Masters2 88277

Model based multi-objective
evaluation of BMP system

configurations for CSO reduction
Case study: Montevideo, Uruguay
Alida Alves Beloqui

MSc Thesis MWI UWEM 2014-43
April 2014
Model based multi-objective evaluation of BMP system
configurations for CSO reduction
Master of Science Thesis

by
Alida Alves Beloqui
Supervisors
Prof. Damir Brdjanovic Ph.D., M.Sc. (UNESCO-IHE)
Prof. Mukand S. Babel Ph.D., M.Sc. (AIT)
Mentors
Solomon Seyoum Ph.D., M.Sc. (UNESCO-IHE)
Arlex Sanchez Torres Ph.D., M.Sc. (UNESCO-IHE)
Examination committee
Prof. Damir Brdjanovic Ph.D., M.Sc. (UNESCO-IHE)
Prof. Mukand S. Babel Ph.D., M.Sc. (AIT)
Solomon Seyoum Ph.D., M.Sc. (UNESCO-IHE)
Arlex Sanchez Torres Ph.D., M.Sc. (UNESCO-IHE)
Assela Pathirana Ph.D., M.Sc. (UNESCO-IHE)
Scholarship donor
Joint Japan World Bank Group Graduate Scholarship Program (JJWBGSP)
This research is done for the partial fulfilment of requirements for the Master of Science degree at the
UNESCO-IHE Institute for Water Education, Delft, The Netherlands and the Asian Institute of Technology, Thailand.
Delft
April 2014
[Optional]
©2014by Alida Alves Beloqui. All rights reserved. No part of this publication or the information contained
herein may be reproduced, stored in a retrieval system, or transmitted in any form or by any means,
electronic, mechanical, by photocopying, recording or otherwise, without the prior permission of the
author. Although the author and UNESCO-IHE Institute for Water Education have made every effort to
ensure that the information in this thesis was correct at press time, the author and UNESCO-IHE do not
assume and hereby disclaim any liability to any party for any loss, damage, or disruption caused by errors
or omissions, whether such errors or omissions result from negligence, accident, or any other cause.
Abstract
Recent research suggests future alterations in rainfall patterns due to climate change that will be affected by
urbanization. These variations in patterns will increase rainfall intensities, affecting public safety and health
in urban areas. Moreover, urban growth is one of the most important drivers of change in the current
century. Traditional approaches for drainage solutions are based on an efficient collection and conveyance
of runoff. This approach offers expensive solutions and low flexibility for future expansions; therefore, low
flexibility for adaptation in front of a changing and uncertain future. New tendencies of urban water
management are focused on decentralised approaches in a long-term planning framework, including the use
of Best Management Practices (BMPs) for stormwater management. These new trends offer sustainable
solutions in an uncertain future scenario under climate change and urban dynamics by adding capacity and
flexibility to drainage systems. Although favourable reasons to use decentralised technologies are
recognized, several implementation obstacles are still identified. Models are useful tools in planning
processes, considering the difficulties experienced during decision making due to the complexity of
drainage systems. This work aims to find a methodology to evaluate different BMP configurations for
retrofitting urban drainage systems; the methodology is model based and uses the storm water management
model (SWMM) and genetic algorithms (GA) for optimization. The proposed approach was applied to
model a highly urbanized watershed in the city of Montevideo, to evaluate the effect of BMPs on runoff
reduction and Combined Sewer Overflows (CSO) discharges. Optimization methods were used to design
solutions considering multiple objectives at a watershed level. Comparisons of impacts between
centralized, decentralized and combined solutions were developed. The proposed approach was proved to
be useful for selection, design and effects assessment of applying BMPs in the studied area. The main
results show the effectiveness of these practices in runoff and CSO reduction and the advantage of
combining centralized and decentralized measures in drainage systems to reach multiple objectives. The
results also highlight the importance of a wide evaluation when assessing BMPs effects including pollution
reduction and liveability improvements; as well as the importance of applying optimization process in
designing to enrich the decision making process.
Keywords: urban drainage systems, best management practices, stormwater management, hydrodynamic
models, multi-objective optimization, combined sewer overflows and runoff reduction.
vii
Acknowledgements
I want to acknowledge my sponsor, the Joint Japan World Bank Group Graduate Scholarship Program for
giving me the opportunity of following this Programme.
I would like to thank my supervisors Prof. Damir Brdjanovic and Prof. Mukand Babel for their useful
recommendations each time that I have presented this work to them. This helped me to improve the global
vision of my research.
I want to acknowledge my mentor, Dr. Solomon Seyoum for guiding me during the period of this research.
Also, I would like to express my gratitude to my co-mentor Dr. Arlex Sanchez for the enthusiasm showed
on this topic, it gave me the necessary motivation to cope with an enormous quantity of work in a short
period of time. I have learnt a lot in this process and enjoyed working in this group. The many discussions
we had were extremely enriching for the development of this study.
My gratitude is also to Dr. Zoran Vojinovic for his valuable help, especially during the topic definition and
proposal development.
This research would not have been possible without the contribution of the Sewerage and Drainage Service
of Montevideo's Municipality's team, who facilitated all the needed information and assisted me during the
case study implementation.
Further, I want to recognize Santiago Urrestarazu due to all the explanation about his work and advices, it
was very helpful for me.
My appreciation is also to all the friends and classmates that I met in this period for their companionship,
they transformed this 20 months in an amazing time. In particular, I want to thank Weeraya and Jovana for
making me feel at home, in Thailand at the beginning and in Europe later.
Finally, I would like to thank my family for their wonderful encouragement along this period and for being
very close despite the distance. Pablo, it is not easy to express my gratitude for your inestimable help and
unconditional support, always.
viii
Table of Contents
Abstract vii
Acknowledgements viii
List of Figures xi
List of Tables xiii
Abbreviations xiv
1. Introduction 1
1.1. Background 1
1.2. Problem definition 2
1.3. Research questions 4
1.4. Hypothesis 4
1.5. Research objectives 5
1.6. Thesis outline 5
2. Literature review 6
2.1. Best Management Practices 6
2.1.1. Definitions, benefits and limitations. 6
2.1.2. Review of different BMPs 7
2.1.3. BMPs effectiveness 10
2.1.4. BMPs selection 14
2.2. Modelling BMPs with SWMM 15
2.3. Optimization of urban drainage systems 15
2.4. Rainfall selection 17
2.5. Discussion 18
3. Methodology 19
3.1. Research approach 19
3.2. Materials and methods 21
3.2.1. Tools used 21
4. Case of study 22
4.1.1. Hydraulic model adjustment 22
4.1.2. Description of the drainage system 24
4.1.3. Rainfall pattern 26
4.1.4. Hydraulic model description 27
5. Data analysis and model set up 29

5.1. Data analysis 29
5.1.1. Data collection 29
5.1.2. Physical data processing 30
5.1.3. Rainfall data processing 32
5.2. Model set up 34
ix
5.2.1. Hydraulic model adjustment 34
5.2.2. Hydrological model addition 37
5.2.3. Calibration 38
5.3. Design storm selection 40
5.3.1. Rainfall-CSO analysis 40
5.3.2. Design storm calculation 43
5.3.3. Evaporation conditions 44
6. Best Management Practices 45

6.1. BMPs selection 45
6.2. BMPs design and costs 48
6.2.1. Pervious pavements 48
6.2.2. Infiltration trenches 49
6.2.3. Green roofs and storages 51
6.2.4. Total present value 51
6.3. BMPs representation in SWMM 51
6.4. BMPs application 53
7. Optimization 57
7.1. Optimization process 57
7.2. Optimization variables 58
7.3. Objective functions 59
7.4. Scenarios to be optimized 61
8. Results and discussion 64

8.1. Comparison of individual measures 64
8.1.1. Design storm simulation 64
8.1.2. Continuous simulation 65
8.1.3. Optimization results 67
8.1.4. Spatial distribution of optimal solutions 70
8.2. Comparison of scenarios 4, 5 and 6 73
8.2.1. Design storm simulation 73
8.2.2. Continuous simulation 74
8.2.3. Optimization results 76
8.2.4. Spatial distribution of optimal solutions 80
8.3. Overview 83
9. Conclusions and recommendations 85

9.1. Conclusions 85
9.2. Recommendations 86
References 88
Appendix A - Rainfall-runoff effect and IDF calculation. 91
Appendix B - Model calibration results. 92
Appendix C - Rainfall-Runoff analysis. 97
Appendix D - BMPs analysis (USEPA 2013b) 99
Appendix E - BMPs applicability analysis. 111
Appendix F - Total present value calculation 116
Appendix G - Continuous simulation results 117
x
List of Figures
Figure 1.1 Impacts of different levels of urbanization, source: (Ellis et al. 2006). .................................... 2
Figure 1.2 Interrelations between the different levels of problems............................................................ 4
Figure 2.1 Runoff hydrographs from five storms simulating different solutions (Damodaram et al.
2010)....................................................................................................................................... 13
Figure 3.1 Different phases of research methodology. ............................................................................ 20
Figure 4.1 Areas of Montevideo, costal sewerage system and Pocitos watershed. .................... 23
Figure 4.2 Watershed for application of the study case (source Google Earth). Orange line
represents the watershed border. Pocitos beach and CSO discharge point are
indicated. ............................................................................................................................... 23
Figure 4.3 Pocitos watershed divided in 14 sub areas according to MM division. ....................... 25
Figure 4.4 Main CSO discharge channel. ........................................................................................... 25
Figure 4.5 Watershed outfall point (green: pluvial pipes; blue: combined pipes; violet:
discharge conduits). (Source Google Earth and MM data base). ................................ 26
Figure 4.6 Annual rainfall distribution in two different meteorological stations.............................. 26
Figure 4.7 Rainfall event used by MM in the model (mm/h - h). ............................................................ 27
Figure 4.8 DWF effect in a runoff hydrograph for 10 years return period rainfall. ................................. 28
Figure 5.1 High resolution picture with streets (left) and contour lines each 2 m (right) for Pocitos
watershed. ............................................................................................................................... 30
Figure 5.2 Result of MLC method (left) and hillshade for slope processing (right) in subwatershed
AYAC..................................................................................................................................... 31
Figure 5.3 IDF curves. ............................................................................................................................. 34
Figure 5.4 Models of total coastal system and Pocitos watershed. .......................................................... 35
Figure 5.5 Hydrographs from subwatershed located upstream Pocitos, modelling total system (grey)
and trough hydrological model (black)................................................................................... 36
Figure 5.6 Real sewerage network (left) and simplified modelled network (right) in Pocitos
watershed. ............................................................................................................................... 36
Figure 5.7 Final hydrodynamic model of Pocitos watershed. .................................................................. 37
Figure 5.8 Runoff hydrograph from EACU subwatershed (black) and runoff considered in MM
model (grey). .......................................................................................................................... 39
Figure 5.9 Analysis of measured rainfall data; darker years are selected for analysis. ............................ 41
Figure 5.10 Design storm........................................................................................................................... 44
Figure 6.1 BMPs selection decision tree for runoff reduction. ................................................................ 46
Figure 6.2 Possible infiltration trenches (red lines) and green................................................................. 47
Figure 6.3 Examples of available spaces for infiltration trenches location.............................................. 49
Figure 6.4 Infiltration trench section (Source: Tetra Tech 2010). ........................................................... 50
Figure 6.5 BMPs representation in SWMM (Source USEPA 2010). ...................................................... 52
Figure 6.6 Subwatershed parameters adjustment after BMPs addition (Source USEPA 2010). ............. 54
Figure 6.7 BMPs application editor in SWMM. ...................................................................................... 55
Figure 7.1 Optimization process. ............................................................................................................. 58
Figure 7.2 Schematization of the system close to the discharge point. .................................................... 61
Figure 8.1 Peak flow in the main channel, total cost and percentage of peak flow reduction;
scenarios 1, 2 and 3. ............................................................................................................... 65
Figure 8.2 Reduction in CSO events during 9 different summers; scenarios 1, 2 and 3.......................... 66
Figure 8.3 Reduction in volume discharged during 9 different summers; scenarios 1, 2 and 3. ............. 66
Figure 8.4 Optimization result for scenario 1. ......................................................................................... 67
Figure 8.7 Pareto fronts comparison for scenarios 1, 2 and 3. ................................................................. 69
xi
Figure 8.8 Spatial distribution of optimal solutions in the GR case. ....................................................... 71
Figure 8.9 Spatial distribution of optimal solutions in the PP case.......................................................... 72
Figure 8.10 Spatial distribution of optimal solutions in the IT case. ......................................................... 73
Figure 8.11 Peak flow in main channel, total cost and percentage of peak flow reduction; scenarios 1,
2 and 3. ................................................................................................................................... 74
Figure 8.12 Reduction in CSO events during 9 different summers; scenarios 4, 5 and 6.......................... 75
Figure 8.13 Reduction in volume discharged during 9 different summers; scenarios 4, 5 and 6. ............. 75
Figure 8.14 Optimization result for the combination of decentralized measures. ..................................... 76
Figure 8.15 Optimization result for storages. ............................................................................................ 76
Figure 8.16 Optimization result for the combination of decentralized and centralized measures. ............ 77
Figure 8.17 Pareto fronts comparison for scenarios 4, 5 and 6. ................................................................. 78
Figure 8.18 Hydrographs in main discharge channel in three possible solutions for scenario 6. .............. 80
Figure 8.19 Spatial distribution of optimal solutions in the case of decentralized measures, scenario 4. . 81
Figure 8.20 Spatial distribution of optimal solutions in the case of centralized measures, scenario 5. ..... 82
Figure 8.21 Spatial distribution of optimal solutions in the case of combined measures, scenario 6. ....... 83
xii
List of Tables
Table 2.1 Processes occurring in different BMPs (P: primary, S: secondary, I: incidental, N/A: not
applicable) (source Urban Drainage and Flood Control District 2010). ................................. 10
Table 2.2 BMPs' functions, qualitative treatment performance and costs (source Urban Drainage
and Flood Control District 2010)............................................................................................ 11
Table 2.3 Effect on hydrological functions of different LID techniques (adapted from EPA 2000)...... 12
Table 4.1 Discharges of coastal sewerage system for 10 years return period rainfall. ........................... 24
Table 4.2 Area and population values for each ...................................................................................... 24
Table 5.1 Different surfaces and coefficients considered. ...................................................................... 31
Table 5.2 Percentages of different surfaces and parameters estimated for each subwatershed. ............. 32
Table 5.3 Maximum intensities expected for different event durations ................................................. 33
Table 5.4 Main characteristics of total and modified models. ................................................................ 34
Table 5.5 Calibration parameters final values and calibration results. ................................................... 39
Table 5.6 Goodness of fit parameters. .................................................................................................... 40
Table 5.7 Rainfall data analysis. ............................................................................................................. 41
Table 5.8 Results of continuous summer rainfall modelling analysis. ................................................... 42
Table 5.9 Design storm development. .................................................................................................... 44
Table 6.1 Pervious pavements characteristics (USEPA 2013b, University of Maryland 2014)............. 48
Table 6.2 Pervious pavements cost calculation. ..................................................................................... 49
Table 6.3 Infiltration trenches cost calculation....................................................................................... 50
Table 6.4 Total present worth for each practice. .................................................................................... 51
Table 6.5 SWMM parameters used in BMPs modelling ........................................................................ 52
Table 6.6 Maximum possible BMPs application. ................................................................................... 55
Table 7.1 Optimization variables............................................................................................................ 59
Table 7.2 Unitary costs calculation. ....................................................................................................... 60
Table 7.3 Scenarios to evaluate. ............................................................................................................. 61
Table 7.4 Percentages of total area covered by BMPs in each scenario. ................................................ 63
Table 8.1 Average values obtained from continuous simulation. ........................................................... 65
Table 8.2 Optimal solutions comparison for scenarios 1 (GR) and 2 (PP)............................................. 70
Table 8.3 Average results from continuous simulation, scenarios 4, 5 and 6......................................... 74
Table 8.4 Optimal solutions comparison for scenarios 4, 5 and 6. ......................................................... 79
xiii
Abbreviations
BMP Best Management Practices DWF Dry Weather Flow
MM Montevideo's Municipality WWF Wet Weather Flow
LID Low Impact Development NSE Nash-Sutcliffe model efficiency
2
WQCV Water Quality Capture Volume R Coefficient of determination
CN Curve Number PEP Runoff peak flow
Tc Concentration Time PEV 6 Runoff volume after 6 hours
USEPA United States Environmental Protection Agency PEV 3 Runoff volume after 3 hours
CSO Combined Sewer Overflow DA Drainage Area
SWMM Storm Water Management Model Sl Slope
MOO Multu Objective Optimization GW Groundwater
EA Evolutionary Algorithms DEDP Dry Extended Detention Pond
GA Genetic Algorithms RP Retention Pond
NSGA Non-dominated sorting genetic algorithm IB Infiltration Basin
EAD Expected Annual Damage IT Infiltration Trenches
DS Design Storm PP Pervious Pavements
CS Continuous Simulation GR Green Roofs
SCADA Supervisory Control and Data Acquisition RH Rainwater Harvesting
UTM Universal Transverse Mercator coordinates NRCS Natural Resources Conservation Services
MLC Maximum likelihood classification VR Voids Ratio
N Manning coefficient V Volume
S Watershed slope O&M Operation and Maintenance
W Watershed width ET Evapotranspiration
IDF Intensity-Duration-Frequency curves St Storages
CP Cumulative Probability A Area
d Rainfall event duration I Imperviousness
Tr Return Period Sc Scenario
i Rainfall event intensity
xiv
CHAPTER 1
Introduction
1.1. Background
Urban growth is an important tendency in the current century. This tendency is more relevant in developing
countries; particularly in small and medium cities (less than 10 million inhabitants). Another characteristic
of this trend is the lack of planning which is leading to in an important increment of impervious areas. As a
result, hydrological patterns are being changed resulting in a rise of runoff volume and peaks that exceeds
drainage system capacity and produce floods (Parkinson and Mark 2005).
Moreover, recent studies suggest future alterations in rainfall patterns due to climate change. These
variations in patterns will cause an increase in rainfall intensities. In addition, changes in rainfall
distribution are related with major urbanisation that affects microclimates. These factors, added to physical
factors, increase flood risk in urban areas (Parkinson and Mark 2005).
If traditional approaches for drainage solutions are analyzed, it can be found that they are based on an
efficient collection and conveyance of runoff. The main objective is to avoid floods through a quick
drainage of runoff flows (USEPA 2000). As an advantage, this method ensures an adequate performance
avoiding changes in the current drainage management. On the other hand, this approach offers expensive
solutions and lower flexibility for future expansions (Chouli et al. 2007).
New tendencies are focused on decentralised approaches in a long-term planning framework including the
use of Best Management Practices (BMPs) for stormwater management. These new trends offer sustainable
solutions in a scenario of an uncertain future based on climate change and urban dynamics. BMPs are
classified into structural and non structural solutions and should be used in a combination among them and
with traditional systems (Ellis et al. 2006). Other types of solution to deal with stormwater management is
Low Impact Development (LID) measures, a relatively new strategy based on decentralized systems, with
the objective of developing an urban design that maintains the original hydrologic conditions. The use of
LID practices in conjunction with BMPs has a positive impact to reach good results on a watershed scale
(USEPA 2000).
An important concept to take into account is Sustainable Stormwater Management (SSM). Sustainable
stormwater management is based on the application of decentralised techniques oriented to runoff
reduction, through the treatment of stormwater as close as possible to the source; for example using
infiltration and evaporation techniques (Hoyer et al. 2011).
Model based multi-objective evaluation of BMP system configurations for CSO reduction 1
Considering the importance of planning processes and the arising difficulties during decision making
processes due to the complexity of stormwater management and associated uncertainties, stormwater
models and optimization techniques are very useful tools.
Stormwater models are especially useful for combined sewer overflows (CSO) management, BMP
selection and evaluation, assessment of impacts of land use changes, etc. (Obropta and Kardos 2007). One
of the most complete and widely utilized models around the world is the Storm Management Model
(SWMM) (Temprano et al. 2006). This model is a dynamic rainfall-runoff tool that simulates water
quantity and quality processes, during single rainfall events as well as for continuous simulations; it is used
mainly for urban catchments. The latest version of SWMM was updated incorporating capabilities to model
five different LID/BMP techniques (USEPA 2010).
Regarding optimization tools, sometimes decision processes include several objectives and disagreements
come from the achievement of these objectives. When one objective is disturbed due to the improvement in
reaching another, multi objective optimization techniques are needed. Under these conditions many
acceptable solutions are possible and its representation is done through a Pareto front (Price and Vojinovic
2011).
1.2. Problem definition

The combination of urban growth with the use of conventional pipe-dependent solutions for urban drainage
has several negative impacts in urban environments. The first impact is the reduction of infiltration and
evaporation. The main effects of this are lower groundwater recharge and higher runoff generation; other
important consequences are warmer and dryer local climates (heat island effect), with the associated effect
of higher rainfall intensities. The combination of these factors increases the risk of flooding and pollution
of receiving waters bodies due to sewer overflows (Hoyer et al. 2011). Figure 1.1 explains the impact of
urbanization over hydrological processes.
Figure 1.1 Impacts of different levels of urbanization, source: (Ellis et al. 2006).
2
An important disadvantage of traditional approaches for stormwater management is the low flexibility for
adaptation in front of a changing and uncertain future, mostly based on urban dynamics and climate
change. Considering the traditional approach, adaptation measures facing environment changes may require
high investments. Thus, these solutions are in general prohibitive in a short term period. As a result,
conventional stormwater approach is not sustainable in an uncertain future scenario (Hoyer et al. 2011).
Conventional combined systems are able to convey sewage and a restricted amount of stormwater. These
systems have combined sewer overflows (CSO) designed to alleviate the runoff flow above the sewer's
capacity. CSOs cause pollution of receiving water bodies; these discharges are relatively regular because
they can occur even during small rainfall events. CSO discharges have possible high impacts depending on
several local factors such as previous weather, land use, traffic conditions, urban planning and drainage
management (Montalto et al. 2007). The degradation of water bodies originated in CSO discharges is a
strong concern that forces authorities to search for corrections. A common solution is the location of
detention ponds to reduce peak flows avoiding overflows, but in general this has high investment costs
(Aad et al. 2010).
Many favourable reasons to use new decentralised technologies are recognized, although several
implementation obstacles are still identified. The main concerns include: financing issues and changes in
institutional arrangements to modify the water management approach; particularly under policies and
regulation frameworks already established. From the technical side, the major constraints include lack of
knowledge, uncertainty of impacts (mainly based in the modelling and measurements difficulties), the
interference of this kind of solution with other uses in particular in high urbanised areas, and the
prioritization of rehabilitating old systems over investments in new technologies (Chocat et al. 2007).
Models represent a valuable method to evaluate the effect of implementing new measures for runoff
quantity and quality control. Elliott and Trowsdale (2007) studied the capabilities of ten models to analyze
stormwater processes; specifically, their capacity of representing LID techniques. They found several gaps
in modelling possibilities of these tools; this has consequences for the ability to represent LID effects over
hydrology and water quality. Even if significant advances have being made during the last years (USEPA
2013a), several limitations still remain for modelling of BMPs and LIDs using dynamic rainfall-runoff
simulations such as SWMM. Despite these limitations, it is clear that a better understanding can be reached
through modelling BMPs impacts; for instance, comparison of different solutions and optimal designs can
be assessed.
The analysis of possible solutions for BMPs implementation in a watershed scale offers multiple adequate
answers. In this case two or more main objectives are faced; for instance, costs reduction and runoff
quantity reduction. The enhancement in one of these objectives will deteriorate the efficiency reached for
the other one.
Summarizing, the main problem can be visualized as traditional solutions not being sustainable for
stormwater management. This lack of sustainability has one of its reasons associated with flood events and
pollution from CSO discharges. These episodes are expected to be more frequent in the future because of
the predicted increment in rainfall intensity and urban dynamics. It is in this scenario where current systems
are expected to fail, due to the lack of adaptability capacity under reasonable investment costs.
The low flexibility has its causes in the use of centralized drainage systems, deficient planning and absence
of integration between urban design and urban drainage. Difficulties in decision making processes due to
the complexity of facing different objectives involved in drainage systems is also influencing the low
flexibility. In addition, problems are indentified in the difficulty of changing institutional arrangements, the
lack of knowledge, and the difficulties of implementing new strategies when a city is already settled.
Finally, focusing on the lack of knowledge, difficulties to evaluate the effect of BMP implementation are
identified. The major problems are based on the difficulties to evaluate the impacts of applying BMP's on a
watershed scale, also related with the necessity of designing systems to reach multiple objectives, the
difficulty to evaluate multiple possible solutions with the objective of selecting the best option and
limitations of representing BMPs using models. Figure 1.2 shows the interrelations between all the
described problems.
Figure 1.2 Interrelations between the different levels of problems.
1.3. Research questions

• Can BMPs alternatives be capable of reducing runoff and CSO discharges?
• Are decentralized and integrated solutions, combining different BMP alternatives, better
than regional solutions to reach several objectives?
• Which are the comparative impacts of applying different combinations of BMP solutions
at a watershed scale?
• Which BMPs options can be applied for runoff reduction in a high urbanized watershed?
• How to improve the design of BMPs to take into account multiple objectives for
retrofitting urban watersheds?
• What are the capabilities of last SWMM version in simulating BMPs?
1.4. Hypothesis
• The use of decentralized solutions, combining local and regional measures, has an important
impact on reducing runoff quantity at watershed scale. These impacts are significant even in an
established and high urbanized watershed.
• Through the decrease of runoff quantity a considerable reduction in CSO discharges for low return
period rainfall events is achieved.
4
• Several different BMPs can be applied to retrofit an urban watershed to achieve runoff reduction.
• It is possible to develop a methodology that uses hydrodynamic models and multi-optimization
tools to improve the design of urban drainage systems using BMP.
• A good representation of BMP effects at a watershed scale can be achieved using SWMM 5.0.022
as modelling tool.
1.5. Research objectives

General objective:
• To develop a framework for multi-objective evaluation of different BMP system configurations for
retrofitting of an established urban watershed.
Specific objectives:
• To analyze different BMP possibilities, their main characteristics, and to select some practices for
evaluation.
• To study characteristics of the hydrodynamic model selected (SWMM5) in simulating BMPs.
• To study the adjustments required in an existing tool for combining the hydrodynamic model with
the optimizer.
• To establish objective functions, optimization variables and constraints for optimization process of
BMPs configurations selection.
• To compare the effects of different BMPs system configurations.
• To apply the proposed approach in a watershed located in Montevideo and to evaluate the effect of
different systems on runoff reduction and CSO discharges.
1.6. Thesis outline

