George Parker Lamptey PDF

QUANTITATIVE ANALYSIS OF APPROXIMATE MODELS
TO THE SAINT VENANT EQUATIONS
By
GEORGE PARKER-LAMPTEY, B.Sc.
THESIS
Presented to the Department of Mathematics,
Kwame Nkrumah University of Science and Technology
in Partial Fulfillment of the Requirements for the Degree of
MASTER OF PHILOSOPHY
College of Science
JUNE 2012
Declaration
I hereby declare that this submission is my own work towards the Master of Philosophy
(M.Phil.) and that, to the best of my knowledge, it contains no material previously pub-
lished by another person nor material which has been accepted for the award of any other
degree of the University, except where due acknowledgement has been made in the text.
George Parker-Lamptey(PG5072510) ..................... ..................
Student Signature Date
Certified by:
Prof. S. N. Odai ..................... ..................
Supervisor Signature Date
Prof. I. K. Dontwi ..................... ..................
Supervisor Signature Date
Mr. K. K. Darkwah ..................... ..................
Head of Department Signature Date
i
Dedication
to my mother, grandmother and two siblings
Naa Odeibaa II, Ayaba Sarpei Cofie, Naa Lamiley and Naa Lamiorkor
with love
Also to the special individuals who have been a blessing to me
Auntie Tsotsoo, Auntie Efua, Naa Koshie, Attoh Dedei, Attoh Korkoi, Yaa Asantewaa,
Naa Adumoyoobi, Whelma and Nana Ama,
ii
Acknowledgements
I am most grateful to Jesus, my Lord and Saviour for his mercies and guidance throughout
graduate school and for His Grace by which I can still hope.
I am extremely grateful to my two supervisors, Prof. I. K. Dontwi and Prof. S. N. Odai
who have both trained and supported me not only academically but also financially and
spiritually. I cannot repay in any form what these men have invested in me.
To Dr. Peter Amoako-Yirenkyi, Mr. K. Darkwa, Mr. J. A. Ackora-Prah and all
the lectures in the Department of Mathematics, I thank you for your encouragement and
support. God bless you.
iii
Abstract
Approximate models such as (nonlinear Burgers’ Equation Model and nonlinear Kinematic
Wave Model) to the normalized Saint Venant Equations are very important models that
can be used in place of the Saint Venant Equations. In addition to these approximate
models, a third order approximate model is proposed and presented in this research. The
constants of the previous models are preserved in the third order approximate model and
the magnitude of the estimated constant of the third derivative is Fo2 /3(1−4/9Fo2 ). The four
point Preissmann is used to discretize the normalized Saint Venant Equations and the three
approximate models (including the third order model). The algorithms are programmed
and the models are simulated. The positive and negative surges are both experimentally
considered in the application of a dam break problem. The quantitative results of the
approximate models are compared to the normalized Saint Venant equations. The third
order derivative is found to equal the Saint Venant equation at lower level than the BEM
iv
Table of Contents
Page
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
Chapter
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Justification of Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.6 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 Mathematical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Channel Routing Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 The Saint Venant Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 12
v
2.4 Kinematic Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 The Burgers’ Equation Model . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.6 The Proposed Third Order Approximate Model . . . . . . . . . . . . . . . 17
2.6.1 Estimation of Model Constants . . . . . . . . . . . . . . . . . . . . 17
3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1 Numerical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.1 Efficiency of a Numerical Scheme . . . . . . . . . . . . . . . . . . . 24
3.1.2 Explicit and Implicit Formulations . . . . . . . . . . . . . . . . . . 26
3.2 Finite Difference Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.1 Application of Finite Difference Schemes to PDEs . . . . . . . . . . 30
3.3 Numerical Scheme for the Mathematical Models . . . . . . . . . . . . . . . 31
3.3.1 Preissmann Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Solution to Systems of Algebraic Equations . . . . . . . . . . . . . . . . . . 34
3.4.1 Linear Systems of Equations . . . . . . . . . . . . . . . . . . . . . . 34
3.4.2 Solution to Nonlinear System of Equations . . . . . . . . . . . . . . 35
3.5 Numerical Solution of Models . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.5.1 Burgers’ Equation Model . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5.2 The Proposed Equation . . . . . . . . . . . . . . . . . . . . . . . . 40
3.5.3 Kinematic Wave Equation . . . . . . . . . . . . . . . . . . . . . . . 42
3.5.4 Saint Venant Equations . . . . . . . . . . . . . . . . . . . . . . . . 44
3.6 Efficiency Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
vi
3.6.0.1 Coefficient of Determination r2 . . . . . . . . . . . . . . . 50
3.6.0.2 Nash-Sutcliffe efficiency E . . . . . . . . . . . . . . . . . . 50
3.6.0.3 Index of Agreement d . . . . . . . . . . . . . . . . . . . . 50
3.6.0.4 Modified forms of E and d . . . . . . . . . . . . . . . . . . 50
3.6.0.5 Relative efficiency criteria Erel and drel . . . . . . . . . . . 50
4 Simulation Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.1 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2 Effect of Weighted Parameter . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2.1 Runtime and Limits of Flow Depth differences . . . . . . . . . . . . 61
4.2.2 Limits of the Flow Depth Differences . . . . . . . . . . . . . . . . . 62
4.3 Quantitative Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.1 The SVE and the BEM . . . . . . . . . . . . . . . . . . . . . . . . 65
4.3.2 The KWM and the SVE . . . . . . . . . . . . . . . . . . . . . . . . 67
4.3.3 The Third Order Approximate Model and the SVE . . . . . . . . . 69
4.3.4 The BEM and the Proposed Approximate Model . . . . . . . . . . 70
4.3.5 The SVE, BEM and the Proposed Approximate Model . . . . . . . 71
4.4 Positive Surge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.5 Efficiency Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.5.0.1 The BEM and SVE . . . . . . . . . . . . . . . . . . . . . 75
4.5.0.2 The KWM and SVE . . . . . . . . . . . . . . . . . . . . . 78
vii
4.5.0.3 The Third Order Model and SVE . . . . . . . . . . . . . . 81
5 Conclusion and Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.3 Recommendation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
viii
List of Tables
4.1 Runtime in seconds of the SVE, BEM, KWM and the third order model . 61
ix
List of Figures
3.1 Conversion of a differential equation to a difference equation . . . . . . . . 30
3.2 Computational stencil for the Preissmann scheme . . . . . . . . . . . . . . 32
4.1 physical plane at time t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2 physical plane at time t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3 The SVE and the BEM respectively . . . . . . . . . . . . . . . . . . . . . . 55
4.4 The KWM and the third approximate model . . . . . . . . . . . . . . . . . 56
4.5 θ = 0.5, θ = 0.65 and θ = 1.00 respectively for SVE . . . . . . . . . . . . . 58
4.6 θ = 0.5, θ = 0.65 and θ = 1.00 respectively for BEM . . . . . . . . . . . . . 59
4.7 θ = 0.5, θ = 0.65 and θ = 1.00 respectively for KWM . . . . . . . . . . . . 60
4.8 Simulation of KWM with extremely high flow depth differences . . . . . . 63
4.9 Distance-time graph for SVE versus BEM . . . . . . . . . . . . . . . . . . 65
4.10 Distance-time percentage error plot for the SVE and the BEM . . . . . . . 66
4.11 Distance-time graph for SVE versus KWM . . . . . . . . . . . . . . . . . . 67
4.12 Distance-time percentage error plot for the SVE and the BEM . . . . . . . 68
4.13 Distance-time graph for the SVE and the third order model . . . . . . . . 69
4.14 Percentage errors for the third approximate model to the SVE . . . . . . . 70
4.15 Distance-time graph for the BEM and the third order model . . . . . . . . 71
4.16 Distance-time SVE, BEM and third order model . . . . . . . . . . . . . . . 72
x
4.17 Distance-time SVE, BEM and KWM . . . . . . . . . . . . . . . . . . . . . 73
4.18 Increasing time nodes t10 ⇒ tn for SVE, BEM and KWM respectively . . . 74
4.19 Efficiency Criteria for the prediction by BEM . . . . . . . . . . . . . . . . 76
4.20 Efficiency Criteria for the prediction by BEM . . . . . . . . . . . . . . . . 77
4.21 Efficiency Criteria for the prediction by KWM . . . . . . . . . . . . . . . . 79
4.22 Efficiency Criteria for the prediction by KWM . . . . . . . . . . . . . . . . 80
4.23 Efficiency Criteria for the prediction by the Third Order Model . . . . . . 82
4.24 Efficiency Criteria for the prediction by the Third Oder Model . . . . . . . 83
xi
Chapter 1
Introduction
The concept of open channel routing has been thoroughly investigated over time. This
chapter focuses on the review of background and research works carried in the area of open
channel routing that supports the presented thesis. It has sections that discuss the back-
ground studies, problem definition, the objectives within the thesis, the general methodol-
ogy employed, justification of the problem and the general structure of the thesis.
1.1 Background Studies
The application of Computational Fluid Dynamics (CFD) in industrial settings currently
span the disciplines of aerospace, automotive, power generation, chemical manufacturing,
polymer processing, petroleum exploration, medical research, astrophysics, etc. CFD is
the science of predicting fluid flow, heat transfer, mass transfer, chemical reactions, and
related phenomena by solving the mathematical equations which govern these processes
using a numerical process (discretization methods, solvers, numerical parameters, and grid
generations, etc.). The employment of CFD as a tool in these industries has led to reduc-
tions in the cost of product, process development and optimization activities. It has also
1
reduced the need for physical experimentation thereby heightening the overall economic
benefit from its usage.
The CFD process begins with a mathematical modeling of a physical problem within
which the exact geometry of the problem is defined. The concept of conservation of matter,
momentum and energy are incorporated into the model by making sure that the model
works within these concepts. The next step is to model the fluid properties empirically and
apply simplifying assumptions in order to make the problem tractable (i.e. steady-state,
incompressible, inviscid, one-dimensional, etc.) Provision is then made for appropriate
initial and boundary conditions for the problem. To be able to solve the problem, the CFD
process now uses numerical methods (generally referred to as discretization) to develop
approximations of the governing equation of fluid mechanics in the fluid region of interest.
The solution arrived at is further post-processed to extract other quantities of interest.
Specific applications of CFD include simulating flow and heat transfer in industrial pro-
cesses (boilers, heat exchangers, combustion equipment, pumps, pipings, etc.), application
in the aerodynamics of ground vehicles, aircraft, missiles, etc. It is also applied in film
coating, thermoforming in material processing applications, heat transfer for electronics
packaging and so many more. CFD is recognized as an engineering method that provides
data that is complementary to theoretical and experimental data. For purely scientific
studies, CFD is a useful tool especially in academic institutions and governmental research
laboratories.
One particular area that comes under the umbrella of CFD is the study of hydraulics;
2
an applied science and engineering dealing with the mechanical properties of liquids. The
branch under hydraulics which deals with free surface flows, occurring in rivers, canals,
lakes, estuaries, and seas is referred to as free surface hydraulics. The application of stud-
ies in this chapter of hydraulics include simulating dam break failures, flood alleviation
schemes, sediment transport and river applications. Open channel flow, a further sub-field
of free surface hydraulics is the subject of discussion in this thesis. It is simply a field that
is centered on describing flow in open channels.
Physical problems in the area of free surface hydraulics and CFD as a matter of fact
are governed by well defined set of physical principles which are translated into mathe-
matical statements mostly partial differential equations. These statements take the form
of equations in which the condition of the particular physical system considered plays the
role of the unknown. Scientists and particularly mathematicians make use of mathematical
models to describe the behavior of an associated physical system as it responds to a given
set of inputs.
An example of models in open channel flow is the Saint-Venant Equations for channel
routing. Also known as the shallow water equations, it is a set of hyperbolic partial differ-
ential equations that describe the flow below a pressure surface in a fluid. The fundamental
basis of almost all CFD problems are the Navier-Stokes equations, which define any single-
phase fluid flow. The Saint-Venant equations are no exception. Particularly important is
the use of non-linear approximate models to the Saint-Venant Equations for overland flow
and channel routing applications; highly motivating are these approximate models in use
3
because of their simplicity of application with fewer parameters and relatively presentable
ease of solutions. Examples of approximate models are the nonlinear Burgers’ Equation
Model (BEM), Kinematic Wave Model, Modified Puls, Muskingum, Muskingum-Cunge,
etc.
With the availability of high speed digital computers which did not happen until the
1960’s, problems which were seem impossible to analyze with respect to the nature of so-
lution are now being modeled and solved via numerical methods . The fact is that most
hydraulic models in particular the Saint-Venant equations and its approximate models can-
not be solved explicitly except by making so many assumptions which are unrealistic for
most applications. Numerical techniques are therefore employed in solving these equations.
Three groups of methods are usually used : finite difference, finite element and finite volume
(Sleigh and Goodwill (2000)). These result in computer programs that help to simulate
the problem at hand. For flow problems, the simulation involves flow variables such as
discharge, velocity, depth cross-section, volume and duration (time). The propagation of
particular interest are peak, time it takes to peak, duration of the hydrograph and at-
tenuation. Applications of channel routing models especially in the management of water
resources (Su Ki Ooi (2005)) has created a high demand for computer programs which
simulate routing variables peculiarly variables based on the one-dimensional Saint-Venant
Equations. Examples of solvers (computer programs) are MIKE - computer program that
includes full solution of the Saint-Venant equations, XPSWM - computer software pack-
age for dynamic modeling of storm water, HEC-RAS - computer program that models
4
hydraulics of water flow (unsteady), MIKE FLOOD, two dimensional MIKE 21, DIVAST,
LISFLOOD-FP, SOBEK, 1D2D, etc.
Approximate models and their dominance is not a new concept especially in an area
like hydraulics with reference to computer programs. The need for approximate models is
deeply rooted in the fact that the ’parent’ models are very complex and need comparatively
more parameters to run or simulate.
In hydraulics, there are two main branches of approximate models for simulating trans-
latory waves in waves in conveyance. These are the diffusive and the kinematic models.
The wide applications of these approximate models are because they are simple and pro-
vide solutions with ease. Price (1994) stated that the nonlinear approximate models serve
as better teaching tools for understanding nonlinear phenomena and have value of rapid
calculations.
Tsai (2003) also discussed the fact that the simplified routing models have fewer pa-
rameters and tend to have advantages in terms of feasibility and practicality when used
for real-time flood operations. An additional advantage is that simplified models need
significantly less computational requirement in terms of time and data (Singh et al 1998).
The simplest of all existing approximate models is the Kinematic Wave Model (Lighthill
and Whitham) although it cannot account for downstream influences such as tides on the
translatory wave [see —Price (1994), Onizuka and Odai (1998), S. N. Odai (1999)].
S. N. Odai (1999) developed an approximate model, the nonlinear Kinematic Wave
Model for predicting open channel flow rate. This was based on the full Saint Venant equa-
5
tions and assumed two constants estimated and expressed in terms of the froude number of
initial flow. The Kinematic Wave Model gave close results to the Saint Venant equations
under subcritical flow and for a given flow depth at the upstream end of the channel, the
approximate magnitude of flow rates at the downstream section is readily obtained in an-
alytical form. Other advantages of the KWM was in its simplicity and ease of application;
requiring less computation time comparatively to the SVE.
Literature documents the diffusive type of models as better compared to the Kinematic
Wave Models [see Akan and Yen (1977), Price (1994)] with which some reliable results
have been developed. They however do not place any significant advantage compared to
the SVE because the diffusive models cannot be solved analytically except for the reduction
in simulation time for numerical solution. The Burgers’ Equation, a form of nonlinear diffu-
sion equation is the simplest integrable mathematical formulation that shows the interplay
between nonlinear and diffusion effects on a propagatory wave [Burgers (1950), Whitham
(1974)].
The Burgers’ Equation Model that approximated the full SVE assuming that flow ve-
locity depends on flow depth and its gradient was presented by Onizuka and Odai (1998)
as a single lower-ordered equation that approximately described unsteady flow in a wide
rectangular open channel. The research was based on the normalized system of the Saint
Venant equations by Kubo and Shimura (1999).
K. Odai Onizuka and Osato (2006) presented an analytical solution of Burgers’ equa-
tion for simulating translatory waves in conveyance channels. In this research, K. Odai
6
Onizuka and Osato (2006) upgraded the previous Burgers’ equation model (Onizuka and
Odai (1998)) for small perturbations in initial uniform flow approximated using the SVE.
The analytical solution of the presented Burgers’ equation was done using Cole-Hopf and
solved via Green’s function and this was compared to the Saint Venant equations solved
numerically by cubic-interpolated pseudo-particle scheme (Yabe and Aoki (1991)) . The
research findings indicated that the Burgers’ Equation Model captures well the character-
istics of translatory wave moving in conveyance channels and hence could be useful for
simulating flood movement and for predicting the flow rate when the flow at an upstream
section is given.
1.2 Problem Definition
The Saint Venant equations are commonly used for prediction and control design for ir-
rigation channels, see Malaterre and Baume (1998) for an overview. Su Ki Ooi (2005)
demonstrated that the Saint Venant Equations can adequately capture the dynamics of
real channels and even concluded in the research that naturally, the model with estimated
parameters is more accurate than the one with physical parameters. Therefore on a qual-
itative scale, the significance of the dynamic model together with the present two class
of approximate models is known. The exact quantum of errors associated with using the
approximate models compared with the dynamic model is however unknown.
It is against this background that the researcher would like to investigate the quanti-
tative comparisons of the approximate models and the dynamic model. As presented in
7
the background studies above, there has been considerable discussion of the qualitative
predictions of the dynamic and its approximate models. Also since the BEM and KWM
are approximated using second order derivative equation and first order derivative equa-
tions respectively, the researcher proposes and presents a third order derivative equation to
examine the effect of adding extra derivative terms as a form of approximating the parent
model.
1.3 Objectives
In the light of the presentation made, the key objective of this study is to estimate a
third order approximate equation to the Saint Venant equation and compare the present
approximate models together with the proposed third order approximate equation to the
Saint Venant Equations.
• estimate a third order approximate model to the Saint Vennat Equations
• use the implicit Preissmann scheme to discretize and solve the SVE, BEM, KWM
and the proposed third order approximate model
• run an error analysis of the approximate models against the dynamic model
• investigate the exact range of the flow depth difference within which these models
present acceptable results
8
1.4 Methodology
This thesis presents both analytical solutions and numerical solutions of the two approxi-
mate models and the dynamic models discussed. The analytical solution of the Saint Venant
equations is by the method of characteristics. More important, is the use of implicit finite
difference methods to present numerical solutions of the considered models. The SVE,
BEM, KWM and the proposed model are all solved using the Preissmann scheme. The
sparse systems of equations that result from the implicit schemes are solved via standard
methods for solving systems of linear algebraic equations. Within the solution for the SVE
and the KWM using the Preissmann scheme, we encounter a nonlinear system of equations
for which the Newton algorithm for nonlinear systems of equations is used to solve. This
reduces to a banded system of linear equations which is taken advantage of Szymkiewicz
(2010).
After solution of the models, we use error analyzing techniques to investigate the per-
formance of the models. The numerical algorithms from the finite difference scheme are
programmed using MATLAB R2007b. The simulation for the models and error analysis
computation is entirely done using MATLAB R2007b.
1.5 Justification of Study
The nonexistence of analytical solutions of the dynamic model and its approximate models
without any assumption or simplification makes any research that is aimed at finding
9
alternate (approximate) solution in relation to the subject very relevant. This thesis will
provide algorithms for the normalized Saint Venant Equations based on the four-point
implicit Preissmann scheme and for the approximate models. The quantitative comparison
also sets the stage for very simple approximate models to be used instead of the dynamic
model with knowledge of thier divergence from the actual solution.
1.6 Structure of the Thesis
The structure of the thesis is given as follows.
• Chapter 2 presents an introduction of the Saint-Venant equations. The two approxi-
mate models are also derived from the normalized set of the Saint-Venant equations.
• Chapter 3 discusses the methodology employed in our study.
• Chapter 4 presents the analysis from the simulation results.
• Chapter 5 presents the conclusion and recommendations for further research
10
Chapter 2
Mathematical Models
2.1 Introduction
In this chapter, we present a general overview of the hydraulic models which are of con-
sideration with respect to this thesis. We begin by looking at the shallow water equations
by presenting it in a term by term procedure and introduce the Burgers’ Equation model
(BEM) and the Kinematic Wave Model (KWM). The derivation and general highlights of
the proposed third order approximate model is presented here.
2.2 Channel Routing Models
Shallow water equations are the commonly used models under open channel flow. They
are placed with the assumption that the flow is shallow relative to the dimensions of the
problem under consideration. The basis for the formation is a continuity equation which
corresponds to conservation of mass and an application of the laws of governing classical
physics which is used to generate the equation of motion. Depending on their construction,
such equations are often written as conservation laws representing the conservation of a
11
particular quantity such has momentum or energy. Then there is an addition of terms to
account for the effects such as friction, geometry variation, viscosity, etc. These additional
terms are often referred to as source terms which generally correspond to some sort of loss
or gain from the system.
In the preceding sections, of this chapter, the Saint-Venant equations are introduced
together with its alternate forms. The set of these partial differential equations in however
form presented called the Saint-Venant equation are also known as shallow water equations.
An introduction of the two approximate models (Burgers’ Equation Model and the Non-
linear Kinematic Wave Model) is also presented. Finally attached to these presentations is
the derivation of the third order approximate model.
2.3 The Saint Venant Equations
These are the most commonly used set of equations that constitute a system in one-
dimensional flows for open channel flow problems. They are used to describe gradually
varied flow of an incompressible inviscid fluid. The Saint Venant Equations (SVE) are a
set of hyperbolic partial differential equations that describe flow below a pressure surface in
a fluid. This model was developed by a mathematician and mechanician named Adhe’mar
Jean Claude Barre’ de Saint-Venant (August 23,1797, Villiers-en-Biere, Seine-et-Marne-
January 1886, Saint-Ouen, Loir-et-Cher). It is a combination of the principles of conti-
nuity (conservation of mass) and momentum (essentially Newton’s second law of motion).
These set of equations are primarily from the Navier Stokes equations.
12
Mathematically, given that Q is the discharge (m3 /sec), A is the cross-sectional area,
q is the lateral flow per unit length of the channel (m3 /sec/m), x is the distance along the
channel, y is the depth of flow, g is the acceleration due to gravity, S0 , the bed slope of the
channel and Sf the friction slope, the governing equations are written as :
∂Q ∂(A)
+ = q (continuity)
∂x ∂t
1 ∂ (Q2 /A)

