperature or concentration gradients may
develop either longitudinally or laterally. Various two- and three-dimensional
water-quality-ecological models have been developed to take into account
horizontal or vertical resolution. There are three broad categories in two-
and three-dimensional water quality modeling: (a) compartment models,
(b) a network of one-dimensional channel models, and (c) two- and three-
dimensional models. In the following discussion the emphasis will be on the
coupling of the hydrothermal and water quality components; however, the
details of the water quality and ecological components will be omitted since
they are discussed in Chapters 3, 4, and 9.
8.4.1. Compartment Models

By ignoring hydrodynamics completely and concentrating on the kinetics of
water quality and biota, modelers have been able to construct models that
310
describe gross changes in the mass balance of well mixed water bodies. The
impoundment is treated as an interconnected system of continuously stirred
tank reactors (CSTR) to give added spatial dimensions and to account in part
for temporal changes. This approach has been adopted in extending laboratory
idealizations of chemical and biological kinetics to two-dimensional systems
in which net transport between reaction cells is derived either by independent
simulation of hydrodynamics or by field measurements.
Snodgrass and O'Melia (1975) developed a model that mass-balances both
particulate and ortho-phosphorus for a two-compartment lake under two sets of
seasonal conditions. The lake is treated as a single, wel1 mixed compartment in
the winter and as two compartments (epilimnion and hypolimnion) in the
summer. The model is formulated in mass balance form as a CSTR for epi-
limnetic and hypolimnetic particulate phosphorus. The CSTR equations
contain vertical transport coefficients for the exchange of mass across the
thermocline.
The Snodgrass-O'Melia model has been applied by the developers to
predict average phosphorus concentrations in lakes with a wide range of
detention times. Comparisons of predicted and observed values show "excellent"
agreement (O'Melia, 1974). The model is recommended as a tool for prediction
of permissible phosphorus loadings in lakes. A similar model for a two-layer
lake was developed by Imboden (1974).
Larsen and Mercier (1975) applied the Snodgrass-O'Melia model to Lake
Shagawa, Minnesota, observing that the model underestimated the amount of
epilimnetic phosphorus. They developed and applied a three-compartment
epilimnetic model, a simplified version of models developed by Baca et al. (1974)
and Thomann et al. (1975). The distinctive features of this model were the
inclusion of algae as a sink for soluble reactive phosphorus and the conversion
of particulate phosphorus to the soluble form so that it would be available for
algal growth.
Detailed models of the eutrophication process have been proposed and
several have been implemented, particularly for Lakes Erie and Ontario. A
noteworthy example is the development of phytoplankton models for Lake
Ontario (Thomann et al., 1975). Three basic models, each with a different level
of detail, were developed and tested. The model Lake 1 simulated the impound-
ment as a three-layer system~epilimnion, hypolimnion, and benthos-and
concentrated on phytoplankton and zooplankton dynamics. Lake 2 provided
additional vertical resolution with seven layers and also considered tempera-
ture, chemistry, and sediment interactions. Lake 3, which was developed only to
the preliminary stage, provided additional spatial resolution with up to seven
layers and 67 segments, and accommodated 10 to 15 variables. A future model
is being planned with up to 5000 compartments (segments· variables).
Each of the models is founded on mass conservation and kinetic principles.
A mass balance equation is written for each constituent and each segment.
311
Solution over a suitable time horizon with appropriate time steps results in the
required space-time description of all variables. The models are driven by
circulations developed from field observations or from models, such as those
of Simons (1973), that have been applied to Lake Ontario.
Verification runs with Lake 1 indicated generally good agreement with
measurements of such constituents as chlorophyll a, zooplankton, carbon, four
forms of nitrogen, and phosphorus. Lake 2 was studied analytically but was not
tested against the lake itself. Lake 3 was tested preliminarily against selected
observations but was not verified.
The plankton model was combined with a three-dimensional circulation
model of Lake Ontario by Simons (1976). The simulations were compared with
three-dimensional data on Lake Ontario collected during 1972. Typical spatial
differences between deep-water and near-shore zones appeared reasonably well
reproduced in the model.
A water quality model to predict phytoplankton production in western Lake
Erie was developed by Oi Toro et al. (1975). The basis of the model is a set of
mass conservation equations that relate the variables to each other. The model
includes effects of biological phenomena (predator-prey relationships),
chemical reactions (nitrification), and other interactions that provide nutrients
necessary for phytoplankton growth. Figure 8.8 indicates the seven spatial
segments used in the model.
83° 30'
42° 00'
/n \
~
\
./
\ -'
~
Pelee
I / Island N
~
/
Kelleys
r
Island
\
41° 30'
82°30'
FIGURE 8.8 Circulation pattern in western Lake Erie, showing prevailing current
directions and spatial segments of the water quality model (Oi Toro et al., 1975).
312
Rates of flow from one compartment to the other were assigned on the basis
of observed and computed flow patterns. The magnitude of mixing of flows
between adjacent segments was calibrated by use of a conservative tracer,
in this case chloride concentration. The model was calibrated by adjusting
parameters that specified the internal kinetics of the seven dependent variables
for the period 1967-1970. Initial values were derived from the literature or
laboratory and "fine-tuned" until agreement between model and lake was
considered acceptable.
8.4.2. Network of One-Dimensional Channel Models

Network models were originally developed for estuarine systems by Chen and
Orlob (1972), Dailey and Harleman (1972), and Najarian and Harleman (1975).
If lakes and reservoirs are connected to each other and they are narrow and
long, a network approximation is advantageous.
In the Chen and Orlob model the water body is subdivided into discrete,
fully mixed volume units, called "nodes." Nodes are characterized by surface
area, depth, volume, and side slopes. All water quality parameters that character-
ize the system are associated with nodes. Nodes are interconnected by channels
or "links," which are defined by length, width, cross-sectional area, hydraulic
radius, and a friction factor. Water is constrained to flow from one node to
another through the defined channels, advecting and diffusing water quality
constituents between nodes.
Mass balance equations are applied to these components, including advection,
diffusion, inflow and outflow, and sink or source terms. Mass balance equations
are coupled with hydrodynamic equations to produce the response of water
quality variables.
Patterson et al. (1975) applied a link-node dynamic water quality model,
similar to one developed by Chen and Orlob (1972), to Green Bay in north-
eastern Lake Michigan. The model represents a shallow lake or estuary (for
which it was originally intended) as a network of one-dimensional channels
(links) and storage elements (nodes). Water movement in the links is usually
simulated by a hydrodynamic model of identical configuration or is derived
from field measurements. In the Green Bay model, current structure was
simulated with a two-dimensional, orthogonal-mesh, finite-difference model
patterned after that of Leendertse (1970). The model simulates temperature,
coliforms, four forms of nitrogen, DO, BOD, phosphorus, two types of algae,
and several conservative constituents.
8.4.3. Two- and Three-Dimensional Models

In lakes or reservoirs, spatial variations of nutrients and plankton or other living
organisms are strongly associated with the development of thermoclines in the
313
vertical direction and horizontal variations between shore zones and deep water.
In such lakes or reservoirs, two- or three-dimensional models are necessary to
simulate the behavior of the whole system. Usually two- or three-dimensional
ecological models are applied and they consist of horizontal or vertical arrange-
ments of volume elements within each of which the state variables change with
time according to a set of biochemical reaction equations. Each volume element
is coupled by water movements that are simulated by two- or three-dimensional
hydrothermal models.
Lam and Simons (1976) devised an advection-diffusion model of Lake Erie
for the transport and dispersion of conservative substances, e.g. mineral salts.
The governing equation for the model is a statement of mass conservation for a
layer bounded by two horizontal planes 1 and 2 (upper and lower layer, re-
spectively) :
oto (hie) = - V' (Vh!e) + V' (KhhSe) + (weh - (we)!
+ (K :~)! - (K :~t + Se'

z z (8.66)
where
e is the concentration of a conservative substance,
h! is the local depth of a model layer,
V is the horizontal component of velocity,
K h' Kz are the horizontal and vertical eddy diffusion coefficients,
w is the vertical water displacement perpendicular to the bounding
surface,
Se represents all external sources.
The model is driven by Simons' multilayer lake circulation model, which
produces the time-varying quantities V and w. In a two-layer model vertical
transport occurs only at the intermediate bounding plane, i.e. at the top of the
bottom layer and at the bottom of the upper layer. For a vertically mixed
°
(homogeneous) system, w = and oeloz = 0, so the right-hand side of (8.66)
reduces to the first two terms and the source term.
Lam and Simons (1976) applied their model to the simulation of chloride
distributions in Lake Erie for conditions corresponding to summer and fall,
1970 and gained generally satisfactory results. Studies were made to determine
the sensitivity of the models to various parameters and parameter levels.
The model developed by Lam and Simons can be applied to simulate non-
conservative water quality parameters, including nutrients. Lam and Halfon
(1978) developed a two-compartment phosphorus model coupled with a hydro-
dynamic model and applied it to Lake Superior. The hydrodynamic model was
developed by Simons (1973) and described in section 8.3.5. A horizontal grid
of size ~x = ~y = 10 km divides the lake so that there are 734 surface points.
314
The vertical structure consists of four layers, separated at 10, 30, and 40 m
below the surface. The ecological model consists of two variables; the trans-
formation from one variable (soluble reactive phosphorus) to the other (particu-
late phosphorus) is controlled by a primary production submodel. The two-
variable model was first calibrated with time-averaged data on the assumption
that the lake was spatially homogeneous. The model was then coupled to the
transport processes computed by the hydrodynamic model.
The primary production submodel is a function of some physical factors,
e.g. water temperature, sunlight, albedo, day length, and water turbidity.
Figures 8.9(a) and (b) show the observed soluble reactive phosphorus distri-
butions in June and September, respectively. The June observations were used
as initial conditions for the model. Figures 8.9(c) and (d) show the predicted
results of soluble reactive phosphorus for September without and with the
transport processes, respectively. The strong interaction of the physical trans-
port processes with the biochemical model is demonstrated by Figure 8.9(d).
The agreement between observed values (Figure 8.9(b» and predicted values
(Figure 8.9(d» is reasonable. Similar models were applied to Lake Ontario by
Simons (1976) and by Simons and Lam (1978). The investigation utilized the
data base accumulated during the 1972 International Field Year on Lake
Ontario. They demonstrated that the effects of water transport processes on
biochemical processes are comparable in magnitude with other physical
processes.
Chen and Smith (1979) developed a three-dimensional ecological-hydro-
dynamic model for Lake Ontario. The model includes mass balance equations
for 15 different classes of biotic and abiotic substances. The biological compart-
ments (algae, zooplankton, and fish) were substantially expanded to give a
(b)
FIGURE 8.9 Maps of soluble phosphorus distribution observed in Lake Superior in

(a) June and (b) September 1970; and maps of the ecological model predictions for Sep-
tember (c) without and (d) with coupling to the transport processes. The contours are
measured in mg m - 3 (from Lam and Halfon, 1978).
315
more complete picture of the ecological behavior of the lake. For example,
the algae compartment was enlarged to include four groups of phytoplankton
and attached algae; four groups of zooplankton were included and four groups
offish in three life stages (adult, young, and eggs or larvae) were represented.
The Chen-Smith model of Lake Ontario was designed to give a three-
dimensional quality characterization of the impoundment by dividing it into
layers and segments, i.e. hydraulic elements, for which separate mass balance
equations can be written. Implementation of the model requires an independent
determination of the flow field in three dimensions and estimates of empirical
diffusion coefficients.
The model was tested against a representation of Lake Ontario that included
41 surface elements and seven vertical layers of varying thickness, a total of 209
elements. Hydraulic inputs were derived from a hydrodynamic model attributed
to Bennett (1974). The hydrodynamic model is of orthogonal form, with 715
surface elements and seven layers. It is wind-driven and considers densimetric
(temperature) effects on circulation. Integration of the hydrodynamic output
to provide flows for the quality model was accomplished by a special interface
program.
The model was simulated for the year 1972, with an hourly time step for
hydrodynamics and a daily time step for water quality and ecology. No rigorous
verification was attempted; however, the correspondence of the modd results
with some limited available water quality data has generally been good.
The 1979 version of the model required about 60 k decimal words of core
storage and 9 s of Univac 1108 CPU time per day of simulation (f1t = 1 day).
By way of contrast, the hydrodynamic program requires 130 k of storage and
120 s of CPU time per day of simulation.

The phytoplankton productivity models of Di Toro et al. (1975) and Thomann
et al. (1975) and the water-quality~ecological models of Chen (1970; Chen and
Orlob, 1972) are fairly representative ofthe state ofthe art ofecological modeling
of impoundments, at least from an engineering viewpoint. However, the very
nature of the modeling exercise that tends to aggregate, average, and smooth
over biological subtleties leaves the more rigorous aquatic biologist somewhat
disconcerted. He would prefer to concentrate efforts on a more correct repre-
sentation of biological interactions, the kinetics of varying life stages, shifts in
grazing preferences, ecological instabilities, and the like. Some trade-offs are
necessary, simply because the model is an approximation of the real system.
These seem to have occurred in the engineer-developed models by gross
simplification of the aquatic ecosystem and in the biologist-developed models
by simplifying the circulatory and exchange processes of the impoundment.
Somewhere between these extremes probably lies the best practical ecological
316
model of a eutrophic lake or reservoir. It seems from our review that the
capability to model these systems has probablY outstripped our understanding
of them. At best, calibrations are rather rough and verifications more so. Part
of the reason is a lack of good data from the field, but there has also been some
over-zealousness on the part of modelers. Sensitivity testing, which has only
been occasionally employed, will no doubt reveal, as it has for models of lake
circulation. that some simplification in ecosystem description is justified. If this
can be accomplished while still satisfying the biologist that his science has not
been unduly compromised, some useful models can probably be produced.
8.5. MATHEMATICAL MODELS OF COOLING IMPOUNDMENTS
Lakes, reservoirs, or shallow artificial impoundments are frequently utilized

to dissipate excess heat from power plants. These heated condenser water
discharges may alter the hydrothermal, chemical, and ecological environment
of the water body and it is necessary to have techniques for the prediction of such
impacts.
Cooling impoundments dissipate waste heat by direct transfer between the
water surface and the atmosphere. A portion of the heat is dissipated by radia-
tion while the remainder is transferred by evaporation and conduction. The
water that is lost in the evaporative process must be replaced by "makeup"
water. This is supplied to the impoundment either by natural inflow or by
pumping from an adjacent water body. The average makeup flow must exceed the
average rate of evaporative water loss so that the content of dissolved solids in
the cooling water can be controlled. The difference between the makeup and
the loss is the "blow-down" or discharge from the impoundment. Cooling
impoundments may be classified into two types: cooling lakes and cooling
ponds. A cooling lake is an existing or man-made water body that impounds a
stream or river (i.e. an "on-stream" reservoir) and provides cooling as part of a
circulating water system. In the terminology of certain regulatory agencies,
this is called a "recirculating cooling water body" and thermal and water
quality standards apply both within the impoundment and for downstream
discharges. A cooling pond is an artificial impoundment that does not intercept
a stream or river (i.e. an "off-stream" or "perched" pond). This is a closed,
circulating water system and, except for the blow-down to an adjacent natural
body, is not normally subject to temperature or other water quality standards.
Both types of cooling impoundment have a number of advantages over
forced- or natural-draft evaporative cooling towers. Because of the lower
pumping head and heat rejection temperature, power production is more
efficient; the thermal inertia of cooling impoundments reduces the diurnal
meteorological temperature fluctuations associated with towers; and their
317
ability to store water on a seasonal basis reduces or eliminates the demand for
makeup water during periods of low flow and increases siting flexibility in
areas that are short of water. Disadvantages include land cost and availability
for off-stream ponds and environmental constraints in siting on-stream impound-
ments. In addition, the use of cooling impoundments has been constrained
by a lack of confidence in the ability to predict hydrothermal performance
and environmental impacts. The disadvantages associated with land cost and
availability for off-stream cooling ponds may sometimes be offset by the
requirement to purchase land in order to obtain" rights" to water usage.
The analysis of thermal discharges in large bodies of water, typical of once-
through systems, has been considerably aided by the development of near- and
far-field zone models. This conceptualization is possible because the far-field
advective and diffusive processes are weakly coupled to the buoyant jet-
induced mixing in the near-field zone. In contrast, transport processes in a
cooling impoundment are dominated by currents induced by the cooling water
discharge and intake. The development and verification of mathematical models
for the hydrothermal structure and performance of cooling ponds and lakes are
summarized below.
8.5.1. Classification of Cooling Impoundments

Existing and proposed cooling impoundments display a wide diversity in
physical features and thermal loading. Thermal loading is defined as the waste
heat rejected by the cooling water system in thermal megawatts per hectare
of water surface. Figure 8.10(a) shows a proposed cooling lake having an area of
620 ha and a thermal loading of 2.4 MW t ha -1 (Lake Merom, Indiana).
Cooling lakes, formed by impoundment of a natural water course, generally
have an irregular shape and relatively large depth (6-30 m). Long, dead-end
side arms are frequently found. Cooling lakes can offer advantages as multi-
purpose sites by accommodating facilities for hydroelectric power, water
supply, and recreation. Environmental regulations generally provide quantita-
tive limits on maximum temperatures or temperature rises within the lake or
otherwise set ecological guidelines on the impact of the induced heat load.
Figure 8.10(b) shows an off-stream cooling pond that does not impede the
flow of a natural stream (Powerton, Illinois). Cooling ponds are usually more
shallow than cooling lakes and, in addition to the external dikes forming the
pond, they may contain internal baffles and dikes to channel flow from dis-
charge to intake. The pond shown in Figure 8.10(b) has an area of 575 ha, a
thermal loading of 4.2 MW t ha - 1, and a mean depth of 3.4 m. Water quality
regulations usually do not apply to cooling ponds. Therefore, the ponds are
usually more heavily loaded (by a factor of 2) than cooling lakes and the major
design consideration is optimization of thermal performance to achieve minimal
intake temperatures and high thermal inertia.
318
Makeup~,
2000 ft
f------I
(a)
3000ft
I
Internal r--
baffles :: Makeup
11/
/"-.. I! Discharge
Levee- canal
Station
1780 MWe
(b) Intake channel (fossil)
FIGURE 8.10 Types of cooling impoundment: (a) a cooling lake, Lake Merom,
Indiana; (b) a cooling pond at Powerton, Illinois (from Jirka and Harleman, 1979).
Table 8.3 (Jirka and Harleman, 1979) compares the physical characteristics
and thermal loadings of some cooling ponds and lakes in the United States.
8.5.2. Distinction Between Stratified and Vertically Mixed Impoundments

The classification into cooling lakes and cooling ponds is useful for legal and
environmental assessment purposes. However, it does not suffice for defining
the internal hydrothermal structure of the impoundment. In order to develop
mathematical models for use in design and prediction of performance, it is
necessary to have an additional classification that distinguishes between
stratified and vertically mixed impoundments. This depends on the thermal
loading, the impoundment depth and shape, and the design and location of intake
and discharge structures.
All natural water bodies have a tendency to stratify thermally under the
action of solar and atmospheric radiation. In a cooling impoundment of average
319
TABLE 8.3 Comparison of Physical Characteristics for Typical Cooling
Lakes and Cooling Ponds in the United States.
Impoundment Area Electricity Waste Heat Thermal Loading
(ha) Generating (MW) (MW ha -')
Capacity
(MW,)
Coo/inq lakes
Lake Anna, VA 5200 3784 (nuclear) 7600 1.5
Clinton Lake. IL 1960 1982 (nuclear) 3750 1.9
Gibbons Creek Res .. TX 920 896 (fossil) 1170 1.3
Lake Merom. IN 620 980 (fossil) 1465 2.4
Lake Robinson. NC 900 135 (fossil) 1750 1.9
+ 730 (nuclear)
Lake Sanchris. IL 865 /232 (fossil) 1930 2.2
Sutherland Reservoir. NE 855 1300 (fossil) 2040 2.4
Cooling ponds
Braidwood.IL 1015 2200 (nuclear) 4520 4.5
Collins.IL 805 2520 (fossil) 3074 3.8
Dresden,IL 510 1600 (nuclear) 2678 5.3
La Salle, IL 825 2156 (nuclear) 4362 5.3
Powerton, IL 575 1780 (fossil) 2437 4.2
size, the kinetic energy input due to the flow-through induced by the condenser
water input and withdrawal is usually large enough to destroy the natural
stratification. Instead, an artificial stratification may be created because of the
forced temperature gradient between the discharge and the condenser intake.
Based on earlier work of Watanabe and Harleman (1977), Jirka and Harleman
(1979) proposed a dimensionless "pond number" P as a measure of whether a
cooling impoundment will be stratified or vertically mixed. The pond number is
defined as
(8.67)
where
1. is the interfacial quadratic-law friction factor,
Qo is the condenser flow rate,
d To is the condenser temperature rise,
fJ is the coefficient of thermal expansion,
g is the gravitational acceleration,
H is the average pond depth,
W is the average pond width (or flow path width in a pond with internal
baffles),
Dv is the dilution ratio for vertical entrance mixing at the plant discharge,
L is the pond length following the mean flow-through path.
320
Horizontal entrainment
_ _.l.--'"-+ +0 J
I
,, ~
T (x,y,t)
Horizontal entrainment
Entrance
Water Imixing
level region
To
--.
Q 0 ---,----!~.,....J--JA----t_---I.-_:f:fi
Vertical H
entrainment T (z, t)
~ Lower-layer
regiOnj
TI
FIGURE 8. I I Diagram of a stratified cooling impoundment.
A stratified cooling impoundment is shown schematically in Figure 8.11.

A derivation of the pond number is given by Jirka et at. (1978). Physically, P
represents the ratio of the thickness of the upper layer, hI' to the total depth H.
If the density of the upper layer, PI (x), is assumed to vary linearly with x because
of surface heat dissipation, then hI (x) = hs = constant and
P = hs (8.68)
H'
In the calculation of pond numbers, the following quantities are useful:

for typical Reynolds numbers, .h = 0.01 (field) and /; = 0.10 (laboratory)
(Jirka et at., 1975); {3117'0 = I1PO/P2' where {3 = 0.0002 °C- I at 20°C; Dv is a
function of the densimetric Froude number of the discharge channel:
Dv = l.2Fro - 0.2, (8.69)

where
(8.70)
321
and ao is the channel exit area if the discharge is along a wall and half the exit
area for a symmetrical discharge. Jirka et al. (1978) recommend a minimum
value of D. = 1.5 for rectangular discharges and D. = 1.2 for special, low-
mixing discharge structures with radial guide vanes. The higher the entrance
mixing parameter, the less likely it is that the pond will be stratified.
Inspection of available field and laboratory data indicates that cooling
impoundments in which P S 0.3 are well stratified "deep" ponds. On the other
hand, if P 20.7 the impoundments are vertically fully mixed "shallow" ponds.
Mathematical models for the prediction of hydrothermal performance are
available for these two cases and will be discussed below. Impoundments in the
range 0.3 < P < 0.7 are partially stratified, having no distinct surface layer and
exhibiting variable degrees of vertical stratification throughout the pond.
The analysis of this class of ponds is difficult, but this range of pond numbers can
be avoided by proper choice of design parameters.
Currents in cooling ponds result from three mechanisms: through flow
(generated by pumping), density differences, and wind stresses. The relative
magnitudes of density and through-flow currents are particularly important
and are closely linked to the thermal structure of cooling ponds; density currents
usually prevail in deep ponds, while shallow ponds may be governed by the
through flow, eddies in the discharge zone, and flow separations at constrictions
or around baffles. Under certain conditions, wind-induced currents may be
stronger than either through-flow or density currents. However, strong winds are
intermittent and surface heat loss during these periods is enhanced.
Deep, stratified impoundments are less subject to short-circuiting between
the discharge and condenser intake than shallow, vertically mixed ponds. In a
stratified pond, buoyancy acts to spread the flow over the surface of the im-
poundment, including dead-end side arms (Brocard et ai., 1977). The thermal
performance of shallow, vertically mixed ponds is highly influenced by the pond
geometry. Short-circuiting and the generation of large ineffective eddies should
be minimized by the use of interior baffles.
8.5.3. Vertically Mixed Cooling Ponds

Cooling impoundments that are characterized by pond numbers of0.7 or greater
are classed as vertically mixed ponds. Thus, the vertically averaged wind-
driven circulation models discussed in section 8.3.1 should be applicable in
combination with an appropriate heat transport equation. However, the
existence of longitudinal temperature gradients and flows generated by the
intake and discharge (which usually dominate the wind-induced currents)
warrants a special category of models.
Yeh et al. (1973) developed a two-dimensional model of a vertically mixed
cooling pond. They made three major assumptions in their transient model.
(a) The flow field is in steady state and the momentum equations contain only
322
FIGURE 8.12 Surface temperature distribution in a vertically mixed cooling pond, as

predicted by Yeh et al. (1973).
pressure and linearized friction terms, leading to a potential flow solution in

analogy to that for viscous flows. (b) There is no entrance mixing, thus reducing
the flow field to a pure source-sink motion. (c) The transient heat transport
equation neglects lateral diffusion and is solved in a Lagrangian sense, i.e.
following the stream tubes as given from the potential flow solution. This
fully mixed model was applied to a cooling pond with a retention time of about
eight days. No analysis or field data have been provided to ascertain that the
pond under study was indeed fully mixed. Results for the predicted temperature
field are given in Figure 8.12. The temperature pattern shows large lateral
temperature gradients at the power plant intake. This is contradictory to usual
observations and could be attributed to the neglect of convective momentum
and lateral heat diffusion terms.
Vasiliev (1978a) presented a model of a vertically averaged cooling pond
with quadratic bottom shear and surface wind shears. Both longitudinal and
lateral heat diffusion were included. At the inflow boundary, volume and
temperature flux conditions were imposed. No numerical results are available.
The importance of well designed internal baffles or dikes in vertically mixed
ponds is shown in Figure 8.13. Figure 8.13(a) shows surface isotherms in the
Powerton, Illinois pond and indicates recirculation eddies in each of the pond
compartments. In the Dresden, Illinois pond (Fig. 8.13(b» the flow is basically
longitudinal. The lateral temperature variations (higher temperatures at the
centerline of each channel and lower temperatures at the boundaries) indicate a
longitudinal dispersion effect in the heat transport.
323
3000 ft
(b)
FIGU RE 8.13 Surface isotherms CF) for two shaIlow cooling ponds: (a) Powerton, IL,
with predominant lateral recirculation behavior, and (b) Dresden, IL, with predominant
longitudinal dispersion behavior (from Jirka and Harleman, 1979).
324
Mathematical models to predict transient thermal performance for both
recirculating ponds and longitudinal dispersion ponds have been developed by
Jirka et al. (1978). Power plant heat loads and meteorological inputs are averaged
over the computational time step (usually between 3 hand 1 day) for prediction
of condenser water intake temperatures.
The pond number for the Powerton cooling pond is 0.6, while for Dresden
P = 0.8. Both ponds may be approximately classed as vertically mixed; however,
the mathematical models for the two differ considerably in their structure because
of the different length: width ratios of the individual compartments in the two
ponds. In Powerton the ratio is about 2, while in Dresden it is approximately 6.
Therefore, as noted above, the Powerton pond contains large recirculating
eddies while Dresden has a predominantly longitudinal flow configuration.
Details of the mathematical model formulations and verification with field data
were given by Jirka et al. (1978). In general, a longitudinal flow pond exhibits
better thermal performance than a recirculating pond. Observed and predicted
intake temperatures during one-and-a-half months are compared in Figure 8.14
for the Dresden pond.
96
90
u.. 84
0
- . -- ,
:J
~
(I)
78
"\
-Discharge
temperature
a.
E ..\~ft',"') "·0 ~.
",'
(I)
f- 72
•r.::J-I '~.
..
.r.N ...•.
,J'~~.
•
:\...-.. .J
~...
_. ~ ~
I ".-Yo,.•-.,..
\ .
\
..-.. ••
v~~y.. Predicted ., t
66 "intake \,,'-
.,.
temperature "-
Observed
60 ':----"--__L------'-_-,-L_"---------'--I _"------L_-'---------'-_--'------' -"-------L--'---------'-1
12 18 24 301 6 12 18 24 31,
September October 1975
FIGU RE 8.14 Predicted and measured intake temperatures for the Dresden. IL, cooling
pond during September and October 1975 (from Jirka et al., 1978). Mean error, O,8°F;
standard deviation, I.QuF.
325
8.5.4. Stratified Cooling Ponds
An extensive survey of techniques for the hydrothermal performance of strati-

fied cooling impoundments has been made by Ryan and Harleman (1973).
The important hydrodynamic and heat transport features of stratified cooling
lakes are as follows. Heated water enters the cooling lake through a discharge
channel; entrance mixing occurs as a result of vertical and lateral entrainment
into the discharge "jet"; beyond the entrance mixing region there exists a
stratified surface layer that is warmer than the underlying water, and surface
heat loss plays a dominant role in gradually decreasing the temperature of this
layer. Ultimately, at distances far from the discharge, the water in the surface
layer down-wells and enters the subsurface region. In this region the water is
advected downward owing to pond through flow. A skimmer wall may be used
at the intake to provide selective withdrawal from the cooler lower layers. The
water from the lower layers finally enters the power plant intake. This process
occurs in a continuous, closed-cycle fashion as the condenser water is discharged
with a temperature rise equal to ~ To above that at the intake. Cooling impound-
ments exhibit highly transient behavior because of variations in the power
plant loading and diurnal and seasonal meteorological fluctuations.
The stratified cooling impoundment model developed by Ryan and Harleman
(1973) explicitly accounts for the three-dimensional hydrodynamic and heat
transport features described above and illustrated in Figure 8.11. This model is
an extension of an earlier one-dimensional vertical reservoir model (Huber and
Harleman, 1968; Ryan and Harleman, 1971). The general applicability of this
model has been established through comparison with field data, notabl:y in the
TVA system (Wunderlich, 1973; Parker et al., 1975).
Ryan and Harleman (1973) used the buoyant jet model developed by
Stolzenbach et al. (1972) as the basis for determining the entrance mixing and the
thickness of the stratified surface layer, hs ' Thus, the upper-layer thickness is a
function only of the discharge channel geometry and densimetric Froude
number. The surface heat loss and, therefore, the temperature in the upper
layer are computed in terms of the fractional surface area but not explicitly
in terms of x-y position. Because of the large depth of the subsurface zone
(in comparison with the surface layer thickness), horizontal flows are neglected
and this zone is assumed to be horizontally stratified.
The intake flow under the skimmer wall takes water selectively from certain
layers, depending on the magnitude of the vertical density gradient. The
thickness of the withdrawal layer is computed using relationships developed by
Kao (1965). River inflows and reservoir outflows can be included at any pond
depth, and entrance mixing or withdrawal layer considerations are included in
these cases.
The transient pond behavior is computed by assuming essentially quasi-
steady state conditions within each time step. Quasi-steady state conditions
that apply to components such as entrance mixing, layer thickness, etc. are
326
taken to be constant within the time step. For this assumption to be correct,
it is necessary that the time step be longer than the characteristic time for
dynamic changes, which is essentially the impoundment length divided by the
internal wave speed. A time step of about one day is appropriate for this reason
and is short enough to allow the study of transient behavior due to plant
loading and meteorology.
Ryan and Harleman (1973) applied the model to laboratory experiments
and field data. Satisfactory agreement was found in most cases, including
surface temperature distribution, vertical profiles, and intake temperatures.
As an example, Figure 8.15 shows a comparison of predicted and measured
intake temperatures over one year for the Hazelwood, Australia cooling pond,
for which P = 0.17; thus, it is well stratified (Jirka et aI., 1978). The overall
accuracy is of the order of 1°C. Another comparison, in Figure 8.16, shows the
vertical temperature distribution in Lake Norman, North Carolina, with heated
discharges from the Marshall power station. This cooling lake also has river
through flow, which is included in the model. In summary, the model by Ryan
and Harleman (1973) seems a reliable tool for transient prediction of stratified
impoundments.
The Ryan and Harleman model was extended by Watanabe et al. (1975) to
take into account horizontal two-dimensional flow and temperature distribu-
tions in the surface layer. Since it is difficult to solve both near-field jet phe-
nomena and far-field stratified flow by using one model, the model is constructed
by integrating separate models for each region.
The major components of the mathematical model (Figure 8.11) are an
entrance mixing region and a surface layer heat loss region with horizontal
temperature gradients overlying a region with a vertical temperature gradient.
The entrance mixing region creates a strong lateral recirculation that significantly
affects the temperature distribution in the far field. In contrast to the Ryan and
Harleman (1973) model, which assumed that the surface layer thickness was a
function only of the discharge channel geometry and densimetric Froude
number, the Watanabe et al. (1975) model includes far-field effects, interfacial
friction, density differences between the surface and the lower layers, and pond
geometry. The surface layer thickness h s is given by eqns. 8.67 and 8.68.
The vertically integrated two-dimensional momentum and heat transport
equations for the surface layer are solved by finite-element methods. Since the
response of the stratified cooling pond is relatively steady over a short time
interval, the transient hydraulic behavior and heat distribution for the surface
layer of the cooling pond are approximated by solving the steady state mo-
mentum and heat equations sequentially with new boundary conditions and
new meteorological conditions at each time step. The lower-layer formulation
is the same as the Ryan-Harleman model. The model has been tested in the
laboratory (Watanabe and Harleman, 1977) and applied to Lake Anna, Virginia
(Jirka et aI., 1977), the site of a large nuclear power station.
327
L.l..
o
~ 75
-...
::J
co
Q)
Cl.
E
-
~
Q)
Q)
co
E 65
Q)
OJ
...
co
Q)
i;; 60
L
co
o
55 L -_ _--l.-_ _ -=-~I L __ _ _I
::_":_c=__------"--___:,--llc_::_--ll
o 100 200 300
Time (days)
FIGURE 8.15 Predicted and measured intake temperatures of the Hazelwood, Australia
cooling pond in 1969 (from Ryan and Harleman, 1973). Full line, mean measured intake
temperature; bold broken line, mean measured temperature at skimmer wall opening;
dotted line, predicted intake temperature; light broken line, predicted temperature at
skimmer wall opening.
20
---- ~
y-
·~140
-
'+- 40 __;
..
days
-
.c
Cl.
Q)
0 60
-..-
-<t
/-+ Marshall
~.
intake
200
80
100
.. days
40 90
FIGU RE 8.16 Predicted and measured vertical temperature profiles in Lake Norman,
North Carolina during August-October 1971 (after Ryan and Harleman, 1973). Full line,
predicted profile (run 1); • measured temperature range, 19 August 1971; • measured
temperature range, 12 October 1971.
328
10000 ft
Dam
FIGURE 8.17 The combined cooling pond (WHTF) and cooling lake of Lake Anna,
Virginia.
Lake Anna is an example of a stratified cooling lake, as shown in Table 8.3.

Actually it is a combined cooling pond and cooling lake. The heated condenser
water is discharged into a portion ofthe lake that is separated from the main lake
by dikes, as shown in Figure 8.17. A large fraction of the total waste heat is
dissipated in this" hot pond" enclosure, which is not subject to environmental
constraints. The flow from this enclosure enters the main lake through sub-
merged conduits in dike III and returns to the condenser water intake. The main
lake is subject to constraints on maximum temperatures and temperature
increases above natural conditions. The finite-element model of Watanabe
et al. (1975) was applied to the main lake and predicted velocity and temperature
distributions were obtained. The enclosed cooling pond portion is thermally
stratified and contains several dead-end side arms. Flows into and out of these
arms occur as buoyancy-driven counterflows. A segmented hydrothermal model
has been developed (Brocard et al., 1977; Jirka et al., 1977; Harleman et al.,
1978) to predict the cooling performance and temperature changes for the
combined Lake Anna impoundment.
Predicted temperatures for Lake Anna were obtained for a ten-year simula-
tion period under both heat loading and natural (preoperational) conditions.
329
1000m
Surface
21.0
f
t
FIGURE 8.18 Surface velocity and temperature CC) distributions in the First Ekibastuz
Thermal Power Plant.
The results from such long-term simulations can be processed to demonstrate

environmental effects such as induced temperatures and induced excess (above
natural) temperatures.
The model presented in section 8.3.5, developed by Vasiliev et al. (1974), was
modified through the elimination of the Coriolis term and the nonlinear inertial
term, and applied to the cooling reservoir of the First Ekibastuz Thermal Power
Plant (surface area 18.5 km 2 , volume 0.86· 108 m 3 , average depth 4.6 m) in the
USSR. Figure 8.18 indicates the predicted behavior of the currents and the
temperature distribution on the water surface under steady state conditions
(Vasiliev, 1978b).

The state of the art of predicting the transient performance of cooling im-
poundments by means of one-, two-, and three-dimensional mathematical
models is fairly well developed. Observations of small-scale cooling impound-
ments under controlled laboratory conditions are useful in the development and
verification of mathematical models. However, because of the different require-
ments for hydrodynamics, stratified flows, and surface heat dissipation and
because laboratory meteorology is essentially steady state, the direct transfer
of laboratory measurements to prediction of the behavior of an impoundment
is not recommended.
330
REFERENCES
Abbott, M. B., A. Danesgaard, and G. S. Rodenhuis (1973). System 21, Jupiter. Journal ofBydraulic
Research 11(1): 1-28.
Baca, R. G., M. W. Lorenzen, R. D. Mudd, and L. V. Kimmel (1974). A generalized water quality
model for eutrophic lakes and reservoirs. Ballel/e Pacific Northwest Laboratories Report PB-
256925.
Bauer, S. W. (1979). Three-dimensional simulation of time-dependent elevations and currents in a
homogeneous lake of arbitrary shape using an irregular-grid finite-difference model. Hydro-
dynamics ofLakes, eds. W. H. Graf and C. H. Mortimer (Amsterdam: Elsevier), pp. 267-276.
Bennett, 1. R. (1974). On the dynamics of wind-driven lake currents. Journal of Physical Ocean-
o.graphy 4( 3): 400-414.
Bonham-Caner, G., and J. H. Thomas( 1973). Numerical calculation of steady wind-driven currents
in Lake Ontario and the Rochester embayment. Proceedinys of 16th Conference on Great Lakes
Research (Ann Arbor, M I: International Association of Great Lakes Research), pp. 640-662.
Boyce, F. M. (1974). Some aspects of Great Lakes physics of importance to biological and chemical
processes. Journalof' Fisheries Research Board ot Canada 31 (5): 689-730.
Brocard, D. N .. G. H. Jirka, and D. R. F. Harleman (1977). A model for the convective circulation
in side arms of cooling lakes. MIT Department of Ciril Engineering, R. M. Parsons Laboratory
Technical Report 223.
Chen, C. W. (1970). Concepts and utilities of ecologic model. Proceedinqs ot American Society of
Civil Enqineers, Journal ot Sanitary Engineerinq Dil'ision 96(SA5): 1085-1097.
Chen. C. W., and G. T. Orlob (1972). Ecologic simulation for aquatic environments. Reportfor
Office of Water Resources Research, US Department of Interior by Water Resources Engineers,
Inc.
Chen, C. W., and D. J. Smith (1979). Preliminary insights into a three-dimensional ecological-
hydrodynamic model. Perspectives on Lake Ecosystem Modelinq,eds. D. Scaviaand A. Robertson
(Ann Arbor, MI: Ann Arbor Science), pp. 249-279.
Cheng, R. T. (1972). Numerical integration of lake circulation around island by the finite-element
method. International Journal of Numerical Methods in Enqineerinq 5: 103.
Cheng, R. T., T. M. Powell, and T. M. Dillon (1976). Numerical models of wind-driven circulation
in lakes. Applied Mathematical Modelling 1(3): 141-159.
Cheng, R. T., and C. Tung (1970). Wind-driven lake circulation by the finite-element method.
Proceedings of 13th Conference on Great Lakes Research (Ann Arbor, MI: International Associa-
tion of Great Lakes Research), pp. 891-903.
Csanady, G. T. (1975). Hydrodynamics of large lakes. Annual Review of Fluid Mechanics 7: 357-
386.
Dailey. J. E., and D. R. F. Harleman (1972). Numerical model for the prediction of transient water
quality in estuary networks. MIT Department of Cil'i/ Engineering, R. M. Parsons Laboratory
Di Toro, D. M., D. J. O'Connor, R. V. Thomann, and J. L. Mancini (1975). Phytoplankton-zoo-
plankton-nutrient interaction model for western Lake Erie. Systems Analysis and Simulation in
Ecology vol. 3, ed. B. C. Pal ten (New York, NY: Academic Press), ch. II.
Edinger, J. E., and E. D. Buchak (1978). Reservoir longitudinal and vertical implicit hydro-
dynamics. Proceedings on Environmental Effects of Hydraulic Enqineerinq Works, Knoxl'ille, TN
(International Association of Hydraulics Research), pp. 160-169.
Fischer, G. (1959). Ein numerisches Verfahren zur Errechnung von Windstau und Gezeiten in
Randmeeren. Tel/us 11(1):60-76.
Fischer, G. (1965). A survey of finite-difference approximations to the primitive equations. Monthly
Weather Review 93(1): 1-10.
Gallagher, R. H., J. A .. Liggett, and S. K. T. Chan (1973). Finite-element shallow lake circulation
analysis. Proceedinqs of American Society of Ciril Enqineers, Journal of Hydraulics Dil'ision
No. HY7: 1083-1095.
331
Gates, W. L. (1968). A numerical study of transient Rossby waves in a wind-driven homogeneous
ocean. Journal 0/ Almo.\pheric Science 25( 1): 3-22.
Gedney, R. T., and W. Lick (1970). Numerical calculation of the steady state wind-driven currents
in Lake Erie. Proceedinqs 0/ l3th Con/i.'rencc on Gre(J{ Lakes Research (Ann Arbor, MI: Inter-
national Association of Great Lakes Research), pp. 829-838.
Gedney, R. T., and W. Lick (1972). Wind-driven currents in Lake Erie. Journal of Geophrsical
Research 77: 2714-2723.
Grotkop, G. (1973). Finite-element analysis of long-period water waves. CompUier MelilOds in
Applied Mechanics and Enqineerinq 2(2): 147-157.
Hansen, W. (1956). Theorie zur Errechnung des Wasserstandes und der Stromungen in Rand-
meeren nebst Auwendungen. Tel/us 8(3): 287-300.
Harleman, D. R. F., G. H. Jirka, D. N. Brocard, K. A. Hurley-Octavio, and M. Watanabe (1978).
Transient simulation of cooling lake performance under heat loading from the North Anna Power
Station. Proceedings of2nd Conference on Waste Heat Managemenl and Ulilization, Miami, FL.
Heaps, N. S. (1969). A two-dimensional numerical sea model. Philosophical Transactions, Series
A 265(1160):93-137.
Hinze, J. O. (1959). Turbulence (New York, NY: McGraw-Hili).
Huber, W. C., and D. R. F. Harleman (1968). Laboratory and analytical studies of thermal strati-
fication of reservoirs. MIT Departmenl o/Civil Engineering, R. M. Parsons Laboratory Technical
Report 112.
Hurley-Octavio, K. A., G. H. Jirka, and D. R. F. Harleman (1977). Vertical heat transport mech-
anisms in lakes and reservoirs. MIT Department o/Cil'il Engineerinq, R. M. Parsons Laboratory
Technical Reporl 227.
Hyden, H. (1974). Water exchange in two-layer stratification waters. Proceedings of American
Society 0/ C;I'il Enqineers, Journal 0/ Hydraulics Dil'ision I00(HY3): 345-361.
Imboden, D. M. (1974). Phosphorus model for lake eutrophication. Limnoloqy and Oceanoqraphy
19(2): 297-304.
Jirka, G. H., G. Abraham, and D. R. F. Harleman (1975). An assessment of techniques for hydro-
thermal prediction. MIT Department 0/ CiI'il Engineerinq, R. M. Parsons Laboratory Technical
Reporl 203.
Jirka, G. H .. D. N. Brocard, K. A. Hurley-Octavia, M. Watanabe, and D. R. F. Harleman (1977).
Analysis of cooling effectiveness and transient long-term simulations of a cooling lake (with
application to the North Anna Power Station). MIT Department 0/ Cil'il Enqineerinq, R. M.
Parsons Laboratory Technical Reporl 232.
Jirka, G. H., and D. R. F. Harleman (1979). Cooling impoundments: Classification and analysis.
Proceedings o/American Society o/CiI'il Engineers, Journal o/Em'ironmental Enqineering Dil'ision
105(EE2): 291-309.
Jirka, G, H., M. Watanabe, K. A. Hurley-Octavio, C. F. Cerco, and D. R. F. Harleman (1978).
Mathematical predictive models for cooling ponds and lakes. Part A: Model development and
design considerations. MIT Department ofCil'il Enqineerinq, R. M. Parsons Laboratorr Technical
Report 238.
Jones, W. P., and B. E. Launder (1973). The calculation of low Reynolds number phenomena with a
two-equation model of turbulence. International Journal 0/ Heat and Mass Tran~/er 16(6): 1119-
1130.
Kao, T. W. (1965). The phenomenon of block in stratified flow. Journal of Geophysical Research
70(4): 815-822.
Karelse, M. (1974). Vertical exchange coefficients in stratified flows. Momentum and Mass Tran.5fer
in Stratified Flows, ed. H. N. C. Breusers. De/fl Hydraulic LaboralOry Report R.880, ch. 2.
Kasahara, A. (1965). On certain finite-difference methods for fluid dynamics. MOllthly Weather
Rel'iew 93(1):27-31.
King, I. P., W. R. Norton, and G. T. Orlob (1973). Finite-element model for two-dimensional
density-stratified flow. Water Resources Enqineers, Inc. Reporl 10 Office 0/ Water Resources
Research.
332
Koh, R. C. T, and L. N. Fan (1970). Mathematical models for the prediction of temperature distri-
butions resulting from the discharge of heated water in large bodies of water. US Environmental
ProteC1iol1 Agency Research Series No. 16130 DWO.
Kullenberg, G. (1974). An experimental and theoretical investigation of the turbulent diffusion in
the upper layer of the sea. Copenhayen Unirl'rsitv, Institutefor Phvsical Oceanography Report 25.
Kvon, V. l. (1979a). Hydrothermal computation. Izvestiya Akademia Nauk SSSR. Energetika i
Transport No.5 (in Russian).
Kvon, V. I. (1979b). A temperature-stratified flow in a running flow water body. MeteoroloC/iia i
Gidrologiia No.6: 74-79 (in Russian).
Lam, D, C. L., and E. Halfon (1978). Model ofprimary production, including circulation influences
in Lake Superior. Journal 01 Applied Mathematical ModelinC/ 2( I): 30-40.
Lam, D. C. L., and T. J. Simons (1976). Numerical computations ofadvective and diffusive trans-
ports of chloride in Lake Erie during 1970. Journal 01 Fisheries Research Board of Canada
33(3): 537-549.
Larsen, D. P., and H. T Mercier (1975). Shagawa Lake recovery characteristics as depicted by
predictive modeling. In: Water Quality Criteria Research of the US Environmental Protection
Agency. US Enl'ironmental Protection Agency, Cormllis, OR Report EPA 660/3-76-079.
Launder, B. E. (1976). Comments on improved form of the low Reynolds number k-f turbulence
models, by G. H. Hoffman. Physics ofFluids 19(5): 765-766.
Lee. K. K .. and J. A. Liggett (1970). Computation for circulation in stratified lakes. Proceedinqs
of American Socien' of Cil·il Engineer." Journal 0/ Hydraulics Dirision No. HY 10: 2089-2/15.
Leendertse, J. J. (1967). Aspects ofa computational model for long-period water-wave propagation.
Rand Corporation, Santa Monica, C A Memorandum RM-5294-PR.
Leendertse, J. J. (1970). A water quality simulation model for well mixed estuaries and coastal
seas: vol. I, Principles of computation. Rand Corporation, Santa Monica, CA Memorandum
RM-6230-RC.
Lick, W. (1976). Numerical models of lake currents. US Enl'ironmental Protection Ayency En-
rironmental Research Laboratories, Duluth, M N Report EPA 600/3-76-020.
Liggett. J. A. (1969). Unsteady circulation in shallow homogeneous lakes. Proceedinqs of American
Societl' 0/ Ciril Enyineers, Journal o/Hydraulics Dirision No. H Y4: 1273-1288.
Liggett, J. A .. and C. Hadjitheodorou (1969). Circulation in shallow homogeneous lakes. Pro-
ceedinys of American Society 0.1 Civil Engineers, Journal 01 Hvdraulics Dirision 95(HY2): 609-620.
Liggett, J. A., and K. K. Lee (1971). Properties of circulation in stratified lakes. Proceedings 0/
American Society ofCit·il Engineers, Journal 0/ Hydraulics Dil'ision No. HY I.
Lilly, D. K. (1961). A proposed staggered-grid system for numerical integration of dynamic
equations. Monthly Weather Rel·ie .... 89(3): 59-65.
Long, R. R. (1970). A theory of turbulence in stratified fluids. Journal 0/ Fluid Mechanics 42(2): 349-
365.
Marchuk, G. I. (1974). Numerical Solution o/the Problems of Atmospheric and Ocean Dynamics
(Leningrad: Gidrometizdat) (in Russian).
Marchuk, G. I. (1975). Methods o/Numerical Mathematics (New York, NY: Springer).
Marchuk, G. I., and V. B. Zalesny (1974). Numerical Model o/theGlobal Ocean Circulation (Siberian
Branch, USSR Academy of Sciences) (in Russian).
Markofsky, M., and D. R. F. Harleman (1971). A predictive model for thermal stratification and
water quality in reservoirs. MIT Department of Cit'il Engineering, R. M. Parsons Laboratory
Techn;cal Report 134.
Markofsky, M .. and D. R. F. Harleman (1973). Prediction of water quality in stratified reservoirs.
Proceedinys o/American Society o/Cil'il Engineers, Journal o/Hydraulics Dil·ision 99(HY5): 729-
745.
McNider, R. T, and 1. J. O'Brien (1973). A multilayer transient model of coastal upwelling. Journal
o/Phvsical Oceanography 3(3): 258-273.
Monin, A. S., and A. M. Yaglom (1965). Statistical Fluid Mechanics: Mechanics/or Turbulence
vol. I (Moscow: Nauka: English translation, 1971. by Massachusetts Institute of Technology).
333
Mortimer. C. H. (1974). Lake hydrodynamics. Mirreihm.'len del' Imernationalen Vereiniqunq
Linmoloqie 20: 124- 197.
Munk, W. H., and E. R. Anderson (1948). Notes on the theory of the thermocline. Journal oj"
Marine Research I.
Najarian. T. 0., and D. R. F. Harleman (1975). A real-time model of nitrogen-cycle dynamics in an
estuarine system. MIT Department oj" Ciril Enqineerin.'l. R. M. Parsons Laboratorr Technical
Report 204; also Prowess in Water Tecll/loloqy (1977) 8:323-345.
Norton. W, R., I. P. King. and G. T. Orlob (1973). A finite-element model for Lower Granite
Reservoir. W{I(er Resources Enqineers Inc.. Walnut Creek, C A Technical Report.
Okubo. A. (1962). A review of theoretical models for turbulent diffusion in the sea. Journal of
Ocean Society oj"Japan, 20th anniversary volume, I: 286-320.
Okubo, A. (1968). A new set of oceanic diffusion diagrams. Chesapeake Bay Institute, Johns
Hopkins Unil'ersity Technical Report 38.
Institute, Ralei.'lh Report 97.
Orlob, G. T., and L. G. Selna (1967). Progress report on development of a mathematical model for
prediction of temperatures in deep reservoirs-Phase 3: Castle Lake investigation. Water
Resources Enqineers, Inc., Walnut Creek, C A Technical Report.
Ozmidov, R. V. (1968). Horbmral Turbulence and Turbulent Exchanye ill the Ocean (Moscow:
Nauka).
Parker. F. L.. B. A. Benedict, and C. Tsai (1975). Evaluation of mathematical models for tempera-
ture prediction in deep reservoirs. US Enl'ironmental Protection A.'Iencr. National Enl'ironmemal
Research Center. Corvallis, OR Report EPA 660/3- 75-038.
Patterson. D. J .. E. Epstein, and J. McEvoy (1975). Water pollution investigations: Lower Green
Bay and Lower Fox River. Report to US Elll'ironmental Protection A.'Iency, No. 68-01-1572.
Phillips, 1. O. (1969). The Dynamics oj the Upper Ocean (London: Cambridge University Press).
Plate, E. 1.. and P. Wengefeld (1979). Exchange processes at the water surface. Hydrodvnamics
()f Lakes, eds. W. H. Grafand C. H. Mortimer (Amsterdam: Elsevier), pp. 277-301.
Platzman, G. W. (1963). The dynamic prediction of wind tides on Lake Erie. Meteoroloqical Mono-
.'Iraphs 4(26): 1-43.
Raney, D. c.. D. L. Durham, and H. L. Butler (1979). Application of a three-dimensional storm
surge model. Hydrodynamics of Lakes. eds. W. H. Graf and C. H. Mortimer (Amsterdam:
Elsevier), pp. 77-86.
Rao. D. B., and T. S. Murty (1970). Calculation of the steady-state wind-driven circulations in
Lake Ontario. Archil' /iir Meteoroloqie, Geophysik. ulld Biok lima, Serie A 19(2): 195-210.
Reid. R. 0 .. and B. R. Bodine (1968). Numerical model for storm surges in Galveston Bay. Pro-
ceedin.'ls oj" American Society ()j" CIl'i! En.'lineers. Journal oj" Waterways and Harhors Dil'ision
94(WWI ):33-37.
Ryan. P. J .. and D. R. F. Harleman (1971). Prediction of the annual cycle of temperature changes
in a stratified lake or reservoir: Mathematical model and user's manual. MIT Departmellt 0/
Ciril Enqineeri/l.'1, R. M. Parsons Lahoratory Technical Report 137.
Ryan, P. J., and D. R. F. Harleman (1973). An analytical and experimental study of transient
cooling pond behavior. MIT Department afCiri! En.'lineerinq, R. M. Parsons Lahoratorr Techni-
cal Report 161.
Schlichting, H. (1968). Boundarr Larer Theory (New York, NY: McGraw-Hili).
Simons. T. J. (1971). Development of numerical models of Lake Ontario. Proceedings of 14th
Conference on Great Lakes Research (Ann Arbor. M I: International Association of Great Lakes
Research), pp. 654-669.
Simons. T. J. (1972). Development of numerical models of Lake Ontario. Part 2. Proceedin.'ls oj"
15th Conference on Great Lakes Research (Ann Arbor, MI: International Association of Great
Lakes Research), pp. 655-672.
Simons. T. J. (1973). Development of three-dimensional numerical models of the Great Lakes.
Canada Cemer/or Inland I-Vaters, Burlin.'lfOn, Omario Science Series No. 12.
334
Simons, T. 1. (1974). Verification of numerical models of Lake Ontario: Part I. Circulation in
spring and early summer. Journal of'Physical Oceano.'l'aphy 4(4): 507-523.
Simons, T. J. (1975). Verification of numerical models of Lake Ontario: Part II. Stratified circula-
tions and temperature changes. Journal of'Phvsical Oceanography 5( I): 98-110.
Simons, T. J. (1976). Analysis and simulation of spatial variations of physical and biochemical pro·
cesses in Lake Ontario. Journal of Great Lakes Research 2(2): 215-233.
Simons, T. J. (1980). Circulation models of lakes and inland seas. Canadian Bulletin of' Fish and
Aquatic Sciences 203, 146 pp.
Simons, T. 1., L. Funkquist, and J. Svensson (1977). Application of a numerical model to Lake
Viinern. Swedish Meteorology and Oceanoqraphy Institute Report RH09.
Simons, T.1., and D. C. L. Lam (1978). Nutrient exchange between volume elements in large lakes.
Modeling the Water Quality of the Hydrological Cycle. Proceedings of Baden Symposium, Sep-
tember. International Association of' Hydroloqical Sciences Publication 125, pp. 185-191.
Snodgrass. W. J., and C. R. O'Melia (1975). Predictive model for phosphorus in lakes. Em'iron-
mental Science and Technolom' 9( I0): 937-944.
Stolzenbach, K. D.. E. E. Adams, and D. R. F. Harleman (1972). A user's manual for three-di-
mensional heated surface discharge computations. MIT Department of' Cil'il Engineering, R. M.
Parsons Laboratory Technical Report 156.
Siindermann, 1. (1979). Numerical modelling of circulation in lakes. Hydrodynamics of Lakes, eds.
W. H. Graf and C. H. Mortimer (Amsterdam: Elsevier), pp. 1-29.
Tatom, F. B., and W. R. Waldrop (1978). Curvilinear unsteady two-dimensional depth-averaged
hydrodynamic model. Proceedings of' Conference on Enl'ironmental Effects of' Hydraulic En-
gineerinq Works. Knoxl'ille, TN (International Association of Hydraulics Research), pp. 329-
339.
Taylor, c., and J. Davis (1972). Tidal and long-wave propagation-A finite-element approach.
Department ofCil'il Engineering, Unitwsity College of'Swansea Report C/Rfl89.
Thomann, R. V., D. M. Di Toro, R. P. Winfield. and D. J. O'Connor( 1975). Mathematical modeling
of phytoplankton in Lake Ontario. US Enl'ironmental Protection Agency, National En!'ironmental
Research Center, Corvallis, OR Report EPA 660/3-75-005.
Tjomsland, T. (1979). Application ofa mathematical hydrodynamic model to Lake Mjosa. Hydro·
dynamics l!fLakes, eds. W. H. Graf and C. H. Mortimer (Amsterdam: Elsevier), pp. 41-49.
Tsvetova, H. A. (1979). Numerical modelling of the dynamics of Lake Baikal. Professional Paper
Vasiliev, O. F. (l978a). Three-dimensional numerical model for hydrothermal analysis of cooling
ponds. Thermal Etftuent Disposal from Power Generation, ed. Z. P. Zaric (New York, NY:
McGraw-Hill), pp. 115-132.
Vasiliev, O. F. (I 978b). Two-dimensional numerical models for hydrothermal analysis of water
bodies. Thermal Ejjluent DisposalFom Power Generation. ed. Z. P. Zaric (New York. NY:
McGraw-Hili), pp. 133-149.
Vasiliev, O. F., and V. I. Kvon (1977). Numerical modeling of flows and heat exchange in cooling
ponds. Proceedings of 16th Conqress of'International Association for Hydraulic Research, Baden·
Baden vol. 2, pp. 1-8.
Vasiliev. O. F., V. I. Kvon. and R. T. Chernishova (1973). Mathematical modelling of the thermal
pollution of a water body. Proceedings of'the 15th Congress of'the International Association Jor
Hydraulic Research, Istanbul vol. 2, pp. 129-136.
Vasiliev, O. F., V. I. Kvon. and R. T. Chernishova (1974). Thermally stratified flow in a water body
of oblong form. Gidroteknicheskoi Stroitel'stl'o NO.4 (in Russian).
Vreugdenhil, C. B. (1973). Computational methods for the vertical distribution of flow in shallow
water. De!f't Hydraulics Laboratory Report WI 52.
Wang, 1. D., and 1. 1. Connor (1975). Mathematical modeling of near-coastal circulation. MIT
Department of' Ci!'il Engineerinq, R. M. Parsons Laboratory Technical Report 200.
Watanabe. M., and D. R. F. Harleman ( 1977). A mathematical model of a stably stratified cooling
335
pond. Proceedings of 16th Congress ()/lntemational Association jor Hydraulic Research. Baden-
Baden vol. 2, pp. 67-74.
Watanabe, M., D. R. F. Harleman, and J. J. Connor (1975). Finite-element model for transient
two-layer cooling pond behavior. MIT Department ojCiL'i1 Engineering, R. M. Parsons Laboratory
Welander. O. (1957). Wind action on a shallow sea: Some generalizations of Ekman's theory.
Tellus 9(1) :45-52.
Wittmiss, J. (1979). Application ofa transient mathematical model to Lake Kosen. Hydrodynamics
oj Lakes, eds. W. H. Graf and C. H. Mortimer (Amsterdam: Elsevier), pp. 31-40.
Wu, J. (1969). Wind stress and surface roughness at air-sea interface. Journal 0/ Geophysical
Research 74(2): 444-445.
Wunderlich, W. (1973). Water temperature prediction model for deep reservoirs. Tennessee Valley
Authority. Water Resources Management Methods Stafr Technical Report A-2.
Yeh, G. T.. A. P. Verma. and F. H. Lai (1973). Unsteady temperature prediction for cooling ponds.
Water Resources Research 9(6): \555-1563.
Yudelson, J. M. (1967). A survey of ocean diffusion studies and data. W. M. Keck Laboratory
for Hydraulics and Water Resources. California Institute 0/ Technology Technical Memorandum
67-2.
CHAPTER 8: NOTATION
dissipation parameter
channel width
C concentration of a conservative substance
Cf resistance coefficient
C interfacial shear stress coefficient; Chezy coefficient
Cw wind stress coefficient
Dv dilution ratio
e kinetic energy (per unit mass)
E v vertical Ekman number
fo or A friction coefficient (Darcy-Weisbach)
h height from datum to the bottom
H depth of water
K heat diffusivity
K H heat transfer coefficient
I mixing length
L length scale of turbulence; pond length
Manning coefficient
pond number
heat source
external sources
shear velocity
average velocities
net entrainment or mixing velocity between layers k and k +
336
W wind velocity; average pond width
f3 coefficient of thermal expansion
( eddy viscosity
YJ elevation of free surface
K von Karman constant
<Pn net heat flux through surface.
Ecological Modeling of Lakes
9 S. E. Jorgensen
9.1. BACKGROUND
9.1.1. Water Quality and Ecological Models of Lakes