Chapter two includes the literature review with the main definitions and characteristics of BMPs and
optimization processes. Several related research are analyzed and summarized.
In chapter three the general methodology applied in this research is described. The tools used are explained
and the case study is presented.
In chapter four the data analysis and model set up processes are described, including the steps followed for
model calibration. Moreover, the analysis implemented for selecting the design storm is explained.
The tool developed for BMPs selection and its application in the case study are presented in chapter five.
The selected BMPs are designed and costs are calculated. Finally, it is explained how the simulation of
these measures is done in SWMM.
In chapter six the optimization process is presented; optimization variables and objective functions are
defined. The different configurations selected for designing using optimization are described.
The results obtained from this research are presented and analyzed in chapter seven.
Finally, in chapter eight the main conclusions deduced from this study and several recommendations to
improve it are presented.
CHAPTER 2
Literature review
To improve the understanding of the context in which the research is developed this chapter summarizes
the most important information about BMPs characteristics, capacities and modelling. Also, information
about optimization processes in urban drainage systems and design storm selection are included.
2.1. Best Management Practices

2.1.1. Definitions, benefits and limitations.
Some authors differentiate between BMPs and LIDs, for example the Urban Drainage and Flood Control
District (2010) defines Best Management Practice (BMP) as:
"A device, practice, or method for removing, reducing, retarding or preventing targeted stormwater runoff
constituents, pollutants, and contaminants from reaching receiving waters".
In addition, the same author defines Low Impact Development (LID) as:
"...a comprehensive land planning and engineering design approach to managing stormwater runoff with
the goal of mimicking the pre-development hydrologic regime. LID emphasizes conservation of natural
features and use of engineered, on-site, small-scale hydrologic controls that infiltrate, filter, store,
evaporate, and detain runoff close to its source".
According to Ellis et al. (2006), US Environmental Protection Agency (USEPA) defines BMPs into two
categories and defines LIDs as wider practices. Non structural BMPs are implemented as source control
measures that minimize runoff and pollution production; structural BMPs are implemented mainly to runoff
storage and pollutants treatment. LIDs practices are considered wider because they take into account the
global development, considering urban planning. Moreover, LID practices combine source control and
small scale solutions looking for decentralized techniques. Current approaches are oriented to integrated
management for urban drainage, combining BMPs with conventional piped systems but in the framework
of LID practices (Ellis et al. 2006).
Young et al. (2009) described LIDs as a different category of BMP that seek to preserve pre-development
hydrology conditions. For them LIDs are an extensive approach because it covers both, some structural
BMPs (i.e. green roofs and rainwater harvesting elements) and non structural BMPs.
Analyzing the different definitions, it can be concluded that in the case of LID approach, the objectives are:
the runoff and pollution control as close as possible to the source, and to integrate urban drainage with
urban planning. But the concept behind these definitions is similar in the fact that both are alternative
6
solutions of the conventional only pipe based system. In this work the terminology BMPs will include
structural BMPs, non structural BMPs and LIDs.
The Urban Drainage and Flood Control District (2010) in accordance with Ellis et al. (2006) recommends
the combination of these measures as a strategy to cope with stormwater management. Firstly, it
recommends the reduction of runoff using LID practices by decreasing impervious areas, increasing time of
concentration, and encouraging onsite storage and infiltration techniques. The second step is the application
of BMPs to provide runoff quality improvement and runoff reduction, depending on the objective followed
in each case. The strategy is oriented mainly to manage stormwater in front of frequent rainfall events,
assuming that large events will be managed trough the traditional drainage infrastructure.
The Urban Drainage and Flood Control District (2010) includes as LID practices: grass swales and buffers,
permeable pavements, bioretentions; but also emphasises in the necessity of planning, preservation of
natural areas, combination of multiple controls and volume reduction as a key target. Different LID
techniques include the conservation of existing facilities, such as trees, stream corridors, wetlands and
infiltration rates; the minimization of impacts using permeable pavements or green roofs; the minimization
of directly connected impervious areas, through the use of grass buffers, grass swales and bioretentions or
rain gardens. BMPs can be implemented in an onsite, sub regional or regional level, even when it is better
to apply regional scale measures for floods control, the management of water quality is enhanced with
distributed practices.
Following the same approach, recommendations of USEPA (2007) base the strategy for water quality
management, applying these elements in three scales: at the site level, at the community level and at the
region or watershed scale. This combination of different solutions applied at different levels is used
because the pollution's influence is present at the three scales, for this reason the highest protection of
receiving waters is reached with combination of solutions approach.
The application of BMPs, or sustainable drainage solutions, in places where a drainage system already
exists is called retrofitting. In general, an important restriction is the availability of space for locating
BMPs. Moreover, lack of knowledge or awareness, costs, lack of experience, public perception, social
acceptability, lack of information and communication, politics and lack of legislation are other barriers to
overcome. But it is important to understand that BMP solutions should be incorporated into current urban
policies and applied by retrofitting existing systems taking into account local characteristics and
considering public awareness (Shutes and Raggatt 2010)
Environmental and economical benefits can be achieved from the application of LID practices for
stormwater management. Firstly, the conservation of natural characteristics and lower disturbance of
original areas; furthermore, this type of measures is less cost intensive and save investments, both from a
long term maintenance and a life cycle cost perspective. The main limitations for LID approach
implementation are spatial restrictions and physical site conditions as for example soil type or slope. Also a
common constraint comes from the community beliefs, that assumes traditional solutions safer (USEPA
2000).
The use of LID techniques impacts reducing runoff generation and pollutants loads; also, this improves the
environment performance and reduces the cost of development in comparison with traditional solutions.
Costs are saved due to the reduction in volume of runoff generation through infiltration and evaporation
(USEPA 2007).
2.1.2. Review of different BMPs

BMP/LID measures offer pollutants removal and runoff quantity control means; some of the most
important practices are described in this section.
Grass Buffer: are strips densely vegetated that allow infiltration and decrease flow velocity; for runoff
treatment these elements permit mainly filtration. This practice is considered as one representative solution
of LID approaches. These elements are designed to manage overland runoff, not concentrated flows, and
can be located in parking areas and roadways. To provide infiltration function these elements need to be in
zones with Hydrologic Soil Groups A or B (Urban Drainage and Flood Control District 2010). This
practice has water quality enhancement as primary objective and cannot be viewed as useful in flood
control strategies (Young et al. 2009).
Grass Swale: are channels with dense vegetation and low lateral slope with the function of conveying the
flow maintaining low velocities. These elements are also inside the category of LID solutions. Other
features of these elements are low longitudinal slopes and wide cross sections; check dams can be added to
decrease the velocity and to improve sedimentation and infiltration (Urban Drainage and Flood Control
District 2010). The main contribution of this practice is to improve stormwater quality applying processes
of sedimentation and filtration; but cannot be visualized as flood control measures. Due to the linear shape
of these elements they are a good option to apply in transport and treatment of highway runoff (Young et
al. 2009).
Bioretention: also called rain gardens, are formed by a depressed area designed to detain the Water Quality
Capture Volume (WQCV: volume of runoff from frequent rainfall events). The main processes occurring
inside these elements are filtration and infiltration. If are designed to capture a major volume these
elements can provide also flood control. Due to its infiltration function, special considerations should be
taken if these elements are built close to structures (Urban Drainage and Flood Control District 2010). The
most important objective of this practice is pollutants removal but also some volume reduction can be
achieved due to infiltration. These measures can be applied through two different methods, basins or filters.
Basins allow infiltration and filtration both as main functions, while filters are located where infiltration is
not possible (Young et al. 2009).
Green roofs: the use of this practice can promote rainfall retention, runoff reduction and water quality
enhancement (Young et al. 2009). This practice is part of the LID concept because it attempts to reduce the
runoff generation close to the source. The effect on pollutants treatment has not been assessed due to lack
of enough water quality data. However, since these measures are efficient in runoff volume decrease
pollutants loads are reduced (Urban Drainage and Flood Control District 2010).
Extended Detention Basins (EDB): these elements are designed to capture runoff for many hours after the
rainfall stops. This BMP is comparable to detention volumes designed for flood control but in this case the
outlet is smaller to achieve longer detention times enhancing contaminants removal of runoff originated in
frequent rainfall events. If flood control is desired, the structure can be designed to provide runoff peak
reduction of stronger rainfall events (Urban Drainage and Flood Control District 2010). Young et al. (2009)
distinguishes two types of EDB, dry EDB and enhanced EDB. Dry EDB are dry structures during non
rainfall periods. Better pollutants removal is offer by enhanced EDB, due to the inclusion of a permanent
marsh area.
Infiltration basins: are structures that store runoff during some time allowing the infiltration of a designed
volume. If designed properly, these structures can provide flood control by reduction of flow peak storing
runoff above the infiltration zone. In this case the concept is the same than for EDB but offering the
possibility of infiltration (Young et al. 2009). The concept is the same than for sand filters, but in filters the
filtered runoff is returning to the conveyance system. In this case the main function is runoff quality
improvement, for this reason filters are used when high runoff pollution is expected (Urban Drainage and
Flood Control District 2010).
8
Retention ponds: are permanent pools with additional capacity to storage the WQCV, to release this
volume slowly. This BMP can be also designed to provide flood control increasing the detention volume
capacity. Groundwater or dry weather flow is required to maintain the pond (Urban Drainage and Flood
Control District 2010). This measure can provide both, water quality improvement and runoff peak
reduction. Retention ponds have some of the highest efficiencies in contaminants removal (Young et al.
2009).
Constructed wetlands: constructed wetland pond (CWP) is a shallow retention pond that allows the growth
of vegetation. It requires permanent inflow to maintain the plants and microorganisms. Runoff quality is
improved by physical, chemical and biological processes. Constructed wetland channels (CWC) is a BMP
designed to conveyance runoff and improves its quality. These channels have dense vegetation that
decreases flow velocity and allows water treatment. This technology requires permanent base flow and low
longitudinal slope (Urban Drainage and Flood Control District 2010).
Permeable pavements: are inside the category of LID practices; this kind of pavement allows the runoff to
enter below the surface and depending on the design this elements can provide volume reduction or
treatment. This technology is an alternative solution for pedestrian areas or areas with low velocity
vehicles; but they are not appropriate for areas with high load of sediments (Urban Drainage and Flood
Control District 2010). The pavement is created by a pervious surfaces located above a gravel reservoir,
where the empty spaces in the gravel reservoir offer storage volume. The runoff can infiltrate into the
surrounding soil. Another option is an underground piped system to collect the runoff, in this case the
storage volume may be increased to guarantee peak flow reduction for frequent rainfall events (Young et al.
2009).
Infiltration trenches: are shallow trenches located over an underground reservoir containing coarse stone
aggregate. Equally than for permeable pavements, empty spaces in this reservoir provide storage volume.
Also in this case, the runoff can infiltrate into the soil or be collected by pipes, in this case extra storage
volume is needed to reduce runoff peaks for low return period events (Young et al. 2009).
Rainwater harvesting: is a LID technique based in the capture of stormwater and the opportunity of its
reuse. The main impacts of this technique are over runoff volume, flow peak and water quality (Young et
al. 2009).
Underground BMPs: these elements offer a solution when surface BMPs are not possible; cases are more
frequent in high urbanized areas where these measures are applied by retrofitting. By implementing these
techniques the environmental benefits of surface BMPs such as infiltration, evapotranspiration and
mitigation of "heat island effect", are not used. For this reason surface BMPs should be always used when
possible (Urban Drainage and Flood Control District 2010).
In a general view, Shutes and Raggatt (2010) distinguish between structural and non structural BMPs;
furthermore, they highlight both as necessaries to reach a more robust and integrated urban water
management strategy. In accordance with them structural BMPs are engineered solutions, while non
structural BMPs are techniques oriented to preserve open and natural spaces using urban planning as a tool.
Young et al. (2009) define non structural BMPs as source control techniques such as the minimization of
impervious areas or the reduction of roadways widths. Other methods are the preservation of natural runoff
patterns omitting curbs and gutters. Also, the disconnection of impervious areas is a non structural measure,
for example sending the runoff from roofs to vegetated areas. The concept behind non structural solutions
is minimum disturbance and minimum maintenance; these techniques have a significant impact over runoff
quality but the impact over runoff quantity is variable. Therefore, structural solutions are always
necessaries, but it is highly probable that applying non structural measures the number and size of needed
structural elements will be reduced.
2.1.3. BMPs effectiveness
Different goals are sought using BMPs, these objectives are reached trough the processes that occur inside
each element. The processes can be separated between hydrologic and pollutant removal or treatment
processes.
Among hydrologic processes flow attenuation, infiltration and evaporation are recognised. Flow attenuation
reduces peak discharges and is achieved with BMPs that detain the Water Quality Capture Volume
(WQCV) and discharge it gradually. Through infiltration processes both, runoff peaks and runoff volumes
are diminished. Evaporation processes are effective reducing runoff volumes (Urban Drainage and Flood
Control District 2010).
Pollutant removal processes include sedimentation, straining, filtration, adsorption/absorption and

biological uptake. Sedimentation consists in the gravitational division of particles from the runoff and this
process occurs in BMPs that capture and slowly discharge runoff. Straining is the physical elimination or
retention of particulates inside the BMP; it can be seen as a coarse filtration. Filtration is the removal of
particles when the flow passes through a media. When a substance is incorporated into another in a
different state the process is absorption (i.e. liquids absorbed by solids); when the physical adherence of
different molecules occurs it is called adsorption. Finally, biological uptake consists of assimilation of
organic and inorganic components by plants and microorganisms. It is not fully understood which
biological processes occur during the short time that stormwater is in contact with biological media inside
BMPs (Urban Drainage and Flood Control District 2010). Table 2.1 shows the diverse processes occurring
in different BMPs. Table 2.2 describes deeper the functions and performances of some BMPs.
Table 2.1 Processes occurring in different BMPs (P: primary, S: secondary, I: incidental, N/A: not
applicable) (source Urban Drainage and Flood Control District 2010).
Hydrologic process Treatment process
BMP Flow Evapo- Adsorption/ Biological
Infiltration Sedimentation Filtration Straining
attenuation transpiration Absorption uptake
Grass Swale I S I S S P S S
Grass Buffer I S I S S P S S
CWC I N/A P P S P S P
Green Roof P S P N/A P N/A I P
Permeable
P P N/A S P N/A N/A N/A
Pavement
Bioretention P P S P P S S P
EDB P I I P N/A S S I
Sand Filter P P I P P N/A S N/A
CWP P I P P S S P P
Retention
P I P P N/A N/A P S
Pond
Underground
Variable N/A N/A Variable Variable Variable Variable N/A
BMPs
10
Table 2.2 BMPs' functions, qualitative treatment performance and costs (source Urban Drainage and Flood
Funtion Effectiveness for Targeted Polluants Other
BMP Volume WQCV WQCV+Flood Total Life-cycle
Sediments Nutrients Bacteria
reduction capture control Metals costs
Grass Swale Yes No No Good Moderate Good Poor Low
Grass Buffer Yes No No Good Moderate Good Poor Low
CWC Somewhat No No Unknown Unknown Unknown Moderate Moderate
Green Roof Yes Yes Yes Unknown Unknown Unknown Unknown Unknown
Permeable
Yes Yes Yes Very good Good Good Unknown High
Pavement
Bioretention Yes Yes Yes Very good Moderate Good Moderate Moderate
EDB Somewhat Yes Yes Good Moderate Moderate Poor Moderate
Sand Filter Yes Yes Yes Very good Good Good Moderate Moderate
CWP Somewhat Yes Yes Very good Moderate Good Poor Moderate
Retention
Somewhat Yes Yes Very good Moderate Moderate Moderate Moderate
Pond
Underground
Variable Variable Variable Variable Variable Variable Variable Moderate
BMPs
As already mentioned, three levels of BMPs application are defined by different authors. Shutes and
Raggatt (2010) also distinguish these three levels and add one more. For them the first level is prevention
with measures such as roads sweeping or minimization of impermeable surfaces. The second level is source
control through the application of measures such as rainwater harvesting, pervious pavements and green
roofs. The third level is site control with for example infiltration methods. The last level is regional control
by implementing for example detention ponds and wetlands.
An additional approach to group LIDs considers the main process applied. USEPA (2007) discriminates
these processes among conservative designs, infiltration practices, runoff storage solutions, runoff
conveyance approach, filtration methods and low impact landscaping. Conservative solutions are based in
the minimization of runoff generation, maintaining open spaces and reducing impervious areas; some
measures are restriction of roads, pavements and parking. Infiltration practices are useful to reduce the
volume of runoff and sometimes offer aesthetic and recreational opportunities; some examples include
infiltration basins and trenches, porous pavements and rain gardens. Storage techniques are efficient in
decreasing the runoff peak and if it includes infiltration can also reduce the runoff volume; advantages can
be obtained designing these structures in empty areas between paved surfaces. Conveyance practices are
composed by elimination of curbs and gutters, creation of grass swales and check dams among other
solutions; the objective is to decrease the velocity of flow and to promote evaporation and infiltration.
Filtration solutions contain bioretentions, rain gardens and vegetated filters. Low impact methods are
formed by reforestation, encourage of native species and grass plantation and enhancement of soil to
increase infiltration.
Regarding to runoff reduction, four components have to be considered to evaluate the effectiveness of
LIDs: curve number (CN), time of concentration (Tc), retention and detention. The influence on these
hydrological components for different LIDs is presented in Table 2.3.
Table 2.3 Effect on hydrological functions of different LID techniques (adapted from EPA 2000).
LID practice Lower CN Higher Tc Retention Detention
Lower slopes X
Increase flow path X
Higher roughness X
Infiltration swales X X
Grass buffers X X X
Disconnection of
X X
impervious areas
Reduction of curbs
X X
and gutters
Rain barrels X X X
Rooftop storage X X X
Bioretention X X X
Revegetation X X X
These effects on hydrologic patterns also have impacts on contaminants removal; by increasing Tc these
practices diminish the capacity of runoff for transporting pollutants and reduce pollutants loads. Pollutants
loads are mainly concentrated in the "first flush" of runoff coming from impervious surfaces; this runoff
transports oils, nutrients, sediments and metals (USEPA 2000). First flush is defined as the phenomena
occurring at the beginning of runoff events characterized by a high concentration of contaminants. These
non continuous discharge events may have a significant pollution impact on receiving waters (Price and
Vojinovic 2011).
Other impacts of these practices are also taken into account to evaluate effectiveness. Shutes and Raggatt
(2010) describe BMPs as an effective and sustainable urban stormwater management, based in its
effectiveness to accomplish with the three indicators of sustainability: flood control, pollutants reduction
and adaptation.
Associated with pollutants reduction is the minimization of CSO discharges. In order to develop urban
pollution management strategies it is important to define objectives about contaminant concentrations and
discharges duration and frequency in accordance with the uses established of receiving waters. This is
particularly important considering that the most dangerous hazard originated in CSO discharges is
bacteriological and pathogenic contamination due to it relation with water related diseases (Price and
Vojinovic 2011). The reduction of CSOs can be achieved through the minimization of runoff volume going
inside the sewer system; one possible measure to reach this objective is by encouraging infiltration. Whit
respect to this, green roofs are effective solutions especially for retrofitting of developed areas (Shutes and
Raggatt 2010). Most of the annual runoff volume is generated by small storms; large storms produce a
minor fraction of average annual runoff discharges. Considering this and the possibility of reducing runoff
volumes applying LID techniques and micro scale BMPs, this approach may have a high positive impact on
receiving waters (Huber et al. 2006).
Stormwater management focus mainly on solving flood and water quality problems. Runoff volume is the
major hydrologic parameter related with water quality and runoff flow peak is for flooding analysis. As a
consequence different rainfall events are used to evaluate these two problems. Small storms generate most
of the annual runoff, while large storms are responsible for small percentages of annual discharges.
However, a criterion to select small design storms is not established, one possible strategy could be to use
continuous simulation to evaluate the overall performance of the system (Huber et al. 2006). Urban
Drainage and Flood Control District (2010) presents a study in which 61% of storms occurring in an
average year, produce almost no runoff; while almost 92% of annual storms have less than 16mm of
rainfall depth.
12
Much research is being done to evaluate the effect of different BMP/LID practices on hydrologic processes
and runoff contaminants removal. Several particular results are presented in this section.
Several authors recommend the combination of different measures to reach better results in runoff quantity
management. For example Damoradam et al. (2010) proposes the use of LID infiltration technologies to
cope with small storms and BMP storage based solutions to manage runoff from strong events, but they
suggest the combination of measures to reach flood solutions in a sustainable approach for stormwater
management. In their case they applied a combination of LIDs and BMPs in a highly urbanized watershed;
they combined the use of permeable pavements, rainwater harvesting and green roofs with detention ponds.
To simulate the hydrologic behaviour of LIDs the Curve Number approach was used; as hydrologic model
the Hydrologic Engineering Centre-Hydrologic Modelling System (HEC-HMS) was used and for hydraulic
simulation the model used was SWMM. This research shows higher efficiency of LID techniques than for
storage based BMPs for small storms and the opposite result when the intensity of rainfall is increased.
From these results, the study concludes that a combination of both techniques is the best approach to reach
sustainability and flood control. Finally, a recommendation to future researches about application of
optimization techniques to find the best combination of LID-BMP is done. Figure 2.1 shows the result of
their research for five different storms.
Figure 2.1 Runoff hydrographs from five storms simulating different solutions (Damodaram et al. 2010).
Burszta-Adamiak and Mrowiec (2013) conducted tests to compare runoff coming from conventional roofs
and from green roofs. They found an increment of approximately 40% in mean retention of runoff using
green roofs. From these results green roofs are recommended as an efficient method to attenuate runoff at
the source, with the capacity of reducing runoff volumes, CSO discharges and flooding problems in urban
areas.
Palla et al. (2008) modeled with SWMM different scenarios of green roofs coverage under different storm
conditions and conclude that green roofs implementation has significant impact on runoff volume reduction
and flow peak attenuation, reducing the risk of local flood events and CSO discharges in urban areas.
Scholz and Kazemi Yazdi (2008) tested the efficiency of an experimental system formed by a vegetated
gravel filter and a detention and infiltration tank. Good pollutants removal efficiencies were registered,
with 77% of biochemical oxygen demand (BOD) removal, 83% of suspended solids (SS) removal, 32% of
nitrate-nitrogen removal and 47% of phosphorus removal.
Aad et al. (2010) used SWMM to model the impact of applying rain barrels and rain gardens to disconnect
impervious areas, mainly roofs, from the drainage system. They found a strong effect on runoff peak and
volume reduction by implementing rain gardens, even for low area coverage; with 3.9% of rooftop area
going to rain gardens the volume was decreased in 38%.
2.1.4. BMPs selection

Several factors influence the selection of BMPs, but the main factors include physical site characteristics,
space constraints, and targeted processes. Physical characteristics include soil, watershed size,
groundwater, base flow and watershed activities (Urban Drainage and Flood Control District 2010).
The type of soil is important to apply infiltration based practices, in this case adequate soils are Hydrologic
Soil Groups A and B. In cases of low permeability, underdrain systems are used; in cases of high swelling
soils, infiltration techniques should be avoided. The size of contributing surface is relevant for BMPs
applied at site level, because some practices need a minimum area (i.e. EDB). At the regional level the
restriction is related with the maximum area, to ensure the adequate treatment of total runoff volume
enough residence time in the BMP has to be achieved. The location of the water table is important because
shallow groundwater offer limitations to implement infiltration solutions and can result in groundwater
occupying volume in storage devices. In other solutions such as constructed wetlands permanent flow is
necessary to maintain vegetation; in these cases shallow groundwater can be useful. Also retention ponds
need a base flow to maintain the wet conditions in dry periods. Watershed condition is another important
factor in cases where for example erosion processes are relevant, in this case selection of BMPs that work
with high loads of sediments required (Urban Drainage and Flood Control District 2010).
Space availability frequently defines the possibility of implementing BMPs; especially in places with high
population density. The major cause of this problem is the lack of urban planning. Finally, it is important to
take into account the objectives when applying BMPs, some practices are more efficient in pollutants
removal and some are more efficient to reduce runoff volume and peak flow (Urban Drainage and Flood
Also, implementation and maintenance costs, aesthetic benefits or liability improvements and public safety
(i.e. mosquitoes generation) should be considered to select which BMPs to apply (Young et al. 2009).
14
2.2. Modelling BMPs with SWMM
Even when modelling can be useful to evaluate the effect of BMPs implementation, different limitations for
this kind of simulations are recognized.
The inclusion of LIDs in the last version of SWMM model is available since July 2010; from that moment
on it has been possible to explicitly simulate some of these solutions at a subwatershed level (USEPA
2013a). SWMM considers LIDs as properties of subcachments. Five different types of these elements can
be represented: bioretention cells, infiltration trenches, porous pavements, rain barrels and vegetative
swales. The impact of LIDs application is reproduced in the general runoff produced by the subwatershed
(USEPA 2010).
The first limitation of SWMM appears when it is needed to simulate practices not included in the list
mentioned above. Another important limitation is its incapacity for reproducing pollutants removal in LID
elements; currently SWMM only can represent the hydrological impact of LIDs (USEPA 2010).
The possibility of modelling storages using SWMM already existed in 2010, evaporation, infiltration and
contaminants treatment processes are included for these elements.
Burszta-Adamiak and Mrowiec (2013) modeled green roofs hydrologic performance using the last version
of SWMM. They used the new capabilities of SWMM modelling LIDs, working with bioretention cells for
green roofs representation. Even when new parameters are introduced that allow better representation of
green roofs layers, these authors could not find available literature about the quality of this model. An
important problem that found was the impossibility of representing rooftop slope and roughness coefficient.
By comparing results obtained from the simulation with results obtained from tests in the field, they
concluded that SWMM5 has restricted capabilities to simulate an accurate runoff hydrograph from green
roofs.
Houston (2006) studied different possibilities for BMPs representation in SWMM including treatment
capabilities. He focuses in simulation of individual BMPs using SWMM, and identified some important
SWMM limitations in BMPs modelling. For example, infiltration in nodes and conduits is not allowed; this
is a restriction when individual BMPs want to be simulated.
Another problem found is the definition of land use as a percentage of total area. If one subwatershed has
part of impervious area and part of pervious area and a land use to represent pollutants buildup want to be
added; SWMM distributes this land use uniformly over the total area. One possible solution is to split the
subwatershed into two. A similar problem is found when pollutants treatment want to be applied at a
subwatershed level, representing street sweeping. (Houston 2006).
Finally, two important limitations where identified by Houston (2006). Firstly, continuity problems in
pollutants mass representation for system routing where found. This research identified significant
pollutants mass balance errors using SWMM, depending on pipes geometry. Another problem includes the
difficulty of representing shot-circuit in treatment devices.
2.3. Optimization of urban drainage systems