1 ∂Q ∂y
+ +g − g(S0 − Sf ) = 0 (momentum)
A ∂t A ∂x ∂x
where the defined variables are given as follows
• 1/A(∂Q/∂t) - local acceleration terms which describes the change in the momentum
due to change in velocity over time
• (1/A)∂ (Q2 /A) /∂x - connective acceleration term which describes the change in mo-
mentum due to change in velocity along the channel
• g∂y/∂x - pressure force term representing the force proportional to the change in
water depth along the channel
The last two terms is a combination of the gravity force and the friction force terms
which describe the force proportional to the bed slope and the friction slope respectively.
The assumptions underlying the Saint-Venant Equations are :
• the bed slope is small resulting in the cosine of the angle between the bed level and
the horizontal begin approximately one.
13
• the pressure is hydrostatic (that is the streamline curvature is small and the vertical
accelerations are negligible).
• the effects of the boundary friction and turbulence can be accounted for by represen-
tations of channel conveyance derived from steady-state flow
The momentum equation presented is expressed in conservative form. It is possible to
substitute uA for Q in the continuity equation and the momentum equation. An expan-
sion of the momentum equation and a further simplification using the continuity equation
produces a mathematically correct non-convective form of the momentum equation. Use of
the non-convective form may however lead to practical difficulties in its numerical solution.
(Nelz and Pender (2009)).
Alternate forms of the Saint Venant equations include:
1. Using Q and h
∂h 1 ∂Q
+ = 0
∂t B ∂x

∂Q ∂ Q ∂h
+ + gA + gA(Sf − S0 ) = 0
∂t ∂x A ∂x
2. Using Q and y where y is the surface elevation of the platform
∂y 1 ∂Q
+ = 0
∂t B ∂x

∂Q ∂ Q ∂y
+ + gA + gASf = 0
∂t ∂x A ∂x
14
3. Using u and h

∂h A ∂u ∂h u ∂A
+ +u = 0
∂t B ∂x ∂x B ∂x h=const
∂u ∂u ∂h
+u +g + g(Sf − S0 ) = 0
∂t ∂x ∂x
4. Using u and y

∂u A ∂u ∂y u ∂A
+ +u + S0 + = 0
∂t B ∂x ∂x B ∂x y=const
∂u ∂u ∂y
+u +g + gSf = 0
∂t ∂x ∂x
Another important alternate form of the Saint-Venant equations considered for wide
rectangular open channels is given by
∂h ∂(uh)
+ = 0 (continuity equation) (2.1)
∂t ∂x
∂u ∂u ∂h 1 u2
+u + + 2 2m − 1 = 0 (dynamic equation) (2.2)
∂t ∂x ∂x Fo h
where h = flow depth; u = flow velocity; x = distance in flow direction; t = time; g =
acceleration due to gravity; S0 = channel bottom slope and Sf = friction slope.
Considering H = initial normal depth; L = reference channel length (= H/S0 ); U =

√ √
celerity of a shallow wave (= gH) and T = time constant (= L/ gH).
Then the flow variables can be normalized with the normalized variables given as follows:
h 0 x 0 u 0 t
h0 = x = u = t =
H L U T
With these conversions and replacement of the variables themselves by dropping the
primes, (2.1) and (2.2) can be converted to a normalized system of Saint Venant equations
15
by Kubo and Shimura (1999) as
∂h ∂(uh)
+ = 0 (2.3)
∂t ∂x
∂u ∂u ∂h 1 u2
+u + = 1 − 2 2m (2.4)
∂t ∂x ∂x F0 h
with all the primes dropped from the variables. F0 is the froude number of initial uniform
flow and it is given by

