Water Supply System Management Design

and Optimization under Uncertainty

Item Type text; Electronic Dissertation

Authors Chung, Gunhui

Publisher The University of Arizona.

Rights Copyright © is held by the author. Digital access to this material

is made possible by the University Libraries, University of Arizona.
Further transmission, reproduction or presentation (such as
public display or performance) of protected items is prohibited
except with permission of the author.

Download date 24/03/2023 08:02:50

Link to Item http://hdl.handle.net/10150/195506

WATER SUPPLY SYSTEM MANAGEMENT DESIGN AND OPTIMIZATION

UNDER UNCERTAINTY

by

GUNHUI CHUNG

_______________________
Copyright © Gunhui Chung 2007

A Dissertation Submitted to the Faculty of the

DEPARTMENT OF CIVIL ENGINEERING AND ENGINEERING MECHANICS

In Partial Fulfillment of the Requirements

For the Degree of

DOCTOR OF PHILOSOPHY
WITH A MAJOR IN CIVIL ENGINEERING

In the Graduate College

THE UNIVERSITY OF ARIZONA

2007
 2

THE UNIVERSITY OF ARIZONA

GRADUATE COLLEGE

As members of the Dissertation Committee, we certify that we have read the dissertation
prepared by GUNHUI CHUNG
entitled WATER SUPPLY SYSTEM MANAGEMENT DESIGN AND
OPTIMIZATION UNDER UNCERTAINTY
and recommend that it be accepted as fulfilling the dissertation requirement for the
Degree of DOCTOR OF PHILOSOPHY

_______________________________________________________________________
Dr. Kevin Lansey Date: December 4, 2006

_______________________________________________________________________
Dr. Juan Valdes Date: December 4, 2006

_______________________________________________________________________
Dr. Larry W. Mays Date: December 4, 2006

_______________________________________________________________________
Dr. Donald R. Davis Date: December 4, 2006

_______________________________________________________________________
Dr. Guzin Bayraksan Date: December 4, 2006

_______________________________________________________________________
Dr. Bart Nijssen Date: December 4, 2006

Final approval and acceptance of this dissertation is contingent upon the candidate’s
submission of the final copies of the dissertation to the Graduate College.

I hereby certify that I have read this dissertation prepared under my direction and recommend
that it be accepted as fulfilling the dissertation requirement.

________________________________________________ Date: December 4, 2006

Dissertation Director: Dr. Kevin Lansey
 3

STATEMENT BY AUTHOR

This dissertation has been submitted in partial fulfillment of requirements for an advanced

degree at The University of Arizona and is deposited in the University Library to be made

available to borrowers under rules of the Library.

Brief quotations from this dissertation are allowable without special permission, provided

that accurate acknowledgment of source is made. Requests for permission for extended

quotation from or reproduction of this manuscript in whole or in part may be granted by the

copyright holder.

SIGNED: Gunhui Chung

 4

ACKNOWLEDGEMENTS

I would like to express my gratitude to all those who help me to complete this dissertation.

I am deeply indebted to my advisor Dr. Lansey for all his support during this work. He was
the one who gave me the opportunity and his restless guides, suggestions and encouragement
made it possible for me to complete this dissertation.

I would like to thank Dr. Bayraksan for her invaluable advices to overcome whenever
seemingly unsolvable problems challenged me. Without her help, this study could not have
been completed.

I also want to thank my other committee, Dr. Valdes, Dr. Mays, Dr. Davis, and Dr. Nijssen
for their generosity and invaluable comments to my humble works. It was my honor to have
such respectful scholars as my committee members.

My special thank goes to Chinmaya for helping me out in C ++ coding. I am also grateful to
all friends of mine, including Jim, Amanda, Richard, Pasha, David, Doo Sun and Tae-
Woong, for their help and friendship. All of you listened to me sincerely, always took my
side and helped me to get over it when I was in trouble.

Especially, I would like to thank Derya, my best friend. She made my closed mind open with
her sincere friendship and helped me survive in unfamiliar environment. I am sure that she
will become a great scholar.

I also want to thank my family, father, mother and two brothers, Chul-Ho and Min-Suk, for
their endless support. Chul-Ho helped me a lot when I started working on computer
programming. Their warm heart is always with me wherever I go.

Finally, I would like to give my special thanks to my husband Inhong whose patient love
gave me strength to confront whatever I was up to.
 5

TABLE OF CONTENTS
ABSTRACT ........................................................................................................................ 9

1. INTRODUCTION ........................................................................................................ 11

1.1 Problem Statement .................................................................................................. 11

1.2 Literature Review.................................................................................................... 12

1.2.1 Water Supply System Design .......................................................................... 12

1.2.2 Deterministic Optimization .............................................................................. 15

1.2.3 Stochastic Optimization ................................................................................... 16

1.3 Summary of Literature ............................................................................................ 18

2. PRESENT STUDY ....................................................................................................... 19

2.1 Dissertation Outline ................................................................................................ 19

2.2 Uniqueness of the Study ......................................................................................... 34

2.3 Conclusions and Future Work ................................................................................ 36

REFERENCES ................................................................................................................. 39

APPENDIX A: A GENERAL WATER RESOURECES PLANNING MODEL USING

DYNAMIC SIMULATION: EVALUATION OF DECENTRALIZED TREATMENT 46

ABSTRACT ...................................................................................................................... 47

1. INTRODUCTION ....................................................................................................... 49

2. LITERATURE REVIEW/BACKGROUND ................................................................ 50

3. MODELING TOOLS ................................................................................................... 53

3.1 Modeling Objective ................................................................................................ 53

3.2 Generic System ....................................................................................................... 54

 6

3.3 General Components ............................................................................................... 57

3.4 Demand/ User Components .................................................................................... 58

3.5 Water Quality Transformations .............................................................................. 62

4. APPLICATIONS .......................................................................................................... 66

4.1 Scenario 1 – Effectiveness of Conservation Practices ............................................ 69

4.2 Scenario 2 – Unavailability of Supply Sources ...................................................... 72

4.3 Scenario 3 – Decentralized Treatment .................................................................... 73

5. CONCLUSIONS........................................................................................................... 77

6. REFERENCES ............................................................................................................. 80

7. TABLES ....................................................................................................................... 86

8. FIGURES .................................................................................................................... 108

APPENDIX B. APPLICATION OF THE SHUFFLED FROG LEAPING ALGORITHM

FOR THE OPTIMIZATION OF A GENERAL LARGE-SCALE WATER SUPPLY

SYSTEM ......................................................................................................................... 126

ABSTRACT .................................................................................................................... 127

1. INTRODUCTION AND BACKGROUND ............................................................... 129

2. PROBLEM DESCRIPTION....................................................................................... 131

3. PROBLEM FORMULATION.................................................................................... 132

3.1 Simple Bound Constraints on System Flows and Component Sizes .................... 136

3.2 Node Constraints................................................................................................... 138

3.3 Arc Related Flow Constraints ............................................................................... 139

3.4 Water Quality Constraints..................................................................................... 141

 7

3.5 Objective Function Penalty Term ......................................................................... 143

3.6 Summary of Formulation ...................................................................................... 144

4. SHUFFLED FROG LEAPING ALGORITHM (SFLA) ............................................ 144

4.1 Global Exploration ................................................................................................ 146

4.2 Local Exploration: Frog Leaping Algorithm (FLA) ............................................. 147

5. APPLICATIONS ........................................................................................................ 151

5.1 Single Wastewater Treatment Plant System ......................................................... 151

5.2 Multiple Wastewater Treatment Plant System ..................................................... 154

6. CONCLUSIONS......................................................................................................... 157

7. NOMENCLATURE ................................................................................................... 159

8. REFERENCES ........................................................................................................... 164

9. TABLES ..................................................................................................................... 166

10. FIGURES .................................................................................................................. 180

APPENDIX C: RELIABLE WATER SUPPLY SYSTEM DESIGN UNDER

UNCERTAINTY ............................................................................................................ 187

ABSTRACT .................................................................................................................... 188

1. INTRODUCTION ...................................................................................................... 190

2. ROBUST OPTIMIZATION FRAMEWORK ............................................................ 192

3. APPLICATION TO WATER SUPPLY SYSTEM .................................................... 199

3.1 Problem Statement ................................................................................................ 199

3.2 Objective Function ................................................................................................ 202

3.3 Simple Decision Bounds ....................................................................................... 204

 8

3.4 Flow Constraints through Nodes .......................................................................... 205

3.5 Flow Constraints through Arcs ............................................................................. 207

3.6 Data Uncertainty and Robust Formulation ........................................................... 209

3.7 Probability Bounds................................................................................................ 215

4. RESULTS AND DISCUSSION ................................................................................. 216

5. CONCLUSIONS......................................................................................................... 220

6. NOMENCLATURE ................................................................................................... 223

7. REFERENCES ........................................................................................................... 226

8. TABLES ..................................................................................................................... 229

9. FIGURES .................................................................................................................... 236

 9

ABSTRACT

A water supply system collects, treats, stores, and distributes water among water

sources and consumers. Increasing population, diminishing supplies and variable climatic

conditions can cause difficulties in meeting water demands; especially in arid and semi-

arid regions where water resources are limited. Given the system complexity and the

interactions among users and supplies, a large-scale water supply management model can

be useful for decision makers to plan water management strategies to cope with future

water demand changes. When this long range water supply plan is developed, accuracy

and reliability are the two most important factors. To develop an accurate model, as much

information as possible on the system has to be considered. As a result, the water supply

system has become more complicated and comprehensive structures. Stochastic search

techniques thus have evolved to find the most accurate solution for the future water

supply plan. Future uncertainty also has been considered to improve system reliability as

well as economic feasibility. This suite of tools can be also useful in deriving consensus

among competing water needs for proposed long-term water supply plans.