Optimization is a very important process to achieve the most effective cost-benefit solution, minimizing
investment and maintenance costs and at the same time maximizing the benefits reached by implementing
those solutions. Multi-objective optimization is especially important to decision making processes to
evaluate several possible options to reach the different desired objectives.
Tools that allow holistic points of view are needed in situations where several goals are pursued; this is
important in decision making process to achieve an optimal assessment between different alternatives. In
this context multi-objective evaluation tools are useful to solve conflicts between different interests. In
particular, available information about the use of these tools for urban drainage is limited. To apply these
techniques for urban drainage the optimizer can generate several scenarios that can be evaluated using
hydraulic models. For the evaluation process of different options, objective functions are needed. Through
repetition of this process optimal solutions are found and can be represented using a Pareto fronts. Using
this result, decision makers are able to choose the best option from a group of optimal solutions (Barreto et
al. 2010).
Multi-objective optimization consists in searching the best possible solutions considering different
objectives, represented by objective functions, in a defined domain and under defined constraints. Conflicts
can arise from the necessity of satisfying several objectives at the same time. In a multi-objective
optimization process it is not possible to improve the achievement of one objective without the
deterioration of another objective. For this reason the result is a set of compromised optimal solutions
visualized through Pareto fronts, these solutions are the alternatives to take into account in decision making
processes. On the contrary, single objective optimization cannot provide a set of optimal solutions to use in
decision making processes (Sahlu 2012).
Evolutionary Algorithms (EA) work with a group of solutions, this approach can be extended to reach
several Pareto optimal solutions in a single simulation. Non-dominated sorting genetic algorithm (NSGA)
was one of the first ones using this concept. NSGA-II was developed as an improved version of NSGA
(Deb et al. 2002). Included as an EA, Genetic Algorithms (GA) need fitness and objective functions, to
evaluate optimal condition and assess each solution performance respectively. The process starts with a
random generation of the first population. Afterwards, in each generation fitness function is used to
evaluate the chromosome. Through this process the best possible solutions are selected. Next generation
consists in creating a new population using these solutions and applying cross over and mutation. The
process continues until the criteria established for ending is reached, this could be a maximum number of
generations or a good enough level of fitness (Sahlu 2012).
Barreto et al. (2010) applied multi-objective optimization in combination with hydrodynamic models to
analyze a case of urban drainage rehabilitation, obtaining valuable information that can be used in decision
making processes. This work also compared the performance of two optimization algorithms, the non-
dominated sorting genetic algorithm (NSGA-II) and the epsilon multi-EA (ɛ-MOEA). They found that
NSGA-II has higher performance with small population sizes, generates more diverse sets of solutions and
use shorter computation time.
When an optimization problem considers two objectives, several optimal solutions are possible from taking
into account the trade-offs from confronted objectives. Pareto representation of optimal solutions is a very
useful tool for making decision processes regarding urban drainage solutions. This is an advantage of
Multi-Objective Optimization (MOO) over one objective optimization analysis (Delelegn et al. 2011).
Delelegn et al. (2011) used MOO tool to support decision making process in a case of flood problems.
They considered two objectives, cost reduction and hydraulic performance of the system. They recognize
as a limitation the use of only one possible solution (storage volumes) due to the necessity of reduced
variables in optimization problems. This research suggests source-control solutions as possible more cost-
benefit efficient solutions.
Damodaram et al. (2010) compares the effect of different BMPs and LIDs on runoff volumes and peak
flows reduction. To select different BMPs/LIDs combinations they tested some arbitrary possibilities for
each practice. This work recommends the utilization of optimization in further research to recognize the
best combination of BMPs and LIDs to achieve sustainable solutions in cases of restricted resources.
16
Jia et al. (2012) studied the effect on runoff reduction of applying BMPs in the Beijing Olympic Village.
They modeled this case coupling SWMM with the BMP Decision Support System (BMPDSS). They found
that a reduction of 27% and 21% of runoff volume and peak flow respectively, can be achieved.
BMPDSS was developed by USEPA and Prince George's County as decision making tool for stormwater
management. This tool is capable of BMPs analysis that simulates the performance of these practices and
optimizes its placement and design (Jia et al. 2012). The optimization component uses evolutionary
optimization techniques that employs scatter search as algorithm. Scatter search is a meta-heuristic search
method used to optimize complex problems (Cheng et al. 2009).
Two examples of optimization frameworks for stormwater management are BMPDSS and the US
Environmental Protection Agency System for Urban Stormwater Treatment and Analysis Integration
(SUSTAIN).These tools are able of optimize systems using several different BMPs to achieve hydrologic
and water quality objectives. In both cases the algorithms used for simulation are similar than those used by
SWMM and both of them can optimize using GA methods. An important limitation presented by these
software is that they use ArcGIS platform to run (Zhang et al. 2013).
Urrestarazu (2013) optimized the size and distribution of SuDS and storages in an urban watershed. He
applied SWMM5 and NSGA-II, coupling these tools with the software SWMM5-EA. For the multi-
objective optimization he used two objective functions: total cost minimization and total benefits
maximization (benefits equal to flood reduction and ecosystem services increment). As optimization
variable he worked with the size of drainage elements. BMPs considered were rain barrels and green roofs.
2.4. Rainfall selection

There are two different approaches of simulation that can be applied in drainage systems modelling; using a
single rainfall events or continuous series. Single events are used based in frequency analysis and are
adequate for designing when the cause have the same frequency than the effect. Continuous series
modelling uses long series of measured rainfall data to evaluate long term effects. Long term analyses are
important when the frequency of the cause cannot be directly related with the frequency of the effect, or
when the result is more connected with the accumulated effect of the cause. The use of long term models is
adequate in cases where the occurrence of the effect is long in comparison with the period in which the
cause occurs (Vojinovic and Abbott 2012).
Several research experiences are reported in the academic literature, they follow different criteria for
rainfall selection in optimization processes and BMPs design. In accordance with Lee et al. (2012) to
analyze BMPs effectiveness it is better to use long-term rainfall data; they use long term hourly
precipitation in BMP systems designing. Jia et al. (2012) evaluates BMPs using hourly rainfall data for one
year that they consider representative in terms of rainfall amount and seasonal precipitation distribution.
Yang and Cui (2012) chooses several rainfall events occurring in the same year and greater than 25 mm
depth for studying integrated stormwater management techniques using BMPs. Urrestarazu (2013) uses
several design events with different return period to optimize the design of BMP systems for floods
reduction; to reach this approach he uses the Expected Annual Damage (EAD) concept to develop a depth-
damage correlation. Also Paredes (2013) uses several design events with different return periods to
optimize the rehabilitation of a drainage system, to reduce floods. Barreto (2012) uses the EAD method in
the process of optimization of drainage systems rehabilitation; using this method different return periods
and its effects are considered.
According to Lucas (2010) CS models can simulate better the response of the system as a whole. Also,
appropriate criteria for defining DS can be achieved from the analysis of CS of rainfall. This possibility
improves the selection of DS models typically used by designers, avoiding the necessity of using CS
simulations that require more effort.
In accordance with (Huber et al. 2006) small storms are in charge of most of the runoff discharge amounts,
while large storms contribute to a small part of annual discharges. Designs to solve flood problems are
based in single events but there are not accepted criteria to select small design storms. One possible option
is the use of continuous simulation (CS) to evaluate how the system works over the time.
2.5. Discussion
Until now, there are not many publications in the academic literature using multi-objective optimization
with Genetic Algorithms to evaluate the use of BMPs and select the best configuration for runoff reduction
in urban drainage systems. The combination of hydrodynamic models in urban drainage, BMPs application
for runoff reduction and optimization tools is currently interesting because it covers several aspects of
planning and designing processes related with decision making for selection and implementation of
adequate measures. Additionally, this work studies the capabilities of the last SWMM version in simulating
BMPs effects and analyzes the results of applying these measures in a highly urbanized watershed under
single rainfall events and continuous series modelling.
This research attempts to help in spreading a methodology that can be helpful in improving current
practices in developing countries, through the application of the obtained results in other cases.
18
CHAPTER 3
Methodology
The general methodology proposed for multi-objective evaluation of different BMP system configurations
is presented in this chapter. In addition, the different tools used to develop this research are enumerated and
explained. Finally, a description of the case study in which this methodology is applied for retrofitting of an
urban watershed is included.
3.1. Research approach

This work aims to develop a methodology to evaluate different BMP configurations for retrofitting urban
drainage systems. This methodology is applied in a case study, to model an urban catchment evaluating the
effect of BMPs on runoff reduction and CSO discharges. Optimization methods are used to design different
configurations considering multiple objectives at a watershed level. Finally, comparisons of effectiveness
between centralized, decentralized and combined solutions are developed.
The research methodology consists of six phases described hereafter:
Phase 1 - Data analysis: The research starts with the collection of the data needed. This data includes
mainly physical characteristics of the case study and rainfall data. Physical and hydrological parameters to
be used in model set up are defined by processing these data. Finally, rainfall data is processed to choose a
design storm and to select continuous time series to be used in results analysis.
Phase 2 - Model set up: The second step includes the development of the hydrological-hydraulic model.
This stage includes the set up of the model. The system is represented by a simplified model including only
the main pipes and subwatersheds in SWMM. Afterwards, hydrological parameters have to be adjusted
(calibration) to obtain adequate runoff results from the hydrologic model.
After the model is ready, it is used to select the design storm. Using historical rainfall data continuous
simulations are run and the effects on CSO are analyzed to define the event to be used in design processes.
Phase 3 - BMPs selection and simulation: The next step consists on a review of different BMPs, analyzing
each option: characteristics, capabilities, limitations and all the different factors that are needed to
understand the available options. Knowing the theory about BMP techniques, the next step is to study local
conditions to identify and select the best techniques to apply in the case study. Afterwards, selected BMPs
are designed and costs are estimated. Finally, the simulation of these selected BMPs with SWMM5 is
implemented, understanding the limitations and main variables to represent these elements using
modelling. Through this procedure the steps needed to vary BMPs quantities in the model can be
understood and variables for optimization process can be selected.
Phase 4 - Optimization: This phase includes the definition of decision variables, constraints and objective
functions to be applied in the optimization process. Afterwards, different possible configurations or
scenarios are developed. To finish, each of these scenarios is optimized in accordance with the objectives
previously defined.
Phase 5 - Results Analysis: The last stage is to analyze and compare the effect of different systems on
runoff quantity reduction under different rainfall events. Continuous simulation is used to analyze the
impact on CSO discharges. Different optimal solutions are evaluated and compared with other optimal
solutions for the same configuration and other scenarios.
Figure 3.1 represents the steps followed in this methodology and the interconnections between the different
phases.
DATA ANALYSIS:
MODEL SET UP
DATA COLLECTION.
HYDRAULIC AND HYDROLOGIC Existing model
MODELS DEVELOPMENT. analysis and
DATA PROCESSING:
simplification.
- Parameters determination
MODEL CALIBRATION.
using ArcGIS analysis.
- Rainfall analysis; time series
selection.
HYDRODYNAMIC
MODEL.
DESIGN STORM
SELECTION
BMP SELECTION
PRACTICES SCREENING. BMP DESIGN

SELECTED MODEL MODIFICATION
AND COST
BMPs SET. FOR BMPs APPLICATION.
SPATIAL CONSTRAINTS ANALYSIS. ANALYSIS.
SUITABILITY ANALYSIS.
COMPARISON OF “DO
NOTHING” AND “MAXIMUM
APPLICATION” SCENARIOS
USING TIME SERIES
MODELLING
OPTIMIZATION
OPTIMIZATION VARIABLES
DEFINITION.
MULTI-OBJECTIVE
OBJECTIVE FUNCTIONS DEFINITION. YES
OPTIMIZATION.
COMPARISON OF BMP
CONFIGURATIONS
SCENARIOS EVALUATING RUNOFF
NO AND CSO REDUCTION
DEFINITION
EFFICIENCY.
Figure 3.1 Different phases of research methodology.
20
3.2. Materials and methods
3.2.1. Tools used
Data analysis:
The analysis and process of data is done using the software ArcGIS. From this analysis different physical
characteristics and hydrological parameters are defined, such as percentage impervious, Manning
coefficients, subwatershed's slopes and widths.
Hydrodynamic model:
As already mentioned, the SWMM version 5.0.022 is used to model the urban drainage system. This is a
dynamic rainfall-runoff model that can be used to simulate single events or continuous simulations of
runoff quality and quantity. The last version of the software is capable of modelling hydrologic, hydraulic
and water quality processes. Among its hydrologic processes, the software is capable of representing
infiltration and rainfall interception, evaporation and five different types of LIDs: bio-retention cells,
infiltration trenches, porous pavements, rain barrels and vegetative swales (USEPA 2010).
Montevideo's Municipality (MM) has a hydraulic model in SWMM for the whole urban area of the city.
This model is reduced to represent only the case study area. Also, it is necessary to verify how the available
model is working. In MM's model runoff hydrographs are entered in different nodes, but the rainfall-runoff
simulation is not done in SWMM. It is necessary to build a hydrologic model in SWMM to represent the
BMP elements that have impacts at source level. After the development of the new model, the adjustment
of different parameters is done to reach the same runoff hydrographs currently used in the existing model.
Rainfall data processing:

Rainfall data available includes 17 years of hourly rainfall; these data are processed using a code developed
with the software Python. Using this code the files provided by MM are processed to obtain rainfall series
of summer periods ready to include into the model or to develop an intensity-duration-frequency analysis.
To evaluate impacts on CSO discharges continuous simulation with summer rainfall series are executed.
The last step in rainfall data analysis is the selection of a storm event to apply in optimization process. To
choose the characteristics of the design storm used continuous rainfall series of summers are run to evaluate
the CSO effects in the watershed.
BMPs selection:
The selection of BMPs is done considering local conditions, limitations and specific objectives. Different
methods for BMPs selection are analyzed to select the most adequate for this research. The method applied
follows the same concept proposed by the Urban Drainage and Flood Control District 2010, which consists
of the application of decision trees to select BMPs based on physical characteristics. To consider
possibilities and limitations preferences of different stakeholders (such as urban planners, drainage and
sewerage system department, operation and maintenance service, traffic regulation service) are taken into
account.
Optimization process:
Multi-objective optimization (MOO) analysis is conducted to select adequate solutions considering
different possible systems. This process is done taking into consideration two objectives: runoff reduction
and minimization of costs. The methodology used to design the best solution is based on evolutionary
methods, Genetic Algorithms (GA) are applied for urban drainage systems optimization in this research.
When evolutionary algorithms are used, the rules of the problem to be optimized are not considered as a
part of the optimization process which makes this method adequate to be used in complex problems. GA
imitates evolutionary theories selecting from a population, the individuals that better fits the requirements
and creating from them new individuals.
The improved version of non-dominated sorting genetic algorithm (NGSA-II) is used as optimization tool.
This algorithm is capable of solving multiple objective optimization problems and can find multiple Pareto-
optimal solutions in one simulation (Deb et al. 2002).
CHAPTER 4
Case of study
4.1.1. Hydraulic model adjustment

The proposed approach will be applied on a study area located in the city of Montevideo, Uruguay. The
case study is a coastal and highly urbanized watershed in the neighbourhood of Pocitos. Pocitos watershed
is located in the southeast zone of Montevideo and it is part of the coastal sewerage system of the city. The
coastal sewerage system covers an area where around 55% of Montevideo's population lives. The areas of
these systems are showed in Figure 4.1.
The city has not wastewater treatment plant. The coastal system has a main collector with a length of 10Km
that follows the line of the coast and conveys the dry weather flow to the final point. This final point is
located in Punta Carretas (Figure 4.1), in this area the system has a pre-treatment plant and a main pumping
station from where the sewage is sent into the Rio de la Plata river through a pipe of 2400m length.
Pocitos watershed has a combined sewer system with 100% coverage, the land use is mainly residential and
it is a formal area with medium to high economic level. The watershed has an area of 575Ha, 70000
inhabitants and one point of CSO discharge in the coast.
This area has one of the highest population densities of the city, with 125p/ha when the average density in
Montevideo is 65p/ha. The area has border with the river Rio de la Plata in the south; in this place one of
the most visited beaches of the city is located. The area covered by the coastal sewerage system has more
than 14 kilometres of coast with more than 12 beaches highly used during summer months.
Due to the proximity between the beach and the CSO discharge point of Pocitos watershed, the main
interest of MM is the reduction of CSO discharges during summer months. The position of these two points
can be observed in Figure 4.2.
Studied solutions carried out by MM include the increase in drainage capacity or the construction of
underground storage tanks to avoid both CSO discharges and to reduce some local pluvial floods that occur
in the area during heavy rainfall events. Studies oriented to reduce runoff at source level have not been
contemplated yet into the planning process of the drainage system of the city.
22
Rio de la
Punta Carretas Plata
Figure 4.1 Areas of Montevideo, costal sewerage system and Pocitos watershed.
Discharge
channel
Pocitos beach
Figure 4.2 Watershed for application of the study case (source Google Earth). Orange line represents the
watershed border. Pocitos beach and CSO discharge point are indicated.
4.1.2. Description of the drainage system
The coastal system has 7 overflow points along the 14 kilometres of coast. Table 4.1 shows peak flows and
volumes discharged in each outfall point from the results obtained with the SWMM model proportionate by
MM; the outfall of Pocitos watershed is called Buxareo. The analysis of these results demonstrates the
importance of Pocitos watershed according to volume discharged. This factor is added to the high
population density and the importance of Pocitos beach, to locate this watershed as a strategic one for the
study of outfall discharges reduction.
Table 4.1 Discharges of coastal sewerage system for 10 years return period rainfall.
Peak flow Volume % of total volume
Discharge point Beaches influenced 3 3
(m /s) (m ) discharged
Buxareo Pocitos/Puerto Buceo 66.3 368813 50.6%
26 de Marzo Puerto Buceo 4.1 12907 1.8%
Malvin Buceo /Malvin 13.1 69700 9.6%
Colombes Buceo /Malvin 18.8 126288 17.3%
Punta Gorda Los Ingleses/Verde 1.6 2966 0.4%
Arroyo del Molino Honda 5.7 21891 3.0%
San Nicolas Mulata/Carrasco 18 126565 17.4%
Pocitos watershed is divided in 14 subwatersheds, using the same division defined by MM to represent the
system in SWMM. Figure 4.3 shows this division.
Using census data, values of population for each subwatershed were calculated. Also, total population for
Pocitos watershed is estimated. Values expected for year 2035 are presented in Table 4.2.
Table 4.2 Area and population values for each

subwatershed and total watershed.
Subcatchment Area (Ha) Population
EACU 81.0 10016
FELI 37.3 2236
FRAN 66.2 5206
BATL 50.1 853
MFRA 21.3 2581
HAED 17.7 3159
CIPR 36.6 7278
ALEJ 21.7 7872
AYAC 82.8 9279
CIAN 23.7 2664
BAUZ 33.0 3508
BUXA 12.5 3792
ECHE 71.5 11389
BUCE 20.1 2033
TOTAL 575.5 71867
24
Figure 4.3 Pocitos watershed divided in 14 sub areas according to MM division.
The watershed counts with two main pipes, one combined and the other one pluvial. The coastal system of
the city has a main collector pipe that follows the coast and receives the sewage coming from the different
watersheds.
The system has two points of discharge into the coast in the catchment area. The main discharge point is an
open channel 660m length (Figure 4.4), connected directly with the main pluvial pipe of the watershed. The
major combined pipe in the catchment is connected through a weir upstream the beginning of this channel.
As a consequence, when it is raining the combined system discharges into the pluvial system and through
this to the coast. The other point of discharge from the system is a weir located in the channel, 150m
upstream the end of the channel, it works as an extra discharge point when the level in the channel is higher
than 2.35m.
Figure 4.4 Main CSO discharge channel.
Another weir is discharging into the main pluvial pipe before the channel's beginning. This weir is coming
from the main coastal collector, which transports water from other watersheds.
Figure 4.5 shows the network configuration in the discharge point of Pocitos watershed on the coast.
Figure 4.5 Watershed outfall point (green: pluvial pipes; blue: combined pipes; violet: discharge conduits).
(Source Google Earth and MM data base).
4.1.3. Rainfall pattern

Mean annual precipitation in Montevideo is 1100 mm/year. This value is well distributed during the year as
can be observed in Figure 4.6. This figure shows monthly mean accumulated precipitation in two
meteorological stations located in different places of the city, based in average data collected during 30
years (Direccion Nacional de Meteorologia 2014).
Annual rainfall distribution

120
100
Rainfall (mm/month)
80
60
Carrasco
40 Prado
20
0
1 2 3 4 5 6 7 8 9 10 11 12
Month
Figure 4.6 Annual rainfall distribution in two different meteorological stations.
26
In Figure 4.6 the annual rainfall distribution can be observed, with an average monthly value of
90mm/month. In this work only summer months (December, January and February) will be consider
because is the period in which the risk of health issues related with pollution coming from sewerage is
highest.
4.1.4. Hydraulic model description

Montevideo's Municipality has a hydraulic model in SWMM for the whole urban area of the city. This
model was used with some modifications to represent only the case study area. The changes done to the
model are explained in the next chapter.
To set up the model the rainfall event used was the same than MM uses for 10 years return period, figure
4.7 shows this event. It is a symmetric storm with total duration of 6 hours. This event is determined using
the Montana's method for intensity calculation, an intensity-duration-frequency relation calibrated for
Montevideo from historical data (Silveira et al. 2013), and using alternating block method with intervals of
5 minutes to create the storm.
In this model rainfall-runoff processes are simulated outside SWMM. MM uses a rainfall-runoff model
based on runoff coefficient to represent runoff generation and unit hydrograph to build a runoff
hydrograph. The concentration time is calculated with Desbordes equation (Intendencia Municipal de
Montevideo 1994).
Rainfall Tr=10 years, d=6 hours

200.0
180.0
160.0
140.0
120.0
i (mm/h)
100.0
80.0
60.0
40.0
20.0
0.0
0 50 100 150 200 250 300 350 400
time (min)
Figure 4.7 Rainfall event used by MM in the model (mm/h - h).
Dry weather flows are calculated by MM considering different types of consumption and infiltration
components from groundwater and from loses in water supply system. The different types of consumption
include residential, commercial, public sector and industry (Intendencia Municipal de Montevideo 1994).
Figure 4.8 shows the low influence of DWF in a runoff hydrograph for 10 years return period event.
Figure 4.8 DWF effect in a runoff hydrograph for 10 years return period rainfall.
Finally, the original model presented some problems related with SWMM version in which it was created crea
and also presents some numerical instability
in problems. The most important instability problem occurs in
the discharge hydrograph. This problem was solved changing the boundary and initial conditions in the
point of discharge in the river. The discharge point in the main channel has a fixed discharge level of 1.5m;
this value was changed to 1m, in accordance with data about the mean water level in the river for the last
three decades (from information obtained from the Port of Montevideo).
28
CHAPTER 5
Data analysis and model set up
In this chapter the steps followed to analyzing and processing data and to develop and calibrate the model
are presented. Among data processing activities the definition of different physical and hydrological
parameters is developed, as well as the study of rainfall data. The calibration of the model is performed and
statistical indicators are used to evaluate the results.
5.1. Data analysis

5.1.1. Data collection
During the data collection as part of the research a field visit was conducted. The main activities performed
were the area visit; system recognition and stakeholders' interviews at the Municipality. The available
information includes topography data, physical characteristics of the network, historical rainfall data,
historical information about CSO discharges and floods events, a hydraulic model in SWMM, water quality
data in the coast, land use and soil information.
• Topography data comprises contour lines of 2m interval and several surface points that contain
manholes and surface elevation of different elements. This information was available as ArcGIS
files.
• Network characteristics: pipes geometry, location, slopes, elevation of different elements and all
the information necessary to know the system is available. This information was provided as
ArcGIS files and the main pipes are already included in a SWMM model.
• Hourly historical rainfall data of 17 years is also accessible. Also, information of 170 different
storms during 15 years, with 5 minutes interval is available. This data was delivered as Excel files.
• MM has information about several years of CSO discharges obtained from the Supervisory Control
and Data Acquisition (SCADA) system, and data about flood events registered in the catchment to
be studied.
• The hydraulic SWMM5 model covering a big area of the city was already built. Separation of the
study area from the rest of the system and verification of how this model was working were
necessary.
• Information about ten years of daily water quality analysis in the coast is available. These analysis
include mainly bacteriological and nutrients data. This information was presented through Excel
files and different reports summarizing results and tendencies.
• Land use, runoff coefficient and different characteristics of subwatersheds included in the area to
be studied were available. This information was provided as ArcGIS files.
• Information about soil was accessible from a geotechnical research done for the city 15 years ago.
5.1.2. Physical data processing
The information available was obtained from the Municipality and it was mainly in ArcGIS format. Next is
presented a list detailing which data was included in this group:
- Information about catchments and subwatersheds for the area covered by the entire
coastal sewerage system. This data includes area, runoff coefficients and dry weather
flows.
- Location and different physical data about drainage inlets, manholes and pipes. For
each manhole the ground level is known. For pipes, section characteristics, invert
level upstream and downstream and length are detailed. Also, a shape indicating flow
direction in pipes.
- One shape with roads data for the entire city.
- Contour lines every two meters for the entire city.
- Some shapes with census information, including population and houses density for
years 1996 and 2004 and estimations for year 2035.
- Data about each property: area, maximum percentage that can be occupied and
maximum height that can be built according with urban planning regulations.
- Low lines where creeks and streams were located before urbanization.
- Soil classification, including infiltration capacity and water table depth in different
parts of the city.
These data were provided in shapes covering the total city area. For this reason the information has to be
processed cutting all the shapes to represent the area of the coastal sewerage system (necessary to work
with the total area contributing to the main coastal collector) and Pocitos watershed. Also, all shapes were
put in the same system of coordinates: UTM (WGS 1984) 21S.
Further, using contour lines every 2m (Figure 5.1), data about surface levels for each manhole and element
of the sewerage system, raster and hillshade files are created (see Figure 5.2). With this information mean
slopes for each subwatershed were estimated using ArcGIS tools.
Figure 5.1 High resolution picture with streets (left) and contour lines each 2 m (right) for Pocitos watershed.
30
In addition, a high resolution satellite image was downloaded and processed using the maximum likelihood
classification (MLC) algorithm (see Figure 5.2). The result was used to estimate percentages of
imperviousness, runoff coefficients and Manning coefficients for each subwatershed. To apply the method
the four different surfaces presented in Table 5.1 were considered. Runoff and Manning coefficients were
defined for each different surface (see Table 5.1); these values were determined in accordance with
theoretical values extracted from Chow et al. (1988) and Chow (1959). Table 5.2 presents surface
percentages and parameters calculated for each subwatershed. The runoff coefficients used by the
Municipality in its model are also presented.
Table 5.1 Different surfaces and coefficients considered.