Chapter 4 surveyed the modeling of ecological processes in streams and lakes.
As elucidated in Figure 4.1, the various submodels must be coupled to make a
total ecosystem model. This has already been demonstrated for models of
streams in Chapter 6.
Nutrient enrichment of lakes has been of major concern in pollution studies,
so it is not surprising that most of the working ecological models of lakes focus
on the problem of eutrophication. The eutrophication models concentrate on
the kinetics of water quality and biota and are therefore, in structure, ecological
models.
The eutrophication of rivers is generally not a problem, because of the short
retention time of water. The major problem in rivers is the concentration of
dissolved oxygen, which depends on the algae producing oxygen, on the nitrogen
cycle consuming oxygen, and on other chemical-biological constituents. The
direct relationship between eutrophication and algal concentration implies,
however, that ecological models have found a much wider use in lake manage-
ment than in river management.
The purpose of this chapter is not to give all specifications of mathematical
models and programs, but to describe some models that are characteristic of
a reasonably wide range of those actually in use. Many are mentioned in the
review (sections 9.2.2 and 9.2.3) and the total systems of equations for five
models are shown in the appendix.
337
338
Among the earliest models of lake systems were those that addressed the
problem of nutrient balance. These nutrient budget models simulated the entire
lake as a mixed tank reactor (Vollenweider, 1965, 1969). Later, this approach
was seized upon by other investigators dealing more explicitly with nutrient
and algal budgets in lakes.
Mathematical models of lakes have evolved along two different lines. First,
there was the extension of the zero-dimensional model to one-, two-, and three-
dimensional models (Chapters 7 and 8); then there were the modeling activities
that focused primarily on a better and more detailed description of the
chemical-biological processes (Park et al., 1974; Chen and Orlob, 1975;
J0rgensen, 1976; and others).
The line represented by CLEAN, CLEANER, and MS CLEANER demon-
strates that the development has been toward inclusion of more biological
constituents, linked by increasingly detailed description and interaction of
biological phenomena. The latest development includes adaptation and bio-
logical feedback mechanisms (Straskraba, 1976, 1979; J0rgensen et aI., 1978;
Park et al., 1979). The tendency is to build more realism into the models by using
more constraints.
9.2. ECOLOGICAL LAKE MODELS
9.2.1. Coupling of Submodels

Over the past decade there has been a clear evolution toward more and more
complex models, not only by adding more constituents but also by making
more complex links (adaptation, feedback, self-organization) between the sub-
models. Ten models have been chosen to demonstrate different levels of com-
plexity and of coupling. The presentation of the models is taken from Park et
al. (1979) with the authors' permission. Figure 9.1 shows the symbols used in
the model descriptions.
Figure 9.2 shows the model of Chen and Orlob (1975). Light and nutrients
are represented by saturation kinetics and temperature by the exponential
van't Hoff equation (eqn. 1, Table 4.12). Nutrient limitation is set by the mini-
mum of the nitrogen, phosphorus, and silicon limitations, while the contri-
butions of light, nutrient, and temperature limitations are multiplicative. Light
extinction is a function of depth and of detritus and phytoplankton concentra-
tions. A Michaelis-Menten expression describes grazing. Di Toro et al. (1977)
use Steele's (1962) equation for light and temperature limitations; grazing and
phytoplankton mortality vary linearly with temperature (Figure 9.3).
In the model of Kremer and Nixon (1978) (Figure 9.4) the limitations are
multiplied, but the functions are the same as in the model of Di Toro et al.
339
DProcess
D
Function
0
Passive storage
~
Sink
~
0
State variable
Harmonic mean
DSwitch
FIGU RE 9.1 Symbols used in Figures 9.2-9.11.
,e-
-§
c>U
._ C
-J 0;:;
)(
Q)
J
Outflow
Phytoplankton
FIGURE 9.2 Flow chart of the phytoplankton model of Chen and Orlob (1975). LO,
N D. T D: light, nutrient, and temperature dependence.
340
Outflow
Temperature Temperature
FIGURE 9.3 Flow chart of the phytoplankton model of Oi Toro <,t al. (1977). FP: factor
dependent on timc.
Diurnal variations of light and a three-day running average for adaptation to

light are included. The model was developed for a marine system, but is directly
applicable to lake systems as well.
The model proposed by Straskraba (1976) is presented in Figure 9.5. Light
limitation is based on the work of Steel (1972); the limiting factors are multi-
plicative; the Stokes law is applied to sinking, and grazing accounts for prey
saturation. Light extinction considers particulate organic matter, dissolved
organic matter, suspended sediment, and phytoplankton biomass. A noticeable
feature is the temperature response of phytoplankton growth with an adaptive
optimum temperature (see also Table 4.12).
The model of Lehman et al. (1975) (Figure 9.6) provides for intracellular
storage of nutrients (discussed in section 4.2.2). Michaelis-Menten expressions
for limitations are used multiplicatively. Mortality is dependent on the number
of days when the growth is suboptimal (section 4.2.4) and sinking is constant.
341
FP
/\
Days
Phytoplankton
FIGURE 9.4 Flow chart of the phytoplankton model of Kremer and Nixon (1978).
Nyholm's (1978) model (Figure 9.7) takes intracellular storage of nutrient

into account (see Table 4.1 and discussion in section 4.2.2). The nutrient limita-
tion terms are combined as resistors in series, the so-called harmonic mean,
suggested by Bloomfield et al. (1974). Light inhibition and adaptation are
incorporated. The van't Hoff equation is used for temperature response and
Michaelis- Menten kinetics for grazing.
The model of Scavia et al. (1976) uses Michaelis-Menten expressions for
light and nutrient limitations and the minimum is used as the limiting factor
(Figure 9.8), which is then multiplied with the nonlinear temperature factor of
Bloomfield et al. (1974). Equations 2 in Table 4.3 are used for grazing and are
modified by temperature and prey preference. Equation 5 of Table 4.6 is used to
express sinking. Respiration is a linear function of temperature and mortality
a function of temperature and nutrient limitation.
The model of Jl2Jrgensen (1976) and Jl2Jrgensen et al. (1978) in Figure 9.9
considers independent nutrient cycles and uses eqns. 14 of Table 4.1 to describe
342
c
_.9
.<:-
c>U
._ C
.....J .~
x
<ll /
L Phytoplankton + DaM + POM + sediment
FP
Outflow
c
o
.;:;
~ /
.~ /
CI:L....-_ _---l
Temperature Viscosity Phytoplankton
FIGU RE 9.5 Flow chart of the phytoplankton model of Straskraba (1976).
algal growth. Light inhibition and adaptation are included, and photosynthesis
determines the intracellular concentration of carbon (photosynthesis being the
uptake of CO 2 ), In addition, respiration (release of CO 2 ) influences this con-
centration, but is controlled by the phytoplankton concentration, temperature,
and the concentration of intracellular carbon. Light extinction accounts for
phytoplankton, detritus, and zooplankton concentrations. Algal growth is con-
trolled by the intracellular concentrations of nutrients (N, P, and C) and the
temperature. A multiplicative expression is used. Outflow, inflow, and grazing
are included; the latter is represented by a Michaelis-Menten expression and a
temperature function. Sinking is controlled by viscosity and a rather complex
sediment model for description of nutrient release is applied.
Figure 9.10 presents the model CLEANER (Youngberg, 1977). It uses
Steele's (1962) equation for light limitation; the harmonic mean of Bloomfield
et al. (1974) to combine nutrient limitations; the nonlinear response of growth
to temperature; and grazing with saturation kinetic formulation, nonlinear
temperature dependence, and prey preference. Respiration, mortality, and
343
Nutrient
FP
Temperature
Excretion Sink
FIGURE 9.6 Flow chart of the phytoplankton model of Lehman el al. (1975).
sinking are temperature-dependent and nutrient limitation influences the

mortality (Scavia et aI., 1976).
MS CLEANER (Figure 9.11) is the most complex of the models presented
in this section. It is claimed that because the parameters have physical and
biological meaning (Desormeau, 1978; Park et aI., 1979a) extensive calibration
is not needed in order to apply the model to a new site. Intracellular storage of
nutrients is used and the uptake processes for phosphorus and nitrogen are
represented by a Michaelis-Menten kinetic (eqns. 14 of Table 4.1). Uptake is,
furthermore, a function of light and temperature. Single-nutrient limitation is
determined by comparing the N:P ratio with thresholds (Droop, 1974; Rhee,
1978).
344
I sat
.. V V
e
.. ..
e•
Days Nutrient supply

FP
Sink
Phytoplankton
FIGURE 9.7 Flow chart of the phytoplankton model of Nyholm (\978).
Light limitation provides for adaptation to varying light intensities (Groden,

1977) and photosynthesis is a linear function of chlorophyll concentration,
which, in phytoplankton, is an exponential function of light intensity. Light
inhibition is represented by Steele's (1974) equation, and Smith's (1936) equation
is used below the inhibitory level. Light extinction is a function of dissolved
organic matter, particulate organic matter, and phytoplankton concentrations.
Temperature limitation does not control photosynthesis directly, but is used
to determine the light saturation and nutrient uptake rate. The equation
developed by Groden (1977) is applied, and temperature adaptation is de-
scribed by using the empirical function suggested by Straskraba (1976). Grazing
differentiates between saturation kinetic feeding, exhibited by copepods and
fish, and feeding at a uniform rate, exhibited by cladocerans that filter at a
constant rate. Grazing is also a function of temperature and prey preference
(Park et al., 1974).
9.2.2. Nutrient Budget Models

Vollenweider (1969, 1975) has suggested a simple nutrient budget model that
considers input, output, and net loss to the sediments (plus a correction factor
345
c
o
.;::
~
00.
III
Q)
c:: " - - - - - - - I
Temperature Temperature Viscosity
TO
FIGURE 9.8 Flow chart of the phytoplankton model of Scavia ('{ at. (1976).
for stratified lakes):

d[PJ
Cit = i p - (Pw - up)[PJ, (9.1)
where
[PJ is the phosphorus concentration in the lake [M L - 3 J,
i p is the phosphorus su pply rate [M L - 3 T - 1J,
Pw is the hydraulic washout coefficient [T-1J,
Up is the sedimentation rate constant [T-1J,
t is time.
This equation is valid ifthe lake is well mixed, its volume is constant, the outflow
is at a concentration equivalent to that of the lake and is equal to the inflow,
346
Outflow
Viscosity
c.. Cl
~ Complex C
';:j
m submodel ~
ClL-_ _--J c.9 L-_---""---'
~ Temperature
]7
Temperature
FIGURE 9.9 Flow chart of the phytoplankton model of Jfilfgensen (1976) and J0rgensen
al. ( 1979).
f't
347
c
o
.~
'c.
til
Q)
cr:L-_ _----L.j
Temperature Temperature
Temperature
FIGURE 9.10 Flow chart or the CLEANER model or Youngberg (1977). Clad.:
cladocerans; Cope. : copepods.
and there is no net supply of phosphorus from sediment. Provided that Ip , Pw,
and tTp are time-independent this equation can be solved:
II'
[pet)] = [P o]exp[ - (Pw - tTp)t] + {I - exp[ - (Pw - tTp)tJ (9.2)
Pw + tTp
Figure 9.12 shows the Vollenweider plot, which is based upon the above-
mentioned considerations.
348
>-
J::
LO TO a.
o
(;
6 -'--/0--===1
A
Nutrient ~
Chlorophyll
~.~ A
/
J::'"
Ole:
.- <Il
--'-
.S
TO
L Phytoplankton
+ DOM .........::...---:..-_-'-'- -'-----J
+ POM 1M
.~~
(lJ
0: - ' -_ _--j
LD Temperatu re
FIGURE 9.11 Flow chart of the phytoplankton submodel of MS CLEANER. FF.

filter feeders: R. raptors; LD, ND, TD: light, nutrient. and temperature dependence: /11,1.
intermediate quantity; G, intermediate factor; Q, flow; SQ, rate of nutrient uptake: FP,
factor dependent on time (days).
349
10
o Lv
5 o GI
Eutrophic
Fu lakes
1
• I
...>
I
-s )2)
r
-G
N
I -p -W -z /
/
/
E 1/ /
/
/
jeMo /'// /
OJ
• Ma +
7
Es (1 )
,?>/, +L
a.. I~\ -~-H >" 8S0 I
0.5 J// I
1
2
:OF
:. T/// ./ '
Oligotrophic
I / /
lakes
k a .
s- /, /
• b·
//3 /'
·c ~..... -
/
•
A
/
/
Ta
/ /'" V
~;/ ,~
/
/
/
5 10 50 100 500 1000

Depth (m)
FIGU RE 9.12 Vollenweider plot of phosphorus concentrations: a, b, and c correspond
to removal of 90. 95. and 99 () (). respectively. of the phosphorus input to Fureso, Results
for 1972 are shown for: GI, Glumso So (Jorgensen e/ ai" 1973); Ly, Lyngby So (Smith,
1973); Es, Esrum So; G. Greifensee; P. Pfiiffiker See; B, Baldegger See; W, Lake Washing-
ton; Z, Zurich See; Ma. Miilaren; H. Hallwiler See; Bo. Bodensee; A. Ageerisee; V,
Viinern; Fu. Fureso; F. Furcso (1954); M. Lake Mendota; T. Turlersee; S. Lake Sebasti-
cook; Mo. Lake Moses; An. Lac d' Annecy; L, Lac Leman; Ta. Lake Tahoe, (I) and (2) are
the limits of the two lake conditions.
Imboden (1974) suggested a two-compartment model for phosphorus con-

tent. The model considers a stratified lake and includes input, output, and
exchange between hypolimnion and epilimnion, as well as sediment exchange.
Four coupled differential equations for dissolved and particulate phosphorus
are applied. The model has been improved (Imboden and Gachter 1978;
Imboden, 1979) by describing nutrient and biomass concentrations as con-
tinuous functions of time and depth and by replacing the first-order kinetic by
Michaelis-Menten kinetics. O'Melia (1974) and Snodgrass and O'Melia (1975)
developed a similar model, but did not include release of phosphorus from the
sediment; however, depth-dependent rates of turbulent diffusion were
considered.
Larsen et al. (1974) found that the Vollenweider and Snodgrass-O'Melia
models underestimated the actual amount of epilimnetic phosphorus, when
applied to Lake Shagawa in Minnesota. They then applied a slightly more
complex model consisting of a three-compartment epilimnetic model, which
350
includes algae as a sink for soluble reactive phosphorus and conversion of
particulate phosphorus to the soluble form. The basic equations for this model
are:
dPA PS
----crt = MYMAX(T)· LIGHT· KP + PS . PA
- (CONRI + SETTLl + Pw)· PA (9.3)
dPS PSIN PS
cit = ---vE - MYMAX(T)· LIGHT· KP + PS· PA
+ CONR2· PP + PwPS, (9.4)
dPP PPIN
cit = ~ + CONRl· PA - (CONR2 - SETTL2 + Pw)· PP, (9.5)
where
is the concentration of algal phosphorus [M L - 3],
PA
is the fractional reduction of MYMAX(T) in the epilimnion due
LIGHT
to the availability of light,
MYMAX(T) is the maximum specific growth rate of phytoplankton as a
function of temperature [T - I],
KP is the half-saturation constant for phosphorus [M L- 3],
CONRI is the rate constant for conversion of algal phosphorus to
particulate phosphorus [T - I],
CONR2 is the rate constant for conversion of particulate phosphorus to
soluble phosphorus [T- 1],
PP is the concentration of particulate (non-algal) phosphorus
[M L - 3],
PPIN is the rate of supply of particulate phosphorus to the epilimnion
[M T-IJ,
PS is the concentration of soluble phosphorus [M L- 3],
PSIN is the rate of supply of soluble phosphorus to the epilimnion
[MT- 1],
SETTLl is the rate constant for settling of algal phosphorus (correspond-
ing to a settling velocity of 0.02 m day - I),
SETTL2 is the rate constant for settling of non-algal particulate phos-
phorus (corresponding to a settling velocity of 0.04 m day-I),
T is the temperature,
VE is the volume of the epilimnion [L 3 ].
Lorenzen et at. (1976) developed a model consisting of two differential

equations only, one for soluble phosphorus and one for exchangeable phos-
351
phorus in the sediment:
dPS = PSIN K 2 · AREA· PSED _ K 1 . AREA· PS _ JL.
PS
dt VL + VL VL VL
(9.6)
dPSED K 1 · AREA· PS K 2 · AREA· PSED
dt VS VS
where
is the lake surface area [L 2 ],
is the rate of transfer of phosphorus to the sediment [L T- I],
is the rate of transfer of phosphorus from the sediment [L T- 1 ],
is the fraction of total phosphorus input to the sediment that is not
available for exchange,
PSED is the total concentration of exchangeable phosphorus in the sediment
[M L -3],
VS is the sediment volume [L 3],
Q is the outflow [L 3 T - I],
VL is the lake volume [L 3 ].
The purpose of the model is to predict long-term changes in lakes that have
undergone significant changes in loading rates. PSIN is therefore understood
as the annual loading of PS, Q the annual outflow, and K 1 and K 2 are measured
in m yr- 1 .
The equations can be solved analytically and the steady state solution of
PS is
(9.7)
A characteristic feature of this model is that, in spite of its simplicity, it

considers the sediment-accumulated phosphorus and that only a fraction of
the total phosphorus input to the sediment is available for the exchange process.
More complex models do not include this important property of the phosphorus
in the sediment, although it is of great importance for the long-term changes in
lakes because a substantial part of the phosphorus in a lake system is
accumulated in the sediment.
The parameters of this model are estimated by the following procedure.
When reasonably good data on loading rates and average aqueous and sediment
concentrations are known:
(1 ) K 3 is estimated.
(2) Since
PSIN - PS oo ' Q
K 1K 3 = PS· AREA ' (9.8)
K I can be calculated.
352
(3) K 2 is calculated from
(9.9)
as the ratio of steady state aqueous to sediment phosphorus concentra-

tions is given by (analytical solution)
PS
oo
--- (9.10)
PSED oo
The model was used on Lake Washington by applying data from 1941-1950
to calculate a consistent set of model constants, based upon K 3 = 0.6. K 3 can
be found on the basis of sediment analysis (a more detailed examination of
mud-water exchange of phosphorus was reported by Kamp-Nielsen, 1975). The
observations during 1955-1970, which showed that the phosphorus loading
increased up to 1964 and decreased thereafter, were well predicted by the model.
However, K 3 = 0.5 gave a better result (Figure 9.13).
Lappalainen (1975) improved Vollenweider's approach by considering the
state of a lake as a function of lake volume, discharge, and phosphorus input.
In this model a regression equation that relates the net sedimentation of phos-
phorus and the oxygen concentration of the hypolimnion is determined. The
model includes a relationship between the sedimentation of phosphorus and
0.1 0 .----~-----.---,----------,--_____,_--___,__--r__-____,
0.08
0.06
l pJ
0.04
0.02
1940 1950 1970 1980

FIGURE 9.13 Calculated and observed (.) annual average total phosphorus con-
centrations (mg 1- I) in Lake Washington.
353
volume, discharge, and phosphorus input. This sedimentation submodel and
the regression expression were used to construct a model for the prognosis of
the oxygen concentration of the hypolimnion, which is used to determine the
boundary phosphorus input, comparable with loads given earlier in the
literature.
Jolankai and Sz6ll6si-Nagy (1978) constructed a simple model of algal
dissolved reactive phosphorus to describe eutrophication in the Bay of
Keszthely, Lake Balaton. The model is based on the specific condition that in
Lake Balaton the main direct source of phosphorus is the sediment. The model
considers three phosphorus state variables: phosphorus in water, in sediment,
and in algae. Temperature and light limitations are not considered.
9.2.3. Multiconstituent Models

The models mentioned above are often described as nutrient budget models.
They all deal with eutrophication in gross terms, i.e. with only one or two layers,
long time scales, and only one governing nutrient. These models give no atten-
tion to short-term hydrodynamics, ecosystem dynamics, or spatial resolution
of quality changes within the impoundment. The result is that such models
have only limited utility in predicting impoundment response to pollution
control strategies, particularly those that may be either of local impact or
short-term.
More comprehensive models of eutrophication have been proposed and
several have been implemented. The models are based upon the principle of
mass conservation. The general mass conservation equation, exemplified below
for a water element j in a one-dimensional model, is applied:
advection diffusion volume change
de
± VJ dt · (9.11)
sources and sinks
Thomann et al. (1975) developed three models with different levels of detail.
Modell conceptualizes the impoundment as a three-layer system (epilimnion,
hypolimnion, and benthos) and considers phytoplankton and zooplankton
dynamics. Model 2 provides additional vertical resolution with seven layers
and considers temperature, chemistry, and sediment interactions. Model
3 is far more comprehensive as it provides additional spatial resolution with
up to seven layers, 67 segments, and 15 state variables. A future model with as
many as 5000 compartments (segments· state variables) is envisioned.
354
Validation of model 1 indicates generally good agreement with observations
for such state variables as chlorophyll a, zooplankton, carbon, four forms of
nitrogen, and phosphorus. Unfortunately model 2 was only studied analytically
and was not validated against observations, and model 3 was only tested pre-
liminarily against selected observations from the field but was not validated.
The Chen-Orlob model (1975) was extended by Chen et al. (1975) for
simulation of Lake Ontario. It included mass balance equations for 15 different
classes of biotic and abiotic substances. Four groups of phytoplankton, four
groups of zooplankton, and four groups of fish in three life stages were repre-
sented. Furthermore, the model divided the lake into segments and layers, for
which separate mass balance equations could be written. Details of the model
are given in the appendix. Implementation ofthe model required an independent
determination of the flow field in three dimensions and estimates of empirical
diffusion coefficients. The hydraulic inputs were derived from a hydrodynamic
model attributed to Bennett (1974). Details of the hydrodynamic model are
not available, except that it is of orthogonal form with 715 surface elements and
seven layers. It is wind-driven and considers effects of temperature on circula-
tion. Successful test simulations for the months of 1972 were carried out with
41 surface elements and seven vertical layers of varying thickness, a total of 209
elements. Results indicated reasonable performance, but no validation was
carried out.
The 1979 version of this lake model required 9 s of Univac 1108 CPU time
per day of simulation (~t = 1 day), which is little compared with the hydro-
dynamic program, which required 120 s of CPU time per day of simulation.
The Chen-Orlob model, in a version known as EPAECO, has also been applied
to a Finnish lake (Kinnunen et aI., 1978).
Patten et al. (1975) developed the Lake Texoma Cove model, which attempts
to consider every conceivable aspect of the cove ecology. Its most distinguishing
feature is that it is an ecologist's model-an ecosystem description drawn by a
relatively large number of scientists under conditions of prolonged, intensive
interaction. The model is further distinguished by its linear construction
(although a few minor nonlinearities occur). The forcing functions are time,
temperature, solar radiation, rain, wind, current, and water level.
The primary producer submodel contains nine state variables: small,
medium, and large phytoplankton, blue-green algae, floating algal mats,
attached algae, aufwuchs, and submergent and emergent vascular plants. The
zooplankton submodel includes those animals retained by a 64 p.m plankton
net, excluding fish larvae and dipteran larvae. Pupal zooplankton was divided
into two groups: small (less than 0.75 mm) and large (0.75 mm or greater).
The benthic invertebrate submodel consists of suspension feeders, deposit
feeders, predators, and scavengers. The vertebrate submodel includes larval
fish, fingerlings, filter-feeding fish, bottom-feeding fish, minnow-like fish, car-
nivorous fish, turtles, and herbivorous and carnivorous harvesters. In the
355
decomposer submodel nine compartments were selected: dissolved organic
matter, particulate organic matter, plant carcasses, animal carcasses, organic
sediment, nitrate and nitrite nitrogen, phosphorus, carbon dioxide, and dis-
solved oxygen. Together, the submodels treat 33 state variables. A connectivity
matrix for the state variables is shown in Figure 9.14. As indicated by the
authors, the model cannot be considered as having been validated. No time-
series data were at hand for comparison with model results. The model repre-
sents one extreme by giving a comprehensive mathematical description of the
ecology.
Di Toro et al. (1975) developed a phytoplankton-zooplankton-nutrient
interaction model for western Lake Erie. They considered seven segments
and seven state variables: inorganic phosphorus, phytoplankton, nitrate
nitrogen, ammonia nitrogen, organic nitrogen, zooplankton, and organic phos-
phorus. Thus the model has 49 compartments. The advective flows are
established primarily by the inflow of the Detroit River and its passage through
the Western Basin into the Central Basin (Figure 8.8, p. 311). In addition to
advective flow, it is necessary to assess the magnitude of mixing of the flow
between adjacent segments. This is accomplished by use of conservative tracer,
the chloride ion.
The Monod kinetic is used to express the rates of growth of phytoplankton
and zooplankton, while the transformations from organic nitrogen to ammonia
and from organic phosphorus to orthophosphate are described by first-order
temperature-dependent kinetics. The constants and their temperature-depen-
dences are found initially by the use of available laboratory experimental data
to set the probable ranges of constants (Di Toro et aI., 1971). Then detailed
comparisons are made between observed and computed data in order to fine-
tune the value (calibration).
The model was finally validated using a composite set of"data from the year
1930. This year was chosen for two reasons: (i) there is a significant data base
and (ii) the limnological conditions that existed in 1930 were far removed from
those in 1970. Reasonably good agreement with observed data resulted for all
systems. However, there was some discrepancy between the observed and the
predicted zooplankton concentrations.
Bierman (1976) constructed a 14-compartment model of inner Saginaw Bay
(Figure 9.15). The idea was to apply a spatially simplified version of the model,
which would be incorporated at a later stage in a spatially segmented version.
A unique feature of the model is that all growth is considered to be a two-step
process, involving separate nutrient uptake and all synthesis mechanisms. This
is in accordance with an increasingly large body of experimental evidence
indicating that the mechanisms of nutrient uptake and all growth are actually
quite distinct (Dugdale, 1967; Fuhs, 1969; Eppley and Thomas, 1969; Fuhs
et al., 1971; Caperon and Meyer, 1972a, b; Droop, 1973; Halmann and Stiller,
1974; Nyholm, 1975).
356
From
1 2 3 4 5 6 7 8 9 10 11 12 13 14 1516 1718 19 20 21 22 2324 2526 27 28 2930 31 32 33
1 0 0 00 00
2 0 0 00 00
3 0 0 00 00
4 0 00 00
5 0 00 00
6 0 00 00
70 00 0 00 00
8 0 00 00
9 0 00
100 0 0 0
1100 0 0 0
120 00 00 000 00 00 00 0 00 0
130 00 00 000 000 00 0 00 0
140 00 00 00 00 00 000 00 0 0
150 00 00 000 00 00 000 0 0 0
160 00 00 00 00 00 000 0 0
To 17 0
00 000 00 00 000 000 0
18 000 00 0 0
19 00 000 000
20 00 00 000
210 0 0 0
22 0 00 0 0 0
23 0 0 00 0
24
250 00 00 00 00 00 00 00 00 000 0 0 0
26 •
27
••• •• •• 0 00 00 00 00 00
••
•
• • 0
28 00 00
29 00 0
300 00 00 00
00 00 00 00 00 00
00 00 00 00 00 000 00
•• 0
310 00 00 00 00 00 00
320 00 00 000 00 00 00 000 000 00 0
330 00 00 000
FIGURE 9.14 Connectivity matrix for the Lake Texoma Cove model. • indicates a
time-varying coefficient and 0 denotes a constant coefficient. There are 992 (32·32 - 32)
possible interactions, 335 are non-null. The compartments are; I, small phytoplankton; 2,
medium phytoplankton; 3, large phytoplankton; 4, blue-green algae; 5, floating algal mats;
6, attached algae; 7, aufwuchs; 8, submergent vascular plants: 9, emergent vascular plants;
10, small zooplankton; II, large zooplankton; 12, fish eggs and larvae; 13, fingerlings; 14,
filter-feeding fish; 15, bottom-feeding fish; 16, minnow-like fish; 17, carnivorous fish; 18,
turtles; 19, herbivorous vertebrate harvesters; 20, carnivorous vertebrate harvesters; 21,
suspension-feeding invertebrates; 22, deposit-feeding invertebrates; 23, invertebrate
predators; 24, invertebrate scavengers; 25, dissolved organic matter; 26, particulate
organic matter; 27, plant carcasses; 28, animal carcasses; 29, organic sediment; 30,
nitrate and nitrite nitrogen; 31, phosphorus; 32, carbon dioxide; 33, dissolved oxygen.
357
Higher predators
- - -- - - -- - - - ------------1
.-------.., i---------~:
Blue-green I Blue-green I I
algae (not I
I aIgae II
I L--..
Green algae
nitrogen- : (nitrogen- ::
fixing) I fixing) I :
L..- --' ~ JI
____ J
Available
silicon
FIG U RE 9.15 Principal constituents of the Saginaw Bay eutrophication model.
The specific phosphorus uptake in the Bierman model is described as a

function of the balance between extracellular and intracellular dissolved
phosphorus:
uptake rate = maximum uptake rate· j(T)I(l)
. (I + P~I . PCA - 1 + PKII . PCM)

(9.12)
PCA = PCAMIN . exp(PC/PO - 1), (9.13)

where
is the uptake constant [L 3 mol- 1],
PKI
PC
is the total amount of phosphorus per cell (PCA plus internal
storage) (mol ceU- 1),
PO is the minimum stoichiometric level of total phosphorus per cell
(mol ceU- 1 ),
PCAMIN is the minimum value of PCA [mol L - 3],
PCA is the internal dissolved phosphorus concentration [mol L - 3],
PCM is the external dissolved phosphorus concentration [mol L -3].
358
FIGURE 9.16 Principal constituents of the Lake Michigan food web model.
An identical approach is used for the nitrogen uptake. The state variable
equations of this model are given in the appendix.
One of the interesting features of the work of Bierman is that the model has
been used to examine which factors are significant for blue-green algae blooms.
It is often important to control the blooms, since these algae cause formation
of a blue-green foam on the water surface and produce toxic substances.
Freedom from grazing and lower sinking rates seem to be the most important
of the competitive advantages, although blue-green algae have a slower maxi-
mum growth rate than all other classes of algae. Relatively small differences in
sinking rates between blue-green algae and other classes showed a significant
effect. Furthermore, it seems essential that blue-green algae have a higher phos-
phorus uptake rate (approximately twice as high) and reach half-saturation at
significantly lower extracellular phosphorus concentrations than other phyto-
plankton classes.
Canale et al. (1976) used a complex food web model for Lake Michigan; the
food web is shown in Figure 9.16. The model distinguishes between raptors,
selective fiIterers, and nonselective fiIterers and uses different equations for the
feeding rates of the three classes of zooplankton. For raptors:
feeding rate = MYZMAX· f(T) LpreyCl (9.14)

Lprey C 1 + K Z '
where
Lprey c 1 is the sum of concentrations of all states that can serve as food for
raptor state Z,
359
MYZMAX is the maximum growth rate of zooplankton [T-lJ,
KZ is the half-saturation concentration for grazing.
Selective filterers obtain food by grazing, but have developed an ability to
vary their filtering rate. A proposed equation for this mechanism is:
KMFM . PHYT + KFLM
feeding rate = MYZMAX, J(T), PHYT + KFLM ,PHYT, (9.15)
where
KFLM is the food level [M L - 3 J,
KMFM is the minimum filtering rate multiplier,
PHYT is the concentration of phytoplankton [M L - 3].
A formulation of the feeding mechanism of nonselective filterers is:
feeding rate = MYZMAX, J(T) L Cl' (9.16)
prey
The nonselective filterers cannot lower their filtering rate when the plankton
content of filtered water increases. Therefore, they operate below maximum
possible efficiency. A simple but unconfirmed formulation suggested by Canale
et at. is:
, 'I'
aSSlml atlOn effi clency
' =
All N " c A24 A24 (9.17)
L.,prey 1 +
CON2
(9.18)
= CONI PHYT + CON2'
where All, A24, CONI, and CON2 are constants.

For the other classes, however, a constant assimilation efficiency that is
related to the growth rate is used:
growth rate = assimilation efficiency, feeding rate. (9.19)
The rate at which state z is consuming prey in state k is, in general, given by
predation of k by z = preference of z for k ' feeding rate, C z • (9.20)
The formulation of the preference factor involves a simple assumption: the
preference of z for k is proportional to the product of the electivity of z for k
and the concentration of k, The electivity (X~ is defined as the fraction of the
diet of z that would be composed of food species k, if all food species were
present in equal concentrations:
preference of z for k = L(X~(X~CC .

i
k
1
(9.21 )
The web of electivities incorporated into the lower food web model is shown in
Figure 9.17; Table 9.1 presents the state of the system.
360
Large
herbivorous
cladocerans
FIGU RE 9.17 Web of eleclivities.
The model applied by Canale et al. has 25 water quality variables and two
vertical layers. The model is calibrated, but cannot be considered validated.
Simulations were conducted with the calibrated model to examine its behavior
when phosphorus concentration and alewife predation became higher than
normal.
A multispecies model has also been developed by Ikeda et al. (1979). The
approach of parameter estimation in this model is interesting, as the main
objective of the model is confined to determining the parameter values that
generate a similar behavior of plankton dynamics rather than finding a best
fit to the observed data. This probably indicates the state of the art of multi-
species models: the model can be calibrated to obtain a similar behavior, but
most data sets are not sufficient in number and accuracy to allow a very good
fit between model and observed values.
J0rgensen (1976) applied a two-stage algal growth process in his lake model
and compared the results of this model with a model based upon general
Monod kinetics. It was concluded that the J0rgensen model gave a better
description of the system response to seasonal variation in nutrient loading
than the generally applied Monod kinetic. Another characteristic feature of
361
TABLE 9.1 The State of the System.
Description Classification
Lt'I>lodora and Polyphemus

} Raptors
Cyclops
Cyclops nauplii
Diaptomus nauplii
Limnoea/anus and Epischura nauplli
Dial'tomus
Liml1o('(J/wlIIs and Episcl11Im
Daphnia
Bosmina and H%pedium
I Selective
filterers
Nonselective
} filterers
Zooplankton
Small diatoms
Large diatoms
Blue-greens
Greens
Detrital nitrogen
Dissolved organic nitrogen
} Ni,mgoo
Ammonia
Nitrate
Detrital phosphorus Epilimnion
Dissolved organic phosphorus } Phosphorus nutrients
Dissolved inorganic phosphorus
Detrital silicon
} Silicon
Dissolved silicon
Total inaccessible nitrogen Nutrient mass held
Total inaccessible phosphorus in hypolimnion
Total inaccessi ble silicon } and sediments
the model is that it contains a more complex submodel for the exchange of
nutrient between sediment and water than was applied in the other models
mentioned. The model distinguishes between exchangeable and nonexchange-
able phosphorus and describes the release of phosphorus as a two-step process:
exchangeable interstitial orthophosphorus
---> ---> ,
phosphorus phosphorus in water phase
with different rate constants for aerobic and anaerobic conditions. Simulation of
reduced phosphorus loading demonstrated that the inclusion of this more com-
plex sediment submodel gives a more pronounced response than the use of a
simple first-order kinetic for phosphorus release, which is used in most models, if
the sediment is considered at all. This is not surprising as a substantial part ofthe
sediment phosphorus cannot be exchanged. The submodel used for nutrient
release is in accordance with comprehensive studies of these processes carried
out on sediment cores from several lakes (J0rgensen et al., 1975; Jacobsen and
J0rgensen, 1975; Kamp-Nielsen, 1975).
The J0rgensen model also considers denitrification. Its application to a
Swedish lake study (S6dra Bergundasj6n) required the modeling of nitrogen
362
fixation (Jorgensen et ai., 1981, 1982). The model, which is a one-layer model
with 17 state variables in the last version (Jorgensen, 1976), has been validated
in studies of two Danish lakes, Glumso So and Lyngby So, with good results. An
examination of several versions of the model is published in the same paper by
Jorgensen et al. (1978). It concludes that distinguishing between two classes
of zooplankton and the introduction of a biologically active layer on top of the
sediment are not important for the validation of the model, and that the use of
daily measured irradiance data rather than average values is essential. The
two models compared are shown in the appendix.
Richey (1977) developed a phosphorus model of Castle Lake, California.
The model includes the following state variables: dissolved inorganic P, dis-
solved organic P, phytoplankton P, bacterial P, zooplankton P, polyphosphate
P, detrital P, ferric P, and sediment P. The changes of processes in pools over
depth and time are considered. The distribution, formation of complexes, and
precipitation of phosphate species as functions of pH and iron concentration
are modeled by the following system of equations:
[POl-J = PT (1 + [H+Jl0- 12 .3 + [H+JIO- 19 .5 + [H+J 3 1O-21.7)

[HPoi-J = P T (1 + [H+Jl0- 12 .3 + [H+JIO- 7.2 + [H+J 21O-9.4)
[H 2POiJ = PT (1 + [H+JIO- 2.2 + [H+JIO- 72 + [H+YIO-19.5)
[H 3 P0 4J = PT ( 1 + [H+Jl0- 2.2 + [H+J 21O-9.4 + [H+J 3 1O-21.7)
[HP0 4SJ = lO-ll[H+J [Fe3+J (HP0 4 removed)
[HP0 4PJ = [HP0 4S][HPOi-J (HP0 4 percentage removed)
[DIPJ = 3.1 . 107(PT - [HPOi-J)(1 - [HP04SJ [HPoi-J)
(phosphate left in solution).
The model CLEANER (Scaviaand Park, 1975; Scavia et ai., 1976) was originally
applied to Lake George, but has been developed for application to a variety
of impoundments, including Sarasota Lake (Florida), Loch Leven (Tayside,
Scotland), several Scandinavian lakes, Lake Balaton (Hungary), and lakes in
Italy and Czechoslovakia. A new version of the model, MS CLEANER, includes
40 state variables and provisions for horizontal and vertical transport. It can be
applied to littoral, pelagic, and profundal zones. A typical solution for a strati-
fied reservoir with 34 state variables requires about 63 s of Univac Ill0 CPU
time for simulation of an annual cycle. The model seems to be applicable to
many case studies that require calibration of only a very limited number of
parameters.
Nyholm (1978) developed an ecological model for shallow lakes, using
seven state variables: phytoplankton, phosphorus in phytoplankton, nitrogen
in phytoplankton, available phosphorus, available nitrogen, detrital phos-
phorus, and detrital nitrogen. The model does not include sediment nitrogen
BOD exertion Dissolved I Reaeration
~ oxygen
model
NH 4 oxidation
~ Benthic
BOD BOD
decay
N0 2 oxidation
Photosy.nthetic Algal
production respiration
Algal Algal uptake -'I
release
.. Anaerobic
release
Algae
Algal release
model
Algal uptake Light attenuation
---------,---------'-----------
Ammonium (Anaerobic) Grazing Uptake

formation denitrification
1
N"mgee Zooplankton I Phosphorus Sediment
P
L release model I release I
FIGURE 9.18 Processes considered in the limnologicaJ model of Baca and Arnett (1976).
l.;J
0"-
l.;J
TABLE 9.2 Lake Model Characteristics.
Model Number of State Nutrients Number of Number of Constant Stoichiometries Calibrated (Cl Numberof
w Variables per Layer Considered Segments Dimensions (OJ (CSjor Independent and/or Case Studies
0\ or Segment or Layers (Ll Nutrient Cycle (NC) Validated (Vl in Literature
.j::o.
Vollenweider I peN) I IL CS C+V Many

Imboden 2 P I 2L,ID CS C+V 3
O'Melia 2 P I 10 CS C I
Larsen 3 P 1 IL CS C I
Lorenzen 2 P I IL CS C+V I
Thomann I 8 P,N.C 1 3L CS C+V I
Thomann 2 10 P, N,C I 7L CS C I
Thomann 3 15 P, N.C 67 7L CS I
Chen and Orlob 15 P, N,C Many 7L CS C At least 2
Patten 33 P, N,C I IL CS C I
Di Toro 7 P. N 7 lL CS C+V I
Bierman 14 P, N, Si I IL NC C I
Canale 25 P, N, Si I 2L CS C I
Jorgensen 17 P. N,C I 1-2L NC C+V 3
CLEANER 40 P, N, C. Si Many Many L CS C Many
Nyholm 7 P, N 1-3 1-2L NC C+V 13
365
and sediment phosphorus, but describes the release of sediment nutrient as a
function of temperature and sedimentation rates. The model accounts for
variable internal storage of nutrients in the cell and describes the specific
growth rate as a function of intracellular nutrient level (eqns. 13, Table 4.1).
The same model has been used to simulate 13 lakes in Denmark, some of
which have been described by two or more compartments. Most parameters
in the model were common to all simulations; a few key parameters were
allowed to vary within a narrow range and if deviations from this range were
necessary a qualitative explanation could usually be given. The simulations
generally gave reasonably good results, especially for lakes with short retention
times. The reason for the poorer agreement between model and observations
for lakes with longer retention times is probably that the model describes the
internal reactions of the lakes very crudely. It is of importance to have a good
submodel of the sediment-water exchange of nutrient when the internal loading
of nutrient is dominating, i.e. for lakes with long retention times.
The model of Baca and Arnett (1976) includes a comprehensive phosphorus
and nitrogen cycle with a few distinctions, related primarily to the recycling of
these nutrients between sediment and water. The model is presented in Figure
9.18 and the governing equations are shown in part (5) of the appendix.
Parker (1978) introduced a model that takes into account the spatial hetero-
geneity of phytoplankton and dissolved nutrient by nutrient-phytoplankton
interaction. Dubois (1975) and Parker (1976) coupled the dynamics of predator-
prey populations with the consequences of physical processes in a turbulent
medium. Halfon and Lam (1978) simulated the spatial movements of phos-
phorus by using the computed currents from a three-dimensional hydrodynamic
model. A biological submodel describes the phosphorus dynamics and primary
production in each grid cell of 20 km . 20 km in Lake Superior.
As this survey shows, the lake models in operation cover a wide range of
complex circumstances, although the list is not complete. Table 9.2 summarizes
some of the model characteristics that can be found in the literature.
9.3. SOME ILLUSTRATIVE CASE STUDIES
9.3.1. Application of Case Studies