√
H m S0
F0 =
nU
where n = Manning’s roughness coefficient; and m = 2/3 for Manning’s formular and
m = 1/2 for Chezy’s formular.
Equations (2.3) and (2.4) will be used throughout this thesis as the Saint Venant equa-
tions.
2.4 Kinematic Wave Equation
The non-linear Kinematic equation is given by
∂h ∂h
+ (α + βh) =0 (2.5)
∂t ∂x
is extracted from the continuity equation (2.3) and a reduced form of the momentum
equation (2.4) assuming small perturbations in the initial normal depth. The constants in
the equation are given as
5 10 5
α = Fo ; β= Fo and α + β = Fo (2.6)
9 9 3
16
2.5 The Burgers’ Equation Model
The Burgers’ equation model is simplest integrable mathematical formulation that shows
the complicated interplay between nonlinear and diffusion effects on a wave (see Burgers
(1950)). Onizuka and Odai (1998) extracted the Burgers’ equation of the form
∂h ∂h ∂ 2h
+ (α + βh) −υ 2 =0 (2.7)
∂t ∂x ∂x
from the Saint Venant equations (2.3) and (2.4) assuming small perturbations in the
initial normal depth. The constants within the Burgers’ equation are given as

5 10 Fo 4 2 5
α = Fo ; β = Fo ; υ = 1 − Fo ; α + β = Fo
9 9 2 9 3
The model is applicable within the range 0 < Fo < 3/2 [see Odai et al (2006)]
2.6 The Proposed Third Order Approximate Model
The proposed third order derivative approximate model is given by
∂h ∂h ∂ 2h ∂ 3h
+ (a + bh) −c 2 +d 3 3 =0 (2.8)
∂t ∂x ∂x ∂ x
The following is the derivation of the coefficient constants within the model
2.6.1 Estimation of Model Constants
From continuity and dynamic equations assuming small perturbations in the initial normal depth. Eqn
(2.8) satisfies the continuity equation if the flow rate per unit width, uh, satisfies the equation
∂h ∂2h
uh = ah + 0.5bh2 − c +d 2 +K (2.9)
∂x ∂x
17
where K is constant. At the initial uniform flow, where h = 1, u = F0 the first and second derivatives of
h with respect to x equals zero
F0 = a + 0.5b + K
⇒K = F0 − a − 0.5b
The flow velocity, u can then be expresed in terms of the flow depth, h as
∂2h

∂h
u= F0 + a(h − 1) + 0.5b(h2 − 1) − c + d 2 /h (2.10)
∂x ∂x
From Eqn (2.10), we can calculate the following variables

2
F0 + a(h − 1) + 0.5b(h2 − 1) −2c F0 + a(h − 1) + 0.5b(b2 − 1) ∂h
u2 = −
h2 h2 ∂x
2
2 2

F0 + a(h − 1) + 0.5b(h − 1) −2c F0 + a(h − 1) + 0.5b(b − 1) ∂ 2 h
+2d −
h2 h2 ∂x2
2 2
2
2 2
2 2
cd ∂h ∂ h c ∂ h d ∂ h
− 2 + 2 + 2 (2.11)
h ∂x ∂x2 h ∂x2 h ∂x2
2
[F0 + a + 0.5bh2 + 0.5b] ∂h c ∂2h d ∂3h ∂2h

∂u c ∂h d ∂h
= 2
− + + 2 − (2.12)
∂x h ∂x h ∂x2 h ∂x3 h ∂x h2 ∂x ∂x2
∂u 1 ∂h
u = [F0 + a(h − 1) + 0.5b(h2 − 1)][−F0 + a + 0.5bh2 + 0.5b]
∂x h3 ∂x
2
c ∂ h d ∂3h
− 2 [F0 + a(h − 1) + 0.5b(h2 − 1)] 2 + 2 [F0 + a(h − 1) + 0.5b(h2 − 1)] 3
h ∂x h ∂x
2 2
c2 + dh(−2F0 − ah + 2a + b) ∂h
2
c ∂ h ∂ h
+ 3 [2F0 − 2a + ah − b] +
h ∂x2 h2 ∂x ∂x2
3 3 2 2
c2 ∂h ∂2h cd ∂ 2 h

cd ∂h ∂ h cd ∂h
− 2 3
− 3 + 3 2
− 2
h ∂x ∂x h ∂x h ∂x ∂x h ∂x2
3 2
d2 ∂ 2 h ∂3h cd ∂ 2 h d2 ∂h ∂2h

+ − (2.13)
h2 ∂x2 ∂x3 h3 ∂x2 h3 ∂x ∂x2
∂ 2 ∂h

∂u 1 ∂h ∂ ∂h
= (a + bh) −c +d 2 −
∂t h ∂t ∂x ∂t ∂x ∂t
∂2h

1 ∂h 2 ∂h
F0 + a(h − 1) + 0.5b(h − 1) − c +d 2 (2.14)
h2 ∂t ∂x ∂x
18
However from the proposed third order approximate model we are able to find an expression for ∂h/∂t
(2.15) and evaluate the first and second derivatives of ∂h/∂t with respect to x as shown in equations (2.16)
and (2.17).
∂h ∂h ∂2h ∂3h
= −(a + bh) +c 2 − (2.15)
∂t ∂x ∂x ∂x3
2 2
∂3h ∂4h

∂ ∂h ∂h ∂ h
= −b − (a + bh) 2 + c 3 − d 4 (2.16)
∂x ∂t ∂x ∂x ∂x ∂x
2 2
2
∂3h ∂4h ∂5h

∂ ∂h ∂ h ∂h ∂ h
= −2b − b − (a + bh) + c − d (2.17)
∂x2 ∂t ∂x2 ∂x ∂x2 ∂x3 ∂x4 ∂x5
Therefore
∂h c[−F0 + a(h + 1) + 0.5b(3h2 + 1)] ∂ 2 h

∂u (a + bh 2
= [F0 − a − 0.5bh − 0.5b] +
∂t h2 ∂x h2 ∂x2
2 2 3 4 2 5
−2h(bd + c ) + d[F0 + a(h + 1) + 0.5b(h − 1)] ∂ h 2cd ∂ h d ∂ h
+ + −
h2 ∂x3 h ∂x4 h ∂x5
2 2
2
3
2
2 3
ac ∂h ad + c ∂h ∂ h cd ∂h ∂ h d ∂ h ∂ h
− 2 + − 2 + 2
h ∂x h2 ∂x ∂x2 h ∂x ∂x3 h ∂x2 ∂x3
2 2
cd ∂ h
(2.18)
h2 ∂x2
Substituting these expressions (2.11), (2.13) and (2.18) into the assumed proposed third order ap-
proximate model (2.18), simplifying and averaging terms with the same derivative coefficients results in
∂h ∂2h ∂3h
A(h) = A0 (h) + A1 (h) + A2 (h) 2 + A3 (h) 3 + · · · (2.19)
∂x ∂x ∂x
where
2
Fo + a(h − 1) + 0.5b(h2 − 1)
A0 (h) = 1 −
Fo2 h2m+2
2
Fo + a(h − 1) + 0.5β(h2 − 1) (a + bh)
A1 (h) = 1 − 2c 2 2m+2
+ [F0 − a − 0.5bh2 − 0.5b]
Fo h h2
1
[F0 + a(h − 1) + 0.5b(h2 − 1)][−F0 + a + 0.5bh2 + 0.5b]
h3
c[−F0 + a(h − 1) + 0.5b(3h2 + 1)] c
A2 (h) = − 2 [F0 + a(h − 1) + 0.5b(h2 − 1)] +
h2 h
2d[F0 + a(h − 1) + 0.5b(h2 − 1)
F02 h2m+2
19
A very important note at this point is the fact that since the proposed model is an approximation of
the Saint Venant equations, not every term in the expression for A(h) (2.19) can be eliminated.
The normalized Saint Venant equations explicitly specify a relation between the first derivatives of the
flow variables. Thus here, the first three derivatives are eliminated for small perturbations of flow depth
about the initial constant.
At h = 1, A0 (h) = 0. Taking the first derivative of A0 (h) with respect to h at h = 1 gives the expression
(a + b) [i.e. from A00 (1)]; Then taking the second derivative of A0 (h) with respect to h at h = 1 gives an
expression for b [i.e. A000 (1)]. c is obtained from A1 (1). The last constant d is obtained from A2 (h) at h = 1.
Thus the coefficients of the third order approximate model to the Saint Venant Equation are given by
F2

5 10 Fo 4 2 5 4 2
a = Fo ; b= Fo ; c= 1 − Fo ; a + b = Fo d = − o 1 − Fo
9 9 2 9 3 3 9
20
Chapter 3
Methodology
3.1 Numerical Methods
Numerical Methods in its entirety is the transformation of mathematical models into an
algebraic, linear or non-linear, system of equations for mesh/grid - related unknown quan-
tities. The transformation procedure is referred to as discretization or space discretization
where a mesh/grid is set up and the continum of space is replaced by a finite number of
points where the numerical of the variable will have to be determined. The accuracy of the
numerical results is critically dependent on the mesh quality.
Numerical methods can generally be classified as:
1. Finite Difference Methods (FDM)
2. Finite Element Methods (FEM)
3. Spectral Methods
4. Finite Volume Methods (FVM)
We shall discuss an overview of all the methods and pay particular attention to FDM
which we will make use of in this thesis.
21
In general terms, finite difference methods represent the problem through a series of
points or nodes. Expressions for the unknown are derived via replacing the derivative
terms in the model equation with truncated or approximate Taylor series expansions. The
earliest numerical schemes were based on finite difference schemes and are conceptually and
intuitively one of the easier methods amongst the given classification to implement. Most
texts describe finite difference as the most traditional and oldest of numerical methods.
However, fundamentally such techniques require a high degree of regularity with the mesh
and so this limits their application to complex problems.
Finite element method employs the use of dividing the domain into elements such as
triangles or quadrilaterals and to place with each element nodes at which the numerical so-
lution is determined. The solution at any position is then represented by a series expansion
of the nodal values within the local vicinity of that position. The nodal contributions are
multiplied by basis functions (also known as shape, interpolation, or trial functions) and
the particular way in which the basis functions are defined determines the choice of variant
of the finite element method. Spectral methods can be considered as a subset of the finite
element method in which the basis functions are defined globally as opposed to the more
common approach whereby the basis functions are local and so one zero outside the neigh-
borhood of the associated node. The original finite element method was developed within
the field of stress analysis and this reflected within the construction and nomenclature of
the approach.
The finite volume method is based on forming a discretization from an integral form
22
of the model equations, and entails subdividing the domain in a number of finite volumes.
Within each volume, the integral relationships are applied locally and so exact conservation
withing each volume is achieved. The resulting expressions for the unknowns often appear
similar to finite difference approximations and depending on the particular method chosen
may be considered as a special case of either the finite difference or finite element tech-
niques. With the emphasis of most fluid modeling problems being based on conservation
principles, the finite volume methods has become more popular approach for general fluid
flow problems
Setting up a numerical scheme therefore involves:
1. Selection of a discretization method for the equations in the model. This implies the
selection between finite difference, finite volume, spectral or finite volume methods
as well as selection of the order of accuracy of the spatial and eventually the time
discretization.
2. Analysis of the selected numerical algorithm. This step concerns the analysis of the
qualities of the scheme in terms of stability and convergence properties as well as the
investigation of generated errors.
3. Selection of a resolution method for the system of ordinary/partial differential equa-
tions in time, for the algebraic system of equations and for the iterative treatment of
eventual non-linearities.
We first take a look at numerical schemes that we have used in this thesis with respect
23
to how they are generally applied and some properties that make their applications wide
in the area considered. Then the preceding sections give a general look at solving linear
and non-linear algebraic systems of equations that come up with regards to finite difference
equations.
3.1.1 Efficiency of a Numerical Scheme
There are four general ways to determine how well a particular numerical technique per-
forms in generating results to a problem being modeled. These are accuracy, consistency,
stability and convergence. The following are summaries of these characteristics.
Accuracy
This is the measure of how well discrete solutions represent the exact solution of the prob-
lem. We use the local / truncation errors; this measures how well the difference equations
match the differential equations. We also use global error which reflects the overall error
in the solution; possible only with the existence of an exact solution. An expression for the
truncation error can be obtained by substituting the known exact solution of the problem
into the discretization, leaving a remainder which is then a measure of the error.
Consistency
Consistency is related to adjusting grid resolution and its effect on the results. Mathe-
matically, a method is said to be consistent if the truncation error decreases as the step
24
size is reduced which is the case when q, p ≥ 0 This is equivalent to saying that as ∆t, ∆x
tends to zero, the discretized equations should tend towards the differential equations. It
is necessary that a scheme be consistent for its practical use.
Stability
A numerical scheme is said to be stable if all of the errors such as round off errors due to
finite arithmetic of the computer that exists within the solution are bounded. An unstable
method results in a solution that tend towards infinity.Most methods have restrictions
placed on the nature of their stability for example is the common use of the ratio of ∆x to
∆t plus a factor.
Convergence
Simply stated, convergence of a method refers to the ability of the method to approach the
exact solutions as the grid spacing is reduced to zero. This is coupled with the global error.
Another very important criterion which must be satisfied in order to produce a valid
solution is the issue of the particular problem being well posed. A problem is considered
well posed if the following conditions are well met:
1. An existence of a solution
2. Uniqueness of the solution
3. A linear dependence of the solution on the data
25
The last condition can be translated to mean that the solution should not be sensitive
to small changes in the initial / boundary data of the problem. If a problem is not well
posed, then a valid numerical solution cannot be generated and any numerical treatment
will either fail or produce poor results. Application of inappropriate boundary conditions
is a way of producing an ill-posed problem.
3.1.2 Explicit and Implicit Formulations
A major division in numerical methods is considering whether a particular method is explicit
or implicit.
If the solution is to be advanced to time level n + 1, the spatial derivatives may be
approximated either in terms of the known values at time level n or the unknown quantities
at time level n + 1.
An explicit method is generated from an approximation for the spatial derivatives using
time level n. Put in a more mathematical way, for time dependent numerical formulations,
the matrix of the unknown variables at the new time is a diagonal whiles the right hand
side of the system is being dependent only on the flow variables in the previous times. This
leads therefore to a trivial matrix inversion and hence to a solution with minimal number
of arithmetic operations for each time step.
An implicit method on the other hand results from using the time level n + 1 as approx-
imations for the spatial derivatives. For time dependent formulations, this is a situation
where the matrix to be inverted is not diagonal since more than one set of the variables
26
are unknown at the same time level.
Explicit methods are generally simpler in terms of the resulting algebraic equations as
compared to implicit methods that require matrix inversions which turn out to be compu-
tationally costly. The advantage of implicit schemes lies in their independence of the times
steps since most explicit methods turn out to be conditionally stable.
The next section is on finite difference schemes and their applications.
3.2 Finite Difference Schemes
The first application of this class of numerical methods is attributed to Leonard Euler
(1707 - 1783) in 1768. This was the first technique to be developed for approximating
ordinary differential equations from which theories regarding their properties and alternate
applications in other subjects have been generated. This section gives a summary of finite
difference methods.
The basis for finite difference methods is the expansion of Taylor Series and substitution
of truncated expressions into the differntial equati. This can be applied to any structured
mesh configuration.
Consider a function u(x), the derivative at point x is simply defined by
du u(x + ∆x) − u(x)