Runoff Manning
Surface Type Nmin Nmax
Coeff. Coeff.
Roof Impervious 75 0.023 0.011 0.035
Road Impervious 85 0.021 0.014 0.027
Green Pervious 40 0.120 0.025 0.500
Other Pervious 50 0.027 0.017 0.036
The value of runoff coefficient considered in the case of green surfaces was high because the city has a
high density of trees seen as green surfaces in the MLC method. In the group "other" surfaces such as
coarse pavements are considered.
Figure 5.2 Result of MLC method (left) and hillshade for slope processing (right) in subwatershed AYAC.
Table 5.2 Percentages of different surfaces and parameters estimated for each subwatershed.
5.1.3. Rainfall data processing

An important step to define the processes of optimization and results evaluation is the selection of the
rainfall event to be used for design purposes. The selection of the rainfall event considers the type of
problem to be solved, the methodology used to design the solutions and the characteristics of the BMPs. It
is also important the selection of events to evaluate the effects obtained from applying these measures.
Based on literature studies and the analysis of different cases presented in Section 2.4 it was concluded that
in general several events with different return periods or long term rainfall data are commonly used to
design and evaluate BMP systems. However, the use of any of these methods in optimization processes
increases the computational time and could be a constraint to evaluate several different scenarios in the
time framework of this research. For this reason it is proposed to use a single event for the design
methodology using optimization.
According with the experiences of the Urban Drainage and Flood Control District (2010) and Huber et al.
(2006) for hydrological studies they define the Water Quality Control Volume analyzing different rainfall
events and its effects. In their study they try to identify one event that includes high percentage of runoff
production events; designing for this event they are covering most of the cases in which runoff is generated.
They divide rainfall events in three groups, the first one with very low depth that is not producing runoff; a
second group with low depth as well that produces a high percentage of runoff events; and the last group
with high depth producing a low percentage of runoff events. They design to cover the second group of
storms that even being micro storms include 80% of runoff producing events.
Following a similar analysis in this research, nine summers of the 17 years of hourly rainfall data were
studied to find out the relation between return periods of rainfall events and their effects on CSO. The first
step consisted in constructing the intensity-duration-frequency curves using the 17 years of rainfall data.
Secondly, a set of summer periods (Dec-Jan) were selected from the 17 years of data to develop the
frequency analysis. Finally, the nine summer periods were studied to identify the duration and return period
of each storm. The SWWM model was run using continuous simulation corresponding to a time series of
three months of rain for each summer period. The outcome of the model was used to analyse the CSO's
discharges frequency. This analysis is presented in next sections.
32
IDF curves development:
To develop the intensity-duration-frequency (IDF) curves a database with 17 years of hourly rainfall data
records was used. It is necessary to work with extreme values of measured rainfall data and fit a probability
distribution to calculate the frequency of rainfall events. Using this extreme values the rainfall intensities
for different return periods and different durations are determined (Pathirana 2013). There are three
different extreme values distributions used to analyze frequency of rainfall events: Gumbel or Type I,
Frechet or Type II and Weibull or Type III (Chow 1964; Maidment 1992). For this study the Gumbel
distribution was used to develop the IDF curves for events with nine different durations.
The followed procedure consisted in the identification of the maximum values of rainfall intensity for each
duration and for each year. After, these maximum values for each year were ranked from minimum to
maximum.
Cumulative probability (CP) of Gumbel distribution was calculated as:
( )/
= , where = (1 − ( > )) is the cumulative probability and ( > ) the probability
of exceeding the extreme rainfall r; a and b are parameters changing for each case and r is the rainfall
intensity.
Taking the double logarithm of CP equation a linear relation between the extreme rainfall r and the double
logarithm of exceeding probability can be achieved:
= − × log {−log [(1 − ( > ) } (5.1)
Annex A shows one example of calculation using this procedure.
Table 5.3 and Figure 5.3 show the result obtained from the process of these data for nine different durations
and seven different return periods.
Table 5.3 Maximum intensities expected for different event durations

(d, in hours) and return periods (Tr, in years).
d (h)\Tr (y) 1 2 5 10 15 20 50
1 15.5 19.8 28.4 40.6 59.1 79.1 93.8
2 13.9 15.8 19.7 25.2 33.6 42.6 49.3
3 12.1 13.4 15.9 19.6 25.2 31.3 35.7
4 9.6 10.7 12.8 15.9 20.5 25.4 29.1
6 6.7 7.6 9.3 11.9 15.8 19.9 23.0
8 5.3 6.1 7.6 9.8 13.0 16.5 19.1
12 4.2 4.8 5.9 7.5 10.0 12.6 14.5
18 3.3 3.7 4.5 5.7 7.4 9.3 10.6
24 2.4 2.7 3.4 4.5 6.0 7.7 8.9
IDF curves
100.0
90.0
80.0
70.0
Tr=1
60.0
Tr=2
i (mm/h)
50.0 Tr=5
Tr=10
40.0 Tr=15
Tr=20
30.0
Tr+50
20.0
10.0
0.0
0 5 10 15 20 25 30
Duration (h)
Figure 5.3 IDF curves.
To continue the analysis for rainfall event selection it is needed to run the hydrodynamic model. In next
Section the model set up is developed and in Section 5.3 the rainfall event is selected.
5.2. Model set up

5.2.1. Hydraulic model adjustment
The first step in processing the hydraulic model obtained from MM was to cut the model to represent only
Pocitos watershed. The original SWMM model provided by MM included the total sewerage system for the
East part of the city. The model had 243 conduits and 257 nodes, covering a total watershed area of
3470ha. After cutting out the model it had 60 conduits and 63 nodes, with a total watershed area of 575ha.
Table 5.4 presents the main characteristics of original and local models and Figure 5.4 shows the total and
local models in SWMM.
Table 5.4 Main characteristics of total and modified models.

Elements Original model Watershed model
Junctions 257 63
Outfalls 13 3
Storages 5 1
Conduits 243 60
Pumps 9 1
Weirs 25 9
Inflow points 95 19
34
Figure 5.4 Models of total coastal system and Pocitos watershed.
del was cut in two nodes of the main collector pipe. To define the boundary conditions of
The complete model
the reduced
ced model the representation of the same situation than for the complete model was established.
Upstream, a hydrograph representing the flow coming from the watershed contributing to the major
collector was added as an inflow in the upstream node.
node Downstream, ann outfall with free discharge was
added replacing the downstream node.
A separate study was developed to define the hydrograph representing the upstream contribution to the
main collector. The total watershed located upstream Pocitos has an area of 2352ha.
2352ha. The percentage of
imperviousness was estimated using the runoff coefficient of each subwatershed provided by MM as data.
The total runoff coefficient was calculated considering each value and the area of each subwatershed, the
average value calculateded was 44%. Peak flow obtained applying the hydrological simulation to this
watershed in SWMM is about 8 times higher than the peak flow entering to Pocitos subwatershed from
upstream in the original model. This is because the hydrological model was not representing
representing several outfall
discharges located in other parts of the coastal sewerage system. To take into account this situation, several
watershed parameters were adjusted until obtaining a similar hydrograph than that obtained in the complete
hydraulic model. Figure 5.5 shows both hydrographs, from hydrological simulation in SWMM (black) and
from modelling the total system (grey).

objective evaluation of BMP system configurations for CSO reduction 35
Figure 5.5 Hydrographs from subwatershed located upstream Pocitos, modelling
modelling total system (grey) and
trough hydrological model (black).
The hydraulic model includes a simplified system through which the total sewerage system in Pocitos
watershed is represented. This part of the city has 100% of sewerage coverage with small pipes pipe in each
street giving the service to every single house; this characteristic defines a complex system that cannot be
easily represented in a hydrodynamic model. In this simplified model only main pipes are considered.
Figure 5.6 shows the real network in the study area and the simplified network used in SWMM.
Figure 5.6 Real sewerage network (left) and simplified modelled network (right) in Pocitos watershed.
36
5.2.2. Hydrological model addition
The original hydraulic model from MM had the runoff hydrographs added as inflows in different nodes, the
rainfall-runoff model is done outside SWMM. In this work it was necessary to build a hydrologic model in
SWMM to represent the BMP elements that have impacts at source level. After the development of the new
model, the adjustment of different parameters was done to reach the same runoff hydrographs used in the
MM's model.
The hydrological model was built using the parameters obtained from data processing and presented in
Table 5.2. Fourteen subwatersheds were added into the model, representing the same subwatersheds used
by MM and presented in chapter 4.
Figure 5.7 shows the final hydrodynamic model in SWMM. In total 15 subwatersheds were added,
including the representation of the upstream system as a watershed explained in section.
Figure 5.7 Final hydrodynamic model of Pocitos watershed.
5.2.3. Calibration
Since MM is not measuring values in the sewerage system, a real calibration process is not possible. For
this reason calibration process in this research consisted in the adjustment of different parameters to have
runoff hydrographs generated in each subwatershed similar to those used by MM in its model. Calibration
procedure was based in trial and error approach changing different parameters to obtain the searched result.
In this process imperviousness percentage, Manning coefficients, slope, width and percentage routed to
pervious area were considered as calibration parameters.
Other parameters needed to build the hydrological model are the depths of depression storage in impervious
and pervious areas, and the infiltration method to be used. Values for depths of depression storage were
taken from USEPA (2010), this manual defines ranges of values between 1.3 and 2.5mm in impervious
cases and from 2.5 to 7.6mm in previous cases. This parameter was not used as a calibration value; fixed
values considered were 2mm for impervious cases and 4mm for pervious surfaces.
The method considered to represent infiltration processes is the Curve Number, values of parameters
needed for this method are taken form USEPA (2010). These parameters were selected considering the
information about soil classes in the study area. Based on the soil information given by MM it was
concluded that soils in the city area have in general a low infiltration capacity, for this reason the values of
curve number used were between 84 and 98, with one value of 70 corresponding to subcatchemnt BATL
that is a park area.
Regarding calibration parameters, final Manning coefficients used are higher than values previously
estimated with the MLC method; the main reason was the necessity of increasing peak times to adjust
results to MM runoff hydrographs.
Slopes were estimated using ArgGIS and contour lines every 2m; final values used after calibration were
lower than calculated slopes, this is in accordance with original values used by MM.
The width values were estimated for each subwatershed dividing the subwatershed area by the longest flow
path (USEPA 2010) for first estimation and defining a range of possible values for calibration process.
Final values of flow width change from initial estimated values; this was expected because this is one of the
least physically-based parameters in SWMM and for this reason one of the main calibration variables
(Tikkanen 2013).
Difficulties were found to adjust the peak time in the adjustment process to obtain the same result from
hydrologic SWMM model than the used by MM. The MM model uses a unitary hydrograph with a delayed
peak. To adjust the SWMM model trying to obtain similar results than those obtained from MM's model,
this effect was represented using the option in SWMM of routing the runoff from impervious areas trough
pervious areas. For this reason the percentage of impervious area that is routed trough pervious surfaces
was also considered as a calibration parameter.
Figure 5.8 shows the result from one adjusted subwatershed and Table 5.5 presents final values of
calibration parameters and hydrographs calibration results. In Annex B results for each subwatershed are
presented. Also, comparative results for the main discharge channel, secondary weir and the pipe
connecting the catchment with the rest of the system downstream are showed in Annex B.
38
Figure 5.8 Runoff hydrograph from EACU subwatershed (black) and runoff considered in MM model (grey).
Table 5.5 Calibration parameters final values and calibration results.

Peak Flow Peak Time Peak Flow Peak Time
Subcatch. % Imp Nimp Nper S (%) W (m) Targeted Targeted Simulated Simulated
(l/s) (hr) (l/s) (hr)
EACU 76 0.090 0.600 3.9 1680 10200 1:33 10205 1:30
FELI 59 0.031 0.169 2.5 350 3860 1:30 3858 1:30
FRAN 64 0.040 0.180 2.3 563 6300 1:36 6301 1:36
BATL 10 0.028 0.160 3.5 780 1630 1:42 1630.5 1:45
MFRA 80 0.061 0.292 3.0 528 3820 1:24 3818 1:15
HAED 83 0.055 0.380 2.2 652 3700 1:21 3700 1:15
CIPR 87 0.040 0.250 3.0 665 7360 1:24 7359 1:15
ALEJ 85 0.050 0.255 3.7 655 5090 1:21 5090 1:15
BUCE 70 0.035 0.185 2.7 281 2960 1:21 2962 1:21
ECHE 72 0.045 0.270 2.9 1464 11460 1:27 11461 1:21
BUXA 80 0.030 0.182 2.9 305 2940 1:18 2939 1:15
BAUZ 67 0.033 0.193 3.3 479 5050 1:24 5052 1:21
CIAN 72 0.038 0.230 1.8 550 3890 1:24 3890 1:21
AYAC 77 0.031 0.177 2.1 774 10470 1:30 10467 1:27
To evaluate results from calibration process, different

di statistics parameters are calculated to compare results
from hydrological model with hydrographs used by MM (Table 5.6).
The first parameter calculated is the Nash-Sutcliffe

Nash model efficiency (NSE).. It compares the mean square
error of obtained values
ues with the variance of reference data. It ranges from -∞ ∞ to 1; NSE=1 represents
perfect match and NSE<0 means the mean of reference values is better predictor than results obtained from
the model. The second parameter calculated is the coefficient of determination
det (R2); it represents the total
variance outcomes that can be explained by the model. This coefficient varies between 0 and 1, values
higher than 0.7 are acceptable (Paredes 2013).
2013) Finally, three parameters representing differences in runoff

peak flow (PEP) and runoff volume (PEV6 and PEV3) are presented; these parameters represent the most
important outputs of the hydrodynamic modelling.
Two different values of volume are considered in this comparison. The first one represents the volume of
runoff after 6 hours moment in which the rainfall finishes. The second one compares runoff volumes 3
hours after the beginning of rainfall; in general at this time around 70% of the runoff volume generated at 6
hours is already produced.
Table 5.6 Goodness of fit parameters.

2
Subcatch NSE R PEP (%) PEV6 (%) PEV3 (%)
EACU 0.77 0.94 0.06 32 19
FELI 0.48 0.90 -0.05 56 34
FRAN 0.52 0.90 0.02 53 28
BATL -0.60 0.76 0.03 99 30
MFRA 0.86 0.94 -0.04 12 8
HAED 0.87 0.94 0.01 3 1
CIPR 0.86 0.92 -0.01 3 0
ALEJ 0.83 0.90 0.00 -5 -6
BUCE 0.70 0.92 0.06 37 27
ECHE 0.86 0.94 0.01 12 6
BUXA 0.85 0.92 -0.03 -1.2 -2
BAUZ 0.78 0.94 0.03 29 20
CIAN 0.81 0.94 0.00 25 17
AYAC 0.79 0.94 -0.03 31 18
Analyzing the results presented in Table 5.6, can be observed that subwatershed BATL has not as good
fitness as other subwatersheds. BATL is formed by a park area that covers around 10% of Pocitos
watershed area. The hydrograph obtained trough hydrological model was accepted as a good result with
higher times in hydrological process than other subwatersheds, which is in accordance with higher
Manning coefficients in a green area. For calibration process in this case the methodology was the same
than in other cases, the adjustment of peak flow value and peak time, accepting higher differences in runoff
volume results.
5.3. Design storm selection

5.3.1. Rainfall-CSO analysis
As already explained in Section 5.1, the next step in the analysis to select the rainfall event for systems
designing is the processing of continuous rainfall data. This procedure takes long time, for this reason some
of the 17 available summers were selected for this analysis. The analysis was done for periods of three
months (summer) using the urban drainage model. The criterion to select which time series to use in this
study consisted in choosing summer periods with average rainfall values. For this selection mean daily
rainfall, rainfall peak and total rainfall values were analyzed for each summer. Grey rows in Table 5.7 and
darker columns in Figure 5.9 indicate the selected summer events.
40
Table 5.7 Rainfall data analysis.
Summer Mean (mm) Peak (mm) Total (mm)
1990/1991 0.12 24.3 261
1991/1992 0.051 6.9 111
1992/1993 0.187 27 405
1993/1994 0.097 47.7 210
1994/1995 0.109 24.2 235
1995/1996 0.073 15 160
1996/1997 0.079 10.5 171
1997/1998 0.181 20 391
1998/1999 0.246 63.1 532
1999/2000 0.050 10 109.9
2000/2001 0.164 40.4 355
2001/2002 0.107 38.5 231
2002/2003 0.131 20.8 283
2003/2004 0.126 51.6 275
2004/2005 0.155 26.1 334
2005/2006 0.123 17.5 265
2006/2007 0.128 33.4 277
Figure 5.9 Analysis of measured rainfall data; darker years are selected for analysis.
For each one of the nine summer periods selected, the developed urban drainage model in SWMM5 was
run using a continuous time series of three months of hourly rainfall data. The period selected for each year
covers the months of December, January and February. The results of these simulations were used to
process each storm. The analysis included the calculation of duration, maximum intensity and return period
of each event. To estimate the return period of each event the IDF curves developed in Section 5.1.3 were
used.
The CSO effects associated with each storm were also analyzed. From this study the time between rainfall
peak and discharge peak, the peak flow discharged and total time of discharge were calculated. The main
discharge channel has a base flow resulting from dry weather flow that varies between 1 and 4 l/s. To
analyze discharges in this channel the minimum value considered as CSO discharge was 10 l/s.
To develop an Intensity-Duration-Frequency analysis the annual maximum events were considered,

consequently the analysis is developed for return periods equal or larger than 1 year. This approach cannot
be used to analyze storms with return period lower than one year, but these small storms are critical in the
analysis of overflows from a combined system (Price and Vojinovic 2011). In this case, through the
analysis of summer rainfall events and CSO discharges, rainfall events were separated in accordance with
return period values. The IDF curves developed are used to evaluate the return period of each storm, and
they were grouped in less than 1 year return period when the frequency cannot be determinate using this
methodology.
Table 5.8 shows the results obtained from the analysis of rainfall-CSO processes for the nine summer
periods considered in this assessment. From these results it can be appreciated that in average 78% of the
cases with CSO are consequence of rainfalls with 1 year or shorter return period. This result shows the
importance of storms with return period lower than or equal to 1 year in the generation of CSO's (in Annex
C one example of storm analysis and the complete study of one summer are presented).
Table 5.8 Results of continuous summer rainfall modelling analysis.

Summer Summer Summer Summer Summer Summer Summer Summer Summer
90/91 93/94 94/95 01/02 02/03 03/04 04/05 05/06 06/07 Average
Number of days with rainfall 24 26 17 24 36 27 25 24 27 26
Number of storms 33 37 22 44 51 35 35 34 33 36
% of events producing CSO
1 58 54 73 45 55 69 46 56 58 57
(>10 l/s)
% of events producing CSO and
2 68 90 69 85 78 79 87 63 84 78
with Tr <= 1 year
Average storms duration (h) 4 4 6 4 4 4 5 4 3 4
Average time between rainfall
2:47 2:35 2:11 2:29 2:17 2:21 2:23 2:23 2:15 2:24
and discharge peaks (h)
Average time of CSO discharge
17 15 18 16 17 16 16 17 15 16
(h)
3
% of days with CSO 18 19 15 18 25 23 16 20 20 19
4
% of raining days with CSO 70 68 76 64 65 88 58 75 72 71
1
Number of events producing CSO / Total number of events during summer
2
Number of events producing CSO and with Tr<=1 / Number of events producing CSO
3
Number of days with CSO / Total number of summer days
4
Number of days with CSO / Number of raining days during summer
42
Other results obtained from this analysis were average storm durations and lag times (average times
between rainfall peak and discharge peak), also presented in Table 5.8. These values are useful to select the
design event duration. The storm duration should be long enough to allow the contribution of the entire
area to the discharge point, for this reason the storm duration is defined based on the time of concentration
of the total watershed (Marek 2011). Another consideration to select storm duration is to follow
specifications defined by the agency having jurisdiction in the area where the project is developed. Finally,
rainfall records can be used to define typical storms durations in the area of work (Durrans and Dietrich
2003). In this case the average duration obtained from rainfall records analysis was 4 hours; this value is
almost two times higher than the average concentration time of the total watershed (presented as time
between rainfall and discharge peaks). MM uses a design storm with 6 hours duration and return period
more than 2 years. The use of these events is focused in designing drainage system with the main objective
of avoiding flooding.
Based on the arguments discussed above, the selected rainfall event for the optimization processes
corresponds to a design storm of 1 year return period and 4 hours of duration.
From the rainfall-CSO analysis it is observed a high influence of accumulated rainfall in CSO discharges.
Based on this observation the evaluation of the designed systems' effects was done using continuous series
of rainfall data. Running continuous simulations with the application of selected BMPs configurations and
comparing with the results obtained running the same time series for the current situation, the impact of
different solutions on CSO during summer periods can be evaluated.
5.3.2. Design storm calculation

The selected design storm has one year return period and 4 hours of duration. To develop this storm the
Alternating Block Method is used. This is a simple method to develop hyetographs from an IDF curve. The
result is a hyetograph with precipitation depths for each successive time interval. The first step is to select
the return period, after that the intensity for each incremental duration can be read from the IDF curve
(Chow et al. 1988).
From the IDF analysis the equation for one year return period curve is estimated as:
= 56.5%(" + 2.65) (5.2)
Where i is the intensity in mm/h and d is the accumulated time in h.
Using this equation and duration periods of 20 minutes the storm is developed as it is showed in the Table
5.9 and Figure 5.10. The total cumulative depth for this storm is 34mm.
Table 5.9 Design storm development.
Accum.Time Intensity Cumulative Incremental Time interval Precipitation
Interval
(min) (mm/h) depth (mm) depth (mm) (min) (mm)
1 20 18.9 6.3 6.3 0-20 1.3
2 40 17.0 11.4 5.0 20-40 1.7
3 60 15.5 15.5 4.1 40-60 2.2
4 80 14.2 18.9 3.4 60-80 2.9
5 100 13.1 21.8 2.9 80-100 4.1
6 120 12.2 24.3 2.5 100-120 6.3
7 140 11.3 26.5 2.2 120-140 5.0
8 160 10.6 28.3 1.9 140-160 3.4
9 180 10.0 30.0 1.7 160-180 2.5
10 200 9.4 31.5 1.5 180-200 1.9
11 220 8.9 32.8 1.3 200-220 1.5
12 240 8.5 34.0 1.2 220-240 1.2
Figure 5.10 Design storm.
5.3.3. Evaporation conditions

Evaporation is considered because it has influence in the BMPs performance obtained from SWMM
modelling. In the case of single event simulation to be used in optimization process, an average summer
value is used. Later, during the evaluation of results with rainfall time series, monthly values will be
considered.
To determine evaporation rates, monthly values from the National Meteorological Direction of Uruguay are
used. The mean values of Tank A evaporation for December, January and February are 6.8, 7.1 and
6.0mm/day respectively. The potential evapotranspiration is estimated from these values using the
evapotranspiration coefficient that has a value of 0.8 for the case study area (Genta and Charbonnier 2013).
The average evaporation value to be used in the single storm simulation is 5.3 mm/day.
44
CHAPTER 6
Best Management Practices
The general tool developed to select BMPs according to local conditions and the steps followed to define
the set of applicable measures in the case study are presented in this chapter. After selecting the applicable
BMPs, they are designed and total values during the lifespan are calculated. Also, the coverage limitations
for each subwatershed in the study area are studied. Finally, the steps followed to include these elements in
the developed hydrodynamic model are described.
6.1. BMPs selection

To develop a methodology to be used in BMPs selection it is necessary to undertake a detailed study
different BMPs options. Through this study the limitations and advantages of each alternative are
understood, this is useful in particular for application in decision making processes. Twelve of the most
used BMPs were studied; this analysis includes: characteristics, application possibilities, functions or
capabilities, constraints, effectiveness and costs. Annex D presents the results of this study.
To define a methodology for BMP selection the different characteristics analyzed were taken into account.
First, the capabilities and functions were considered to group the different possible BMPs in accordance
with the main pursued objective. The main objective can be: runoff reduction, water quality improvement
or both together. Second, the major physical constraints that limit BMPs application were considered: the
size of the drainage area, slope of the placement place, type of soil and depth of groundwater. Third, other
considerations were analyzed, such as space availability, implementation and maintenance costs, aesthetic
benefits and public safety. These elements were not included in the decision tool to avoid complexity.
Figure 6.1 illustrates the BMP selection method in the case of runoff reduction as main objective. In the
same figure, the path followed in this case can be observed. As a result, the possible BMPs to be applied in
this case study are infiltration trenches with underdrain system, pervious pavements with underdrain
system, green roofs and rainfall harvesting.
Figure 6.1 BMPs selection decision tree for runoff reduction.
To recognize each BMP in the diagram the numbers in the figure can be connected with the numbers used
in Annex D. The first criterion is drainage area (DA); the second one is the slope (Sl); the next one is the
type of soil according to NRCS hydrologic soil groups; the final factor considered is the depth of
groundwater (GW). Cases marked with asterisk indicate the necessity of an underdrain system.
46
After the application of this tool a deeper analysis has to be done to define exactly which measure can be
applied in each subwatershed and to what extent. As a general criteria after interviews with some
stakeholders (people from MM involved in drainage projects and urban planning) particular measures in
every single house were avoided due to difficulties in regulation and control processes. For this reason the
methodology followed to identify BMP applicability consists in searching for big areas for green roofs
development, and public places to locate infiltration trenches and pervious pavements. Rainfall harvesting
devises were discarded because its implementation in individual houses requires high regulation and
control measures.
As an example, the study of one subwatershed is presented, the rest of the cases are explained in Annex E.
The area selected is the subwatershed EACU with a surface of 82ha, where 70% of the area is private and
30% is public (this is a common characteristic in residential neighbourhoods in Montevideo). To apply the
BMP selection methodology it is observed that the area is highly urbanized and not areas bigger than 2ha
are expected to be drained into BMP systems. The average slope is 2.8%; the soil has low permeability; and
no shallow groundwater is detected in this zone. Applying the method described in the diagram, four
different BMPs are possible: infiltration trenches or pervious pavements with underdrain system, green
roofs and rainwater harvesting. As in the rest of the cases, rainwater harvesting was discarded due to local
conditions related with regulations and controls needed.
In this particular subwatershed, the presence of 2600m length of wide roads with a central linear green
space was identified; in these places infiltration trenches can be located. Using this practice 10% of the
total area can be drained into this BMP. Also, several big roofs were identified covering 7% of the area as
first estimation, where the application of green roofs is possible centralizing the maintenance and regulation
of this practice. Figure 6.2 shows the location of these BMPs in the studied area.
Figure 6.2 Possible infiltration trenches (red lines) and green

roofs (orange areas) locations in subwatershed EACU.
6.2. BMPs design and costs
6.2.1. Pervious pavements
Three different types of pervious pavements can be considered: porous asphalts, permeable interlocking
concrete pavement or pervious concrete. The main designing characteristics of each case are presented in
Table 6.1.
Table 6.1 Pervious pavements characteristics (USEPA 2013b, University of Maryland 2014).
Permeable Interlocking
Porous Asphalt Pervious Concrete Pavement
Concrete Pavement
Thickness (mm) 50 - 100 60 (pedestrian) - 80 (roads) 100 - 200
Voids of pavement (%) 16
Chocke course thickness (mm) 25 - 50 50 25 - 50
Chocke course material Small sized aggregate. Small sized aggregate. Small sized aggregate.
Base reservoir thickness (mm) 75 - 100 75 - 100 75 - 100
Base reservoir material Crushed stones of 5 to 20 mm Crushed stones of 5 to 20 mm Crushed stones of 5 to 20 mm
Subbase reservoir thickness Depends on storage. Not Depends on storage. Not Depends on storage. Not
(mm) required in pedrestian app. required in pedrestian app. required in pedrestian app.
Subbase reservoir material
Stones of 20 to 60 mm. Stones of 20 to 65 mm. Stones of 20 to 65 mm.
(mm)
Underdrain Perforated pipe. Perforated pipe. Perforated pipe.
Geotextile Optional. Optional. Optional.
Slopes > 2%, terracing base is Slopes > 2%, terracing base is Slopes > 2%, terracing base is
Limitations
needed. needed. needed.
Maintenance Vacuum sweeping. Vacuum sweeping. Vacuum sweeping.
Lifespan (years) 30 20 - 40
The design of the pervious pavement to be used in this case considers a thickness of 10cm for the
pavement, 0.3m of aggregate for base and sub base (storage) and perforated pipes as underdrain system.
Existent pavement removal was considered in the cost calculation because the measure is applied in an
urbanized area.
To facilitate the drainage of filtration media, perforated pipes of underdrain system should not be separated
more than 1.5m between centres (Melbourne Water 2005). In this case one meter of 100mm pipe for each
square meter of pervious pavement was considered.
To calculate the unitary investment cost for each BMP, current costs proportionate by MM were used.
Pervious pavements cost was not included in this information. In accordance with California Stormwater
Quality Association (2003) pervious pavements are 15% cheaper than normal pavements, but from
Metropolitan Area Planning Council (2010) permeable pavements costs more than conventional
pavements. For this study the same cost than conventional pavements was considered. Table 6.2 shows
different items with unitary costs and total investment cost for each square meter of pervious pavements to
be considered in this application case.
48
Table 6.2 Pervious pavements cost calculation.
PERVIOUS PAVEMENTS
2
Task Cost Quantity/m Unitary cost (Eu)
2 2
Pavement remotion 8 Eu/m 1m 8
3 3
Excavation 25 Eu/m 0.3 m 8
3 3
Aggregate base 23 Eu/m 0.3 m 7
Underdrain 20 Eu/m 1m 20
2 2
Pavement 43 Eu/m 1m 43
2
Unitary Pervious Pavement Cost (Eu/m ) 85
The annual maintenance cost of pervious pavements is estimated between 15% of investment cost
(California Stormwater Quality Association 2003) and 2% of investment cost (Metropolitan Area Planning
Council 2010). In this case the maintenance cost to be considered is 5% of the investment cost.
6.2.2. Infiltration trenches

To design infiltration trenches an average drainage area contributing to each meter of trench is estimated.
From this estimation and the total rainfall depth of the design storm, the needed volume to be stored is
calculated.
The available area to locate this kind of BMPs is in the linear green spaces placed in the middle of the wide
main roads (Figure 6.3 shows examples of these areas). Considering a total of 30m width for these roads,
with 67% of pavement and 33% of sidewalks, and runoff coefficients of 0.85 and 0.70 for road and
sidewalk respectively, the runoff coefficient of the drainage area is 0.80.
Figure 6.3 Examples of available spaces for infiltration trenches location.
The total rainfall depth of the design storm is 34mm, and the void ratio (VR) of aggregate reservoir is 67%
(Tetra Tech 2010) (calculated using the equation: Porosity = VR / (VR + 1)).
The storage volume for each meter of trench can be calculated as:
V% = ) * + + ,- .,/01 × 0-,22 , 22 -+ % (6.1)

m 3, " +,
3411%
V% = [4 10008 × 301 × 0.80<
m
0.67 = 1.22 1 %1
;
The infiltration trench was designed with 1m depth and 1.2m width based in the calculation of the storage
volume needed. Figure 6.4 shows the section of an infiltration trench (without underdrain).
The design of the underdrain system follows the same criteria used in the pervious
perviou pavement case. The
result is one meter of 100mm perforated pipe for each meter of trench. In this case the element has an
external drainage area, in the case of pervious pavement the drainage area is equal than the pavement area.
As a result the flow generated
erated for each meter of pipe is higher in the case of trenches, for
f this reason the
capacity
acity of the pipe is verified. The
The maximum runoff generated for each meter of pipe is 13 1 l/s (30m2 of
drainage area and runoff coefficient of 0.8). Considering 100mm diameter,
diameter, 1% of slope and 0.013 of
Manning coefficient, the maximum capacity of the pipe is 26 l/s that represents the double of the capacity
needed.
Figure 6.4 Infiltration trench section (Source: Tetra Tech 2010).