During the preparation of this survey of models, it was found that although a
great number of aquatic models are described in the literature, only a limited
number of lake models have been calibrated and validated. J0rgensen (1978)
has paid attention to this problem and encouraged studies in which ecological
models are tested with field data to determine their applicability to environ-
mental management. Two case studies have been selected to illustrate the
applicability of ecoiogicallake models.
366
9.3.2. Ovre Heimdalsvatn, Norway
The lake has a surface area of 0.775 km 2 and is 3 km long; the maximum depth
is 13 m and the mean depth 4.7 m. The lake is subalpine and ultra-oligotrophic.
It is a natural lake and is a good example, because the retention time is 2 months
but can be as short as 2-3 days during the spring runoff. A version of CLEANER
without intracellular storage of nutrients was used but it predicted a phyto-
plankton peak that was 3 weeks before the observed peak. However, MS
CLEANER yielded an excellent fit to the data with an intracellular nutrient
submodel. These observations are in accordance with the experiences published
by Jl2lrgensen (1976).
This case study shows that MS CLEANER is capable of modeling the
combined effect of light, temperature, and nutrient in- and outflow. Figure 9.19
illustrates this capability for phytoplankton and zooplankton; these results
0.150
•
0.125 !\
.....
0.100
l
•
..r:::.
OJ
.iii
~
~ 0.075
"0 •
OJ
(/)
• •
E
(/)
0.050
.Q
co
• •
0.025 c:i
•••
• .0
...o.~OO"·~·D·. • •• cr'"
0,----,----,---,---L...L---"---,--'-'--'--.J-.L-L...L---"---,---'--..J'----'---L...1....-L...L--'---L--!.......J
1.00 141.00 281.00 365.00
Time (days)
FIGU RE 9.19 Simulation of biomass in Ovrc Heimdalsvatn. Norway, 1972. The results
arc for predicted (0) and observed (.) phytoplankton and for predicted (D) and observed
(.) zooplankton (from Desormeau, 1978). Unpublished data are by courtesy ofP. Larsson.
367
were obtained after the calibration of two parameters only. For all other
parameters general values were applied.
9.3.3. Glums0 S0, Denmark

Glums0 S0 is a small (250000 m 2 ), shallow lake (maximum depth 2 m) with
advanced eutrophication. Four versions of the J0rgensen model (details are
given in part (4) of the appendix) were calibrated for the 1973-74 data and
validated for the 1974-75 data. The validation was based on a comparison of
soluble phosphorus, soluble nitrogen, phytoplankton concentration, zoo-
plankton concentration, and productivity. The validation shows that version
II is preferable to version I.
The details in the appendix cover both versions. Version II uses measured
irradiance data while version I uses average data based on measurements over
the last two decades. Version III, which accounts for light adaptation, and
version IV, which considers diurnal variation of irradiance, gave poorer valida-
tion than version II. The result of the validation is summarized in Table 9.3,
which shows Y and the correlation coefficient for comparison of predicted
and measured values:
(
"~_ (Xi _ Xi )2jx m ) 1/2
Y = L,I-l l m
(9.22)
n '
Xl being the predicted values of the model, X m the measured data, xm the mean
of the X m , and n the number of compared pairs of figures. Figures 9.20-9.24
illustrate the validation by comparison of predicted and measured values.
3.0
OJ 2.5
E
~ 2.0
.... -0,
o
~
en
1.5
o
.L:
Q. 1.0
-co
o
f- 0.5
15110 1/12 1/2 1/4 1/6 1/8 1110

1974 1975
FIGU RE 9.20 Concentration of total phosphorus in Glumso So: observed data, X;
calculated data, D.
368
TABLE 9.3 Results of the Validation of Four Versions of the J0rgensen Model
Applied to Glumso So, Denmark.
Model Version Y Correlation Coefficient
I 0.574 0.78
II 0.420 0.79
III 0.644 0.74
IV 0.768 0.63
2.5
U'
'c -;,
~E
.-
g; :::J
_Q) 0...
.o.c
:::J c.
- (f)
o 0
(/).c
c.
15/10 1/12 1/2 1/4 1/6 1/8 1/10

1974 1975
FIGURE 9.21 Concentration of soluble inorganic phosphorus in Glumso So: observed
data. x ; calculated data, O.
15
,-
Cl
E 10
c
Q)
Cl
.......
0
'c ;;P,
5
....rc0
I-
I I
15/10 1/12 1/2 1/4 1/6 1/8 1/10

1974 1975
FIGURE 9.22 Concentration of total nitrogen in Glumso So: observed data, x;
calculated data, O.
369
OJ 6
E
ai 5
OJ
o...
'~ 4
,~
fi5 3
...OJ
g 2
Q)
:0
:::J 1-
o
(/)
15110 1112 1/2 1/4 1/6 1/8 1110

1974 1975
FIGURE 9,23 Concentration of soluble inorganic (nitrate and ammonia) nitrogen in
Glums0 S0: observed data, x; calculated data, D,
80
I
OJ
E 60
c::
0
+-'
-"
c:: 40
co
a.0
+-'
~
>-
a.. 20
15110 1/12 1/2 1/4 1/6 1/8 1/10

1974 1975
FIGURE 9.24 Concentration of phytoplankton in Glums0 S0: observed data, x (on
the basis of suspended matter, 1-80 jlm) and 0 (on the basis of chlorophyll); calculated
data, D,
370
9.4. CONCLUDING COMMENTS
9.4.1. Discussion and Conclusions

The very nature of the modeling exercise, which tends to aggregate, average,
and smooth over the biological subtleties, leaves the more rigorous aquatic
biologist somewhat disconcerted. He would prefer to concentrate efforts on a
more correct representation of biological interactions, the kinetics of varying
life stages, shifts in grazing preferences (Canale et al., 1976), ecological in-
stabilities, etc. Some trade-offs are necessary simply because the model is an
approximation of the real system. These seem to have occurred either by
simplification of the aquatic ecosystem or by simplification of the circulatory
and exchange processes of the impoundment (compare with Table 9.2). How-
ever, the scope is not to include more and more details and to build as complex
a model as possible, but rather to build a model that gives a quantitative de-
scription of what is in focus-to meet the aims of the model. There is no such
thing as a general ecological lake model, but in every case study the goals and
the resources available must be balanced so that the right model can be selected
(see also Chapter 2).
The experience with CLEANER and MS CLEANER, however, has shown
that some generality exists. These models have been applied in many case
studies and have obviously a certain general applicability, although some case
studies (e.g. Lake Balaton) have demonstrated that even these complex models
cannot be used completely in general. The reason is probably that although
the models include a reasonably wide range of processes they cannot include
all possible processes. In the Lake Balaton study the crucial process is probably
the resuspension of sediment caused by wind stirring the body of water, and
this process is not included in CLEANER or in MS CLEANER.
In this context it must be stressed that a model cannot be better than the
data on which it is based. A very complex model will contain more parameters
to be calibrated, requiring more observations. Furthermore, validation will
require another independent set of observations. It is therefore not surprising
that the very comprehensive models are not validated or sufficiently well
calibrated.
Model structure depends also on the accuracy required, so it is important
to consider the accuracy with which it is possible to simulate a specific eco-
system. This brings up the question of how much we can rely on the observed
data. It is assumed that ecological observations normally will have a standard
deviation of 10-25 %, which must be taken into consideration when the ac-
curacy of the model is estimated.
Many lake models have been calibrated to fit limnological data collected
with the usual frequency of two sets of measurements per month. Such data
can hardly be used to describe the dynamics of the system, as demonstrated in
371
c
o
.~
E
Q)
u
c
o
u
c
£
~
c
CD
C.
o
>-
..c.
0..
1 April 15 April 1 May 15 May
FIGURE 9.25 Comparison of measuring frequencies: x every second week; 0 every

second day. The two frequencies give completely different pictures of phytoplankton
dynamics. and will lead to different calibration results.
Figure 9.25. The data that are used for calibration should always be carefully
collected with a frequency corresponding to the dynamics of the calibrated
system or subsystem (Jorgensen, 1979; Jorgensen et al., 1981). Only little work
has been carried out to compare the results of different models, but it seems
more important to include a detailed description of the nutrient uptake by
phytoplankton and the sediment-water exchange of nutrient than to implement
details of the higher trophic levels (Jorgensen, 1976; Jorgensen et al., 1978;
Nyholm, 1978). Nevertheless, acceptable validation results were obtained by
several studies. This does not imply that the models can be used generally;
rather, they were selected properly to solve specific problems.
A determination of sufficient model complexity enters the modeling process
at two stages (Beck, 1978):
(1) during the initial stage, when the analyst must choose a certain level
of complexity before attempting to verify the model against field data;
and
(2) during the final phases, when the analyst must decide whether the model
has been verified and has sufficient complexity for its intended
application.
Jl2Jfgensen and Mejer (1977) suggested the use of a quantitative index for the
selection of model complexity. The idea is to use a concept of sensitivity for
identifying the model structure. Basically it is an inverse "submodel sensitivity,"
called the ecological buffer capacity, which measures the influence (sensitivity)
that additional suggested submodels have on a particular state variable (e.g.
372
phytoplankton concentration for eutrophication models) to see whether any-
thing is changed by increasing the complexity.
Tapp (1978) examined and compared the use of simple and complex eutro-
phication models. He concluded that simple models can be used for first-
approximation analysis, but where data exist to establish a basis for a more
complex model these should be used. This conclusion is in accordance with
the state of the art (10rgensen, 1979).
Lake modeling has developed along two lines. The development from
CLEAN to CLEANER to MS CLEANER seems to aim at a general model
that can be used on any new case study after minor changes and a calibration
of only a few crucial parameters. The other development is to start with
knowledge of the lake ecosystem to construct a model that is balanced in
complexity. The experience gained from previous case studies is applied in
selecting the model for each new study. A general uncritical application of a
complex model seems dangerous, but if the models are used with some critical
sense and the necessary modifications and calibrations are carried out from
case to case, it seems a proper course to follow. Improvements will evidently
result from this development, as the experiences gained steadily are included
in the models.
However, our knowledge of ecosystem processes is limited and submodels
included in a complex model may be valid in one case study but give wrong
results in many others. This is avoided by using the other strategy, whereby
the characteristic features of the modeled ecosystem form the basis for a more
or less specific model, which will contain far fewer parameters and be more
certain in the sense that the model components are selected in accordance with
knowledge of the ecosystem. The amount of data required for calibration
of the general model seems to be higher than for the more specific model pro-
vided that the more general model is used critically, which implies that it must
be calibrated and to a certain extent validated. Blake and Gentil (1979) pub-
lished an interesting sensitivity analysis of a discrete lake model, and concluded
that a considerable amount of data is needed to calibrate a model for prediction
purposes.
The conclusion of this discussion is that both approaches should be en-
couraged. Both developments will probably lead to a better understanding of
the lake ecosystem and to better predictive models with a wider applicability.
There are impoundment models available today that are able to give an
acceptable quantitative description of the eutrophication process. Except for
some special cases, where phosphorus (or nitrogen) is limiting and a more
simple model might be sufficient, it seems that a management model must in-
clude the main biogeochemical mass flows, including the sediment-water
exchange processes (J0rgensen and Mejer, 1977), as well as a reasonably ac-
curate hydrodynamic description. Furthermore, algae succession can be
described fairly accurately (Ahlgren 1973a,b; Bierman, 1976; Park et al.,
373
1979b), although more knowledge of parameter values for different algal
species or classes is desirable.
9.4.2. Future Research Needs

Several problems are still unsolved, and further research is needed. In particular,
it is necessary (1) to examine the sensitivity of the models to the use of different
submodels; (2) to gain more experience in the application of models that
include ecology as well as hydrodynamics; and (3) to validate the existing
models, preferably over a period that includes a change of loading.
Present-day models are based upon an ecological structure of lake eco-
systems that might limit the possibility of using the models as predictive tools,
since the impact of pollution might require the development of a different
ecological structure. Future models will probably attempt to include such
changes in ecological structure. Some promising but primary approaches have
already been published, using catastrophe theory (Duckstein et aI., 1977;
Dubois, 1979), thermodynamics (Jl!Jrgensen and Mejer, 1979; Mejer and
Jl!Jrgensen, 1979; Mauersberger, 1979), or self-organization and adaptation
(Straskraba, 1979). The next generation of models will probably contain some
basic ecological principles evolved from these efforts, which implies that
changes in structure can be predicted.
APPENDIX. FIVE LAKE MODELS
(1) Model by Bierman

State variables
PSA(L), NSA(L)moles of phosphorus and nitrogen per milligram cell dry
weight for phytoplankton, L,
PCM, NCM, SCM moles of phosphorus, nitrogen, and silicon per liter of
solution,
A(L) milligrams dry weight per liter of phytoplankton, L,
Z(K) milligrams dry weight per liter of zooplankton, K,
TOP, TON, TOS moles of total unavailable phosphorus, nitrogen, and
silicon per liter of solution.
For each state variable the model equations are written in the form of a mass
balance differential equation:
rate of change of state variable = rate of change due to water circulation, Q
+ rate of change due to interactions in system volume, V
374
State variable equations
For each phytoplankton type, L, the rate of change of intracellular phosphorus
is given by:
dA(L) . PSA(L)
V dt = Q(ABD(L)· PSABD(L) . A(L)· PSA(L))
+ V A(L)· RIPM(L)· f(T)f(l)· 0.322· 1O-4(mol mg- 1 )
.C+ PKl(~)' PCA(L) - 1 + PKl~L)' PCM)

- V A(L) . PSA(L) . (RAGRZD(L) + RLYS(L)
. T. TCROP ASINK(L)) (A1.l)

+ DEPTH .
Expanding by the chain rule for derivatives gives:
V dA(L)· PSA(L) = V(A(L) dPSA(L) + PSA(L) dA(L)). (A 1.2)

dt dt dt
For each phytoplankton type the rate of change of biomass is given by:
V d~~L) = Q(ABD(L) - A(L))
+ V A(L)(SPGR(L) - RAGRZD(L)
- RLYS(L). T· TCROP - ASINK(L)) (AU)

DEPTH .
In (A 1.3) it is assumed that the contribution to the derivative due to changes

in intracellular phosphorus is negligible.
Setting the right-hand sides of (A 1.1) and (A 1.2) equal and substituting
(AU) gives the following equation for the state variable PSA(L):
V A(L) dPSd~(L) = Q. ABD(L) . (PSABD(L) - PSA(L))
+ V A(L)· RIPM(L)· f(T)f(l) -0.322.10- 4
.C+ PKl(L. PCA(L) - 1 + PKl~L)' PCM)

- V A(L)· PSA(L)· SPGR(L). (AlA)
An identical approach is used for the state variable NSA(L).
375
The equation for phosphorus concentration PCM is
dPCM
V -dt- = Q(PCMBD - PCM) - V f(T)f(I) ·0.322· 10- 4
. N~~:lRIPM(L)' A(L)C + PKl(~)' PCA(L)

- 1 + PKl:L). PCM)] + V N~~:C [(RLYS(L) . TA(L) . TCROP
+ RAEXC(L)· A(L»(PSA(L) - PSAMIN(L»]
+ V· RDCMp· T . TOP + WPCM. (AU)
The equation for nitrogen concentration NCM is functionally identical to the

equation for PCM.
The equation for silicon concentration SCM is
V dS~M = Q(SCMBD - SCM)
- V L (A(L)· SPGR(L) . SSA(L»

Diatoms
+ V . RDCMP . T . TOS + WSCM. (A1.6)
The equation for the concentration Z of zooplankton K is
V dd~) = Q(ZBD(K) - Z(K» + VZ(K)(RZ(K) - ZDETH(K». (A 1.7)
The equation for the total unavailable phosphorus concentration TOP is
V dTOP = Q(TOPBD - TOP)

dl
Naspec
+ VT· TCROP L (RLYS(L) . A(L) . PSAMIN(L»
L=l
Nzspec
+ V L (RZPEX(K)· Z(K»
K=l
TOSINK)
- V. TOP ( RDCMp· T + DEPTH + WTOP. (A1.8)
The equations for TON and TOS are functionally identical to (A1.8).
376
Process rate equations
The specific growth rate of phytoplankton L, SPGR(L), is equal to the minimum
of the fo Howing three functions:
maximum growth rate· f(T)f(l){l - exp[ -0.693(P/PO - I)]},
maximum growth rate· f(T)f(l)· (N - NO)/(KNCELL + N - NO),
maximum growth rate· f(T)f(l)· SCM/(KSCM + SCM).
The rate of growth of zooplankton K is expressed by:
RZ(K) = RZMAX(K) . ZASSIM(K)
Naspec ZEFF(K, L) . A(L)

. L~l KZSAT(K, L) + Lt~fec ZEFF(K, L)· A(L)'
The rate at which phytoplankton L is ingested by zooplankton is:
Nzspec RZMAX(K) . Z(K) . ZEFF(K, L)
RAGRZD(L) = K~l KZSAT(K, L) + L~,;,~ec ZEFF(K, L)· A(L)'
The rate at which phytoplankton L is excreted by zooplankton (used to calculate

the amount of phosphorus excreted to the pool of available phosphorus) is:
_ Nz~ec (1 - ZASSIM(K» . RZMAX(K) . ZEFF(K, L) . Z(K)
RAEXC(L) - K~l KZSA T(K, L) + Lt~fec ZEFF(K, L) . A(L)
The rate at which phosphorus is excreted to the unavailable pool by zooplankton

K is:
RZPEX(K) = (I - ZASSIM(K» . RZMAX(K)

Naspec
ZEFF(K, L) . A(L) . PSAMIN(L)
'L
L=l KZSA T(K. L) + Lt~fec ZEFF(K, L) . A(L)'
Miscellaneous functions
where e = 1.07 for diatoms, 1.08 for greens, and 1.10 for blue-greens.
T = TMAX[0.50 - 0.50 SINE(6.23'T;~E + 4»].

where 4> is chosen such that SINE = 0 on 1 November.
1
f(l) = [exp( -txl)-exp( -txO)],
ke·DEPTH
377
where
Ia
exl = Is exp( - ke . DEPTH),
0- Ia
ex - Is'
ke = 1.9jSecchi depth + 0.17· TCROP
= 0.633 + 0.17· TCROP,
Ia = 2000 foot-candles, Is is the irradiance at the surface, and the photoperiod
is 0.50.
Notation
is the phytoplankton concentration (mg dry wt 1- I),
A
ASINK is the phytoplankton sinking rate (m day- I),
BD is a suffix that refers to the boundary value of a particular
variable,
f(l), f(T) are the light intensity and temperature reduction
factors,
KNCELL is the intracellular half-saturation constant for nitrogen-
dependent growth (mol cell- I),
KSCM is the Michaelis constant for silicon-dependent growth
(moll-I),
KZSAT(K,L) is the half-saturation concentration of phytoplankton
L for grazing by zooplankton K,
Naspec is the number of phytoplankton species,
Nzspec is the number of zooplankton species,
P, N are the moles of phosphorus and nitrogen per phyto-
plankton cell,
PCA, NCA are the intracellular phosphorus and nitrogen concen-
trations (mol 1- I(cell volume)-I),
PCM, NCM, SCM are the nutrient concentrations of phosphorus, nitrogen,
and silicon in solution (mol 1- I),
PSA, NSA are phosphorus and nitrogen storage in phytoplankton
cells (mol (mg dry wt) - I),
PKl, NKI are affinity constants for phosphorus and nitrogen
uptake mechanisms (I mol- I),
PO, NO are the minimum stoichiometric levels of phosphorus
and nitrogen per phytoplankton cell (mol cell-· I),
Q is the water circulation rate [L 3 day - I],
RlPM, RlNM are the maximum phosphorus and nitrogen uptake
rates (day-I),
378
RAEXC(L) is the rate at which phytoplankton L is excreted by zoo-
plankton (day-I),
RAGRZD(L) is the rate at which phytoplankton L is grazed (ingested)
by zooplankton (day - I),
RDCMP is the decomposition rate from unavailable to available
nutrient pools (day- I DC- I),
RLYS is the algal death rate [(day DC mg l-I)-IJ,
RZ is the zooplankton specific growth rate (day - I),
RZMAX is the zooplankton maximum ingestion rate (day-I),
RZPEX(K), are phosphorus, nitrogen, and silicon excreted by zoo-
RZNEX(K), plankton K to the unavailable nutrient poll
RZSEX(K) [mol(mg zooplankton)- I day- IJ,
SPGR is the phytoplankton specific growth rate (day-I),
SSA is the silicon stoichiometry for diatoms
[mol(mg dry wt)-IJ,
T is the temperature (DC),
TCROP is the total phytoplankton biomass (mg dry wt 1- I),
TOP, TON, TOS are the concentrations of unavailable phosphorus,
nitrogen, and silicon (mol 1- I),
TOSINK is the sinking rate of nonliving organic material
(m day-I),
V is the system volume,
WPCM, WNCM, are the external point loading rates of available phos-
WSCM phorus, nitrogen, and silicon (mol day- I),
WTOP, WTON, are the external point loading rates of unavailable
WTOS phosphorus, nitrogen, and silicon (mol day - I),
Z is the zooplankton concentration (mg dry wt 1- I),
ZASSIM is the zooplankton assimilation efficiency,
ZEFF(K, L) is the efficiency of zooplankton K in ingesting phyto-
plankton, L
ZDETH is the zooplankton death rate (day - I).
(2) Model by Canale et al.

Summary of the model equations
c. = [ growt h J - [predation by other] -[predation]

- [resPiration]
Z Z zooplankton Z by alewife z loss z
-AI4lt)·c z (z=I,2, ... ,9)

natural
death rate
. for copePhodJs [predation] [resPiration] [sinking]
c = [growt - - -
P P loss p loss p loss p
(p = 10, 11, 12, 13)
379
CI4 = NCR· L
z
(I - [aSSim~lation]
effiCiency z
)[eatin g] Cz + NCR.
rate z
L[death
z
natural]
rate z
- AI8 . TC I4 - A23 . SINK(t)· CI4
·
CIS . L.p [reSPiration]
= NCR" I
oss p
+ NCR·
A " [reSPiration]
21· L. I
z oss z
+ A 18· TC I4
- A20· TC I 5 + LOAD IS
·
CI6 = NCR· (l - A21)· Lz [reSPiration]
I + A20· Tc A22· TC 16
lS -
oss z
ANH3 . CI 6 )"
- NCR· ( ANH3. C 6 + (I _ ANH3)C
I I7
[growth]p 7 + LOAD I6
C· I7 = A22· TC I 6 - NCR· ( (1 - ANH3)CI7 )"

L [growth]p
ANH3· CI6 + (I - ANH3)C I7 p
+ LOAD 17
C23 = A23· SINK(t)· CI4 + NCR· L [ Sinking]

I . VOLEP
p oss p
+ (LOAD IS + LOAD I6 + LOAD 17) . VOLHY

CI8 = PCR· L (I - [aSSim~lation] )[eating] Cz + PCR. L [ natural]
z effiCiency z rate z z death rate z
- A17· TC I8 - A23· SINK(t)· CI 8
·
CI 9 = PCR· L [reSPiration]
I + A17· TC I8 - A19· TC I9 + LOAD I9
p oss p
c20 = PCR· Lz [ reSPiration]

I
oss z
+ A19· TC l9 - PCR· L [growth]p + LOAD 20
p
C2 4 = A23· SINK(t)· CI 8 + PCR· L[ Sinking]

I . VOLEP
p oss p
+ (LOAD I9 + LOAD 20 )' VOLHY

e21 = SCR. L [reSPiration] + [predation] - A16· TC21
diatoms loss p loss p
- A23· SINK(t)· C21

Cn = A16· TC21 - SCR· L [growth]p + LOAD 22
dialoms
Sinking]
C2S = A23· SINK(t)· C21 + SCR· L [ I . VOLEP
diatoms OSS p
+ LOAD 22 · VOLHY.
380
Model coefficients
Table I Zooplankton-Related Coefficients.
Symbol Definition Unit Value

Used
A 7Iept.-pol~'.
A7eye. } Maximum snatching rates at 20 c e } mg food-C
-----
mg z-C day
0.70
0.43
A 7,ye. rt 2.6
A7diap.n 6.5
A7Iim.-ep. n
A 7d ,"p.
A 7Iim._ep.
} Maximum filtering rates at 20'T
mg z-C day
5.2
1.0
1.25
l
A 7d ,ph. 4.0
Filtering rates at 20 c C
A7ho~m.-hoh 3.5
KFOOD,epl._po,y.
KFOOD,ye.
Half-saturation food level for raptors } mg food-C 1- I
0.2
0.2
A9 Minimum filtering rate multiplier None 0.1
AIO Food level where multiplier is (I + A9)/2 mg food-C 1- 1 0.2
AlIR Assimilation efficiency of raptors mgz-C 0.4

AIlS
AIIN
Assimilation efficiency of selectives
Maximum assimilation efficiency of nonselectives
} -----
mg food-C
0.7
0.8
A24 Half effective food level for nonselectives I1lg food-C 1- I 0.2
AI2'du",
A 12nauPlii
Respiration rates at 20'C } day-l 0.06
0.04
A 14"e.
A 14d;,p.
Al4 IimA'"r1.
} Natural deatb rates for copepods
} day-l
0.005
0.005
0.003
Table II Phytoplankton-Related Coefficients.
Symbol Definition Unit Value Used
A 1"m.dialom~ 2.1
A Jlg.diatoms
A] bluc.greens
A 19rcens
} Maximum growth rates at 20 0
e
} day-l 2.0
1.6
1.9
ISsm,diatoms 225
ISlg. diatom ...
I S blue-greens
ISgree ns
} Optimum light intensities } Iy d,y , 225
600
Inn
A6 sm . diatom.. . 0.05
A6 1g . diaLOms
A6blue_greens
A6green'
}Maximum sinking rates } d,y' 0.Q3
0
0.02
KN
KP
KS
} Michaelis constants for phytoplankton growth} mg nutrient 1-
m
I
0.015
0.0025
0.030
DEPTH Depth of eupbotic zone 20
A3 Respiration rate at 20°C for phytoplankton day-l 0.03
381
Table III Nutrient-Related Coefficients.
NCR
PCR
SCR
Al8
} Nutrient-to-carbon ratios } mg nutrient
------
mgC
0.2
0.001
Detrital nitrogen ---+ organic nitrogen
A20 Organic nitrogen ---+ ammonia 0.0012
A22 Ammonia ---+ nitrate 0.008
Al7 Detrital phosphorus ---+ organic phosphorus } d'Y'C' 0.01
Al9 Organic phosphorus ---+ inorganic phosphorus 0.01
AI6 Detrital silicon ---+ dissolved silicon 0.0015
A23 Maximum detrital sinking rate day-I 0.05

A21 Fraction of zooplankton-respired nitrogen that is organic None 0.7
Q Net flow through Lake Michigan I day- I 1.37· lOll
VOLEP Volume of Lake Michigan epiJimnion I 1.218.10 15
VOLHY Volume of the hypolimnion I 3.654.10 15
ANH~ Concentration of ammonium ion
LOAD Input
(3) Model by Chen and Orlob
Ecological model formulations
TDS
BOD
DO
TEMP
ALGAE
ZOO
Q
I FISH
Volume
382
General mass balance equation for abiotic substances:
advection diffusion input output settling
reaeration decay transformation uptake

NH) --+ N0 2 --+ NO) byproduct
respiration release
General mass balance equation for biota:

t
dVC
+" dC
j
d t = L. QjC j
" "
L. EA dz + L. QinCin - L QOUlCj
+ (Ill - Rt - SI - M)VC I - 112 VC 2 F2,l.
grazing
Phytoplankton (algae):
C N P
K +C KN +N Kp + P'
)12' C 2 = zooplankton.
Zooplankton:
A(JT-20 algae
III = 11 KA + algae'
MI = (J. + fJ· toxicity 112, O 2 = fish.
Fish:
AIJT- 20 zooplankton
III = l1u
K z + zooplankton'
MI = (J. + fJ· toxicity 112' C 2 = harvest.
383
(4) Model by JeJrgensen
Table IV Equations, Modell!.
The model has 17 state variables:

CC Carbon in algal cells (g m - 3),
FNF Proportion of nitrogen in fish,
FNZ Proportion of nitrogen in zooplankton,
FPF Proportion of phosphorus in fish.
FPZ Proportion of phosphorus in zooplankton.
NC Nitrogen in algal cells (g m - 3),
NO Nitrogen in detritus (g m - 3),
NSEO Nitrogen in sediment (g m - 3),
NS Soluble nitrogen (g m - 3),
PB Phosphorus released biologically from sediment (g m - 3),
PC Phosphorus in algal cells (g m - 3),
PO Phosphorus in detritus (g m - 3),
PE Exchangeable phosphorus in sediment (g m - 3),
PHYT Phytoplankton biomass (g m - 3),
PI Phosphorus in interstitial water (g m - 3).
PS Soluble phosphorus (g m - 3),
ZOO Zooplankton biomass (g m -3).
Levell. the differential equations:

dCCjdr = (UC- RC)· PHYT - (SA + GZ/Y + Q/V)· CC
dFNF/dr = (PREO/y)· (FNZ - FNF)
dFNZ/dr = MYZ . (FNA - FNZ)
dFPF/dr = (PREO/y)(FPZ - FPF)
dFPZjdr = MYZ·(FPA - FPZ)
dNC/dr = UN· PHYT - (SA + GZ/Y + Q/V)· NC
dNO/dr = L· GZ· NC + MZ· NZOO + L· PREO· NFISH - (KON + SO + Q/V)· NO
+ QNDIN
dNSEO/dr = (SA· NC + SO· NO - NREL)/AE
dNS/dr = KON . NO + RZ + NZOO + PREO· NFISH + NREL - UN· PHYT
+ QNSIN - (Q/V + OENIT)· NS
dPB/dl = QSEO/AB - QBIO - QOSORP
dPC/dr = UP· PHYT - (SA + GZ/Y + Q/V)· PC
dPO/dr = L· GZ· PC + MZ· PZOO + L· PREO . PFISH - (KOP + SO + Q/V). PO
+ QPDIN
dPE/dr = [(12/29)· SA· PC - QSEO + SO· PO]/AE - KE· PE
dPHYT/dr = (COR - SA - GZ/Y - Q/V)· PHYT
dPI/dr = (AE/AI)· KE· PE - QDIFF/AI
dPS/dr = KOp· PO + RZ· PZOO + PREO· PFISH - UP· PHYT + QDIFF +
+ QPSIN - (Q/V). PS + AB· (QBIO + QOSORP)
dZOO/dr = (MYZ - RZ - MZ - Q/V). ZOO - PREO· FISH/Y
QNDIN (QPDIN) and QNSIN (QPSIN) represent the inflows of detrital and soluble nitrogen
(phosphorus)
Level [I. rates:

COR = COR m .,,·FTI·FN3·FC3·FP3
GZ = MYZ ·FZP
KON = KON lo ' FT3
KOP =KOP,o·FT3
KE = KE 20 • FT2
MYZ = MYZ m " . FPH . FTI (continued over)
384
Table IV (continued)
NREL = FTS ·(KREL· NSED + 0.08)/(1000D)

PRED = PREDm.,,· FTI . FZ
QBIO = 0.563· FT6· (PBI 1800)/( I000· DB)
QDIFF = FT4·r12I·(PI - PS) - 1.701(1000D)
QDSORP = (0.60 Ig PS - 2.27)/(1000· DB)
QSED = min(SA ·PC ·5.06·10·-')
RC = RC m,,· FC4· FTI
RZ = RZ",.,,· FTI
SA = (SVS/D) ·(FT2)'2
SO = (SVDjD) . (FT:!) I 2
UC = UC",ax· FC! . FC2· FRAD
UN = UN max . FN I . FN2
UP = UP"",,·FPI·FP2
Level III, limiting factors:
FCA = CC/PHYT
FC! = (FCA m" - FCA)/(FCA",,, - FeA min )
FC2 = C/(KC + C)
FC3 = I - CC",,,/CC
FC4 = (CC/CC mi ,,)23
FNA = NC/PHYT
FNI = (FNA m" - FNA)/(FNA max - FNA mi ,,)
FN2 = NS/(NS + K N)
FN3 = ! - NCmin/NC
FPA = PC/PHYT
FPH = max(O, (PHYT - 0.5)/(PHYT + KA»
FPI = (FPA",ax - FPA)/(FPA max - FPA min )
FP2 = PSj(PS + KP)
FP3 = I - PC mi"/PC
FRAD = Ig[(RAD + KL)/(RAD· BEER + KL)]/n
BEER = exp(-n)
n = (11 + (3 ·PHYT)D
FTI = exp[ -2.311 - 16.51/15]
FT2 = ()'- 20
FT3 =,p' 10
FT4 = (t + 273)/280
FTS = exp(0.151 I)
FT6 = exp(0.2031)
FZ = max(O, (ZOO - KS)/(ZOO + KZ»
FZP = ZOO/PHYT
Level IV, other equations:
FISH = FISHo·{1 + 0.8 sin[0.017453 (DAY + 150)])
NWAT = NC + NO + NS + NZOO
PW AT = PC + PO + PS + PZOO
NTOT = NC + NO + NS + NZOO + NFISH + AE· NSED
PTOT = PC + PO + PS + PZOO + PFISH + AE· PE + AI· PI + AB· PB
NZOO = FNZ· ZOO
NFISH = FNF· FISH
PZOO = FPZ· ZOO
PFISH = FPF· FISH
PROD = CDR· PHYT
AB = (DB/D)· DMU
AE = LUL·DMU/D
AI = LUL·(I - DMU)D
385
Table V Parameter Values.
Symbol Definition Unit Values Values Based

Upon:
Modell Model II
ex Extinction coefficient of m- I
0.27 0.27 Chen and Orlob
water (1975)
Specific extinction m2 g- I 0.18 0.18 Chen and Orlob
fJ
coefficien t of (1975)
phytoplankton
C Concentration of mg I-I 100 100 Measurements
inorganic carbon
CORm:J' Maximum growth rate of day-l 2.3 2.53 Calibration
phytoplankton
D Depth m 1.8 1.8 Measurements
DB Depth of biologically m 2.10- 3 Measurements
very active layer
DENIT Denitrification coefficient day-l 0.02 0.03 Nitrogen
balance
DMU Dry matter in sediment 0.07 0.07 Measurements
FISH o Concentration of fish mg I-I 0.55 0.3 Calibration
KA Michaelis constant for mg 1-' 0.5 2.0 Chen and Orlob
zooplankton grazing on (1975)
phytoplankton
KC Michaelis constant for mg I-I 0.5 0.5 Chen and Orlob
carbon uptake (1975)
KDN lo Decomposition rate of day-I 0.1 0.1 Calibration
detritus nitrogen at 10°C
KDP lo Decomposition rate of day-I 0.25 0.4 Calibration
detritus phosphorus at
lOT
KE 10 Decomposition rate of day-I 2.5· 10-· 2.2· 10- • Calibration
D
PE at 20 C
KL Michaelis constant for kcal m- 1 day-I 400 400 Gargas (1975)
light
KN Michaelis constant for mg 1-' 0.2 0.2 Lehman el at.
nitrogen uptake (1975), Chen
and Orlob (1975)
KP Michaelis constant for mg 1- 1 0.02 0.02 Lehman el at.
phosphorus uptake (1975), Chen
and Orlob (1975)
KREL Rate constant for release day- I 0.0040 0.0040 Jacobson and
of nitrogen Jorgensen (1975)
KS Theshold zooplankton mg I-I 0.75 0.75 Steele (1974)
biomass
KZ Michaelis constant for fish mg I-I 0.35 0.35 Calibration
feeding on zooplankton
LUL Unstable layer of sediment m 0.1 0.1 Measurements
OS LUL· (I - DMU) m
MA Mortality of day -I 0.09 Calibration
phytoplankton (continued over)
386
Table V (conrinued)
Symbol Definition Unit Values Values Based

Upon:
Modell Model II
MYZll\;JX Maximum growth rate of day - I 0.2 0.188 Calibration

zooplankton
MZ Mortality of zooplankton day- I 0.025 0.033 Calibration
NH 4 P Ammonia concentration mg I-I 0.2 0.2 V. Jorgensen
in rainwater (1972)
N0 3 P Nitrate concentration in mg I-I 0.16 0.16 V. Jorgensen
rainwater (1972)
<jJ Temperature coefficient 1.072 1.072
for degradation of detritus
PREDm ", Maximum feeding rate of day- I 0.06 0.012 Calibration
fish on zooplankton
PP Phosphorus concentration mg 1- J 0.0015 0.0015 V. Jorgensen
in rainwater (1972)
RC m " Maximum respiration day- J 0.13 0.088 Calibration
rate of phytoplankton
RZ miLx Maximum respiration day- , 0.035 0.028 Calibration
rate of zooplankton
SVD Settling rate of detritus m day-J 0.002 0.0019}
SVS Settling rate of m day-I 0.06 0.19 Jorgensen (1976)
Scenedesmus
a Temperature coefficient day-J 1.03 1.03 Chen and Orlob
for decomposition of PE (1975)
UC m " Maximum rate of carbon day- J 0.65 0.55 Calibration
uptake
UN m " Maximum rate of day-' 0.03 0.015 Calibration
nitrogen uptake
UP min Maximum rate of day-J 0.003 0.0014 Calibration
phosphorus uptake
FCA min Minimum kg C per kg 0.15 0.15
phytoplankton biomass
FCA m " Maximum kg C per kg 0.6 0.6
FNA min Minimum kg N per kg 0.015 0.015
phytoplankton biomass Jorgensen (1976)
FNA m" Maximum kg N per kg 0.10 0.10
FPA min Minimum kg P per kg 0.001 0.001
FPA m " Maximum kg P per kg 0.013 0.013
V Volume of the lake m3 420000 420000
Y Yield of feeding 0.63 0.63 Chen and
zooplank ton and fish Orlob (1975)
L IjY - I 0.59 0.59
387
Table VI Forcing Functions (all given as tables).
QTRI m J day - I tributaries

NTOTRI Total N (mg 1- ') tributaries
NH 4 TRI mg NH; -N 1 1 tributaries
NO~TRI mg NO.1 -N 1- , tributaries
PTOTRI Total P (mg 1-') tributaries
PTRI mg PO~ - .p 1- 1 tributaries
T Temperature of lake water
QWAS m J day -, wastewater
NTOWAS Total N (mg 1- I) wastewater
NH 4 WAS mg NH; -N 1- 1 wastewater
NOJWAS mg NO.1 -N I" wastewater
PWAS Total P (mg 1-') wastewater
RAD Global irradiance (kcal m - 2 day - I)
Q Outflow (m J day' ')
QPREC Precipitation (m J day- ')
Table VII Additional Equations, Model III.
dZOOl/dl = (MYZI - RZI - MZI - QiV)' ZOOI - PREDI . FISHI Y

PREDI = PREDI"u, . FTI . FZ
MYZI = MYZI",,,,' FD· FTI
RZI = RZI",n . FTI
FD= DET/(DET + 2.0)
DET= (ND/FNA + PD/FPA)/2
MZI is the mortality
ZOO I is the concentration of zooplankton species I.
(5) Model by Baca and Arnett

aps
at
phytoplankton sediment sediment gain by
uptake release uptake organic decay
and
apo
--at = (D p - CgZ)PA pp + RzZA pz 12 , PO,
recycled from recycled from sediment decay
dead phytoplankton zooplankton trapping
where
PS is the concentration of inorganic phosphorus,
PO is the concentration of organic phosphorus,
PE is the concentration of inorganic phosphorus in sediment,
P is the phytoplankton concentration,
are yield coefficients,
is the net phytoplankton growth rate,
388
Dp is the phytoplankton death rate,
Cg is the zooplankton grazing rate,
Rz is the zooplankton respiration rate,
/ I is the sediment uptake rate,
/2 is the organic phosphorus decay rate,
/3 is the sediment release rate,
/4 is the sediment trapping rate,
Z is the zooplankton concentration.
The terms in braces apply to the hypolimnion and the term in square brackets
designates processes that depend on anaerobic conditions.
For the nitrogen submodel, the equations are:
aCI CI
-:1- -JIC l - PGpA npc C + J 4 C4 + {JsCd,
ut I + 3
ammonia phytoplankton organic anaerobic
oxidation uptake nitrogen sedimenl
recycled release
ammonia nitrite
oxidation oxidation
C3
- PGpA np C C
I + 3
nitrite phytoplankton anaerobic
oxidation uptake denitrification
RzZA nz
organic recycled from recycled from sediment
decay dead phytoplankton zooplankton uptake
where
CI is the ammonia nitrogen concentration,
C2 is the nitrite nitrogen concentration,
C3 is the nitrate nitrogen concentration,
C4 is the organic nitrogen concentration,
Cs is the sediment nitrogen concentration,
JI is the ammonia oxidation rate,
J2 is the nitrite oxidation rate,
J3 is the denitrification rate,
J4 is the organic nitrogen decay rate,
Js is the sediment nitrogen decay rate,
J6 is the sediment uptake rate,
Anp , A nz are the nitrogen: carbon ratios for phytoplankton and zooplankton.
In these formulations the rates of phytoplankton uptake of ammonia (C I) and
389
nitrate (C 3 ) are governed by their relative concentrations in the water column.
Denitrification and release of nitrogen (ammonia) from the sediment depend
on the existence of anaerobic conditions.
REFERENCES
Ahlgren. I. (1973a). Limnologiska studier av Sjl'>n Norrviken. Ill. Avlastningens elfekter. Scripta
Lil11l1olo,(fica Up.l'aliellcia No. 333.
Ahlgren, I. (1973b). Limnological studies of Lake Norrviken, a eutrophicated Swedish lake.
Nutrients and phytoplankton before and after sewage diversion. Scripta Limllolo.'lica Upsaliellcia
No. 334.
Baca. R. G .• and R. C. Arnett (1976). A finite-element water quality model for eutrophic lakes.
Fillite Elemellts ill Water Resources vol. 4, eds. W. G. Gray, G. Pinder, and C. Brebbia (Plymouth,
Devon:Pentech), pp. 125-147.
Beck. M. B. (ed.) (1978). Mathematical modeling of water quality. Summary Report of a IlASA
Workshop, September 13-16, 1977. Collaboratire Proceedill.'!s CP-78-1 0 (Laxenburg, Austria:
International Institute for Applied Systems Analysis).
Bennett, J. R. (1974). On the dynamics of wind-driven lake currents. Journal of Physical Oceallo-
.'!raphy 4(3):400-414.
Bierman, V. J. (1976). Mathematical model of the selective enhancement of blue-green algae by
nutrient enrichment. Modelillg Biochemical Processes in Aquatic Ecosystems, ed. R. P. Canale
(Ann Arbor, MI: Ann Arbor Science), pp. 1-32.
Blake, G .. and S. Gentil (1979). A discrete lake model and its sensitivity study. State of the Art ill
Ecoloqical Modellinq. ed. S. E. Jorgensen (Copenhagen: International Scoiety for Ecological
Modelling), pp. 527-544.
Bloomfield, J. A., R. A. Park. D. Scavia, and C. S. Zahorcak (1974). Aquatic modeling in the
eastern deciduous forest biome, US International Biological Program. Modelill.iJ the Eutro-
phication Process, eds. E. Middlebrooks, D. H. Falkenberg, and T. E. Maloney (Ann Arbor, MI:
Ann Arbor Science), pp. 139-158.
Canale, R. P.. L. M. DePalma, and A. H. Vogel (1976). A plankton-based food web model for Lake
Michigan. Modelin.iJ Biochemical Processes in Aquatic Ecosystems, ed. R. P. Canale (Ann Arbor,
MI: Ann Arbor Science). pp. 33-74.
Caperon, J., and J. Meyer (1972a). Nitrogen-limited growth of marine phytoplankton -I. Changes
in population characteristics with steady-state growth rate. Deep Sea Research 19: 60 I.
Caperon, J., and J. Meyer (1972b). Nitrogen-limited growt~. of marine phytoplankton --II. Uptake
kinetics and their role in nutrient limited growth of phytoplankton. Deep Sea Research
19:619.
Analysis and Simulation in Ecolo.'!y vol. 3, ed. B. C. Pallen (New York, NY: Academic Press),
pp.476-588.
Chen. C. W., M. Lorenzen, and D. J. Smith (1975). A comprehensive water-quality-ecologic model
for Lake Ontario. Report to Great Lakes Enrironment Research Laboratory, National Oceanic
and Atmospheric Administration by Tetra Tech, Inc., 202 pp.
Desormeau, C. J. (1978). Mathematical modeling of phytoplankton kinetics with application to
two alpine lakes. Cellter fill' Ecologiml Modeling, Rensselaer Poll'technic In.witute, Troy. NY
Report 4, 2 I pp.
Di Toro, D. M., D. J. O'Connor, and R. V. Thomann (1971). Nonequilibrium systems in natural
water chemistry. American Chemical Society, Admnced Chemistry Series 106: 131-180.
Di Toro, D. M., D. J. O'Connor, R. V. Thomann, and J. L. Mancini (1975). Phytoplankton-
zooplankton-nutrient interaction model for western Lake Erie. 51'stems Analysis and Simulation
in Ecolo.lf.l' vol. 3, ch. II, ed. B. C. Pallen (New York. NY: Academic Press). pp. 423-474.
390
Di Toro, D. M., R. V. Thomann, D. J. O'Connor, and J. L. Mancini (1977). Estuarine phytoplankton
biomass models- Verification analyses and preliminary applications. The Sea vol. 6 Marine
modeling, eds. E. D. Goldberg et al. (New York, NY: Wiley), pp. 969-1020.
Droop, M. R. (1973). Some thoughts on nutrient limitation in algae. Journal 01' Phycology 9: 264.
Droop, M. R. (1974). The nutrient status of algal cells in continuous culture. Memoranda 01' the
Marille Biologicol Associati(!I1, L J\. 54:825-855.
Dubois, D. M. (1975). A model of patchiness for prey-predator populations. Ecological Modelling
1: 67-80.
Dubois, D. M. (1979). Catastrophe theory applied to water quality regulation of rivers. State ol'the
Art in Ecological Modelling, ed. S. E. Jorgensen (Copenhagen: International Society for Ecologi-
cal Modelling), pp. 751-759.
Duckstein, L., J. Casti, and J. Kempf (1977). Modeling phytoplankton growth in small eutrophic
ponds with catastrophe theory. Proceedings 01' 13th American Water Resources Conference,
Tucson, AZ.
Dugdale, R. C. (1967). Nutrient limitation in the sea: Dynamics, identification, and significance.
Limnology and Oceanoqraphy 12: 685.
Eppley, R. W., and W. H. Thomas (1969). Comparison of half-saturation constants for growth and
nitrate uptake of marine phytoplankton. Journal 01' Phycology 5: 375.
Fuhs, G. W. (\969). Phosphorus content and rate of growth in the diatoms Cyclotella nana and
ThalossiosiraffUl·iatilis. Journal 01' Phycology 5: 312.
Fuhs, G. W., S. D. Demmerle, E. Canelli, and M. Chen (1971). Characterization of phosphorus-
limited planktonic algae, nutrients and eutrophication: The limiting nutrient controversy.
Proceedings of Symposium of American Society of Limnology and Oceanography and Michigan
State Unil'ersity, East Lansing, MI, February 1/-12, p. 113.
Gargas, E. (1975). A Manual/or Phytoplankton. Primary Production Studies in the Baltic (Baltic
Marine Biologists in cooperation with the Danish Agency of the Environment).
Groden, T. W. (1977). Modeling temperature and light adaptation of phytoplankton. Center /or
Ecological Modeling, Rensselaer Polytechnic Institute, Troy. NY Report 2,17 pp.
Halfon, E., and D. C. L. Lam (1978). The effects of advection-diffusion processes on the
eutrophication of large lakes. A hypothetical example: Lake Superior. Ecological Modelling 4:
119-132.
Halmann, M., and M. Stiller (1974). Turnover and uptake of dissolved phosphate in freshwater. A
study in Lake Kinneret. Limnology and Oceanography 19: 774.
Ikeda, S., Y. Inoue, and S. Iwai (1979). Multispecies of plank tons and nutrients model of lake eutro-
phication. A simulation study in Lake Biwa. State 01' the Art in Ecological Modellinq. ed. S. E.
Jorgensen (Copenhagen: International Society for Ecological Modelling), pp. 501-526.
19:297-304.
Imboden, D. M. (1979). Modelling of vertical temperature distribution and its implication on bio-
logical processes, in lakes. State oftile Art ill Ecological Mode/ing, ed. S. E. Jorgensen (Copenhagen:
International Society for Ecological Modelling), pp. 545-561.
Imboden, D. M., and R. Giichter (1978). A dynamic lake model for trophic state prediction.
Ecological Modellinq 4: 77-98.
Jacobson, O. S., and S. E. Jorgensen (1975). A submodel for nitrogen release from sediments.
Ecoloqical Modellin" I: 147-151.
Jolankai, G., and A. Sz6116si-Nagy (1978). A simple eutrophication model for the Bay of Keszthely,
Lake Balaton. Modeling the Water Quality of the Hydrological Cycle. Proceedinqs of Baden
Symposium, September. International Association 0/ Hydrological Sciences Publication 125,
pp. 137-150.
Jorgensen, S. E. (1976). A eutrophication model for a lake. Ecological Modelling 2: 147-165.
Jorgensen, S. E. (1978). Focus on ecological modeling. Ecological Modelling 4: 1-2.
Jorgensen, S. E. (ed.) (1979). State of the art of eutrophication models. State 01' the Art in Ecoloqical
Modellinq (Copenhagen: International Society for Ecological Modelling), pp. 293-298.
391
Jorgensen, S. E., L. Bengtsson, and M. Friis (/982). A case study of lake modeling. Siidra Bergun-
dasjiin. International Society/or Ecological Modelling Journal 4.
Jorgensen. S. E., and D. R. F. Harleman (eds.) (1978). Hydrophysical and ecological modelling
of deep lakes and reservoirs. Summary Report of a IIASA Workshop, December 12-15, 1977.
Collaborati"e Proceedings CP-78-7 (Laxenburg, Austria: International Institute for Applied
Systems Analysis).
Jorgensen, S. E., L. Kamp-Nielsen, and O. S. Jacobsen (1975). A submodel for anaerobic mud-
water exchange of phosphate. Ecological Modelling I: 133-146.
Jorgensen. S. E., and H. F. Mejer (1977). Ecological buffer capacity. Ecological Modellillg
3:39-61.
Jorgensen, S. E., and H. F. Mejer (1979). A holistic approach to ecological modelling. Ecological
Jorgensen, S. E., H. F. Mejer, and M. Friis (1978). Examination ofa lake model. Ecological Model-
Iin,lj 4:253-278.
Jorgensen, S. E., H. F. Mejer, L. A. Jorgensen, and L. Kamp-Nielsen (1981). Parameter estimation
for a eutrophication model. Ecological Modelling 13: 111- [29.
Jorgensen, S. E., O. S. Jacobsen, and I. Hoi (1973). Prognosis for a lake. Vatten 29: 383-404.
Jorgensen, V. (1972). Nedbarens indhold ofplantluaringsstof. Statens Plantlal'lsmode 22 November,
15pp.
Kamp-Nielsen, L. (1975). A kinetic approach to the aerobic sediment-water exchange of phos-
phorus in Lake Esrom. Ecological Modelling I: 153-160.
Kinnunen, K., J. S. Niemi, and J. Eloranta (1978). Adaptation of the EPAECO model to a lake
in central Finland. Modeling the Water Quality or the Hydrological Cycle. Proceedings orBaden
Symposium, September. International Association or Hydrological Sciences Publication 125.
pp.115-127.
Kremer, J. N., and S. W. Nixon (1978). A Coastal Marine Ecosystem Simulation and Analysis
(Berlin: Springer), 217 pp.
Lappalainen, K. M. (1975). Phosphorus loading capacity of tubes and a mathematical model for
water quality prognoses. Proceedings or 10th Nordic Symposium on Water Research, Vocrlose,
May 20-22, 1974 (Helsinki: ., Entrofierung" NORFORSK).
Larsen, D. P., H. T. Mercier, and K. W. Malveg (1974). Modeling algal growth dynamics in
Shagawa Lake, Minnesota. Modeling the Eutrophication Process, eds. E. J. Middlebrooks, D. H.
Falkenberg, and T. E. Maloney (Aim Arbor, MI: Ann Arbor Science), pp. 15-33.
Lehman, J. T., D. B. Botkin, and G. E. Likens (1975). The assumptions and rationales ofa computer
model of phytoplankton population dynamics. Limnology and Oceanography 3: 343-
364.
Lorenzen, M. W., O. J. Smith, and L. V. Kimmel (1976). A long-term phosphorus model for lakes:
Application to Lake Washington. Modeling Biochemical Processes in Aquatic Ecosystems, ed.
R. P. Canale (Ann Arbor, MI: Ann Arbor Science), pp. 75-92.
Mauersberger, P. (1979). On the role of entropy in water quality modelling. Ecological Modelling 7:
19/-199.
Mejer, H. F., and S. E. Jorgensen (1979). Energy and ecological buffer capacity. State 01" the Art in
Ecological Modelling, ed. S. E. Jorgensen (Copenhagen: International Society for Ecological
Modelling). pp. 829-846.
Nyholm, N. (1975). Kinetics studies of phosphate-limited algae growth. Thesis, Technical Unirersity
or Copenhagen.
Nyholm, N. (1978). A simulation model for phytoplankton growth and nutrient cycling in eutro-
phic, shallow lakes. Ecological Modelling 4: 279-310.
Institute, Raleigh Report 97, 45 pp.
Park, R. A., C. Collins, and C. Desormeau (1979a). Modeling combined effects of environmental
factors on phytoplankton. Center/or Ecological Modeling, Rensselaer Polytechnic Institute, Troy,
NY Report.
392
Park. R. A .. T. W. Groden. and C. J. Desormeau (1979b). Modification to model CLEANER,
requiring further research. Per.lpeui/'es on Lake Ecosystem Modelinq. eds. D. Scavia and A.
Robertson (Ann Arbor. MI: Ann Arbor Science). pp. 87-108.
Park. R. A.. er al. (1974). A generalized model for simulating lake ecosystems. Siml/lario/l
23: 33-50.
Parker. R. A. (1976). The influence of eddy diffusion and advection of plankton population systems.
International Journal oj'Sysrems Science 7: 957-962.
Parker. R. A. (1978). Spatial patterns in some nutrient-plankton models. Ecological Modelling 6:
78-90.
Patten. B. c.. D. A. Egloff. and T. H. Richardson (1975). Total ecosystem model for a cove in Lake
Texoma. Systems Analysis and Simulation in Ecology vol. 3. ed. B. C. Patten (New York, NY:
Rhee, G. Yu. (19n). Effects of N: P atomic ratios and nitrate limitation on algal growth, cell com-
position, and nitrate uptake. Limnology and Oceanowaphy 23: 10-25.
Richey, J. E. (1977). An empirical and mathematical approach toward the development of a phos-
phorus model of Castle Lake. Ecosystem Modeling in Theory and Practice, eds. C. A. S. Hall and
I. W. Day, Jr. (New York, NY: Wiley-Interscience). pp. 267-287.
Scavia. 0 .. B. J. Eadie. and A. Robertson (1976). An ecological model for the Great Lakes. Great
Lakes Enl'ironmental Res!!Urch Laboratorv. National Oceanic and Atmospheric Association.
Ann Arbor, MI Contribution 64.
Scavia, 0 .. and R. A. Park (1975). A user's manual for the aquatic model CLEANER. Federal
Water Industrr. TrOI', NY Report 75-16.
Smith, E. L. (1936). Photosynthesis in relation to light and carbon dioxide. Proceedings of National
Academy oj'Sciences 22: 504-511.
Smith, M. T. (1973). Lynqby Kommune Report on Lyngby Lake, Copenhagen. 165 pp.
Snodgrass, W. J.• and C. R. O'Melia (1975). Predictive model for pbosphorus lakes. Sensitivity
analysis and applications. Enl'ironmenllll Science and Technolo.qy 9(10): 937-944.
Steel, J. A. (1972). The application of fundamentallimnological research in water supply system
design and management. Journal o/Zoology 29: 41-67.
Steele, J. H. (1962). Environmental control of photosynthesis in the sea. Limnology and Oceano-
graphy 7: 137-150.
Steele, J. H. (1974). The Strtlcture oj'Marine Ecosystems (Oxford: Blackwell), pp. 74-91.
Straskraba. M. (1976). Development of an analytical phytoplankton model with parameters
empirically related to dominant controlling variables. Umll'eltbiophysik, eds. R. Glaser. K.
Unger. and M. Koch (Berlin: Akademie-Verlag), pp. 33--65.
Straskraba, M. (1979). Natural control mechanisms in models of aquatic ecosystems. Ecological
Tapp, J. S. (1978). Eutrophication analysis with simple and complex models. Journal of Water
Pollurion Control Federation 50: 484-492.
Thomann. R. V.. D. M. Di Toro. R. P. Winfield. and D. J. O'Connor (1975). Mathematical model-
ling of phytoplankton in Lake Ontario. US Enrironmental Protection Agency. National Em'iron-
mental Research Center. COrl'allis. OR Report EPA 660/3-75-005.
Vollenweider, R. A. (1965). Calculation models of photosynthesis-depth curves and some impli-
cations regarding day rate estimates in primary production. Memorie dell Istiruto Italiano di
Idrohioloqia 18 Suppl. :425-457.
Vollenweider. R. A. (1969). Miiglichkeiten und Grenzen elementarer Modelle der Stoffbilanz von
Seen. Archil' Hydrohiologie 66: 1-36.
Vollenweider. R. A. (1975). Input-output models with special reference to the phosphorus loading
concept in limnology. Schll'eizerische Zeitschrififur Hydrologie 37: 53-83.
Youngberg. B. A. (1977). Application of the aquatic model CLEANER to a stratified reservoir
system. Center for Ecological Modeling. Rensselaer Polvtechnic Institute. Troy, NY Report I,
22 pp.
393
CHAPTER 9: NOTATION
AREA lake surface area