ux = = lim = (3.1)
dx x→0 ∆x
The removal of the limit in the above equation results in a finite difference, which ex-
plains the name given to it. When ∆x is small this formula can be used as an approximation
27
for the derivative of u at x.
Then the numerical solution ui can be thought of as point values where ui = u(i∆x).
With reference to the notation discussed the approximate first derivative of u with respect
to x can be written alternatively as:
1. Forward difference
ui+1 − ui
(ux )i = + O(∆x)
∆x
2. Backward difference
ui − ui−1
(ux )i = + O(∆x)
∆x
3. Central difference
ui+1 − ui−1
(ux )i = + O((∆x)2 )
2∆x
These can all be used to approximate the first derivative of u with respect to x. In
the area of truncation errors, the forward and backward differences are both first order
approximations whereas the central difference is second order, as can be shown by Taylor
series analysis. These formulae depends on the problem being modeled. In the case of
PDEs for example which we will be taking particular look at, most schemes are based upon
straight forward, backward and central difference formula.

The examples considered above therefore constitute the fundamental concepts that hold
finite difference methods in place. The complexity that unfolds in this branch of compu-
tation is solely based on how these equations are proved. Considering u now as a function
of two variables, i.e. u(x, t), then the following are derivatives taken with respect to the
independent variables and the associated errors attached to the derivatives. Considering
28
u(xi , tj ) and the fact that a value of u(x, t) at a grid point (xi , tj ) can be denoted by
uji = u(xi , tj ), we can write
1. Forward difference for ux :

uji+1 − uji
ux (xi , tj ) = + O(∆x)
∆x
2. Backward difference for ux :

uji − uji−1
ux (xi , tj ) = + O(∆x)
∆x
3. Centered difference for ux

uji+1 − uji−1
ux (xi , tj ) = + O((∆x)2 )
2∆x
4. Centered difference for uxx
uji−1 − 2uji + uji+1

uxx (xi , tj ) = + O((∆x)2 )
(∆x)2
5. Forward difference for ut :

uj+1
i − uji
ut (xi , tj ) = + O(∆t)
∆t
6. Centered difference for ut :

uj+1 − uij−1
ut (xi , tj ) = i
+ O((∆t)2 )
2∆t
7. Centered difference for utt :
uj−1 − 2uji + uj−1

utt (xi , tj ) = i i
+ O((∆t)2 )
(∆t)2
These formulae are most useful for solving PDEs. The notation used to represent the
derivatives is known as the compact subscript-superscript and order of magnitude notation.
29
3.2.1 Application of Finite Difference Schemes to PDEs
Finite difference schemes are used to develop efficients methods for approximating solu-
tions to partial differential equations. Suppose that a partial differential equation (PDE)
is expressed symbolically as L[u] = S, holds on some region Ω of the x − t plane. Let us
cover with a finite difference grid (xi , tj ) = (i∆x, j∆t). The figure below is an illustration
of the procedure. If all the derivatives in the PDE are replaced by difference quotients, the
results is a finite difference equation(FDE). This is known as the method of discretization
or differencing the continuous PDE problem to obtain a discrete FDE problem. Differenc-
ing therefore entails using difference quotients to approximate derivatives in the manner
discussed.
Figure 3.1: Conversion of a differential equation to a difference equation
The solution of the FDE Uij approximates uji = u(xi , tj ), the solution of the PDE at
the grid points. If the discretization is to provide a useful approximation, the solution
of the PDE should nearly satisfy the FDE when the grid spacings ∆x and ∆t are taken
sufficiently small. The conditions for efficiency of a numerical scheme should apply here.
30
3.3 Numerical Scheme for the Mathematical Models
This thesis takes into consideration the Preissmann scheme which is an implicit finite dif-
ference scheme. The researcher uses this weighted scheme to solve all the models discussed
in this thesis.
3.3.1 Preissmann Scheme
The Preissmann scheme is the most widely applied implicit finite difference scheme because
of its simple structure with both flow and geometrical variable in each grid point. This
implies a simple treatment of boundary conditions and a simple incorporation of structure
and bifurcation points. Also it has the advantages that steep wave fronts may be properly
simulated by varying the weighting coefficient, α.
The scheme was first proposed by Alexander Preissmann in 1961 when he was a hy-
draulic engineer. It has been used since then for hydraulic modeling in various distinctions
with the hydraulic setting. [Preissmann and Cunge 1961; Ligget and Cunge, 1965, Cunge,
Holly and Versey, 1980]. Also known as the the box scheme, the reasons for it’s widespread
use are as follows:
• it works on non-staggered grid, which allows us to calculate both unknowns in the
same nodes. This is important in natural rivers.
• it relates the variables coming from neighboring nodes only, what allows us using
variable space interval ∆x without affecting the accuracy of approximation.
31
• it ensures approximation of 1st order of accuracy and for particular case of 2nd order.
• it gives exact solution of the linear wave equations for properly chosen values of ∆x
and ∆t, making possible the comparison of exact and numerical solutions.
• it is implicit and absolutely stable so it does not require limiting of the value of time
step, whereas the imposed boundary conditions are introduced readily.
The four-point implicit Preissmann finite difference scheme uses a forward difference
and weighted parameter, α as shown below:
Figure 3.2: Computational stencil for the Preissmann scheme
A point up is therefore approximated as
1 1
αuk+1 + (1 − α)uki + αuk+1 k

up = i i+1 + (1 − α)ui+1
2 2
with k, i begin the time and spatial indices respectively. The weighted parameter α lies
32
between 0 and 1. Derivatives are approximated as follows:
uk+1 k+1

∂u i+1 + ui − uki+1 + uki
=
∂t 2∆t
k+1 k+1

∂u ui+1 − ui uki+1 − uki
= α + (1 − α)
∂x ∆x ∆x
This scheme is second order if α = 0.5, and first oder accurate otherwise. Linear
stability analysis shows that the centered scheme is unconditionally stable for α ≥ 0. This
feature makes it very interesting for practical applications, since contrarily to the case of
explicit scheme, it is not subject to the Courant-Friedrichs-Levy stability condition that
constraints the time to small values.
The Preissmann for these likely reasons has become the standard method for one-
dimensional numerical modeling in the field of hydraulic engineering. A stability analysis
including convective and friction terms has shown another condition is necessary however
for the stability of the Preissmann scheme in addition to the conditions already discussed.
The Vedernikov number γ, must be smaller than 1 where γ is defined by
a A dR
γ= F0
b R dA
where a is the exponent of the hydraulic radius and b the exponent of the velocity in
the evaluation of the friction slope, A is the cross-sectional area of flow, R is the hydraulic
radius and F0 is the Froude number.
The Preissmann scheme is the discretization method used in the Institute for Software
Integrated Systems (ISIS) package for the unsteady solver simulation which solves Saint
Venant based equations for free surface flows(open channels)
33
3.4 Solution to Systems of Algebraic Equations
One of the very basic problems that occur in numerical analysis of partial differential equa-
tion with finite difference especially is the solution of linear systems of algebraic equations.
Most times, nonlinear systems of partial differential equations are converted using existing
methods of solution to linear systems of equations before they are solved. There exists two
main branches of solution with regards to the subject of linear systems: i.e. direct methods
and iterative methods. The sections that follow highlight general solution to system of
equations and also discusses how to handle non-linear systems of equations by conversion
to linear systems.
3.4.1 Linear Systems of Equations
Consider a system of n linear equations with n unknowns for which in matrix form we can
write as
AX = B (3.2)
where
     
 a11 a12 · · · a1n   x1   b1 
     
     

 a21 a22 · · · a2n 


 x2 

 b2 
A=  X=  B= 
 .. .. ... ..   ..   .. 

 . . . 


 . 


 . 

     
     
an1 an2 · · · ann xn bn
where A = (aij ) (i, j = 1, 2, . . . , n) is a real square matrix of system coefficients,
34
X = (xi ) (i = 1, 2, . . . , n) is a column vector of unknowns and B = (bi ) (i = 1, 2, . . . , n)
is a column vector of constants.
It is assumed that the matrix A is nonsingular so that the system (3.2) has a solution.
The matrix A mostly occurs in practice as diagonal matrices, lower triangular ma-
trices, upper triangular matrices,sparse matrices, banded matrix or in some cases
symmetric matrices.
For the cases listed, advantage can be taken of the special structure of the matrix to
generate solution algorithm much more efficient. The choice of either a whether to use a
direct method or iterative method directly depends on these factors. The nature and size
of the matrix are factors that influence the choice of methods of solution. Also important
is the kind of solutions that these methods bring on board. Direct methods are capable of
providing exact solutions after executing a finite number of mathematical operations (on
condition that round off errors do not occur). Iterative methods on the other hand make
use of an initial guess of the solution and provide a series of better approximations for the
solution. In the process of doing this, if the series of solution that the iterative method
present converges, i.e. if it tends to a limit, then the limit is referred to as the solution of
the considered system.
3.4.2 Solution to Nonlinear System of Equations
For systems that are non-linear, there are a number of techniques available for the solution
of the unknown vector. The methods of solution for non-linear systems are all iterative.
35
Consider a system of algebraic equations
AX = B (3.3)
If in modeling a problem, this results in the coefficients matrix being functions of the
unknowns i.e. aij = aij (x1 , x2 , . . . , xn ) or/and bi = bi (x1 , x2 , . . . , xn ), the system (3.3) is
referred to as a nonlinear system. This is the occurrence as we will see when the Saint
Venant equations are discretized using the four point Preissmann scheme. The nonlinear
system of equation is usually represented by
F(X) = 0 (3.4)
where
F(X) = AX − B
X = (x1 , x2 , x3 , . . . , xn )T
F = (f1 (x), f2 (x), f3 (x), . . . , fn (x))T
The following is a brief note on the Newton Method which is a standard method of
solution for nonlinear systems.
Newton Method
Assume that the approximate solution of the nonlinear system is known. Then we can
write
X = X(k) + ∆X(k) (3.5)
36
where
X = exact solution of the nonlinear of the system
X(k) = solution approximation i iteration k
∆X(k) = vector of differences between the exact and estimate
k = index of iteration
Assume that the functions being components of the system are continuous and differ-
entiable with regards to X. Using the fact the F(X) = 0 and using Taylor series expansion
around X(k) , we obtain

(k) (k) (k) ∂F X(k)
∆X(k) ≈ 0

F X + ∆X ≈F X + (3.6)
∂X
with the correction vector expressed as
∆X(k) = X(k+1) − X(k) (3.7)
Since X(k) estimates the exact solution X. It is therefore possible to write:

∂F X(k)
X(k+1) − X(k) = −F X(k)

(3.8)
∂X
where  
(k) (k) (k) (k) (k) (k)
∂f1 x1 ,x2 ,...,xn ∂f1 x1 ,x2 ,...,xn
 ∂x1
··· ∂xn 
 
(k) (k) (k) (k) (k) (k)
 ∂f2 x1 ,x2 ,...,xn ∂f2 x1 ,x2 ,...,xn 
···
(k)
 
∂F X  ∂x1 ∂xn 
= J(k) =  (3.9)
∂X  .. .. .. 