).
Table 6.3 shows the calculation

tion of investment cost for each linear
linear meter of infiltration trench.
trench
Table 6.3 Infiltration trenches cost calculation.

INFILTRATION TRENCH
2
Task Cost Quantity/m Unitary cost (Eu)
3 3
Excavation 25 Eu/m 1.2 m 30
3 3
Aggregate base 23 Eu/m 1.2 m 28
Underdrain 20 Eu/m 1m 20
2
Unitary Pervious Pavement Cost (Eu/m ) 78
The annual maintenance cost for infiltration trenches is estimated between 5 and 20% of the investment
cost. Higher values are recommended to ensure long-term
long functionality (USEPA 2013b).
2013b) The value
assumed for this case was 15% of investment cost.
ifespan of infiltration trenches is estimated to be between 10 and 15 years (U.S. Department of

The lifespan
Transportation - Federal Highway Administration 2013).
2013)
50
6.2.3. Green roofs and storages
In the case of green roofs and storages, values used by Urrestarazu (2013) were taken as reference. He used
extensive green roofs with 104 Eu/m2 of unitary investment cost; annual maintenance cost equal to 5% of
investment cost and lifespan of 30 years.
Regarding to storages, current costs of projects for implementing underground storages in the city were
considered. The investment cost for storages is 1446 Eu/m2 and the annual maintenance cost considered
here is 0.5% of investment cost (in accordance with suggestions from MM based on present experiences).
Following the recommendation of Gironás et al. (2010) the methodology used in this work to design
storages consist in the use of SWMM to simulate storages with orifices as outlets and weirs as overflows.
In each case the depth of storages was fixed in 3m (except in BATL case where it is 1.5m because the
storage is not underground), and the surface area was fixed equal to the maximum available area for each
subwatershed. The dimensions of the orifices were determined running SWMM several times and changing
the orifices size iteratively until the reduction of peak flow for the design storm was maximum (without
using the weirs).The sizes of the weirs were determinate for the event of 10 years return period, running
SWMM several times while changing the weir size and height, avoiding the fullness of the tank.
6.2.4. Total present value

The total present value of a project is the sum of investment cost and the present worth of operation and
maintenance (O&M) costs required during the working period considered for the element. The present
value of annual O&M costs is the quantity that should be invested now at a specified interest rate to pay the
O&M costs every year during the life of the element (Sharma et al. 2013). The calculation of total present
value (Giddens 1997) considers the same period of 20 years for the four different practices. The interest
rate used is 5%, that is the real recommended value in Uruguay (Urrestarazu 2013). Table 6.4 shows the
results of total present value per square meter for each practice. In Annex F the detailed calculation is
presented.
Table 6.4 Total present worth for each practice.
2
BMP Total Present Value (Eu/m )
Pervious Pavement 138
Infiltration Trench 141
Green Roof 169
Storage 1536
6.3. BMPs representation in SWMM

BMPs in SWMM are simulated as a combination of layers and its properties are defined per unit of area
(USEPA 2010), Figure 6.5 shows this representation.
Figure 6.5 BMPs representation in SWMM (Source USEPA 2010).
Several parameters are needed to simulate BMPs using SWMM. To define the parameters for each practice
and for each layer, the values recommended by USEPA (2010) and Tetra Tech (2010) were used. Also the
values defined during the design of the BMPs (for instance the storage volume) and particular values for
the case study (such as very low soil infiltration capacity) were considered. Table 6.5 presents the different
parameters used in each case.
Table 6.5 SWMM parameters used in BMPs modelling

(Tetra Tech 2010; USEPA 2010; University of Maryland 2014).
PERVIOUS PAVEMENTS INFILTRATION TRENCHES GREEN ROOFS
Surface Surface Surface
Storage depth (mm) 5 Storage depth (mm) 5 Storage depth (mm) 8
Vegetation Volume Fraction 0 Vegetation Volume Fraction 0 Vegetation Volume Fraction 0.8
Surface Roughness (Manning) 0.02 Surface Roughness (Manning) 0 Surface Roughness (Manning) 0
Surface Slope (%) 2 Surface Slope (%) 0 Surface Slope (%) 0
Pavement Storage Soil
Thickness (mm) 100 Height (mm) 1000 Thickness (mm) 100
Void ratio 0.2 Void ratio 0.82 Porosity 0.453
Impreviouss fraction 0 Conductivity/Filtration rate 5 Fiels capacity 0.19
Permeability (mm/h) 200 Clogging factor 0 Wilting point 0.085
Clogging factor 385 Underdrain Conductivity/Filtration rate 11
Storage Drain Coeff. 2.17 Conductivity slope 10
Height (mm) 300 Drain Exponent 0.5 Suction Head 110
Void ratio 0.65 Drain Offset Height 0 Storage
Conductivity/Filtration rate (mm/h) 5 Height (mm) 50
Clogging factor 0 Void ratio 0.65
Underdrain Conductivity/Filtration rate 0
Drain Coeff. 1.1 Clogging factor 0
Drain Exponent 0.5 Underdrain
Drain Offset Height 0 Drain Coeff. 1.9
Drain Exponent 0.5
Drain Offset Height 0
To determine the voids ratio the equation suggested by USEPA (2010) is used. In accordance with which
the voids ratio (VR) is depending on porosity values, following the equation: , ,= +> = 3 /(3 + 1).
52
To define the underdrain parameters the equation recommended by USEPA (2010) is used:
? = ) × (ℎ − A)B (6.2)
, where q is the outflow in mm/hr, c is the drain coefficient, h is the height of stored water in mm, H is the
drain height in mm and n is the drain exponent.
The parameters required by SWMM are c and n; a typical value of n is 0.5 in which case the drain acts like
an orifice. To calculate c, a value of q equal than the maximum rainfall intensity from the design storm was
considered. In the case of pervious pavements, (h - H) is equal than the storage depth plus the pavement
depth minus the drainage pipe height (considered as 100mm). The value obtained for this depth is 300mm
and the value for c was 1.1 mm/h. In the case of green roofs the storage depth considered from the design
was 100mm, and the value of c obtained was 1.9 mm/h.
In the case of infiltration trenches a different criterion was followed. The difference in this case is that
while for pervious pavements and green roofs the drainage area is the area of the BMP; in the case of
infiltration trenches the area of runoff contribution is external. A different equation suggested by USEPA
(2010) is used in this case:
B
) = 2 × C %D, (6.3)
where D is the depth of stored water and T the time required to drain this water.
Considering a total rainfall depth of 34mm, 30m2 the drainage area per meter of trench, a runoff coefficient
of 0.8 and 1.2m2 of surface trench area per linear meter; the total depth of stored water is 680mm.
According to Idaho Department of Environmental Quality (2005), infiltration trenches should be
completely drained in 24 hours to ensure aerobic conditions. The drainage coefficient was calculated using
this time period, the result obtained was 2.17 mm/h.
To define green roof characteristics, as thickness and storage height, the recommendations of Conservation
Technology Inc. (2008) for extensive green roofs were followed. Soil characteristics were taken from
USEPA (2010) for sandy loam soils.
6.4. BMPs application

The incorporation of BMPs in the model of the study area is done adding combinations of the four practices
considered (pervious pavements, infiltration trenches, green roofs and storages) in each one of the 14
subwatersheds that integrate the Pocitos drainage model. Pervious pavements (PP), green roofs (GR) and
infiltration trenches (IT) can act in parallel treating different portions of the runoff generated.
After the application of these practices some subwatershed parameters have to be changed to compensate
the original area that is replaced by BMPs, Figure 6.6 explains this changes.
Figure 6.6 Subwatershed parameters adjustment after BMPs addition (Source USEPA 2010).
The parameters that need to be changed are the percentage impervious and the width; equations below
show how to calculate these new parameters.
IJKL G H %F × I MNO − (IPP + IJM + IQR ) (6.4)

%FG H = =
I MNO G H I MNO − (IPP + IJM + IQR )
I MNO G H I MNO G H I MNO I MNO − (IPP + IJM + IQR ) (6.5)

SG H = = × = ×S
T I MNO T I MNO
Where %I and %INew are the original and new values of percentage impervious. ATot and ATot-New are the
original and new values of total subwatershed area. AImp-New is the new impervious area of the
subwatershed. APP, AIT and AGR are the areas occupied by pervious pavement, infiltration trenches and
green roofs respectively. W and WNew are the original and new values of subwatershed width. Finally, L is
the value of maximum overland flow length.
Through the analysis presented in Annex E the maximum values of BMPs that can be located in each
subwatershed were identified. These maximum values are presented as percentage of total area in the case
of pervious pavements and green roofs, as length in the case of infiltration trenches and as maximum
surface area in the case of storages.
Table 6.6 shows these values for each subwatershed and for each BMP. This table also presents the original
and new values of percentage impervious and subwatershed width. The new values of these parameters are
calculated for the case of maximum application, in which all the possible BMPs and the maximum number
of each BMP of them applied.
54
Table 6.6 Maximum possible BMPs application.
Area % W PP PP Total Area IT Total Area GR Total Storages
Subcat. 2 IT (m) 2 GR (%) 2 % ImpNew Wnew
(ha) Imper (m) (%) (m ) (m ) Area (m )
v 2
EACU 81.8 76 1680 10 81800 2600 3120 7 57260 1400 71 1388
FELI 37.3 59 350 13 48490 950 1140 7 26110 0 49 279
FRAN 66.2 64 563 10 66200 2400 2880 5 33100 2800 57 476
BATL 50.1 10 780 9 45090 2600 3120 0 0 20000 0 705
MFRA 21.3 80 528 10 21300 200 240 5 10650 0 76 448
HAED 17.7 83 652 10 17700 200 240 5 8850 0 80 553
CIPR 36.6 87 665 10 36600 1000 1200 5 18300 0 85 563
ALEJ 21.7 85 655 10 21700 0 0 5 10850 0 82 557
AYAC 82.9 77 774 12 99480 250 300 7 58030 1800 72 627
CIAN 23.7 72 550 12 28440 950 1140 6 14220 0 66 448
BAUZ 33 67 479 12 39600 900 1080 7 23100 3600 59 386
BUXA 12.5 80 305 10 12500 0 0 5 6250 0 76 259
ECHE 71.5 72 1464 11 78650 1100 1320 5 35750 8400 67 1227
BUCE 20.1 70 281 13 26130 1400 1680 5 10050 0 63 228
The definition of PP, GR and IT coverage in each subwatershed in SWMM requires the designation of
different parameters. Figure 6.7 shows the editor and values of these parameters for the three BMPs in the
case of subwatershed EACU. To introduce these BMPs in one subwatershed it is necessary to specify the
surface area of each unit and the number of units. From these data the percentage of total subwatershed area
occupied with the practice is calculated automatically.
automa
Figure 6.7 BMPs application editor in SWMM.

The unitary area of pervious pavements was assumed equal to 1000m2; in the case of infiltration trenches
the unitary area used was 120m2 (1.2m width and 100m length); in the case of green roofs the area
considered for each unit was 600m2 (an average value for big roofs in the watershed).
In the case of PP it is needed to define the top width of overland flow surface for each unit because this
practice uses overland flow to convey surface runoff (USEPA 2010).
Finally, in the case of IT it is needed to define the percentage of impervious area that is treated. This value
is equal to the length of trenches multiplied by the unitary width of drainage area (30m in this case).
Similar than in the case of percentage impervious and subwatershed width, this parameter has to be
changed each time that this BMP coverage is changed; the equation that represents this modification is:
TJM × 120%
30 × 1.2
(6.6)
%IUKL VWP = 100 ×
%F × I MNO
Where 30m2 is the drainage area per meter of trench; LIT is the length of trenches; 120m2 is the area of each
unit of trench (1.2m wide and 100m long); 1.2 is the width of trenches; %I is the percentage of impervious
area and ATot is the total area of the subwatershed.
After all the parameters had been set the model was ready to run continuous time series simulations under
maximum BMPs application. Also, the optimization processes started, coupling this model with the
optimizer.
56
CHAPTER 7
Optimization
This chapter contains the description of the optimization process followed in this study to design the
different BMPs systems configurations. The optimization variables and objective functions are defined in
accordance with the case characteristics and the outlined targets. Finally, the different scenarios of BMPs
configurations for which the designing process is applied are defined.
7.1. Optimization process

The optimization of the BMPs was performed using a Genetic Algorithm technique. This technique
considers populations formed by individuals that are evaluated using a fitness function and ranked in
accordance with this. After, the best individuals are selected, and a new population is created through
crossover and mutation mechanisms. This routine is repeated until the stopping condition is satisfied.
Figure 7.1 shows this process, different colours indentify different stages in the process. The first stage is
the initialization and generation of the first population using random methods, the population number is
predefined. The first population is the only one randomly generated. The second stage uses problem-
specific criteria (defined by the objective functions); using these criteria the evaluation of each individual is
performed. Finally, the best individuals are selected and from them a new population is created. The
process is finished when the number of generations (also predefined) is accomplished. The result is a Pareto
front with the best individuals, containing the values of each variable.
In this case the optimization process was run with a population of 120 and 80 generations, using the
improved version of non-dominated sorting genetic algorithm (NGSA-II).
Figure 7.1 Optimization process.
7.2. Optimization variables

As was discussed in Section 5.4, to introduce PP, GR and IT in one subwatershed it is necessary to specify
the surface area of each unit and the number of units. The unit area of each BMP is fixed and the number of
units is used as optimization variable in this case. From maximum coverage areas for each BMP defined in
Annex E, and the unit area of each element, the maximum number of units that can be applied in each
subwatershed was defined. These are the upper limits for the optimization variables (presented in Table
7.1).
In the case of storages the variable is the unitary surface area, selected as 1 m2. Maximum values in this
case are the maximum available areas, also established in Annex E. SWMM5 can simulate storages using a
functional or a tabular storage curve. In this case functional storage curve is selected; this curve uses three
parameters: I = I × C *+ℎV + . The parameter considered as optimization variable for storages is A,
with the limits shown in Table 6.6; the parameters B and C are considered zero.
58
Table 7.1 shows the 48 optimization variables and its limit values. The column "% Imp treated" represents
the percentage of impervious area that is sent to IT in each subwatershed. This parameter and the method
applied for its calculation were defined in Section 6.4.
Table 7.1 Optimization variables.

GR Max GR Max IT % Imp A Max PP Storages
Subcat. Var. IT (m) Var. PP (%) Var. 2 Var.
(%) Units Units treated Units (m )
ALEJ 5 18 v1 0 0 v15 0 10 22 v29 0
AYAC 7 97 v2 250 3 v16 1 12 99 v30 1800 v43
BATL 0 0 v3 2600 26 v17 78 9 45 v31 3000 v44
BAUZ 7 39 v4 900 9 v18 12 12 40 v32 3600 v45
BUCE 5 17 v5 1400 14 v19 30 13 26 v33 0
BUXA 5 10 v6 0 0 v20 0 10 13 v34 0
CIAN 6 24 v7 950 10 v21 17 12 28 v35 0
CIPR 5 31 v8 1000 10 v22 9 10 37 v36 0
EACU 7 95 v9 2600 26 v23 13 10 82 v37 1400 v46
ECHE 5 60 v10 1100 11 v24 6 11 79 v38 8400 v47
FELI 7 44 v11 950 10 v25 13 13 48 v39 0
FRAN 5 55 v12 2400 24 v26 17 10 66 v40 2800 v48
HAED 5 15 v13 200 2 v27 4 10 18 v41 0
MFRA 5 18 v14 200 2 v28 4 10 21 v42 0
After defining the variables, the functions to change the percentage of imperviousness and the watershed
width, whenever the BMPs coverage is changed, can be written in a general form for each subwatershed.
For example, in the case of subwatershed EACU these equations are:
0.76 × 818000 − (1000 × .;X + 120 × .Y; + 600 × .Z ) (7.1)

%FG H =
818000 − (1000 × .;X + 120 × .Y; + 600 × .Z )
81800 − (1000 × .;X + 120 × .Y; + 600 × .Z ) (7.2)

SG H = × 1680
81800
Where 0.76 is the original fraction impervious, 818000 is the total area in m2, 1680 is the original width in
m, 1000 and v3 are the PP unitary area and number of units respectively, 120 and v23 are the unitary area
and number of units of IT, 600 and v9 are GR unitary area and number of units.
7.3. Objective functions

The objective functions selected for the optimization process are the minimization of total cost and peak
flow in the main CSO discharge channel.
To define the total cost, the costs for each unit of BMP has to be calculated. Table 7.2 presents the cost per
square meter of each BMP calculated as present value. From these values and the area of each unit, the
unitary costs are estimated, Table 7.2 shows the results.
Table 7.2 Unitary costs calculation.
Total Cost Area per Unit Cost per Unit
BMP 2 2
(Eu/m ) (m ) (Eu/Unit)
Pervious Pavement 138 1000 138000
Infiltration Trench 187 120 22440
Green Roof 169 600 101400
Storage 1536 1 1536
From these results, the cost function is:
,=+ [0-). = \ -( ]BUO^ × 138000 + FD]BUO^ × 22440 + _ ]BUO^ × 101400 + `+, a bc d × 1536) (7.3)
Where:
]BUO^ = .YZ + .;e + ⋯ + .gh + .gY (7.4)
FD]BUO^ = .hi + .hj + ⋯ + .YX + .Yk (7.5)
_ ]BUO^ = .h + .Y + ⋯ + .h; + .hg (7.6)
`+, a bc d = .g; + .gg + ⋯ + .gX + .gk (7.7)
The second objective function represents the peak flow minimization in the main channel and can be
written as:
`l [0-). = \ -( m [/,n - * * [Fo IoIT_\I 7) (7.8)
The peak flow values in the main discharge channel are obtained from the output of SWMM simulations,
using the Design Storm presented in Section 5.3.2 as rainfall event.
Figure 7.2 schematizes the interconnections in the drainage network close to the point of CSO discharges
into the river. Three main pipes are related at this point: the main coastal collector that conveys the water
coming from the upstream watersheds; the main pipe of Pocitos combined system that transport the dry
weather flow mixed with wet weather flow from Pocitos watershed; and the main pluvial pipe of Pocitos
watershed with only wet weather flow. The main collector and the combined system pipe discharge through
weirs into the main pluvial pipe when certain water level is reached, these are the CSO flows that are going
into the main channel and later into the river.
The reduction of runoff achieved due to the application of BMPs in Pocitos watershed decreases the flows
in the main combined pipe and the main pluvial pipe. For this reason the peak flow reduction is used in
optimization process to evaluate the effect in the main channel, where both pipes (combined and pluvial)
are discharging during rainfall events. This point is represented as point 1 in Figure 7.2.
Figure 7.2 also shows points where evaluation of the BMPs effects are done, for instance the point where
CSO frequency reduction is measured (point 2) and where runoff volume and peak flow are checked to
assess BMPs performance (point 1).
60
Figure 7.2 Schematization of the system close to the discharge point.
Boundary and initial conditions connecting the system to the upstream system, the downstream system and
in the discharge point into the river were defined in Section 5.2.
7.4. Scenarios to be optimized

The optimization process applied to design the different configurations of BMPs consists of six different
scenarios. Table 7.3 presents these scenarios in which the different practices are applied alone and in
combination. The selection of these scenarios seeks to compare the effect of centralized and decentralized
measures, as well as the effect of applying uniform or combined BMPs.
Table 7.3 Scenarios to evaluate.

Scenario 1 Application of only Green Roofs (GR).
Scenario 2 Application of only Pervious Pavements (PP).
Scenario 3 Application of only Infiltration Trenches (IT).
Scenario 4 Combination of GR, PP and IT.
Scenario 5 Application of only Storages
Scenario 6 Combination of GR, PP, IT and Storages.
The first assessment of BMPs application is done by comparing the situations of "Do nothing scenario" and
the "Maximum number of BMPs units" scenario. The last one represents the situation in which the BMPs
are applied at the maximum number of units corresponding to each scenario. These two situations are the
limits, in which the cost is minimum but runoff is maximum ("Do nothing"); and the cost is maximum but
runoff is minimum ("Maximum BMPs application").
The limitations identified from physical and local constraints analysis presented in Annex E determinate
low BMPs application that can derive in very little effect in cases in which BMPs are not combined
(Scenarios 1, 2 and 3). With the objective of having appreciable effects for all scenarios and to allow
comparison between alternatives, these limitations were modified for the scenarios 1 and 2 to reach higher
BMPs application.
Before these changes, GR application was defined between 5 and 10% of total area using big roofs only,
mainly to avoid high application due to regulation and control problems from MM. In the same approach
PP application was between 10 and 15% of total area, these values were calculated based on availability of
parking and sidewalks areas.
In the new approach for scenarios 1 and 2 maximum GR application was increased to between 15 and 20%
of total area. To reach this value the use of normal roofs is also considered. It is verified that the new
coverage of GR is lower than the total roofs coverage. The watershed has approximately 70% of private
area and from the analysis of satellite images it is estimated that about 70% of this area corresponds to
roofs. From this calculation the maximum capacity of green roofs application is 50% of total area, in this
study between 30 and 40% of total roofs were used for GR application.
The maximum coverage of PP is changed from between 10 and 15% to between 15 and 20% of total area.
In this case the maximum limit was defined by physical constraints. In the study area around 30% of total
surface is public area; of this public area an average of 60% corresponds to sidewalks and 40% to roads. As
a result, 12% of total area is covered by roads and 18% by sidewalks. Due to the restrictions of pervious
pavements to not be applied in high traffic surfaces, the limit of PP application was considered as 18% of
total area.
Finally, the length of IT was maintained in those watersheds where this practice is applicable. The reason
for this is that this practice is limited by the existence of wide avenues, so physical constraints are not
allowing the increment of coverage for this scenario.
The maximum application of storages was not modified because its limits were already defined in
accordance with the biggest area available.
Table 7.4 shows the percentages of total area used to locate BMPs in each scenario. For scenarios 3 and 4
the coverage defined in Annex E is kept the same. For scenarios 1 and 2 the new alternative that seeks
higher efficiencies is applied.
62
Table 7.4 Percentages of total area covered by BMPs in each scenario.
%Atotal to %Atotal to %Atotal to %Atotal to
BMPs Sc1 BMPs Sc2 BMPs Sc3 BMPs Sc4
ALEJ 15% 15% 0% 15%
AYAC 17% 17% 2% 21%
BATL 0% 9% 8% 17%
BAUZ 17% 17% 8% 27%
BUCE 15% 18% 21% 39%
BUXA 15% 15% 0% 15%
CIAN 16% 17% 12% 30%
CIPR 15% 15% 8% 23%
EACU 17% 15% 10% 27%
ECHE 15% 16% 5% 21%
FELI 17% 18% 8% 28%
FRAN 15% 15% 11% 26%
HAED 15% 15% 3% 18%
MFRA 15% 15% 3% 18%
TOTAL
15% 15% 7% 23%
COVERAGE (%)
Scenario 5 and 6 are not presented in Table 7.4; Scenario 5 includes only storages and coverage is not
represented by percentage of area treated. In scenario 6 the coverage of BMPs is the same than for scenario
4 but storages are added.
As can be seen in Table 7.4 the percentage of area treated in the case of scenario 3 (only infiltration
trenches) is a half than the percentage treated in the cases of scenarios 1 and 2.
The results obtained after running optimization for each scenario are presented in Chapter 8.
CHAPTER 8
Results and discussion
The results obtained from continuous simulations and optimization process, are presented in this chapter.
With the objective of comparing among different configurations the discussion is divided in two parts.
First, the result obtained from assessing scenarios 1, 2 and 3; these scenarios apply individual measures:
green roofs, pervious pavements and infiltration trenches respectively. Secondly, the results obtained from
the last three scenarios are analyzed. These scenarios include centralized and decentralized measures and
the combination of them. Scenario 4 combines green roofs, pervious pavements and infiltration trenches;
scenario 5 uses storages as centralized measures; and scenario 6 combines centralized and decentralized
measures.
8.1. Comparison of individual measures

The aim of comparing these scenarios is to evaluate the efficiency of each BMP separately. The analysis is
done separately for the results obtained from modelling the design storm and the continuous time series,
and comparing for both cases the situation "do nothing" with "maximum application". The last condition
corresponds to the application of the maximum possible quantity of units in each subwatershed.
8.1.1. Design storm simulation

Figure 8.1 shows the peak flow in the main channel obtained by simulating the design storm event, for the
"do nothing" condition and with the maximum possible number of practices in each scenario. Second and
third plots in Figure 8.1 present the total costs and peak flow reduction percentages respectively, for each
case under the same conditions.
The case of maximum application of green roofs is more efficient in terms of peak flow reduction but at the
same time presents higher total cost than the case of maximum application of pervious pavements. The case
of infiltration trenches presents much lower peak reduction and cost; this is directly related with the lower
application possibility for this practice, due to space constraints.
64
20000 Do nothing IT 1.6E+08 20%
GR GR
GR PP PP PP
15000 1.2E+08 15%
10000 8.0E+07 10%
5000 4.0E+07 5%
IT
IT
0 0.0E+00 0%
1 2 3 4 1 2 3 1 2 3
Peak Flow (l/s) Cost (Eu) Peak flow reduction
Figure 8.1 Peak flow in the main channel, total cost and percentage of peak flow reduction; scenarios 1, 2 and
3.
8.1.2. Continuous simulation

Continuous simulations are run for the nine selected summers (see section 5.3.1). CSO frequency reduction
from the main combined pipe, reduction in volume discharged to the river and reduction of the peak flow
discharged to the river are measured in each case. The points where these parameters are controlled were
presented in Figure 7.2.
Table 8.1 shows the results obtained from this analysis as average values calculated for the total number of
summers modelled. In Annex G the complete results are presented. Percentages of reduction are always
measured taking the "do nothing" condition as reference.
Table 8.1 Average values obtained from continuous simulation.