c concentration
Cj concentration of constituent in water element j
CONRI rate constant for conversion of algal to particulate phosphorus
CONR2 rate constant for conversion of particulate to soluble phosphorus
E eddy diffusion coefficient
K 1 specific rate of phosphorus transfer to sediment
K 2 specific rate of phosphorus transfer from sediment
K 3 fraction of total phosphorus input to sediment that is not avail-
able for exchange
KFLM food level
KMFM minimum filtering rate multiplier
KP half-saturation constant for phosphorus
KZ half-saturation constant for grazing
lp phosphorus supply rate
LIGHT fractional reduction in MYMAX(T) in epilimnion due to avail-
ability of light
MYMAX(T) maximum specific growth rate of phytoplankton as a function
of temperature
MYZMAX maximum specific growth rate of zooplankton
[P] phosphorus concentration in lake
PA concentration of algal phosphorus
PC phosphorus concentration per cell (PCA plus internal storage)
PCA internal dissolved phosphorus concentration
PCAMIN minimum value of PCA
PCM external dissolved phosphorus concentration
PHYT concentration of algal phytoplankton
PKI uptake constant
PO minimum stoichiometric level of phosphorus per cell
PP concentration of particulate (non-algal) phosphorus
PPIN rate of supply of particulate phosphorus to epilimnion
PS concentration of soluble phosphorus
PSED concentration of soluble phosphorus in sediment
PSIN rate of supply of soluble phosphorus to epilimnion
Q outflow rate
SETTLl settling rate constant for algal phosphorus
SETTL2 settling rate constant for non-algal particulate phosphorus
J,j volume of water element j
VE volume of epilimnion
VL lake volume
VS sediment volume
394
Xm measured value
Xl predicted value
Y objective function
IX k electivity coefficient
Pw hydraulic washout coefficient
up sedimentation rate constant.
Modeling the Distribution and
10 Effect of Toxic Substances in
Rivers and Lakes
S. E. Jorgensen
10.1. INTRODUCTION
Only a few models attempting to describe quantitatively the distribution and

effect of toxic substances have been published. However, since the impact of
toxic substances on man and aquatic ecosystems is well recognized, it is im-
portant to develop such models and use them as environmental management
tools.
Passage of the Toxic Substances Act of 1976 in the US, unprecedented
fines, and continual development of data on lethal and sublethal effects attest
to the expansion of control on the production and discharge of such substances.
As a result, considerable effort has been devoted in recent years to the develop-
ment of predictive schemes that would permit a judgment of the effects of a
given compound on the environment.
Many water constituents are toxic at certain concentrations and interact
directly with the biota of the ecosystem, causing death or severe stress and
limiting the use of water resources. With growing industrialization and a
steadily increasing number of new toxic compounds, there is indeed a great
need for development of water quality models that can be used for predicting
safety levels and establishing water quality criteria. This modeling effort should
take account of: (i) heavy metals; (ii) oils and chlorinated hydrocarbons; (iii)
pesticides; (iv) other organic toxic compounds; and (v) radionuclides. The basis
for modeling the distribution and effect of toxic substances in an aquatic eco-
system is knowledge of the processes in the ecosystem. Section 10.2 is devoted
to a survey of our present knowledge of these processes and the related co-
efficients.
395
396
The few models in use can be divided into three classes.
I. Food chain or food web models, similar to the eutrophication models
presented in Chapter 9, with additional state variables to describe
the concentrations of toxic substances, such as compartment models,
provide great flexibility and incorporate reasonably complicated
food webs. However, such models often become unwieldy and require
comprehensive sets of data for their calibration and validation. If
there are m ecological variables (or components) at n spatial locations,
and if we consider p toxic substances, then there are mnp equations
(differential or algebraic) to be solved. Thousands of equations can
easily result.
II. Other modeling approaches attempt to simplify the procedure. Such
models require less data and can be applied more generally, but are
less accurate or give less information. Often, for instance, the model
does not describe seasonal variations. However, this is not significant
when the model is used as a management tool, as the relevant problem
most often is to map the worst-case situation.
III. The simplifications are determined by the scope of the model. A
typical environmental management problem is the relation between
the amount of toxic substance discharged and the concentration in fish.
A model focusing on this problem does not need to include all trophic
levels, but it might be possible to solve the problem by setting up a
model for the concentration of a toxic compound in fish and con-
sidering only the processes shown in Figure 10.1.
The three classes of model are surveyed in section 10.3, and their advantages and
disadvantages discussed.
Concentration factor
Feed
Feed -.:::---t- for next
level
Not
digested Mortality Excretion
(respiration)
FIGURE 10.1 Modeling the concentration of a toxic substance in a trophic level.
397
The model should predict a concentration or level of toxic compound in
water, fish, phytoplankton, etc. that can be related to the effect, such as in-
creased mortality or decreased growth rate. Since both the model and our
knowledge of effects of toxic compounds are uncertain it is necessary to apply
high safety factors for the assessment of water quality criteria. The relation
ofthis type of model to other ecological models was mentioned in section 4.2.10,
where the modification of mortality and growth rate by toxic substances was
demonstrated.
10.2. PROCESSES AND RELATED PARAMETERS
10.2.1. General Considerations

Ecosystem processes in which toxic substances occur can be divided into three
groups:
(1) physical processes, such as adsorption and volatilization;

(2) chemical processes, such as oxidation, photolysis, hydrolysis, ioniza-
tion, and complexation; and
(3) biological processes, such as biodegradation, uptake, and excretion.
The processes are included in the total models. This section contains some
general ideas about these processes and the parameters used to describe them.
Detailed information is available for a limited number of toxic compounds
(J0rgensen, 1979b); when it is not obtainable, general rules have to be used.
10.2.2. Volatilization from Water

The theory of this process has been developed by several people (Liss and Slater,
1974; MacKay and Cohen, 1976; Smith et aI., 1977). Figure 10.2 illustrates the
major features of a two-film model of mass transfer, which is generally applied
in chemical engineering. The water phase is assumed to be well mixed so that
any volatile compound is at a uniform concentration Cs except in the vicinity
of the interface. A stagnant liquid film of thickness 6 1 separates the bulk of the
water phase from the interface. A volatile component moves through this film
by diffusion. The concentration decreases across the film from Cs to CSi ' and the
rate at which the component is transported across this film, N s , is given by
Ns = K~(Cs - C Si ), (10.1)
where K~ is the liquid film mass transfer coefficient (m h - 1). A stagnant gas
film ofthickness 6g is on the air side. The partial pressure PSi on this side is related
398
Interface
~
:::l
IJ)
IJ)
Water
• Air
~ Convective transport
Q.
Psi
co
"f
co
Q.
o IPS
r:
C I
o
'+:< C si :
....~c Liquid film ~ Gas film
OJ
<.l
C 1 _1.,_ _-,,::
o : (51 (59 I
U
1 1
Distance
FIG U RE 10.2 The two-film model of volatilization from the surface of a water body,
to C Si (molar concentration on the water side) in accordance with Henry's law:
PSi = HeCsi = HXSi' (10.2)
where He and H are Henry's law constants expressed in moles and as a mole
fraction, respectively:
He = 1.8·10;-3 H. (10.3)
The rate of transport across the gas film may be expressed by
KS
Ns = R; (PSi - P s), ( lOA)
where K~ is the gas film mass transfer coefficient (m h - 1). By continuing this
equation, we obtain
KS
v -
- A(l
V K S+
_
HSK S
RT)-l '
(10.5)
leg
where
K~ is the overall transfer coefficient (h - 1),
A is the interfacial area (m 2 ),
V is the liquid volume (m 3 ),
T is the temperature eq.
399
A similar equation can be used for oxygen transport. We assume that
D
K) = b;' (10.6)
where D is the molecular diffusion coefficient, and similarly

D
Kg = b. (10.7)
g
It has been shown that, if molecules are spherical, molecular diffusion

coefficients in solution are inversely proportional to molecular diameters d,
so that
(10.8)
where S indicates the toxic substance and 0 oxygen. If data on the diffusion
coefficients or molecular diameter for the component are not obtainable, the
molecular diameter can be estimated from the critical volume v., , since
nd 3 v., v.,
6 = 2N or 3N' (10.9)
where N is the Avogadro number; the molecular diameter for oxygen, dO, is
0.298 nm.
H~ can be estimated from solubility and vapor pressure:
HS=~ (10.10)
c Swo'
where P s is the vapor pressure ofS in pure form and Swo is the solubility in water.
When data for the considered component are not available, data for a related
component can be used.
10.2.3. Sorption
Sorption of toxic components on to suspended matter, sediment, and biota is a
very important process. Available data might fit either the Langmuir or
Freundlich adsorption isotherms. The latter are more generally used (Smith
et al., 1977; and for heavy metals, Reimer and Krenkel, 1973):
(10.11)
where
Ss is the weight of the component sorbed per gram of sorbent,
Sw is the weight in each liter or milliliter of solution,
K, n are constants.
400
10 6
...c
.~
u
i:Q)
10 4
0
u
c
0
.~
.~
"-
co
a..
o
10 2
o
10- 8 10- 6
1
Solubility (g ml- )
FIGURE 10.3 Partition coefficient plotted against solubility of Coyote Creek sediments
(K p • 0 measured) and of a mixed population of bacteria (K b .... measured).
At low substrate concentrations, n is often close to unity and K becomes a

partition coefficient.
Smith et al. (1977) have shown, in a limited number of case studies, that for a
given sorbent the logarithm of the partition coefficient and the logarithm of
the solubility are linearly related (Figure 10.3). Although this relationship seems
to be generally valid, compounds that interact by ion exchange probably
would not fit this plot.
10.2.4. Chemical Oxidation

Oxidation of toxic organic components may be important under some en-
vironmental conditions. Where it is of importance, a first-order reaction scheme
seems to give an acceptable, accurate description.
10.2.5. Photolysis
Photochemical transformation is a significant process for many toxic com-
ponents (Wolfe et al., 1976; Zepp et al., 1977). The rate of absorption of light,
401
I., by a chemical is determined from
(10.12)
where
S is the concentration of the chemical (mol 1- 1),

l is the molecular light extinction coefficient,
I A is the intensity of the incident light,
k. = (1..
By multiplying I. by the quantum yield 4>, which is the efficiency of converting

absorbed light into chemical energy, we find the rate of direct photolysis:
(10.13)
The photochemical transformation is a first-order reaction, where k p is

dependent on the intensity of the incident light. Zepp et al. (1977) have demon-
strated how the half-lives for photolysis vary with the season because of varia-
tion in lA' Wolfe et al. (1976) have suggested that ( and 4> be measured in
laboratory experiments so that k p = f(I A) can be calculated from these values
as a function of the hour, day, season, and latitude.
10.2.6. Hydrolysis
Hydrolysis of organic compounds usually results in the introduction of a
hydroxyl (-OH) group:
RX + H 20 -----+ ROH + HX. (10.14)
The kinetics of hydrolysis can be expressed as:
where kh , kB , kA' and kN are rate constants. Few data on kh as a function of pH

are available. However, Wolfe et al. (1977) give kinetic data for methoxychlor
and DDT and other information can be found in Wolfe et al. (1976).
10.2.7. Ionization and Complexation

These processes are rapid and can only be included in the model by use of
equilibrium expressions. The equilibrium constant can often be found in one of
the several handbooks containing the relevant data.
402
10.2.8. Biodegradation
As mentioned in section 4.1.2, the rate of biodegradation can be described by
means of the Monod kinetic equation:
S
P = Pmax K +S (10.16)
s
_ dS = !!.- X = Pmax SX (10.17)
dt Y Y Ks + S
dX
Cit = pX, (10.18)
where
S is the concentration of substrate,
P is the specific growth rate,
Y is the cell yield,
X is the biomass per unit volume,
Ks is the half-saturation constant.
For many common substrates, K s is of the order of 10- 1 pg ml- 1 . If this is

considerably higher than S, the disappearance of substrate is a first-order
process in both X and S. The maximum growth rate Pmax and K s are known for
some of the more important processes (J0rgensen, 1979b).
10.2.9. Interaction Between Biota and Environment

Pollutant uptake is related to the flow of energy into the organism in the forms
of food and oxygen. The balanced energy equation states that the energy of the
ingested ration, corrected for fecal and non fecal losses, er R, is equal to the meta-
bolic rate RESP plus the growth rate d W /dt:
erR = RESP + dW/dt. (10.19)
It is generally accepted that
(10.20)
where a and b are constants and W is the weight of the organism; b is close to
0.75-0.8 and independent of the level of metabolism (Norstrom et ai., 1976).
The growth rate can be written as a constant MY(n) times wg, where 9 is a
constant.
Bioaccumulation of toxic compounds, dTC/dt, is proportional to the total
respiration Q (uptake from water) and to the uptake from food:
dTC
----cIt = eprCprR + epwC pw V, (10.21)
403
where
Cpf is the concentration of pollutant in the food,
epf is the efficiency of pollutant uptake from food,
C pw is the concentration of the toxic compound in the water,
epw is the efficiency of pollutant transfer from water,
R is the ingestion rate,
V is the respired volume of water.
V is inversely proportional to the oxygen concentration and proportional to

RESP.
Body clearance of toxic compounds has been shown to follow a first-order
reaction scheme (Matida et al., 1971). The rate of toxicant clearance can be
written as:
dTC
-dt= k cI ·TC·Wf , (10.22)
where the clearance coefficient kcl depends mainly on the type of toxic substance,
and f is a constant between -0.2 and -0.8. All the parameters mentioned
above are dependent upon temperature and oxygen concentration (Jorgensen,
1976).
Since many parameters are required in this submodel to represent the great
number of toxic and biotic substances, some parameter generalization would be
very useful where the parameters are not available. For ef , an average value of
0.67 can generally be used; epf and e pw are highly dependent on the toxic sub-
stance, but only slightly dependent on the biota. The growth rate and metabolic
rate per unit weight can be found as functions of weight from Figures 10.4 and
10.5 if there is a lack of data.
The excretion rate as a function of length often follows the metabolic rate,
which is confirmed by Thomann (1978) for polychlorobiphenyl (PCB)
(Figure 10.6) and by Jorgensen (1979a) for cadmium (Figure 10.7). There seems,
furthermore, to be a proportional relationship between kc1 and water solubility,
although the amount of statistical material to support this observation is not
yet sufficiently large to allow generalization.
10.2.10. Migration of Trace Metals Across Water-Sediment Interface

Heavy metals accumulate in sediment, and it is therefore of special interest to
include a detailed submodel for migration of such trace metals across inter-
faces between water and sediment. Brooks et at. (1968) studied the equilibrium
between sediment and interstitial water in basins of Southern California border-
land and found that none of the metals considered could exist in solution in the
amounts measured if they were bound as simple sulfides (Table 10.1). It seems
404
0
Cll
+'"
...
co
..c
~
...Cl0
Cl -4
-12 -8 -4 4 8
Ig W
FIGU RE lOA The relationship between the intrinsic rate of natural growth (cal g- 1 h - 1)
and weight (g) for various organisms.
• ~ Unicellular
.""".... ...
2
~.
~
... ~ ...
Cl 0 .. ':~ Heterotherms
·~6
~
, , Homeotherms
-2 ,
,0
0', ,
,
'--.0
-16 -12 -8 -4 0
Ig W
FIGURE 10.5 The relationship between metabolic rate r (day- 1) and weight (g) for
various organisms.
TABLE 10.1 Solubilities of Metals (Brooks et at., 1968).

Solubility in Solubility in Seawater Actual
S2- -Free Water with IS < 10- 3M Concentration
(ppb) (ppb) (ppb)
CdS 3.6. 10- 29 8.10- 4 5.10- 12 0.2-8

CoS 3.0. 10- 20 0.01 2. 10- 9 0.4-3
CuS 8.5· 10- 45 6.10- 12 6.10- 23 0.7-14
FeS 3.7. 10- 29 39 3.10- 2 1-69
NiS 1.4.10- 24 0.08 1 . 10- 7 2-16
ZnS 1.2.10- 23 0.3 9.10- 7 1-152
405
2
10
9(DDT)
I I-
>
ro
1:l
..:.:: c
It-- o
....11l ''::;
U
~ :::J
1:l
c ~
0 l-
''::;
- 50 <f2..
~
u -100 g
CO
U
c...
X
11l 10- 2 1-
-o
I--
I I I I I
10- 3 10- 2 10-1 1 10
Organism size (mm)
FIGURE 10.6 Variation of PCB excretion rate with organism size.
'.6
Organism size (mm)
FIGURE 10.7 Rate of excretion of cadmium (day - 1) plotted against the length of
various organisms. I, man; 2, mice; 3, dogs; 4, oysters; 5, clams; 6, phytoplankton.
TABLE 10.2 Major Soluble Metal Species in Interfacial Seawater from Equilibrium Calculations.
Trace En vironmental Assumed Solubility Major Soluble Complex (at pH 7.5)
Metal Condition Controlling Product K,p
Solid (In pK,p)" Major Species Formation Percentage of Total
Constant Soluble Concentration
fJ (In Ii)'
Cd(lI) Oxidizing CdCO, 13.59 CdCI+ 2.69 56.5
CdCI~ 2.69 15.2
CdCl,i- 2.91 10.0
.j::>. CdCl6-
0 2.25 9.1
0"- CdCl] 2.78 9.0
Cd(ll) Reducing CdS 26.96 Cd(HS)~ 15.21 97.2
Cd(HS), 17.09 2.2
Cd(HS),i- 19.26 0.1
Cd(HS)+ 7.94 <0.1
CdCI+ 2.69 <0.1
CrOll) Reducing Cr(OHh 33.52 Cr(OH). 33.12 90.2
Cr(OHj; 19.15 9.2
Cr(OH)2+ 10.63 0.1
Cr(citric acid)" - 5.55· <0.1
Cr(serine)2 + 8.79 <0.1
Cu(lI) Oxidizing CU2CO,(OH)2 33.16 Cu[B(OH)4J~ 12.55 33.6
CuCO'i 6.77 30.1
Cu[B(OH)4J + 7.13 27.7
CuC!+ 1.58 4.6
Cu(CO,)~- 10.01 3.1
Cu(lI) Reducing CuS 36.38 Cu(HS), -4.40· 99.4
CuS(HSW -4.40· 0.5
Cu(histidine) + 1l.71 <0.1
Cu(histidine)~ 19.73 <0.1
Cu[B(OH)4J~ 12.55 <0.1
Fe(III) Oxidizing Fe(OHh 39.29 Fe(OH). 35.29 97.1
Fe(citric acid)- 28.14 1.5
Fe(OH){ 20.37 1.2
Fe(p-OH benzoic acid) + 15.2 <0.1
Fe[B(OH),E 15.6 <0.1
Fe(II) Reducing FeS 16.90 Fe(citric acid)2. 18.64 86.6
FeHPO~ 7.03 9.0
Fez; 2.4
FeCI; 0.38 0.9
FeCI~ 0.81 0.7
Hg(II) Oxidizing HgCl z 15.10 HgCll+ 15.6 80.4
HgO 26.24" HgCI; 14.8 15.4
HgCI~ 13.9 4.1
Hg(OH)~ 22.4 <0.1
HgCIBrO 2.01" <0.1
Hg(ll) Reducing HgS 53.89 HgS~- 0.57" 98.5
~
0 Hg(cysteine)o 46.2 1.2
-.l -3.70"
HgS(HS)~- 0.2
Hg(HS); -3.50" <0.1
HgS(HzS)~ -4.25" <0.1
Mn(Il) Oxidizing Mn(OH)z 12.19 MnCl-t 1.06 38.1
MnC0 3 9.20 MnCI~ 1.52 34.0
MnO z 0.92" Mn H 17.2
MnCI; 1.11 7.4
MnHCO; 1.8 2.1
Mn(II) Reducing MnCI+ 1.08 38.2
MnC0 3 9.20 MnCI~ 1.52 34.1
MnS (pink) 9.70 Mn z + 17.2
MnS (precipitate) 15.70 MnCI; 1. II 7.4
MnHCO; 1.8 1.7
Ni(II) Oxidizing NiC0 3 8.2 NiH 47.0
Ni(OH}z 14.81 NiCI+ 0.72 38.4
(fresh) NiCI~ 0.70 10.0
Ni(OH)z 17.31 NiSO~ 2.18 4.2
(aged) Ni(OH)+ 4.70 0.3
Ni(II) Reducing NiS(tl) 18.5 Ni(cysteine)" 10.48 99.0
NiS(i') 25.7 Ni(tyrosine)~ 10.36" <0.1
Ni(histidinet 9.79 <0.1
Ni(hjstidine)~ 17.76 <0.1
NiH <0.1
(conrilllled over)
Trace Environmental Assumed Solubility Major Soluble Complex (at pH 7.5)

Metal Condition Controlling Product K,p
Solid (1n pK,p)" Major Species Formation Percentage of Total
Constant Soluble Concentration
{i (In {i)'
Pb(lI) Oxidizing PbCO, 13.30 PbCO~ 7.4 64.6

PbCI~ 3.02 28.7
+:0- PbCI+ 1.60 2.4
9.89 2.0
o00 Pb(CO,W
PbCI~ 1.78 1.0
Pb(lI) Reducing PbS 27.65 PbS(HS)- ·-6.9 b 83.9
PbS(H 2S)o -6.8 b 16.0
PbCO~ 7.4 <0.1
PbCI~ 3.02 <0.1
PbCI+ 1.60 <0.\
Zn(Il) Oxidizing ZnCO, 10.65 Zn H 38.1
ZnSiO, 21.03 b Zn(OH)~ 12.89 29.6
ZnCI+ 0.43 18.0
ZnCJ~ 0.61 6.6
ZnSO~ 2.27 4.3
Zn(lI) Reducing ZnS 22.60 Zn(HS)i -3.0 b 99.2
Zn 2 + 0.2
Zn(OH)~ 12.89 0.1
ZnCI+ 0.43 <0.1
ZnCI~ 0.6\ <0.1
e
" All values of pK. p and {J are corrected to the conditions of 1 = 0 and T = 12 C using the Davies modification of the Debye-Huckel
expression and the van't Hoff equation. When !!.H, (enthalpy of reaction) was unknown. K. p and {J values for the other temperature
(usually T = IS-25°C) were used.
b HgO(s) + H 20 ¢ HgH + 20H-: Mn0 2(s) + 2H+ ¢ Mn 2 + + t02 + H 20: ZnSiO,(s) + H 20 ¢ Zn H + 20H- + Si0 2(s);
Cr J + + citric acid (H,L) ¢ CrL ° + 3 H +: CuS(s) + H 2S(aq) + HS - ~ Cu(HS)i : CuS(s) + 3 HS - ¢ CuS(HS)~ - ;
HgH + 0- + Br- ~ HgCIBrO; HgS(s) + S2- ¢ HgS~-: HgS(s) + 2HS- ~ HgS(HS)~-: HgS(s) + H 2S(aq) + HS- ¢ Hg(HS)i;
HgS(s) + 2 H 2S(aq) ¢ HgS(H2S)~; Ni 2 + + tyrosine - (or HL -) ¢ Ni(HL)~; PbS(s) + H 2S(aq) :0= PbS(H 2S)o;
PbS(H 2S)o ¢ PbS(HS)- + H + ; ZnS(s) + H 2S(aq) + HS - ¢ Zn(HS)i .
409
most probable that complexes form and solubilize the metals even in the
presence of sulfide.
In a reducing environment Fe and Mn are released to the interfacial seawater
in significant quantities, while concentrations of other metals, such as Cd, Cu,
Ni, Pb, and Zn, decrease in comparison with the original seawater background.
These changes are related to the formation of sulfides. Lu and Chen (1977)
suggested that these sulfide solids (except Cr, which might form Cr(OHh)
would regulate the free metal ion concentration. They set up an equilibrium
model controlled by the sulfides (except for chromium, which was controlled by
Cr(OHh). The results of the model are shown in Table 10.2 for T = 12°C,
I = 0.7, where I is the ionic strength. In the model the total concentrations of
ligands in seawater were used to calculate the concentrations of complexes
formed. Table 10.3 lists the typical total concentrations of known ligands for
which data are available on stability in interstitial water. These concentrations
were used in the Lu and Chen model. The model was validated for three types of
sediment: silty clay, silty sand, and sandy silt, with different concentrations of
heavy metals. After long-term incubation, the concentrations of Cd, Hg, Mn,
Ni, Pb, and Fe in the clay sediment were found to be close to the model calcula-
tions. Therefore, it seems that the formation of solids and the complex formation
control the solubility of trace metals. By use of this model as a base, it has been
shown that sulfide complexes are the most important soluble species for Cd,
TABLE 10.3 Concentrations of Ligands in Interstitial Seawater

(Reducing Conditions). A factor of 20 was applied because the concentra-
tions of Nand P compounds released into interfacial water are about 20
times higher than those in average seawater.
Ligand Concentration (M) Ligand Concentration (M)
Total soluble carbonate 8.10- 3 Glutamic acid 1.09.10- 6

Total soluble borate 6.10- 4 Glycine 4.0 . 10- 6
Total soluble silicate 5.10- 4 Glycollic acid 7.89. 10- 6
Ammonium ion 4.10- 4 Histidine 2.58 . 10- 7
Nitrite 7. 10- 7 p-hydroxybenzoic acid 4.35 . 10- 7
Nitrate 1.4.10- 6 Hydroxyproline 3.05 . 10- 7
Orthophosphate 2.5.10- 5 Lactic acid 1.11.10- 7
Sulfide 5.10- 4 Lencine 7.63 . 10- 7
Sulfate 2.8.10- 2 Lysine 6.85 . 10- 7
Fluoride 8.10- 4 Malic acid 1.49.10- 5
Chloride 0.5.10- 6 Methionine 1.34 . 10- 7
Bromide 8.10- 6 Ornithine 8.47.10- 7
Iodide 5. 10- 7 Proline 1.74.10- 7
Acetic acid 2.10- 4 Serine 1.90. 10- 6
Alanine 1.12.10- 6 Threonine 8.40.10- 7
Arginine 1.15.10- 7 Tryptophane 9.80.10- 8
Aspartic acid 1.2.10- 6 Tyrosine 5.24. 10- 7
Citric acid 1.04.10- 6 Valine 5.13.10- 7
Cysteine 1.65.10- 7
410
Hg, and Pb, organic complexes for Fe and Ni, chloride complexes for Mn, and
hydroxide complexes for Cr.
Under oxidizing conditions the controlling solids may change gradually
from metallic sulfides to carbonate, hydroxide, oxyhydroxide, oxide, and
silicate solids, resulting in a change of the solubility of trace metals. The solubility
of Cd, Cu, Ni, Pb, and Zn increases owing to the formation of more soluble
solids, while the solubility of Fe and Mn might decrease because of the formation
of high oxidation states, less soluble oxides, or hydroxides. Through thermo-
dynamic calculations using one-twentieth of the concentration in the reducing
conditions of organic ligands and [NH 3 ] = 0, [NO;] = 0, [S2 -] = 0, [total
carbonate] = 1O- 2,[totalsolublesilicate] = 4·1O- 4 ,and[l:POJ-] = 2.10- 6,
it was found that only Zn and Cr solubilities are close to the experimental data.
Cd, Cu, Ni, and Pb were far below the equilibrium concentrations and Fe and
Mn far above. This discrepancy may be explained by adsorption.
From these considerations, it seems that metal migration is mainly controlled
by the chemistry of the water, with the redox condition as one of the principal
factors, except for Cr and ionic Hg. The released amounts of Cd, Cu, Ni, Pb, and
Zn increase as the redox condition becomes more oxidizing, while the opposite
is true for Fe and Mn. Equilibrium thermodynamic calculations seem to hold
under reducing concentrations, but equilibrium constants for formation of
humic complexes must be included for Cu, Zn, and Fe.
TABLE 10.4 Stability Constants (- pK) of

Complexes in Water (Zitko and Carson, 1976)
Cation Ligand
Glycine ATP Glutathione Acetic Acid
Ca 2 + 1.31 3.60 0.39

Mg 2 + 3.44 4.00 0.82
Zn 2 + 5.52 4.85 8.30 1.57
Cd 2 + 4.80 )0.50 1.70
Cu 2 + 8.62 6.13 2.24
If data are lacking, typical seawater concentrations for ligands might be used
(Table 10.3). Furthermore, these calculations can be used under oxidizing
conditions for Zn and Cr, while further examination is necessary for other
metals under these conditions to find the equilibrium description, as sorption
phenomena constitute a major factor. Lu and Chen (1977) considered only
seawater, but other ligand formations may playa role for other aquatic eco-
systems. Other possible complex formations are listed in Table 10.4.
411
10.3. MODELS
10.3.1. Food ChainjWeb Models

The literature on the transport of hazardous substances generally begins by
assigning to the ecological system a series of compartments positioned in space
and time. The concept of a compartment arises from a grouping of ecological
properties, species, and types (e.g. phytoplankton, zooplankton, fish). The
continuum of the environment is replaced by finite, discrete, interacting trophic
levels. The details of each compartment must be specified, and attention is often
directed toward a part of the ecosystem, depending on the aims of the model.
The ecological concepts of compartment analyses have been reviewed by
Patten (1971) and have already been applied in conjunction with eutrophica-
tion models. Figure 4.2 (p. 118) is based upon these concepts, used here for
modeling the processes of nitrogen in lake systems.
Thomann et al. (1974) modeled the distribution of cadmium in western
Lake Erie, applying a food chain model with a basic structure similar to that of
eutrophication models. The equations are mass balances around each discrete
trophic level at some position in space. The mass of toxicant per unit biomass at
the considered level is used as the state variable. The rates of uptake of heavy
metals directly from water by zooplankton and fish are not considered. The
model includes, in addition to the model for cadmium, a general eutrophication
model.
Miller's (1979) model for mercury in water considers six compartments:
water, sediment, suspended matter, invertebrates, plants, and fish. Other
compartment models have been constructed in food chain studies of hazardous
substances. Gillett et ai. (1974) modeled the movement of pesticides; Hill et ai.
(1976) considered the dynamic behavior of vinyl chloride in aquatic ecosystems;
Lassiter et al. (1976) focused on the fate of mercury in aquatic systems; and
Lassiter (1978) considered the dynamics of methyl parathion and benzothio-
phene. These models take into account the following processes (having rates V).
(A) Volatilization Vy is described as a first-order reaction:

Vy = ky[COMP],
where ky is the rate constant and COMP the concentration of the
component.
(B) Photolysis Vp is described as a first-order reaction:
Vph = kph[COMP],
where kph is the rate constant.
(C) Oxidation Vo is described often as a second-order reaction:
v" =ko [02][COMP],
where ko is the rate constant and [0 2] the oxygen concentration.
412
(D) Hydrolysis VH can proceed with hydrogen ions, water, and hydroxyl
IOns:
where k H , k H20 , and kOH are rate constants.

(E) Microbial biodegradation Vb is described by means of a Michaelis-
Menten expression:
= kb[COMP] [M] ( 10.23)

Vb ks + [COMP] .,
where kb is a rate constant, [M] the activity of the microbiota, and k s

a half-saturation constant.
(F) Adsorption is considered a fast reaction. The definition may vary
according to the situation as well as to the chemical substance. Fast
reactions can be considered to be equilibria for which the computations
are algebraic, whereas slow reactions (A to E) are best described by
differential equations. A Freundlich adsorption isotherm is often used
to indicate the equilibrium:
[CAD] = a[COMp]b, (10.24)
where [CAD] is the concentration (e.g. mg per kg dry matter) of the

compound adsorbed on suspended matter or sediment, and a and b
are constants.
(G) Ionization and complexation are fast chemical reactions, for which
the computations are algebraic. Equilibrium is described by a mass
equation. As an example (Lassiter, 1978), let us consider a chemical
that can exist in three ionic states: uncharged, and singly and doubly
positively charged. In contact with particles (P) (suspended or sedi-
mentary) the material can adsorb physically, or either charged form can
attach to particles by cation exchange. Thus, the chemical CE can
exist in six states:
Equilibrium
constant
+ H+
CE ~-- CEH+ K1
CE + P ~
CEP Ks
CEH+ + H+ CEH~+ K2 (10.25)
CEH+ + P ~-- CEH-P K st
CEH~+ + P ~ CEH 2 -P K S2
413
An algebraic equation can be written to represent equilibrium con-
ditions for each reaction. An equation representing conservation of
mass is also used, so that an algebraic computation for the six forms
is possible. One convenient form for computation is the expression of
each state as a fraction of the total chemical present; e.g., the expression
for the unionized fraction lX o is
IX O = 1 + Ks[P] + K1[H+] + K1Ks1[H+][P] + K 1K 2[H+]2
+ K 1 K 2K s2 [H+]2[P], (10.26)
and so on for the other five fractions. Most chemicals do not present as
difficult a problem as the one in the example. The importance of
computing the ionized species is that kinetic reactions vary with species.
Processes A to G contain explicit, quantifiable reference to the environ-
ment, such as pH, concentration of suspended matter, microbial
activity, water velocity, dispersive mixing, and geometry of the water
body.
(H) Interaction between the biota and the environment must also be
considered. Generally, the models are structured in trophic levels 2-4,
but it is also possible to consider a food web.
The processes determining the concentrations of toxic substances
in an organism are shown in Figure 10.1. The organisms ~ake up
toxic substances through feed and directly from the water. The first
process is described as Jlmaxf(temp)' YT· j(feed), where Jlmax is the
maximum growth rate, j(temp) is a temperature function, YT is the
utility coefficient for the toxic compound in feed, and j(feed) represents
the concentration of feed. j(temp) and j(feed) were described in
detail in Chapter 4. YT is strongly dependent on the toxic component.
The uptake directly from water is often described by a concentration factor,

which is, however, dependent on the size of the organism. The excretion rate
v" can be described by a first-order reaction:
v" = ke[COMP]o, (10.27)
where the rate constant k e depends on the size of the organism and [COMP]o is
the concentration of the toxic component in the organism, e.g. based upon dry
weight. The other processes will not change the concentration of the toxic
component in the organism, and the equations mentioned in Chapter 4 can be
applied directly.
A compartment model requires, as demonstrated, detailed knowledge of the
processes of transportation of toxic components in an aquatic environment.
We know that the equations shown give a good description of the individual
processes, but we know the equation parameters for only a limited number of
processes. However, some general rules can be set up, as shown in section 10.2.
414
The state of the art is that, by means of this type of model, we can give an
approximate description of the fate of a toxic component in an aquatic eco-
system, although more needs to be known about parameter values, and especially
about the relationships between environmental factors and these values.
Furthermore, it is noticeable that no feedback mechanisms have been taken into
consideration, as our present knowledge of these is rather limited.
10.3.2. Simplifications
As mentioned in the introduction, the models described above are unwieldy
and can only be calibrated and validated against comprehensive sets of data,
which are available only in few cases. Consequently, a management model needs
a higher degree of simplicity, in one way or another.
Thomann (1978) reduced the number of equations by discretizing ecological
space. He considered the trophic position as one-dimensional:
ljJ = Sf(L, t), (10.28)
where ljJ is measured in milligrams of toxicant per liter and per unit length of
organism, S is in milligrams of toxicant per liter, L is the trophic length, and I is
time. This modeling approach uses up to three spatial dimensions plus one for
the ecological dimension. The governing equation for ljJ in a completely mixed
water volume is
v(aai + a~~ljJ) = QljJin - QljJ + W'(L) - S'(L). (10.29)
We then make several assumptions. (I) The uptake of toxicant is proportional

to the water concentration C pw and (2) to the biomass along L; (3) excretion
of toxicant occurs from the entire food chain according to first-order kinetics
on ljJ; and (4) ljJin = 0, i.e. no input of toxicant mass is associated with the food
chain. The governing equation then becomes:
V(aljJ + aVLljJ) = -Q(L)ljJ - K(L)VljJ + kJL)m(L)VC pw (10.30)

at aL
and for the water phase:
V dC
PW
d = W - QC pw - ,1,VC pw + V foo K(L)ljJ dC pw - ve pw fookJL)m(L)dL.
t L, L1
(10.31)
This model seems able to simplify models of toxic substance distribution in

aquatic ecosystems.
415
Baccini and Imboden (1977) developed a model using a thermodynamic
approach by considering formation of certain stable metal species in a set of
possible inorganic and organic ligands. The proposed model treats the plankton,
allochthonous particles, and chemically undefined dissolved substances as
uniform ligands in a given set of conditions.
A third simplification was suggested by J0rgensen (1979a). This approach
includes the processes shown in Figure 10.1 for the considered trophic levels.
The following differential equations for biomass and the concentration of a
toxic substance in the biomass can be evaluated (the notation is in Table 4.13,
pp. 140-142) (taken from J0rgensen, 1979a).
dBIO(n)
dt = BIO(n)(MY(n)· YF(n) - MORT(n) - RESP(n) - MY(n + 1»
(10.32)
dTOX(n)
dt = BIO(n)(MY(n)· YT(n) . yen - 1) - MORT(n) . yen)
- EXCR(n)· )'Cn) - MY(n + I)· yen) + UT(n)· TOX(O». (10.33)
YF is the efficiency for uptake of food and UT (day - I) is the rate of uptake
from water. Since
TOX(n)
yen) = BIO(n) (10.34)
and
'( ) _ TOX'(n)· BIO(n) - BIO'(n) . TOX(n)
(10.35)
y n - (BIO(n»2 '
we have
dy(n)
~ = MY(n)(y(n - 1)· YT(n) - y(n)· YF(n»
+ y(n)(RESP(n) - EXCR(n» + UT(n) . TOX(O). (10.36)

Only a few data, however, are available on the uptake rate, while many references
give information about the concentration factor CF at steady state. dBIO(n)jdl,
dTOX(n)/dt, and dy(n)/dt are all equal to zero, and under the circumstances of
the experiments on which the CF value is based, MORT(n), MY(n + I), and
yen - 1) are zero as well:
MY(n)· YF(n) - RESP(n) = 0 (10.37)

and
- EXCR(n) . yen) + UT(n)· TOX(O) = 0 (10.38)
416
or
y(n) _ UT(n) _ CF
(10.39)
TOX(O) - EXCR(n) - .
n = 0 corresponds to the water phase in these equations. From eqn. 10.39,

UT(n) or EXCR(n) can be found if EXCR(n) or UT(n) and CF are known.
One purpose of modeling the distribution of toxic substances is to find a
relationship between the input of toxic substances to the aquatic ecosystem
and the approximate concentrations at different trophic levels. However, the
concentration will show seasonal variations, probably reaching a maximum
at the highest growth rate (summer) (Gallegos and Whicker, 1972; Betzer and
Pilson, 1974). As the object of the model is to find the maximum concentration
rather than to simulate the seasonal variations, it is suggested that, in addition
to modeling the conditions of maximum growth, the concentration should be
determined for different growth rates.
The following state variables are used.
(1) The concentration of soluble toxic substances in water, TOX(O):

the differential equation is based upon:
dTOX(O) = I sources - I sinks + inflow - outflow. (10.40)

dt
(2-5) The concentration of toxic substances in trophic levels 1-4: the

equations are as indicated above (see eqn. 10.36). J0rgensen (l979a)
discussed the dependence of concentration on temperature.
(6) The equilibrium between water and the six state variables: the
concentration of adsorbed material on suspended matter, y(O), can
be described by use of an adsorption coefficient KA, which must be
found for each case study:
y(O) = KA . TOX(O), (l0.41)

although some indication can be given on the basis of the literature.
In this context, y(n - 1) for filter feeders (see eqn. 10.36) should be
a weighted average of y(l) and yeO). An estimated ratio of the con-
centrations of phytoplankton and suspended matter should be used
for weighting.
(7, 8) The last two state variables cover the sediment: the concentrations
y(s) in the sediment and y(i) in the interstitial water. y(s) and y(i) are
related by equilibrium equations, which must be established in every
case study, although general expressions are available for some heavy
metals (Lu and Chen, 1977). If it is critical whether the conditions
are oxidizing or reducing, an oxygen model must be superimposed on
417
the heavy metal model. y(s) is found on the basis of:
dy(s) = y(s) + rat(O) . 1'(0) + rat(1) . 1'(1) _ SOL
dt BIOL ' (10.42)
where rat(O) is the mass ratio of suspended matter that has settled to
active sediment and rat(1) the mass ratio of settled phytoplankton
to active sediment, based upon dry matter. BIOL is the amount of
dry matter in the sediment. It is not constant, since the organic
matter decomposes at a rate dependent on the composition of the
sediment and the temperature. SOL is the rate at which the toxic
compound dissolves. The expression for SOL depends on the
equilibrium equation relating y(s) to y(i). Finally, y(i) and TOX(O)
must be related by a diffusion expression. However, the submodel
for the water-sediment exchange of toxic substances requires further
studies before a satisfactory description of the processes in the sedi-
ment can be set up.
10.3.3. Models Considering a Single Trophic Level

As mentioned in the introduction, it may often be possible to answer essential
management questions by considering only the trophic level in focus, e.g. fish.
The model for methyl mercury by Fagerstr0m and Asell (1973) illustrates this
modeling approach. The principles of this model are summarized below.
(1) Roach and pike are recruited to the model at an age of one year with a
prescribed body burden of methyl mercury (MM). At a prescribed age
the fate of the fish is no longer taken into account.
(2) Methyl mercury is gained via food according to:
fOOd) . (concentration) . ( assimilation ),
( intake of MM in food efficiency of MM
and via respired water according to:
volume of ) (concentration) ( withdrawal )

( respired water . of MM in water . efficiency of MM .
Food intake and volume of respired water are in turn calculated from
body weight and growth rate according to Winberg (1960).
(3) MM is lost by excretion, which is assumed proportional to body burden
and specific metabolic rate.
(4) All metabolic rates are adjusted with respect to temperature.
(5) A pike is assumed to have a maximum and a minimum length of prey
and both are linear functions of the length of the pike (Domanewski,
418
1962; Lawler, 1965). The optimum length is assumed to be the arith-
metic mean of the two limiting lengths and the frequency distribution
of prey is assumed normal, with the interval between maximum and
minimum lengths equaling four standard deviations. Superimposed on
this preference function is the survival function of the prey, so the final
sampling function is skewed. A prescribed fraction of the prey is
assumed to be pike, i.e. a certain amount of cannibalism is presupposed.
(6) The driving force for the model is the growth of the fish. Parameters for
both growth and survival are treated as constants, i.e. stable size and age
distributions are assumed. Deviations from this situation can be brought
about by letting the parameters be, for example, time-dependent,
temperature-dependent, or affected by feedback from the MM con-
centrations.
It was concluded that the model is able to mimic the static picture that is
common in nature: (a) MM concentrations in fish increase with age; and
(b) MM concentrations in predatory species exceed those in prey species by a
factor of 2-5.
Aoyama et al. (1978) set up a model for predicting heavy metal concentrations
in fish. The model considers the concentration factor (water/fish), the excretion
rate, and bioaccumulation through the food chain. The model has been calibrated
and validated; laboratory data were applied, with good results.
Leung (1978) developed a model of accumulation of pesticides in fish.
This model considers that the following processes increase the concentration of
pesticides in fish: (a) uptake associated with prey, which is ingested, (b) pesticide
entering through gills, and (c) pesticide adsorbed on to the body. The model
contains the following processes that decrease the concentration: (1) defecation
of pesticide, (2) excretion of pesticide, (3) loss due to release of products, (4)
transformation to a daughter product, (5) loss due to ingestion of fish by higher
predators, and (6) loss due to nonpredatory death. The model has been applied
successfully for DDT and methoxychlor.
Seip (1978) modeled the uptake of heavy metals by algae. The basis of his
model is the following equation for the amount of zinc, Zi' accumulated in age
class i during the time interval ~t:
(10.43)
where
Ui is the uptake rate in age class i,

Si is the secretion rate from age class i,
CA is the concentration of the heavy metal in algae,
Cw is the concentration of the heavy metal in water,
419
Ni is the biomass of age class i,
q is a constant.
The concentration of zinc in the algae is given by:
(10.44)
where C1 is the initial concentration of the heavy metal in algae. Data from
literature and from observations at a polluted and an almost unpolluted locality
were used in parameterization. The model was used to predict concentration
factors along a natural gradient of zinc concentrations in a Norwegian fjord.
The calculated values were in good agreement with observations.
Orlob et al. (1980) developed a model to simulate the fate of copper in a
marine environment. The model is based on the fact that the ionic form has the
toxic effect. As little as 10 /lg 1- I or even less of ionic copper has sublethal
effects on sensitive marine biota. Consequently, the model includes a quantitative
description of the processes that determine the concentration of ionic copper:
(1) formation of copper ion complexes, (2) adsorption of ionic copper on sus-
pended matter, and (3) adsorption of complexed copper on suspended matter.
The three forms, ionic copper, adsorbed copper, and complexed copper, are in
an equilibrium described by three equations.
A model developed for description of the chromium distribution in a Danish
firth (Mogensen, 1979) also demonstrates the use of simplifications based on a
knowledge of the actual processes in the ecosystem. Since Cr(IlI) has a very
low solubility, most of the chromium is accumulated in the sediment. The
model is able to describe the distribution of chromium in the water on the basis
ofthe analysis of several sediment cores. This pattern ofdistribution can again be
used to relate the discharge of chromium to the seqiment as a function of the
distance to the point ofdischarge (Mogensen and Jergensen, 1979). The accumu-
lations of chromium in algae and in benthic animals were included in the model;
other biological processes with chromium were considered insignificant.
Lam and Simons (1976) applied a model oflead nitrate spill to Lake Ontario,
using an approach similar to the two mentioned above. They represent the
free ion concentration, the concentration of precipitated lead, and the ability
of water to precipitate lead by state variables. The distribution is described by
means of a hydrodynamic model. The concentration of free lead ions is related to
the toxicities of algae, invertebrates, and fish.
Gromiec and Gloyna (1973) considered the transport of radionuclides in
water. The governing equation is the same as that used by Mogensen and
Jergensen (1979):
OC = E a c _ U ac
2
I. ;; K i( G(C) - C I ), (10.45)
ot ox 2
ax i= I
420
where
C is the concentration of the toxic substance,