 . . . 

 
(k) (k) (k) (k) (k) (k)
 ∂fn x1 ,x2 ,...,xn ∂fn x1 ,x2 ,...,xn 
∂x1
··· ∂xn
37
(3.9) is the Jacobian of the nonlinear system (3.4). Then the vector form of the Newton
Method is given as:

−1
X(k+1) = X(k) − J(k) F X(k)

(3.10)
Because of the time and memory consuming nature of this structure () i.e. the building
of the Jacobian matrix of dimension n × n in each iteration and the inversion of the matrix,
it is rather preferable to use the system given below:
J(k) ∆X(k) = −F(k) (3.11)
This system (3.11) is a linear system that can be solved in terms of the correction vector
∆X(k) . Afterwards the new estimate of X is calculated as
X(k+1) = X(k) + ∆X(k) (3.12)
When the matrix J happens to be banded, the approach is more efficient as discussed
under systems of linear of equations. Note that once the Jacobian matrix is inverted, it
looses its banded form and has all non-zero elements.
3.5 Numerical Solution of Models
This section presents the solution to the discussed routing models using the Priessmann
Scheme also known as the Box Scheme. They include solutions for the Burgers’ Equation
model (BEM), Kinematic Wave Equation (KWM), The proposed third order approximate
model and the normalized Saint Venant Equations.
38
3.5.1 Burgers’ Equation Model
The Burgers’ equation is given by
∂h ∂h ∂ 2h
+ (α + βh) −c 2 =0 (3.13)
∂t ∂x ∂t
Using the box scheme the derivatives are given as follows:
∂h 1 1
hk+1 k+1
hki+1 + hki−1

= i+1 + hi−1 −
∂t 2∆t 2∆t
∂h θ (1 − θ) k
hk+1 − hk+1 hi+1 − hki

= +
∂x ∆x i+1 i
∆x
∂2h hk − 2hki + hki−1 hk − 2hki + hki−1
= (1 − θ) i+1 + θ i+1
∂x2 ∆x 2 ∆x2
Substituting the derivatives into the BEM, we get

" #
hk+1 k+1

k+1 k ∆t hki+1 − hki−1 i+1 − hi−1
hi − hi + α (1 − θ) +θ +
2 ∆x ∆x
∆t k k+1
hi+1 hi+1 − hki−1 hk+1

β i−1 −
2
∆t
(1 − θ) hki+1 − 2hki + hki−1 + θ hk+1 k+1
+ hk+1

γ i+1 − 2hi i−1 = 0
∆x2
γ∆t ∆t
let s= , r=
∆x2 ∆x
αr αr
hk+1 + hki + (1 − θ) hki+1 − hki−1 + θ hk+1 k+1

i i+1 − hi−1
2 2
βr k k+1 k+1
hi+1 hi+1 − hki−1 hi−1

+
2
−s(1 − θ) hki+1 − 2hki + hki−1 − sθ hk+1 k+1

+ hk+1

i+1 − 2hi i−1 = 0
This can be simplified further as
a4 + a6 hki−1 + a9 hk+1 k+1

+ a3 + a5 hki+1 + a9 hk+1

i+1 + (1 + a0 ) hi i+1
= − (a2 + a7 ) hki−1 − (a8 − 1) hki − (a1 + a7 ) hki+1
where the introduced variables are defined by
a1 = αr(1 − θ)/2 a3 = αθ/2 a5 = βr/2 a7 = −s(1 − θ) a9 = −sθ
a2 = −αr(1 − θ)/2 a4 = −αθ/2 a6 = −βr/2 a8 = 2s(1 − θ) a0 = 2sθ
39
anj a4 + a6 hki−1 + a9

=
bnj = (1 + a0 )
cnj a3 + a5 hki+1 + a9

=
dnj = − (a2 + a7 ) hki−1 − (a8 − 1) hki − (a1 + a7 ) hki+1
then the result is a triangular system of equations as follows
n+1
anj hj−1 + bnj hn+1
j + cnj hn+1 n
j+1 = dj (3.14)
illustrated in matrix form as

    
 • •  h22
  − d12 a12 h21 
    
    
 • • •   h2   • 
  3   
  =  (3.15)
    
  h24  
• • •  •
    
 
    
    
• • h15 d15 − c15 h26
We solve this system using appropriate methods of solving systems of algebraic equations. The proce-
dure is then repeated for calculations of h at n = 2, 3, . . . , 5
3.5.2 The Proposed Equation
The third order equation equation model is given by
∂h ∂h ∂ 2h ∂ 3h
+ (α + βh) −c 2 +d 3 =0 (3.16)
∂t ∂x ∂t ∂t
Using the box scheme the derivatives are given as follows:
∂h 1 1
hk+1 k+1
hki+1 + hki−1

= i+1 + hi−1 −
∂t 2∆t 2∆t
∂h θ (1 − θ) k
hk+1 k+1
hi+1 − hki

= i+1 − hi +
∂x ∆x ∆x
∂2h hki+1 − 2hki + hki−1 hk+1
i+1 − 2hi
k+1
+ hk+1
i−1
= (1 − θ) + θ
∂x2 ∆x2 ∆x2
∂3h hk − 2hki+1 + 2hki−1 − hki−2 hk+1 k+1 k+1 k+1
i+2 − 2hi+1 + 2hi−1 − hi−2
= (1 − θ) i+2 + θ
∂t3 2∆x3 2∆x3
40
Substituting the derivatives into the proposed third order equation model, we get
" #
hk+1 k+1

∆t hki+1 − hki−1 i+1 − hi−1
hk+1
i − hki +α (1 − θ) +θ +
2 ∆x ∆x
∆t k k+1

β i−1 −
2
∆t
(1 − θ) hki+1 − 2hki + hki−1 + θ hk+1 k+1 k+1

γ 2 i+1 − 2hi + hi−1
" ∆x #
d∆t hki+2 − 2hki+1 + 2hki−1 − hki−2 hk+1 k+1 k+1
i+2 − 2hi+1 + 2hi−1 − hi−2
k+1
+ (1 − θ) +θ = 0
3∆x3 2∆x3 2∆x3
γ∆t ∆t d∆t
let s= , r= p=
∆x2 ∆x 2∆x3
αr αr
hk+1 + hki + (1 − θ) hki+1 − hki−1 + θ hk+1 k+1

i i+1 − hi−1
2 2
βr k k+1

+ i−1
2
−s(1 − θ) hki+1 − 2hki + hki−1 − sθ hk+1 k+1

+ hk+1

i+1 − 2hi i−1
+p θ(hk+1 k+1 k+1 k+1 k k k k

i+2 − hi+1 + 2hi−1 − hi−2 ) + (1 − θ)(hi+2 − hi+1 + 2hi−1 − hi−2 ) = 0
This can be simplified further as
b4 hk+1
k+1 k+1
k
+ b2 + a3 + a9 + a5 hki+1 hk+1 k+1

i−2 + b3 + a9 + a4 + a6 hi−1 hi+1 + (1 + a0 ) hi i+1 + b1 hi+2
= −b8 hki−2 − (b7 + a2 + a7 ) hki−1 − (a8 − 1) hki − (b6 + a1 + a7 ) hki+1 − b5 hki+2
where the introduced variables are defined by
a1 = αr(1 − θ)/2 a3 = αθ/2 a5 = βr/2 a7 = −s(1 − θ) a9 = −sθ
a2 = −αr(1 − θ)/2 a4 = −αθ/2 a6 = −βr/2 a8 = 2s(1 − θ) a0 = 2sθ
b1 = pθ b2 = −2pθ b3 = 2pθ b4 = −pθ
b5 p(1 − theta) b6 = −2p(1 − θ) b7 = 2p(1 − θ) b8 = −p(1 − θ)
41
xk0 = b4 xk1 = b3 + a4 + a9 + a6 hki−1

xk2 = (1 + a0 ) xk3 = b2 + a3 + a9 + a5 hki+1

xk4 = b1 xk5 = −b8
xk6 = −(b7 + a7 + a2 ) xk7 = 1 − a8
xk8 = −(b6 + a7 + a1 ) xk9 = −b5
dki = xk5 hki−2 + xk6 hki−1 + xk7 hki + xk8 hki+1 + xk9 hki−2
then the implicit system can be written as
xn0 hk+1 n k+1 n k+1

i−2 + x1 hi−1 + x2 hi + xn3 hk+1 n k+1 n
i+1 + x4 hi+2 = dj (3.17)
Expressed in matrix form, a system of 8 unknowns will conform to

    
 • • •  h23   d23 − x10 h21 − x11 h22 
    
    
 •
 • • • 
 h24  
  d24 − x20 h22 

    
    
 •
 • • • • 
 h15  
  • 

  =  (3.18)
    

 • • • • • 

 h26  
  • 

    
    

 • • • • 

 h27  
  d27 − x54 h210 

    
    
• • • h28 d28 − x63 h29 − x64 h210
We solve this system using appropriate methods of solving systems of algebraic equations. The proce-
dure is then repeated for calculations of h at n = 2, 3, . . . , 10
3.5.3 Kinematic Wave Equation
Using the Preissmann implicit scheme, we approximate the derivatives and points as follows:
∂h 1 1
hk+1 k+1
hki+1 + hki

= i+1 + hi −
∂t 2∆t 2∆t
∂h α (1 − α) k
hk+1 − hk+1 hi+1 − hki

= +
∂x ∆x i+1 i
∆x
α k+1 (1 − α) k
hi+1 + hk+1 hi+1 + hki

hp = i +
2 2
42
By substituting these derivatives into the the KWM we have
c1 hk+1 + c2 hk+1 k+1 k+1

+ a5 hki+1 + a6 hki bhp + c3 hki + c4 hki+1 = 0

i i+1 + a3 hi+1 + a4 hi (3.19)
where
a1 = ∆x a3 = 2α∆t a5 = 2(1 − α)∆t
(3.20)
a2 = −∆x a4 = −2α∆t a6 = −2(1 − α)∆t
and
c1 = (a1 + aa4 ) c3 = (a2 + aa6 )
(3.21)
c2 = (a1 + aa3 ) c4 = (a2 + aa5 )
This is a nonlinear system which can be solved using the Newton Method.
We therefore find the Jacobian matrix associated with a system of equations of for example k = 1 : 6.
It is given by a lower triangular matrix

 
 J11 
 
 
 J J22 
 21 
 
 
J=
 J32 J33 
 (3.22)
 
 

 J43 J44 

 
 
J54 J55
The unknown vector X which contains the unknown variables is solved in the system
Jk ∆X = −Fk (3.23)
where    

 h22
  F1 h22 , h23 , h24 , h25 , h26 
   
   
 h2   F 2 2 2 2 2 
h2 , h3 , h4 , h5 , h6 
 3   2
   
   
X =  h4 
 2
; F=
 F3 2 2 2 2 2 
h2 , h3 , h4 , h5 , h6  (3.24)
   
   
 h2   F 2 2 2 2 2 
h2 , h3 , h4 , h5 , h6 
 5   4
   
   
2 2 2 2 2 2
h6 F5 h2 , h3 , h4 , h5 , h6
43
3.5.4 Saint Venant Equations
The numerical solution of the Saint Venant Equations is discretized as follows: First the
continuity equation is considered-
∂h ∂ (uh)
+ =0 (3.25)
∂t ∂x
Using the Preissmann scheme the partial derivative of a variable with respect to x and t
are given as follows
hk+1 k+1

∂h i+1 − hi hk − hki
= α + (1 − α) i+1 (3.26)
∂x ∆x ∆x
k+1 k+1

∂h 1 hi + hi+1 1 hki + hki+1
= − (3.27)
∂t 2 ∆t 2 ∆t
Substituting these finite difference into the continuity equation and arranging coefficients of similiar terms
we have
φ11 hk+1
i + φ12 uk+1
i + φ13 hk+1 k+1
i+1 + φ14 ui+1
+µ11 hki + µ12 uki + µ13 hki+1 + µ14 uki+1 = 0
where the introduced constants in the problem are defined as follows :
a1 = ∆x a3 = 2α∆t a5 = 2(1 − α)∆t