Scenario Sc. 1 (GR) Sc. 2 (PP) Sc. 3 (IT)
Average CSO frequency reduction: 10% 7% 0%
Average volume discharged reduction: 9% 13% 0%
Average peak flow discharged reduction: 19% 21% 6%
From the analysis of these results it can be deduced that green roofs have higher efficiency than pervious
pavements in CSO frequency reduction, but lower efficiency in volume reduction. A higher efficiency in
volume reduction from PP application is explained because this measure has infiltration capacity besides
storage and evaporation capabilities; while in GRs only storage and evaporation process occur. In this
context, the capacity of volume reduction presented by GR is justified by evaporation processes and from
CSO volume reduction coming from the combined system
The effect in reducing the peak flow in the main discharge channel is evaluated because it is directly related
with the capacity of these measures to diminish floods and decrease CSOs. Measuring this parameter in the
main channel helps to evaluate the impact on the whole system (the point of CSO evaluation is affected for
only one part of the system). Although GR had better performance reducing peak flow in the case of a
single event modelling, for continuous simulations the efficiency of GR and PP is almost the same with a
small improvement under PP practices.
The effects of infiltration trenches cannot be evaluated due to the low percentage of application. Figures 8.3
and 8.4 show the results obtained for each summer in terms of decreasing CSO and volume reduction
respectively.
Both practices have similar effects on decreasing CSO frequency and the volume discharged (Table 8.1).
However, pervious pavements are more effective in decreasing runoff volume (Figure 8.3). The
effectiveness in CSO reduction is not consistent over the period when evaluating results for each year
(Figure 8.2); this behaviour cannot be visualized using average values. During some summers (01-02, 04-
05 and 05-06) the reduction in frequency is high but in other cases (90-91 and 02-03) this measures have
not impact decreasing the number of CSO events.
One possible explanation for the variation of results in CSO frequency reduction for different years is the
strong effect of rainfall patterns (rainfall intensity, frequency and duration) on BMPs performances. In
particular, the highest difference in CSO reduction is observed if summers 01-02 and 02-03 are compared
(Figure 8.2). This is in accordance with the strong dissimilarity in rainfall days and number of storms
between these two summers (Table 5.8); the summer 02-03 has 50% more of raining days and 15% more of
storm events than the summer 01-02.
30%
25%
CSO events reduction.
20%
15%
GR
10%
PP
5% IT
0%
90-91 93-94 94-95 01-02 02-03 03-04 04-05 05-06 06-07
Summer.
Figure 8.2 Reduction in CSO events during 9 different summers; scenarios 1, 2 and 3.
16%
14%
12%
Volume reduction.
10%
8%
GR
6%
PP
4%
IT
2%
0%
90-91 93-94 94-95 01-02 02-03 03-04 04-05 05-06 06-07
Summer.
Figure 8.3 Reduction in volume discharged during 9 different summers; scenarios 1, 2 and 3.
66
From the analysis of results obtained comparing GR and PP as possible practices to reduce runoff and to
decrease the CSO frequency, it can be concluded that PPs present better global performance. This BMP
presents better performance in reducing runoff by allowing infiltration and also has an impact on CSO
reduction. In the case of GR implementation a lower impact in runoff reduction is achieved while the
reduction in CSO discharges is higher, but this effect cannot be ensured for every summer.
8.1.3. Optimization results

The three scenarios were optimized. In each case 9600 solutions were evaluated (80 generations of 120
individuals). Figures 8.4, 8.5 and 8.6 present the results obtained in each case. All solutions evaluated and
the best 120 found solutions are presented in the next Pareto fronts.
Pareto front, Scenario 1 (GR).

18500
All solutions
Best solutions
18000
17500
Peak Flow (l/s)
17000
16500
16000
15500
0.00E+00 2.00E+07 4.00E+07 6.00E+07 8.00E+07 1.00E+08 1.20E+08
Cost (Eu)
Figure 8.4 Optimization result for scenario 1.
Pareto front, Scenario 2 (PP).
18,500
All solutions
Best solutions
18,000
17,500
Peak Flow (l/s)
17,000
16,500
16,000
0.00E+00 2.00E+07 4.00E+07 6.00E+07 8.00E+07 1.00E+08
Cost (Eu)
Pareto front, Scenario 3 (IT)

18,400
All solutions
Best solutions
18,350
18,300
Peak Flow (l/s)
18,250
18,200
18,150
18,100
0.00E+00 5.00E+05 1.00E+06 1.50E+06 2.00E+06 2.50E+06 3.00E+06
Cost (Eu)
Pareto fronts of GR and PP scenarios present linear shapes while the solution in the case of IT is not linear.
A possible explanation for this result is based in that the main difference between these BMPs is that
whereas green roofs and pervious pavements treat the water falling above them, infiltration trenches are
receiving water from an external drainage area. As a result the effect of applying GR or PP is directly
proportional with the coverage area; while in the case of infiltration trenches this effect is not directly
68
linear. The idea for IT modelling is that when the length of trenches is changing the drainage area
contributing to these elements is also changing. After analysing the Pareto front obtained, it is concluded
that this area is fixed in the optimization process. For this reason the plot has a first region in which the
efficiency is improving substantially with investment growth, but after reaching the point in which the
trenches have the necessary volume to store the flow coming from the fixed drainage area, almost no
improvements are obtained even if the investment is increased.
By analyzing the Pareto fronts obtained the importance of the optimization process for design purposes can
be visualized. For instance in the case of GR and PP, to obtain a peak flow of 17000 l/s a range of between
15 and 18% of total cost is covered for different possible solutions. The best solution at this level is
showing the cheapest alternative among several with the same result in flow reduction.
In the case of scenario 3 the range of cost variation for the same flow reduction result is even higher. To
obtain a peak flow of 18200 l/s the range of possible alternatives has a cost variation of about 70%.
Figure 8.7 compares the best solutions for the three scenarios. In this plot it can be observed that IT
alternative is the dominant solution in the initial range of low costs and low peak flow reduction, offering
better efficiencies with minor costs. Due to the low application considered in this case it is not known if
this tendency can be maintained for higher investments. When costs and peak flow reduction are higher
GRs offers better solutions than PPs.
Pareto fronts of GR, PP and IT systems.

18,500
GR
PP
IT
18,000
17,500
Peak Flow (l/s)
17,000
16,500
16,000
15,500
0.00E+00 2.00E+07 4.00E+07 6.00E+07 8.00E+07 1.00E+08 1.20E+08
Cost (Eu)
Figure 8.7 Pareto fronts comparison for scenarios 1, 2 and 3.
The comparison between optimal solutions for GR and PP configurations can be better visualized in Table
8.2, where optimal solutions with equal price are presented for both cases (these points are highlighted in
Figure 8.7). The cost reduction is comparing the cost of maximum application of BMPs with the cost of
optimal solutions. The flow reduction is comparing the peak flow in the case of "do nothing" with the peak
flow of optimized solutions.
Table 8.2 Optimal solutions comparison for scenarios 1 (GR) and 2 (PP).
SCENARIO 1
Do
Solution 1 Solution 2 Solution 3 Max BMPs App
Nothing
Cost 2.0.E+07 6.0.E+07 1.0.E+08 0 1.4.E+08
Peak Flow 17885 16898 16016 18409 15167
Cost reduction 86% 58% 30%
Flow reduction 3% 8% 13%
SCENARIO 2
Do
Solution 1 Solution 2 Solution 3 Max BMPs App
Nothing
Cost 2.0.E+07 6.0.E+07 1.0.E+08 0 1.2.E+08
Peak Flow 17872 16982 16195 18409 15719
Cost reduction 84% 51% 19%
Flow reduction 3% 8% 12%
The results are very similar but better efficiencies of GRs are detected when the investment is increased
(solution 3). This tendency can be clearly observed in the comparison of Pareto fronts, where the GR
solution is more dominant when the investment is higher. The efficiency of GR, regarding to peak flow
reduction, is even higher than in the case of PP if it is consider that GR has higher cost per square meter
than PP; so in a case of equal investment the surface of GR is smaller than the surface of PP.
8.1.4. Spatial distribution of optimal solutions

To analyze the spatial distribution of optimal solutions, for each configuration, the best solutions obtained
from the optimization process were considered. Average values of the number of units applied in the best
120 solutions were calculated for each subwatershed. These average values, the standard deviations and the
maximum units applicable in each subwatershed are presented in Figures 8.8, 8.9 and 8.10, for the cases of
GR, PP and IT respectively. Also subwatershed areas and percentages of imperviousness are showed;
darker colours represent higher number of BMP units applied.
70
±
Figure 8.8 Spatial distribution of optimal solutions in the GR case.
Figure 8.8 shows the distribution in the case of green roofs, it can be observed that the higher application of
measures occurs in the upper part of the watershed. In the lower part of the watershed less units are applied,
even when they are available (visualized through the maximum number of units applicable, symbolized
with the darkest column). One reason that could be influencing this result is that the areas located in the
south-western part of the watershed are contributing directly to the downstream system, not to the main
channel.
Also, the area of the subwatersheds can be associated with the percentage of available units applied; in
bigger watersheds a higher fraction of units is applied (visualized through the relation between the first and
the last column). This case presents the exception of the park area where green roofs cannot be used; it is
the subwatershed with almost circular shape and 50ha size in Figure 8.8.
Figure 8.9 presents the results obtained after mapping the optimal solutions obtained from the optimization
process applied in the design of pervious pavements systems.
±
Figure 8.9 Spatial distribution of optimal solutions in the PP case.
In the case of PP systems the spatial distribution presented in Figure 8.9 is similar to the case of GR. Again
most of available units are used in the upper part of the catchment, although the tendency is less strong in
this case. For this scenario, even if the use of PP is considered in the park area the optimized solutions are
not allocating many units in this area. This can be explained due to the low impervious percentage of this
subwatershed.
Figure 8.10 presents the spatial distributions of optimized configurations for the case of ITs. As can be
observed in the figure this measure cannot be applied in all the subwatersheds, in particular in the northeast
area.
72
±
Figure 8.10 Spatial distribution of optimal solutions in the IT case.
Finally, examining Figure 8.10 a different tendency in solutions distribution is observed in the case of ITs.
In this case the availability of units applicable is not well distributed along the watershed, and in two of
upper subwarsheds this practice is not allowed. In this case the main trend is not the use of upper areas and
it is surprising how the most south-western subwatershed is used to allocate practices when this area in not
contributing to the main channel.
8.2. Comparison of scenarios 4, 5 and 6

Scenarios 4, 5 and 6 represent the application of combined BMPs representing decentralized measures,
storages as centralized measures and combined BMPs mixed with storages to represent the mixture of
different solutions. Comparing these scenarios the efficiency of each different configuration can be
evaluated.
8.2.1. Design storm simulation

The first plot in Figure 8.11 shows the peak flow in the main channel obtained from the simulation of the
design storm event. The cases presented are the "do nothing' situation and maximum BMPs application for
scenarios 4, 5 and 6. The second and third plots in Figure 8.11 show total costs and peak flow reduction
percentage for each scenario under the same conditions.
25000 2.0.E+08 GR+PP+ 50%
Do IT+ St GR+PP+
GR+PP+
20000 Nothing 1.6.E+08 40% IT+St
IT
GR+PP+IT GR+PP+IT+ St
15000 St 1.2.E+08 30%
St GR+PP+
10000 8.0.E+07 20% IT
5000 St
4.0.E+07 10%
0 0.0.E+00 0%
1 2 3 4 1 2 3 1 2 3
Peak Flow (l/s) Cost (Eu) Peak flow reduction
Figure 8.11 Peak flow in main channel, total cost and percentage of peak flow reduction; scenarios 1, 2 and 3.
The cases of maximum application of scenarios 4, 5 and 6 are more efficient in peak flow reduction than
scenarios 1, 2 and 3. The main reason is because the percentage of area treated is higher in scenario 4 (see
Table 7.4); this also causes higher cost for scenario 4 comparing with scenarios 1, 2 and 3. In the case of
scenario 5, only storages application, the efficiency is high even when the space constraints limit the
number of tanks to 6 in a case of 14 subwatersheds influencing the results. Also the cost of maximum
application for scenario 5 is much lower than in other scenarios. Finally, the efficiency of scenario 6 is
much higher because it is the combination of scenarios 4 and 5, but this also induces higher costs.
8.2.2. Continuous simulation

Average results from continuous simulations in this case are presented in Table 8.3. Analyzing these results
it can be observed a low impact of scenario 5 in CSO events and volume discharged reductions. A higher
efficiency in CSO reduction was expected due to storage tanks have as main objective the peak flow
reduction and this effect has a direct impact in decreasing the discharge of CSOs through weirs. The cause
of low impact in CSO reduction is due to the location of the storages, most of them are not influencing the
main combined pipe and for this reason are not decreasing the flow in that pipe, from where CSO are
originated. The low impact in volume reduction is because this measure has the reduction of peak flow as
main objective but no infiltration or evaporation processes are occurring, for this motive the volume of
water remains the same.
The combination of distributed measures in scenario 4 has medium to high impact in terms of CSO events
and volume discharge reduction, with higher efficiency in volume reduction. This effect is reasonable
considering that a mixture of infiltration, evaporation and storage processes is occurring when these
decentralized measures are applied. The effect in runoff volume reduction is an important point considering
the results obtained for Cambez et al. (2008), in accordance with which reductions in pollutants loads
discharged are very similar than reductions in overflowed water volumes.
The combination of centralized and decentralized measures (scenario 6) has high efficiency in the
accomplishment of both objectives. As can be observed in Table 8.3 the percentage of volume reduction is
the addition of effects of scenarios 4 and 5, but the effect in CSO reduction is more than this addition.
Table 8.3 Average results from continuous simulation, scenarios 4, 5 and 6.

Scenario Sc. 4 (GR+PP+IT) Sc. 5 (St) Sc. 6 (GR+PP+IT+St)
Average CSO frequency reduction: 10% 4% 17%
Average volume discharged reduction: 14% 3% 17%
Average peak flow discharged reduction: 22% 34% 46%
74
Figures 8.12 and 8.13 present the results of CSO frequency and volume discharged reduction for each
summer. In general it is observed a higher impact of measures over volume reduction than for the decrease
of CSO frequency; the results are better distributed among different years in the case of volume reduction.
Regarding this, the enhanced effect of scenario 6 on CSO reduction compared with scenarios 4 and 5 is
again observed by analyzing every year. The configuration of combining centralized and decentralized
measures has a more constant effect for every summer in CSO reduction, having effects even in cases
where nor decentralized neither centralized measures have impact (summers 90-91 and 93-94).
A higher effect of storages in CSO frequency reduction is expected because the good performance of these
measures in peak flow reduction. This high efficiency in peak flow reduction can be observed in results for
the main discharge channel, presented in last row of Table 8.3. Due to some singularities of this particular
system this effect cannot be visualized: most of the storages located in the system have no effect on the
main combined pipe from where CSOs are discharged into the river.
30%
25%
CSO events reduction.
20%
15%
Scenario4
10% Scenario5
Scenario6
5%
0%
90-91 93-94 94-95 01-02 02-03 03-04 04-05 05-06 06-07
Summer.
Figure 8.12 Reduction in CSO events during 9 different summers; scenarios 4, 5 and 6.
25%
20%
Volume reduction.
15%
10% Scenario4
Scenario5
5% Scenario6
0%
90-91 93-94 94-95 01-02 02-03 03-04 04-05 05-06 06-07
Summer.
Figure 8.13 Reduction in volume discharged during 9 different summers; scenarios 4, 5 and 6.
8.2.3. Optimization results
Scenarios 4, 5 and 6 are optimized following the same criteria explained in the case of scenarios 1, 2 and 3.
Results are shown in Figures 8.14, 8.15 and 8.16.
Pareto front, Scenario 4 (GR+PP+IT)

18,000
All solutions
Best solutions
17,500
17,000
Peak Flow (l/s)
16,500
16,000
15,500
0.00E+00 2.00E+07 4.00E+07 6.00E+07 8.00E+07 1.00E+08 1.20E+08
Cost (Eu)
Figure 8.14 Optimization result for the combination of decentralized measures.
Pareto front, Scenario 5 (Storages)

18,500
All solutions
Best solutions
18,000
17,500
17,000
Peak Flow (l/s)
16,500
16,000
15,500
15,000
14,500
0.00E+00 5.00E+06 1.00E+07 1.50E+07 2.00E+07 2.50E+07
Cost (Eu)
Figure 8.15 Optimization result for storages.
76
Pareto front, Scenario 6 (GR+PP+IT+Storages)
17,500
All solutions
17,000 Best solutions
16,500
16,000
15,500
PEak Flow (l/s)
15,000
14,500
14,000
13,500
13,000
12,500
1.00E+07 3.00E+07 5.00E+07 7.00E+07 9.00E+07 1.10E+08 1.30E+08
Cost (Eu)
Figure 8.16 Optimization result for the combination of decentralized and centralized measures.
The results of optimization for scenario 4 present linear shape. This is expected considering that it is the
combination of scenarios 1, 2 and 3 where scenario 3 (presented in Figure 8.6) is the only one with non
linear shape but has low influence comparing with the other two solutions.
The shape of the Pareto front in the case of storages has similar shape than infiltration trenches solution.
Both measures receive water from a fixed external drainage area and present a range in which the
efficiency improves noticeably for small increments of investments, but after reaching the necessary
volume for storing the runoff coming from the fixed area, the efficiency decreases perceptibly (observed as
a decrease in the slope of Pareto front).
Also in this case the importance of optimization can be visualized analyzing the rage of different possible
costs for the same flow reduction result. For instance in the case of scenario 4 to reach a peak flow of
16500 l/s the range of costs of possible solutions varies 14%. In the case of scenario 5 to reach the same
peak flow, the range of costs varies 46%. Finally, in the case of scenario 6, to have a maximum flow of
14000 l/s the possible solutions fluctuate in a range of 48% of maximum cost.
Figure 8.17 presents the three Pareto fronts of scenarios 4, 5 and 6. In this graph can be observed that
scenario 5 is the dominant solution for low costs and peak flow reduction until 20%. To reach higher peak
flow reductions scenario 6 is the best option. The configuration presented in scenario 4 shows lower peak
flow decreases for equal costs.
Pareto fronts of decentralized, centralized and combined configurations.
19,000
18,000
17,000
Peak Flow (l/s)
16,000
Sc4
Sc5
15,000
Sc6
14,000
13,000
12,000
0.00E+00 2.00E+07 4.00E+07 6.00E+07 8.00E+07 1.00E+08 1.20E+08
Cost (Eu)
Figure 8.17 Pareto fronts comparison for scenarios 4, 5 and 6.
Table 8.4 shows the cost reduction (comparing with the maximum possible cost for each scenario) and the
peak flow reduction (comparing with the "do nothing" situation) for different optimal solutions. Three
solutions are showed in blue (highlighted in Figure 8.17); these cases are optimal solutions with similar
cost for each configuration. Analyzing these results it can be verified the better efficiency of solutions
composed by storages, which achieve better peak flow reduction with the same investment. This result is in
accordance with results obtained for Urrestarazu (2013), who concludes that the use of storages is more
cost-effective that the use of BMPs, and BMPs use is justified when ecosystem services are considered
among the benefits.
78
Table 8.4 Optimal solutions comparison for scenarios 4, 5 and 6.
SCENARIO 4
Max BMPs
Solution 1 Solution 2 Solution 3 Solution 4 Do Nothing
App
Cost 1.6.E+07 2.4.E+07 6.4.E+07 1.2.E+08 0 1.4.E+08
Peak Flow 17854 17688 16726 15641 18409 15150
Cost reduction 89% 83% 55% 16%
Flow reduction 3% 4% 9% 15%
SCENARIO 5
Max BMPs
App
Cost 1.8.E+04 5.3.E+06 1.1.E+07 2.2.E+07 0 3.2.E+07
Peak Flow 18381 17899 15066 14649 18409 13153
Flow reduction 0.2% 3% 18% 20%
SCENARIO 6
Max BMPs
App
Cost 2.4.E+07 2.7.E+07 7.6.E+07 1.2.E+08 0 1.7.E+08
Peak Flow 16043 14454 13492 12942 18409 11544
Flow reduction 13% 21% 27% 30%
An interesting result is visualized comparing the cases of scenarios 4 and 6; with the same investment the
case that combines centralized and decentralized solutions (scenario 6) shows much higher efficiency
decreasing the maximum flow. Adding another advantage of combined solutions to the result obtained from
continuous simulation analysis, from which the combination of different measures affects both CSO
frequency reduction (mainly from storages implementation) and volume discharged reduction (mostly from
BMPs performance).
An additional discussion can be obtained from Table 8.4, solutions in green show two cases where the
application of optimization processes in design has strong effects in investments reduction. In the case of
scenario 5 the optimized solution 3 is reaching a reduction of 18% in peak flow (from a maximum
reduction possible of 29% with this configuration) investing only the 35% of maximum cost (needed to
reduce the 29% the flow). In the case of scenario 6 a similar situation is observed, 27% of peak flow
reduction is achieved (that represents the 73% of maximum possible peak flow reduction: 37%) investing
only 44% of the maximum cost.
Figure 8.18 compares the flow in the main discharge channels in the cases of "do nothing", optimal solution
(green one in Table 8.4) and maximum BMPs application for the configuration presented under scenario 6.
In this hydrograph the high efficiency of the optimized solution can be observed, achieving a high flow
reduction compared with blue case, but using only the 44% of the investment needed to reach the green
case.
20000
18000
Do nothing
16000
Op.Solution
14000 Max.App.
12000
Flow (l/s)
10000
8000
6000
4000
2000
0
0 100 200 300 400 500 600 700
Time (min)
Figure 8.18 Hydrographs in main discharge channel in three possible solutions for scenario 6.
8.2.4. Spatial distribution of optimal solutions