In this study, a general large-scale water supply system that is comprised of modular

components including water sources, users, recharge facilities, and water and wastewater

treatment plants was developed in a dynamic simulation environment that helps users

easily understand the model structure. The model was applied to a realistic hypothetical

system and simulated several possible 20-year planning scenarios. In addition to water

balances and water quality analyses, construction and operation and maintenance of

system components costs were estimated for each scenario. One set of results
 10

demonstrates that construction of small-cluster decentralized wastewater treatment

systems could be more economical than a centralized plant when communities are

spatially scattered or located in steep areas where pumping costs may be prohibitive.

The Shuffled Frog Leaping Algorithm (SFLA), then, was used to minimize the total

system cost of the general water supply system. Sizing decisions of system components’

capacities – pipe diameter, pump design capacity and head, canal capacity, and water and

wastewater treatment capabilities – are decision variables with flow allocations over the

water supply network to meet water demands. An explicit representation of energy

consumption cost for the operation of satellite wastewater treatment facilities was

incorporated into the system in the optimization process of overall system cost. Although

the study water supply systems included highly nonlinear terms in the objective function

and constraints and several hundred decisions variables, a stochastic search algorithm

was applied successfully to find optimal solutions that satisfied all the constraints for the

study networks.

An accurate water supply plan is achieved. However, the system reliability is not

assured. A robust optimization approach, hence, was introduced into the design process

of a water supply system as a framework to consider uncertainties of the correlated future

data by applying a new robust optimization approach. The approach allows for the

control of the degree of conservatism which is a crucial factor for the system reliabilities

and economical feasibilities. The system stability is guaranteed under the most uncertain

condition. It was found that the water supply system with uncertainty can be a useful tool

to assist decision makers to develop future water supply schemes.

 11

1. INTRODUCTION

1.1 Problem Statement

Providing sufficient water of appropriate quality and quantity has been one of the

most important issues in human history. Most ancient civilizations were initiated near

water sources. As populations grew, the challenge to meet user demands also increased.

People began to transport water from other locations to their communities. For example,

the Romans constructed aqueducts to deliver water from distant sources to their

communities.

Today, a water supply system consists of infrastructure that collects, treats, stores,

and distributes water between water sources and consumers. Limited new natural water

sources, especially in the southwest region of the USA, and rapidly increasing population

has led to the need for innovative methods to manage a water supply system. For

example, reclaimed water has become an essential water resource for potable and non-

potable uses. Structural system additions including new conveyance systems and

treatment and recharge facilities and operation decisions, such as allocating flow and

implementing conservation practices, are made with the present and future demands in

minds. As additional components and linkages between sources and users are developed,

the complexity of the water supply system and the difficulty in understanding how the

system will react to changes grows. The inherent uncertainty in climate and water

demands and supplies further raises the difficulty in interpreting the system. These

concerns raise the need for generalized design tools for decision makers and the public to
 12

plan structural changes and manage the water supply system to adapt to future water

demands and supplies. Such tools can be simulation or optimization models and may

directly or indirectly account for system uncertainties.

1.2 Literature Review

Many efforts on the development of a water supply system have been made through

for sustainable water supply. However, the complexity of system limited the site specific

application at the first era. As water demands pressures raise increasingly on the existing

water supply system, many studies attempted to develop a general water supply system to

assist decision makers to design more reliable systems for a long range operation period.

These attempts also include the optimization of total system construction and operation

cost. Under given situations such as limited computational technology, complicatedness

and uncertainty in water supply systems, the ultimate goal of these studies are to supply

water sources to users reliably in a more sustainable and timely manner as a long-term

plan.

1.2.1 Water Supply System Design

Computer-based models together with their interactive interfaces are typically called

as decision support systems (DSS) (Loucks 1995). Despite the limitation of software and

hardware of technology in 1970s and 1980s, many site-specific river basin models were

developed and used by engineers in water management organizations for an operational

planning of their basins (Zagona et al. 2001) such as the U.S. Bureau of Reclamation’s
 13

(USBR), Colorado River Simulation System (CRSS), Tennessee Valley Authority’s

(TVA) Daily Scheduling Model, and the Potomac River Interactive Simulation Model

(PRISM). CRSS representative of reservoir system models has been used to establish

complicated operating policies to balance end-of-water-year storage in Lakes Powell and

Mead (USBR, 1987). PRISM was originally developed and implemented to a regional

water supply system for the Washington metropolitan area (Palmer et al. 1980).

To overcome the deficiencies of hard-wired models, several well-supported, general

river and reservoir models such as HEC-5 (Zagona et al. 2001) and HEC-3 (Wurbs 1993)

have been developed to apply policy options that modelers can parameterize and/or

prioritize to represent the operations for a specific system. HEC ResSim (US Army 2003)

and its predecessor, HEC-5, are the two of the most widely used and well documented

reservoir-system simulation models for the operation simulation of a reservoirs system in

a river network for flood control, water supply, hydropower, and instream flow

maintenance for water quality (Wurbs 1993 and Mays and Tung 1992). The HEC-3

Reservoir System Analysis for Conservation program is much simpler than HEC-5 and

does not have the component for the comprehensive flood-control (Wurbs 1993).

Generalized mathematical water management models were also developed by many

researchers. Ocanas and Mays (1981a), Huang and Loucks (2000), and Yang et al. (2000)

applied their reservoir/river management model to hypothetical river networks and

evaluated total cost and benefit of the network. Other applications have a specific

application network such as South Taiwan (Yen and Chen 2001), the Aral Sea basin of

central Asia (Cai et al. 2003), the Rio Grande river from Elephant Butte, New Mexico to
 14

Fort Quitman, Texas (Ejeta et al. 2004), Syr Darya River basin (Cai et al. 2001), and

water supply system in southern Israel (Cohen et al. 2004). However, these models could

not be generalized because they did not consider all possible components and thus the

generality and flexibility were not sufficient to allow end-users to easily modify the

model. RiverWare (Biddle 2001, Magee and Goranflo 2002, Gilmore et al. 2000, Fulp

and Harkins 2001) is a general object-oriented model but limited only to reservoir

management.

All previous works mentioned above generally did not incorporate water quality

parameters into the models. Exceptions are the studies conducted by Ocanas and Mays

(1981a and 1981b) that considered biochemical oxygen demand (BOD) and total

suspended solid (TSS) as the water quality parameters. Cai et al. (2001 and 2003) and

Cohen et al. (2004) modeled salinity and Ejeta et al. (2004) incorporated a total dissolved

solid (TDS) component.

Recently, more generalized and object-oriented system dynamics simulation models

have been developed. As one of the earliest applications, Palmer et al. (1993) tailored a

dynamic simulation software application to represent the Portland water supply system.

Other applications include river basin planning (Palmer et al. 2000), long-term water

resource planning and policy analysis (Simonovic et al. 1997; Simonovic and Fahmy

1999), reservoir operation (Ahmad and Simonovic 2000), sustainability of a water

resource system, and water supply planning and management (Nandalal and Simonovic

2003). System dynamics modeling was also used to model sea level rise in a coastal area

by Ruth and Pieper (1994).

 15

Simonovic and Bender (1996) applied a dynamic simulation approach to a collaborative

planning-support system by adding environmental issues, e.g. fish habitat, to

hydroelectric power generation. Stave (2003) developed a system dynamics model for the

Las Vegas water supply system to increase public understanding about the value of water

conservation. Passell et al. (2002) presented a computerized dynamic model to simulate

the hydrology, ecology, demography and economy in the Middle Rio Grande Basin using

the commercially-available application, Powersim Studio. Water sustainability and

groundwater storage in San Pedro River Basin, AZ was evaluated by Sumer et al. (2004).

1.2.2 Deterministic Optimization

Little research has been conducted in optimizing water supply system planning.

Ocanas and Mays (1981a and b) formulated and solved a water reuse planning

optimization model using non-linear programming under steady and dynamic conditions.