E is the eddy diffusion coefficient,
U is the advection flow,
G is decay, a function of C.
The first two terms defined mixing characteristics and dilution, and the third
term covers uptake and release by various aquatic surfaces. This approach
excludes chemical processes such as those mentioned in points A to E and G and
simplifies the processes F and H. These simplifications are often allowed in
transport models of heavy metals, since the governing processes for heavy
metals are sorption on plants, including algae, on suspended matter, and on
sediment. Gromiec and Gloyna demonstrated how such a model can be applied
to the transport of radionuclides. A laboratory investigation of the sorption
processes is used as the basis for parameter estimation. However, generally a
degree of knowledge of the release of heavy metals from the sediment seems
necessary, since a substantial part of the heavy metals will accumulate in the
sediment.
The case studies by Gromiec and Gloyna and by Mogensen and J0rgensen
can be used to predict concentrations in the considered species but can be
included as submodels in total models as well. No case was studied by the use
of two or more models, so no result of comparison is available. However, from
the survey of the three classes of model it is apparent that the more compre-
hensive models, class I, will often require so much data and such detailed
knowledge of the individual processes in the environment that this type of
model will very often be omitted because of limited resources.
Whether class II or class III models should be selected for a given problem
cannot be answered generally. If the problem is limited to one trophic level, or
to one species or group of species, a class III model will suffice and will give
more details too, while a problem that requires the entire ecosystem to be
considered demands a class II model.
10.4. CONCLUDING DISCUSSION
We are today able to model the distribution and effect of only a small number of
the overwhelmingly many compounds that have or might have impact on the
environment. Workable models have been set up for some pesticides (DDT,
methoxychlor), PCB, and some heavy metals (mercury, lead, and chromium).
The models are able to describe discharge in the worst situation with acceptable
accuracy. The concentrations of toxic substances in fish and the food chain
accumulation can also be predicted fairly well. In other words, models of
421
toxic substances can be used as management tools for a limited number of
compounds, although some of the processes are not known in sufficient detail.
However, research has intensified during the last few years, so that more
knowledge has been gained about increasing numbers of compounds and
processes, as well as the parameters used to describe them.
Although a general model cannot be developed for all compounds, more
research must be devoted to general principles, if the task of modeling the
distribution and effects of the more important toxic substances is not to be
insuperable. This chapter has demonstrated some basic procedures, but more
experience in the application of models of toxic substances is required before
we can develop models for classes of components (the same type of equation
for all components, but with different parameters that depend on physico-
chemical data, such as solubility). It is not too optimistic to expect that this point
will soon be reached, but simultaneous efforts in the development of total
models and research into the individual processes are needed.
REFERENCES
Aoyama, I., Yos. Inoue, and Yor. Inoue (1978). Simulation analysis of the concentration process
of trace heavy metals by aquatic organisms from the viewpoint of nutrition ecology. Water
Research 12:837-842.
Baccini, R., and D. Imboden (1977). A copper model for an aquatic ecosystem. Paper presented at
SlL Con/aenee, Copenhagen.
Betzer. S. B.. and M. E. Q. Pilson (1974). The seasonal cycle of copper concentration in Busyeon
cilnaliculatum. Biological Bulletin 142: 165-175.
Brooks, R. R., B. J. Presley, and I. R. Kaplan (1968). Trace elements in the interstitial waters of
marine sediment. Geoehimica et Cosmoehimica Acta 32: 397-414.
Domanewski, L. N. (1962). Biulleten lnstituta Biologii Vodochranilischtsch 12: 50 (German trans-
lation no. 160 from Swedish Natural Science Research Council).
Fagerstrom, T., and B. Asell (1973). Methyl mercury accumulation in an aquatic food chain. A
model and implications for research planning. Ambio 2: 164-171.
Gallegos, A. F .. and F. W. Whicker (1972). Radio cesium retention by rainbow trout as affected by
temperature and weight. Report of National FNF Series 100, 115642: 1-25.
Gillett, J. W .. et al. (1974). A conceptual model for the movement of pesticides through the environ-
ment. National Eneironmental Research Center, US EI11'ironmental Protection Agencr, Corvallis,
OR Report EPA 660/3-74-024, p. 79.
Gromiec, M. J., and E. F. Gloyna (1973). Radioactivity transport in water. Final Report No. 22 to
US Atomic Energy Commission, Contract AT(I I-I )-490.
Hill, J., et al. (1976). Dynamic behavior of vinyl chloride in aquatic ecosystems. Eneironmental
Research Laboratory, US EnL'ironmental Protection Agency, Athens, GA Report EPA 600/3-76-001,
p.64.
Jorgensen, S. E. (1976). A model of fish growth. Ecological Modelling 2: 303-313.
Jorgensen, S. E. (1979a). Modelling the distribution and effect of heavy metals in an aquatic
ecosystem. Ecological Modelling 6: 199-223.
Jorgensen, S. E. (ed.) (l979b). Handbook of Eneironmental Data and Ecological Parameters (Copen-
hagen: International Society for Ecological Modelling).
422
Lam, D. C. L., and T. J. Simons (1976). Computer model for toxicant spills in Lake Ontario.
Enrironmental Bioyeochemistrr vol. 2 Metals transfer and ecological mass balances, ed. J. O.
Nriago (Ann Arhor, M I: Ann Arbor Science), pp. 537 549.
Lassiter, R. R. (I97S). Principles and constraints for predicting exposure to environmental
pollutants. US EIlI'ironmental Protection Aqency, Carrallis, OR Report EPA II X-I27519.
Lassiter, R. R .. J. L. Malanchuk, and G. L. Baughman (1976). Comparison of processes deter-
mining the fate of mercury in aquatic systems. Proceedinys (II Conference on Enrironmental
Modeliny and Simulation. US Enl'ironmental Protection Aqency, Washinqton. DC Report EPA
600/9-76-016, pp. 619-623.
Lawler, H. G. (1965). Toxic substances in fish. Journal 01 Fisheries Research Board 0( Canada
23: 1358-1377.
Leung, D. K. (l978). Modeling the bioaccumulation of pesticides in fish. Center for Ecoloyical
Modeliny, Rensselaer Polytechnic Institute, Troy, NY Report 5.
Liss, P. S" and P. G. Slater (1974). Flux of gases across the air-sea interface. Nature 247: IS I-I S4.
Lu, J. C. S., and K. Y. Chen (I 977}. Migration of trace metals in interface of seawater and polluted
surficial sediments. Enrironmental Science and Technoloqy 11 : 174-182.
MacKay, D .. and Y. Cohen (1976). Prediction of volatilization rate of pollutants in aqueous
systems. Proceedings olS.l'mposillm on Nonhi%gical Transport and TransfOrmation oj'Pollllt<Jnts
on Land and Water, 11-13 May, National Bureau olStandards, Gaithershury, MD.
Matida. Y, H. Kumada, S. Kimura, Y. Saiga, T. Nose, M. Yokote, and H. Kawatsu (1971).
Toxicity of mercury compounds to aquatic organisms and accumulation of the compounds by
the organisms. Bulletin 01 Freshwater Fisheries Resmrch Lahomtorv, Tokyo 21 : 197 -225.
Miller, D. R. (1979). Models for total transport. Principles oIE('otoxicoloq.l' Scope vol. 12, ed. G. C.
Butler (New York, NY: Wiley), pp. 71-90.
Mogensen, B. (1979). Chromium pollution in a Danish fjord. Licentiate Thesis, ROl'al Danish School
of Pharmacy, Copenhagen.
Mogensen, B., and S. E. Jorgensen (1979). Modelling the distribution of chromium in a Danish
firth. Proceedinys 01 ist international Conterence on State 0( the Art in Ecoloqical Model/iny,
Copenhaqen, i978, ed. S. E. Jorgensen (Copenhagen: International Society for Ecological Model-
ling), pp. 367-377.
Norstrom, R. L et al. (1976). A bioenergetics-based model for pollutant accumulation by fish.
Simulation of PCB and methyl-mercury residue levels in Ottawa River yellow perch. Journal 01
Fisheries Research Board of Canada 33: 248-267.
Orlob, G. T., D. Hrovat, and F. Harrison (1980). Mathematical model for simulation of the fate
of copper in a marine environment. American Chemical Society, Ad1'Gnces in Chemistry Series
189: 195-212.
Patten, B. C. (ed.) (1971). Systems Analysis and Simulation in Ecoloyy vol. 1 (New York, NY:
Academic Press), 607 pp.
Reimer, R. S" and P. A. Krenkel (1973). Sorption phenomenon in the organics of bottom sedi-
ments. Heary Metals in the Aquatic Em'ironment, ed. P. A. Krenke! (Oxford: Pergamon), pp.
117-132.
Seip, K. L. (1978). Mathematical model for uptake of heavy metals in benthic algae. Ecological
Model/iny 6: 183-198.
Smith, J. H" et 01. (1977). Environmental pathways of selected chemicals in freshwater systems,
Part I. US Enrironmental Protection Ayency, Corml/is. OR Report EPA 600/7-77-113.
Thomann, R. Y. (1978). Size-dependent model of hazardous substances in aquatic food chain.
US EllI'ironmental Protection Aqency, Cor/'{/I/is, OR Report EPA 600/3-78-036.
Thomann, R. Y., et al. (1974). A food chain model of cadmium in western Lake Erie. Water Researt'h
8: 841-851.
Win berg, G. G. (1960). Fisheries Research Board 01 Canada Translation Series No. 194 (Nauchnye
Trudr Beloru.l'skol'O Gosut!arstrennol'O Unirersiteta imeni v.1. Lenina, pp. 1-253 (1956».
423
Wolfe, N. L., R. G. Zepp. G. L. Baughman. R. C. Fincher. and J. A. Gordon (1976). Chemical and
photochemical transformation of selected pesticides in aquatic systems. US Enl'ironmenlal
Proleclion Aqencr, Corral/is, OR Report EPA 600;3-76-067.
Wolfe. N. L., el ul. (1977). Methoxychlor and DDT degradation in water rates and products.
Enl'ironmental Sdenl'e and Technoloql' 11: 1077-1081.
Zepp, R. G .. el ul. (1977). Photochemical transformation of DDT and methoxychlor degradation
products DDE and DMDE by sunlight. Archil'es oFEnl'ironmenlal Conlamina/ion und Toxicology
6: 305-314.
Zitko, Y.. and W. G. Carson (1976). The mechanism of the effect of water hardness on the lethality
of heavy metals 10 fish. Chemisphere 11: 299-303.
CHAPTER 10: NOTATION
A interfacial area
a,b constants (adsorption isotherm)
(C t) C A (initial) concentration of heavy metal in algae
Cpf concentration of pollutant in food
C pw concentration of pollutant in water
Cs concentration in the liquid phase
C Si concentration at interface
Cw concentration of heavy metal in water
CAD concentration of adsorbed component
COMP concentration of component
dO diameter of oxygen molecule
DS, D, DO molecular diffusion coefficients
ef correction for fecal and nonfecal losses
epf efficiency of pollutant uptake from food
epw efficiency of pollutant uptake from water
E eddy diffusion coefficient
G(C) decay, a function of concentration
H, He Henry's law constants
1 ionic strength
1. rate of light absorption
1;, intensity of light
kc1 clearance coefficient
ke rate constant for excretion
ko rate constant for oxidation
k ph rate constant for photolysis
k, rate constant for volatilization
KS gas film mass transfer coefficient
K~1 liquid film mass transfer coefficient
Ks half-saturation constant
K,S overall film mass transfer coefficient
424
L trophic length
[M] activity of microbiota
MY(n) growth rate coefficient
n constant
N Avogadro number
N; biomass of age class i
Ns transportation rate
Ps vapor pressure of S
PSi partial pressure at interface
Q uptake from water
R ingestion rate
RESP respiration rate
S concentration of chemical or substrate S
Sj secretion rate
S. weight of component sorbed per unit of sorbent
Sw concentration in aqueous solution
Swo solubility in water
TC toxic compound concentration
uptake rate in age class i
flow velocity
rate of uptake from water
microbial biodegradation rate
critical volume
excretion rate
rate of hydrolysis
rate of oxidation
rate of photolysis
rate of volatilization
weight
x distance
X biomass per unit volume
Y cell yield
YF efficiency for uptake of food
YT utility coefficient
Zi amount of heavy metal (zinc) in age class i taken up during time
interval fl.1
thickness of gas film
thickness of liquid film
molecular extinction coefficient
biodegradation rate (growth rate)
quantum yield
mass of toxicant per liter and per unit length of organism.
11 Sensitivity Analysis, Calibration,
and Validation
M. B. Beck
11.1. INTRODUCTION
In Chapter 2 a procedure for modeling was introduced. The chapter showed that
the modeling procedure divided essentially into two parts: first, the development
of a model from existing general theory and basic principles; and then the
evaluation ofthe model against observations ofthe behavior of the field system.
Throughout the intervening chapters of the book much has been said about
model development, but there has been little discussion of model evaluation.
The objectives ofthis chapter are to expand upon some of the topics introduced
in Chapter 2, for example, sensitivity analysis, calibration, and validation, and
thus to complete the discussion of model development with a discussion of
model evaluation. This chapter, therefore, is strongly methodological in content.
However, it is difficult to discuss model calibration, for instance, in the absence
of experimental data, and the value of any method is best judged by its applica-
tion in practice. Accordingly, we shall use illustrative case studies to support
the principal themes of the chapter.
Section 11.2 deals with (a priori) sensitivity analysis; it resumes the discussion
of section 2.4.1 of Chapter 2. The mathematical treatment of section 11.2 is
introduced using the Streeter-Phelps model of BOD-DO interaction and then
followed by a case study of a more complex model for stream quality of the
Berkel River in the Netherlands (van Straten and de Boer, 1979). Sensitivity
analysis can, of course, be carried out without the constraints of having to use
field observations. In that sense, as indicated in Chapter 2, it may technically
belong to the a priori phase of the modeling procedure. Section 11.3 examines
the problem of model calibration, which covers the individual subproblems of
model structure identification, parameter estimation, and verification (see
section 2.6). This section provides a review of previous applications of parameter
425
426
estimation algorithms in water quality modeling. It also discusses briefly the
development of one particular estimation algorithm, the extended Kalman
filter (EKF) (see, for example, Jazwinski, 1970), and gives two examples of the
use ofthis algorithm: for the River Cam in England (Beck and Young, 1976) and
for the Jordan River, Utah, USA (Bowles and Grenney, 1978a,b). Section 11.4
deals with model validation, using a case study of Lyngby Lake in Denmark to
illustrate the long-term prediction of the response of a lake to substantially
changed nutrient loading conditions (Jorgensen et al., 1978).
11.2. SENSITIVITY ANALYSIS
Sensitivity analysis addresses the problem of examining the relative magnitudes

of changes in the model predictions with respect to changes in the values of the
model parameters ~. The main distinction between a priori sensitivity analysis
(section 2.4.1) and a posteriori sensitivity analysis (section 2.7.1) is that the latter
requires knowledge of the variance-covariance structure of the calibrated
parameter estimation errors. One can easily apply a priori sensitivity analysis
(as will be discussed in this chapter) to a calibrated model; the Berkel River
study is one example of such an application. Hence the distinction of" a priori"
is dropped from the title of this section, although it should be noted that
(a posteriori) sensitivity analysis in the sense of section 2.7.1 will not be discussed.
Equation 2.1 of section 2.4.1 defined a sensitivity coefficient sij for the change
.1Ci in the ith state variable of the model resulting from a change .1/3j in the value
of the jth parameter:
.1Ci/C;
(11.1)
sij = .1/3)P j '
where we have normalized the relationship by including a nominal reference
value Pj for the parameter, which would give a nominal reference value ci for
the state variable. In general, .1c; and .1/3j are understood as small changes in the
neighborhoods of c; and Pj' As stated earlier, a definition of the type given by
(11.1) enables the analyst to investigate whether a certain percentage change in a
parameter has no real significance (sij ;:::: 0), whether /3j is a dominant parameter,
or whether a small change in /3j induces instability in the model structure.
We can develop this intuitive notion somewhat further in order to discuss
parameter influence coefficients or parameter sensitivity functions (Tomovic,
1964; Eykhoff, 1974). Let us take the example of stream BOD distribution from
the linear classical Streeter-Phelps model (see also Rinaldi and Soncini-Sessa,
1978) :
dc(r)
co=c(r) for r=O, (11.2)
dr
427
in which c(r) is the BOD of the river at a point corresponding to a travel time r,
and K 1 is the first-order kinetic rate constant for BOD decay (stream flow is
assumed to be constant with time and distance). Our objective is to compute the
difference between the solution c(r) of( 11.2), given the initial condition (co + L\co)
and the parameter value (K 1 = K 1 + L\K I)' and the nominal solution c(r)
of (11.2), given Co and K I. Therefore, the method that we shall develop allows
for analysis of the sensitivity of the model performance to changes in the initial
conditions; each case is treated independently here.
First, we shall be more specific and define c(r) to be a function not only of
time of travel but also of the parameter K I' i.e. c(r; K I). The solution of (11.2)
can now be written as a function of the reference solution c(r; K I) and of small
perturbations in the vicinity of the reference solution by taking a first-order
Taylor series expansion:
_
c(r; K d = c(r; K d
- + [OC(~r; K
oK l
I)] KI
L\K 1 + .... (11.3)
Here, we shall define the sensitivity coefficient as
. K ) ~ [oc(r; K l ) ]
s( r, 1 oK _. (11.4)
1 K,
The notation [. ]KI represents evaluation of the partial derivative at the point
K 1 = K I; thus this derivative is clearly the .. instantaneous" gradient of the
solution c with respect to K I' to which gradient L\clL\K 1 would be an approxi-
mation-compare, therefore, (11.1) and (11.4). The significance of (11.3) is
that it shows how the family of solutions c(r; K I) in the neighborhood of the
reference solution may be quickly computed once the sensitivity coefficient of
(11.4) is known. Partly for this reason we wish to obtain an expression for
generating s(r; K I)'
If we differentiate the model relationship of (11.2) with respect to K 1 and
evaluate the result for K 1 = K I' then
o2 c(r; KI)] _ - [oc(r; K 1)] _ [ c(r,. K 1 )]_K • (11.5)

[ oKlor KI - -K 1 oK I Kl
I
After interchanging the order of differentiation in the second partial derivative

on the left-hand side of(11.5) and applying the definition of(11.4), we have
ds(r;K 1 ) _ - _ -
-d~ = -K1s(r; K 1) - c(r; K 1); (11.6)
where the initial condition for this equation has been derived from (11.4) and
the analytical solution of(11.2). We have, therefore, a differential equation whose
solution provides the variations of the sensitivity coefficient with respect to the
independent variable for a chosen nominal reference value of the parameter,
i.e. K I. We shall discuss later why this might be important.
428
To complete the picture, however, it is also possible, given a small change
Llco in the initial condition of (11.2), to obtain likewise a differential equation for
an associated sensitivity coefficient, namely
ds(,; co) - _
d, = - K 1 s(,; co); s(O; co) = 1, (11. 7)
in which, once again, the initial condition is derived through differentiation of the
analytical solution to (11.2).
Inspection of (11.7) shows that the sensitivity of the model response (i.e. the
BOD along the river) to changes in the upstream BOD decays exponentially
as " the time of travel, increases. In other words, as expected, for points very
distant from the reference point (of discharge) the equilibrium BOD is essentially
independent of the in-stream BOD at the reference point. The parameter
sensitivity function of (11.6) is rather more interesting. Figure 11.1 shows the
evolution of this equation for a nominal BOD decay rate coefficient ([(1) of
0.3 day-1 and a BOD at the reference point ('D) of 10.0 g m- 3. The sensitivity
coefficient is always negative-an increase in BOD decay rate coefficient
decreases the remaining BOD-and shows a peak negative value. The minimum
is the result of a balance between two opposing effects: (i) a change in K 1 has a
greater effect on the remaining BOD, the more the BOD has decayed, i.e. at
points further from the reference point; (ii) a change in K 1has a relatively greater
effect, the higher the absolute value of BOD, i.e. at points close to the reference
point. We may say that the model response is most sensitive to the chosen value
of K 1at, = 3 (days); alternatively, when, = 1 (day) a change of +0.05 (day-1)
in K 1, i.e. K 1 = 0.35 (day- 1), would lead to an approximate change of -0.37
(g m - 3) in the remaining BOD at that location (from eqn. 11.3). However,
0.0
I~
Time of travel T (days)

FIGURE 11.1 Coefficient SeT; [(I) expresses the sensitivity of the stream BOD to
changes in the BO D decay rate constan t K I (see also eq n. 11.6). The uni ts of s( T; K 1) are
gm-Jday-I.
429
such results are only approximate, and are only valid in the locality of the
nominal reference solution because of the linearization inherent in the analysis.
What can be concluded from the preceding analysis of parameter sensitivity
functions? Probably the most important factor would be the concise representa-
tion of how the model solution is influenced by its parameter values at each point
in space (or time), i.e. how model solutions differ from the nominal solution at
different spatial locations. In fact, if our intuitive definition of the coefficient
of sensitivity in eqn. 11.1 is recalled, it is seen to be ambiguous because it does
not state how ~Ci is defined: for example, is it a difference in peak response or a
difference in equilibrium response, and so on? Thus relationships of types
(11.6) and (11.7) will demonstrate where the model solution is most sensitive
to the parameter value, whether at a nonequilibrium (or transient, or near-
field) position for 1: -+ 0, or whether at an equilibrium (steady state, or far-field)
position, 1: -+ CD. It would, of course, be possible to generate equivalent results
by recomputing and comparing the model solutions of (11.2) for each change of
the parameter value or initial condition. In this context we could also approxi-
mate these same model solutions by solving (11.6) or (11.7) simultaneously with
(11.3), though it is not obvious whether there would generally be any compu-
tational advantage offered by such an alternative.
The method is not confined to representations of linear systems. In general,
for the nonlinear function
(11.8)
with n state variables and q parameters, the equation for the sensitivity Si/1:)
of state Ci to parameter f3j is
+ [o/;{. }] for i = 1. 2.... , n; j = 1,2.... , q. (11.9)

of3 j c
The notation [']c indicates that all partial derivatives are computed by sub-
stitution of those values for the state vector c and parameter vector p that are
defined for the nominal reference solution of the model. Inspection of the
subscripts i andj indicates that in principle there are q such equations as (11.9)-
where pis a vector of q parameters-to be solved simultaneously for each of the
n state variable equations. The initial conditions of (11.9) are given by silO) = 0,
provided that the initial conditions of the state vector Co are not considered as
model parameters.
We shall further generalize the relationships of the type given by (11.9). Let
us suppose that an n-state-variable (lumped-parameter) model is described by
430
the following vector differential equation:
dc(r)
~ =f{c, r; ~}, (11.10)
in which!{c, r; ~} is an n-element, vector-valued function. If we define an n-

element sensitivity vector s(r; f3) for the jth model parameter [3j' then the re-
sulting n parameter sensitivity functions may be written concisely as:
ds(r; [3j) = [D!{C, r; p}] s(r,.[3.) + [Of{C,"[3r; ~}] _.

J (11.11)
dr c c Uj ~
The (Jacobian) matrix

onc, r; P}]
[ oC c
is an n . n matrix with elements i, 1:
OJ;{C, r; p}] for i, I = 1, 2, ... , n.
[ oc,
The vector
O!{ c, r; ~}J
[ o[3j jj
is an n-element column vector. Hence, for models more complex than the simple
example of eqn. 11.2, eqn. 11.11 gives the simultaneous solution for the sensitivity
of each state variable equation to changes in the value of [3j'
Eykhoff (1974), quoting Meissinger (1960), lists several further possible
applications of parameter sensitivity functions. One of these applications,
specifically based on the interpretation of the sensitivity coefficient as a gradient,
and implicitly as the gradiellt of a model error function with respect to a param-
eter value, will become important later when we discuss algorithms for parameter
estimation in section 11.3. In section 2.4.1 the notion of parameter idelltifiability
was introduced in relation to the insensitivity of the model responses to the
assumed parameter values. When one is working with large, complex models,
in which it is not necessarily self-evident how each parameter affects each state
variable response, sensitivity analysis may yield important insights into the
properties of the model and into the likelihood of successful model calibration.
11.2.1. A Case Study: Berkel River, Netherlands

Van Straten and de Boer (1979) carried out a sensitivity analysis of a water
quality model for the Berkel River in the Netherlands. This section reports some
of their results.
431
Although only a 10 km reach of the river was chosen for the study, the reach
has three weirs and, at the time of the experiment, the time of travel along the
reach was 108 h. A discharge of mechanically treated sewage is located approxi-
mately 2.5 km downstream of the upstream boundary of the reach. As is typical
of such lowland rivers, the growth of floating algae during low-flow, summer
conditions can be considerable. For this reason a model was developed to
characterize the relationships between five state variables: dissolved oxygen
(DO), carbonaceous BOD (C-BOD), nitrogenous BOD (N-BOD), algae
(expressed as chlorophyll a), and soluble reactive phosphorus (SRP). The model
assumes that dispersive properties of the river can be neglected so that the partial
differential equations may be solved along the characteristic trajectories of
stream flow (e.g. Di Toro, 1969; see also Chapter 6). De Boer (1979) calls this a
"moving cell model," since it idealizes the behavior of the system as a sequence
of batch reactors moving with the flow. Thus the model has five equations
written in the form of (11.1 0). Estimates pof the parameter values for the model
were obtained largely by trial-and-error fitting of the model to field observations.
The model combines expressions for the various source and sink terms that are
frequently found in stream dissolved oxygen models (e.g. O'Connor and Di Toro,
1970) and in lake phytoplankton models (e.g. Di Toro et al., 1971).
A sample of the solutions to the sensitivity functions of eqn. 11.11 is given in
Figures 11.2 and 11.3. Here the sensitivity of the state variable responses to
-;:;
.I:
OJ
'Qi 0.8
~
~
"'C
M
I 0.4
E
OJ
Q)
'"
OJ 0.0
-.g
(ij
0
t:
-0.4
B
...
~
t:
Q)
u
t: -0.8
0 0 48 72
u
Time of travel (h)
FIGURE 11.2 Solutions of the sensitivity functions of eqn. 11.11 for the Berkel River
model (adapted from van Straten and de Boer, 1979). The curves represent (for a given
time of travel, r) the change in algal concentration for: A, a 10 % increase in the value of
the algal growth rate coefficient; B, a 10 0<, increase in the value of the light extinction
coefficient.
432
0.8
(")
0.4
I
E
Cl 0.0 I""----~
o
oco
-0.4
-0.8
o 48 72
Time of travel (h)
FIGURE 11.3 Solutions of the sensitivity functions of eqn. 11.11 for the Berkel River
model (adapted from van Straten and de Boer, 1979). The curves represent (for a given
time of travel, r) the change in carbonaceous BOD for: A, a 10 % increase in the value of the
algal death rate coefficient; B, a 10/0 increase in the value of the carbonaceous BOD decay
rate coefficient.
changes in the estimated parameter values is expressed as an absolute change in

the state variable for a relative increase of 10 % in the parameter value. The two
trajectories of Figure 11.2 represent the sensitivity of the algal concentration to
changes in: A, the algal growth rate coefficient; and B, the light extinction co-
efficient. A diurnal pattern is clearly evident and the growth rate coefficient
is seen to have a more dominant effect than the extinction coefficient. Toward
the end of the reach the 10 %increase in the growth rate coefficient would alter
the simulated algal concentration by 0.7 g m - 3 (dry weight), an increase over the
nominal concentration of about 35 %. Figure 11.3 shows two trajectories for the
sensitivity of the carbonaceous BOD to changes in the values of: A, the algal
death rate coefficient; and B, the carbonaceous BOD decay rate coefficient.
The sensitivity function for the former parameter is especially interesting. A
higher algal death rate coefficient would initially lead to a higher BOD, as a
result of the increased production of BOD from the greater amount of dead
algal material. However, as T increases less algae remain in the system and thus
at downstream locations a higher algal death rate means that less BOD is
produced from dead algal matter. The effect of increasing the algal death rate
coefficient is, therefore, to decrease BOD levels at the downstream end of the
reach. Trajectory B in Figure 11.3 shows that BOD is particularly sensitive to
the value of the BOD decay rate coefficient between the point of wastewater
discharge and the first weir. In fact it was assumed that a higher rate of BOD
decay would occur in this reach because of the easily degradable substances
433
present in an effluent that has not received any biological treatment. Curve B,
therefore, reflects a 10 %increase in a relatively higher value for the BOD decay
rate coefficient over that stretch of river.
All the trajectories of Figures 11.2 and 11.3 exhibit a trend toward higher
sensitivity of the model responses as the distance downstream increases. One
may conclude, therefore, that: (a) if the estimated parameter values are not
correct, then predictions from the model become increasingly unreliable at
higher travel times; and (b) a greater experimental sampling effort should be
allocated to the lower reaches of the system, for it is here that the model suggests
the behavior of the system will be most sensitive with regard to subsequent
calibration of the parameter values.
11.3. CALI BRAnON AND VERIFICAnON
The subject of model calibration is subdivided into the two problems of

(a posteriori) model structure identification and parameter estimation. In
Chapter 2 we pointed out that the application of a parameter estimation
algorithm, particularly a recursive estimation algorithm, is frequently a part of
the process of solving the model structure identification problem (see also
Beck, 1979a). In this section, therefore, methods of model structure identification
will not be discussed as a separate issue in itself. Several methods of model
structure identification can be found in the literature (e.g. Box and Jenkins,
1970; Akaike, 1974; Van den Boom and Van den Enden, 1974; Unbehauen and
G6hring, 1974; Wellstead, 1976, 1978), although almost invariably they address
the problem in terms of black box model representations. It was also suggested
in Chapter 2 that model structure identification, at least for the internally
descriptive models that are the principal subject matter of this book, is the
crucial problem in calibration and technically very difficult to solve. It will be
apparent that one of the case studies used to illustrate the application of a
parameter estimation algorithm is, strictly speaking, a study of the problem of
model structure identification.
In general, it is beyond the scope of this book to discuss estimation algorithms
in detail. The texts by Eykhoff (1974) and Gelb (1974) provide suitably in-
structive introductions to the subject; the book by Mehra and Lainiotis (1976)
gives a somewhat more specialized treatment of later developments. The
review section of this chapter, section 11.3.4, will in addition give references to
several other illustrative applications of estimation algorithms. The discussion
of parameter estimation begins, however, with the recall of some basic intro-
ductory concepts from Chapter 2. This is followed by a brief summary of the
principal components of off-line estimation algorithms and then a more detailed
treatment of recursive parameter estimation algorithms.
434
Unmeasured Measurement
disturbances ~ (r) errors" ( r )
Process +
+
Measured Measured
inputs d (r) States c(r) outputs cOl r)
Parameters PI r)
FIGURE 11.4 Definition of the system and its associated variables.
11.3.1. Parameter Estimation

Figure 11.4 gives the general definition of the (water quality) system and its
associated variables, as discussed in Chapter 2. Already, however, we have
restricted the discussion to lumped-parameter models (either dynamic or steady
state) by specifying only one independent variable r (meaning either time or time
of travel). This is partly for clarity and simplicity, although to a great extent this is
indicative of the limitations of currently published studies of parameter estima-
tion in water quality models. Our treatment of parameter estimation focuses,
therefore, on the problem as specified by the following model form.
State vector variations:

e(r) = j{e(r), per), d(r)} + ~(r). (l1.12a)
Output response observations:
eO(r) = m{e(r), pcr)} + 1)(r). (11.12b)
The dot notation in (11.12a) denotes differentiation with respect to r.
d(r) is a vector of measured input forcing variables,

e(r) is a vector of state variables (the symbol e is used because, in general, the
state variables are here represented by the concentrations of dissolved
and suspended materials in the water body),
eO(r) is a vector of measured output response variables,
pcr) is a vector of model parameters,
~(r) is a vector of unmeasured system disturbance variables,
1)(r) is a vector of measurement errors,
j, m are nonlinear vector-valued functions.
The basic requirement of parameter estimation is that estimated model output

responses cO(r), computed when the input measurements d(r) are given, i.e.
CO(r) = /,{c(r), ~(r), d(r)}, (11.13)
435
can be compared with observed values for the outputs cO('r). Estimates of the
outputs are required because, in general, the measured outputs are not straight-
forward error-corrupted observations of the state variables. Minimization of
some function of the errors between observed and estimated outputs, i.e.
(11.14)
leads to a set of "best" estimates pfor the parameter values according to the
chosen error function. In (11.14) the functional arguments pand c have been
selectively retained in order to illustrate two points. First, since the objective is
parameter estimation the errors £ are indicated as being dependent only upon
c
the parameters. Second, however, the errors depend also upon since the esti-
mated outputs CO are functions of the state variable estimates C, which may not
necessarily be uniquely determined by the choice of parameter estimates and
input measurements alone. This implies, then, solution of the related problem of
state estimation (as we shall see in section 11.3.2). Essentially, the input-output
field data d and CO are the fixed basis for model calibration and parameter
estimation; they are, as it were, the fixed external description of the system
behavior. The objective of calibration is to make plausible hypotheses about the
internal description (c and P) of the behavior (Figure 11.4) according to a chosen
criterion and to certain assumptions about the random processes ~ and 1).
To reiterate the discussion of section 2.6.4, let us start with the method of
least-squares parameter estimation, which minimizes the error (loss) function,
(11.15)
The fundamental role of least-squares estimation as a solution to the problem
of evaluating parameters by reference to field data is undisputed; it is nearly
always quoted as the basis for further development of more intricate methods
(for example, Draper and Smith, 1966; Astrom and Eykhoff, 1971; Young, 1974;
Soderstrom et al., 1978). Least-squares estimation does, however, suffer from an
important disadvantage, namely the problem of bias in the estimates, i.e.
(11.16)
where E{· } is the expectation operator. In other words, since we are estimating
the parameters on the basis offield data that are a sample realization of randomly
distributed variables, the parameter estimates can also be characterized as having
probability distributions. Furthermore, if Gaussian distributions are assumed,
the properties of the parameter estimates can be specified by the values of their
means and covariances. Thus (11.16) states that the mean or most probable
values of the parameter estimates are not equal to the true values of the param-
eters. Least-squares estimators, however, are not always biased, although they
are in many cases of practical interest. Such a discrepancy between the true
parameter values and the least-squares estimates will occur when, as frequently,
the statistical properties of the random sequences ~ and 1) do not conform to
436
those of zero-mean sequences that are not correlated with each other and that
are not correlated with themselves in time, i.e. ~ and 'I are not white noise
sequences.
There are many other estimation algorithms, few of which, however, depart
essentially from the principle of least squares. We shall mention only two other
approaches. The first is the maximum likelihood method (e.g. Astrom and
Bohlin, 1966; Box and Jenkins, 1970; Kallstrom et al., 1976; van Straten, 1983).
In this case, if the loss function of (11.15) is modified to give a weighted form of
the squared loss function,
(11.17)
where W is a matrix of weighting coefficients, then maximum likelihood
estimation usually corresponds to minimizing the loss function (11.17) with a
particular choice of W. This choice is that W is set equal to the inverse of the
variance-covariance matrix of the associated output response observation
errors. That is to say, maximum likelihood estimation assumes some statistical
knowledge about the system under investigation and weights the squared errors
in accordance with the accuracy of the system response measurements. If the
measurement errors are assumed to be independent of each other, in which case
W becomes a diagonal matrix with nonzero elements on its leading diagonal
only, the minimization of (11.17) is sometimes referred to as weighted least-
squares estimation.
The second method, known as the method of instrumental variables (e.g.
Kendall and Stuart, 1961; Johnston, 1963; Young, 1976), seeks to generate a set
of variables, the instrumental variables, which, it is hoped, conform to certain
statistical requirements that ensure the prevention of bias in the parameter
estimates. Unlike the method of maximum likelihood, an instrumental variable
estimator does not necessarily require the analyst to qualify formally the
structure of the correlated properties of the sequences ~ and 'I.
These are thus some of the more important approaches to the problem of
parameter estimation. It is now appropriate to recall from Chapter 2 the
distinction between off-line and recursive methods of parameter estimation.
11.3.2. Off-Line Estimation Algorithms

The basis of an off-line method is that, during each iteration, the model output
responses are compared with the field observations, while the parameter
estimates are held constant, and at the end of each iteration a value for the error
function, for example (11.15), is computed. The algorithm assesses the shape of
the error function surface for the particular values of the parameters and error
function at each iteration, and then attempts to descend toward the minimum
of the error function by specifying a corrected set of parameter values for the
next iteration. The form of this iterative search for the" best" set of parameter
437
values may be defined, for example, by the following class of gradient algorithms:
(11.18)
Here superscript i denotes the revised parameter estimates at the end of the ith
iteration; A is a positive scalar step length, VI\(J) denotes a vector of gradients of
the loss function J with respect to the parameters p, and 'I' is a matrix of co-
efficients to be determined. It is not always necessary to solve the problem
numerically. For instance, some models may have a form that, when substituted
into the error function (11.15), allows differentiation of J with respect to the
parameters to yield explicit analytical solutions for the best estimates. It may
also be observed from (11.18) that when VI\(J) = 0, W= w-
1; in other words, the
Iterative numerical algorithm has converged on the set of parameter estimates
that minimizes the error function.
Without going into the extensive field of methods for finding the extremum
of a function, i.e. optimization, we may note from (11.18) that in implementing
an off-line estimation algorithm the following three factors require consideration.
(i) There is the determination of the gradients VI\(J) and, most probably,
the computation of numerical approximations thereof. If we look at the
form of J in (11.15), and thence recall (11.14), we can see how the
gradients required for (11.18) will be dependent upon the sensitivity
of the model responses C"{P} to the parameter values. More specifically,
J is a function of the errors £, which are a function o( the estimated
output responses eo, which in turn are a function ofthe state estimates e.
P,
Differentiation of J with respect to therefore, implies differentiation
of c, the state vector, with respect to p, i.e.
This Jacobian matrix is clearly the matrix of sensitivity coefficients,

as defined in (11.4), but evaluated for the current set of state and
parameter estimates c, p. It can now be seen, through (11.18), how
sensitivity coefficients are directly linked with the performance of
parameter estimation algorithms.
(ii) There is also the choice of the matrix '1': to name but two basic al-
ternatives, we have the method of steepest descent in which, simply,
'I' = I,
that is,
for i = j
for i =I j,
438
where I is the unit identity matrix; or the N ewton-Raphson method,
where
aJ
2
l/Jij = aPi ap j ,
i.e. elements that are second partial derivatives of the error function
with respect to the parameters.
(iii) Lastly, after determination of the direction in which to reduce the error
function surface toward its minimum, there is the problem of specifying
the magnitude A. of the step to be taken in that direction.
From these basic ingredients arises a variety of methods that attempt to

combine and exploit the better features of each individual algorithm in terms of
its stability and convergence properties (for example, Wilde, 1964; Wilde and
Beightler, 1967; Shastry et al., 1973).
The off-line schemes of parameter estimation are, therefore, seen to be
essentially developed from the explicit solution of an optimization problem.
The routine that searches for the optimum, however, need not be coupled to an
intelligent appraisal of the shape of the loss function surface, as it was in the
example algorithm of (11.18). A much easier routine to implement would be
one that searches in a random fashion for the minimum of the errors between
the model and the field data (e.g. Halfon, 1979). For such a random search
routine the parameter values must usually be specified a priori to lie between
certain bounds:
~min ~ ~ ~ ~max' (11.19)
For a large number of iterations i through the field data, the combination of
parameter values that yields the lowest value of the loss function is chosen for
the best parameter estimates.
11.3.3. Recursive Estimation Algorithms: An Example

We have chosen to discuss recursive estimation algorithms in greater detail than
off-line algorithms and to discuss, in particular, the extended Kalman filter
(EKF) algorithm (for example, Jazwinski, 1970). The application of this
algorithm will then be illustrated by two case studies in section 11.3.5. The
presentation is somewhat limited in its breadth, therefore, although there will be
advantages in terms ofdemonstrating the estimation of parameters that may vary
in time and space.
The Extended Kalman Filter

Unfortunately there is no concise derivation of the EKF algorithm since it is
implicit that one is first familiar with the form of the linear Kalman filter (LKF)
439
•T, .----------------~
T2 TN
FIGURE 11.5 Conceptual picture of a recursive estimation scheme. The notation Tk

denotes the kth discrete sampling instant in a time series with N samples.
(Kalman, 1960; Kalman and Bucy, 1961; Gelb, 1974). This in turn implies a
reasonable appreciation of the notion of state estimation, which is especially
important for the present purposes because the key link between the LKF and
EKF will be the interpretation ofthe parameter estimation problem as a problem
of state estimation. We may recall, therefore, the ambivalent attitude toward use
of the terms "state" and "parameter," which was suggested in section 2.3.
It will be helpful in the discussion here to visualize a parameter as merely a
special case of a constant state variable.
In Chapter 2 the basic idea of a recursive parameter estimator was introduced.
Figure 11.5 repeats the diagram shown before as Figure 2.8(b). Attention will be
confined to a revised form of the output observations of (11.12b), that is,
(11.20)
in which 'k is the kth diserete point in time (or time of travel). This merely
reflects the more realistic situation in which discretely sampled observations of
water quality are available. A close study of Figure 11.5 allows one almost
intuitively to write down the structure of a recursive estimator: thus the newly
revised estimate of the parameter vector at the current time, say P('k)' would be
a function of its immediate past estimate P('k _1) and a correction term that is
based on the error between (he observation and model prediction at time 'k, i.e.
P('k) = P('k-l) + G('k}€('k)' (11.21)
new old weighting prediction
estimate = estimate + factor error
The error vector £( 'k) is the error between a model prediction (estimate) of the
system response at time 'k and the noise-corrupted measurements of that
output response. The gain matrix of weighting factors, G('k)' is essentially a
function of the available field data. It may be noted in passing that (11.21) is
structurally similar to the off-line algorithm of (11.18); Young (1974) has
indicated how some forms of recursive algorithm are themselves also gradient
algorithms. It is not difficult to see that a recursive state estimator can be
constructed along lines exactly analogous to (11.21), that is,
c(,:) = c(,;;) + G(,;;}€(,;;). (11.22)
440
In this case G('k-) is a function partly of the model parameter values; c(,t)
denotes the newly updated state vector estimate after receipt of all the measured
information at time 'b whereas c(,;;) represents a best forward extrapolation
(in time) of the state c prior to receipt of the current measurements. E(,;;) is
likewise based upon this same extrapolation c(,;;). The terms" prior," "current,"
and" after" are intended here to be indicative of the sequential procedure of the
algorithm.
An important difference between (11.21) and (11.22) lies in the arguments of
c and p. As one would expect, the state of water quality will change over the
interval from ' k - l to 'k; it is therefore prudent to use a model to make some form
of extrapolated prediction over this interval for comparison with the next
measurements obtained at 'k' In contrast, the assumed model of parameter
variations would be that, in fact, they remain constant. Thus the best prediction
of the parameter values at a later instant is that they have the same values as
estimated at present. With this in mind, it is appropriate now to develop a
conceptual picture of the EKF algorithms; this is shown in Figure 11.6. A
model of "reality" is embedded in the filter. Predictions, of the kind c(,;;) in
(11.22), are computed from the model by using the measured input disturbances
Unmeasured Measurement
disturbances ~ errors"
Measured +
states c m + Measured outputs CO
"Reality"
' - - - - - - - ' Unmeasured

states C u
,- --- ----,
Respective levels of
I Extended Kalman filter
uncertainty (error) in the I
I model, the disturbances, I
I and the measurements
I
I
CO
I
I Predictions
Model
I d
I
I I
I I
Corrections
I I
I I
L _ __ J
FIGURE 11.6 Conceptual picture of the extended Kalman filter. cm ' cu ' and pare,
respectively, estimates of the measured state variables, the unmeasured state variables,
and the parameters_
441
of the system, d(r,.), and fed forward to the corrector algorithm together with
the current observations CO( r k) of the system output response. For the corrector
algorithm, i.e. (11.22), it is apparent that additional (and parallel) algorithms
are required that compute the evolution of the uncertainty, or error bounds,
associated with the state estimates. The results from the corrector algorithm
computation are .the updated estimates; these in turn are fed back to the model
for revision of subsequent predictions.
Two qualitative interpretations of the filter may be helpful at this point.
First, the filter can be seen as an algorithm for" translating" information about
the observed input-output behavior of the real system into model-related
estimates of the state variables and parameters. Second, the name "filter" also
suggests the intuitive idea that the algorithm has the objective of filtering from
the given field data the unwanted influences of measurement noise and un-
certain disturbances.
The following development of the EKF depends on two key items:
(i) the formulation and outline solution of the combined state-parameter

estimation problem; and
(ii) the method of computation ofthe gain matrix G(r;), and the dependence
of this matrix on specified measures of the levels of uncertainty in the
model as a representation of reality, in the input system disturbances,
and in the output response observations.
A formal derivation of the EKF is given in the source reference, Jazwinski (1970).
Alternatively, Young (1974) provides an outline of how the EKF algorithms
can be obtained from an extension of linear regression analysis; this same
outline is treated in depth by Beck (1979b).
Combined state-parameter estimation. Let us assume that the type of model to

be used is the linear form of (11.12a) and (11.20):
e(r) = Ac(r) + Bd(r) + ~(r), (11.23a)
with discretely sampled, noise-corrupted output response measurements co:
(l1.23b)
in which the dot notation refers to differentiation with respect to time r. The
matrices A, B, and H have the dimensions n· n, n· m, and p. n, respectively; n, m,
and p correspond to the dimensions of the state (c), input (d), and output (CO)
vectors. Equation 11.23b expresses the fact that in general the outputs are not
necessarily simple error-corrupted observations of the state variables (the
outputs may be linear combinations ofthe state variables). ~ is an n-dimensional
vector of unmeasured, random process disturbances, 11 is a p-dimensional vector
of chance measurement errors, and both ~ and 'I will be assumed to approximate
442
zero-mean, white, Gaussian sequences (formal definitions are given below).
For the linear system of (11.23) the linear Kalman filter would thus provide
recursive estimates c(rklrk) of the state vector c(r k), where the argument of c
indicates that these are estimates conditioned upon all available measurements
up to and including those at the current time r k • The restriction of specifying
(11.23) as linear is primarily a convenience for illustrative purposes; the actual
problem at hand will soon become nonlinear in any case.
Let us suppose now that some of the unknown, or imprecisely known,
elements of the matrices in (11.23), that is a vector of parameters ~, say, are
required to be estimated simultaneously with the state vector. One approach
to realizing a simultaneous state-parameter estimator is to augment the state
vector c with the parameter vector and accordingly to postulate a set ofadditional
differential equations representing the parameter "dynamics." (In this sense the
method of quasilinearization is similar to the EKF since it, too, sets up the
parameter estimation problem by interpreting the parameters as additional
state variables (Bellman and Kalaba, 1965; Lee, 1968).) If the augmented state
vector c* is defined by
c* ~ [-;-]
the state-parameter dynamics and observation equation are given In the
following general nonlinear form:
c*(r) = /*{c*(r), d(r)} + ~*(r), (11.24a)
cO(r k) = m*{c*(r k )} + tt(rk)· (11.24b)
The functions /*{.} and m*{·} are vector-valued; they are nonlinear because
of the product terms involving elements of ~ with elements of c and d. ~*(r)
denotes that the vector of stochastic disturbances in (11.24a) is now of a different
order to that defined for ~(r) in (11.23a).
Let us consider the problem of specifying the dynamics of the parameters p.
Of particular importance to the subsequent discussion are two such specifica-
tions: (a) we might assume that the parameters are constant, that is, time-
invariant:
p(r) = 0; (11.25)
or (b) it might be proposed that they vary in an unknown "random walk"
fashion:
p(r) = ~(r). (11.26)
in which ~ is a vector of zero-mean, white, Gaussian disturbances. Were there
to be more a priori information on the parameter variations, it would be ap-
propriate, for instance, to define the dynamics as oscillatory in accordance with
some diurnal or seasonal fluctuation.
443
The algorithms and computation of the gain matrix. The EKF is a linear
approximation of the nonlinear filter that would ideally be needed to provide
estimates of c* in (11.24). Briefly, there are three main steps in its derivation.
(i) Linearization of the nonlinear augmented state eq uations by means of a
first-order Taylor series expansion about a deterministic nominal
reference trajectory for the state variables. From this a set of linear
equations is obtained for the dynamic variations of small perturbations
about the reference trajectory.
(ii) Application of a linear Kalman filter for estimation of the small-
perturbations vector.
(iii) Substitution of the current augmented state-parameter vector
estimates as the choice of nominal reference trajectory. If this specific
choice is not made, the state estimates can be obtained by combining
the filter estimates of the small perturbations with the known de-
terministic reference trajectory. However, as a consequence of such a
choice it is possible to formulate the EKF directly in terms of the
augmented state-parameter vector c*, rather than in terms of the small
perturbations; and by this choice it is also more probable that the
perturbations about the reference trajectory are in fact sufficiently
small to justify linearization as a valid approximation. Figure 11.7
gives further explanation of this outline; a more complete interpre-
tation of the derivation is given in Beck (l979b).
Small perturbation
about nominal
reference trajectory
*u
:c«l
Q) ---~----
------\:: /' .--"
'':
«l
> Small perturbation
....«l
Q)
in the extended Kalman

....
(/)
filtering algorithm
I I I I I I
'1 '2 '3 '4 '5 '6 '7 '8 '9 '10 '11
Time or time of travel ( , )
FIGURE 11.7 An example showing how the relinearization procedure of the extended
Kalman filter is capable of preserving only small perturbations about the reference trajectory
(in the EKF the current state estimates are substituted for the reference trajectory). Con-
tinuous line: the true state; • observed values of t he state: chain line: a nominal reference
trajectory; broken line: the state estimates from the filtering algorithm.
444
We have lastly to introduce some precise notation. Thus, let us define the state-
parameter estimation error variance-covariance matrix as
(11.27)
in which E {. } is the expectation operator and the argument (, I'k) denotes an

estimate, or error variance, of the state variable (or parameter) at time, based
upon all past measured information up to and including that available at 'k'
The EKF algorithms can now be stated summarily in the following form.
State estimates
(i) For prediction between sampling instants,
C*('k !'k- 1) = C*('k-1 I'k - 1) + Cf*

.. tk - 1
{c*(, I'k - 1), d( r)}dr. (11.28a)
(ii) For correction at the sampling instant,
c*('kl'k) = c*('kl'k-1) + G*('k)[CO('k) - m*{c*('kl'k-1)}]' (11.28b)
Error covariances
(i) Prediction,
P('kl'k-1) = cJ)('k' 'k-1)P('k-11'k-1)«J)T('b 'k-1) + Q. (11.28c)

(ii) Correction,
P('kl'k) = [I - G*('k)M('k)]P('kl'k-1)[1 - G*('k)M('k)]T

+ G*('k)RG*T('k)' (11.28d)
Gain matrix computation,
T
T
G*('k) = P('kl'k-1)M ('k)[M('k)P('kl'k-1)M ('k) + Rr 1. (11.28e)
There are some additional definitions for (11.28).
(a) Superscript -1 denotes the inverse of a matrix.
(b) The matrices Q and R are, respectively, the variance-covariance
matrices of the stochastic processes ~* and 11 in (11.24), where these
processes are also assumed to have zero means and be normally
distributed:
and
445
with
for k =I- t
for k = t.
(c) The model-predicted response errors (residual errors) are given by
£(tk) = CO(tk) - m*{c*(tk/'k-I)} = CO(t k) - cO(tkl'k-I)' (11.29)
(d) The gain matrix G*('k) refers to the combined state-parameter vector.
(e) The matrices <1»( 'b 'k -I) and M('k) refer (implicitly) to the (augmented)
state transition matrix (e.g. Dorf, 1965) and the observations matrix
for the linearized small-perturbations system obtained at the first step
of the derivation shown above, i.e.
cD('b 'k-l) ~ exp{F( tk-l)[ 'k - 'k- I]}
with
F(tk-l) ~ [an{c*(,), d(,)}/acj]I~.«)=~.«k-ll<k_tl
d(r)=d«k-l)
and
M('k) ~ [amt{c*('k)}/acj]I~.(tk)=~.(rklrk_tl
in which it and mt are individual elements of the nonlinear vector-
valued functions f* and m* given by (11.24).
Since the EKF algorithms are recursive it is clear that a set of initial values at
time '0 must be specified for the state-parameter estimates c*(,o I'0) and their
associated error variances P('0 I'0)' Also implied by the algorithms is the fact
that values have to be specified for the covariance matrices Q and R, which, as
indicated elsewhere (Beck, 1979a), is an especially difficult task with respect to Q.
The importance of algorithms (11.28), however, is not to be able to recall their
exact form but to notice the following two items.
(i) Even though the EKF is a linear approximation designed to treat the
system of nonlinear equations 11.24, the consequences of the lineariza-
tion procedure appear only as the matrices cD and M in the error co-
variance and gain matrix computation algorithms. For the state
estimation computation. (11.28a) and (11.28b), the original nonlinear
functions are preserved.
(ii) Apart from their use as a measure of the confidence bounds on the
accuracy of the estimates,t the covariance algorithms essentially
t However, in view or the linearization approximation one should be very cautious in making this
interpretation for the parameter estimates, although such caution is not necessary with a recently
proposed modified form or the EKF (Ljung, 1979).
446
serve the purpose of making an intelligent choice of gain matrix. One
can observe that (qualitatively) when the measurement error variance,
i.e. R, is large, the gain matrix tends to be small. In other words, the
filtering algorithm will ignore large errors €( rk) between predicted and
observed responses since it attributes these to measurement error.
Conversely, when the filter is uncertain of the model performance, i.e.
P(rklrk-l) is large, the gain matrix is relatively large and significant
correction of the estimates will follow if large prediction errors are
perceived.
Figure 11.8 shows some additional schematic features of the EKF algorithms;
further details of the operation of the algorithms, especially with respect to the
solution of the model structure identification problem, are given in Beck (1979a).
L
State - parameter
estimates
,-----<
-
,... Model
(eqn.ll.28al
c* (rki Tk_1)
~
Comparison
and correction
leqn.11.28b)
~
6*(T,.,I" ,I 6*(T,/T,)
I Delay
I
I I _. - _.
- "--_._-- _.- - - ---" - - - - -
Estimation error
covariance solution
<J> (T" T, 11 M(T,)
R
Q ()
()
I
~
Filter gain
~
L:: Prediction
computation Correction ~
leqn 11.28c) G*(T,I ,.- (eqn.11.28d)
I
leqn.11.28el
P(T,lr, 11
'I PIT"IT,,} , Delay