(3.28)
a2 = −∆x a4 = −2α∆t a6 = −2(1 − α)∆t
and
φ11 = a1 φ13 = a1 µ11 = a2 µ13 = a2
(3.29)
φ12 = a4 hk+1
i φ14 = a3 hk+1
i+1 µ12 = a6 hki µ14 = a5 hki+1
From the momentum equation, we have
∂u ∂u ∂h 1 u2
+u + + 2 2m − 1 = 0 (3.30)
∂t ∂x ∂x F0 h
44
Taking the derivatives in the equation and using the Preissmann scheme to discretize them and then
finally subsituting them into the momentum equation (3.30), we have
φ21 hk+1
i + φ22 uk+1
i + φ23 hk+1 k+1
i+1 + φ24 ui+1
+µ21 hki + µ22 uki + µ23 hki+1 + µ24 uki+1 + fi = 0
where the introduced constants are given by
φ21 = a4 h2m
p φ22 = (a1 + a4 up ) h2m
p φ23 = a3 h2m
p
φ24 = (a1 + a4 up ) h2m

p µ21 = a6 h2m
p µ22 = (a2 + a6 up ) h2m
p
(3.31)
1
µ23 = a5 h2m
p µ24 = (a2 + a5 up ) h2m
p a7 = F02
2∆t∆x
a8 = −2∆t∆x fi = a7 u2p + a8 h2m

p
The general scheme therefore is to solve the system of equations below
φ11 hk+1
i + φ12 uk+1
i + φ13 hk+1 k+1 k k k k
i+1 + φ14 ui+1 + µ11 hi + µ12 ui + µ13 hi+1 + µ14 ui+1 = 0
φ21 hk+1
i + φ22 uk+1
i + φ23 hk+1 k+1 k k k k
i+1 + φ24 ui+1 + µ21 hi + µ22 ui + µ23 hi+1 + µ24 ui+1 = −fi
In the system of equations, the nodal values of the water levels hk+1
i and hk+1
i+1 as well as the flow
velocities uik+1 and uk+1

i+1 are unknowns.
Given the way they appear, a pair of equations can be written for each space interval for i = 2, 3, . . . , m−
1 where m here is the number of nodes in the spatial computation. We then obtain a set of 2(m−1) algebraic
equations with 2m unknowns representing the nodal values of the function h and u. Two boundary
conditions allow us to complete the system of equations. for i = 2
δ0 hk+1
2 + (1 − δ0 )hk+1
2 = δ0 h1 (tn+1 ) + (1 − δ0 )u1 (tn+1 ) (3.32)
for i = m − 1
δL hk+1 k+1
m−1 + (1 − δL )hm−1 = δL hm (tn+1 ) + (1 − δL )um (tn+1 ) (3.33)
45
In the above equations δ0 and δL are integer values that can take values of 0 or 1. The value of 1
designates the boundary condition imposed in the form of water stage (function h1 (t) or hm (t)) whereas
the value of 0 corresponds to the flow velocity (function u1 (t) or um (t)). In this research, we use the water
stage condition
For example, given an equally spaced grid with 6 nodes, we start with the calculation of the time driven
unknowns hk+1
i , uk+1
i , hk+1 k+1
i+1 and ui+1 as follows:
for k = 1, for j = 2
φ11 h22 + φ12 u22 + φ13 h23 + φ14 u23 + µ11 h12 + µ12 u12 + µ13 h13 + µ14 u13 = 0
φ21 h22 + φ22 u22 + φ23 h23 + φ24 u23 + µ21 h12 + µ22 u12 + µ23 h13 + µ24 u13 = −f
for j = 3
for j = 4
There are a total of 6 equations with 8 unknowns i.e. h22 , u22 , h23 , u23 , h24 , u24 , h25 and u25 . In order
to resolve this we expand the equation at the boundary conditions using the water stage conditions i.e.
δ0 = 1 and δL = 0. This gives us the two additional equations to complete the solution. The additional
equations are given by
for j = 2; h22 = h21
for j = 5; h25 = h26
46
Adding the two boundary conditions, we have the system of equations in the form:
h22 − h21 = 0
h25 − h26 = 0
In matrix form, we can write the system of equations above as

  
 1 0 0 0 0 0 0 0  h22 
  
  

 φ11 φ12 φ13 φ14 0 0 0 0 
 u22 

  
  

 φ21 φ22 φ23 φ24 0 0 0 0 
 h23 

  
  

 0 0 φ11 φ12 φ13 φ14 0 0 
 u23 

  +
  

 0 0 φ21 φ22 φ23 φ24 0 0 
 h24 

  
  

 0 0 0 0 φ11 φ12 φ13 φ14 

 u24 

  
  

 0 0 0 0 φ21 φ22 φ23 φ24 

 h25 

  
  
0 0 0 0 0 0 1 0 u25
47
    
 0 0 0 0 0 0 0 0  h12   h12 
    
    

 µ11 µ12 µ13 µ14 0 0 0 0 
 u12  
  0 
    
    

 µ21 µ22 µ23 µ24 0 0 0 0 
 h13  
  −f 
    
    

 0 0 µ11 µ12 µ13 µ14 0 0 
 u13  
  0 
  = 
    

 0 0 µ21 µ22 µ23 µ24 0 0 
 h14  
  −f 
    
    

 0 0 0 0 µ11 µ12 µ13 µ14 

 u14  
  0 
    
    

 0 0 0 0 µ21 µ22 µ23 µ24 

 h15  
  −f 
    
    
0 0 0 0 0 0 0 0 u15 h26
We can alternatively represent the system by
G(X) = AX(k+1) + BX(k) + F = 0 (3.34)
The system G(X) is nonlinear in terms of the unknown vector Xk+1 . Therefore we implement an
iterative scheme. For this purpose we use the Newton method presented in earlier sections. The iteration
process has the following form:
J(k) ∆X(k+1) = −G(k) (3.35)
where
∆X(k+1) = X(k+1) − X(k) − correction vector
J(k) − Jacobian matrix
k − index of iteration
The Newton method requires that the system of linear algebraic equations is solved in each iteration.
To this purpose a modified Gauss elimination or LU decomposition algorithm should be used, which takes
advantage of the banded matrix.
48
3.6 Efficiency Criteria
There are many efficiency criteria for measuring the margin of error between observed
variables and predicted values by mathematical models. Whether the observed values are
real catchment or are in themselves variables from existing models, established efficiency
criteria can be used to compare the observed and predicted values.P. Krause and F. (2005)
Efficiency criteria are defined as mathematical measures of how well a model simulation
fits available observations Beven (2001). In this thesis, the researcher makes use of the
Coefficient of determination r2 , the Nash-Sutcliffe efficiency E, the index of agreement d,
Nash-Sutcliffe efficiency with logarithmic value ln E, Modified forms of E and d and Relative
efficiency criteria Erel and drel . The Coefficient of determination r2 is the squared of
value of the coeffcient of correlation and ranges between which means no correlation and
1 where is a perfect correlation between observed and predicted values. The Nash-
Sutcliffe, E measures performance for values between 1.0 (perfect fit) and −∞ using 1 minus
the sum of absolute squared difference between predicted and observed values. Both r2 and
E are insensitive to over- or underprediction. The efficiency criteria that overcomes the
sensitivity to over-underprediction is the index of agreement, d. It has a range similar
to r2 with 0 (no correlation) and 1 (perfect fit). The formulae for the different efficiency
criteria used in this research are defined mathematically in the below as: The following are
formula for efficiency criteria used in this thesis. The symbols O and P used to calculate
the different represnt the Observed values and Predicated values respectively.
49
3.6.0.1 Coefficient of Determination r2
 2
Pn
Oi − Ō P i − P̄
r2 =  qP i=1 qPn
 (3.36)
n
i=1 Oi − Ō i=1 Pi − P̄
3.6.0.2 Nash-Sutcliffe efficiency E

Pn
(Oi − Pi )2
E = 1 − Pni=1 2 (3.37)
i=1 Oi − Ō
3.6.0.3 Index of Agreement d

Pn 2
i=1 (Oi − Pi )
d = 1 − Pn (3.38)
Pi − Ō + Oi − Ō 2

i=1
3.6.0.4 Modified forms of E and d

Pn
|Oi − Pi |j
Ej = 1 − Pi=1
n
j with j ∈ N (3.39)
i=1 Oi − Ō

Pn j
i=1 |Oi − Pi |
dj = 1 − Pn with j ∈ N (3.40)
Pi − Ō + Oi − Ō j

i=1
3.6.0.5 Relative efficiency criteria Erel and drel

Pn Oi −Pi 2
i=1 Oi
Erel = 1 − P 2 (3.41)
n Oi −Ō
i=1 Ō
Pn Oi −Pi 2
i=1 Oi
drel = 1 − 2 (3.42)
Pn |Pi −Ō|+|Oi −Ō|
i=1 Ō
50
Chapter 4
Simulation Results and Discussion
4.1 Introduction
This chapter is a presentation of the simulation using the algorithms based on the implicit
Preissmann scheme that has been used to solve the normalized Saint Venant Equations
(SVE), the Burgers’ Equation Model (BEM), the Kinematic Wave Equation (KWM) and
the presented third order approximate model to the Saint Venant Equations. The al-
gorithms have been programmed using MATrix LABoratory (MATLAB 7.8.0 (R2007a))
together with the formulas used for the efficiency criteria.
Using graphical techniques, the researcher visually examines the approximate models
and the nature of solutions as compared to the SVE. The Graphical Technique provide a
visual comparison of simulated and measured constituent data and first overview of model
performance (ASCE, 1993)
Additionally, since the third order approximate model is an upgrade of the Burgers’
Equation model, both models are compared graphically. The two approximate models (The
BEM and the third order approximate model) are also evaluated using relative percentage
deviations from the SVE. The researcher investigates the effect of the weighted parameter
51
that is found in the Preissmann scheme and its variational effect on the hydraulic models
considered in this research. The limits of the flow depth difference (the difference between
the initial and final flow depth) within which each of the models discussed in this thesis
generate acceptable results are also investigated.
Finally, the researcher uses efficiency criteria to test the efficiency of the approximate
models as compared to how well they mimic the parent model. The efficiency criteria are
used to support the quantitative analysis.
4.1.1 Application
The hydraulic models presented in this research and the numerical algorithms provided
in Section (3.5 - 3.7) have been applied to a series of hypothetical examples, simulating
mainly the flow depth, h within considered distance and duration in a flow channel.
We consider the problem of a Dam Break and use the hydraulic models presented to
simulate flow in this occurrence. A dam break refers to any problem where a barrier
separating water at two different levels is suddenly removed. This can also be likened to
the experiment of the sudden opening of a sluice or gate. The main objective in a dam
break is to determine flow which results from a sudden destruction of the dam. The water
is assumed to be at rest on both sides of the dam initially. At t = 0, the dam breaks and
our problem is to determine the subsequent motion of the water for all x and t.
Fig. 4.1 and Fig4.2 are two situations that occur when water is staged at two different
levels. Suppose that a rectangular canal having a horizontal bed with an initial depth,
52
hinitial is contained by a lock or wall. If the lock gate or wall is suddenly removed, Fig. 1
displays the physical plane at time t for the propagation. The water level moves from the
initial depth, hinitial indicated in the diagram to a final depth, hf inal .
If however, the gate or wall is removed relatively further and faster (instantaneously),
the water level moves to the floor bed before it rises to the final depth, hf inal . The descrip-
tion is captured in Fig 4.2 on page 54.
Figure 4.1: physical plane at time t
53
Figure 4.2: physical plane at time t
Negative Surge Case
Consider a 50m horizontal channel in which water stays at rest with constant depth h(x, t) =
1.50m. At the boundary x = 0, the following function is given in terms of the flow depth,
h.