The steady state model consisted of a nonlinear objective function and linear and

nonlinear constraints for a single period. A large-scale generalized reduced gradient

technique was used to solve the optimization problem (Ocanas and Mays 1981a). In the

follow-up paper, a similar large-scale generalized reduced gradient technique and

successive linear programming methods were applied to a dynamic water reuse planning

model for single-period and multi-period models (Ocanas and Mays 1981b). Water

quality was considered in both studies. In the dynamic model, the capacity expansion of

treatment facilities was considered at the beginning of the period and operation costs

were calculated over the period. Their model provided a basis on the optimization
 16

structure for water supply management systems. Conveyance systems were considered as

lumped units without detailed representations of energy loss and capacity.

Ejeta et al. (2004) applied a general approach to the Rio Grande in New Mexico and

Texas including a total dissolved solid (TDS) constraint. The objective here was to

maximize total net benefit using Generalized Reduced Gradient technique interfacing the

Simulated Annealing Algorithm.

1.2.3 Stochastic Optimization

Stochastic optimization methods have been applied in the water supply system design

to deal with supply and demand uncertainties. Most work has focused on two-stage

and/or multistage linear or nonlinear stochastic programming. The objectives in design

and/or operation of water supply system were to minimize total cost for the water

transfers to spot-markets (Lund and Israel 1995), to develop for long-term and short-term

management options (Wilchfort and Lund 1997), to manage supply capacity and develop

operation protocols for water shortage management (Jenkins and Lund 2000), and to

develop design criteria for the future operation or the system responses to the design

(Elshorbagy et al. 1997).

The above applications have optimized the system with respect to the expected values

of the objective function but did not consider the aspects regarding either risk-averse

behavior or trade-offs between sub-optimality and infeasibility. Despite the consideration

of constraint penalty functions in these expected valued optimizations, decisions that

hedge against risk were considered and addressed in a few researches. Fiering and
 17

Matalas (1990) posed water supply planning robustness with respect to global climate

change, and Watkins and McKinney (1997) reformulated a two-stage stochastic model

developed by Lund and Israel (1995) embedded in the robust optimization framework

(Mulvey et al. 1995) by including the standard deviation of cost over all scenarios,

representing the risk and showed decreasing risk of incurring extremely high costs in the

event of a severe drought. However, this robust formulation is another way of the

unconstrained optimization problem minimizing the standard deviation of cost as much

as possible.

Bertsimas and Sim (2004) presented a framework to find robust solutions that are not

affected by data uncertainty. In this dissertation, their methodology is applied in a water

supply system. This paradigm in robust optimization was introduced by Soyster (1973)

and, however, the solutions are too conservative in a sense that much of optimality is lost

for the system robustness. A conservative design usually leads to a high-cost, which

might not be desirable in practice. Ben-Tal and Nemirovski (1999 and 2000), El-Ghaoui

and Lebret (1997), and El-Ghaoui et al. (1998) considered uncertain linear problems with

ellipsoidal uncertainties, and proposed a new approach for robust optimization in order to

overcome the conservatism. In order to control the conservatism, these approaches

introduced nonlinear problems into the system, which are computationally intractable.

This difficulty motivated Bertsimas and Sim (2004) to suggest another approach for

robust optimization. Their approach retains linearity of Soyster (1973) and allows for the

control of the degree of conservatism at the same time.

 18

1.3 Summary of Literature

Although dynamic modeling and computer technology has advanced to the point

where comprehensive water supply systems can be realistically represented, simulation

studies to date have been limited to site-specific applications. Few studies have

considered and none have explicitly evaluated energy consumption and cost. Limitations

in simulation are also inherent in deterministic optimization models. Published models

have not fully integrated water quality, energy use and costs into a comprehensive

planning and operation model.

Finally, although uncertainties have been examined in various water supply related

problems, most of these studies adopted two-stage and/or multistage linear or nonlinear

stochastic programming and focused on the optimization of expected values of the

objective function. No studies attempted to apply robust optimization methods in order to

deal with the risk of system failure due to uncertainty.

Thus, all aspects of simulation and optimization have significant shortcomings that

affect the quality and usability of the resulting solutions. This dissertation attempts to

begin to fill the gaps in the state of modeling and optimization of water supply system

using a combination of existing and new technologies and methods.

 19

2. PRESENT STUDY

2.1 Dissertation Outline

The goal of this dissertation is to develop a suite of tools to assist water policy makers

better understand the impacts of alternative water policies and plan for a long-term

management of general large-scale water supply systems. Since long-term water supply

planning includes comprehensive present and future predicted data, hydraulic and

hydrology information, accurate and uncomplicated water supply system is needed to be

developed. Modules for composing a water supply system, thus, were developed,

expected to be tailored and replicated by users. Water sources, users, and treatment

facilities have a different module developed by dynamic simulation software can be

applied to arbitrary locations by assigning area specific data. Multiple sources, users, and

treatment facilities can be combined by duplicating a module. Input parameters are

defined explicitly and users could interchange parameters and create a water supply

system network. Energy and cost calculations were embedded in the modules by

literatures’ equation. If results are verified or modules are needed to be changed, internal

equations can be researched without difficulty as dynamic simulation model have

transparent structure. Various long-term supply plans, for example, water saving

measures and different source conditions and geometry are defined and implemented

using multi-users, sources, and treatment plants.

Water supply system plans, however, is needed to be developed rather than merely

tested. Large numbers of decisions have to be made before developing a water supply
 20

strategy, for instance, the size of transporting and treatment facilities and flow

distributions. To build up a better strategy, an optimization technique is needed.

Decisions considered are the development and expansion of system infrastructure,

expansion and operation strategies. Like simulation model, hydraulic energy relationships

through transporting systems are embedded. Couple hundred variables and nonlinear

objective function and constraints cause difficulties to apply linear or non-linear

programming. Random search technique, Shuffled Frog Leaping Algorithm (SFLA) is

applied to solve different scales of water supply system. Despite of the huge size of water

supply system, SFLA solve successfully and find an optimal solution. Consideration of

uncertainty, however, is raised as an issue to be concerned.

To consider uncertain factors, stochastic formulation is needed. Water supply system

is pre-processed to simplify nonlinear equations. As a method of stochastic optimization,

robust technique was adopted and GAMS/BARON is used to solve the water supply

system. Robust optimization ensures the system’s safety in the worst case.

GAMS/BARON is a commercial solver for global optimization using branch-and-bound

method. A new approach on robust optimization is applied in this study. The new

approach employs the degree of conservatism which is a level of consideration of

randomness to control solution’s conservatism. This approach is useful because,

sometimes, economical benefit is sacrificed to robustness of the system. Different degree

of conservatisms is applied to investigate total cost rising. From the total cost of change,

the appropriate level of the degree of conservatism could be suggested. The following

section outlines the dissertation goals and general approaches.

 21

The three goals of this dissertation are (1) to develop a set of modules representing

various water uses and treatment options that can be easily combined to describe a

general water supply system (Appendix A), (2) to formulate and solve a deterministic

optimization model that minimizes the cost for construction and operation of a system

that consists of water transport facilities such as pipes, pumps, and canals as the set of

decision variables using explicit representation of energy consumption costs and

evaluating the tradeoff of multiple satellite wastewater treatment facilities (Appendix B)

and (3) to consider uncertainty in future conditions while optimizing a simplified system

to provide decision makers and understanding of the tradeoff between cost and risk

(Appendix C).

A General Water Resources Planning Model using Dynamic Simulation: Evaluation

of Decentralized Treatment (Appendix A)

A general tool to evaluate water supply systems is developed within a dynamic

simulation framework. This model can assist decision makers in planning long range

strategies for the system management by evaluating water quality, possible future water

supply condition, energy conservation and system hydraulics. The model is written in

modular form with sub-models developed for various system components. These

modules can easily be linked to construct a model for a general system without

developing each component from basic relationships.

To demonstrate how the model can be used, three scenarios are considered on a

hypothetical system for a twenty years planning period (2000-2020). The first scenario
 22

examines the impact of conservation measures aimed at reducing domestic and industrial

demands. The second condition represents the inability to secure additional external

water sources and its impact on long-term system storage. Finally, the cost effectiveness

and impact on water quality of decentralized treatment is analyzed for a disperse supply

system that covers a range of topography.

The hypothetical system is comprised of five sources (precipitation, imported water,

uncontrolled river, regulated river with reservoir, groundwater), twelve users (four

domestic areas, an industrial area, four agricultural areas, two large outdoor turf areas, an

environmental and recreational area), five treatment systems (two water treatment and

three wastewater treatment systems), and a recharge facility.

Piping is main transporting means, but open channel gravity flow transports untreated

water when permitted to reduce cost. However, pumping through pipelines is required in

some cases such as extracting agricultural irrigation water from groundwater. Water and

wastewater treatment plant can be chosen to the lumped or conventional systems.

Lumped treatment plant has representative removal efficiency for a total treatment

system, while conventional system has unit processes having different removal

efficiencies and capacities, then total efficiency is calculated by summing them up.