To analyze the spatial distribution of optimal solutions, the best solutions obtained from optimization
process in each scenario are considered. In the case of scenarios 4 and 6 several different practices are
applied. For these cases how the average values of units applied in the best 120 solutions are distributed
among these different practices in each subwatershed is presented in Figures 8.19 and 8.21. In figure 8.20 a
similar representation than in scenarios 1, 2 and 3 is followed, because also in this case one only practice is
applied. Darker colours represent higher number of BMP units applied and softer colours represent areas
with lower application of measures.
Figure 8.19 presents the spatial distribution of the different BMPs considered in optimized solutions for the
case of scenario 4. For each subwatershed the relative quantities of each measure are represented, also the
area and percentage of imperviousness of each subwatershed are presented.
80
±
Figure 8.19 Spatial distribution of optimal solutions in the case of decentralized measures, scenario 4.
From analyzing the spatial distribution in the case of combining decentralized measures (Figure 8.19), the
higher use of practices in the upper part of the watershed is observed. In most of the cases the three possible
practices (GR, PP and IT) are used when they are available. The practice less used is IT which is in
accordance with the lower maximum coverage possible comparing with the other practices. GR and PP are
in general equally used among different watersheds. Differently than in cases presented before, in this case
the maximum number of possible applicable units and the standard deviation of the average used are not
presented. The reason of this is that there are three applicable BMPs and the graphical representation of
those values is not possible.
Figure 8.20 shows the distribution of the average usage of storages for optimal solutions. This measure can
be applied only in six watersheds. In this case, because only one measure has to be represented the
maximum number of units applicable and the standard deviation are presented.
±
Figure 8.20 Spatial distribution of optimal solutions in the case of centralized measures, scenario 5.
The optimal distribution of centralized measures is presented in Figure 8.20; the six subwatersheds where
this measure is applicable can be visualized. The optimal solutions are not using the storage capacity
located in the park area (50ha size). This tendency was detected before in the case of other measures and it
is related with the low percentage of imperviousness of this area which determines low runoff production.
Among the other five areas with storage capacity, the subarea located lower in the watershed has the
highest fraction of storage available used. The reason of this is because this storage is allocated in the main
pluvial pipe coming from the entire watershed, just before the beginning of the main channel, for this
reason the structure has a direct impact in peak flow reduction in this channel and is affecting the flow
coming from the total pluvial system.
In Figure 8.21 the distribution of the combination of centralized and decentralized solutions is presented.
Following the same approach than for the case of scenario 4, the relative distribution among measures is
presented for each subwatershed.
82
±
Figure 8.21 Spatial distribution of optimal solutions in the case of combined measures, scenario 6.
Finally, analyzing the spatial distribution of combined configurations, the higher use of units in the upper
part of the watershed in combination with the use of the storage volume available in the lowest
subwatershed can be visualized.
The solution of mixing centralized and decentralized measures (scenario 6) presents better performances in
runoff volume and CSO frequency reduction. Also, this configuration presents better efficiencies under
equal investments than just decentralized measures. Finally, this scenario shows a better distribution of
measures application over the whole area.
8.3. Overview
The proposed approach for the design of configurations shows advantages to select the type and number of
BMP units to be applied, reducing the necessary investments. This approach also presents advantages to
determine the best spatial location for the distribution of these measures to reach higher efficiencies in
runoff and CSO control.
The results of this research have shown that the best long term strategy in the case analyzed is the
application of combined centralized and decentralized measures, applying decentralized measures mainly
in the upper part of the watershed and centralized measures in the lower part of the area.
By analysing the results obtained in this work the low influence that some subwatersheds have on the CSO
and runoff volume produced in the main channel can be understood. This allows the simplification of the
design process mainly by reducing the number of variables.
The reduction achieved in terms of runoff volume and CSO events going into the river shows the
importance of considering these measures as a tool when planning the drainage system for the city. The
main objective of conveying the dry weather flow from the coastal system to the subaquatic emissary that is
going 2.5Km into the river is to protect the beaches from pollution coming with the wastewater. By
applying decentralized measures all over the total area of the coastal sewerage system this objective is
enhanced by decreasing the pollutants loads discharged from CSOs during rain periods.
The results also show that benefits regarding to water quality for recreational proposes can be achieved by
applying decentralized measures. These improvements could be better evaluated adding the simulation of
water quality to the hydrodynamic model. Moreover, the addition of this goal as an objective function in
the optimization process can provide different results in the design of drainage systems. Through this
analysis the importance of decentralized measures could be visualized since the enhancement of water
quality in the coast is one of the main objectives of the Municipality.
Currently the city is updating the Master Plan of Sewerage and Drainage for the next 25 years. The
consideration of mixing decentralized and centralized measures as a long term option may widen the range
of alternatives to be analyzed. To include these alternatives in any future scenario, they should to be taken
into account in the process of urban planning and regulations development that will be included in this
Master Plan.
84
CHAPTER 9
Conclusions and recommendations
9.1. Conclusions
In this research a framework for multi-objective evaluation of different BMP configurations to be applied
in an urban watershed is proposed. A general methodology was developed and applied in a case study area
located in Montevideo, Uruguay. This methodology has proven to be a valuable tool in the decision making
process for stormwater management in this urban area. The proposed approach and its application include
the phases of BMPs selection according to particular conditions. Also, it contains the design of BMPs and
the optimization of these systems to maximize the efficiency when addressing multiple objective problems.
The optimization is performed with the aim of balancing the level of investments and the amount of
wastewater discharged to the coast. The main conclusions of this study are:
• Best management practices are able to efficiently reduce runoff and CSO frequency in a high
urbanized area such as Pocitos watershed. This case study did not allow proving that decentralized
measures are more efficient than centralized ones in accomplishing several objectives. On the other
hand, this research shows that in this case the combination of decentralized and centralized
measures offers better results reaching higher efficiencies in the framework of runoff reduction -
cost minimization.
• This research highlights the relevance of research and innovation applied to real-case situations.
This is shown by the use of hydrodynamic models to evaluate alternative measures affecting
hydrological processes. The inclusion of the hydrologic model in SWMM improved the system
representation and shows the usefulness of representing the rainfall-runoff processes through the
hydrodynamic model. Moreover, this study includes the use of measures affecting the hydrologic
processes for runoff reduction. This approach is different than the presently used by the
Municipality that only considers centralized structural measures such as storages, to cope with
problems in the drainage system. The results obtained can be the starting point of a new approach
including the hydrological processes evaluation in the drainage system planning.
• The use of a single event and rainfall time series to evaluate the results are complementary
methods that allow the assessment of complex systems in a constrained time framework. Under
this approach the evaluation of systems under different conditions is possible, resulting in a wide
assessment of results. In particular, the use of time series simulation to evaluate the effects of
BMPs enables a better understanding of their performance. Consequently, it shows different results
over time depending on rainfall patterns that are affecting the BMPs performance.
• One of the most important steps in the developed methodology consists in BMPs suitability
analysis considering particular characteristics and limitations of each case. From this analysis real
possibilities of practices application are developed, taking into account local physical constraints
and stakeholders preferences. This phase is focused in the improvement of drainage system's
sustainability in the long term, requiring less controls and maintenance efforts to support the
system in the future.
• Optimization processes applied in drainage configurations design achieved the expected goals.
This approach offers several possible solutions with the same runoff reduction effect, allowing the
selection of the cheapest option. The use of this tool presents a range of optimal solutions, while
saving important investments. Furthermore, the study of optimal solutions enables the
understanding of the importance of measures distribution over the watershed. Better results can be
obtained under similar investments if the measures are distributed on key areas, having higher
effects on the drainage system.
• In this case centralized solutions composed by storages are more efficient reducing peak flows and
CSO frequencies with lower costs. This result can present some changes if additional benefits are
considered, for instance liveability improvements and heat island effect reduction. Enlarging the
range of objectives and possible advantages from BMPs application could provide a wider view of
trade-offs between investments and benefits obtained.
• BMPs performed better than centralized solutions in reducing the volume discharged to the
recipient. This is an important point if reductions in pollutants loads discharged are considered
proportional to reductions in overflowed water volumes.
• Green roofs and pervious pavements have similar effects, but pervious pavements performed better
in runoff volume reduction due to its infiltration capacity. The results of both measures in CSO
frequency reduction during summer months are variable for different years. Considering both
objectives (runoff and CSO reductions) PPs present better global performance. This is because PPs
presents better performance in reducing runoff and also have an impact on CSO reduction. While
under GR implementation a lower impact in runoff reduction is achieved whereas the reduction in
CSO discharges is improved, but this effect cannot be ensured for every summer. In addition,
better efficiencies in peak flow reduction are obtained from applying GR when optimized solutions
are evaluated.
• The results are consistent enough to develop the general conclusions about the acceptability of this
methodology as a stormwater management tool for the case studied. However, some results are
different than expected, such as the performance of storages application regarding CSO frequency
reduction or the low efficiency of infiltration trenches. This is mainly due to some particularities
and complexities of the case study applied (low soil infiltration capacity and complex
interconnections between pipes).
9.2. Recommendations
Despite the promising results obtained from this study there is a wide range of opportunities for future
research. These recommendations are oriented to improve the results through a more detailed analysis of
the problem and possible solutions.
• The reliability of results could be improved by calibrating and validating the hydrodynamic model.
To achieve this goal it is essential to perform measurements in the network (flow, water level).
86
These observations will enhance the understanding of the current system's behaviour and the
accuracy of the model in representing it. Additionally, the evaluation of BMPs representation using
SWMM can be achieved through direct observations, after implementing these practices.
• The use of a single event in the optimization process allowed the assessment of several
configurations in a reduced period of time. The development of optimization processes using
several design events (Expected Annual Damage concept) and continuous simulation instead of
using a single design storm could be useful to analyze how optimal solutions change. Moreover,
the analysis of continuous series would allow the evaluation of effects such as the frequency of
CSO, adding a new objective function in the optimization process.
• The drainage system configuration strongly influences optimization results. This should determine
the definition of the objective function regarding to runoff and CSO reductions. When the
objective is the runoff reduction the minimization of runoff volume in the main channel should be
considered. In the case of CSO frequency reduction the peak flow should be minimized in the main
combined pipe. To obtain configurations capable of reducing both, these two objective functions
should be considered in the optimization process.
• Even if particular conditions were considered to define the BMPs to be applied, and the limits of
coverage, a deeper study of the drainage system should be developed before defining which BMPs
to use and how to distribute them. This study before the definition of optimization variables can
save efforts in this process, allowing the use of less variables and the definition of lower coverage
constraints in subwatersheds with stronger impact on the system.
• The design and simulation of infiltration trenches should be improved in order to compare the
effects of this practice with those obtained from green roofs and pervious pavement application.
This would allow a better comparison of three different BMPs applied as single measures. This is
interesting due to the better effectiveness showed by infiltration trenches when optimal solutions
were compared in a low range of investments.
• Stronger consideration of social, legal and institutional conditions should be taken into account in
the phase of BMPs selection. In this case physical limitations and preferences of some stakeholders
were considered but a deeper study can be developed to ensure an adequate selection of the best
configuration in each particular case, considering for instance public participation.
• Inclusion of water quality simulation to study BMPs effects in pollution reduction should be done.
Addition of water quality as an optimization objective, such as the minimization of pollutants loads
discharged, could add a key target and a broader perspective of measures effectiveness.
• The application of the same methodology in other case studies to obtain results influenced by other
particular conditions will allow better general conclusions and verification of the obtained results.
Furthermore, this would allow the evaluation of other BMPs and different combinations of
centralized and decentralized measures to evaluate if the improvement of effectiveness achieved
using combined solutions is maintained under the usage of other measures.
References
Aad MPA, Suidan MT, Shuster WD (2010) Modeling Techniques of Best Management Practices : Rain Barrels
and Rain Gardens Using EPA SWMM-5. 434–443.
Barreto W (2012) Multi-Objective Optimization for Urban Drainage Rehabilitation. 250.
Barreto W, Vojinovic Z, Price R, Solomatine D (2010) Multiobjective Evolutionary Approach to Rehabilitation
of Urban Drainage Systems. 547–554.
Burszta-Adamiak E, Mrowiec M (2013) Modelling of green roofs’ hydrologic performance using EPA's
SWMM. Water Sci Technol 68:36–42. doi: 10.2166/wst.2013.219
California Stormwater Quality Association (2003) California Stormwater BMP Handbook. pp 1–10
Cambez MJ, Pinho J, David LM (2008) Using SWMM 5 in the continuous modelling of stormwater hydraulics
and quality. 1–10.
Cheng M-S, Zhen JX, Shoemaker L (2009) BMP decision support system for evaluating stormwater
management alternatives. Front Environ Sci Eng China 3:453–463. doi: 10.1007/s11783-009-0153-x
Chocat B, Ashley R, Marsalek J, et al. (2007) Toward the Sustainable Management of Urban Storm-Water.
Indoor Built Environ 16:273–285. doi: 10.1177/1420326X07078854
Chouli E, Aftias E, Deutsch J-C (2007) Applying storm water management in Greek cities: learning from the
European experience. Desalination 210:61–68. doi: 10.1016/j.desal.2006.05.033
Chow V te (1959) Open-channel hydraulics. 680.
Chow V te (1964) Handbook of Applied Hydrology. 1563.
Chow V te, Maidment D, Mays L (1988) Applied Hydrology. 572.
Conservation Technology Inc. (2008) Green Roof Handbook. Green Roof Handb 25.
Damodaram C, Giacomoni MH, Prakash Khedun C, et al. (2010) Simulation of Combined Best Management
Practices and Low Impact Development for Sustainable Stormwater Management1. JAWRA J Am Water
Resour Assoc 46:907–918. doi: 10.1111/j.1752-1688.2010.00462.x
Deb K, Member A, Pratap A, et al. (2002) A Fast and Elitist Multiobjective Genetic Algorithm : 6:182–197.
Delelegn SW, Pathirana A, Gersonius B, et al. (2011) Multi-Objective Optimization of Cost-Benefit of Urban
Flood Management using a 1D2D Coupled Model. Water Sci Technol 63:1054.
Direccion Nacional de Meteorologia (2014) http://meteorologia.gub.uy/index.php/estcli. In: Estad. Meteorol.
Durrans R, Dietrich K (2003) Stormwater Conveyance Modeling and Design. 668.
Elliott a, Trowsdale S (2007) A review of models for low impact urban stormwater drainage. Environ Model
Softw 22:394–405. doi: 10.1016/j.envsoft.2005.12.005
Ellis B, Revitt M, Scholes L, et al. (2006) Sustainable Water Management in the City of the Future Deliverable 2
. 1 . 1 : Review of the adaptability and sensitivity of current stormwater control technologies to extreme
environmental and socio-economic drivers.
Genta JL, Charbonnier F (2013) Agua en la atmosfera. Notas del curso Hdrologia e Hidraulica Aplicadas. 39.
Giddens C (1997) Rawlinsons 1997 New Zealand Construction Handbook.
Houston EB (2006) The use of stormwater modeling for design and performance evaluation of best management
practices at the watershed scale.
Hoyer J, Dickaut W, Kronawitter L, Weber B (2011) Sustainable Water Management in the City of the Future
Water Sensitive Urban Design Principles and Inspiration for Sustainable Stormwater Management in the
City of the Future - Manual - Water Sensitive Urban Design.
Huber WC, Cannon L, Stouder M (2006) BMP Modeling Concepts and Simulation BMP Modeling Concepts
and Simulation. Environ Prot 166.
Idaho Department of Environmental Quality (2005) Catalog of Stormwater BMPs for Idaho Cities and Counties.
663.
Intendencia Municipal de Montevideo (1994) Plan Director de Saneamiento del Departamento de Montevideo.
Jia H, Lu Y, Yu SL, Chen Y (2012) Planning of LID–BMPs for urban runoff control: The case of Beijing
Olympic Village. Sep Purif Technol 84:112–119. doi: 10.1016/j.seppur.2011.04.026
88
Lee JG, Selvakumar A, Alvi K, et al. (2012) A watershed-scale design optimization model for stormwater best
management practices. Environ Model Softw 37:6–18. doi: 10.1016/j.envsoft.2012.04.011
Lucas WC (2010) Design of Integrated Bioinfiltration-Detention Urban Retrofits. J. Hydrol. Eng. 15:
Maidment D (1992) Handbook of Hydrology. 1143.
Marek M (2011) Hydraulic Design Manual. Texas
Melbourne Water (2005) Water Sensitive Urban Design, Engineering Procedures, Stormwater. 304.
Metropolitan Area Planning Council (2010) http://www.mapc.org/resources/low-impact-dev-toolkit/permeable-
paving. Massachusetts LID Toolkit, Fact Sheet 6 Permeable Paving 4.
Montalto F, Behr C, Alfredo K, et al. (2007) Rapid assessment of the cost-effectiveness of low impact
development for CSO control. Landsc Urban Plan 82:117–131. doi: 10.1016/j.landurbplan.2007.02.004
Obropta CC, Kardos JS (2007) Review of Urban Stormwater Quality Models: Deterministic, Stochastic, and
Hybrid Approaches1. JAWRA J Am Water Resour Assoc 43:1508–1523. doi: 10.1111/j.1752-
1688.2007.00124.x
Palla A, Berretta C, Lanza LG, Barbera P La (2008) Modelling storm water control operated by green roofs at
the urban catchment scale. 1–10.
Paredes D (2013) Multi-Objective-Rehabilitation of Urban Drainage Systems within the Flood Risk Framework
Case study : Quito - Ecuador. 109.
Parkinson J, Mark O (2005) Urban Stormwater Management in Developing Countries, 1st ed. 218.
Pathirana A (2013) Urban Drainage Course Material.
Price RK, Vojinovic Z (2011) Urban Hydroinformatics: Data, Models and Decision Support for Integrated
Urban Water Management., 1st ed. 520.
Sahlu SB (2012) Robust Optimization of urban drainage systems under uncertainties. 105.
Scholz M, Kazemi Yazdi S (2008) Treatment of Road Runoff by a Combined Storm Water Treatment, Detention
and Infiltration System. Water Air Soil Pollut 198:55–64. doi: 10.1007/s11270-008-9825-6
Sharma JR, Najafi M, Qasim SR (2013) Preliminary Cost Estimation Models for Construction, operation and
Maintenance of Water Treatment Plants. J Infrastruct Syst ASCE 19:451–464.
Shutes B, Raggatt L (2010) Sustainable Water Management in the City of the Future Development of generic
Best Management Practice ( BMP ) Principles for the management of stormwater as part of an integrated
urban water resource management strategy .
Silveira L, Terra R, Usera G, et al. (2013) ESTUDIOS DE BASE Y ASESORAMIENTO PARA LA
ACTUALIZACIÓN DEL PLAN DIRECTOR DE SANEAMIENTO DEL DEPARTAMENTO DE
MONTEVIDEO. 45.
Temprano J, Arango O, Cagiao J, et al. (2006) Stormwater quality calibration by SWMM : A case study in
Northern Spain. 32:55–63.
Tetra Tech (2010) Practices ( BMP ) Performance. 2010:232.
Tikkanen H (2013) Hydrological modeling of a large urban catchment using a stormwater management model (
SWMM ).
U.S. Department of Transportation - Federal Highway Administration (2013)
http://environment.fhwa.dot.gov/ecosystems/ultraurb/3fs1.asp. In: Stormwater BMPs an Ultra-Urban
Setting - Fact Sheet Infiltration Trench.
University of Maryland (2014) https://extension.umd.edu/sites/default/files/_docs/programs/master-
gardeners/Howardcounty/Baywise/PermeablePavingHowardCountyMasterGardeners10_5_11%20Final.pd
f. Permeable Pavement Fact Sheet Inf Howard Cty , Maryl Homeown 12.
Urban Drainage and Flood Control District. 2010. Urban storm drainage. Criteria Manual. Volume 3 – Best
Management Practices. Water Resources Publications, LLC. Denver, CO. 589pp
Urrestarazu S (2013) Sustainable drainage systems as ecosystem services - Case study: urban catchment in the
city of Montevideo, Uruguay. 130.
USEPA http://cfpub.epa.gov/npdes/stormwater/menuofbmps/index.cfm?action=results - Accessed
7/10/2013
USEPA (2000) Low Impact Development ( LID ).
USEPA (2010) STORM WATER MANAGEMENT MODEL USER ’ S MANUAL. 285.
USEPA (2013a) SWMM Updates. www.epa.gov/nrmrl/wswrd/wq/models/swmm/epaswmm5_updates.txt.
USEPA (2007) Reducing Stormwater Costs through Low Impact Development (LID) Strategies and Practices.
USEPA (2013b) http://cfpub.epa.gov/npdes/stormwater/menuofbmps.
Vojinovic Z, Abbott MB (2012) Flood Risk and Social Justice. 563.
Yang R, Cui B (2012) Framework of Integrated Stormwater Management of Jinan City, China. Procedia Environ
Sci 13:2346–2352. doi: 10.1016/j.proenv.2012.01.223
Young KD, Dymond RL, Kibler DF (2009) Virginia ’ s Stormwater Impact Evaluation : Developing an
Optimization Tool for Improved Site Development , Selection and Placement of Stormwater Runoff BMPs
By.
Zhang G, Hamlett JM, Reed P, Tang Y (2013) Multi-Objective Optimization of Low Impact Development
Designs in an Urbanizing Watershed. 2013:95–108.
90
Appendix A - Rainfall-runoff effect and IDF calculation.
Next table and graph show the procedure followed in IDF calculation for the case of events with duration
of 6 hours.
IDF calculation procedure (Pathirana 2013), example with 6 hours duration.

Year Max i(mm/h) Rank Total RP Pe P (CPcorrect) - ln(P) -ln(-ln(P)) i (mm/h)
1996 6.6 1 17 1.06 0.94 0.06 2.89 -1.06 6.7
1994 6.8 2 17 1.1 0.9 0.11 2.20 -0.79 7.6
1991 8.8 3 17 1.2 0.8 0.17 1.79 -0.58 8.2
2003 8.8 4 17 1.3 0.8 0.22 1.50 -0.41 8.8
1997 9.9 5 17 1.4 0.7 0.28 1.28 -0.25 9.3
2001 10.1 6 17 1.5 0.7 0.33 1.10 -0.09 9.8
2000 10.2 7 17 1.6 0.6 0.39 0.94 0.06 10.3
1990 11.4 8 17 1.8 0.6 0.44 0.81 0.21 10.8
1995 11.6 9 17 2.0 0.5 0.50 0.69 0.37 11.4
1998 11.9 10 17 2.3 0.4 0.56 0.59 0.53 11.9
2004 12.0 11 17 2.6 0.4 0.61 0.49 0.71 12.5
1993 12.1 12 17 3.0 0.3 0.67 0.41 0.90 13.1
2006 13.8 13 17 3.6 0.3 0.72 0.33 1.12 13.8
1999 14.9 14 17 4.5 0.2 0.78 0.25 1.38 14.7
2002 16.6 15 17 6.0 0.2 0.83 0.18 1.70 15.8
2005 17.0 16 17 9.0 0.1 0.89 0.12 2.14 17.2
1992 19.0 17 17 18.0 0.1 0.94 0.06 2.86 19.6
20 0.1 0.95 0.05 2.97 19.9
50 0.0 0.98 0.02 3.90 23.0
Linearization of exceeding probability and maximum intensity relation,

example for 6 hours rainfall duration.
Appendix B - Model calibration results.
92
Figure A2.1: Hydrologic model calibration results: rainfall, MM hydrographs (target) and SWMM
modelling results (simulated).
94
Figure A2.2: Hydrologic model calibration result: simulated discharge in main channel (computed) and
result from MM model (observed).
Figure A2.3: Hydrologic model calibration result: simulated discharge in secondary weir (computed) and
result from MM model (observed).
Figure A2.4: Hydrologic model calibration result: simulated hydrograph in downstream pipe (computed)
and result from MM model in hte same pipe (observed).
96
Appendix C - Rainfall-Runoff analysis.
Next figure illustrates as an example a case of rainfall and its effect in terms of CSO in the main channel.
Analysis of 20/12/2006 storm and its effect in CSO discharge.
Next table presents the results from the analysis of one summer period. Events with values of one year or
less of return period are grouped together. In this case for example, the summer has 24 rainy days with 35
storms. Analysing these 35 storms only 24 are producing discharges (CSO's) higher than 10 l/s. From these
24 storms, 19 have return period equal or smaller than 1 year. As a result, if discharges are avoided for
storms with return period equal to 1 year the problem will be solved in 79% of cases during this summer.
Rainfall-CSO analysis results for summer 2003/2004.
Summer 2003/2004
Day d (h) imax (mm/h) Tr (years) Max Disch (l/s) Tc (h) Disch. Time (h)
3 12 2.8 <1 1700 2:55 20
4 1 0.6 <1 < 10
7 12 6.7 8 6223 1:20 22
8 2 1 <1 117 2:40 13
10 2 9 <1 3197 1:45 15
13 2 51.6 > 50 54438 1:10 15
14 2 0.8 <1 < 10
2 1 <1 15 3:00
15
1 1.2 <1 21 2:10 12
20/21 2 8 <1 3507 2:05 15
2 0.5 <1 < 10
26
2 2.2 <1 245 4:40
4 2.5 <1 805 2:10 16
27
2 0.4 <1 < 10
29 1 3.9 <1 105 2:40 13
34 1 1.3 <1 8.6 3:45 13
1 0.2 <1 < 10
40
4 12.1 4 11058 1:20 18
1 0.4 <1 < 10
41
1 0.1 <1 < 10
1 0.1 <1 < 10
44
1 1.5 <1 11 3:20 14
1 0.3 <1 < 10
61 9 5.1 1 4689 2:35
2 8.4 <1 4549 1:30
62 4 1.8 <1 1692 1:15 18
1 0.2 <1 < 10
74
7 13.5 12 12865 1:15
74/75 8 7.6 4 7451 2:15 22
77 2 6.2 <1 1699 2:05 13
80 1 2.3 <1 20 2:50 14
2 1.1 <1 15 3:45
81
1 2.4 <1 30 2:10
3 3.9 <1 1359 2:00 16
82
1 0.1. <1 < 10
98
Appendix D - BMPs analysis (USEPA 2013b)
1. DRY EXTENDED DETENTION POND