I Pir,IT,1
I I
FIGURE 11.8 Block diagram of the extended Kalman filter showing the computation
both of the state-parameter estimates and of the covariance matrix. The functional depen-
dence of the matrices <J) and M on the corrected and predicted state estimates, respectively,
is defined in the text.
447
11.3.4. Some Recent Applications of Parameter Estimation
Table 11.1 summarizes the literature on recent applications of parameter
estimation algorithms in the development and assessment of stream and lake
water quality models. The applications can be broadly classified according to
whether they used off-line (0) or recursive (R) estimation schemes in the manner
described previously. The models are classified according to the choice of:
(i) independent variable(s), i.e. time and/or one-dimensional space;

(ii) quality characteristics, i.e. state variables; and
(iii) model equation form, i.e. differential or difference equations.
A few remarks are necessary to qualify the contents of Table ILL For instance,
the paper by Ivakhnenko et al. (1977) is primarily concerned with the problems
of model discrimination and model structure identification rather than with the
problem of parameter estimation. In fact in the Group Method of Data Handling
(GMDH) the implicit problem of parameter estimation is treated with a least-
squares estimator. Other references (Shastry et al., 1973; Beck and Young,
1976; Jolankai and Sz6ll6si-Nagy, 1978; Halfon et al., 1979) are similarly
oriented toward model structure identification.
Four papers, those of Huck and Farquhar (1974), Beck (1975), Gnauck al. et
(1976), and Halfon et al. (1979), deal with models and estimation procedures that
are representative of input-output (i.e. black box) time-series analysis techniques.
Halfon et al. (1979) also make use offrequency response methods (see section
2.4). Earlier examples of similar approaches are given by Thomann (1967), Fuller
and Tsokos (1971), and Edwards and Thomes (1973). Other contemporary
studies along the same lines include the use of: correlation analysis (Schurr
and Ruchti, 1975); a recursive instrumental variable (IV) approach (Beck, 1978a);
and further applications of the Box and Jenkins (1970) maximum likelihood
methods (e.g. Mehta et al., 1975). Thus, although Table ILl focuses on internally
descriptive model calibration, this does not suggest that there has been any lack
of attention given to input-output model identification.
Lastly, we may observe that only one study (The, 1978) has addressed the
particularly difficult problem of parameter estimation in distributed-parameter
models with the use of field data.
11.3.5. Two Case Studies

The two case studies in this section (Beck and Young, 1976; Bowles and Grenney,
1978a,b) illustrate the application of the extended Kalman filter algorithm
discussed in section 11.3.3. Both examples assume a lumped-parameter form of
model; one of them represents a steady state condition, while the other treats
TABLE 11.1 Summary of Recent Applications of Parameter Estimation Algorithms in Water Quality Modeling. Model
description includes a definition of independent and dependent variables. 0 and R denote off-line and recursive estimation
algorithms, respectively.
Authors Field Data Algorithm Type of Model
Stream "'at'" quality modelinq
Koivo and Phillips (1971) Stochastic approximation Time and space; BOD. DO; analytical
.j:;. (least squares); R solution to first-order partial
.j:;. differential equation.
00
Koivo and Phillips (1972) Least squares; 0 Space; BOD. DO; steady state
analytical solution to first-order partial
differential equation.
Koivo and Phillips (1976) Linear Kalman filter; R Time and space; BOD. DO; difference
equations.
Koivo and Koivo (1978) Least squares; R (state Time and space; BOD. DO; first-order
estimation only) partial differential equation.
Lee and Hwang (1971) Quasilinearization Space; BOD. DO; ordinary
(least squares); 0 differential equation.
Shastry et al. (1973) Sacramento River (1962) Weighted least squares; Space; BOD. DO; ordinary
maximum likelihood; 0 differential equation.
Huck and Farquhar (1974) St. Clair River (1971) Maximum likelihood; 0 Single-point spatial location. time
variations; DO. chloride; black box
time-series model.
Beck (1975) River Cam (1972) Maximum likelihood; 0 Time; BOD. DO; ordinary differential
equation; also black box time-series
model.
Beck and Young (1976) River Cam (1972) Extended Kalman filter; R Time; BOD. DO; ordinary differential
equation.
Whitehead and Young (1975) Bedford-Ouse River (1973) Multivariable instrumental Time; BOD. DO; difference equations.
variable--approximate maximum
likelihood (MIVAML); R
Young and Whitehead (1977) River Cam (1972), MIVAML; R Time; BOD, DO; difference equations.
Bedford-Ouse River (1973)
Lettenmaier and Burges (1976) Extended Kalman filter; R Space; BOD, DO; ordinary
differential equations.
Erni and Ruchti (1977) Aare River Differential approximation Single-point spatial location, time
method; 0 variations; DO; difference equations.
Ivakhnenko el 01. (1977) River Cam (1972) Group Method of Data Handling Single-point spatial location, time
(GMDH);O variations; BOD:DO; difference
equations.
Stehfest (1978) River Rhine (1971) Quasilinearization Space; BOD, DO; ordinary differential
t
\0 (least squares); 0 equations.
Stehfest (1978) River Rhine (1971) Quasilinearization Space; easily degradable organic
(least squares); 0 matter, slowly degradable organic
matter, bacterial mass, protozoan mass,
DO; ordinary differential equations.
Bowles and Grenney (1978a) Jordan River, Utah Extended Kalman filter; R Space; BOD, DO, NH 3 -N, NOrN,
algal N, organic N; ordinary
Moore and Jones (1978) River Cam (1972) Coupled Bayesian-Kalman Time; BOD, DO; ordinary differential
filter; R equations.
Rinaldi el al. (1979) Bormida River Least sq uares; 0 Space; BOD, DO; analytical solution
to first-order ordinary differential
equations.
Tamura (1979) Linear Kalman filter Time and space; BOD, DO;
(and others); R difference equations.
The (1978) River Rhine Linear Kalman filter; R Time and space; conductivity;
second-order partial differential
equation (finite-difference
approximation solution).
Whitehead (1980a) River Cam (1972) Instrumental variable; R Time; BOD, DO; ordinary differential
equations.
Whitehead (1980b) Bedford-Ouse River (1978) Extended Kalman filter; R Time; DO (in four river reaches);
NHrN (in two river reaches); ordinary
(col/til/ued over) differential equations.
Authors Field Data Algorithm Type of Model
Lake wafer qualify modelinq
Di Cola ef al. (1976) Leopold's Park Pond, Brussels Least squares; 0 (solved as an Time; autotrophs, herbivores,
(1973-75) optimal control problem) carnivores; ordinary differential
equations.
oj:>. Gnauck ef al. (1976) Saidenbach Reservoir, Least squares; R Time; DO, chlorophyll a, particulate
VI GDR (I 966··70\: organic matter; regression relationship.
o
Klic;ava Reservoir, CSSR (1963- 72)
lolankai and Sz6116si-Nagy (1978) Lake Balaton, Hungary Maximum likelihood; R Time; soluble reactive phosphorus.
(197 I-77) chlorophyll a, exchangeable
phosphorus in sediment; ordinary
Lewis and Nir (1978) Greifensee, Switzerland (1973) Weighted least squares; 0 Time; soluble reactive phosphorus.
particulate phosphorus; ordinary
Halfon ef al. (1979) Small lake ecosystem Least squares; 0 (also Time; solu ble phosphorus, particulate
frequency-domain analysis) phosphorus, a low-molecular-weight
form of phosphorus, colloidal
phosphorus; ordinary differential
equations.
Benson (1979) Lake Placid, British Columbia, Least squares; 0 Time: phytoplankton biomass;
Canada ordinary differential equations.
Di Toro and van Straten (1979), Lake Ontario (1972) Maximum likelihood; Time; 16 state variables divided
van Straten (1983) weighted least squares; 0 between epilimnion and hypolimnion
layers; ordinary differential equations.
Scavia (1980) Saginaw Bay, Lake Huron Extended Kalman filter; R Time; eight state variables; ordinary
451
a dynamic situation. These examples reflect problems and constraints that are
typical of current studies in model calibration: problems of too few data; data
with too high levels of error and uncertainty; and hence the technical difficulties
that have in general restricted the application of estimation algorithms to small
models with only a few state variables.
Case Study I: River Cam, England

Like the Berkel River discussed in section 11.2.1, the River Cam is a slow,
lowland river that is susceptible to significant growth of algae during summer.
This particular study concerns a field experiment carried out in summer 1972
on a 4.5 km stretch just downstream from Cambridge. The field data and a
complete set of results for the model development exercise are given in Beck
(1978a).
The dynamic behavior of BOD-DO interaction in a reach may be approxi-
mated by a lumped-parameter, ordinary differential equation form of model.
The model assumes that the mixing properties of the reach can be idealized
as those of a continuously stirred tank reactor (CSTR) (see also Thomann,
1963 and Chapter 6, section 6.3). If the assumptions of Dobbins (1964) are
adopted for the definition of BOD-DO interaction, the model takes the form:
This equation for the state vector dynamics corresponds to (11.23a) of section
11.3.3. The output observations equation for the error-corrupted measurements
of downstream BOD and DO, corresponding to (11.23b), is given by
C~('k)] [C 1('k)] ['11 ('k)]

[C'2('k) =
ci'd + '12('k) ,
(11.30b)
where 'krepresents the kth day of the experiment. Other terms in (11.30) are
defined as follows.
C 1(,), ci,) are state variables representing concentrations of BOD and DO,
respectively, at the downstream end of the reach (g m - 3),
d 1(,), die) are measured input variables representing concentrations of BOD
and DO, respectively, at the upstream end of the reach (g m - 3),
Q(,) is the stream discharge in the reach (m 3 day-I),
V is the (constant) volume of water in the reach (m 3 ),
K1 is the BOD decay rate constant (day- I),
452
K2 is the reaeration rate constant (day-l),
C( r) is the saturation concentration of DO (g m - 3),
LA(r) is the rate of addition of BOD to the reach by, for example,
local surface runoff (g m - 3 day - l),
DB(r) is the net rate of addition of DO to the reach by the combined
effects of photosynthesis, respiration, and decomposition of mud
deposits (g m- 3 day-l),
¢l(r), ¢ir) are unknown stochastic disturbances of BOD and DO, re-
spectively (g m - 3 day - l),
IJl(rk), IJir k ) are random measurement errors associated with the downstream
BOD and DO observations, respectively (g m- 3).
In this example, given input-output data for d l , d 2 , c~, and c~, we wish to
estimate the parameters K l' K 2, L A ( r), and DB ( r) in addition to the state variables
cl(r) and cir). The augmented state-parameter vector is thus defined by
(11.31)
and the augmented state-parameter dynamics become
c!(r) - (c!(r) + Q(r)/V)c!(r) + (Q(r)/V)d 1 (r) + c~( r) ¢!(r)
-(c!ef) + Q(r)/V)c!(r)-c!(r)c!(r) + (Q(r)/V)d 2 (r»)
c!(r) ( ¢!(r)
+ c!(r)Cs(r) + c~(r)
c!(r) o + o
c!(r) o o
c~(r) o ¢~(r)
c~(r) o ¢~(r)
(11.32a)
Clearly the observations equation (11.30b) is modified to

cO(r k) = [I : O]c*(rk) + TJ(r k), (11.32b)
in which 0 is the null matrix. Equations 11.32 are now in a form that corresponds
directly with the general representation of the combined state-parameter
estimation problem in (11.24). For estimation purposes K land K 2 are assumed
to be time-invariant, while LA(r) and DB(r) are idealized as random walk
parameters. The specifications of the covariance matrices P( r 0 Iro), Q, and R
(section 11.3.3) for this example are given in Table 11.2.
The results of processing the field data with the extended Kalman filtering
algorithms are shown in Figure 11.9; the field data cover the period from 6 June
to 25 August 1972, inclusive. In Figure 11.9(a) the peak estimate of down-
stream BOD at day r 58 is a consequence of both a high stream discharge and
high upstream BOD caused by a thunderstorm. Inspection of Figure 11.9(e)
453
TABLE 11.2 Specification of Leading Diagonal Elements for the Covariance
Matrices P(,o I '0), Q, and R for the River Cam Case Study. All other elements in
the matrices are assumed to be zero.
State/Parameter A priori Estimation Unmeasured Input Measurement Units
Error Covariance Disturbance Error
P(ToIT o ) Covariance Q Covariance R
1.0 0.4 0.4 (gm- 3)2

1.0 0.4 0.2 (g m - 3)2
0.005 o (day-I)2
0.005 o (day-I)2
2.0 0.05 (gm-"day-I)2
2.0 0.05 (g m -.1 day - 1)2
indicates that a significant addition of BOD to the reach occurs over the period
from '40 to '50 and from about '65 onward, even though this particular section
of the River Cam receives no direct local surface runoff and these intervals were
periods of warm, sunny weather. The estimates of DB (Figure 11.9(f)) likewise
suggest a net addition of DO to the stream during '40-' 50; but for the initial
20-25 days of the experiment there exists the opposite apparent effect of a net
removal of DO from the water body (a result primarily, it is thought, of biased
BOD measurements). The recursive estimation trajectories for the parameters
K 1 and K 2 in Figures 11.9(c) and (d) are more or less stationary, although the
estimates [(2 of the reaeration rate constant undergo substantial modification
between '40 and, 50'
In section 2.6.3 of Chapter 2 we tried to illustrate the qualitative features of
model structure identification. Figure 2.9 attempted in a conceptual fashion to
show how recursive parameter estimation algorithms can' yield useful diag-
nostic information about this problem. Figure 11.9(d) is a specific realization of
that earlier conceptual picture. It was also stated earlier that" for much of the
time ... model structure identification ... is confronted with the need to offer
plausible hypotheses about 'unexplained' relationships in a set of field data."
Why, for example, is there an apparent addition of BOD to the reach at certain
times of the experiment? The primary hypothesis to emerge from the analysis is
that significant growth of a floating algal population is responsible for part of
the observed dynamic BOD-DO interaction. It is further postulated that the
growth rate of algae is governed by the prevailing sunlight conditions, an
assumption that thus requires further systematic evaluation (Beck and Young,
1976; Beck, 1978b). The results of Figure 11.9 are representative merely of a
part of the difficult process of model structure identification.
Case Study 2: Jordan River, Utah

The second case study is concerned with a steady state water quality modeling
problem (Bowles and Grenney, 1978a). We shall see how it illustrates different
8.0
(a)
Downstream 6.0
BOD (g m-3 ) 4.0
Downstream
DO (g m-3 )
(c)
BOD decay rate
constant (day-1) 0.2
0.0
0.4 (d)
Reaeration rate
constant (day-1) 0.2
0.0
Net rate of 2.0
addition of BOD
to reach 1.0
(g m- 3 day-1) 0.0
-1.0
2.0 (f)
Net rate of 1.0

addition of DO
0.0
to reach
(g m- 3 day-1)
10 20 30 40 50 60 70 80
Time (days)
FIG U RE 11.9 Results from the application of the EKF in the River Cam case study
(adapted from Beck and Young. 1976): (a) observations and state estimates. ct. for
downstream BOD; (b) observations and state estimates, c!, for downstream DO; (c)
estimates,cr, for the BOD decay rate constant (Kd; (d) estimates. ct. for the reaeration
rate constant (K 2 ): (e) estimates, i'~. for the net rate of addition of BOD to the reach
(LA): (f) estimates. ct. for the net rate of addition of DO to the reach (DB)'
454
455
aspects of the EKF algorithm, principally at the stage of formulating the system
equations. The state vector variations are again provided in the form of lumped-
parameter, ordinary differential equations, though here distance along the river
(or time of travel) is the independent variable:
cl(r) K*(Ll(r)-cl(r»-Klcl(r)
cir) K*(Lir)-cir»+Ks2c5Cr)-K23cir)- f2{c 2(r), cir)}c4(r)
c3(r) K*(L 3(r)-cir»+K 23 cir)- f3{cir), cir)}c 4(r)
c4(r) K*(L 4(r)-c4(r»-K 45 c4(r)+ f4{c2(r), cir)}c 4(r)
i' 5(r) K*(L 5(r) - c5(r» + K 45 C4( r) - K 52 C 5(r)
c6(r) K*(L 6 (r) - c6(r» - DB + K 2(C(r) - c6(r» - Klel(r) - yK 23 c 2(r)
~l (r)
~2(r)
~ir)
+ ~4(r)
(11.33)
~5(r)
~6(r)
cl(r) to c6(r) are the respective stream concentrations of carbonaceous

BOD, ammonia nitrogen, nitrate nitrogen, algal nitrogen,
organic nitrogen, and dissolved oxygen (g m - 3),
K* is a rate constant for the addition of components by lateral
inflow (day-l, in travel time),
K l , K 2 , e,er) are as defined for (11.30),
Ll(r) to L 6(r) are the respective concentrations of each component in
lateral inflow (g m - 3),
K 23 is the first-order rate constant for conversion of ammonia
nitrogen to nitrate nitrogen (day-l),
K 45 is the death rate constant for algae (day- 1),
K 52 is the rate constant for decomposition of organic nitrogen
into ammonia nitrogen (day-l),
DB is the rate of oxygen consumption by bottom mud deposits
(g m- 3 day-l),
y is a coefficient for the rate of oxygen consumption by
nitrification (dimensionless),
f2 {. },f3 {. }, f4 {. } are nonlinear functions describing, respectively, the rates
of uptake of ammonia nitrogen and nitrate nitrogen by
algae and the growth rate of algae (day-l).
For this second case, therefore, the state vector behavior is nonlinear. More-
over, the observations are not as straightforward as in the River Cam study:
one ofthe observations is available as the linear sum of two ofthe state variables,
456
organically bound nitrogen and nitrogen bound as algal cell material. The noise-
corrupted output observations are thus given by
100 000
o 1 0 0 0 0
CO(Tk) = 0 0 1 0 0 0 C(Tk) + T\(Tk), (11.34)
o 0 0 1 1 0
o 0 0 001
where these are measurements defined at each spatial location Tk •
(a)
r~
'7 15.0
E
Cl
10.0 • ••
o
oII)
5.0 -/
.-r.
0.0
(b)
(Y)
15.0
I
~"........
E
10.0
..-.-.-.-
Cl
o
o 5.0 ~ • •
0.0
u -
'c '7 (c)
~ E
o
.....
Cl
C 2.0 •
,.--..~.
o Q)
c Cl
o 0....
•
.~
co .'!::
;;; C 1.0
C
~
[tJ
~ .2'
C [tJ •• ••
8 + 0.0
60 50 40 30 20 10 0
Distance from Great Salt Lake (km)
FIGU RE 11.10 Results from the application of the EKF in the Jordan River case study
(adapted from Bowles and Grenney. 1978a): (a) observations and state estimates. ci • for
BOD; (b) observations and state estimates. c2 , for DO; (c) observations and state esti-
mates, C4 + (-5. for organic nitrogen plus algal nitrogen.
457
Some of the results quoted by Bowles and Grenney are shown in Figures 11.10
and 11.11. Figure 11.10 represents a typical set of estimates for BOD, DO, and
algal and organically bound nitrogen concentration when the EKF is applied
to the problem of state estimation only with eqns. 11.33 and 11.34. In contrast
to the trajectories of Figure 11.9, Figure 11.10 gives the patterns of both pre-
dicted, c('k I'k _ 1), and corrected, c('k I'k), state estimates generated by the
prediction and correction algorithms of the filter (eqns. 11.28a and b, respec-
tively). Some of the changes evident in the estimated states are, however, due to
the effects of point-wise addition of constituents from discharges coincident with
the measurement location. The persistently poor correspondence between the
estimates and observations for algal and organically bound nitrogen is described
by Bowles and Grenney as a consequence of the large errors attributed to these
measurements. This, then, is a case of the filter" believing" it has an accurate
model for this part of the system (recall the discussion of section 11.3.2) so that
large errors are ignored as the spurious consequences of chance.
Lastly, Figure 11.11 provides estimates for the parameter L 2 (,), the ammonia
nitrogen concentration in lateral inflow, when the augmented state vector is
defined as
and when Li,) is idealized as a random walk parameter. The principal objective
of this analysis, therefore, is not model calibration, in the sense of model
structure identification or parameter estimation. Rather, such an application of
the EKF is directed toward the combined use of model and field data for
reconstruction of information about variables that may be of interest from the
point of view ofwater quality management. Similar and more extensive results for
the estimation of spatially varying parameters in this case study can be found in
Bowles and Grenney (1978b).
C")
I
_ c E 2.0
o ~ Cl
C 0 - 1.5
.2 ;; ~
ro 'c
.... t't1 ~
0
1.0
E· c .~
a> 0 Cii
g E Qi 0.5
uo CllE ~
.... 0 0 L::::L:==r__..L-_ _..L-_ _, l -_ _, l - _ - - - l
C • 60 50 40
Distance from Great Salt Lake (km)
FIGURE 11.11 Results from the application of the EKF in the Jordan River case study
(adapted from Bowles and Grenney, 1978a): estimated concentration of ammonia nitrogen
in lateral inflow (L 2 ).
458
C oncll/ding Remarks
The two case studies have illustrated some of the potential benefits of using
recursive estimation algorithms in solving the problem of model calibration.
There are, however, limitations on the performance of these algorithms, and of
the extended Kalman filter in particular. For example, the choices of variance-
covariance matrices P( '0 I'0), Q, and R, which are unavoidably subjective
choices, determine in part the behavior of the filtering algorithms. It is also
necessary to have reasonable a priori estimates P(,o I'o)ofthe model parameters,
since the EKF does not guarantee convergence to a globally" optimum" set of
estimates. Nevertheless, if the analyst is aware of these and other limitations,
then the analysis will be all the more effective, and less likely to lead to erroneous
conclusions. However in this respect a recent development of particular
significance is a modified form of the EKF, proposed by Ljung (1979), which
overcomes the difficulties of convergence associated with the original EKF.
11.3.6. Verification
Most ofwhat needs to be stated about verification was introduced in section 2.6.5.
The essence of model verification, as understood here, is the problem of checking
the statistical properties of the predicted model response errors, i.e.
(11.35)
where CO('k) is a model-related estimate of the vector of measured outputs
(see also Figure 2.10). Typically, according to the definition of the system and its
variables in Figure 11.4, assumptions are made about the statistical properties of
the stochastic sequences ~ and 1'1 (system disturbances and measurement errors,
respectively). Usually these assumptions require ~ and 1'1 to be zero-mean,
white noise sequences, or to have been generated by simple manipulations from
such sequences. It is customary also to assume that the sequences are drawn
from random variables with Gaussian distributions. If all these assumptions are
valid, the model response errors will be required to conform, in the majority of
cases, with the statistical properties of zero-mean, Gaussian, white noise
sequences. Stated more specifically, it is required that:
E{£('k)} = 0 (11.36a)
E{l:;(rk)l:j{'/)} = 0 for all k, I and for i =I j; i, j = 1,2.... , p (11.36b)
E{l:;('k)l:;(r/)} = 0 fork=ll;i=I,2, ... ,p (11.36c)
E{l:i('k)d j (,/)} = 0 for all k, I and for i = 1, 2, ... , p; j = I, 2, ... , m.
(11.36d)
Taken in turn, the conditions of (11.36) state that: the errors have sample mean
values of zero (11.36a); are not cross-correlated among themselves (11.36b);
459
are not autocorrelated (l1.36c); and are not correlated with the measured
input forcing functions (l1.36d). Results illustrative of this kind of analysis are
given in Beck (l978b).
11.4. VALIDAnON
Findeisen et al. (1978) define validation in the following terms: "a model can
never be completely validated; we can never prove that its results conform to
reality in all respects; it can only be invalidated." Models are working hypotheses
about the nature of the behavior of a system. While the analyst may seek con-
firmation of his hypotheses in the process of validation, the basic purpose of
validation is in fact to find invalid hypotheses; and knowledge of invalid hy-
potheses should ultimately lead to revised and better approximations (models) of
reality. That, we presume, is a goal of every modeling exercise.
1.4 ,------,---------,--x--,-----,----,----------,--,-------,----,--------,
1.2 1952 1953 195t\ 1955
\
1.0 \
0.8
M
E 0.6
Cl 0.4
~ 0.2 - -x
~ 0.0147101471014710147101471014710147101471014710
a.
~ 1961 1962 1963 1964 1965 1966 1967 1968
0.8
..r::.
a.
(1J0.6
'0 0.4 j\
~ 0.2 x Y \'x-,x
x-x -x V
§
'';:; 0.0147101471014710147101471014710147101471014710
~ 1.0 ~ 1970 1971 1972 1973 1974 1975

53 0.8
u
o 0.6
c:: x
x x
U 0.4:>il<~ xxi
0.2 ~V
0.0 1 3579111 357 9111 3 5 791113 579111 3 57 9111 3579 11
FIGURE 11.l2 Validation of a model for the concentration of total phosphorus in
Lyngby Lake (adapted from Jorgensen el al.. 1978). The continuous line represents the
observed data; the broken line represents the values predicted by the model. The model was
calibrated with data up to and including the end of 1958; diversion of the sewage discharge
was implemented during the first quarter of 1959.
460
In precise mathematical terms no further remarks on model validation will be
made here. Suffice it to say that verification, as discussed in the previous section,
is not a particularly rigorous test of whether a model is "good" or "bad." It is
possible to fit a model to a set of field data with almost arbitrarily small error,
provided that there is a sufficient number of parameters in the model. Such a
model would probably be of trivial value in terms of solving problems or in terms
of understanding system behavior. It is of greater importance to learn whether
a model that has been calibrated and verified with one set of data gives a
reasonable approximation of behavior observed in a second (and independent)
set of field data. As an example of this latter type of validation procedure, let
us take the case of Lyngby Lake in Denmark (J0rgensen et aI., 1978). A model was
calibrated and verified for the period 1952-58, during which time the lake was
receiving an effluent of wastewater that had been treated both mechanically and
biologically. After 1958 the wastewater discharge was diverted to a coastal
location. Figure 11.12 shows a comparison of the field data for 1952-75 with
the estimated responses of the model. Having calibrated the model for the
period 1952-58, 10rgensen et al. made no further adjustment of the model
parameter estimates. Thus 1959-75 is a period used for model validation; it is
also a period in which we may assume that the behavior of the lake was sig-
nificantly different from the observed behavior (1952-58) used for calibration.
The reader is left to draw his own conclusions about the results of Figure 11.12-
validation, like many other topics of this chapter, is a matter of personal
judgment.
11.5. SUMMARY AND CONCLUSIONS
Chapters 2 and 11 are complementary chapters. The objective of this chapter

has been to provide a more detailed discussion of the problems and methods of
sensitivity analysis, calibration, and validation. There are clearly limits to the
power of our current methods of model calibration, as this chapter has demon-
strated. It is true that elegant and efficient algorithms of parameter estimation
are available in theory, but at the interface between practice and theory in water
quality model calibration it is a robust, workable algorithm that the analyst
really requires. When dealing with field data it is rare, if ever, that the analysis
yields elegant solutions with great efficiency.
The chapter has not touched upon many of the philosophical matters that
inevitably surround the development of mathematical models and their
calibration. Perhaps this is a serious omission, for there are some fundamental
questions that need to be asked about the relationship between model develop-
ment and model evaluation (Young, 1978; Beck, 1981; Lewandowski, 1981).
Thomann, for example, urges that more attention be given to model verification
(Thomann and Winfield, 1976):" ... it is no longer of great moment if hundreds
461
of sets of nonlinear equations are successfully solved on a large computer.
What is of significance, however, is whether the numerical computations are
reasonable representations of the real world."
Thomann's concern over the rapidly increasing size of water quality models
without a consistent increase in the verified capabilities of these models is a
concern to which others would probably subscribe. But what is "reasonable"
to one person may well be quite different from what is "reasonable" to another
person. Some recent attempts at a formal resolution-by reference to experi-
mental observations-of these vexed questions of model verification and suf-
ficient model complexity provide sobering evidence for the builder of large
models (Maciejowski, 1978, 1979). On the other hand, a small model calibrated
accurately against limited historical field data may not be capable of predicting
responses to substantially altered future input disturbances. No doubt there will
always be a gap between what can be simulated, in theory, and what can be
verified in practice. One hopes that the size of the gap will not increase.
REFERENCES
Akaike, H. (1974). A new look at statistical model identification. IEEE Transactions on Automatic
Control AC-19:716-722.
Astrom, K. J., and T. Bohlin (1966). Numerical identification of linear dynamic systems from
normal operating records. Theory o( Self~adaptil'e Control Systems, ed. P. H. Hammond (New
York, NY: Plenum). pp. 96-111.
Astrom, K. J., and P. Eykhoff(l971). System identification-A survey. AUlOmatica 7: 123-162.
Beck, M. B. (1975). The identification of algal population dynamics in a non tidal stream. Com-
puter Simulation of Water Resources Systems, ed. G. C. Vansteenkiste (Amsterdam: North-
Holland), pp. 483-494.
Beck, M. B. (l978a). A comparative case study of dynamic models for DO-BOO-algae interaction
in a freshwater river. Research Report RR-78-19 (Laxenburg, Austria: International Institute for
Applied Systems Analysis).
Beck, M. B. (I 978b). Random signal analysis in an environmental sciences problem. Applied
Mathemu.ical Modelling 2(1) :23-29.
Beck, M. B. (l979a). Model structure identification from experimental data. Theoretical Systems
Ecology, ed. E. Halfon (New York, NY: Academic Press), pp. 259-289.
Beck, M. B. (l979b). System identification, estimation and forecasting of water quality-Part I:
Theory. Working Paper WP-79-31 (Laxenburg, Austria: International Institute for Applied
Systems Analysis).
Beck, M. B. (1981). Hard or soft environmental systems? Ecological Modelling II :233-251.
Beck, M. B., and P. C. Young (1976). Systematic identification of DO-BOD model structure.
Proceedings of American Society of Cil'il Engineers, Journal of Environmental Engineering
Division I02(EE5): 909-927.
Bellman, R. E., and R. E. Kalaba (1965). Quasilinearization and Nonlinear Boundary Value Problems
(New York, NY: American Elsevier).
Benson, M. (1979). Parameter fitting in dynamic models. Ecological Modelling 6: 97-115.
de Boer, B. (1979). A moving cell method of the dissolved oxygen and phytoplankton dynamics in
rivers. Hydrological Sciences Bulletin 24(1): 199-211.
462
Bowles, D. S., and W. J. Grenney (I 978a). Steady-state river quality modeling by sequential
extended Kalman filters. Water Resources Research 14:84-95.
Bowles, D. S., and W. J. Grenney (1978b). Estimation of diffuse loading of water quality pollutants
by Kalman filtering. Applications o.lKalman Filter to Hydrology, Hydraulics and Water Resources,
ed. c.-L. Chiu (Pittsburgh, PA: University of Pittsburgh, Stochastic Hydraulics Program),
pp. 581-597.
Box, G. E. P., and G. M. Jenkins (1970). Time-series Analysis, Forecasting and Control (San Fran-
cisco, CA: Holden-Day).
Di Cola, G., L. Guerri, and H. Verheyden (1976). Parametric estimation in a compartmental
aquatic ecosystem. Preprints of Fourth IFA C Symposium on Identification and System Parameter
Estimation, Institute ol Control Sciences, Moscow. Part 2, pp. 157-165.
Di Toro, D. M. (\ 969). Stream equations and method of characteristics. Proceedings ol American
Society 0.1 Civil Engineers, Journal ol Sanitary Engineering Division 95(SA4): 699-703.
Di Toro, D. M., D. J. O'Connor, and R. V. Thomann (1971). A dynamic model of the phyto-
plankton population in the Sacramento-San Joaquin Delta. American Chemical Society,
Advances in Chemistry Series 106: 131-180.
Dj Toro, D. M., and G. van Straten (1979). Uncertainty in the parameters and predictions of
phytoplankton models. Working Paper WP-79-27 (Laxenburg, Austria: International Institute
Dobbins, W. E. (1964). BOD and oxygen relationships in streams. Proceedings of American Society
ol Cil'il Engineers, Journal ol Sanitary Engineering Division 90: 53-78.
Dorf, R. C. (1965). Time-domain Analysis and Desigll olControl Systems (Reading, MA:Addison-
Wesley).
Draper, N. R., and H. Smith (1966). Applied Regression Analysis (New York, NY: Wiley).
Edwards, A. M. c., and J. B. Thomes (1973). Annual cycle in river water quality: A time-series
approach. Water Resources Research 9: 1286-1295.
Erni, P. E., and J. Ruchti (1977). Der Sauerstoffhaushalt von Fliessgewiissern: Kritische Priifung
eines mathematischen SauerstoffmodeHs an hand der Identifikation seiner Parameter in Fliissen
und kiinstlichen Rinnen. Schweizerische Zeitschrilifilr Hydroloqie 39(2): 26 I-276.
Eykhoff, P. (1974). System Identification-Parameter and State Estimation (Chichester: Wiley).
Findeisen, W.. A. Iastrebov. R. Lande, J. Lindsay, M. Pearson, and E. S. Quade (1978). A sample
glossary of systems analysis. Working Paper WP-78-12 (Laxenburg. Austria: International
Fuller, F. c., and C. P. Tsokos (1971). Time-series analysis of water pollution data. Biometrics
27: 1017-1034.
Gelb, A. (ed.) (1974). Applied Optimal Estimation (Cambridge, MA: MIT Press).
Gnauck, A., J. Wernstedt, and W. Winkler (1976). On the use of real-time estimation methods for
the mathematical modelling of Iimnological ecosystems. Preprints of Fourth IFAC Symposium
on Identification and System Parameter Estimation, Institute o.lControl Sciences, Moscow, Part 2,
pp. 124-133.
Halfon, E. (1979). On the parameter structure of a large-scale ecological model. Contemporary
Quantitative Ecologr and Related Ecometrics, eds. G. P. Patil and M. L. Rosenzweig (Fairland,
MD: International Cooperative Publishing House), pp. 279-293.
Halfon, E., H. Unbehauen, and C. Schmid (1979). Model order estimation and system identifica-
tion theory and application to the modeling of 32p kinetics within the trophogenic zone of a small
lake. Ecological Modelling 6: 1-22.
Huck, P. M., and G. T. Farquhar (1974). Water quality models using the Box-Jenkins method.
Proceedings oj' American Society of Civil Engineers, Journal o.l Em'ironmental Engineering
Dil'ision I OO( EE3): 733-752.
Ivakhnenko, A. G., G. I. Krotov, V. I. Cheberkus, and V. N. Vysotskiy (1977). Identification of
dynamic equations of a complex plant on the basis of experimental data using self-
organization of models, Part I-One-dimensional problems. Soviet Automatic Control
10(2): 24-30.
463
Jazwinski, A. H. (1970). Stochastic Processes and Filtering Theory (New York, NY: Academic
Press).
Johnston, J. (1963). Econometric Methods (New York. NY: McGraw-Hill).
Jolankai, G., andA. Szollosi-Niigy (19]g). A simple eutrophication model for the Bay of Keszthely,
Lake Balaton. Modeling the Water Qualin' or the Hydrological Cycle. Proceedings or Baden
Symposium, September. International Association or Hydroloqical Sciences Publication 125,
pp.137-150.
Jorgensen, S, E.. H. Mejer, and M. Friis (1978). Examination ofa lake model. Ecological Modelling
4: 253-278.
Kallstrom, c. G., T. Essebo, and K. J. Astriim (1976). A computer program for maximum likeli-
hood identification of linear multivariable stochastic systems. Preprints o( Fourth IF AC
Symposium on Identification and System Parameter Estimation, Institute or CO/1/rol Sciences,
Moscow, Part 2, pp. 508-521.
Kalman, R. E. (1960). A new approach to linear filtering and prediction problems. Transactions
or American Society of Mechanical Engineers, Series D: Journal or Ba,ric Engineering 82: 35-45.
Kalman, R. E., and R. S. Bucy (1961). New results in linear filtering and prediction theory.
Transactions o(American Society o(Mechanical Enyineers, Series D: Journal or Basic Engineering
83:95-108.
Kendall, M. G., and A. Stuart (1961). Ad1'Onced Theory o(Statistics (London: Griffin).
Koivo, A. J., and G. R. Phillips (1971). Identification of mathematical models for DO and BOD
concentrations in polluted streams from noise corrupted measurements. Water Resources
Research 7(4): 853-862.
Koivo, A. J., and G. R. Phillips (1972). On determination of BOD and parameters in polluted
stream models from DO measurements only. Water Resources Research 8(2): 478-486.
Koivo, A. J., and G. R. Phillips (1976). Optimal estimation of DO, BOD and stream parameters
using a dynamic discrete time model. Water Resources Research 12(4): 705-711.
Koivo, H. N.. and A. 1. Koivo (1978). Least-squares estimalor for polluted stream variables in a
distributed parameter model. Adl'ances in Water Resources I: 191-194.
Lee, E. S. (1968). Quasilinearization and In1'Oriant Imbedding (New York, NY: Academic Press).
Lee, E. S., and I. Hwang (1971). Stream quality modeling by quasilinearization. Journal of Water
Pollution Control Federation 43(2): 306-317.
Lettenmaier, D. P., and S. 1. Burges (1976). Use of state estimation techniques in water resource
system modeling. Water Resources Bulletin 12( I): 83-99.
Lewandowski, A. (1981). Issues in model validation. Working Paper WP-81-32 (Laxenburg,
Austria: International Institute for Applied Systems Analysis).
Lewis. S., and A. Nir (1978). A study of parameter estimation procedures ofa model for lake phos-
phorus dynamics. Ecological Modelling 4:99-117.
Ljung. L. (1979). Asymptotic behavior of the extended Kalman filter as a parameter estimator for
linear systems. IEEE Transactions on Automatic Control AC-24: 36-50.
Maciejowski, J. M. (1978). The modeling of systems with small observation sets. Lecture NOles in
Control and Information Sciences vol. 10 (Berlin: Springer).
Maciejowski, J. M. (1979). Model discrimination using an algorithmic information criterion. Auto-
matico 15: 579-593.
Mehra. R. K., and D. G. Lainiotis (eds,) (1976). System Identification: Adl'ances and Case Studies
Mehta. B. M., R. C. Ahlert, and S. L. Yu (1975). Stochastic variation of water quality in the Passaic
River. Water Resources Research 11(2): 300-308.
Meissinger, H. F. (1960). The use of parameter influence coefficients in computer analysis of
dynamic systems. Proceedings of Western Joint Computer Conference, San Francisl'o, pp. 181-192.
Moore, R. J., and D. A. Jones (1978). Coupled Bayesian-Kalman filter estimation of parameters
and states of dynamic water quality model. Application of Kalman Filter to Hydrology, Hydraulics
and Water Resources, cd. c.-L. Chiu (Pittsburgh, PA: University of Pittsburgh, Stochastic
Hydraulics Program), pp. 599-635.
464
O'Connor, D. 1., and D. M. Di Toro (1970). Photosynthesis and oxygen balance in streams.
Proceedings or American Society of Cit'il Engineers, Journal of Sanitary Engineering Division
96(SA2): 547-571.
Rinaldi, S., and R. Soncini-Sessa (1978). Sensitivity analysis of generalized Streeter-Phelps models.
AdL'ances in Water Resources I: 141-146.
Rinaldi, S., R. Soncini-Sessa, H. Stehfest, and H. Tamura (1979). Modeling and Control of Ril'er
Quality (New York, NY: McGraw-Hill).
Scavia, D. (1980). Uncertaintyanalysisofa lake eutrophication model. Ph.D. Dissertation, Unjz'ersity
or Michigan, Ann Arbor, MI.
Schurr, 1. M., and 1. Ruchti (1975). Kinetics of oxygen exchange, photosynthesis and respiration in
rivers determined from time-delayed correlations between sunlight and dissolved oxygen.
Schwei::erische Zeitschri/i./ur Hydrologie 37( I): 144-174.
Shastry, 1. S., L. T. Fan, and L. E. Erickson (1973). Nonlinear parameter estimation in water
quality modeling. Proceedings of American Society of Civil Engineers, Journal or Em'ironmental
Engineering Dit'ision 99(EE3): 315-331.
Soderstrom, T., L. Ljung, and I. Gustavsson (1978). A theoretical analysis of recursive identifica-
tion methods. Automatica 14:231-244.
Stehfest, H. (1978). On the monetary value of an ecological river quality model. Research Report
RR-78-1 (Laxenburg, Austria: International Institute for Applied Systems Analysis).
van Straten, G. (1983). Maximum likelihood estimation of parameters and uncertainty in phyto-
plankton models. Uncertainty and Forecasting of Waler Quality, eds. M. B. Beck and G. van
Straten (Berlin: Springer), pp. 157-171.
van Straten, G., and B. de Boer (1979). Sensitivity to uncertainty in a phytoplankton-oxygen model
for lowland streams. Working Paper WP-79-28 (Laxenburg, Austria: International Institute
Tamura. H. (1979). On some identification techniques for modeling river quality dynamics with
distributed lags. Handbook or Large-scale Syslems Engineering Applicalions, eds. M. G. Singh
and A. Titli (Amsterdam: North-Holland), pp. 274-294.
The, G. (1978). Parameter identification in a model for the conductivity of a river based on noisy
measurements at two locations. Modeling, Identification and Control in Environmental Systems,
ed. G. C. Vansteenkiste (Amsterdam: North-Holland), pp. 823-830.
Thomann, R. V. (1963). Mathematical model for dissolved oxygen. Proceedings of American
Society or Civil Engineers, Journal or Sanitary Engineering Dil'ision 89(SA5): 1-30.
Thomann, R. V. (1967). Time-series analysis of water quality data. Proceedings a/American Sociely
or Cir,il Engineers, Journal 0/ Sanilary Engineering Dil'ision 93(SA 1): 1-23.
Thomann, R. V., and R. P. Winfield (1976). On the verification of a three-dimensional phyto-
plankton model of Lake Ontario. Proceedings of Conference on Em·ironmental Modeling and
Simulation. US Environmental Protection Agency, WashinglOn, DC Report EPA 600/9-76-016,
pp. 568-572.
Tomovic, R. (1964). SenSilil'ily Analysis or Dynamic Syslems (New York, NY: McGraw-Hill).
Unbehauen, H., and B. Gohring (1974). Tests for determining model order in parameter estimation.
Automatica 10: 233-244.
Van den Boom, A.1. W., and A. W. M. Van den Enden (1974). The determination of the orders of
process and noise dynamics. AUlomatica 10: 245-256.
Wellstead, P. E. (1976). Model order testing using an auxiliary system. Proceedings of Institution or
Eleclrical Engineers 123: 1373-1379.
Wellstead, P. E. (1978). An instrumental product moment test for model order estimation. AUIO-
matica 14: 89-91.
Whitehead, P. G. (l980a). An instrumental variable method of estimating differential-equation
models of dispersion and water quality in nontidal rivers. Ecological Modelling 9: 1-14.
Whitehead, P. G. (1980b). Real-time monitoring and forecasting of water quality in the Bedford
Ouse river system. Hydrological Forecasting. Proceedings or Oxford Symposium. Inlernalional
Associalion of Hydrological Sciences Publication 129. pp. 333-342.
465
Whitehead, P. G., and P. C. Young (1975). A dynamic-stochastic model for water quality in part
of the Bedford Ouse river system. Computer Simulation of Water Resources Systems, ed. G. C.
Vansteenkiste (Amsterdam: North-Holland), pp. 417-438.
Wilde, D. J. (1964). Optimum Seeking Methods (Englewood Cliffs. NJ: Prentice-Hall).
Wilde, D. J., and C. S. Beightler (1967). Foundations 01' Optimization (Englewood Cliffs, NJ:
Prentice-Hall).
Young, P. C. (1974). A recursive approach to time-series analysis. Bulletin ofthe Institute of Mathe-
math's and Its Application 10: 209-224.
Young, P. C. (1976). Some observations on instrumental variable methods of time-series analysis.
International Journal 01' Control 23: 593-612.
Young, P. C. (1978). General theory of modelling for badly defined systems. Modelling, Identification
and Control in Em'ironmental Systems, ed. G. C. Vansteenkiste (Amsterdam: North-Holland),
pp. 103-135.
Young, P. c., and P. G. Whitehead (1977). A recursive approach to time-series analysis for multi-
variable systems. International Journal of Control 25 :457-482.
A system matrix for state variable relationships in a linear

system
B input matrix for relationships between input variables and
state variables in a linear system
vector of state variables
augmented vector of state variables and parameters
vector of measured output response variables
vector of state estimates
vector of model-predicted values for the output response
variables
nominal deterministic reference trajectory for the state
vector
saturation concentration of dissolved oxygen
vector of measurable input disturbances
net rate of addition of DO to reach of river by the combined
effects of photosynthesis, respiration, and decomposition
of mud deposits
DB rate of oxygen consumption by mud deposits
E{· } expectation operator
f{. } nonlinear vector-valued function for variations in the state
variables
nonlinear functions for, respectively, the uptake of
ammonia nitrogen and nitrate nitrogen by algae, and the
growth rate of algae
system matrix for relationships between small perturbations
in a linearized system
466
G('k) gain matrix in the correction procedure of the Kalman

filter
G*('k) gain matrix in the correction procedure of the extended
Kalman filter
H matrix for the relationships between observations in a
linear system
I unit identity matrix
J (squared error) loss function
K1 BOD decay rate constant
K2 reaeration rate constant
K 23 first-order rate constant for conversion of ammonia
nitrogen to nitrate nitrogen
K 4S rate constant for death of algae
K S2 rate constant for decomposition of organic nitrogen to
ammonia nitrogen
K* rate constant for the addition of components by lateral
inflow
LA rate of addition of BOD to reach by, for instance, local
surface runoff
Li concentration of water quality component i in lateral
inflow
m number of input variables
m{· } nonlinear vector-valued function for relationships between
state and output variables
M('k) matrix for the relationships between observations of a
linearized small-perturbation system
n number of state variables
p number of output variables
P(, I'k) (state-parameter) estimation error covariance at time "
given all measured information up to and including that
available at time 'k
q number of parameters
Q system noise covariance matrix
Q(,) stream discharge
R measurement error covariance matrix
s sensitivity coefficient
V volume of water in reach of river
W matrix of weighting coefficients
~ vector of parameters
P nominal reference vector of parameter values
~ vector of parameter estimates
y coefficient for rate of oxygen consumption by nitrification
<'lkl Kronecker delta function
467
predicted model response errors (or residual errors)
vector sequence of disturbances for random walk param-
eters
'I vector of (stochastic) measurement errors
.A. scalar step length in an off-line parameter estimation
algorithm
vector of (stochastic) unmeasured input disturbances of
system
independent variable of time or of time of travel
kth discrete sample in time or in time of travel
state transition matrix of a linearized small-perturbation
system
'I' matrix of coefficients in an off-line parameter estimation
algorithm.
Models for Management Applications
12 D. P. Loucks
12.1. INTRODUCTION
Water quality management planning involves the identification and evaluation

of various management alternatives for achieving economic and water quality
goals. Economic goals are often expressed in terms of cost effectiveness (cost
minimization) and in terms of the distribution of the cost among those who
should pay. Water quality goals are usually expressed as wastewater effluent
standards or water quality standards in waste-receiving water bodies, or both.
The effectiveness of any management alternative may be measured in terms of
how well it accomplishes these goals. Water quality management models can
assist planners in identifying and evaluating possible management alternatives
in order to determine which alternative is best.
Water quality management models are usually extensions of some of the
simulation or predictive models discussed in the previous chapters. In addition
to the predictive equations, management models include as unknowns the
design and operating policy variables of each management alternative. Relation-
ships are included that describe the resulting effluent or water quality, and the
cost of each management alternative as a function of the design and operating
policy variables. Also included in these models are the constraints defining the
desired effluent and/or water quality standards.
Most water quality management models are optimization models. As such,
the relationships that define economic costs and the resulting effluent or water
quality as functions of the unknown design or operating variables must conform
to what is required for model solution using particular optimization solution
procedures. If simplifications or modifications are necessary purely for model
468
469
solution, then, as emphasized throughout this chapter, the resulting solution
should be checked using more accurate water quality simulation models.
Management models are used for a preliminary evaluation of various alterna-
tives and for identifying what data are important and needed prior to the
implementation of a more expensive data collection and simulation study.
12.2. MANAGEMENT ALTERNATIVES FOR

WATER QUALITY CONTROL
The first and most obvious method of water quality control is to limit the
amount of waste discharged into water bodies. This type of control can take
on numerous forms, some of which are described below.
(1) Each waste producer is required to discharge less waste, for example
through process changes or removal of at least some minimum specified
fraction of the waste prior to releasing the remainder into natural waters
or land waste disposal areas. Removal of waste can be accomplished by
a variety of physical, biological, and chemical processes.
(2) The portion of the treated wastewater effluent that, if released into the
natural water body, would result in a lower water quality than desired
is stored instead. Ponds or tanks can be used for effluent storage. The
quantity and timing ofdischarge of stored effluent to land or water should
depend in part on the assimilative capacity of the receiving body.
(3) Waste is piped, either prior to or following some treatment, to areas
within or outside of the region for additional treatment and/or disposal
at land or water sites having greater assimilative capacities. This
alternative also permits the processing of wastes at larger regional
facilities that benefit from economies of scale in construction and
operating costs, as well as from increased operating efficiencies.
(4) In-stream quality can be improved by artificial aeration or flow
augmentation. The deficit of dissolved oxygen can be decreased by
injection of air into the water. Increasing the stream flow in periods of
low flows by releasing water from upstream reservoirs may also improve
the stream quality by dilution and by changing velocities and tempera-
tures, which, in turn, affect the reaction rates of various quality
constituents.
Each of these means of water quality management will be discussed in

greater detail later in this chapter. Prior to this discussion some remarks on
management objectives and the criteria that are used for evaluating alternative
combinations of the various central options are appropriate.
470
12.3. MANAGEMENT OBJECTIVES AND QUALITY STANDARDS
Water quality management objectives are multiple and conflicting. Those in

control of activities that generate wastes would naturally prefer to dispose of
their wastes at no cost to themselves and, if possible, to others as well. This
policy leads to higher profits if income is being derived from the waste-making
activities, or to less taxes if the wastes are derived from human settlements such
as cities and municipalities.
However, if the discharge of wastes does result in added costs elsewhere, i.e.
in environmental damages, those who incur these costs and damages would
also prefer not to incur them. They can argue that dischargers of waste into
water bodies, for example, should pay for the environmental damage. Yet
because those who discharge waste are usually not affected by the damage
caused by that waste, there is no economic incentive to control that discharge.
Water pollution is said to be an externality, i.e. it is imposed on individuals
other than those who cause it. This is the central conflict involving water quality
management in river basins throughout the world.
Because the private market system fails to charge each polluter an amount
equal to the damages resulting from his waste discharge, regulatory action is
often required. The types of incentive that water quality regulatory agencies
have used to compensate for the failure of individual polluters to consider the
damages they impose on others are: (1) legislative, including direct regulation,
the establishment of effluent or stream quality standards, licensing, and zoning;
(2) legal, including compensation for damages and fines for violation of law;
and (3) economic, including effluent charges or taxes, subsidies, accelerated
depreciation allowances, and the like. Whatever the methods used, the objective
should be to achieve a more efficient and equitable allocation of natural re-
sources from the standpoint of society as a whole.
One of the difficulties in finding a plan that is both efficient and equitable
is the problem of quantifying water quality benefits or damages. This problem
is similar to that of attributing a monetary benefit to such things as aesthetics
and clean air. There is also the problem of determining equitable distributions
of costs and benefits. Thus, the selection of the desired water quality and the
determination of who will pay for it often become political decisions. This
political aspect is reflected in both the water quality management objectives
and the quality standards. Political systems have clearly demonstrated their
sensitivity to these multiobjective aspects of water quality management
problems.
Those who develop regional water quality management models usually
assume the actual or potential existence of some governmental institution that
has the authority to control water quality within its region, by economic in-
centives such as effluent charges and/or by legal means such as effluent
standards. The main purpose of most regional water quality models is to
471
examine alternatives that will reduce both the private costs and public damages
resulting from water pollution.
To begin quantifying a rather general objective for regional water quality
management, let us consider a river basin in which there are numerous indi-
viduals or groups s that discharge pollutants into the natural water courses.
Included among those individuals or groups of individuals are organizations,
such as state and federal pollution control agencies, that have financial as well
as political interests in the quality of the natural water within the basin. Water
quality control alternatives such as waste water treatment and effluent storage
impose costs on private agencies and, because of cost-sharing programs, on
public or governmental bodies as well. Quality control alternatives such as
flow augmentation and artificial aeration may only add to the cost paid by one
or more public agencies. Regardless of who pays, the cost to each individual
or group s can, for the moment, be denoted as a function of the scale of all
alternatives used for water quality control.
If Si is the scale of some waste reduction alternatives at site i costing q(SJ
for each group s, a cost-effective objective without regard to cost distribution
or to the political influence of each group can be written as:
minimize I I q(SJ (12.1)
,. ....
Such an objective mayor may not result in an acceptable solution. This depends
in part on the quality standards imposed. Usually the minimum allowable
quality standards are not intended to represent the desired quality. Environ-
mental protection goals and allowances for future growth and uncertainties
often result in planned or desired qualities that are higher than specified by
minimum quality standards. One way to achieve a quality that comes closer to
the target is through the proper allocation of effluent taxes Tf to the group s
discharging wastes at site i. The tax could be dependent on the waste released
at each site i.
The fraction Pi = Pi(SJ of the waste reduced or removed at each site i is, of
course, a function of the scale Sj of the waste reduction measures employed at
each site. If It; is the constant quantity of waste available at site i prior to the
implementation of any waste reduction measures, then It;(l - PJ is the re-
maining quantity that will be discharged. Ideally any tax on the amount of
waste discharged should reflect the external damages attributable to that
discharge. The purpose of the tax or subsidy is to provide an economic incentive
for reducing the external damages, if any, that result from the discharge of
wastes.
Effluent charges typically are paid to the appropriate river basin authority
or other public agency responsible for water quality management. The task of
such an agency in establishing and implementing effluent charges is to set
them in an equitable manner so as to cover the agency cost, not paid from other
sources, of measures taken by the agency to achieve a desired water quality
472
and for agency administration and operation. The dischargers themselves are
interested in minimizing their costs of waste treatment or reduction plus the
tax they must pay for the wastes that they discharge.
Let Ci(S;) represent the annual agency costs of measures S; taken at site i
to improve water quality, and let CHS;) be the individual or private costs, as
before. The aim of the control agency is to establish effluent tax rates TT per
unit of waste discharge U';(1 - P;) in such a way as to minimize the total cost
of water quality control:
minimize ~ ( C;(S;) + ~ q(S;)) (12.2)
agency other private

costs and/or public
costs
while ensuring that
(12.3)
s ;
total effluent total agency
tax income cost
and that the desired quality is maintained at all sitesj:

(12.4)
quality at site} minimum desired
quality at}
The individuals who must pay the cost of waste reduction or treatment and/or
an effluent tax on the waste U';(1 - P;) discharged at sites i are of course in-
terested in minimizing their total costs:
minimize L: [q(S;) + TT U';(1 - Pi)] \:Is. (12.5)
waste reduction effluent charge
cost
These objectives do not attempt to quantify the benefits or damages associ-
ated with the resulting water quality, except through the establishment of
effluent charges and, perhaps, quality standards. The charges and standards
would have to be defined prior to their incorporation into water quality models.
One of the advantages, however, of model construction and solution prior to
the final establishment of charges and standards is the ability to estimate the
costs of each group s and the resulting water quality associated with various
proposed combinations of charges and standards.
This problem, defined by eqns. 12.2-12.5, is an example of a multilevel-
multiobjective planning problem. The controlling agency objective (12.2) is to
be minimized subject to the minimization of a number of other objectives
(12.5), which are not controlled by the agency except through the establishment
473
of the effluent tax rates n in those objectives. Research has only begun to
identify some approaches to solving the overall multilevel-multiobjective
problem. In the absence of satisfactory solution procedures, the agency-level
objective by itself (the minimization of total cost of all measures implemented
for water quality control, eqn. 12.2) is typically chosen for a preliminary evalua-
tion of water quality management alternatives.
Yet with or without effluent taxes, the agency objective of cost minimization
has not been generally accepted. This is in part because there are, indeed, other
water quality management objectives at other levels of planning and decision
making, and it is not obvious how these objectives should be combined to
simulate, in a specific situation, what might be the actual response to a decision
at the central agency level.
The political process of establishing effluent charges and minimum accept-
able qualities, in the form of either effluent or stream quality standards, involves
the participation of each group of interested individuals within the river basin.
Some groups have more political influence than others. This depends not only
on their political skills but also on how strongly they feel about certain issues.
To include the effect of this influence in water quality models, it is often assumed
that relative weights can be defined and used in the objective function. Each
weight Ws reflects the relative influence that group s exerts compared with all
other groups defined by the model. If the proper weights, charges, and standards
are used, a socially or politically equitable and efficient water quality manage-
ment policy might result from the following objective:
minimize L Ws L [q(SJ + n ~(1 - Pi)]. (12.6)
The difficulty here, of course, is that the relative political weights are
unknown, even to the decision makers, until the final decision is made. By
varying the relative weights, however, an analyst can define some of the efficient
alternatives from the infinite set of possible alternatives. If the objective is
piecewise linear, the number of efficient alternatives defined by this procedure
will be smaller than the number of possible efficient alternatives. Efficient
alternatives can also be defined by setting upper bounds on all but one of the
terms within the square brackets in (12.5) and minimizing the other. Clearly,
both the former weighting approach and this latter target approach, and even
other more efficient iterative multiobjective approaches, merely define possible
solutions, not necessarily a best solution.
If all the relative weights are assigned values of unity, the objective function
(12.6) will represent the minimization of total costs and effluent charges without
regard to any redistribution of the costs and charges among various polluters.
As the relative weights or target levels change, so will the alternatives associated
with those weights or targets. Relatively high weights or low targets will corre-
spond to those polluters having a relatively strong political position and
474
interest, which in effect will reduce their share of the total cost. Varying the
weights or targets permits an examination of the stream quality that is likely
to be associated with various cost distributions.
Other means have been used to incorporate considerations of equity into
otherwise strictly cost-effective models (e.g. those having objective functions
of the type (12.2». These include constraints specifying equal scales of various
alternatives or some function of these scales, such as equal treatmen t efficiencies
or equal costs per capita contributing to the total waste at various sites. If Si is
the scale of quality control alternatives employed at site i, then constraints
requiring equal. scales of control could be written as
Sj=Sk ViEZk , (12.7)
where Zk is the set of sites in zone k of the region. Zones within a region may
be defined geographically or by types of polluter.
Although numerous analysts have included equity within the constraint set
of their models, it could be argued that equity is an objective-one of many
that water quality planners consider. The relative weight given to an equity
objective depends in part on the economic costs of achieving it, as well as on
the administrative and political costs of not achieving it, i.e. the unquantifiable
cost associated with implementing a plan that calls for a wide range of quality
control requirements within a region and that minimizes only the total economic
costs. Equity within any zone of a river basin can be expressed as an objective
by defining and then minimizing the absolute difference between the minimum
and maximum scales of water quality control within the zone. Let s~nin and Sk'"'
denote variable lower and upper bounds on the scale Sj of a single control
alternative within zone k so that
(12.8)
Part of the equity objective can involve minimizing the sum of the weighted
differences between Sk'in and Sk'"' over all zones k:
minimize L wk(Sk'ax - Sk'in). (12.9)
k
Constraints on water quality can be defined as (a) effluent standards restricting

the waste W;(1 - Pi) released at site i to be no greater than the maximum
allowab Ie quantity:
(12.10)
or as (b) receiving-water quality standards requiring the quality Qj(S) at a site
or reach i within the water body to be no less than some minimum allowable
quality:
Qj(S) ~ Q7'in. (12.11)
These objectives and constraints will be defined in greater detail below.
475
12.4. WATER QUALITY CONTROL ALTERNATIVES
In each of the models to be described, wastewater treatment and/or reduction

through process changes (i.e. the removal of some fraction of the total waste
load prior to discharging the remainder into the receiving water bodies) will be
included as a method of controlling water quality. Models have been developed
for assisting in the design of wastewater treatment facilities and in the design
of treatment processes within these facilities (Loucks, 1967; Lawrence and
McCarty, 1970; Fan et al., 1974; Middleton and Lawrence, 1974, 1976; Van Note
et al., 1975; Grady, 1977). Such models are useful for defining the costs of
wastewater treatment. Since the design and cost of alternative treatment
facilities are fairly well defined and known, the discussion in this section will
begin with an examination of how to dispose of the treated wastewater effluent.
12.4.1. Wastewater Disposal on Land

There are two alternatives for disposing of wastewater effluent from a treatment
plant. These consist of discharging the effluent either into a receiving water body
or on to land for further waste reduction prior to drainage into a water body.
Wastewater disposal on land, or land application as it is commonly called, may
be attractive in certain areas where land is available. Land application permits
further removal of nitrogen, phosphorus, organics, pathogens; and traces of
heavy metals from the wastewater effluent. Of all these constituents, nitrogen
is the most mobile in soils, and hence its removal usually controls the rate of
land application (Haith, 1973; Koenig and Loucks, 1977). Nitrogen in the
drainage waters from well aerated soils exists mostly in the form of nitrate
nitrogen, N0 3 -N.
Outlined in this section is a simple simulation model to assist in evaluating
irrigation land application alternatives that do not generate more than the
predefined maximum allowable N0 3 -N concentrations in the drainage waters.
Effluent disposal on land is commonly accomplished using spray irrigation
methods. Nitrogen is removed from the soil mostly by plant uptake and sub-
sequent harvest or consumption by grazing.
Figure 12.1 illustrates a typical land application system. Of interest are the
storage lagoon capacity V, the land application area A, the irrigation volumes
Q21 in each period t (of length ~t) within a year, and the maximum discharge
rate Q!'
m= max (Q2r/~t),
t
(12.12)
that minimize the total annual cost C( V, A, Q!). This cost can then be compared
with the additional cost of advanced wastewater treatment required to meet
the same NOrN and other constituent effluent standards.
476
Effluent
storage
Wastewater
treatment plant lagoon
1
Waste
source 0t
Receiving
Land application water body
Area A
irrigation area
Subsurface
drainage water
FIGURE 12.1 Components of the land application system of wastewater disposal.
The relationships between the components of the land application system

can be defined by a series of mass balance equations. The wastewater volume
mass balance for the storage lagoon equates the final storage volume St + 1 to
the initial storage volume St plus the difference between the inflow volume Qlt
and the outflow volume Q2t in each period t of the year:
'<ft. (12.13)
If t is the last period within the year, then t + 1 = 1. The lagoon storage
capacity V equals the maximum of all the storage volumes S,:
V = max
, (St) '<ft. (12.14)
The inflow volumes are known, but the outflow volumes Q2t and lagoon
capacity V are unknown decision variables. Equation 12.13 can be modified
to include evaporation losses if desired. Such losses could be based on the
average lagoon volume Sf in each period:
'<ft. (12.15)
Nitrogen in the lagoon may be removed by ammonia volatilization, denitri-
fication, algal uptake, and settling. If a first-order decay process is assumed, the
mass of nitrogen (the product of concentration n2 • t + 1 and volume S,+l) in the
lagoon at the end of each period t equals the initial lagoon nitrogen mass S, n 2t
477
plus the difference between the mass input Qltn 1r and the combined mass out-
flow Qz,(nz,,+ 1 + n zt )j2 and decay k,St(nz. t +1 + nz t)/2, where k t is the
temperature-dependent nitrogen removal rate in period I [T- I]:
St+ InZ., ... 1 StnZt + Qltnlt Qz,(nz.,+ 1 + nz r)j2
mass at end mass at mass in mass in effluent
of period I beginning influent
of period r
k,S,(nz.,+ 1 + nz,)/2. (12.16)

mass decay in
period r
This completes the equations involving the storage lagoon. What remains to
be described is the spray irrigation site.
If it is assumed that the soil moisture content, expressed as a depth [L], is
maintained at field capacity M throughout the year (since otherwise additional
lagoon storage capacity may be required), the water balance for the irrigated
area is defined by equating the irrigation rate Qz,/A with the evapotranspiration
rate £, plus the drainage rate d, less the average precipitation rate P, in period t:
"It. (12.17)
Each of these terms is expressed in units of length. To prevent surface runoff,
the irrigation rate, allowing for precipitation and loss by evaporation, should
not exceed the maximum drainage capacity d, or
"It. (12.18)
Drainage occurs, i.e. d, > 0, when the application rate exceeds that required
to just maintain soil moisture content at field capacity.
Soil nitrogen relationships can be approximated by separately defining mass
balance equations for organic and inorganic nitrogen. Average organic nitrogen
levels in the soil must reach an equilibrium value F [M L - Z] if the waste
disposal system is to be operated at a steady state. This value may be determined
by the native fertility of the soil, or if an objective of land application is to build
up soil productivity, F will be a desired equilibrium value. Soil organic nitrogen
levels will deviate from the equilibrium value during any period owing to
mineralization of some fraction m, of the organic nitrogen and addition of
organic nitrogen XO, [M L - Z] from wastewater irrigation during the period.
If 0, is the deviation [M L - Z] at the beginning of period t then the total organic
nitrogen level at the beginning of period t + 1 is F + 0, + 1 :
F + 0,+ 1 = F + 0, - m,(F + 0,) + XO,
or
(12.19)
478
The nitrogen addition per unit land area is some fraction IX of the nitrogen in
the lagoon effluent Qz/n z. t + 1 + n zr )/2 that is in the organic form, divided by
the total land area A :
XO r = IXQz,(n z. r+ 1 + n zt )/2A "It. (12.20)
The soil inorganic nitrogen content per unit land area, I, + 1 [M L - 2], at the
end of each period t equals the sum of the initial inorganic nitrogen content I"
the fraction mr of organic nitrogen that was mineralized in period t, and the
inorganic nitrogen addition Xlt in the wastewater effluent, less that leached
from the soil by drainage, L t , and that removed from the soil by plant growth,
N" during the period:
"It. (12.21 )
The addition of inorganic nitrogen is that contained in the lagoon effluent
divided by the whole irrigation area A:
"It. (12.22)
In the preceding equations, soil nitrogen losses from ammonia volatilization,
denitrification, and surface runoff were assumed insignificant. The effect of this
assumption will be conservative. The mineralization fraction mt will depend
on the average soil temperature during each period t. Of course, when the soil
is frozen, essentially no mineralization or drainage takes place (d t ~ 0).
The inorganic nitrogen loss from leaching, L, [ML - 2], depends on the
average inorganic nitrogen concentration (1,+ 1 + I,)/2M in the soil water
when leaching or drainage occurs, i.e. when d, > 0:
L t = dt(1t + I t + 1)/2M "It. (12.23)
Plant uptake of inorganic nitrogen, Nt, will depend on the type of cover crop
grown and harvested or consumed as well as on the available inorganic nitrogen
in the soil. If N'('ax is the upper limit of the nitrogen uptake (which will depend
on the type of plant) and if up to 70% of the soil nitrogen is available to the
irrigated crops, then Nt equals the smaller of these two maxima:
Nt = min(0.7[It + m/F + 0,) + Xlt], N;'ax] "It. (12.24)
The final constraint applies to the quality of the drainage water. If n'('ax is the
maximum allowable nitrate nitrogen concentration, then
"It. (12.25)
Model Solution Procedure

A simulation procedure can be defined to solve this model. Input data include
wastewater inflows Qlt to the storage lagoon and their nitrogen concentrations
479
nil' the nitrate nitrogen decay rate k" the hydrological parameters P, and E"
and the soil data parameters M, d, and m"~ Then a particular set of lagoon
volume discharges QZt can be selected whose total equals the sum of the lagoon
inflows Qlt. The maximum of all Qz, determines the capacity Q! of the pumps
and pipes required to connect the lagoon to the irrigation area. The outflows
Qz, together with the known inflows QIt determine the storage capacity V of
the lagoon as found from the simultaneous solutions of eqns. 12.13 and then
solution of eqn. 12.14. (Since each of eqns. 12.13 is linearly dependent on the
others, each S, in the solution to (12.13) can be adjusted by a constant amount
to ensure any desired minimum storage volume for increased detention times
and nitrogen removal.)
When each lagoon storage volume St has been determined, and since the
influent nitrate nitrogen concentrations n l , are known, the simultaneous
solutions of eqns. 12.16 yield the nitrate nitrogen concentrations n z, of the
lagoon effluent. Finally, ( 12.17) and (12.18) can be used to compute the minimum
irrigation area required for the effluent discharges QZt.
Equation 12.18 ensures that for any particular land area A, the drainage d t
cannot exceed the drainage capacity d. Knowing the effluent nitrate nitrogen
concentrations n2' permits the solution of (12.20) and (12.22) for the nitrogen
additions per unit land area, XO, and XI,. It is then possible to solve (12.19).
The simultaneous solutions of (12.19) can provide an estimate of deviations
in soil organic matter per unit area, at, at the beginning of each period t associa-
ted with particular values of A and QZt. Knowledge of each 0, permits the
simultaneous solutions of (12.21), (12.23), and (12.24), if it is assumed that the
plant uptake of nitrogen, Nt, equals 0.7[/, + m,(F + a,) + XI,]. After the
solution of these equations for each inorganic nitrogen mass per unit area, I"
if any Nt exceed N,;,ax, those Nt are set equal to N,;,ax and the equations must
be solved again. This procedure is continued until eqns. 12.24 are satisfied.
It only remains to check that the constraint eqns. 12.25 limiting the nitrate
nitrogen concentration in the drainage water are satisfied. If not, the irrigation
area A mus" be increased if any actual nitrate nitrogen concentration exceeds
n,;,ax. The area can be decreased if all nitrate nitrogen concentrations are less
than n';'ax. Once the minimum area has been found, the total annual cost
C(V, A, Q!) can be determined (Haith et ai., 1977).
This simulation procedure can be repeated for different combinations of
volume discharges QZt in an effort to identify the least-cost design. As described
above, the simulation procedure involves nothing more complex than the
simultaneous solution of several sets of linear equations, some of which may not
even be necessary if, say after a period when the soil is frozen, the initial storage
volume and nitrate nitrogen concentration in the lagoon are known. When the
simulation equations are used for any particular problem, some unit conversion
coefficients may be necessary to maintain the desired units.
480
12.4.2. Regional Treatment and Transport
An alternative to numerous separate treatment facilities is the transport of
partially treated wastewater to one or more regional advanced wastewater
treatment facilities for further removal of constituents prior to discharge into
a water body or on to a land area. This usually results in increased treatment
efficiencies and reliabilities and possibly lower total costs because of economies
of scale in wastewater treatment. Nevertheless, added to the cost of advanced
treatment at any regional facility is the cost of wastewater transport.
Figure 12.2 illustrates a possible situation in which a regional treatment
facility might be considered. Each existing treatment plant needs to be up-
graded to meet new effluent standards and increasing volumes of wastewater
flow. Let us assume that the types and concentrations of wastes in the effluent
of each plant are approximately the same. Therefore, the annual cost Cj(QT)
of increased waste removal capacity at each treatment site i can be defined as
a function of the treated wastewater volume QT at that site.
An alternative to increasing the efficiency of each plant is to transport
wastewater effluent from one or more treatment plants to one or more regional
advanced waste treatment facilities that could be located at various existing
or new treatment plant sites. The wastewater volume QT that is to be treated
at each site i will be the difference between the total wastewater inflow to that
site and the total wastewater outflow. The inflow is the volume Q? collected at
D Wastewater source site

o Existing plant site
,: \ Potential regional plant site
\ .. _ j '
Influent flows
Plant sites
Transported
flows
Waste-receiving stream
FIGURE 12.2 Existing and potential wastewater treatment sites.
481
the site plus the volume F ji transported to the site from adjacent sites j. The
outflow from any site i is the volume Fij transported to adjacent sites j. Hence
at each of sites 1, 2, 3, and 4 shown in Figure 12.2 the total inflow must equal
the total outflow plus the volume treated, if any.
Q~=QT+F12 (12.26)
Q~ + F 12 + F 32 = Q1 + F 23 (12.27)
Q~ + F 23 = Ql + F 32 + F 34 (12.28)
F 34 = QI· (12.29)
Because site i = 1 is not a potential regional wastewater treatment site, but

can be expanded to meet the required effluent standards for the existing influent
volume Q~, the treated flow Qi cannot exceed Q~. This condition is satisfied
by eqn. 12.26.
The annual pipeline construction and pumping costs CiiFij) associated with
the transport of wastewater from site i to adjacent site j will be some function
of the flow F ij . Because of the difference in elevation between those sites, these
costs may depend on the direction of flow. The capacity of each pipe segment
between sites i and i + 1 must equal at least the flow volume F i. i + 1 or F i + 1. i'
If flow in either direction is possible, as between sites 2 and 3, one or both of
these flows will undoubtedly be zero, as will its cost, in the model solution. In
the example illustrated in Figure 12.2, the capacity of the pipeline between
sites i = 1 and 2 must equal F 12' The pipeline capacity between sites 2 and 3
will equal F 23 + F 32 (since one will be zero), and the pipeline capacity between
sites 3 and 4 will equal F 34'
Let us assume a cost minimization objective:
minimize ,i (Ci(QD + ~
,= 1 J
Cij(FiJ). (12.30)
The two components of the objective include annual wastewater treatment,

pipeline construction, and pumping costs. The sum over sites j includes only
those adjacent to site i.
If the cost functions exhibit fixed costs and economies of scale, i.e. decreasing
marginal costs for certain ranges of QT and Fij , then the problem can be formu-
lated as a linear mixed-integer programming model. In this case it is probable
that only one regional plant, if any, will be in the solution, even though the
model constraints (defined by eqns. 12.26-12.29) allow for more than one
regional facility. When the treatment plant cost functions Ci(Qi) are being
derived, the waste concentration in the influent as well as the maximum allow-
able concentrations in the effluent must be known and considered. More
detailed analyses can be found in Roman (1970) and Chi (1972).
482
12.4.3. Multiple Point Source Waste Reduction
to Meet Water Quality Standards
The models presented so far have assumed the existence of quality standards
specifying the maximum allowable constituent concentrations in the waste-
water effluent that can be discharged into a water body. There may also
exist stream quality standards specifying the maximum allowable constituent
concentrations within the water body. These maxima may vary with the location
within the water body, e.g. along a stream or estuary. Numerous models have
been proposed for use in estimating the degree of waste removal at various
point source sites along a water body that will meet both effluent and water
quality standards. The most common of these models apply to the management
of dissolved oxygen concentrations in streams and estuaries.
The oxygen required for the decomposition or assimilation of any particular
quantity of biodegradable water is expressed as biochemical oxygen demand,
BOD. The oxygen demand of a waste can be separated into two components,
the amount required for the assimilation of the carbonaceous waste material,
BODe, and that required for the assimilation of the nitrogenous waste material,
Boon. This division permits a more accurate description of the oxygen demand
at any point in the stream, lake, or estuary than would the total BOD, because
the rates of deoxygenation associated with the two components differ. Another
reason for explicitly considering the nitrogenous component of BOD is that
as the percentage of carbonaceous BOD that is removed increases, say to 80 or
SoJ %, the percentage of the nitrogenous component in the remaining waste-
water effluent increases (Loucks and Jacoby, 1972). As water quality standards
require increasingly high waste removals or treatment efficiencies, the nitro-
genous wastes discharged into natural waters become increasingly important
for the prediction of dissolved oxygen concentrations.
The depletion of dissolved oxygen by the metabolic processes of waste-
consuming organisms, plant respiration, benthic deposits, and the like is offset
by the absorption of oxygen from the atmosphere, from plant photosynthesis,
and possibly from other natural and artificial means. Differential equations
describing the processes of oxygen depletion and replacement were described
in Chapter 6. The solutions of these differential equations, subject to the
appropriate boundary and initial conditions, represent the temporal and
longitudinal distributions of BODe, Boon, and dissolved oxygen concentra-
tion along a water course. If both natural and wastewater flows are constant,
steady state can be assumed. For water quality control alternatives that are
inflexible with respect to time, it is often reasonable to base the scale of these
alternatives on some critical steady state conditions that can occur at specified
locations during certain times of the year.
Let each waste source site along a nondispersive river be denoted by the
index i and each quality-monitoring site in the water body by the index j. The
483
oxygen demand BOD j [M L- 3] at any quality site j resulting from the dis-
charge of a mass of BODf and BODi' per unit time [M T- 1] at all source sites
i can be predicted by using equations such as
(12.31)
where each parameter bij is the BOD mass at sitej resulting from unit mass of
BOD discharged at site i and BOD j is the BOD [M L - 3] at sitej resulting from
all sources other than at sites i. Similarly, the dissolved oxygen deficit D j at any
quality site j can be written as
1 ~ _
D·] Qj L.,(dc.BODc
=~.
i I) l
+ d~.BOD~) + D·
lJ l ]
(12.32)
if the water body has a flow volume Qj at site j. In (12.32), Dj is the dissolved
oxygen deficit [M L - 3] at site j resulting from all BOD sources other than at
sites i. Each parameter dij is the mass of dissolved oxygen at site j resulting
from unit BOD discharge at site i.
In the two equations above, the masses of BODf and BODi' discharged into
the water body at each site i in each period may be unknown decision variables.
If Wf and Wi' [M T- 1] denote the total masses of carbonaceous and nitro-
genous oxygen-demanding waste produced per unit time at site i, and if Pf is
the fraction of the carbonaceous waste removed by treatment, then
BODf = Wf(1 - PD. (12.33)
The amount of nitrogenous BOD removed is some functionJi(Pf) of the carbon-
aceous BOD removal efficiency. Hence,
BODr = Wi(1 - Ji(PD). (12.34)
Equations 12.31-12.34 can be combined to form a mathematical model whose
solution can identify various com binations of wastewater treatment efficiencies
along a river or estuary that will satisfy both effluent and water quality stand-
ards. The planning problem illustrated in Figure 12.3 involves four point
sources of waste and numerous quality-monitoring sites. The problem is to
determine the degree of treatment, Pf, at each site i that satisfies effluent
standards BODr ax at wastewater discharge sites i and stream quality standards
BODTax and DOt n for both BOD and dissolved oxygen concentration at
various sites j. Since there are usually many alternative combinations of waste-
water treatment efficiencies Pf that will meet the standards, the objective of the
analysis will be to identify those that minimize the sum of the costs of waste-
water treatment Cj(Pf) at all sites i:
4
minimize L Cj(Pf). (12.35)
j= 1
484
Waste source ~.
-t
o
Wastewater discharge ~. 11-Pi )
Wastewater treatment ~. Pi
FIGU RE 12.3 Wastewater discharge sites along a river system,
Constraint equations associated with the BOD effluent standards

BODiax [M L - 3] at each site i are
_1 (BOD?
Qi' I
+ BOD~) <
& -
BOD,:"ax
"
(12.36)
where Q7 is the wastewater volume discharged from site i. For any water
quality site j in the river the quality standards for BOD and DO can be
expressed as
(12.37)
DO~ - Qj
~I i
(dtBODi + dijBODi) + Dj ~ out"· (12.38)
The term DO~ in (12.38) is the saturation dissolved oxygen concentration at

site j. An alternative formulation of this model is given elsewhere (Loucks
et al., 1981, ch. 10).
A sufficient number of quality sitesj must be selected to ensure that the maxi-
mum BOD or dissolved oxygen deficit within the entire river section of interest
is not greater than the maximum acceptable. An alternative to the initial
selection of numerous quality sites is to select a few such sites, solve the model,
485
determine where the maximum BOD and DO deficit occur, and, if these con-
centrations are unacceptable, constrain the concentrations at these critical
sites and repeat the procedure. For long river systems this iterative trial-and-
error procedure may be less costly than solving a model that includes a large
number of quality sites. Two or three iterations are usually all that is required.
Equations 12.35-12.38 can be made piecewise linear for solution by linear
programming or linear mixed-integer programming algorithms. The unknown
variables are the waste removal efficiencies Pi of the treatment plants at the
waste source sites i. The coefficients of the model, bij and dij' must reflect the
design flow conditions between sites i andj. They can be determined by methods
discussed in Chapter 6.
Some of the more detailed nonlinear models, discussed in Chapter 6, that
simulate many water quality constituents can also be converted into economic
management models. For example, the application of finite-difference or finite-
element techniques permits the conversion of the QUAL II simulation model
to an optimization or management model. A large number of water quality
constituent concentrations can then be considered. However, to do this, the
relationship between, say, carbonaceous BOD removal efficiencies Pi and the
removal efficiencies for nitrogen, phosphorus, and all other constituents in the
wastewater influent must be defined. The removal efficiencies (and costs) are
not independent, just as the nitrogenous BOD removal efficiency {;(Pi) was
not independent of the carbonaceous BO 0 removal efficiency Pi in the BO 0-
DO model described above.
The interdependence (or joint-product effect) of removal efficiencies may
be defined by dividing the range of carbonaceous BOD removal efficiencies
Pi into several segments h. This permits the consideration of piecewise linear
cost functions,
Ci(Pi) ~ L CihPih' (12.39)

h
as well as piecewise linear efficiency functionsf7(Pf) for any waste constituent k:
f7(Pf) ~ Lf7hP~h' (12.40)

h
In (12.40), coefficient f~h is the fraction of constituent k removed per unit

fraction of carbonaceous BOD removed in segment h at site i. If linear program-
ming is used to solve the resulting model, one must be sure that each Pih is at
its maximum value if Pi.h+ 1 > O. Otherwise, the cost and efficiency functions
will be in error. Mixed-integer programming or separable programming
algorithms that ensure this condition may be necessary. Other ways of modeling
these joint-product effects are also available (Loucks et al., 1981).
In addition to the linear programming models just discussed, the literature
contains numerous examples of water quality management models based on
486
other optimization algorithms. Dynamic programming (Liebman and Lynn,
1966), along with linear programming, is perhaps one of the most commonly
used approaches (for example, see Thomann and Sobel, 1964; Sobel, 1965;
Loucks et at., 1967; Deininger, 1969, 1970; Gunther, 1970; Hass, 1970;
Chia Shun Shih, 1970; Lehmann, 1971,1974; Dorfman et ai" 1972; Milaszewski
and Roman, 1972, 1978; Roman, 1974; Hock, 1978). A very comprehensive
study, using dynamic programming, was reported by Newsome (1972) and
Warn (1978) for the River Trent in England. Others have used nonlinear
programming (Bayer, 1977) and geometric programming (Ecker, 1975), to
name only some of the numerous techniques proposed and applied.
12.4.4. Flow Augmentation

The models discussed above are all based on a critical design flow volume or
cross-sectional area, denoted as Qj or A j at each quality site j. If the discharge
rate for BOD and other wastes from one or more sources is constant, the waste
concentration and dissolved oxygen deficit at various sites may either decrease
or increase, depending on the magnitude of the stream flow.
Increasing the stream flow has four primary effects on water quality. First,
the volume of water increases. If the increased flow is of higher quality than the
base flow, the increased flow dilutes the waste constituent concentrations and
increases the minimum dissolved oxygen concentration. Second, the water
velocity increases and, in turn, usually increases the reaeration rate and length-
ens the distance over which an oxygen-demanding pollutant causes an oxygen
deficit. Third, if base flows are augmented with cooler water, the deoxygenation
rate decreases and the saturation concentration of dissolved oxygen increases,
Conversely, if higher-temperature waters are used for augmentation, the de-
oxygenation rate increases and the saturation dissolved oxygen concentration
decreases, Finally, increased stream flow may increase the BOD addition
through runoff and scour of benthic deposits. All of these factors may well
result in flow augmentation being beneficial at some sites and detrimental at
others (Jaworski et al., 1970; Loucks and Jacoby, 1972).
Let us consider the example of a single waste discharge site upstream from
some quality sites. If the dissolved oxygen deficit and BOD of the flow upstream
of the discharge site are less than those in the discharged wastewater, and if the
increased stream flows do not pick up too much additional BOD from increased
runoff and scour, then the additional dilution of the wastewater flow will
decrease the BOD and, therefore, will increase the minimum dissolved oxygen
concentration downstream from the waste discharge site. Though the minimum
dissolved oxygen concentration is increased, these same conditions may lower
the actual dissolved oxygen concentration at one or more specific quality sites,
as illustrated in Figure 12.4.
487
2 3
I I
9.0
I 8.0
Cl 7.0
E
- 6.0
oo 5.0
4.0
o ~ ~ 00 ~ 1001~1~100
Cl 3.0
.§ 2.0
o
oco 1.00 L----L~_l_====:::E§§:E§§3~~___l...,_
o ~ ~ 00 ~ 1001~1~ 100
Distance (km)
FIGURE 12.4 BOD and dissolved oxygen concentration downstream from a single
waste source under increasing flow conditions: QI < Q2 < Q3.
In Figure 12.4 each pair of functions corresponds to the oxygen "sag"

concentration and the BOD resulting from a single waste source at site i = 0
for three stream flows QI < Q2 < Q3. For the conditions stated above, the
minimum dissolved oxygen concentration increases with the stream flow. At
a site j = 1, an increase in flow results in a decrease in the dissolved oxygen
deficit and BOD, i.e. the water quality increases with flow. The opposite may
occur at quality sites j = 2 and 3, where the quality decreases as the stream flow
increases, at least up to a certain flow volume. Hence there can exist situations,
as illustrated in Figure 12.4, in which the waste removal efficiencies at upstream
treatment facilities designed to meet both dissolved oxygen and BOD stream
488
quality standards at low design stream flows (e.g. the minimum average seven-
day consecutive flow expected once in ten years) are not sufficient to meet
these same quality standards at higher stream flows. This was illustrated very
clearly in an optimization-simulation study of the water quality of the
Saint John River in the United States and Canada (H.G. Acres Ltd., 1971). As
10'
G
2
FIGURE 12.5 Comparison of dissolved oxygen concentrations predicted by analytical

model for low-flow conditions and by a daily simulation model for the Saint John River in
the United States and Canada (after H. G. Acres Ltd.. 1971). Vertical scale, DO (mg 1- 1).
489
seen in Figure 12.5, daily flows typically higher than the seven-day, ten-year low
design flow condition resulted in lower predicted DO from increased runoff
and scour from a large point source of BOD near Grand River.
Since stream flows in excess of the commonly chosen low design flow occur
much more frequently, for a considerable portion of the time the stream quality
at particular downstream sites may be less than what would occur in a critical
low design flow condition. The selection of the critical design flow Qj' therefore,
may become an important consideration in the determination of waste removal
efficiencies at treatment facilities.
The critical stream flow Qj can be augmented by the release of additional
waters from reservoirs or by reduction of water withdrawals. The increased
flow may be a means of improving the critical flow conditions (e.g. by in-
creasing the volume of dilution water, reducing temperatures, or increasing
the reaeration rates), thereby reducing the required treatment capacities.
The net cost of flow augmentation can be defined as the minimum cost
necessary to increase the flows during the period of low flows for the sole
purpose of water quality management. In other words, the cost of augmentation
is that required over and above the costs needed to maximize the net benefits
of flow regulation for uses other than for quality management. If such con-
ditions do not exist at the time flow augmentation is being considered, then the
benefits derived from flow augmentation, apart from improved quality, should
be subtracted from the gross costs of augmentation. This net cost C q (L1Q) of
the flow increase L1Q can then be included in a cost minimization objective
function (Loucks and Jacoby, 1972).
If the temperature and velocity of the flow between any pair of sites i and j
change because of flow augmentation, the parameters and variables that define
the two transfer coefficients bij and dij' and BOO j and V j will change. Hence
bij(Qij), di/Qi), BOOiQ), and Vj(Q) are functions of the stream flow Qij or Qj.
These functions would take the place of the constants bij' dij' BOOj , and Vj ; in
addition, Qj + L1Q would replace the term Qj in (12.37) and (12.38).
12.4.5. Artificial In-Stream Aeration

In-stream aeration is another method of increasing the dissolved oxygen con-
centration in rivers. This is usually accomplished by injecting air into water
through a network of perforated pipes or by rotating devices that cause surface
turbulence, thereby increasing the area over which oxygen transfer can occur.
These methods may be particularly efficient for the temporary improvement
of near-anaerobic conditions, i.e. at sites where the dissolved oxygen deficits
are relatively high, but they require energy and may cause excessive noise.
The oxygen transfer rate due to artificial aeration varies directly with the
dissolved oxygen deficit D, with the water quality and temperature, and with
the flow. If it is assumed that the rate of oxygen transfer per unit of power input
490
is constant over a period t and that the aerators are operating at full capacity,
the rate of oxygen transfer OJ [M T- 1 J at site j is a function of aerator power
capacity PA j [L M T- 1J and dissolved oxygen deficit Dj [M L- 3J at site j:
(12.41)
Aeration devices are often the most cost-effective means of meeting dissolved
oxygen standards that would otherwise be violated during rarely encountered
extreme low-flow, high-temperature conditions (Pinaldi et ai., 1979). The
typical dissolved oxygen control problem consists of determining the least
expensive number of aeration units to be used, their location along the stream,
and their design capacity measured in terms of power.
The cost of aerators operated at full capacity can be expressed as a function
of their capacity. The function can account for the expected down time necessary
for maintenance and repair. If the aerators are not operated at capacity, addi-
tional equations will be needed to define separately the annual capacity and
the operating costs (Ortolano, 1972). Let C/PA) be the cost of each aerator
unit of power capacity PA j at location j along a stream having a specified
extreme design flow of Qj and known upstream oxygen-demanding waste
discharges. Such costs, together with equations for predicting the oxygen
deficit and oxygen addition from artificial aeration, can be used to develop an
optimization model for finding the optimum capacities PA j at various sites j.
If the objective is one of total cost minimization, this can be written as:
minimize L CiPA). (12.42)

j
This objective is to be met subject to the maintenance of stream quality standards

for dissolved oxygen. If a stream is divided into homogeneous reaches and all
reach junctions and potential aerator sites j are numbered successively in the
downstream direction, then the dissolved oxygen deficit Dj just upstream of
site j, prior to any artificial addition of oxygen at that site, is a function of the
BOD mass Bj _ l' the initial deficit mass Dj _ 1 Qj _ 1, and the mass of added
oxygen OJ_I at the next upstream sitej - I:
(12.43)
The parameter dj _ ',j is the dissolved oxygen deficit at site j resulting from unit
BOD mass at site j - 1, and ej-l,j is the dissolved oxygen deficit at site j
resulting from a unit deficit mass at site j - 1.
Dissolved oxygen standards would apply to each Dj prior to any artificial
reaeration. For a given dissolved oxygen saturation concentration DO~ at
site j,
DOS) - D.) >
-
Domin
.I Vj, (12.44)
491
in
where DOT is the minimum allowable dissolved oxygen concentration at site
j. The final set of constraints defines the production of oxygen at each site j as
a function of the power capacity of the aerator (eqn. 12.41).
Of course, there may also be a constraint on the total number of aerators
permitted, but this constraint tends to be redundant, or infeasible, if the sites j
are chosen at locations of maximum deficit and the cost functions exhibit
economies of scale. To estimate the annual cost functions CiPA) some esti-
mates of operating times are needed. These estimates have to be based on an
analysis of the hydrological record of stream flow conditions, some assumptions
regarding the oxygen-demanding waste discharges during critical flow con-
ditions, and the cost of energy.
The control problem defined by (12.42-12.44) can be structured for solution
by discrete dynamic programming in which the stages, the possible locations
of the aerators, are variable and dependent in part on the location and capacity
of aerators installed upstream (Rinaldi et aI., 1979).
12.5. LAKE QUALITY MANAGEMENT
There are a number of alternatives for managing lake water quality. For man-
made lakes these include the selection of the impoundment site if it is not
already fixed, the design of the outlet structure and the release policy, the
control of constituents in the inflow, artificial destratification by such means
as diffused air or mechanical pumping, dredging, and other ways ·of altering
the normal physical, biological, and chemical processes that affect water
quality (Symons, 1969). Simulation models that are able to predict with any
reasonable accuracy the impact on water quality of any of these management
alternatives are indeed just beginning to appear. Optimization models that
incorporate water quality prediction together with various management
alternatives have not yet been developed for lakes and reservoirs, with the
exception of fully mixed impoundments. However, just as multiparameter
water quality simulation models such as QUAL II (Chapter 6) are being
adapted for optimization (management) modeling, so will the multiparameter
simulation models for stratified lakes and reservoirs.
12.5.1. Constant-Volume Well Mixed Lakes

For reasonably well mixed lakes of constant volume, the one-dimensional
models discussed in Chapter 7 (section 7.2) can be developed and include as
unknown decision variables the extent and cost of waste discharge reduction.
If it is assumed that a lake has a constant volume V [V], an outflow rate
Ql [L 3 T- 1], and a nutrient or waste input rate Nt [M T- 1] having a net decay
rate constant K [T- 1], the concentration C Nt [M L -3] of N at the end of
492
period t after an initial concentration C N.r- 1 can be found by integrating the
differential equation that defines the rate of change in CN:
dCN _ N, QtCN KC
d"t-v----v- N'
(12.45)
This yields
C N1 = Qt :~V - (Qr :~V - CN,r-l )ex p [ -~t(~ + K) J (12.46)
The term ~t is a particular interval within period t. Hence if ~t is zero, the

concentration of N is C N, 1- 1, i.e. what it was at the end of period t - 1 or at the
beginning of period t. If the inflow rates Nt and Qt are constant for a long time,
i.e. as ~t approaches infinity, the concentration CN approaches the equilibrium
concentration, Nt/(Qr + KV).
To illustrate the use of (12.46) in a lake quality management model, let us
consider the discharge N a [M T- 1 ] of a mass of nutrient in each period t at
various sites j along a lake. The rate of discharge is assumed constant over each
period (having a time interval ~t equal to 1), but the mass of nutrient discharged
can be altered by removing a fraction Xi at a cost C;(XJ Thus the mass of
nutrient discharged is Na(l - Xi) at each site i. Equation 12.46 can now be
written for each period within a year:
"It, (12.47)
where each known et can be expressed by
"It (12.48)
and the variables Xi are bounded:
X ;nin :5: Xi :5: x:"ax Vi. (12.49)
The variables C Nt and Xi are unknown. If the rates Nt and Qt repeat themselves
over each cycle of periods t, the initial concentration C ND can be set equal to
the concentration C NT at the end of the final period T to yield a steady state
solution of concentrations CN1 for all periods t = 1,2, ... , T This eliminates the
need to specify arbitrary or observed initial concentrations C ND , unless of
course a dynamic model is desired for a predefined number of periods t. In
addition, the model could be made more realistic by the inclusion of changes
in the decay rates K t in each period that could result from changes in water
temperature.
493
A simple economic objective might be to minimize the total cost:
minimize L ClX;), (12.50)
i
subject to (12.47) and (12.49) and to constraints ensuring that the concentrations
C Nt do not exceed some acceptable level C'iJ"x:
C Nr :-::;; CN'ax \:ft. (12.51)
Once again, it should be emphasized that these models are extremely
simplified for most lake quality management problems. Nevertheless, with some
judgment as to the appropriate values of Qt and V, they can sometimes be used as
a preliminary means of establishing average concentrations of nutrients or
pollutants in well mixed lakes and of identifying the mix of water quality con-
trol alternatives that should be further analyzed using more detailed simulation
models (Russell, 1975; Canale, 1976; Pavoni, 1977).
12.6. PREDICTING ALGAL BLOOM POTENTIAL
One of the major concerns of lake management in nutrient-rich areas is the

potential for algal blooms. Large algal populations or blooms can seriously
depress dissolved oxygen concentrations at night when algal respiration con-
tinues but oxygen production has stopped. Particularly disastrous can be the
sudden collapse of a bloom, resulting in a tremendous oxygen demand for de-
composition of the dead algal cells. In either case the depression of dissolved
oxygen levels can kill fish and other aquatic life and produce bad odors and
appearance. In addition, large algal populations can clog filters in water supply
systems and often produce substances toxic to fish, shellfish, and animals that
drink the water.
A set of optimization models can be used to predict the potential for algal
blooms and thus to help estimate the effectiveness of alternative management
strategies (diversion of nutrient-bearing waste and thermal discharges or lake
destratification). This is done by determining the largest probable bloom that
could occur under the conditions placed on algal growth by temperature and
by nutrient and light availability.
Such models were developed and applied by Bigelow et at. (1977) to saltwater
lakes and estuaries in the Netherlands. At any given temperature, the maximum
potential algal bloom is limited by nutrient and light availability. The nutrient
constraints in the model limit the total concentration of critical nutrients to
that actually available. Let x j be the unknown concentration of living algae in
species j [M L - 3], aij the mass of nutrient i contained in unit mass of species
j [M M - 1], Yi the concentration [M L - 3] of nutrient i that is temporarily held
in nonliving matter (dead algae and decomposable organic matter) and that
494
will become available for algal growth, and Wi the unknown concentration
[M L - 3] of dissolved nutrient i immediately available for algal growth. The
total concentration N j [M L - 3] of nutrient i potentially available for algal
growth is expressed as:
Ni = Wi + Yi + L aijxj , (12.52)
j
Upon the death of an algal cell, the nutrients in the cell enter the pool (y;) of
nutrients bound up in nonliving material. Upon decay ofthe nonliving material,
the nutrients become available for further algal growth. The rate of change of
the nutrient concentration in the nonliving material, dY;/dt, is the difference
between the rate at which algae die and the rate at which nutrient i is released
from nonliving material into the water column and joins the available nutrient
(w;). Let dj be the death rate [T- 1] of speciesj and U i the rate of release of nutrient
i from nonliving material (the mineralization rate); then
(12.53)
If algal blooms build up rather slowly compared with the rate at which nutrients
are released from nonliving material, the concentration of nutrients, Yi' will be
reasonably close to equilibrium (dy/dt ~ 0) at the peak of the algal bloom. Then,
a reasonable estimate of Yi is the equilibrium concentration y{, obtained by
setting (12.53) to zero:
(12.54)
From eqns. 12.52-12.54 it follows that the equilibrium concentration N{ of

nutrient i is
This relationship specifies how the total biomass in all algal species is con-
strained by the available nutrients if (12.54) applies.
In warm summer months, algal blooms can build up very rapidly such that
the concentration of nonliving material lags behind the buildup of the algal
populations. If, at the peak of the bloom, the amount of nutrient i in nonliving
matter is approximately kiy{ for some empirically determined k j , 0 :os; k i :os; 1,
then
(12.56)
495
Of course, k i = 0 corresponding to Yi =0 would result in the restriction that
the total mass of nutrient i in all living algae must be less than N;, the total
nutrient available. While this yields the physically possible upper bound on
algal biomass, it may not be sufficiently restrictive to allow determination of
the maximum probable or realistically achievable algal population density.
The latter is the more important quantity from the point of view of lake
management.
The constraints on light availability affect the capacity of an algal species
to grow and compete with other algal species. The amount of light that algae
receive depends on the incident solar radiation throughout the day, Io{t), the
fraction of the radiation transmitted across the air-water interface, (J, and the
light extinction rate '1 [L- 1] in the water column. In general the light intensity
at depth z is given by
(12.57)
An algae population can grow if its individuals receive light of an adequate
intensity throughout the day as they move throughout the epilimnion or upper
layers of a lake or estuary so that production exceeds respiration and mortality.
If PlIz{t)] is the rate of production of algal species j at light intensity I it), for
species j to participate in a bloom the average production rate of algal biomass
throughout the 24 h must be no less than the minimum required for growth, p'J'in:
-
1 f24 - 1 fz=ax Pl{Jlo{t)e-~Z]dzdt ~ prt", (12.58)
24 0 Zmax 0
where Zmax is the maximum mixing depth. For any temperature, the production
rate PlI] peaks at some characteristic value for which the algal species is best
adapted. Thus for each species j there will be a feasible range,
(12.59)
within which speciesj can thrive and outside of which the average light intensity
is either too low or too high.
The extinction coefficient '1 depends on the natural color or turbidity of the
water, resulting in a background extinction rate '10' and on the densities of algae
(self-shading) and nonliving matter in the water column. If the contribution
to the total extinction coefficient of a unit concentration of algal speciesj is "Ii'
then the extinction coefficient resulting from background and living algae is:
(12.60)
To this must be added the contribution from nonliving algae. This is primarily
an effect of the chlorophyll remaining in dead algae because reflection is not
nearly as important as the absorption of light. If the decay rate [T- 1] of the
light-absorbing capacity of dead algae is v, then the differential equation
496
describing the behavior of the contribution 1Jd of dead algae to the total extinc-
tion coefficient 1J is
d1Jd
dt
= Ij
1Jdx - V1Jd'
J J J
(12.61)
Again, if the bloom grows slowly so that the concentration of chlorophyll in

dead algae and hence its contribution 1Jd to the extinction coefficient stay in
relative equilibrium, then d1Jd/dt ~ 0, and the equilibrium extinction coefficient
contribution from dead algae is approximately
1J* '" 1J·d·

, _ J_J X.
d ......... ~ V JO (12.62)
J
However, if blooms grow rapidly, 1Jd may lag behind the value given in (12.62)
and the total extinction coefficient at the peak of the bloom is best given by:
1J = 1Jo +I 1JjXj + k1J:

j
kd.)
= 1Jo + ~ 1Jj ( I + --; xj (12.63)
for some empirically determined k between and 1. °

For incorporation into the linear programming model, the acceptable
extinction coefficient interval (1Jjln, 1Jjax) for each species is used to determine
a set of subintervals (1J;,in, lJ,:ax). Typically a number of subintervals s will be
contained within any particular species interval. Each subinterval between
1J,:in and 1J,:ax will include acceptable extinction coefficient values for a set Ss
of algal species j.
A set of linear programming models for finding the maximum probable or
reasonable biomass or chlorophyll concentration of an algal bloom can be
developed using these nutrient and light availability constraints. To find the
maximum potential biomass or chlorophyll concentration one would maximize
for each s either biomass:
(12.64)
or chlorophyll:
cs = I CjXj' (12.65)
jeS.
where C j is the chlorophyll content [M M- 1 ] per unit mass of algal species j.