Conservation practices are means to stretch current water supplies. Population growth

controls may also be employed to limit demand increases and to avoid constructing new

infrastructure and/or water supplies. Here, a scenario is posed in which a community

desires to examine the impact of these types of measures on the hypothetical water supply

system.
 23

Incentives and ordinance programs are defined. Since the amount of savings depends

upon the age of the existing fixtures/appliance, to reduce indoor use, incentives for (1)

faucet, shower head, and toilet replacement with more efficient fixtures and (2) front

loading clothes washer purchases can be provided. Generally outdoor use reductions are

made by ordinance for new homes. For example, new homes may be required to have

grey water reuse system in which water from bathtubs, faucets, clothes washers, and

showers is collected for outdoor purposes. Grey water retrofit construction costs for

existing homes are prohibitive. All of the above measures will reduce demand for

pumping but will also reduce the amount of reclaimed water available for large area

irrigation. Other ordinances can include prohibiting fountains and evaporative coolers in

new homes and requiring existing homes to remove fountains.

Swimming pools are a major consumptive use. In the climate conditions of the

hypothetical community, the average evaporation loss from a swimming pool is about 5

m/yr (229 in/year). Evaporation losses can be reduced by one or more set of measures.

An incentive program could be introduced for pool cover purchases that would be used in

the off-season. Other ordinances for reducing pool use are reducing pool draining

frequencies, restricting future swimming pools construction, filling existing swimming

pools, or reusing swimming pool drain or back-flush water as grey water.

Other water saving programs that can be evaluated are briefly described below.

During the growing season, outdoor water use restrictions prohibit irrigation on a

particular day or time by ordinance to reduce evaporation volumes. An irrigation

efficiency program requires high efficient irrigation system. Landscaping standards

 24

require turf irrigation systems to be replaced by drip irrigation or xeriscaping. Rainfall

harvesting can replace treated water usage to reduce outdoor water use in new or existing

homes. Lastly, a water wasting ordinance imposes fines for wasting or unreasonable use

of water. These losses are assumed to be zero after the ordinance implement.

Similar conservative measures, such as toilet retrofit, improvement of water drip

irrigation efficiency, outdoor water use restriction, water loss due to violations and

swimming pool savings, can be applied to the industrial sector. Only purchasing pool

covers is supported with a government incentive. All other programs are implemented by

ordinance. In addition, water audits can be completed on these large users.

Conservation measures is implemented one at a time in the middle of simulation and

total cost and water use saving are investigated. When a conservation measure is

implemented, facility construction cost remains the same, while operation cost declines

with respect to water use decrease. Operational cost for pumping and piping is reduced as

the amount of water conveyance decrease. Some conservation measures decrease water

use by reducing demand through high efficient fixtures. Water and wastewater treatment

operational cost also decreases due to the water demand reduction. The other

conservation programs decrease fresh water usage by using reclaimed water through grey

water recycling, rainfall harvesting, or pool discharges, while water demands remain the

same.

Some conservation measures, such as reusing pool discharges and restricting outdoor

water use do not have perceptible benefit on water and cost saving. This is because the

amount of pool discharges to wastewater treatment systems is small. Restricting outdoor

 25

water use does not change the volume of water applied since turf water demands are not

adjusted although the irrigation efficiency is improved. Industrial conservation has

similar, but smaller effects since most of the industrial water use is used for indoor

purposes.

When all conservation measures including the installation of rainwater harvesting

system on water supply are simultaneously implemented, the water use in the domestic

areas decreased by more than 70%, however total system cost only decreases by 2.6%.

This decrease is not significant to a long-term water management plan. However, because

the primary goal of the plan is ordinarily safe-yield, the positive effect on groundwater

storage makes the conservation measures meaningful application.

As water resources become more stressed and climatic variability increases, the

potential for reduced supply is more likely. In addition, environmental systems may also

require additional water. In this scenario, it is assumed that the imported water source is

no longer available and that the existing use of reclaimed water as the riparian zone’s

water source precludes its use for other purposes. Serious repercussions are anticipated in

the water balance and groundwater mining must occur.

Groundwater is used to replace the unavailable supplies for three cases: without

imported water, without reclaimed water, and without imported and reclaimed waters.

Nearly 19% of the groundwater storage is depleted when imported and reclaimed waters

were not available demonstrating the need in many communities for these supplies in

order to maintain a sustainable system.

 26

Decentralized wastewater treatment has become a topic of interest as a cost effective

means to treat and recycle effluent. For small communities, unsewered communities, and

communities covering a range of elevations, cluster wastewater treatment system may be

an appropriate treatment option. Cluster wastewater treatment has also been called

community-wide decentralized wastewater management (Lombardo Associates, Inc.

2004).

Here, we consider a distributed wastewater scheme in which multiple satellite water

and wastewater treatment plants (WTPs and WWTPs) are located throughout a

community with the ability to treat and distribute reclaimed water to nearby users.

Economies of scale suggest that, under many conditions, a single large WWTP would be

less expensive than multiple plants. However, when the community covers a range of

elevations and/or a long distance, pumping and piping costs for reclaimed water may be

more expensive than construction cost of multiple wastewater treatment plants.

The efficacy of constructing up to two water treatment systems and three wastewater

treatment systems in the hypothetical system is examined in this application. Costs are

computed for treatment plant construction and operation, pumping and piping for water

transfers, system expansion, operation and maintenance of treatment system and energy

needed for pumping. Pipes between sources, treatment systems, and users were assumed

to be laid to cover the least distance and elevation and determined by a trial and error and

engineering judgment.

The optimal treatment distribution in this example for both lumped and conventional

cases was 2 WTPs and 3 WWTPs which make reclaimed water users be in the vicinity of
 27

the WWTP. Since the transportation cost is dominated by piping/pumping, a remote

system is much more expensive.

The overriding factor in the cost effectiveness of multiple treatment plants is the cost

for transporting water through the system. These costs are determined by the distance and

elevation changes in the region. Therefore, construction of two water treatment plants and

three wastewater treatment plants is the least expensive treatment system for the

community.

Differences in water quality are examined between plant alternatives at key locations

and within the system for one plant configuration. Concentration of four water quality

indicators (BOD, TSS, hardness, and Giardia) remains under the minimum quality

requirement in sources during the simulation. Influent qualities to users are satisfied by

the minimum usable quality.

Application of the Shuffled Frog Leaping Algorithm for the Optimization of a

General Large-scale Water Supply System (Appendix B)

Developing an optimal strategy is difficult, if not impossible, to determine for a

complex water supply system using simulation alone. The second part of this dissertation

(Appendix B) develops a deterministic optimization model for a generic water supply

system. Given the nonlinear, discrete and discontinuous nature of the problem, a

stochastic search algorithm is used to minimize the costs for new transport and treatment

facility construction and for operating and maintaining the overall system.
 28

The optimization problem has been formulated for the water supply systems for 20-

year planning periods. New structural component construction is permitted at the outset

(year 1) and new components or existing component expansion may be added after 10

years. Biochemical Oxygen Demand (BOD) is used as the representative water quality

parameter.

The first system to be optimized consist single water and wastewater plants, multiple

sources (imported water, groundwater aquifer, and surface water) and two demands

centers (domestic and agricultural). Three types of water transport structures are used

depending upon the connection: canal, pipe and/or pump. All canal flows for the

conveyance of imported and raw water sources are driven by gravity. Agricultural areas

and water treatment plant may directly pump groundwater from aquifers available nearby

so do not require a pipe link. Other flows are transported through pipes that may require a

pump station to supply the energy necessary to pass flow through the pipeline and satisfy

the minimum pressure head requirement at the outlet. Groundwater replenishment

through recharge basins and seepage losses from users is assumed.

The network consists of six canal depth construction decisions (6) and seven

pump/pipeline arcs with their three design decisions (pipe diameter and pump flow

capacity and head) for a total of 21 decisions. The network also includes two pump links

with decision pump design flow and head (four total design decisions) and two treatment

facilities with plant capacity decisions (total of 2 decisions). Thus, the total number of 33

design decision variables result for each of the two planning periods (total 66).

Independent control decision variables (10) distribute water through the network.
 29

Therefore, the final optimization problem contains a total 86 of decision variables for the

two design periods (66 structural and 20 control decision variables).

Stochastic random search technique, SFLA, is used to solve the deterministic water

supply system. After 5 minutes of calculation time, total construction and operation cost

for the single treatment plant system for the 20-year period is $771 million (present value

for year 0) or an annual cost of $47 million.

The optimal component design and the optimal network solution are found by SFLA.

Since the study system is influenced by only population growth, most system component

sizes do not change over the planning period; owing to economies of scale over the

delayed cost of expansion. Special mechanisms between increasing population, water

demand, water source availability, transportation capacity, and water quality requirement

caused complexity and the algorithm found the optimal water supply plant satisfying all

constraints.