• Characteristics: Dry structure that temporary retains runoff. Stormwater has to be detained at least
24 hours to allow pollutants removal.
• Application: Water quality improvement (sedimentation) and peak runoff reduction. Has little
control over runoff volume. Can be easily designed for flood control. Has little control over runoff
volume. Provides moderate pollutants removal if it is considered in designing stage; it is not
efficient in soluble pollutants removal.
• Functions: Volume reduction: Low; GW recharge: None; Peak rate control: High; WQ
improvement: Low.
• Constraints: Soil group A is not adequate due to high infiltration. It needs at least 4Ha of drainage
area to allow residence time for pollutants removal avoiding a too small outlet. Does not have
slope or soil limitations more than soils group A in which case impermeable liner is needed. If
ground water is too superficial the pond can be permanently wet causing public safety issues
regarding mosquitoes breeding. Even when maintenance minimization can be considered in design
annual and semi-annual maintenance is needed.
• Observations: Soils group D are beneficial. Low aesthetic liability. It is one of the most used BMP
but requires space and could be not the most appropriated if the objective is water quality
improvement. It is good option for retrofit. It needs relatively continuous area per unit, but this
value is not large comparing with treated area (2-3% of drainage area). One variation of this
practice is the extended detention pond enhanced; it is a detention pond with permanent marsh area
to improve pollutants removal. It has better pollutants removal through sedimentation, absorption
and biological treatment. It can have public safety issues regarding mosquitoes breeding.
• Effectiveness: TSS 61%; Phosphorous 19%; Nitrogen31%, Metals 26/54%.
• Cost: An average construction cost is 150 Eu/m3. Maintenance cost is estimated in 3-5% of
construction cost. Dry ponds are long live facilities, typically more than 20 years. These structures
can detract the value of adjacent properties in 3-10%.
2. RETENTION POND
• Characteristics: Permanent pools with additional capacity to storage runoff, it releases this
volume slowly.
• Application: Water quality improvement (high pollutants removal) and peak runoff
reduction.
• Functions: Volume reduction: Low; GW recharge: Low; Peak rate control: High; WQ
improvement: Medium
• Constraints: Needs large contribution areas (minimum of 4Ha) because it needs an

adequate base flow. Maximum drainage area is 2600Ha. Shallow groundwater is
favourable as source of base flow. Soil groups A and B are not adequate due to high
infiltration. Public safety issues. Large areas are needed to place these structures.
• Observations: Soils group D are beneficial. High aesthetic liability. Between 1 and 3% of
contributing area is needed for the pond construction. High pollutants removal is reached
when the pond is long and narrow (minimum of 2:1 ratio is required, 4:1 ratio as ideal). It
is a good option in places with shallow groundwater.
• Effectiveness: Retention basins are capable of removing sediments, BOD, organic

nutrients and trace metals from runoff. Also, significant peak reduction can be achieved
with this practice. Biological processes allow the reduction of soluble nutrients.
• Cost: It adds value to the site where is applied because aesthetic value improvements.
100
3. INFILTRATION BASIN
• Characteristics: Extended pond that stores runoff and infiltrates a designated portion.
• Application: Serves as peak mitigation facility for flood control. Reduces runoff volume trough
infiltration and evaporation. High pollutants removal efficiency (filtering and biological treatment).
• Functions: Volume reduction: High / GW recharge: High / Peak rate control: Medium-High / WQ
improvement: High
• Constraints: No suitable for shallow groundwater depths (0.7 to 1.7m from groundwater table) and
bedrock. Soil group A is not adequate because too high infiltration; soil group D is not adequate
because low infiltration (rate: 13 to 76/300 mm/h). Slope lower than 5%, preferably flat. Space
limitations and low infiltration in highly urbanized areas. Infiltrated water interferes with existing
infrastructure and water supply wells. No allowed to receive stormwater hotspot. Applicable for
less than 2-4Ha area size. No easily applicable as stormwater retrofit in developed areas. High
maintenance needed to avoid clogging.
• Observations: Allows groundwater recharge. High failure rates (clogging). Aesthetic practices and
public safety can be reached if clogging is avoided (complete drained in 72hrs).
• Effectiveness: TSS 75%; Phosphorous 60/70%; Nitrogen 55/60%, Metals 85/90%, Bacteria 90%
• Cost: Relatively cost-effective practices. Construction: 50 Eu/m3 Maintenance: 5-10% of

construction cost.
4. INFILTRATION TRENCHES
• Characteristics: Shallow trenches underdrain by an underground reservoir filled with coarse stone
aggregate.
• Application: Water quality BMP (filtering) that can provide peak runoff reduction (1 or 2 years of
return period) by increasing its volume if is equipped with underground piping.
• Functions: Volume reduction: Medium; GW recharge: High; Peak rate control: Medium; WQ
improvement: High
• Constraints: No suitable for shallow groundwater depths and bedrock (separation 0.65-1.65m).
Soil group A is not adequate because too high infiltration*. Soil group D is not adequate because
low infiltration* (between 12 and 76 mm/h). Slope lower than 5% and preferably in flat areas; the
drainage area can be steep. In urbanized areas the main restrictions are interference of infiltrated
water with existing infrastructure and low permeability of urban soils. Applicable for small
drainage areas, less than 2-4Ha with high impervious cover to avoid clogging. High maintenance;
high failure rates that can be decreased with frequent maintenance, adequate location and design.
• Observations: Allows groundwater recharge. It does not provide visual enhancement or aesthetic
improvements. During small storm events the runoff volume reduction can be significant.
• Effectiveness: TSS 75%; Phosphorous 60/70%; Nitrogen 55/60%, Metals 85/90%, Bacteria 90%
• Cost: Expensive compared with other BMPs if cost per area treated is considered. Construction
cost 130 Eu/m3.
102
5. POROUS PAVEMENT
• Characteristics: Pervious surface placed over a gravel reservoir sustained by highly porous
(permeable) soil. There are three types: porous asphalt (PA), permeable interlocking concrete
pavement (PICP) and pervious concrete pavement (PCP).
• Application: Mainly for water quality but provides peak rate and volume runoff reduction (1 or 2
years of return period) by increasing storage volume in the reservoir if is equipped with
underground piping.
• Functions: Volume reduction: Medium / GW recharge: Medium / Peak rate control: Medium / WQ
improvement: Medium
• Constraints: Parking areas or areas with light traffic (no high traffic). No suitable for shallow
groundwater depths and bedrock. Soil group A is not adequate because too high infiltration*, soil
group D is not adequate because low infiltration* (rate: 13 to 300 mm/h). Slope lower than 2-5%.
It needs to be protected from high sediment loads. Infiltrated water interferes with existing
infrastructure and water supply wells. Maintenance is needed to avoid clogging. Applicable for
less than 4Ha area size for good cost effectiveness.
• Observations: Allows groundwater recharge. Used mainly for pedestrian and parking lots.
Municipal use to reduce combined sewer overflows and localized flooding. It is needed 40% of
void spaces in the stone underdrain bed, and 0.3-0.9 depth.
• Effectiveness: Volume retention: 25/97%(PA), 45/100%(PICP), 100%(PCP). Pollutants removal:

TSS=95%(PA), 70/80%(PICP) / 90%(PCP); Metals=75/95%(PA), 65/90%(PICP), 75/90%(PCP);
Nutrients=40%(PA, 35/65%(PICP)
• Cost: High variability in costs. PA material: 4.5-9 Eu/m2. PICP material: 35-70 Eu/m2. PCP
material: 20-80 Eu/m2. In general 10-20% more expensive than standard pavements.
6- GREEN ROOFS
• Characteristics: Sheet of vegetation that covers an otherwise conventional roof. Two types of
green roofs: extensive with less than 15cm of growing medium (stormwater management as main
function) and intensive with more than 15cm of growing medium (also aesthetic and architecture
objectives).
• Application: If it is properly installed provides rainfall retention, runoff reduction and water quality
improvement.
• Functions: Volume reduction: Medium/High; GW recharge: None; Peak rate control: Low; WQ
improvement: Medium
• Constraints: Structures have to be able to support the loading of green roof materials under
saturated conditions. The slope of the roof can range between 0 and 40 degrees. Low maintenance
is required, more frequent at the beginning until establishment of vegetation is reached.
• Observations: Can be used with shallow groundwater. Efficient in CSO reduction strategies and
reduction of urban heat island effect. Good alternative for high urbanized areas and retrofit of
developed areas. Drainage media and underdrain piping should be designed to face 2 year return
period frequency events.
• Effectiveness: Extensive green roofs can absorb 50mm of rainfall. Intensive green roofs can absorb
100mm of rainfall.
• Cost: Extensive green roofs: 40-170 Eu/m2. Intensive green roofs: 170-680 Eu/m2. It is an
attractive option when long term maintenance costs are considered. Annualized costs are lowered
if energy savings and recreational space opportunities are considered.
104
7- BIORETENTION
• Characteristics: Is an infiltration area filled with specific vegetation. Different categories: basins
and filters (with and without infiltration); and off-line and on-line.
• Application: Primary function: water quality improvement (adsorption, filtration, volatilization,

ion exchange and biological decomposition). Can provide runoff reduction by infiltration.
• Functions: Volume reduction: Medium; GW recharge: Medium/High; Peak rate control:

Low/Medium; WQ improvement: Medium/High.
• Constraints: No areas greater 2Ha. Basins case is no suitable for shallow groundwater depths and
bedrock; it case also needs soil sufficiently permeable to allow infiltration. Slope has to be lower
than 5%. Requires landscaping maintenance. Not designed to provide flood control but can restore
pre-development hydrology.
• Observations: Allows groundwater recharge. Applied to small areas; can be used in highly
urbanized sites. Occupy about 5% of the drained area. Good retrofit option but expensive to retrofit
the entire watershed (this is general for BMPs treating small areas). Adds aesthetic value.
• Effectiveness: P: -65/95%; N: -15/65%; Metals: -65/95%.
• Cost: Relative expensive BMP but it is lower if it is considered that the area has to be landscaped
anyway. The cost function is: C=7.3*V^0.99; with C=investment cost, V=volume of water treated.
8- SAND FILTERS
• Characteristics: Adequate for water with highly polluted, works as pre-treatment and temporary
runoff storage. It stores the runoff temporarily and after percolates the runoff through the filter
media.
• Application: Primary it is a water quality BMP but can provide runoff peak control because its
storage capacity.
• Functions: Volume reduction: Low; GW recharge: Low; Peak rate control: Low; WQ
improvement: High
• Constraints: Soil group A is not adequate due to high infiltration. Slope lower than 20%. Should be
located 0.6 to 1.2m above water table. The main cause of failure it the clogging of the media. This
practice is adequate for small drainage areas; the maximum area depends of the filter type but
values range between 1000m2 and 4Ha.
• Observations: Sand filters are adequate to treat runoff from highly urbanized areas (67 to 100% of
imperviousness). It is a good option in places where aesthetic issues are relevant.
• Effectiveness: Water quality is improved through physical, chemical and biological mechanisms.
Peak rate control of lower than 10 years return period rain events is possible.
• Cost: If it is applied for large areas the cost effectiveness of the filter is reduced.
106
9- GRASS SWALE / VEGETATED SWALE
• Characteristics: Surface depression grassed which collects and conveys runoff. Three cases:
grassed channels (GCH), dry swales (DS) and wet swales (WS).
• Application: Water quality improvement facility (sedimentation and filtration), if it is properly

designed. High pollutant removal potential.
• Functions: Volume reduction: Low-Medium / GW recharge: Low-Medium / Peak rate control:

Medium-High / WQ improvement: Medium-High
• Constraints: Suitable option for highway runoff treatment because are linear practices. Cannot be
considered as flood protection measures. Can be used with shallow groundwater is it is wet type, if
not 0.6m of distance are required. Soil below this practice needs some infiltration capacity to allow
filtration, if not a filtration bed has to be added (at least 0.75m). Soil group A is not adequate due
to high infiltration. Slope lower than 2-6%; if not check dams are needed. Low velocities are
required to allow pollutants removal. No easily applicable as stormwater retrofit in developed
areas because space constraints. No adequate to receive stormwater hotspot. Applicable for less
than 4Ha area. Public safety issues in wet case. Regular maintenance (annual).
• Observations: Allows groundwater recharge. One retrofit opportunity is the use to modify existing
drainage ditches (for ex. including check dams to improve infiltration). In developed areas can be
expensive due to the quantity of units needed. If possible should convey 10-year flows with 0.15m
of freeboard.
• Effectiveness: TSS: 60/80%(GCH), 80/99%(DS), 70/80% (WS); TP: 5/45%(GCH), 20/99%(DS),

20/40%(WS); N: 40%(GCH), 84/99%(DS), 40%(WS); Metals: 15/70%(GCH), 40/99%(DS),
6/70(WS); Bacteria: <0
• Cost: Are one of the lowest cost BMP options. An approximated cost is 13 Eu/m2.
10- VEGETATED FILTER STRIP / GRASS-VEGETATED BUFFER
• Characteristics: Densely vegetated strip of land. The type of vegetation ranges from native species
to forest.
• Application: Primary water quality improvement practice that works extremely well as pre-
treatment measure for other BMPs (filtering, sedimentation and infiltration). This practice has not
high pollutant removal capacity.
• Functions: Volume reduction: Low-Medium / GW recharge: Low-Medium / Peak rate control:

Low / WQ improvement: High
• Constraints: Cannot be considered as flood protection measures. At least 0.6-1.2m of separation

with groundwater is needed. Soil group A is not adequate due to high infiltration. Soil below this
practice needs some infiltration capacity (between 1.5-15mm/hr). Slope between 2-6%. If water is
stagnant due to low slope or bad design, can cause public safety issues. Difficulty to maintain sheet
flow and avoid short circuits from concentrated flow. It needs large amount of space, typically
equal to the impervious area they treat, so are impractical in ultra-urbanized areas. No adequate to
receive stormwater hotspot. Cannot treat large drainage areas, no more than 0.4-1Ha and 200m of
length, to avoid concentrated flow. Maintenance is very important to avoid short circuit.
• Observations: Are suited to treat water from roads. Water velocity should be lower than 0.3m/s.
Residence time needs to be higher than 5-9min. If this practice is integrated into a landscaping
plan, concerns about area requirements are negligible.
• Effectiveness: TSS: 55/85%; P: -25/40%; N: -27/20%; Metals: -16/55%
• Cost: As investment cost the land cost is considered because the large area needed, as a
consequence the cost is relatively high for urban areas. Maintenance cost is estimated in 670
Eu/Ha/Year.
108
11- CONSTRUCTED WETLANDS
• Characteristics: Structural BMP seeded with diverse and emergent aquatic vegetation.
• Application: Improves water quality by physical (sedimentation, adsorption and filtration),

chemical (metals precipitation) and biological (nutrients removal and degradation) means.
• Functions: Volume reduction: Low; GW recharge: Low; Peak rate control: Low; WQ
improvement: High
• Constraints: Needs large contribution areas, no less than 4Ha. Baseflow is needed to ensure the
adequate practice performance. Shallow groundwater is favourable as source of baseflow. Soil
group A is not adequate due to high infiltration, but in general it can be accomplished in the
presence of a variety of soil types. Should not be constructed closer than 15m of surfaces steeper
than 10%. Baseflow, runoff and groundwater have to be greater than losses due to evaporation and
infiltration. Significant land area is required for constructed stormwater wetlands.
• Observations: High aesthetic liability when are properly designed and maintained. Public safety
issues. Should not be used as runoff quantity control practice because it causes rapid fluctuations
in water levels producing deterioration of vegetation.
• Effectiveness: This practice improves runoff quality in the same manner as retention and enhanced
extended detention basin.
12- RAINWATER HARVESTING
• Characteristics: Devices that intercept precipitation and store it, allowing its use. These devises
can be cisterns, rain barrels and vertical storages.
• Application: It is a runoff control strategy that impacts on runoff volume, peak rate and quality.
• Functions: The use of captured runoff volume has a positive impact on the volume, peak rate and
quality of stormwater.
• Constraints: Rainwater demand for reuse application has to be determined before the application of
this practice. The location of the rainwater harvesting device is important due to evaporation
losses.
• Observations: Can be used with shallow groundwater. Stored water can be used for irrigation, fire
protection or flushing toilets.
• Effectiveness: In combined systems the runoff volume reduction has strong benefit in reducing the
frequency of CSO. Several factors influence the pollutants removal performance of this system.
110
Appendix E - BMPs applicability analysis.
EACU
• Infiltration trenches (2600m)
• Pervious pavements (10%)
• Green roofs (7%)
• 82 ha
• 2.9Ha of big roofs
• 70% Private area / 30% public area
• 2600m of wide roads with 10% of total area
FELI
• Storage (1400m2)
• 37.2 ha
• 11 Ha Hospital area
FRAN
• 66.2 ha
BATL
• Infiltration trenches (2200m/460m outside park)
• Infiltration basin (8000m2)
• Surface storage (15000m2)
• 50.1 ha
• Public area
MFRA
• 21.3 ha
HAED
• 17.7 ha
112
CIPR
• 36.6 ha
ALEJ
• 21.7 ha
AYAC
• 82.8 ha
CIAN
• 23.7ha
BAUZ
• 33 ha
BUXA
• 12.5 ha
114
ECHE
• Surface storage (8400m2)
• 71.5 ha
BUCE
• 20.1 ha
Appendix F - Total present value calculation
PERVIOUS PAVEMENTS INFILTRATION TRENCHES GREEN ROOFS STORAGES

Construction Cost Construction Construction Cost Construction Cost
2 85 2 49 2 104 2 1446
(Eu/m ) Cost (Eu/m ) (Eu/m ) (Eu/m )
Annual O&M Cost Annual O&M Annual O&M Cost Annual O&M Cost
2 4.3 2 7.4 2 5.2 3 7.2
(Eu/m ) Cost (Eu/m ) (Eu/m ) (Eu/m )
1 4.0 1 7.0 1 5.0 1 6.9
2 3.9 2 6.7 2 4.7 2 6.6
3 3.7 3 6.3 3 4.5 3 6.2
4 3.5 4 6.0 4 4.3 4 5.9
5 3.3 5 5.8 5 4.1 5 5.7
6 3.2 6 5.5 6 3.9 6 5.4
7 3.0 7 5.2 7 3.7 7 5.1
8 2.9 8 5.0 8 3.5 8 4.9
9 2.7 9 4.7 9 3.4 9 4.7
10 2.6 10 4.5 10 3.2 10 4.4
11 2.5 11 4.3 11 3.0 11 4.2
12 2.4 12 4.1 12 2.9 12 4.0
13 2.3 13 3.9 13 2.8 13 3.8
14 2.1 14 3.7 14 2.6 14 3.7
15 2.0 15 3.5 15 2.5 15 3.5
16 1.9 16 3.4 16 2.4 16 3.3
17 1.9 17 3.2 17 2.3 17 3.2
18 1.8 18 3.1 18 2.2 18 3.0
19 1.7 19 2.9 19 2.1 19 2.9
20 1.6 20 2.8 20 2.0 20 2.7
Total O&M Cost 53.0 Total O&M Cost 91.6 Total O&M Cost 64.8 Total O&M Cost 90.1
Total Present Total Present Total Present Total Present
2 138 141 2 169 2 1536
Worth (Eu/m ) Worth (Eu/m) Worth (Eu/m ) Worth (Eu/m )
Total Present 141
2
116
Appendix G - Continuous simulation results
Max BMPs Reduction

SCENARIO 1: GR Rainfall Do nothing
application (%)
Events 24 8 8 0%
SUMMER 3
Volume (m ) 821095 746987 9%
90/91
Max Flow (l/s) 18620 15212 18%
Events 26 8 7 13%
SUMMER 3
Volume (m ) 575782 518876 10%
93/94
Max Flow (l/s) 44285 41717.9 6%
Events 17 10 9 10%
SUMMER 3
Volume (m ) 693651 618616 11%
94/95
Max Flow (l/s) 15396 9849 36%
Events 24 11 8 27%
SUMMER 3
Volume (m ) 625699 569859 9%
01/02
Max Flow (l/s) 41429 35629.9 14%
Events 36 13 13 0%
SUMMER 3
Volume (m ) 708182 625634 12%
02/03
Max Flow (l/s) 12324 8267 33%
Events 27 15 13 13%
SUMMER 3
Volume (m ) 786032 712630 9%
03/04
Max Flow (l/s) 48674 42691 12%
Events 25 13 11 15%
SUMMER 3
Volume (m ) 1326918 1276933 4%
04/05
Max Flow (l/s) 32951 31974 3%
Events 24 14 12 14%
SUMMER 3
Volume (m ) 695472 632174 9%
05/06
Max Flow (l/s) 10117 8355 17%
Events 27 18 17 6%
SUMMER 3
Volume (m ) 693026 648724 6%
06/07
Max Flow (l/s) 25045 18230 27%
Max BMPs Reduction
SCENARIO 2: PP Rainfall Do nothing
application (%)
Events 24 8 8 0%
3
SUMMER 90/91 Volume (m ) 821095 717344 13%
Max Flow (l/s) 18620 15060 19%
Events 26 8 8 0%
3
SUMMER 93/94 Volume (m ) 575782 508706 12%
Max Flow (l/s) 44285 35887 19%
Events 17 10 9 10%
3
SUMMER 94/95 Volume (m ) 693651 590651 15%
Max Flow (l/s) 15396 10323 33%
Events 24 11 9 18%
3
SUMMER 01/02 Volume (m ) 625699 541410 13%
Max Flow (l/s) 41429 34468 17%
Events 36 13 13 0%
3
SUMMER 02/03 Volume (m ) 708182 595876 16%
Max Flow (l/s) 12324 8759 29%
Events 27 15 14 7%
3
SUMMER 03/04 Volume (m ) 786032 679844 14%
Max Flow (l/s) 48674 39659 19%
Events 25 13 11 15%
3
SUMMER 04/05 Volume (m ) 1326918 1205944 9%
Max Flow (l/s) 32951 29968 9%
Events 24 14 12 14%
3
SUMMER 05/06 Volume (m ) 695472 595092 14%
Max Flow (l/s) 10117 8392 17%
Events 27 18 18 0%
3
SUMMER 06/07 Volume (m ) 693026 590449 15%
Max Flow (l/s) 25045 19097 24%
118
Max BMPs Reduction
SCENARIO 3: IT Rainfall Do nothing
application (%)
Events 24 8 8 0%
3
SUMMER 90/91 Volume (m ) 821095 812475 1%
Max Flow (l/s) 18620 17704 5%
Events 26 8 8 0%
3
SUMMER 93/94 Volume (m ) 575782 567376 1%
Max Flow (l/s) 44285 42750 3%
Events 17 10 10 0%
3
SUMMER 94/95 Volume (m ) 693651 567376 18%
Max Flow (l/s) 15396 13032 15%
Events 24 11 11 0%
3
SUMMER 01/02 Volume (m ) 625699 616536 1%
Max Flow (l/s) 41429 40483.8 2%
Events 36 13 13 0%
3
SUMMER 02/03 Volume (m ) 708182 699889 1%
Max Flow (l/s) 12324 10030 19%
Events 27 15 15 0%
3
SUMMER 03/04 Volume (m ) 786032 779913 1%
Max Flow (l/s) 48674 47036 3%
Events 25 13 13 0%
3
SUMMER 04/05 Volume (m ) 1326918 1314579 1%
Max Flow (l/s) 32951 32507 1%
Events 24 14 14 0%
3
SUMMER 05/06 Volume (m ) 695472 692669 0%
Max Flow (l/s) 10117 9426 7%
Events 27 18 18 0%
3
SUMMER 06/07 Volume (m ) 693026 686992 1%
Max Flow (l/s) 25045 23133 8%
Max BMPs Reduction
SCENARIO 4: GR + PP + IT Rainfall Do nothing
application (%)
Events 24 8 8 0%
3
SUMMER 90/91 Volume (m ) 821095 712016 13%
Max Flow (l/s) 18620 14693 21%
Events 26 8 8 0%
3
SUMMER 93/94 Volume (m ) 575782 503248 13%
Max Flow (l/s) 44285 37108 16%
Events 17 10 9 10%
3
SUMMER 94/95 Volume (m ) 693651 585840 16%
Max Flow (l/s) 15396 9958 35%
Events 24 11 8 27%
3
SUMMER 01/02 Volume (m ) 625699 538571 14%
Max Flow (l/s) 41429 34023 18%
Events 36 13 13 0%
3
SUMMER 02/03 Volume (m ) 708182 590687 17%
Max Flow (l/s) 12324 8431 32%
Events 27 15 13 13%
3
SUMMER 03/04 Volume (m ) 786032 675895 14%
Max Flow (l/s) 48674 39734 18%
Events 25 13 11 15%
3
SUMMER 04/05 Volume (m ) 1326918 1214023 9%
Max Flow (l/s) 32951 30242 8%
Events 24 14 12 14%
3
SUMMER 05/06 Volume (m ) 695472 594563 15%
Max Flow (l/s) 10117 8215 19%
Events 27 18 17 6%
3
SUMMER 06/07 Volume (m ) 693026 577321 17%
Max Flow (l/s) 25045 18294 27%
120
Max
Reduction
SCENARIO 5: St Rainfall Do nothing applicatio
(%)
n
Events 24 8 8 0%
3
SUMMER 90/91 Volume (m ) 821095 798545 3%
Max Flow (l/s) 18620 12704 32%
Events 26 8 8 0%
3
SUMMER 93/94 Volume (m ) 575782 554614 4%
Max Flow (l/s) 44285 26787 40%
Events 17 10 10 0%
3
SUMMER 94/95 Volume (m ) 693651 671933 3%
Max Flow (l/s) 15396 8535 45%
Events 24 11 9 18%
3
SUMMER 01/02 Volume (m ) 625699 606966 3%
Max Flow (l/s) 41429 26027 37%
Events 36 13 13 0%
3
SUMMER 02/03 Volume (m ) 708182 684152 3%
Max Flow (l/s) 12324 7190 42%
Events 27 15 14 7%
3
SUMMER 03/04 Volume (m ) 786032 767110 2%
Max Flow (l/s) 48674 30080 38%
Events 25 13 13 0%
3
SUMMER 04/05 Volume (m ) 1326918 1308970 1%
Max Flow (l/s) 32951 32887 0%
Events 24 14 13 7%
3
SUMMER 05/06 Volume (m ) 695472 678552 2%
Max Flow (l/s) 10117 7701 24%
Events 27 18 18 0%
3
SUMMER 06/07 Volume (m ) 693026 661623 5%
Max Flow (l/s) 25045 13394 47%
Max BMPs Reduction
SCENARIO 6: GR + PP + IT + St Rainfall Do nothing
application (%)
Events 24 8 6 25%
3
SUMMER 90/91 Volume (m ) 821095 681294 17%
Max Flow (l/s) 18620 10975 41%
Events 26 8 6 25%
3
SUMMER 93/94 Volume (m ) 575782 476493 17%
Max Flow (l/s) 44285 20424 54%
Events 17 10 9 10%
3
SUMMER 94/95 Volume (m ) 693651 562188 19%
Max Flow (l/s) 15396 6733 56%
Events 24 11 8 27%
3
SUMMER 01/02 Volume (m ) 625699 521509 17%
Max Flow (l/s) 41429 19688 52%
Events 36 13 13 0%
3
SUMMER 02/03 Volume (m ) 708182 564747 20%
Max Flow (l/s) 12324 5943 52%
Events 27 15 13 13%
3
SUMMER 03/04 Volume (m ) 786032 652220 17%
Max Flow (l/s) 48674 21467 56%
Events 25 13 11 15%
3
SUMMER 04/05 Volume (m ) 1326918 1192649 10%
Max Flow (l/s) 32951 30126 9%
Events 24 14 11 21%
3
SUMMER 05/06 Volume (m ) 695472 572544 18%
Max Flow (l/s) 10117 6556 35%
Events 27 18 16 11%
3
SUMMER 06/07 Volume (m ) 693026 545117 21%
Max Flow (l/s) 25045 11242 55%
122

Masters2 88277

Uploaded by

Copyright:

Available Formats

Masters2 88277

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

Masters2 88277

Uploaded by

Copyright:

Available Formats

Model based multi-objective

evaluation of BMP system

Alida Alves Beloqui

Master of Science Thesis

List of Tables xiii

5. Data analysis and model set up 29

6. Best Management Practices 45

8. Results and discussion 64

9. Conclusions and recommendations 85

1.2. Problem definition

Figure 1.2 Interrelations between the different levels of problems.

1.3. Research questions

1.5. Research objectives

1.6. Thesis outline

2.1. Best Management Practices

2.1.2. Review of different BMPs

Pollutant removal processes include sedimentation, straining, filtration, adsorption/absorption and

2.1.4. BMPs selection

2.3. Optimization of urban drainage systems

2.4. Rainfall selection

3.1. Research approach

The research methodology consists of six phases described hereafter:

PRACTICES SCREENING. BMP DESIGN

Figure 3.1 Different phases of research methodology.

Rainfall data processing:

4.1.1. Hydraulic model adjustment

Table 4.2 Area and population values for each

Figure 4.4 Main CSO discharge channel.

4.1.3. Rainfall pattern

Annual rainfall distribution

Figure 4.6 Annual rainfall distribution in two different meteorological stations.

4.1.4. Hydraulic model description

Rainfall Tr=10 years, d=6 hours

Figure 4.7 Rainfall event used by MM in the model (mm/h - h).

Data analysis and model set up

5.1. Data analysis

Table 5.1 Different surfaces and coefficients considered.

5.1.3. Rainfall data processing

= − × log {−log [(1 − ( > ) } (5.1)

Annex A shows one example of calculation using this procedure.

Table 5.3 Maximum intensities expected for different event durations

Figure 5.3 IDF curves.

5.2. Model set up

Table 5.4 Main characteristics of total and modified models.

Model based multi-objective

Figure 5.7 Final hydrodynamic model of Pocitos watershed.

Table 5.5 Calibration parameters final values and calibration results.

To evaluate results from calibration process, different

The first parameter calculated is the Nash-Sutcliffe

Model based multi-objective

Table 5.6 Goodness of fit parameters.

5.3. Design storm selection

To develop an Intensity-Duration-Frequency analysis the annual maximum events were considered,

Table 5.8 Results of continuous summer rainfall modelling analysis.

5.3.2. Design storm calculation

= 56.5%(" + 2.65) (5.2)

Where i is the intensity in mm/h and d is the accumulated time in h.

Figure 5.10 Design storm.

5.3.3. Evaporation conditions