This maximization would be subject to nutrient constraints,
(12.66)
497
to the light limitations,
(12.67)
and to nonnegativity,
Solving the model for each set Ss with the corresponding extinction coefficient
interval gives a set of potential concentrations of biomass {B S } or chlorophyll
{CS}. The largest of these,
B max = max B S (12.68)
and
e max = , e,
s
max (12.69)
provide estimates of the potential or maximum probable values of these

parameters.
The peak biomass of a bloom could be estimated with a dynamic ecosystem
simulation model. This requires considerably more information than needed
for this simple optimization model, as well as the ability to identify the environ-
mental conditions that would give rise to the maximum possible bloom.
This optimization model makes no attempt to describe the time dynamics
of the aquatic system. Rather it uses the available nutrient constraints and
information about light requirements and self-shading characteristics of algal
species to predict the maximum biomass or chlorophyll concentration that
might reasonably be achieved during a particular week or month. Using a
management period of this length should ensure that sufficient time is available
for potential blooms to materialize, as long as algal populations are already
present in reasonable densities in natural environments.
There may be no feasible solution to the optimization model for some s
because an algal bloom may be unable to develop at a particular temperature
and in corresponding light conditions. In winter and early spring, when the
water is cold, there may be no feasible solution for any value of s. Feasibility
could also be affected by small changes in some of the parameters. From the
dual variables of the nutrient and light availability constraints, one can obtain
some indication of the impact on bloom potential caused by a change in the
available nutrient concentrations N; and algal extinction coefficient limits.
As formulated, the model considers only the effect of a bloom in terms of
total biomass or chlorophyll. The oxygen demand from decomposition of dead
algal cells, which would occur over a short time were the maximum bloom to
498
collapse suddenly and all the algae die, can also be determined. The effect of
zooplankton has been omitted from the model but could be incorporated in an
approximate way. The model is relatively simple and serves as an approximate
description of a very complex system. Its potential value lies in its simplicity,
its ability to estimate answers to some important management questions, and
its relatively modest data requirements compared with more complex and
realistic dynamic algal models.
12.7. CONCLUDING REMARKS
It would be ideal if one could say that water quality models such as those
discussed in this chapter could be used directly to identify optimum, or at
least improved, solutions to water quality management problems. Perhaps
they can in situations where there exist sufficient data and sufficient modeling
expertise to develop, calibrate, and verify models and their solutions and
where the objectives of those responsible for water quality planning and
management are clearly defined. Where ideal conditions do not exist, these
models can serve another purpose.
Water quality management models, such as those outlined in this chapter,
are most commonly used for developing an understanding of the relative
impacts on water quality of alternative management practices and for deter-
mining the significance or importance of having more accurate or more detailed
data. These insights can guide the development of effective plans and decisions,
if not actually identify the best plan or decision.
Water quality modeling, if done well, can give an understanding of why
some management alternatives are better than others for a particular river
basin. Modeling can provide estimates of how the river system will respond, at
least in a relative sense, to different waste discharges. In addition, models can
be used to help identify preferred management plans, based on various manage-
ment objectives and assumptions concerning future resource costs, technology,
and social and legal requirements.
In acknowledging the role that water quality models can and should play in
the planning process, one must recognize the inherent limitations of models as
representatives of any real problem. The input data, including management
objectives and assumptions concerning the physical, biological, and chemical
processes in the water body, may be controversial or uncertain. Of course, the
input affects the output. While the input data and model may be the best avail-
able, one's knowledge about the actual water body and about how future
events may alter its behavior will always be limited. In addition, since public
water quality objectives change, water quality models must be viewed as
flexible tools that can adapt to changing circumstances as they are perceived
and to changing data as they become available.
499
REFERENCES
Acres, H. G., Ltd. (1971). Water quality management methodology and its application to the Saint
John River. Appendix G. H. G. Acres Ltd., Niaqara Falls, Ontario Report (unpublished).
Bigelow, J. H., J. G. Bolten, and J. C. DeHaven (1977). Protecting an estuary from f1oods-A policy
analysis of the Oosterschelde. Vol. IV, Assessment of algal blooms, a potential ecological dis-
turbance. Rand Corporation, Santa Monica. C A Report R-2121/4-NETH.
Bayer, M. B. (1977). A modeling method for evaluating water quality policies in nonserial river
systems. Water Resources Bulletin 13(6): 1141-1151.
Canale, R. P. (ed.) (1976). Modelinq Biochemical Processes in Aquatic Ecosystems (Ann Arbor, MI:
Ann Arbor Science).
Chi, T (I 97::!). Water conveyance models. Models for Managing Regional Water Quality, eds. R.
Dorfman, H. D. Jacoby, and H. A. Thomas, Jr. (Cambridge, MA: Harvard University Press),
ch.8.
Chia Shun Shih (1970). System optimization for river basin water quality management. Journal of
Water Pollution Control Federation 42(10): 1792-1804.
Deininger, R. A. (1969). Uber die Planung eines wirtschaftlich optimalen Systems von Kliiranlagen.
Gas- und Wasserfach 110(52): 1443-1445.
Deininger, R. A. (1970). Les calculatrices et la recherche operationelle pour la planification des
installations d'epuration. Centre Be~qe d'Etude et de Documentation des Eaux Rapport 314.
Dorfman, R., H. D. Jacoby, and H. A. Thomas, Jr. (eds.) (1972). Modelsfor Manaqing Regional
Water Quality (Cambridge, MA: Harvard University Press).
Ecker, J. G. (1975). A geometric programming model for optimal allocation of stream dissolved
oxygen. Management Science 21 :658-668.
Fan, L. T, M. T Kuo, and L. E. Erickson (1974). Effect of suspended wastes on system design.
Proceedinqs of American Society ofCivil Engineers, Journal ofEnvironmental Enqineerinq Division
IOO(EE6): 1231-1247.
Grady, C. P. L., Jr. (1977). Simplified optimization of activated sludge process. Proceedings of
American Society of Cillil Engineers, Journal of Environmental Engineering Division
103(EE3): 413-429.
Gunther, W. (1970). Optimierungsmodelle zur Planung und Betreibung von Anlagensystemen zur
Abwasserbehandlung in Flussgebieten. Wasserwirtschaft- Wassertechnik 20(3): 81-84.
Haith, D. A. (1973). Optimal control of nitrogen losses from land disposal areas. Proceedinqs of
American Society of Civil Engineers, Journal of Enl'ironmental Engineering Division
99(EE6): 923-937.
Haith, D. A., A. Koenig, and D. P. Loucks (1977). Preliminary design of wastewater land applica-
tion systems. Journal oj' Water Pollution Control Federation 49( 12): 2371-2379.
Hass, J. E. (1970). Optimal taxing for the abatement of water pollution. Water Resources Research
6(2): 353-365.
Hock, B. (1978). Water quality modeling as a tool for decision making in Hungary. Modeling the
Water Quality of the Hydroloqical Cycle. Proceedinqs of Baden Symposium, September. Interna-
tional Association of H ydroloqical Sciences Publication 125, pp. 312-321.
Jaworski, N. A., W. J. Weeber. and R. A. Deininger (1970). Optimal reservoir releases for water
quality control. ProceedinqsofAmerican Society ofCil'i1 Engineers, Journal ofSanitary Enqineerinq
Division 96(SA3): 727-742.
Koenig, A., and D. P. Loucks (1977). A management model for wastewater disposal on land. Pro-
ceedings of American Society of Ch·iI Enqineers. Journal of Enl'ironmental Engineerinq Dil'ision
103(EE2): 181-196.
Lawrence, A. W., and P. L. McCarty (1970). Unified basis for biological treatment design and
operation. Proceedin.qs of American Society of Ci"iI Engineers. Journal of Sanitary Engineerinq
Didsion 96(SA3):757-778.
Lehmann, H. (1971). Ein Beitrag zur Schaffung optimaler Beschaffenheitsverhiiltnisse in Fliessge-
wiissern unter Beriicksichtigung der Abwasserbehandlung, der Wasseraufbereitung und sonstiger
Nutzungen. Wasserwirtschajr- Wassertechnik 21 (12): 404-41 O.
500
Lehmann, H. (1974). Planung der Wasserbeschaffenheit in Fliessgewiissern bei Beriicksichtigung
okonomischer Kriterien. Wasserwirtscha/I- Wassertechnik 24(8): 262-267.
Liebman, 1. c., and W. R. Lynn (1966). The optimal allocation of stream dissolved oxygen. Water
Resources Research 2(3): 581-591.
Loucks, D. P. (1967). Risk evaluation in sewage treatment plant design. Proceedings ot American
Society otCivil En9ineers, Journal otSanitary En9ineering Division 93(SAI):25-39.
Loucks, D. P., and H. D. Jacoby (1972). Flow regulation for water quality management. Models
for Managin9 Regional Water Quality, eds. R. Dorfman, H. D. Jacoby, and H. A. Thomas, Jr.
(Cambridge, MA: Harvard University Press), ch. 9.
Loucks, D. P., C. S. Revelle, and W. R. Lynn (1967). Linear programming models for water
pollution control. Management Science 14(4):B-166-B-181.
Loucks, D. P., J. R. Stedinger, and D. A. Haith (1981). Water Resource Systems Planning and
Analysis (Englewood Cliffs, NJ: Prentice-Hall).
Middleton. A. c., and A. W. Lawrence (1974). Cost optimization of activated sludge systems.
Biotechnology and Bioengineering 16: 807-826.
Middleton, A. c., and A. W. Lawrence (1976). Least cost design of activated sludge systems.
Journal 0/ Water Pollution Control Federation 48(5) :889-905.
Milaszewski, R., and M. Roman (1972). Optymalizacja stopnia oczyszczania sciekow dla systemu
oczyszcralni. Gospodarka Wodna 36(1):23-29.
Milaszewski, R., and M. Roman (1978). Some physical and economic aspects of optimization of
the degree of wastewater and water treatment. Modeling the Water Quality 0/ the Hydrological
Cycle. Proceedi'lgs of Baden Symposium, September. Internarional Associution of Hydroloyical
Sciences Publication 125, pp. 373-382.
Newsome, D. H. (1972). The Trent River model-An aid to management. Modeling (it IVater
Resource Systems vol. II, ed. A. K. Biswas (Montreal: Harvest House), pp. 490-509.
Ortolano, L. (1972). Artificial aeration as a substitute for wastewater treatment. Models for
Mana9ing Re9ional Water Quality, eds. R. Dorfman, H. D. Jacoby, and H. A. Thomas, Jr.
(Cambridge, MA: Harvard University Press), ch. 7.
Pavoni, J. L. (ed.) (1977). Handbookfor Water Quality Managemelll Planning (New York, NY:
Van Nostrand Reinhold).
Rinaldi, S., R. Soncini-Sessa, H. Stehfest, and H. Tamura (1979). Modeling and Control 0/ River
Quality (New York, NY: McGraw-Hili).
Roman, M. (1970). Podstawy stosowania grupowych oczyszcralni sciek6w. Nowa Technika w
Inzynierii Sanirarnej, Seria Wodociegi i Kanalizacja (Warsaw: Arkady).
Roman, M. (1974). Kompleksowa optymalizacja stopnia oczyszczania sciek6w i wody. Prace
Naukowa Politechniki Warszawskeij, Seria Budo Wniuwo No. 33 (Warsaw: Warsaw Polytechnic),
pp.47-60.
Russell, C. S. (ed.) (1975). Ecological Modeling in a Resource Management Framework. Resources
for the Future Working Paper QE-I (Baltimore, MD: Johns Hopkins University Press).
Sobel, M. J. (1965). Water quality improvement programming problems. Water Resources Research
1(4) :477-487.
Symons, J. M. (1969). Water quality behavior in reservoirs. US Department 0/ Health, Education
and Weltare, Cincinnati, OH, Public Health Sen'ice Publication 1930.
Thomann, R. V., and M. J. Sobel (1964). Estuarine water quality management and forecasting.
Proceedings ot American Society ot Ch·i/ Engineers, Journal ot Sanirary Engineering Dit·ision
90(SA5):9-36.
Van Note, R. H., P. V. Herbert, R. M. Patel, C. Chupek, and L. Feldman (1975). A guide to the
selection of cost-effective wastewater treatment systems. US Environmental Protection Agency,
Washington, DC Report EPA 430/9-75-002.
Warn, A. E. (1978). The Trent mathematical model. Mathematical Models in Water Pollution
Control, ed. A. James (New York, NY: Wiley), ch. 7.
501
mass of nutrient i in unit mass of algal species j

land area
BOD mass at site j resulting from unit mass of BOD discharged at
site i
B' maximum potential biomass concentration
BOo j BO0 at site j resulting from all sources other than at sites i
ClSJ agency cost of waste reduction
q(SJ private cost of waste reduction for group s
C' maximum potential chlorophyll concentration
Cj chlorophyll content per unit mass of algal species j
dij mass of dissolved oxygen at site j resulting from unit BOD discharge
at site i
death rate of algal species j
dissolved oxygen deficit at site j resulting from unit BOD mass at
site j - 1
drainage rate
dissolved oxygen deficit at site j
dissolved oxygen deficit at site j resulting from all other BOD sources
other than at sites i
saturation dissolved oxygen concentration at site j
dissolved oxygen deficit at site j resulting from unit deficit mass
at site j - 1
evapotranspiration rate
fraction of constituent k removed per unit fraction of carbonaceous
BOD removed in segment h at site i
F equilibrium level of organic nitrogen in soil
Fij wastewater volume transferred from site i to adjacent sites j
It initial inorganic nitrogen content
I it) light intensity at water depth z
kt temperature-dependent nitrogen removal rate in period t
K nutrient or waste decay rate constant
Lt inorganic nitrogen leached by drainage
mt mineralized fraction of organic nitrogen
M field capacity
n concentration of nitrogen
Nt inorganic nitrogen removed by plant growth during period t;
nutrient or waste input or discharge rate
OJ rate of oxygen transfer at site j
°t
PlSJ
deviation of level of organic nitrogen at start of period t
fraction of waste reduced at site i
Pj rate of production of algal species j
502
Pt average precipitation rate
PA j aerator power capacity at site j
Qi wastewater volume discharged from site i
Qj flow volume at site j
iJi quality of water at site i
Qi maximum discharge rate
QZt irrigation volume in each period t (of length Ilt) within a year
s subinterval of algal species interval
Si scale of waste reduction measures at site i
Ss set of algal species j
St storage volume at start of period t
Ti tax rate per unit of waste discharge
Ui rate of release of nutrient i from nonliving material (mineralization
rate)
v storage lagoon capacity
W weighting constant
Wi concentration of dissolved nutrient i immediately available for algal
growth
w; constant quantity of waste available at site i
Wi, wr total masses of carbonaceous and nitrogenous oxygen-demanding
waste produced per unit time at site i
Xj concentration ofliving algae in species j
X fraction of nutrient or waste
XIt inorganic nitrogen addition
XO, organic nitrogen addition
(yf)Yi (equilibrium) concentration of nutrient i temporarily held in non-
living matter
set of sites in zone k of a region
fraction of nitrogen in lagoon effluent
fraction of radiation transmitted across air-water interface
light extinction rate
decay rate of light-absorbing capacity of dead algae.
Future Directions
13 G. T. Orlob
13.1. ACHIEVEMENTS IN MODELING WATER QUALITY
There is little doubt that mathematical models have reached a level of acceptance
in the scientific and engineering communities that makes them viable in research
and problem solving. Over roughly two decades, from the early 1960s, modeling
of water quality has moved forward-albeit, with difficulty at times-from mere
intellectual exercise of the mathematically inclined to creative synthesis of
scientific principles and empirical observations aimed at quantification of
natural aquatic phenomena. One has only to skim the most recent journals of
science and engineering to appreciate that" model," .. modeling," "systems,"
"systems analysis," etc. are solidly embedded in the lexicon oftoday's researchers
and practitioners. The hardware to facilitate the use of models has advanced so
rapidly, especially in the design of inexpensive high-capability minicomputers,
that the machine seems less of a restriction than ever. Moreover, to an important
degree models of hydraulic, hydrological, water quality, and ecological be-
havior of the earth's water resources have become useful tools in problem
solving, not merely ends in themselves. It is the development of such models that
we have tried to chronicle here.
If we consider that the" modern era" in modeling of water quality com-
menced with the now classic Streeter-Phelps formulation of the DO sag, then
the chronology of our accomplishments since 1960 has been roughly as follows.
1960-65
The Streeter-Phelps approach was extended by discretization of the stream
system to allow spatial variability in load patterns, geometry, and kinetic
coefficients. Temperature was introduced as a state variable in one-dimensional
stream and primitive reservoir models that considered heat exchanges at the
air-water interface. Temporal variability was accommodated by providing
503
504
independently derived solutions of the hydrodynamic equations for one-
dimensional systems. The advection-diffusion equation was adopted as the
basic description for water quality transport processes. The first simple models,
essentially one-dimensional and steady state, began their use in water quality
management.
1965-70
Discretized one-dimensional models of stream and reservoir systems were
extended to include additional sources and sinks, e.g. nitrogenous oxygen
demands and photosynthesis. One-dimensional networks were used to represent
two-dimensional vertically mixed systems. Thermal stratification was simulated
for deep reservoirs and lakes that could be represented as one-dimensional
continua. Nutrient budgets and primary productivity of lakes were simulated
for simple, fully mixed systems. Water quality models began to figure promi-
nently in assessments of the impacts of pollution.
1970-75
Multilayered two-dimensional models of wind-driven circulation were de-
veloped, providing a basis for description of water quality in large lakes.
Water-quality-ecological interactions were incorporated in multisegment
models. The kinetics of primary productivity and higher trophic levels were
developed for model application. The Michaelis- Menten- Monod limiting
nutrient concepts became prominent in modeling of eutrophying aquatic
systems. Finite-element models were developed for two-dimensional systems,
adding flexibility to the already well accepted finite-difference techniques.
Sensitivity testing assumed greater importance in assessing model reliability.
Increased dimensionality of models heightened concern for adequacy of data
for calibration and validation.
1975-80
During the last five years or so, increasing attention has been given to improve-
ments in model reliability and assessment of capability. Reassessments were
made of trade-offs between model detail and the required data base. Automatic
calibration and parameter estimation techniques were devised. Improvements
were made in the description of important hydrodynamic processes essential
to the characterization of water quality, for example, the effects of internal
mixing and surface wind shear. The coupling of hydrodynamics and water
quality, as in the case of thermal stratification, was given increased attention
with the development of improved two- and three-dimensional lake models.
Complexities of the aquatic ecosystem were the subject of an increasing number
of biologically oriented modelers, leading to improved descriptions of the
ecological relationships and the process of eutrophication. A general prolifera-
tion of models gave rise to concern for lack of documentation of useful models
and practical mechanisms for technology exchange.
505
These two decades, which account for virtually all of the history of water
quality modeling as we know it, have not been without some negative aspects.
Where these developed, they were attributed generally to overzealousness on
the part of practitioners of the modeling" art" and to lack of effective com-
munication between the modeler and the potential user of his product. There was
a general tendency in the mid-1960s to oversell the potentials of modeling,
especially in the field of water quality management. This was encouraged by the
rapid advances in the sixties of systems technology and computer sciences. Also,
there was a natural tendency in academia torise to the intellectual challenges of
modeling, leading to the overproduction of models, as evidenced in the litera-
ture. However, despite these general difficulties, water quality modeling is
apparently a viable part of our technology, still offering challenges to the
modeling enthusiast.
13.2. AREAS FOR IMPROVEMENT
At the end of each of the preceding chapters the author has endeavored to sum-
marize the major accomplishments within the frame of his topic, at the same time
indicating the direction toward needed improvement. Certain improvements
refer specifically to a particular relationship or model and need not be repeated
here, but there are some areas of concern that seem to apply to modeling
generally, at least to the modeling of water quality. To bring our book to a
conclusion on an optimistic note, it seems worth while to identify a few areas
where the efforts of future modelers may be productive.
13.2.1. Model Reliability

Among the most perplexing of questions is "How good is the model?" A clear
and compelling need exists to answer this question in explicit and quantitative
terms. However, this must be done in the context of the intended use of the
model and the person or persons who will interpret the results.
In the context of water quality management, it appears generally to be the
case that most models are far too complex to serve as useful aids in the decision
process. Yet, simplification of the model (and concomitant reduction in the
data base required) may be at the sacrifice of credibility to the decision maker.
One solution may be the utilization of hierarchical models, which at one
extreme of detail preserve some essential attributes of the water body, while at
the other assure rapid, inexpensive, and more informative exposition of the
consequences of alternatives.
Quantitative measures of reliability include statistical measures of the
processes of calibration, verification, and validation. Each of these processes is
506
much in need of formalization in the modeling process, perhaps even of stan-
dardization to some degree. It is apparent that much of the difficulty with model
credibility stems from lack of accepted criteria for ascertaining reliability.
Development of such criteria should be beneficial to the modeling art.
13.2.2. Data and Data Management

An almost universal complaint by modelers and users of model output is aimed
directly at the quantity and quality of data. Generally, the data are not sufficient
and often they are of the wrong kind. There is a need for improved methods
for structuring the data collection program that supports modeling. The data
management activity should be developed in parallel with the model if possible
or, if an existing model is to be employed, it should be developed with the aid
of the model.
Design of the data collection scheme depends on the question of model
reliability, discussed above. Of course, if data collection is more costly than
model development (usually the case) then it would seem prudent, if not
imperative, that model reliability assessment and data system design be carried
out simultaneously. Research aimed at improving the relationship of these two
crucial modeling activities should have a high priority.
13.2.3. Models and Submodels

Water quality models described in the previous chapters are largely descriptive
and deterministic in structure, of the distributed-parameter type. In addition,
many of these are dynamic, an added complexity that exacerbates the problems
of parameter estimation, reliability estimation, calibration, etc. While such
models often have considerable merit as management tools or aids in research,
they may prove demanding of data, costly to operate, and even conducive, be-
cause of their complexity, to increasing human error. There is a need to obviate
these weaknesses, perhaps with supporting models or "submodels" of simpler
design.
Possibilities for simplification include:
(1) reducing spatial segmentation;

(2) changing from dynamic to steady state;
(3) reducing the number of parameters, i.e. "lumped" instead of dis-
tributed parameters;
(4) introducing inputs stochastically instead of deterministically.
Development of hierarchical models, compatible submodels, or complementary

models of different conceptual form (e.g. lumped versus distributed, or
stochastic versus deterministic) should increase the utility of modeling generally.
507
13.2.4. First Principles Versus Empiricism
The nature of modeling suggests that there will always be some dependence on
empiricism. However, to the extent practical, empirical representation should
be supplanted by first principles (see Chapter 3). The argument here is for a
fundamental understanding of the water body whenever possible rather than
succumbing to the temptation to mount increasingly demanding programs to
collect data in support of empirical modeling.
A case in point is the development of an improved hydrodynamic model to
describe wind-induced mixing in the hypolimnion of a stratified lake. This
comparatively recent innovation in lake modeling promises to improve
substantially the description of internal mixing processes that previously were
treated almost entirely as analogous to molecular diffusion.
13.2.5. Hydrodynamics in Relation to Water Quality

There has been a tendency, especially among the more biologically oriented,
to concentrate on rigorous development of ecological relationships in water
quality modeling, occasionally at the expense of comparable rigor in character-
izing the hydromechanical behavior of the water body. While there exist cases
where hydrodynamics is of small consequence, it is more often the case that
the major mechanisms of water quality and ecological change are closely linked
to movement of the water and internal mixing processes. A successful modeling
effort will depend on sound description of these motions, notwithstanding an
elegant development of the pertinent water-quality-ecological relationships.
Examples of the importance of giving appropriate weight to both aspects of
water quality modeling are seen in efforts to characterize eutrophication of
large lakes, e.g. the Great Lakes, where wind-driven circulations determine
the fate of nutrients discharged in shoreline areas.
13.2.6. One, Two, and Three Dimensions

Capability to model the water quality changes in relatively small, deep
impoundments that are strongly stratified seems to be well developed, although
some improvements are possible yet in describing the processes of internal
mixing over the annual cycle. Also, two-dimensional vertically mixed systems,
i.e. shallow lakes and reservoirs without strong stratification, are accommodated
fairly well by existing models. However, long, narrow reservoirs or fjord-type
lakes and large, deep lakes are still prime candidates for modeling effort. A
first essential step is a rigorous model of two-dimensional stratified flow in
which the internal mixing is represented in terms of the traditional parameters
describing the flow.
508
Attempts to design such a model using the finite-element method appear
promising, primarily because of the flexibility the method provides for dealing
with irregular physical boundaries. Incorporation in such a model of an
improved description of turbulent mixing, such as the Launder K -( representa-
tion, is a necessary direction of future research.
13.2.7. Kinetics and Stoichiometry

The long tradition in water quality analysis of the universality of certain
kinetic coefficients and stoichiometric constants has encouraged modelers to
extend these notions into the realm of ecological modeling. While there are
some indications that this confidence is justified for very simple or lumped-
parameter models, there remains considerable reservation (especially among
the aquatic biologist modelers). Additional research and modeling experience
are required to assure the reasonable extension of existing models to new
situations.
13.2.8. Documentation
Lack of complete documentation is probably the greatest obstacle to user
acceptance, despite the existence of a sizable number of proven water quality
models. Since model development, itself, is more or less an "art," there is
considerable variability in the quality of in-program comment and user
manuals. There is a need, as yet not addressed, to standardize or to provide
some measure of consistency in this important step in model development.
13.2.9. Register for Software

An initial motivation for the development of this book was the apparent lack
of an effective mechanism for technology transfer in the field of mathematical
modeling of water resource systems. Although there has been substantial pro-
gress to improve this situation by just such organizations as IlASA, as this
volume itself attests, there is still a need to formalize an institutional structure
for technology exchange. This structure is envisioned as an international
register of software (perhaps with counterparts at a national level), which
would document proven models and, if feasible, serve as a clearing house for
dissemination of programs and essential documentation. This is a role already
assumed by IlASA on a limited basis. In the future it is expected that the scope
of this effort will be expanded and that other projects, such as that represented
by this volume, will continue to enhance technology transfer in the field of
mathematical modeling.
509
13.3. CONCLUSION
During the course of preparing this book, a product of the collective effort of
an international team of dedicated experts in the field of water quality modeling,
it became painfully apparent that we had little hope of capturing that elusive
"state of the art." Modeling is itself dynamic, often changing so rapidly that
a manuscript completed one week can be out of date the next. Moreover, we
continued to be plagued by the lag between creation of the model and its
appearance in the technical literature, most often a year or more. Finally, we
were continually reminded that the art we were seeking was not really found
in the technical press after all, but appeared most often in reports of very
limited distribution. If we were lucky we found such a document in the files
of one of our group, if not we had to depend on the assistance of our friends
and advisors, whose help was generously given.
Considering all these difficulties in gathering sufficient useful information
to justify this book, we decided that we could not in good conscience consider
our effort as actually characterizing the state of the art. However, we hope we
have come fairly close to the target within our own frame of reference. This
has called for identification and discussion of a selected group from among the
many models we have reviewed. Our criteria for inclusion, admittedly arbitrary
on our part, embraced such subjective determinations as "practical," "well
documented," "widely used," "transferable," etc. We recognize that some impor-
tant models may have been excluded, either inadvertently or by our subjective
judgment. Nevertheless, we hope that our sampling of the literature and
available reports has been sufficiently comprehensive to have stimulated the
reader to examine our treatise carefully, to point out to us its deficiencies
(which we acknowledge before the fact), and, above all, to contribute to
further advancing the "state of the art."
Index
Absorption, 80 Bioaccumulation, 402, 418

Activity coefficients, 78 Biochemical oxygen demand,
Adsorption, 80, 397,412 carbonaceous, 57,178,183,189,190,
of copper, 419 204,483
Advection (see also Convection), 44, 69, decay rates, 219-220, 427, 428
234, 237, 238 estimation of rate constant, 451-454
Advection-diffusion equation, 57-58, 254, interaction with dissolved oxygen,
259 178-179,426-430,451-453,455
Aeration, in-stream, 469, 489 nitrogenous, 431 if
Affinity 78, 98, 103 Body clearance, 403
Algae: see Phytoplankton Bottom stress (see also Shear. stress), 290
Alternative-direction implicit method, 293 Boussinesq approximation, 60, 276, 295
Ammonium ion, 118
Assimilation efficiency, 359 Cadmium, 403, 405, 411, 412
Atmospheric radiation, calculating, Calibration (see also Parameter
172-173 estimation), 11-14,23-35, 167,425,
433-458
Baca-Arnett model, 233, 245, 259, 363, Cannibalism, 418
387-389 Carrying capacity, 87, 105
solution method, 259 Causal models, 116
Bacteria, 118 Cell yield, 402
biomass, 205 Characteristic trajectories, 431
growth,205 Chen-Orlob model, 117, 233, 309, 339,
Nitrobacter sp., 135,209,214 354,381-382
Nitrosomonas sp., 135,209,214 Chlorophyll a, 139,431,495,496
role in degradation, 205, 412 Chromium, 410, 419
Balance equations, general statement of. Circulation in lakes,
45-51 classification of, 278
Benthic oxygen demand, 186,200 equations of state, 59-63, 215,
Berkel River, Netherlands, 425, 430-433 276-277,301,307
Biased estimates, 435 wind-driven, models of, 288-309
Bifurcation, 101, 110 CLEAN,251
51l
512
CLEANER, 4, 119,338,342,347 Detritus, 132
Competition, 109 Diffusion, 47, 50, 56, 75, 92, 97, 234, 237
Complexation, 401, 412 air-water interface, 397
of copper, 419 eddy, 284
Complexity, of models, 44,110 effective, 240-242, 263
Components, defined, models, 282
biological, 46, 55, 59, 71 moleCUlar, 240,399
chemical, 45, 55, 71 turbulent, 241,281
physical, 45 vertical coefficient of, 286
Computational representation, 13-14, Dispersion, 48, 154
15-17 Dissipation, 43, 67, 97
Concentration factor, of toxic substances, Dissolution, 83
396,413,415 Dissolved oxygen: see Oxygen
Conceptualization, defined, 13 14, 15-17 Diversity, 92
Conservation, DOSAG, 3, 188
of energy, 44, 155,237,301 Dynamic programming model, 491
of mass, 55-57, 236, 255, 353 DYRESM, 234, 242,261-263,266
of momentum, 59-60, 215, 276-277,
301,307
Continuity equation (see also Mass Ecological models (see also Lake models),
balance), 55, 215, 301 364
Continuously stirred tank reactor (CSTR) Baca-Arnett,387-389
idealization, 18,451 Bierman, 357, 373-378
Convection (see also Advection), 47, 56, Canale et ai" 358, 378-381
59, 217 Chen-Orlob, 117,233,309,339,254
Cooling impoundments, 316 381-382
case studies of, 322-329 CLEANER, 4, 338, 342, 347
classification, 317, 319 Di Toro et al., 340
stratified, 320, 325 EPAECO, 248, 354
vertically mixed, 321 Jorgensen, 346, 360, 382-387
Coriolis force, 60,278,301 Kremer and Nixon, 341
parameter, 62 LAKECO, 5, 233, 256-258, 339
Correlation analysis, 447 Lehman et al., 340, 343
Cost, of water quality management, MS CLEANER, 5, 338, 348
effectiveness, 468 Nyholm, 344
minimization, 491 Patten et al., 354
Courant-Friedrichs criterion, 292 Scavia et aI., 345
Critical deficit, 4, 180 Straskraba, 340, 342
Cybernetic models, 93 WQRRS, 233, 243
Ecosystem,
described, 43, 72, 74, 109
Decomposer, 71, 77, 97 of a lake, 117,258,356-358
Decomposition, rates of, 131,219-220, of a river, 200
427--428,451--454 Eddy diffusivity, 50, 56, 60, 69, 282, 284
Deep Reservoir Model, 249, 261 Eddy viscosity, 283
conceptual representation of, 235 Effluent storage, 469
Delaware Estuary Comprehensive Study Egestion,
(DECS),3 process of, 70, 77, 88
Denitrification, process of, 133 rate of, 73
Deoxygenation, 81, 178 Ekman models, 296
Desorption, 80 Energy, 66
Destratification, 234 balance, 44, 66, 155,237
513
dissipation, 262 of PCB, 405
heat: see Heat exchange at air-water of toxicants, 403, 418
interface rate of, 73, 413
internal, 67, 102 Experimental design, 21, 24, 28-30
mechanical, 67 normal operating conditions, 28
of chemical reactions, 77 sampling frequency, 29
radiative, 67,171-175 sampling programs, 29, 433
total, 66, 261-263 extended Kalman filter (EKF) algorithm,
Enthalpy, 68 426,438--446
Entrainment rate, 262 case studies using, 447--457
Entropy, 91 conceptual picture of, 440
balance, 44, 96 gain matrix in, 439, 441,444--446
density, 95 statement of, 444
flow density, 96 Extinction coefficient, 69, 237, 495
principle, 43
production, 43, 78, 82, 97
Feedback, 119
specific, 103
Feeding rate, 71, 87
Enzymatic catalysis, 102
Filter feeders, 129,358,416
EPAECO, 248, 354
Finite-difference method, 20, 51-53
Epilimnion, 63, 117
alternative-direction implicit, 293
Equilibrium, 82-100
explicit, solution by, 244-245
approach,82
implicit, solution by, 243-244
constant, 79
Finite-element method, 53-55, 233, 245,
local, 83, 95
temperature, 155,288 259
Equity objective, 474 solution by, 259
Error function, Fish,
effect on water quality, 138
definition of, 32-34, 435-436
growth, 354,418
gradient of surface of, 430, 437--438
Flow augmentation, 469
Error variance-covariance,
Flows,
in extended Kalman filter, 444,
generalized, 97
453,458
types of, 59, 61, 276
of measurement errors, 436, 453
Food chain, 73, 77
of parameter estimates, 426, 453
Forces, 60
of state estimates, 453
generalized, 43, 97
Errors,
Free enthalpy, 78,96, 103
estimation, 12,37-38
in measurements, 12, 16,434,441,458 Freeze-thaw cycle, 242
in modeling, 12 Frequency,
in parameter estimates, 29 of measurement, 371
in state estimates, 441 response analysis, 20, 447
linearization, 12 Freundlich isotherm, 399
lumping, 12 Friction, 60, 67, 97
residual (model response), 36,445 Froude number, 65
Euler number, 65 densimetric, 228-229, 295, 320
Eutrophication model, 337, 357
Evaporative heat loss, calculation of, Galerki method, 260-261
172-174 Gibbs free energy approach, 78, 83
Evolution theorem, 99 Gibbs potential, 82, 94
excretion, 70, 88 Grazing, 71, 87, 97
by zooplankton, 117, 132 of algae by zooplankton, 71, 127,
of cadmium, 405 358-359
514
Group Method of Data Handling Lake models,
(GMDH) algorithm, 447, 449 Baca-Arnett, 245, 259, 363, 387-389
Growth, Deep Reservoir Model, 249, 261
intrinsic rate for animals, 404 DYRESM, 234, 242, 261-263, 266
laws of, 106-109 EPAECO, 248,354
maximum specific rate, 81, 119 LAKECO (Chen-Orlob), 5,233,
rate limitations, 105, 127,202,357 256-258,339,381-382
MIT temperature, 230, 243, 328
Half-saturation constant, 81, 119, 402 MS CLEANER, 3,127
Heat, Lake quality management, 491
balance, 66, 237, 238, 277 Lakes, case studies of,
conduction equation, 68, 92 Anna (USA), 328
dilfusivity,69 Balaton (Hungary), 363
exchange at air-water interface, 69, Castle (USA), 362
170-175,286 Erie (USA), 227, 294,311
supply, 67 George (USA), 363
transport, 69 Glums0 (Denmark), 367, 368, 369
Heavy metals, Lyngby (Denmark), 426, 459, 460
solubility products, 404, 406-408 Michigan, Green Bay (USA), 5, 312
uptake by algae, 411, 418-419 Norman (USA), 327
Henry's Law, 398 Ontario (USA), 297, 302, 303, 306, 310
Hydrodynamic models, 59-60 Gvre Heimdalsvatn (Norway), 366
Ekman-type, 269 Paijanne (Finland), 248-249
one-dimensional stream, 215 Shagawa (USA), 310
three-dimensional, 63, 300 Vanern (Sweden), 5, 300, 303
two-dimensional stratified flow, 298, Washington (USA), 264, 265, 352
300 Zurich (Switzerland), 4, 227
two-layer, 252, 294, 310 Lambert-Beer law, 69
Hydrolysis, 72, 75, 97, 401, 412 Laterally averaged models, 298
Hydrostatic approximation, 276, 290, 295 Least-squares estimation, 32, 435,
Hypolimnion, 63, 117 447-450
Ligands, in sea water, 409
Identifiability, 30, 38, 430 Light,
Inhibition, 81 attenuation, 123-127,495
allosteric, 139, 207 diurnal variation, 123
competitive, 139, 207 extinction coefficient, 69, 431, 432
In situ experiments, 145 limitations, 123, 127
Instrumental variables, method of, 436, Linear Kalman filtering (LKF) algorithm,
447-449 439,442-443
Interactions, trophic, 44, 72, 77 Linear programming model, 496
Ionization, 401, 412 Linearization, 65, 427
Irradiance, 69, 84, 106 in extended Kalman filter, 443
Link-node models, 216, 312
Jacobian matrix, 430, 437 Lotka-Volterra equations: see Predator-
Jordan River, Utah, 426, 453-457 prey models,
Lyapunov function, 100
Kalman filter: see Extended Kalman
filter Mass balance, 55, 56, 236, 255, 353
Mass fraction, 55
Lagrangian solution technique, 189 Mass transfer, 230
LAKECO (see also Chen-Orlob model), coefficient of a gas, 398
5,233,256-257,381-382 reaeration, 178, 221-223, 254, 451-454
515
two-film model, 397~399 MS CLEANER, 5,127,338,348
Maximum likelihood estimation, 436, Multiobjective planning, 472
447--450 Multiplicity of solutions, 83, 110
Measured input disturbances,
definition of, 16
Measured output (response) variables, Navier-Stokes equations, 59, 276
definition of, 16 Newton-Raphson algorithm, 438
Mercury, 410,411 Nitrification, process of, 133
methyl, 417, 418 Nitrogen,
Metabolic rate, relationship to growth, allochthonous, 131
404 ammonia, 117,118, 135,213,455,457
Michaelis-Menten kinetics, 80, 84, dissolved organic, 117, 133, 214,
119, 127 455--457
bacterial growth, 206 fixation, 133, 358
light limitation, 123, 126, 127 nitrate, 117, 135,213,455
phytoplankton growth, 201, 202 nitrite, 117, 135,213
Migration, of trace metals, 403 particulate organic, 135,214
Mixing: see Diffusion, Dispersion phytoplankton, 135, 213, 455--457
Model discrimination, 31, 447 zooplankton, 135,214
Model evaluation, 425--461 Nitrogen cycle, 118
Model structure identification, 24, 30-32, in streams, 196,212
33-35,36,425,433,446--447,453, Michaelis-Menten models, 209
457 nitrogen models, 199
Model types, Nonlinearity, 46, 80, 94, 99, 110
black box (input-output), 20-21, 32, No-slip condition, 296
433,447 Numerical stability, 244, 292, 293
compartment, 309-313, 338 Nutrient budget models, 344, 353
deterministic, 19,46, 109-111 Castle Lake, 362
distributed-parameter, 18 Imboden, 349
dynamic, 20 Larsen et al., 350
internally descriptive, 20, 32,433,447 Snodgrass-O'Melia, 349
linear, 18 Vollenweider, 344,349
link-node, 312 Nutrients, 75, 80, 85, 105
lumped-parameter, 18,434 balance in rivers, 203
moving cell, 431 degradation, 206
multilayer, 294 intracellular, 120
multisegment, 252 limitations, 105, 127, 357
nonlinear, 18 release from sediment, 135, 136
steady state, 20
stochastic, 19
vertically averaged, 289 Objective function, 471, 472
Modeling procedure, 11-39,425 Ohio River Commission, 2
diagram of, 13 Onsager relations (see also parameter
Molar fraction, 89, 104 estimation), 99
Momentum, conversation of, 59~60, 215, Optimization models, 468
276-277,301,307 Oxidation,
Monod kinetics, 81, 85, 119, 355, 402 of ammonia, 203, 254
Monte Carlo simulation, 38 of chemical compounds, 400, 411
Mortality, of detritus, 254
of phytoplankton, 138, 202, 432 of nitrite, 203, 254
of zooplankton, 131 Oxygen,
rate, 59, 70, 75, 89 balance formulation, 254
516
dissolved, 57, 81,179 Radioactivity, transport of, 419-420
prediction of, 178, 181, 189, 190, 194, Random search algorithm, 438
195,431 Random walk parameter, 442, 452, 457
primary production of, 83-86. 105 Raphael model, 157,231
sag equation, 178, 179 Reaction rates, 44, 56, 71, 73, 75, 77-83,
transfer formulation, 221-223 116
Parameter estimation, 11,22,24, 32-35, Reaeration (see also Mass transfer)
425,433--435 80, 178, 221-223, 254, 451--454
applications of, 447-450 Recursive parameter estimation: see
off-line scheme, 32-33,433,436--438 parameter estimation
recursive (on-line), 32-34,433,436, Regional wastewater management,
438--446,451--458 transport, 480
Parameters, definition of, 17 treatment, 480
Particulate matter, 71. 76 Reservoirs, case studies of,
Partition coefficient, 400 Fontana (USA), 245-247
PCB, 403, 405 Hungry Horse (USA), 230, 247
Phosphorus, 116, 120,203,310 Lake Hartwell (USA), 267, 268
balance, 57, 361, 362 Lake Koocanusa (USA and Canada),
model validation, 459 242
models of, 344, 349, 350, 362 Lake Roosevelt (USA), 230, 299
particulate, 459 Lower Granite (USA), 300
solu ble reactive, 431 Ross Lake (USA), 248, 250
Photolysis, 400, 411 Sutton (USA), 300
Phytoplankton, Wellington (Australia), 263, 264, 266
blue-green, 358 Respiration,
growth rate, 200, 354, 402, 431,432, process of, 72, 75, 89, 97, 202,206
451,453 rate of, 59, 106,402
mineralization, 131 Reynolds number, 65, 295
mortality, 138, 202, 432 Reynolds stresses, 276
prediction of blooms, 493 Rhine River model, 205, 210-211
respiration, 130, 254 Richardson number, 66, 173,262,284,
settling, sinking, 130-132 285, 308
Photosynthesis, 83,120-123,254,220 Rigid lid approximations, 281, 296
Photosynthetic oxygenation, 186,254 River flow, 49,57,69,215
Pond number, 319-321 Rivers, case studies of,
Potential, of biological components, 95, Cam (England), 426, 451--453
104 Jordan (USA), 426, 453--457
Prandtl number, 284 Moselle (France), 164, 169, 170
Predator-prey models, 71-73,119,129, Rhine (West Germany), 208
233, 358 San Antonio (USA), 194, 195
Preference factor (coefficient), 87, 359 Rossby number, 65
Primary productivity, 44, 70-72, 75-76,
83-86,101-108 Saint John River model, 488
Principle of superposition (in linear Saturation, 80
models), 18 Second law of thermodynamics, 43, 91,
Principles of modeling, general, 42 97,98
Protozoa, role in degradation, 205, 208 Sediment,
migration of heavy metals, 403
QUAL, I, 188, 191 sorption by. 399
QUAL II, 4, 8, 119,200 Selection of model type, 13-14, 17-22
Quasilinearization method, 442 Self-purification factor, 180
Quasistatic approximation, 62 Sensible heat, 69, 170, 174
517
Sensitivity, Stratified reservoir, conceptual
analysis, II, 13, 30, 426--433 representation, 235
a posteriori, 24, 37-38,426 Stream models,
a priori, 21-22, 425--426 Camp-Dobbins, 5, 184
case study of, 430--433 DOSAG, 3, 188
coefficient, 22, 437 Frankel-Hansen, 181
definition of, 426--427 QUAL 1,188
functions, 426, 428--432 QUAL 11,200
of nonlinear systems, 429 River Cam, 426
Settling, of phytoplankton, 130-132 River Rhine, 210
Shear stress, Saint John River, 488
interfacial, 295 Streeter-Phelps,
solid boundary, 290 equation, 80-81,178
water surface, 285 model, 3,179-188,426-320,488
Similarity, 65 Strouhal number, 65
Simulation-optimization, 491 Structure, dissipative, 109
Small perturbations: see Linearization Submodels, of water quality models, 118
Solar radiation, 171,237,238 System,
Solubility, identification, 12,21
of gases, 399 open, 90
of metals, 404, 406--408 self-organizing, 109
Sorption, of toxics, 399
Space differencing, 292 Temperature effects, on process rates,
Spline interpolation, 53 136-137,201-202
Stability, 110 Temperature models,
analysis, 22 equilibrium, 155,288
constant of complexes, 410 one-dimensional, lake, 230, 243
criteria, 99-10 I one-dimensional, nonsteady state,
numerical, 292 stream, 157-159
State transition matrix, 445 one-dimensional, steady state,
State variables, stream, 155, 156, 159
definition of, 16 two-dimensional, lake, 230, 312
estimation of, 435, 439--446, 457 Tennessee Valley Authority, 161, 171,231
States, Thermocline, 69, 117, 247
metastable, 83 Thermodynamic relations, 96
stable, 99, 10 I Thermodynamics, of irreversible
standard, 79 processes, 80
steady, 99, 101 Threshold concentration, 87, 127
Steepest descent algorithm, 437--438 Time differencing, 293
Stochastic processes, 93 Toxic substances, 395
Stoichiometric, Act of 1976,395
models, 70 in sediments, 409
ratios, 78, 139-144 modeling principles, 396,414
Storage, intracellular, 86 Toxicity, to fish, 117, 138
of energy, 103 Trial-and-error model fitting, 431
Stratifica ti on, Trophic,
effects of, 285 interactions, 87
of cooling ponds, 325 interrelationships, 72
process of, 228-230, 283 Turbulence,
Stratified flow models, exchange coefficients, 307
finite-difference, 298, 300 kinetic energy of, 262
finite-element, 298, 300 models. 281-282
518
shear, 281 Chen-Orlob, 117, 233, 309, 339, 354,
transport of heat, 277 381-382
transport of momentum, 277 EPAECO, 248, 354
LAKECO, 233, 256-258
Uncertainty (see also errors), 26-27, 29, MS CLEANER, 5, 127, 338, 348
37-38,441 Thomann et al., 353, 264
Unmeasured input disturbances, 434, 441, two- and three-dimensional, 310
458 WORRS, 233, 243, 256-258
definition of, 16 Water-quality--ecological relationships,
Unsteady flow, 117,258,356-358
in lakes, 288-309 Water quality management, 468
in streams, 215-217 objectives, 475
Uptake, 70, 75, 80, 86, 97 quality standards, 470
Weighted least-squares estimation, 436
Validation, 11-14,35,36-38,426,459-460 WESTEX, 263, 268
Verification, of models, 24, 35-36,425, White noise sequence, 35, 436, 442, 458
458-459 Wind mixing, 262, 285
Volatilization, of toxic substances from Wind stress, 287, 295
water, 397 WQRRS (see also LAKECO), 233, 243
Waste disposal,
on land, 475 Zero-dimensional models, 252, 344, 349
to receiving water, 484 Zooplankton,
Wastewater control, biomass, 129
flow augmentation, 486 excretion, 131
multiple sources, 482 grazing, 127-129,359
storage lagoon, 476 grazing preference, 359
Water-quality--ecologicaJ models, growth, 71, 202,359
Baca-Arnett, 233, 245, 259, 363, mortality, 131
387-389 respiration, 131
(contents continued)
6 Stream Quality Modeling

7 One-dimensional Models for Simulation of Water
Quality in Lakes and Reservoirs
G. T.Orlob
8 Two- and Three-dimensional Mathematical Models
for Lakes and Reservoirs
M. Watanabe, D. R. F. Harleman, and
O. F. Vasiliev
9 Ecological Modeling of Lakes
S. E. J~rgensen
10 Modeling the Distribution and Effect of Toxic
Substances in Rivers and Lakes
S. E. J¢rgensen
11 Sensitivity Analysis, Calibration, and Validation
M. B. Beck
12 Models for Management Applications
D. P. Loucks
13 Future Directions
G. T. Orlob
THE WILEY IIASA INTERNATIONAL SERIES ON APPLIED SYSTEMS
ANALYSIS
Conftieting Objectives in Decisions

Edited by David E. Bell, University ofCambridge; Ralph L. Keeney, Woodward-Clyde
Consultants, San Francisco; and Howard Raiffa, Harvard University
Material Accountability: Theory, Verification, and Applications
Rudolf Avenhaus, Nuclear Centre Karlsruhe and University of Mannheim
Adaptive Environmental Assessment and Management
Edited by C. S. Holling, Institute ofAnimal Resource Ecology, University of British
Columbia
Orxanization (or Forecasting and Planning: Experience in the Soviet Union and the
United States
Edited by William R. Dill, New York University; and G. Kh. Popov, Moscow
StIlte University
Management of Energy/Environment Systems: Methods and Case Studies
Edited by Wesley K. Foell, University of Wisconsin-Madison
Systems Analysis by Multilevel Methods: With Applications to Economics and
Management
Yvo Diricltx, Twente University of Technology; and L. Peter Jennergren, Odense
University
Connectivity, Complexity, and Catastrophe in Large-Scale Systems
John L. Casti, New York University
Pitfalls of Analysis .
Edited by G. Majone and E. S. Quade, International Institute for Applied Systems
Analysis
Control and Coordination in Hierarchical Systems
W. Findeisen, Institute of Automatic Control, Technical University, Warsaw;
F. N. Bailey, University of Minnesota; M. Brdys, K. Malinowski, P. Tatjewski, and
A. Wozniak, Institute of Automatic Control, Technical Institute, Warsaw
Computer Controlled Urban Transportation
Edited by H. Strobel, Institute for Transportation and Communication' Friedrich List',
Dresden
National Perspectives on Management of Energy/Environment Systems
Edited by Wesley K. Foell, University of Wisconsin- Madison and Loretta A. Hervey,
International Institute for Applied Systems Analysis, Laxenburg
Mathematical Modeling of Water Quality: Streams, Lakes, and Reservoirs
Edited by Gerald T. Orlob, University of California, Davis
Forthcoming
The Electronic Oracle
D. Meadows and J. Robinson (both of USA)
JOHN WILEY & SONS

Chichester· New York· Brisbane· Toronto· Singapore
A Wiley-Interscience Publication
ISBN 0 471 10031 5

Mathematical Modeling of Water Quality PDF

Uploaded by

Copyright:

Available Formats

Mathematical Modeling of Water Quality PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mathematical Modeling of Water Quality PDF

Uploaded by

Copyright:

Available Formats

What topics are covered in the book?

What topics are covered in the book?

What mathematical techniques are discussed for modeling water quality?

What mathematical techniques are discussed for modeling water quality?

INTERNATIONAL SERIES ON

APPLIED SYSTEMS ANALYSIS

International Institute for

This book is the first to deal comprehensively with the

CONFLICTING OBJECTIVES IN DECISIONS

3 ADAPTIVE ENVIRONMENTAL ASSESSMENT AND

4 ORGANIZATION FOR FORECASTING AND

5 MANAGEMENT OF ENERGY/ENVIRONMENT SYSTEMS:

6 SYSTEMS ANALYSIS BY MULTILEVEL METHODS

7 CONNECTIVITY, COMPLEXITY, AND

10 COMPUTER CONTROLLED URBAN TRANSPORTATION

11 MANAGEMENT OF ENERGY/ENVIRONMENT SYSTEMS:

12 MATHEMATICAL MODELING OF WATER QUALITY:

JOHN WILEY & SONS

All rights reserved.

No part of this book may be reproduced by any means. or transmitted,

Lihrary ofConRress CataloRing in Puhlication Data:

Mathematical modeling of water quality: streams,

British Lih"ary Cataloguing in Puhlh'ation Data:

Mathematical modeling of water quality: streams,

Filmsel al Composition House Limited. Salisbury. Wiltshire.

Printed and hound in Greal Brilain al Ihe Pitman Press, Bath.

GERALD T. ORLOB S. E. J0RGENSEN

M. BRUCE BECK DANIEL P. LOUCKS

M. J. GROMIEC PETER MAUERSBERGER

DONALD R. F. HARLEMAN OLEG F. VASILIEV

J. JACQUET MASATAKA WATANABE

A capability to represent quantitatively the responses of the aquatic environ-

Notation for Book xix

1.1 Water quality in perspective I

2 A Procedure for Modeling II

3 General Principles in Deterministic Water Quality Modeling 42

4 Modeling the Ecological Processes 116

5 Simulation of the Thermal Regime of Rivers 150

6 Stream Quality Modeling 176

7 One-dimensional Models for Simulation of Water Quality in Lakes and

8 Two- and Three-dimensional Mathematical Models for Lakes and

8.1 Introduction 274

8.3.3 Ekman-type models, 296

9 Ecological Modeling of Lakes 337

10 Modeling the Distribution and Effect of Toxic Substances in Rivers and

10.2.5 Photolysis, 400

II Sensitivity Analysis, Calibration, and Validation 425

12 Models for Management Applications 468

13.1 Achievements in modeling water quality 503

A cross-sectional area; algal biomass

IX lip, specific volume

Chapters 2-4 and 6-12 have individual lists of extra notation.

1.1. WATER QUALITY IN PERSPECTIVE

Water as a natural resource has been a primary concern of man throughout

1.2. THE BEGINNINGS OF MODELING

Stimulated by the need to control pollution of the major sources of freshwater

t Or .. science:" if the reader prefers.

1.3. GROWTH OF WATER QUALITY MODELING

1.4. PROLIFERATION OF MODELS AND THE NEED FOR

d(V:{t» = G(F, t) - S(t), (3.6)

a(~F) + Vh ' (HvhF + S) = HG, (3.7)

a(AF) + ~ (QF + AS) = AG, (3.14)