For expansion of analysis, multiple water users and wastewater treatment plants are

included in the multiple wastewater treatment plant system in order to investigate a more

generalized system. This network is greater than the Single wastewater treatment plant

system and consists of six users - three domestic areas, one industrial, one agricultural,

and one large outdoor area – and three wastewater treatment plants. In general, input

parameters used for the multiple wastewater treatment plant system are the same as for

the Single wastewater treatment plant system except for the initial population at the

domestic areas. The design variables include 6 canals depths, 29 pipe sizes (29

parameters for each pipe diameter, pump design capacity, and pump head), 2 pump
 30

capacities (pump design capacity and head), and 4 treatment plant capacities for each of

the two planning periods. Total structural design variables are 101 (= 6 + 3 × 29 + 2 × 2 +

4) for each design period. This network has 44 arc connections and flow allocations

through twenty-three arcs out of forty four are defined as operation decision variables

while the remaining twenty-one arcs are dependent variables that are computed by mass

balance equations. A total of 248 decision variables (2 design periods × 124 decision

variables (101 and 23 for structural design and operation variables, respectively) are

established for the multiple wastewater treatment plant system application.

The SFLA optimization process of the multiple wastewater treatment plant system

takes 70 minutes and nearly 582 thousands function evaluations with a Dell Inspiron

computer system with Centrino Duo T2300 1.6GHz CPU and 1GB of RAM. The optimal

cost for the system was $837 million as the present value in the starting year of the

planning period and the estimated annual cost was $51 million. Although the optimal

solution found may not be the global optimal solution due to high discrete nonconvexity

associated with the study system, the optimization process demonstrates the improvement

in overall system cost and reduction in the penalty term.

Since existing surface and subsurface water sources within the system are enough to

meet water user demands, no external water is purchased. Domestic areas are supplied

from groundwater which has clear enough quality as a drinking water, and industrial and

agricultural area are mainly supplied from upstream river through water treatment plant.

Reclaimed water is used for large outdoor area. Like the single treatment plant, a few
 31

expansions are needed for the multiple wastewater treatment plant system in the second

design period because of economies of scale.

Reliable Water Supply Network Design under Uncertainty (Appendix C)

Water supply plans are based upon forecasted demands and supplies. Deterministic

optimization, although valuable, does not consider the impact of the uncertainty in

forecasts. The third paper in this dissertation applies a new approach for robust solution

(Bertsimas and Sim 2004) in attempt to understand the tradeoff between cost and the

degree of conservatism. This approach controls the degree of conservatism through a

straightforward parameter. In practical application, the most conservative solution taking

into account the most extreme case is not usually applicable because of its high cost.

Through multiple deterministic optimization solutions the tradeoff between the

conservatism and cost is investigated.

The robust optimization method is applied to minimize the total cost of construction,

expansion, and operations and maintenance of a hypothetical water supply system. The

system includes subsurface (aquifer), surface, and imported water sources, domestic and

agricultural irrigational users, and water and wastewater treatment plants. Unlike

previous applications, a 15 year planning period that is divided into two design periods

and 10 operation periods is considered.

Water supply system has infrastructures before applying for new and optimized

facilities. Groundwater is the main source to supply water demand of domestic and

agricultural areas at year 0. As the result, groundwater is depleted and water

 32

sustainability becomes an issue in the hypothetical community. Groundwater storage at

year 0 is 9.50 km3, which is below the requirement for sustainable water source. The

optimized water supply plan for the next fifteen years would be able to recover

groundwater storage up to 9.93 km3.

Another issue is that, particularly in semi-arid regions, surface water is insufficient to

sustain environment and subsurface water source is being depleted as it supply for water

users. Wastewater effluent is often discharged to a normally dry or low flow channel.

Over time, a downstream riparian habitat developed that the effluent continues to sustain.

If communities move to using reclaimed wastewater effluent for nonpotable and,

potentially, potable uses, this water would no longer be released to the environment as it

is today. Thus, communities will face serious water depletion from both surface and

subsurface sources and the decision to maintain environmental flows and sustainable

groundwater storage have to be made.

Since studied water supply network needs new water structures and sources to

preserve environmental water in river stream and aquifer, as an alternative source,

external source (imported water) can be applied.

The water system’s arcs are five canals, four pipelines, and two pump stations. The

associated design decision at each design epoch are the canal depths, d, (5 canals), pipe

diameters, κ, (4 pipelines), pump design discharges, χ, (4 pump/pipelines + 2 pumps),

pump design heads, H, (4 pump/pipelines + 2 pumps), and water and wastewater

treatment plant capacities, w, (2 plants). Thus, the total number of design decision

variables is 46 (23 × 2 design periods). In addition, the flows on 11 independent arcs

 33

must be determined for each of 10 operation periods. Lastly, the number of binary

variables for pipe flowrate, x, (4 pipelines × 10 operational period) and pump flowrate, µ,

{(4 pump/pipelines + 2 pumps) × 10 operational period} is 100. Thus, the optimization

problem includes a total number of 256 decision variables with 100 binary variables. The

continuous mixed-integer nonlinear problem were solved using GAMS/BARON global

optimization solver with the relative termination tolerance of 0.05 (Sahinidis and

Tawarmalani, 2005).

Compared to flowrates when probabilities of violation (P) of constraints are 0.1, the

amount of water purchased at year 1 (7.17 cms) in the nominal problem is smaller than

10.02 cms (when P = 0.1). When probability of violation is 0.1, the solution ensure that

the constraint remains feasible at least 90%. Imported water purchasing of which price is

incredibly high ($0.81/m3) happens only at year 1. Total pumping water from aquifer

when probability of violation is 0.1 is also smaller than one in nominal problem to

preserve sustainable water in aquifer. Both cases use reclaimed water from wastewater

treatment plant as an agricultural purpose. Increased domestic and agricultural demands

lead large amount of imported water and more supply when P = 0.1.

Domestic area demand increases along time by population growth and causes more

supply to domestic area, while agricultural demand decrease after 5th year of operation.

Most design decisions, therefore, are not suggested to expand after 5th year. However,

pump capacity and head in flows ‘to’ and ‘from’ domestic area are expanded in nominal

condition to supply increasing demand. Increasing demand of domestic area is supplied

mostly from upstream river through water treatment plant, which leads to reduce water
 34

transporting to agricultural area from upstream river. Reclaimed water from wastewater

treatment plant is chosen as an alternative to supply agricultural area after 5th operational

period and pump capacity is expanded.

When probability of violation is 0.1, only pump capacity from wastewater treatment

plant to agricultural area is expanded from 5.78 cms to 7.46 cms because of increasing

uncertainty in precipitation. Groundwater storage requirement constraints have increasing

number of uncertain parameters ( J i ) depending on operational period. Uncertainty in

yearly precipitation is generated independently and total uncertainty increases along time,

which inflow to agricultural area from precipitation decrease, thus water requirement of

agricultural area increases along time.

The system is optimized with probability of violation from 0.1 (the most

conservative) to 1.0 (nominal). As conservatism increase, total cost increase as well to

insure system reliability. Total cost increase dramatically between probability of violation

0.7 and 0.5 of which shape is the same as the amount of external water purchased. Water

purchasing cost cause large increasing in total cost.

2.2 Uniqueness of the Study

Results from the three noted papers demonstrate the potential of improving water

supply system design and planning long-term operation. To date, tools to achieve this

goal are lacking. The suite of models developed in this dissertation provides a

comprehensive set of models to analyze and optimize water supply systems.

 35

The following unique contributions have resulted from this work:

(1) Dynamic simulation has been applied in a number of water resources applications

but not to represent a complete of a water supply system. The modular structured

tool is general and can be used to model a general system that includes multiple

sources, users, and transport components and treatment systems and account for

the spatial allocation of system components.

(2) Within the dynamic simulation model, water quality components and

conventional treatment systems are included and can be easily incorporated using

a transparent structure.

(3) Application of the dynamic simulation model demonstrates the potential benefits

of decentralized treatment within a recycle/reuse system

(4) The large-scale deterministic optimization model extendes previous efforts by

incorporating the full system including water users and reuse rather than only the

source and simple user nodes.

(5) The deterministic optimization model also demonstrates the ability of the Shuffled

Frog Leaping Algorithm to solve optimization problems with over than 250

decision variables (albeit without guarantee of finding a global optimum).

(6) The robust optimization method of Bertsimas and Sim (2004) is applied to

incorporate uncertainties in water supply and demand and the cost/risk of system

failure tradeoff is investigated. The new approach is easy to implement and to

interpret results and trade-offs.

 36

2.3 Conclusions and Future Work

Given the complexity of water supply systems, decision makers may have difficulty

in understanding the impacts of water management policies. In this study, a large-scale

generalized water supply simulation model is developed to assist decision makers better

understand the policy impacts. Model components include detailed domestic usage,

industrial, agricultural, and environmental water demands, multiple supplies, conveyance

system from sources to users, surface and groundwater storage, and conservation

practices. Each model component is modularized to facilitate the transplantation of its

component into another system. Water quality and energy loss through the conveyance

system are also considered in the developed system. The developed system is applied to

hypothetical water communities for the system evaluation.

Scenarios include investigating the effect of various conservation measures, assessing

the impact of the unavailability of a water source, and evaluating the effect of the spatial

distribution of the system components. The total cost for the system expansion, operation

and maintenance and water source sustainability are investigated. Conservation measures

and supplemental water source reduce total operation cost of water supply system.