 1.50 t ≤ 0;

h(x = 0, t) = (4.1)

 1.00 t > 0

The experiment is done for 35 seconds within which the the spatial and duration nodes
are stretched on a numerical grid of 250 equal nodes. The value of the weighted parameter
for this experiment was θ = 0.75 and the Froude Number of initial flow was Fo = 0.549.
The condition is applied across all the models and the results of the simulation are
54
shown in Fig. 4.3 and Fig. 4.4. From the top of Fig 4.3 is the SVE and the down subfigure
is the BEM. Fig. 4.4 has the top subfigure as the KWM and the proposed model is at the
down subfigure on page 56.
Figure 4.3: The SVE and the BEM respectively
The hydrographs displayed in Fig 4.3 show distance (x) in the horizontal axis and the
predicted flow depth, h on the vertical axis with varying time, t in 10 equal intervals from
55
t = 10 to t = 120 units represented by the different colored straight and dotted lines.
Figure 4.4: The KWM and the third approximate model
The prediction of the SVE in Fig. 4.3 of the propagation through the channel shows
smooth diffused waves. The BEM for the same problem of initial flow depth of 1.5m and
final depth of 1.0m shows a generally close shape to the SVE with the same pattern of
diffusion. The nature of the presented third apprximate model to the SVE is very similar
56
to the propagation displayed by the BEM. Of the four models, the KWM displays evident
less diffused waves as expected because of the absence of the diffusive term in the model
that is present in the three other models (i.e. the SVE, BEM and the third approximate
model to the SVE).
The BEM solution produces less steeper waves compared to the SVE with mean flow
depth averagely 0.7% short of the output from the SVE. The KWM produces evidently
steeper waves compared with both the SVE and BEM. The average flow depth throughout
is very small (averagely 2.7% short of the SVE and 2.5% of the BEM).
4.2 Effect of Weighted Parameter
A number of experiments were carried out to determine the general effect of the weighted
parameter θ in the Preissmann scheme used in discretizing the models. The results are
presented in Fig 4.5, Fig. 4.6 and Fig. 4.7 for the SVE, BEM and KWM respectively. The
values of θ were varied as θ = 0.5, θ = 0.65 and θ = 1.00.
The results shows that as the weighted parameter increased, there is a general corre-
sponding increasing diffusive effect in the solution presented by all the models. The SVE
in Fig. 4.5 on page 58 for θ = 0.5 shows some level of distortions along the waves produced
representing the individual times especially at the initial flow depths and the final flow
depths. For the other weights greater than 0.5 (0.5 < θ ≥ 1), the distortions disappear
and there is the occurrence of smooth waves. The BEM in Fig. 4.6 page 59 shows smooth
diffused waves which are more diffused as θ increases.
57
Figure 4.5: θ = 0.5, θ = 0.65 and θ = 1.00 respectively for SVE
58
Figure 4.6: θ = 0.5, θ = 0.65 and θ = 1.00 respectively for BEM
59
Figure 4.7: θ = 0.5, θ = 0.65 and θ = 1.00 respectively for KWM
60
Mean (h1 − h0 ) SVE BEM KWM New Model
0.1 6.209053 0.447079 0.390647 0.457752
0.2 6.250272 0.468406 0.378322 0.469059
0.3 6.347221 0.447802 0.37436 0.463257
0.5 6.389163 0.447907 0.382700 0.458149
0.8 6.501417 0.454828 0.378633 0.460398
1.2 6.737718 0.454211 0.394196 0.459429
1.5 6.787165 0.459494 0.363316 0.466864
2.0 7.209245 0.454090 0.361972 0.495859
Table 4.1: Runtime in seconds of the SVE, BEM, KWM and the third order model
The same pattern is given off in the steep-natured solutions by the KWM. These steep
waves in Fig. 4.7 on page 60 becomes slightly diffused for increasing values of the weighted
parameter.
4.2.1 Runtime and Limits of Flow Depth differences
Table 4.1 on page 61 is the summary of the experiment of determining the runtime (mea-
sured in seconds) for the simulation of the hydraulic models for different mean heights.
The runtime is recorded in seconds it takes Matlab to simulate the models.
For each mean difference in height (h1 − h0 ), 15 different experiments were considered.
Therefore the runtime for the difference in flow depth 0.1 for the SVE in Table 4.1 on page
61
61 given as 6.209053 seconds represents the average for 15 different flow depth difference
of 0.1.
This investigation was conducted using a uniform grid of dimension 140 × 140 for the
spatial and duration domains. The researcher observed that for every 100 node increment,
the runtime for the SVE increased by averagely 364%, the runtime for the BEM increased
by averagely 125% and the runtime for the KWM also increase by 206%. The presented
third order approximate model to the SVE also has a runtime increment of averagely 112 %
for every 100 node increment. There is a general increment in the runtime for the SVE and
the KWM as the difference in between the initial flow depth and final flow depth increases.
This observation is not so evident in the case of the BEM and the presented third order
approximate model of the SVE.
4.2.2 Limits of the Flow Depth Differences
As one of the specific objectives of this thesis, the research seeks to determine the limits
(minimum and maximum) range of the flow depth difference (between the final flow depth,
h0 and the initial flow depth, h1 ) for which the parent model together with its approximate
models provide acceptable results. This is done by first considering how small(minimum)
the flow depth difference (h1 − h0 ) can be in order for acceptable results to be obtained
and then how large(maximum) difference between the initial and final flow depth can be
in order to obtain acceptable results for the independent models.
The SVE when simulated with a final flow depth of h = 1.0m has a minimum range of
62
up to h = 1.06m and a maximum range of up to h = 3.97m within which acceptable results
are attainable. Any specified value outside this range results in unacceptable values for the
simulation. When the SVE is simulated with a final flow depth of h = 2.0m however, there
is a minimum range of h = 2.009m and a maximum range of up to h = 7.3m within which
acceptable results are given. A simulation of the SVE with a final flow depth of h = 5.0m
has a minimum of h = 5.01 and a maximum range of up to h = 16.2m. For values higher
than h = 5.0m however, there is a similar trend of the minimum range within 2% of the
specified flow depth and a maximum of up to 3 times the specified flow depth.
Interestingly noted was the fact that there seemed to be no maximum range for which
the KWM can be simulated. The results as the range is increased is a melon-like structure
as shown in Fig. 4.8.
Figure 4.8: Simulation of KWM with extremely high flow depth differences
The Kinematic Wave equation when simulated using the final flow depth of h = 1.0m
63
has the minimum range of up to h = 1.06 for the generation of acceptable results. Similarly,
when simulated using a final flow depth of h = 2.0m and h = 5.0m, the minimum range
results of up to h = 2.06 and h = 5.09m respectively for acceptable results.
The BEM and the third order approximate model to the SVE also deliver acceptable
results irrespective of extremely small minimum values, for example specifying a final flow
depth of h = 1.0m results in a minimum of as low as h = 1.00000000009m for which
acceptable results are generated for any specified value within. This holds for the BEM
and the third order approximate model to the SVE.
4.3 Quantitative Comparison
This subsection provides experiments on the point by point prediction of the approximate
models (The KWM, the BEM and the presented third order approximate model to the SVE)
in comparison to the Saint Venant equations (SVE). There is the use of graphical techniques
for comparison supported with relative percentage errors that measure the deviations of
the solutions of the considered approximate models to the SVE.
The BEM is compared to the SVE and the KWM is also compared to the SVE by
examining exactly the quantum of errors generated by using these models in place of the
SVE. The BEM is then placed on the same grid with the third order approximate model.
Finally, all the approximate models are compared to the parent model, the SVE by first
considering the BEM with the third order approximate model and then the BEM along
with the KWM also on the same scale.
64
4.3.1 The SVE and the BEM
The figure below (Fig 4.9) shows the hydrograph for the SVE (— straight line) and the
BEM (- - - dashes) on a 250 × 250 grid structure for an initial flow depth of h = 1.5m and
a final flow depth of h = 1.0m.
Figure 4.9: Distance-time graph for SVE versus BEM
Fig 4.9 shows that the BEM simulates well the results shown by the SVE. It results
in producing almost an exact mimic to the SVE. At t = 20, the prediction of the BEM is
0.1% lower than the SVE between a mean distance of about 15m. It then predicts exactly
as the SVE and then starts to overpredict after about 15m.
At t = 20 and beyond however, the BEM starts with over predicting the SVE estimates
as low as 0.09% and then predicts exactly as SVE along the channel for a mean distance of
5.0m and then overpredicts the SVE after that. This overprediction spans till the end of the
channel. The minimum percentage underprediction of the BEM throughout the channel
65
is 1.5%. The maximum pecentage overprediction is 2.9%. To demonstrate the consistent
nature of the errors generated the another problem of grid structure 200 × 200 is simulated.
The percentage errors given throughout the channel for time steps of 20 units is given in
Fig 4.10 for the specified simulation.
Figure 4.10: Distance-time percentage error plot for the SVE and the BEM
66
4.3.2 The KWM and the SVE
Figure 4.11: Distance-time graph for SVE versus KWM
Fig. 4.11 is the 2-D plot of the KWM and the SVE propagation prediction throughout
the channel for selected time values. It starts from t = 20 to t = 120. The way the KWM
predicts propagation is very different from the BEM. It constantly over predicts from the
start of the channel after the dam breaks and then equals the SVE for a relatively short
distance of 2m. For most time calculations, however the intervals for which the predictions
of the KWM equal the SVE are points along the channel after which which it under predicts
the SVE. The minimum over prediction given by the KWM relative to the SVE is 5.0 %
whiles the maximum over prediction is up to 30%.
Fig. 4.12 on page 68 is a percentage-distance plot indicating the errors of the KWM
relative to the SVE. This indicates the high quantum by which the KWM under predicts
the SVE.
67
For the purposes of consistency, a different structure is considered being 200 × 200. Fig
4.12 is therefore the results for the relative percentage errors generated for the problem of
initial flow depth being h = 1.5m and a final flow depth of h = 1.0m
Figure 4.12: Distance-time percentage error plot for the SVE and the BEM
68
4.3.3 The Third Order Approximate Model and the SVE
The figure below (Fig 4.913) shows the hydrograph for the SVE (— straight line) and the
third order approximate model to the SVE (- - - dashes) on a 250 × 250 grid structure for
an initial flow depth of h = 1.5m and a final flow depth of h = 1.0m. This look very similar
to the simulation of the BEM.
Figure 4.13: Distance-time graph for the SVE and the third order model
The relative percentage errors demonstrate the difference between the third order ap-
proximate model to the SVE and the results obtained from the BEM. Fig 4.14 on page
70 shows that the maximum over prediction tapper just around 1% with the maximum
relative percentage deviation almost 2.5%. Fig. 4.14 is 200 × 200 grid scale demonstration
of the relative percentage error comparison of the third order approximate model to the
SVE.
69
Figure 4.14: Percentage errors for the third approximate model to the SVE
4.3.4 The BEM and the Proposed Approximate Model
Fig. 4.15 shows the comparative hydrograph of the simulation by the BEM and the third
order approximate model on the same grid. The effect of the third order derivative addition
shows a smooth slight over prediction by the third order model away from the initial flow
depth h = 1.5m. When the flow depth averages h = 1.474m, the two models (The BEM and
the third order approximate model) present exactly the same solution for varying intervals.
This pattern is consistent throughout the propagation. The rest of the propagation however
shows that the third order derivative model under predicts what the BEM model displays.
Towards the end of the channel, averagely when h = 1.01m, the two models predict the
same results.
70
Figure 4.15: Distance-time graph for the BEM and the third order model
4.3.5 The SVE, BEM and the Proposed Approximate Model
To access how well the third order approximate model mimics the results by the SVE
compared to the dominance of the BEM, the three models (the third order approximate
model (- - -), the BEM (**) and the SVE(—)) are examined on the same structure. Fig
4.16 shows the hydrograph displaying a 250 node solution of water staged at an initial flow
depth of h = 1.5m with a final flow depth of h = 1.0m
From the nature of the results shown in Fig 4.16, the advantage of the third order
approximate model can be established. Apart from the relative error comparison demon-
strating the minimum over prediction by the third order model, it is also observed that
the two approximate models equal the SVE at different points along the propagation. The
BEM solution equals the SVE at averagely h = 1.37m. The presented third order model on
the other hand equals the SVE at past midway averagely when the flow depth, h = 1.19m
71
Figure 4.16: Distance-time SVE, BEM and third order model
The comparison of the SVE, BEM and KWM show similar patterns by the fact that
the approximate model equals the parent model at different points. The plot of the three
models for the same problem is given in Fig. 15.
72
Figure 4.17: Distance-time SVE, BEM and KWM
4.4 Positive Surge
A positive surge results from a sudden change in flow that increases the depth. It is an
abrupt wave front. The unsteady flow conditions may be solved as a quasi-steady flow
situation (Chanson (2004)).
This can result when a dam breaks from the downstream point. The wave therefore
rises from the lower initial point and steadily rises to a higher final flow depth.
Consider a channel that has water stage and is released at an initial constant depth
h(x, t) = 1.0m. At the boundary x = 0, the following function is given in terms of the flow
depth, h.
The dynamics of the SVE, BEM and KWM are given in Fig. 4.18 respectively.
73
Figure 4.18: Increasing time nodes t10 ⇒ tn for SVE, BEM and KWM respectively
74
4.5 Efficiency Criteria
The efficiency criteria for hydraulic models discussed in Chapter 3 is evluated for the
models. Since there is a comparison of the two approximate models against the SVE, the
observed values Oi are the values of the output from the SVE for any given simulation
while the predicted value, Pi are those of either the BEM, the third order model or the
KWM.
4.5.0.1 The BEM and SVE
A channel of length 50 discretized into 200 equal nodes is considered. The solution for
the dynamics of the flow depth is therefore in a matrix of dimension 200 × 200. The
progression of the prediction by the BEM is considered first relative to the SVE. The plots
of the various efficiency criteria is calculated from the initial point to half the channel (i.e.
t = 0 − t = 100), then from (t = 0 − t = 150),from t = 0 − t = 170) and then from for the
overall channel is considered (t = 0 − t = 200).
Fig. 4.19 and 4.20 are two dimensional graphs that represent the same progression of a
wave within a channel of length 50m discretized into 200 nodes. The graphs are presented
in 2 × 2 array. In explaining, we make use of the notation fij to mean the ith row graph
and the jth column graph.
As seen in f11 and f12 , Fig 4.19 shows the same progression from the initial point to
halfway the channel for the different discussed efficiency criteria. The prediction by the
BEM is a bit distorted for almost all criteria up to t = 10.
75
Figure 4.19: Efficiency Criteria for the prediction by BEM
76
Figure 4.20: Efficiency Criteria for the prediction by BEM
77
After t = 10 however, the prediction is stable, averagely 0.99 for all criteria till halfway
the channel. r2 = 0.9989, E = 0.9985,d = 1.000, ln E = 0.999, Ej = 0.9605,dj = 0.9649,
Erel = 0.9942 and drel = 0.9979.
f21 and f22 from Fig 4.19 displays the prediction from the initial up to t = 150. It is
observed that the models starts to under predict will all measures starting to drop except
for ln E which remains close to 1.
After t = 150, there is a massive change in all measures including ln E with dj responding
very sensitively to the difference in SVE and BEM values. Towards the end of the channel,
the coefficient records low fit between the SVE and BEM. (i.e. r2 = 0.0278 with E = −39.87
and dj = 0.0526). This is shown in Fig 4.20.
4.5.0.2 The KWM and SVE
The efficiency criteria is again used to investigate the predictive performance of the KWM
relative to the SVE for the same problem with grid dimension of 200 × 200.
The graphs are constructed in progression from initial to halfway (t = 0 − t = 200),
then from t = 0 − t = 150, t = 0 − t = 170 and then from t = 0 − t = 200.
f21 and f22 of Fig 4.21 displays (t = 0 − t = 100). There is a general decline in the
measures except for ln E which remains equal to 1. There is however almost no distortion
to initial from from t = 0 − t = 10. The prediction gives a range or 0.9-0.5 averagely for
the efficiency measures. f11 and f12 of Fig 4.22 displays a general decline in the measures
78
Figure 4.21: Efficiency Criteria for the prediction by KWM
79
Figure 4.22: Efficiency Criteria for the prediction by KWM
80
but very steep. Ej and Erel measures as far as -7.62 and -9.1 respectively.
f11 and f12 of Fig 4.22 shows the overall prediction of the KWM for the SVE. Towards
the end of the channel specifically between t = 195 to t = 200, there are very low measures
ranges as follows: r2 = 0.0002 − 0.30, E = −198.9121 − (−2.3164),d = 0.0195 − 0.2212,
ln E = 0.83 − 0.99, Ej = −98.97 − (2.422),dj = 0.0099 − 0.1117, Erel = −199.9129 −
(−0.0632) and drel = 0.0446 − 0.3754.
4.5.0.3 The Third Order Model and SVE
The efficiency criteria is finally used to investigate the predictive performance of the third
order approximate model relative to the SVE for the same problem with grid dimension of
200 × 200.
f11 and f12 of Fig 4.23 displays (t = 0 − t = 100). There is a large correspondence
between the values here the BEM. For almost all the efficiency measures except Er el, there
is recording of between 0.9-1.0.
f21 and f22 of Fig 4.24 shows the overall prediction of the third order approximate
model relative to the SVE. Towards the end of the channel specifically between t = 196 to
t = 200, there are very low measures ranges as follows: r2 = 0.082 − 0.30, E = −162.9121 −
(−2.3164),d = 0.0385 − 0.8, ln E = 0.91 − 0.99, Ej = −40.61 − (1.54),dj = 0.0239 − 0.1133,
Erel = −213.89 − (−0.1632) and drel = 0.0556 − 0.5783.
81
Figure 4.23: Efficiency Criteria for the prediction by the Third Order Model
82
Figure 4.24: Efficiency Criteria for the prediction by the Third Oder Model
83
Chapter 5
Conclusion and Recommendations
5.1 Introduction
The key objectives of this research were to consider the Saint Venant Equations (SVE) and
its approximate models i.e. the Burgers Equation Model (BEM) and the Kinematic Wave
Equation (KWM) for which are second order and first order approximate respectively.
In addition to this, a third order approximate model is proposed and presented in this
thesis. The third order approximate model to the SVE and the two existent approximate
models are therefore quantitatively analyzed with the inclusion of graphical techniques.
The approximate models together with the parent model are solved using finite differences
(specifically the Preissmann scheme).
The researcher uses MATLAB to program the algorithms to simulate the models.The
quantitative analysis is performed using percentage errors and then using efficiency criteria
for hydraulic models. Also considered was the range within which acceptable results are
generated for all the individual models.
84
5.2 Conclusion
The conclusions in this research are summarized as follows:
• The estimated third order approximate model to the SVE has preserved forms of
the constant coefficients of the KWM and the BEM for the first and second orders
respectively
• The estimated constant of the third derivative has a magnitude of
Fo2