Conservation measures stretch current water use further. Additional imported and/or

reclaimed water have the benefit of fresh water usage. Decentralized water and

wastewater treatment plants are suggested as an economical approach in the spatially

distributed water community. The results showed that the distance and elevation changes

between demand centers in the example system are such that multiple distributed

treatment facilities are more cost effective than single centralized plants.
 37

To optimize system decisions, two approaches, deterministic heuristic algorithm

(Shuffled Frog Leaping Algorithm, SFLA) and stochastic robust optimization technique

are examined. The former approach allowed us to deal with highly nonlinear and discrete

terms in the objective and constraint functions. Single and multiple wastewater treatment

systems were established and their total system costs were evaluated in order to

investigate the system applicability to an arbitrary network. The optimized solutions

satisfy all the constraints including water quality, pressure, and demand requirements.

The latter robust approach is adopted to take into account the uncertainty factors for

the system optimization. Uncertainties in water demands and availability, and their

correlation with precipitation are considered as stochastic parameters. The degree of

conservatism is introduced to examine the tradeoff between the system reliability and

economic feasibility. Probability bound which means the probability of violation of a

constraints is introduced as a way to present the degree of conservatism. Overall total

system cost increases with the degree of conservatism. Since infrastructure exit in year 0,

construction cost of treatment and transportation facilities does not have apperent effect

in total operation cost, while the cost of purchasing external water source to supply

insufficient internal sources cause total cost increased. This result could be a useful tool

to support decision makers.

As computational technology continues to improve, the developed system can be

extended to include more detailed process models to more realistically simulate a water

supply system. Further research efforts are needed to collect data associated with water
 38

supply system for the system validation to increase the reliability and accuracy of the

system representation.

1) application to a real system

2) alternative deterministic optimization schemes to confirm SFLA result

3) extension of deterministic optimization to include water and wastewater treatment

plants relationships

4) additional water quality parameters to consider quality constraints on water

supply system

5) Extension of stochastic optimization to include water and wastewater treatment

plants relationships

6) Consideration of temporal correlation in stochastic optimization

7) Additional pre-processes such as linear relaxation for stochastic optimization

 39

REFERENCES

Ahmad, S., and Simonovic, S. P. (2000). “System dynamics modeling of reservoir

operations for flood management.” Journal of Computing in Civil

Engineering, 14(3), 190-198.

Bertsimas, D. and Sim, M. (2004). “The price of robustness.” Operation Research, 52(1),

35 – 53.

Biddle, S. H. (2001). “Optimizing the TVA reservoir system using RiverWare, bridging

the gap: meeting the world’s water and environmental resources challenges.”

Proceedings of the World Water and Environmental Resources Congress,

ASCE.

Branson, F. A., Gifford, G. F., Renard, K. G., and Hadley, R. F. (1981). Rangeland

hydrology, range science series 1. Society for Range Management, Denver,

CO.

Cai, X., Mckinney, D. C., and Lasdon, L. S. (2001). “Piece-by-Piece approach to solving

large nonlinear water resources management models.” Journal of Water

Resources Planning and Management, 127(6), 363-368.

Cai, X., McKinney, D. C., and Lasdon, L. S (2003). “Integrated hydrologic-agronomic-

economic model for river basin management.” Journal of Water Resources

Planning and Management, 129(1), 4-16.

 40

Cohen, D., Shamir, U., and Sinai, G. (2004). “Sensitivity analysis of optimal operation of

irrigation supply systems with water quality considerations.” Irrigation and

Drainage Systems, 18, 227-253.

Ejeta, M. Z., McGuckin, T., and Mays, L. W. (2004). “Market exchange impact on water

supply planning with water quality.” Journal of Water Resources Planning

and Management, 130(6), 439-449.

Elshorbagy, W., Yakowitz, D., and Lansey, K. (1997) “Design of engineering systems

using a stochastic decomposition approach.” Engineering Optimization, 27(4),

279-302.

Fiering, M. B. and Matalas, N. C. (1990). Decision making under uncertainty, climate

change and U.S. water resources, P. E. Waggoner, ed., John Wiley & Sons,

Inc., New York, N.Y.

Fulp, T. and Harkins, J. (2001). “Policy analysis using RiverWare: Colorado river interim

surplus guidelines.” Proceedings of ASCE World Water & Environmental

Resource Congress, Orlando, FL.

Gilmore, A., Magee, T., Fulp, T., and Strezepek, K. (2000). “Multi-objective

optimization of the Colorado river.” Proceedings of the ASCE 2000 Joint

Conference on Water Resources Engineering and Water Resources Planning

and Movement, Minneapolis, MN.

Huang, G. H. and Loucks, D. P. (2000). “An inexact two-stage stochastic programming

model for water resources model for water resources management under

uncertainty.” Civil Engineering and Environmental Systems, 17(2), 95-118.

 41

Jenkins, M. W. and Lund, J. R. (2000). “Integrating yield and shortage management

under multiple uncertainties.” Journal of Water Resources Planning and

Management, 126(5), 288 – 297.

Lombardo Associates, Inc. (2004). National decentralized water resources capacity

development project – cluster wastewater systems planning handbook,

Newton, Massachusetts.

Loucks, D. P. (1995). “Developing and implementing decision support systems: a

critique and a challenge.” Water Resources Bulletin, 31(4), 571-582.

Lund, G. R. and Israel, M. (1995). “Optimization of transfer in urban water supply

planning.” Journal of Water Resources Planning and Management, 121(1),

41-48.

Magee, T. M. and Goranflo, H., M. (2002). “Optimizing daily reservoir scheduling at

TVA with RiverWare.” Proceedings of the Second Federal Interagency

Hydrologic Modeling Conference, Las Vegas, NV.

Mays, L. W. and Tung, Y. (1992). Hydrosystems Engineering and Management,

McGraw-Hill, New York.

Minnesota Pollution Control Agency. (2000). Wastewater treatment and collection

systems.

Mulvey, J. M., Vanderbei, R. J., and Zenios, S. A. (1995). “Robust optimization of large-

scale systems.” Operations Research, 43(2), 264 – 281.

Nandalal, K. D. W., and Simonovic, S. P. (2003). “Resolving conflicts in water sharing: a

systemic approach.” Water Resources Research, 39(12), 1362-1372.

 42

Ocanas, G. and Mays, L. W. (1981a). “A model for water reuse planning.” Water

Resources Research, 17(1), 25–32.

Ocanas, G. and Mays, L. W. (1981b), “Water reuse planning models: extensions and

applications.” Water Resources Research, 17(5), 1311-1327.

Palmer, R. N., Wright, J. R., Smith, J. A., Cohon, J. L., and ReVelle, C. S. (1980). Policy

analysis of reservoir operation in the Potomac river basin, volume I, executive

summary, Johns Hopkins University, Baltimore, MD.

Palmer, R. N, Keyes, A., and Fisher, S. (1993). “Empowering stakeholders through

simulation in water resources planning.” Proceedings of the ASCE Water

Management in the 90s Conference, Seattle, Washington, 451-454.

Palmer, R. N., Mohammadi, A., Hahn, M. A., Kessler, D., Dvorak, J. V., and Parkinson,

D. (2000). “Computer assisted decision support system for high level

infrastructure master planning: case of the city of Portland Supply and

Transmission Model (STM).” Proceedings of the ASCE 2000 Joint

Conference on Water Resources Engineering and Water Resources Planning

and Movement, Minneapolis, MN.

Passell, H., Tidwell, V. and Webb, E. (2002). “Cooperative modeling: a tool for

community-based water resource management.” Southwest Hydrology, 1(4),

26.

Ruth, M. and Pieper, F. (1994). “Modeling spatial dynamics of sea-level rise in a coastal

area.” System Dynamics Review, 10(4), 375-389.

Sahinidis, N. and Tawarmalani, M. (2005) GAMS/BARON Solver Manual

 43

(http://www.gams.com/dd/docs/solvers/baron.pdf)

Simonovic, S. P. and Bender, M. J. (1996). “Collaborative planning-support system: an

approach for determining evaluation criteria.” Journal of Hydrology, 177,

237-251.

Simonovic, S. P., Fahmy, H., and El-shorbagy, A. (1997). “The use of object-oriented

modeling for water resources planning in Egypt.” Water Resources

Management, 11, 243-261.

Simonovic, S. P. and Fahmy, H. (1999). “A new modeling approach for water resources

policy analysis.” Water Resources Research, 35(1), 295-304.

Soyster, A. L. (1973). “Convex Programming with set-inclusive constraints and

applications to inexact linear programming.” Operations Research, 21, 1154-

1157.

Sumer, D. Y., Lansey, K. and Richter, H. (2004). “Evaluation of conservation measures

in the Upper San Pedro Basin.” The 2004 World Water and Environmental

Resources Congress, Salt Lake City, UT, ASCE

Stave, K. A. (2003). “A system dynamics model to facilitate public understanding of

water management options in Las Vegas, Nevada.” Journal of Environmental

Management, 67, 303-313.

US Army Corps of Engineers (2003). HEC-ResSim User’s Manual, V. 2, September.

(www.hec.usace.army.mil/software/hec-ressim/hecressim-hecressim.htm)

U.S. Bureau of Reclamation (1987). Colorado River Simulation System: Overview,

Denver, Colorado.
 44

Walski, T. M., Brill, E. D., Gessler, J., Goulter, I. C., Jeppson, R. M., Lansey, K., Lee,

H., Liebman, J. C., Mays, L., Morgan, D. R., and Ormsbee, L. (1987). “Battle

of the network models: epilogue.” Journal of Water Resources Planning and

Management, 113(2), 191-203.

Watkins, D. W. and McKinney, D. C. (1997). “Finding robust solutions to water

resources problems.” Journal of Water Resources Planning and Management,

123(1), 49-58.

Wilchfort, G. and Lund, J. R. (1997). “Shortage management modeling for urban water

supply systems.” Journal of Water Resources Planning and Management,

123(4), 250 - 258.

Wurbs, R. A. (1993). “Reservoir system simulation and optimization models.” Journal of

Water Resources Planning and Management, 119(4), 455-472.

Yang, S., Sun, Y., and Yeh, W. W-G. (2000). “Optimization of regional water

distribution system with blending requirements.” Journal of Water Resources

Planning and Management, 126(4), 229-235.

Yen, K. H., and Chen, C. Y. (2001). “Allocation strategy analysis of water resources in

South Taiwan.” Water Resources Management, 15, 283-297.

Zagona, E. A., Fulp, T. J., Shane, R., Magee, T., and Goranflo, H. M. (2001).

“Riverware: a generalized tool for complex reservoir system modeling.”

Journal of the American Water Resources Association, 37(4), 913-929.

 45

APPENDICES
 46

APPENDIX A: A GENERAL WATER RESOURECES

PLANNING MODEL USING DYNAMIC SIMULATION:

EVALUATION OF DECENTRALIZED TREATMENT

A. Graph
 47

A General Water Resources Planning Model using Dynamic Simulation:

Evaluation of Decentralized Treatment

G. Chung1, K. Lansey2, P. Blowers3, P. Brooks4, W. Ela5, S. Stewart6 and P. Wilson7

ABSTRACT

Increasing population, diminishing supplies and variable climatic conditions can cause

difficulties in meeting water demands; especially in arid regions where water resources

are limited. Given the complexity of the system and the interactions among users and

supplies, a large-scale water supply management model can be useful for decision makers

to plan water management strategies to cope with future water demand changes. It can

also assist in deriving agreement between competing water needs and consensus and buy-

in among users of a proposed long-term water supply plans. The objective of this paper is

to present such a general water supply planning tool that is comprised of modular

components including water sources, users, recharge facilities, and water and wastewater

treatment plants. The model was developed in a dynamic simulation environment that

helps users easily understand the model structure.

1
Graduate Student, Department of Civil Engineering and Engineering Mechanics, The University of Arizona, Tucson,
AZ 85721, USA (Tel: 1-520-360-9554, E-mail: gunhui@email.arizona.edu)
2
Professor, Department of Civil Engineering and Engineering Mechanics, The University of Arizona, Tucson, AZ
85721, USA (Tel: 1-520-621-2512, Fax: 1-520-621-2550, E-mail: lansey@engr.arizona.edu)
3
Assistant Professor, Department of Chemical and Environmental Engineering, The University of Arizona, Tucson,
AZ 85721, USA (Tel: 1-520- 626-5319, E-mail: blowers@engr.arizona.edu)
4
Assistant Professor, Department of Hydrology and Water Resources, The University of Arizona, Tucson, AZ 85721,
USA (Tel: 1-520- 621-3424, E-mail: brooks@hwr.arizona.edu)
5
Associate Professor, Department of Chemical and Environmental Engineering, The University of Arizona, Tucson,
AZ 85721, USA (Tel: 1-520- 626-9323, E-mail: wela@email.arizona.edu)
6
Research scientist, Department of Hydrology and Water Resources, The University of Arizona, Tucson, AZ 85721,
USA (Tel: 1-520- 626-3892, E-mail: sstewart@hwr.arizona.edu)
7
Professor, Department of Agricultural and Resource Economics, The University of Arizona, Tucson, AZ 85721, USA
(Tel: 1-520- 621-6258, E-mail: pwilson@ag.arizona.edu)
 48

The model was applied to a realistic hypothetical system and simulated several

possible 20-year planning scenarios. In addition to water balances and water quality

analyses, construction and operation and maintenance of system components costs were

estimated for each scenario. One set of results demonstrates that construction of small-

cluster decentralized wastewater treatment system could be more economical than a

centralized plant when communities are spatially scattered or located at steep areas where

pumping costs may be prohibitive.

 49

1. INTRODUCTION

Increases in water demands have led to the need for innovative supply and demand

management to economically and efficiently operate a system within budget while

meeting user demands. A broad range of concerns resulting from modifying supplies and

demands must be considered in devising a water supply plan. The complexity of the

water supply system, however, makes it problematical to understand the interactions

between components; even for those intimately involved in the planning process. The

complicated system also causes difficulties in educating the public, improving existing

system operations, and finding low cost designs.

Thus, modeling tools that can represent a water supply system and demonstrate the

effects of management decisions can be extremely valuable. Several such tools have been

developed to simulate water supply systems. These models (Ocanas and Mays, 1981a and

1981b; Yen and Chen, 2001; Huang and Loucks, 2000; Cai et al., 2001 and 2003; Ejeta et

al., 2004; Yang et al., 2000; and Cohen et al.; 2004) tend to be system specific and are

generally inflexible in easily adapting to other systems and do not have user-friendly

interfaces.

This paper presents an integrated object-oriented dynamic simulation approach to

develop water supply system model that can be applied to design a long range plan. The

generality allows systems composed of multiple sources, users, and transportation and

treatment systems to be relatively easily organized for specific locations. The dynamic

simulation approach allows users and the general public to look inside the model and
 50

understand the relationships that comprise the model. This open architecture is

particularly useful in situations where several conflicting goals are to be addressed. In

addition to water balances, the model can also track water quality and costs over the

planning period duration.

2. LITERATURE REVIEW/BACKGROUND

Computer-based models together with their interactive interfaces are typically called

decision support systems (DSS) (Loucks, 1995). Despite software and hardware

limitation during the 1970s and 1980s, many site-specific river basin models were

developed and used by engineers in water management organizations for operational

planning of their basins (Zagona et al. 2001) such as the U.S. Bureau of Reclamation’s

(USBR) Colorado River Simulation System (CRSS), Tennessee Valley Authority’s

(TVA) Daily Scheduling Model, and the Potomac River Interactive Simulation Model

(PRISM). CRSS is representative of reservoir system models and captures a complicated

set of operating policies that balance end-of-water-year storage in Lakes Powell and

Mead (USBR 1987). PRISM was originally developed and implemented for a regional

water supply system for the Washington metropolitan area (Palmer et al. 1980).

To overcome the deficiencies of hard-wired models, several well-supported, general

river and reservoir modeling tools such as HEC-5 (Zagona et al. 2001) and HEC-3

(Wurbs 1993) have been developed that apply policy options that modelers can

parameterize and/or prioritize to represent the operations for a specific system. HEC
 51

ResSim (US Army 2003) and its predecessor, HEC-5, are two of the more widely used

and well documented reservoir-system simulation models for simulating the operation of

a system of reservoirs in a river network for flood control, water supply, hydropower, and

instream flow maintenance for water quality (Wurbs 1993 and Mays and Tung 1992).

The HEC-3 Reservoir System Analysis for Conservation program is much simpler than

HEC-5 but does not have the comprehensive flood-control capabilities of HEC-5 (Wurbs

1993).

Generalized mathematical water management models were also developed by

individual researchers. Ocanas and Mays (1981a), Huang and Loucks (2000), and Yang

et al. (2000) applied their reservoir/river management model to hypothetical river

networks. Other applications have a specific application network such as South Taiwan

(Yen and Chen 2001), the Aral Sea basin of central Asia (Cai et al. 2003), the Rio Grande

river from Elephant Butte, New Mexico to Fort Quitman, Texas (Ejeta et al. 2004), Syr

Darya River basin (Cai et al. 2001), and water supply system in southern Israel (Cohen et

al. 2004). However, these models were not generalized as they did not consider all

possible components and model generality and flexibility were insufficient to allow end-

users to easily modify the model. RiverWare (Biddle 2001, Magee and Goranflo 2002,

Gilmore et al. 2000, Fulp and Harkins 2001) is a general object-oriented model but is

limited to reservoir management.

All previous works mentioned above generally did not incorporate water quality

parameters in the models. The exceptions are Ocanas and Mays (1981a and 1981b) who

simulated biochemical oxygen demand (BOD) and total suspended solid (TSS). Cai et al.
 52

(2001 and 2003) and Cohen et al. (2004) modeled salinity and Ejeta et al. (2004)

incorporated a total dissolved solid (TDS) component.