4
1 − Fo2
3 9
• The third order approximate model equals the SVE at a lower level consistently
compared to the BEM at a higher flow depth level
• The third order approximate model is most efficient model quantitatively taking away
other advantages of the BEM
• The KWM has a minimum range of simulation but can be used to simulate problems
with extremely wide flow depth differences.
• The weighted parameter within the box scheme has a positive relationship with the
diffusive nature of the waves. An increment in the weighted parameter increases the
diffusive nature of all the models considered in this research.
• The BEM in its given form simulates well the SVE with far less computional cost
of time of as low as 3,000% less time than it takes to run one simulation for the
85
SVE. The result is based on implicit schemes employed to solve both equations and
therefore has established an insurance of stability. The range of over prediction and
under prediction for the BEM relative to the SVE is 2.34% and 1.5% respectively.
• The KWM therefore apart from it’s steep wave producing form provides no advantage
over the BEM relative to the SVE. The BEM with its extremely low runtime can well
be used in place of the SVE.
5.3 Recommendation
The researcher recommends an investigation into the distortion that occurs at the end of
the prediction by the KWM. This qualitative observation also occurs in implicit schemes
for all the models considered when the problem is solved in the region of dt > dx.
This occurrence may account for the instantaneous movement of the dam wall in the
Dam Break problem.
The third order approximate model can be used together with the BEM since they
predict equal to the SVE at a lower and higher flow depth point respectively
86
Bibliography
Akan, A. O., & Yen, B. C. (1977). A nonlinear diffusion-wave model for unsteady
open-channel flow. Proc., 17th Congress, Int. Association for Hydraulic Research,
2 , 181190.
Beven, J. K. (2001). Rainfall-runoff modelling. The Primer, John Wiley and Sons Ltd,
Chichester , 319.
Burgers, J. M. (1950). Correlation problems in a one-dimensional model of turbulence.
Proc., Royal Netherlands Academy of Science, Amsterdam, 53 , 247-260.
Chanson, H. (2004). Environmental hydraulics of open channel flows. Elsevier Butterworth-
Heinemann, Oxford, UK , 483.
Kubo, N., & Shimura, H. (1999). Theoretical analysis of transient phenomena of con-
veyance waves with small froude number. Trans. Jpn. Soc. Irrig. Drainage Reclama-
tion Eng.(In Japanese), 138 , 25-36.
Lighthill, M. J., & Whitham, G. B. (1955). On kinematic waves. i: Flood movement in
long rivers. Proc. R. Soc. London, Ser. A, 229 , 281316.
Malaterre, P. O., & Baume, J. P. (1998). Modeling and regulation of irrigation canals: Ex-
isting applications and ongoing researches. IEEE international conference on systems,
man and cybernetics, San Diego, CA, 3850-3855.
Nelz, S., & Pender, G. (2009). Desktop review of 2d hydraulic modelling packages. En-
87
vironment Agency, Rio House, Waterside Drive, Aztec West, Almondsbury, Bristol,
BS32 4UD.
Odai, K., Onizuka, & Osato. (2006). Analytical solution of the burgers equation for simu-
lating translatory waves in conveyance channels. Journal of Hydraulic Engineering,
194-199.
Odai, S. N. (1999). Nonlinear kinematic-wave model for predicting open-channel flow rate.
J. Hydraul. Eng, 125 (8), 886889.
Onizuka, K., & Odai, S. N. (1998). Burgers equation model for unsteady flow in open
channels. J. Hydraul. Eng, 124 (5), 509512.
P. Krause, D. B., & F., B. (2005). Comparison of different efficiency criteria for hydrological
model assessment. Advances in Geosciences, 5 , 89-97.
Price, R. K. (1994). Flood routing models. computer modeling of freesurface and pres-
surized flows, m. h. chaudhry and l. w. mays, eds. Kluwer Academic Publishers,
Dordrecht, The Netherlands, 375-407.
Sleigh, P. A., & Goodwill, I. M. (2000). The saint venant equations.
Su Ki Ooi, E. W., M.P.M Krutzen. (2005). On physical and data driven modeling of
irrigation channels. Control Eng. Practice, 461-471.
Szymkiewicz, R. (2010). Numerical modeling in open channel hydraulics numerical inte-
gration of the saint venant equations faculty of civil and environmental engineering.
Gda sk University of Technology, 310-375.
Tsai, C. W. S. (2003). Applicability of kinematic, noninertia, and quasisteady dynamic
88
wave models to unsteady flow routing. Journal of Hydraulic Engineering, 129(8),
613-627.
Whitham, G. B. (1974). Linear and nonlinear waves. Wiley, New York , 45-50.
Yabe, T., & Aoki, T. (1991). A universal solver for hyperbolic equations by cubic-
interpolation. i. one-dimensional solver. Comput. Phys. Commun., 66 , 219232.
89

George Parker Lamptey PDF

Uploaded by

Copyright:

Available Formats

George Parker Lamptey 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

George Parker Lamptey PDF

Uploaded by

Copyright:

Available Formats

QUANTITATIVE ANALYSIS OF APPROXIMATE MODELS

TO THE SAINT VENANT EQUATIONS

GEORGE PARKER-LAMPTEY, B.Sc.

Presented to the Department of Mathematics,

Kwame Nkrumah University of Science and Technology

in Partial Fulfillment of the Requirements for the Degree of

George Parker-Lamptey(PG5072510) ..................... ..................

Student Signature Date

Prof. S. N. Odai ..................... ..................

Supervisor Signature Date

Prof. I. K. Dontwi ..................... ..................

Supervisor Signature Date

Mr. K. K. Darkwah ..................... ..................

Head of Department Signature Date

to my mother, grandmother and two siblings

Also to the special individuals who have been a blessing to me

Naa Adumoyoobi, Whelma and Nana Ama,

I am extremely grateful to my two supervisors, Prof. I. K. Dontwi and Prof. S. N. Odai

To Dr. Peter Amoako-Yirenkyi, Mr. K. Darkwa, Mr. J. A. Ackora-Prah and all

support. God bless you.

1.1 Background Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.5 Justification of Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.6 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 Channel Routing Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.3 The Saint Venant Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.5 The Burgers’ Equation Model . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.6 The Proposed Third Order Approximate Model . . . . . . . . . . . . . . . 17

2.6.1 Estimation of Model Constants . . . . . . . . . . . . . . . . . . . . 17

3.1 Numerical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.1.1 Efficiency of a Numerical Scheme . . . . . . . . . . . . . . . . . . . 24

3.1.2 Explicit and Implicit Formulations . . . . . . . . . . . . . . . . . . 26

3.2 Finite Difference Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.2.1 Application of Finite Difference Schemes to PDEs . . . . . . . . . . 30

3.3 Numerical Scheme for the Mathematical Models . . . . . . . . . . . . . . . 31

3.3.1 Preissmann Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.4 Solution to Systems of Algebraic Equations . . . . . . . . . . . . . . . . . . 34

3.4.1 Linear Systems of Equations . . . . . . . . . . . . . . . . . . . . . . 34

3.4.2 Solution to Nonlinear System of Equations . . . . . . . . . . . . . . 35

3.5 Numerical Solution of Models . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.5.1 Burgers’ Equation Model . . . . . . . . . . . . . . . . . . . . . . . . 39

3.5.2 The Proposed Equation . . . . . . . . . . . . . . . . . . . . . . . . 40

3.5.3 Kinematic Wave Equation . . . . . . . . . . . . . . . . . . . . . . . 42

3.5.4 Saint Venant Equations . . . . . . . . . . . . . . . . . . . . . . . . 44

3.6 Efficiency Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3.6.0.2 Nash-Sutcliffe efficiency E . . . . . . . . . . . . . . . . . . 50

3.6.0.3 Index of Agreement d . . . . . . . . . . . . . . . . . . . . 50

3.6.0.4 Modified forms of E and d . . . . . . . . . . . . . . . . . . 50

3.6.0.5 Relative efficiency criteria Erel and drel . . . . . . . . . . . 50

4 Simulation Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4.2 Effect of Weighted Parameter . . . . . . . . . . . . . . . . . . . . . . . . . 57

4.2.1 Runtime and Limits of Flow Depth differences . . . . . . . . . . . . 61

4.2.2 Limits of the Flow Depth Differences . . . . . . . . . . . . . . . . . 62

4.3 Quantitative Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4.3.1 The SVE and the BEM . . . . . . . . . . . . . . . . . . . . . . . . 65

4.3.2 The KWM and the SVE . . . . . . . . . . . . . . . . . . . . . . . . 67

4.3.3 The Third Order Approximate Model and the SVE . . . . . . . . . 69

4.3.4 The BEM and the Proposed Approximate Model . . . . . . . . . . 70

4.3.5 The SVE, BEM and the Proposed Approximate Model . . . . . . . 71