More recently, general system dynamics simulation object oriented models have been

developed. In one of the earliest applications, Palmer et al. (1993) tailored a dynamic

simulation software application to represent the Portland water supply system. Other

applications include river basin planning (Palmer et al. 2000), long-term water resource

planning and policy analysis (Simonovic et al. 1997; Simonovic and Fahmy 1999),

reservoir operation (Ahmad and Simonovic 2000), sustainability of a water resource

system, and water supply planning and management (Nandalal and Simonovic 2003).

System dynamics modeling was also used to model sea level rise in a coastal area by

Ruth and Pieper (1994).

Simonovic and Bender (1996) applied dynamic simulation in a collaborative

planning-support system to relate environmental issues, e.g., fish habitat, to hydroelectric

power generation. Stave (2003) prepared a system dynamics model of the Las Vegas

water supply system to increase public understanding of the value of water conservation.

Passell et al. (2002) presented a computerized dynamic simulation model of the

hydrology, ecology, demography and economy of the Middle Rio Grande Basin. Water

sustainability and groundwater storage in San Pedro River, Basin (AZ) was simulated by

Sumer et al. (2004).

In this paper, a set of modules are discussed that model various components of a

water supply system including water treatment and groundwater recharge. These modules

can be linked to represent the complete general water supply system and allow users to
 53

evaluate alternative water management options. Total construction and operations costs

and water quality and availability are computed in the model. Given the current interest

in decentralized treatment, a hypothetical system is analyzed to evaluate the cost-

effectiveness of multiple treatment facilities within a community.

3. MODELING TOOLS

Various object-oriented dynamic simulation modeling tools are available including

Stella (http://www.isi.edu/isd/LOOM/Stella/), Dynamo (http://www.cs.auc.dk/

~normark/dynamo.html), Vensim (http://www.vensim.com/software.html), and Power-

sim (http://www.powersim.no). The power of object-oriented simulation is the ease of

constructing “what if” scenarios and tackling large, messy, real-world problems

(Nandalal and Simonovic, 2003). Powersim has flexibility in linking to other software

like Visual Basic, Visual C++, and Web program using Powersim SDK

(http://www.powersim.no). This feature makes Powersim Studio more powerful than

other object-oriented modeling languages. In this study, Powersim Studio 2003 was used

to develop water supply system modules and was linked to Visual Basic Studio for input

processing.

3.1 Modeling Objective

The goal of this effort is to develop a set of modules representing various water uses

and treatment options that can be easily combined to describe a general water supply
 54

system. The modules can be tailored to the specific location by varying the module

parameters. Combining modules requires limited programming ability and promotes

rapid development of water resources management tools. Inclusion of a water quality

component is a unique feature of the tools and allows for examination of decentralized

treatment for specific waste streams or within a recycle/reuse system. In addition to the

modular structure, advantages of a dynamic simulation approach are the simple interfaces

and transparency in equations and relationships that comprise the model. This paper

describes the overall approach, the relationships comprising each module, and an

application to a hypothetical southwest US water supply system.

3.2 Generic System

Figure A.1 shows a general water supply system that includes all of the modeled

water supply, demand, and treatment components. Water supply components are

imported water, river and reservoir, subsurface, precipitation, and reclaimed water. Water

demand components are relationships describing the amount of water needed for various

purposes. Agricultural, domestic, industrial, large outdoor uses, such as parks, schools,

and golf courses, and environmental and riparian area are represented. The first four

sectors’ demands are computed by determining individual user or unit area demands and

aggregated over the sector. Environmental and recreational uses are based on estimated

in-stream water requirements. In Figure A.1, one of each supply/demand type is shown.
 55

However, multiple components of the same type can be included in a model. For example,

each community within a watershed may have separate treatment facilities.

Water is conveyed to various users by pipes or canals that have defined capacities.

Mass balances within the system are computed accounting for appropriate connections

between uses and sources/sinks. A simple mass balance relationship is applied for most

systems:

S t − S t − ∆t = ∆S t = ∑Q
inflow
i ,t − ∑Q
outflow
o ,t (A.1)

where St is the storage in the system at time t and Qi,t and Qo,t are the inflows and

outflows during the time interval, ∆t, respectively.

Some interactions are described with more complex relationships. For example, flows

between surface and groundwater sources are based on hydraulic head differences.

Incidental discharge from the water distribution and sewer systems and planned aquifer

recharge from recharge basin are accounted for in mass balance relationships as

consumptive evaporative uses and flows into/out of the system. For planning purposes,

the dynamic simulation model performs calculations on a seasonal time step. Alternative

time steps can also be examined.

Simplified (lumped) and conventional water and wastewater treatment facilities are

included in the DSS. Lumped water and wastewater treatment facilities have constant

removal efficiencies and calculate effluent quality by a mass balance equation.

Conventional plant models are based upon current environmental engineering literature

and provide more detail on unit operation removal efficiencies. Rapid mixing and

flocculation, disinfection with chlorine, sedimentation, filtration, and sludge handling

 56

using drying beds comprise a conventional water treatment system and a conventional

wastewater treatment system is composed of primary settling, aeration tank, secondary

settling, gravity thickening, anaerobic primary digestion, anaerobic secondary digestion,

and vacuum filter. Each unit operation has a user-defined capacity.

Water quality is measured in terms of biochemical oxygen demand (BOD), total

suspended solids (TSS), hardness and Giardia (as a bacteria indicator). Depending upon

the use, surface water can be sent to directly to users or through a water treatment plant.

Groundwater is assumed to only require disinfection thus it can be delivered directly to

all users. Reclaimed water can be applied for agricultural and large outdoor irrigation, if

it is of acceptable quality.

Effluent from users is collected and sent to wastewater treatment plants (WWTP). An

on-site decentralized wastewater treatment plant may be added to an industrial use before

effluent is discharged to the WWTP. Additional modules are being developed that will be

applicable to other specific contaminants that may be more economically removed on-site

rather than at the wastewater plant.

The spatial distribution of users or elevation differences may suggest that

decentralized water and wastewater treatment plants may be appropriate to reduce total

system cost and can be represented in the simulation model. Outflow from the WWTP or

advanced WWTP can be discharged to a downstream river directly or indirectly

recharged to an aquifer (Figure A.1). As noted, multiple components of the same type can

be included in the system model by reproducing the modules and applying the

appropriate parameters for that process. Total system cost considers construction and
 57

expansion of new and existing structures, operation and maintenance of water supply

system, and water consumption by users. In addition to meeting water demands

operational and construction costs are computed based upon literature relationships (US

EPA, 1976). The following sections provide more details for each system component

including their decisions and the parameters describing each process.

3.3 General Components

Population/Households/Businesses

Population is the primary driving factor for water demand. Population growth is

assumed to follow a power function with an annual percentage increase. Other models

can be substituted. The number of households and businesses are estimated by assuming

an average number of persons per household and households per business, respectively.

Mass balance

The overriding governing principle in a large-scale basin model is conservation of

mass (Eq. A.1). Surface and subsurface sources, flow transportation facilities, and users

must preserve mass balance. Only reservoirs and aquifers have storage. Inflows and

outflows must balance for all other components (i.e., ∆S equals 0).

Inflows to a component can be provided from an upstream component (e.g.,

residential demand center sends water to a WWTP). In addition, five external flows can

supply water to a system: precipitation, unregulated rivers, regulated river flows

 58

(reservoir), imported water, and natural groundwater recharge. Water leaves the basin as

groundwater pumping, streamflow or evaporation/transpiration.

Aquifer/Recharge basin

Aquifers are modeled by a mass balance relationship and may have a maximum

capacity. Flows to and from the aquifer are related by Darcy’s Law for unconfined

conditions. The one-dimensional Dupuit equation can be derived from Darcy’s Law

under the assumptions: 1) the velocity of flow (Q, m3/s/m) is proportional to the tangent

of the hydraulic gradient and 2) the flow is horizontal and uniform in a vertical section or:

Q=K
2
h1 − h2
2
( 2
h − h2
=K 1
2
) (A.2)
2( x 2 − x1 ) 2∆x

where K is soil conductivity (m/s), ∆x is the flow distance between two components, and

h1 and h2 are the component’s hydraulic heads. This relationship is applied to flows

between river and aquifer, and reservoir and aquifer. Pumped groundwater is withdrawn

to meet user demands. Natural inflows are supplied as mountain front recharge and are

not related to head differences.

3.4 Demand/ User Components

Agricultural/Large outdoor

Agricultural and large outdoor consumptive use (evaporation/transpiration and crop

storage) varies by crop type (including turf) and season (Table A.1). Typical southwest
 59

US crops (Alfalfa, cotton, lettuce and wheat) are available to be selected in the model. An

override option allows users to provide data for alternative crops or a lumped water

demand for entire agricultural area. Large outdoor turf uses include schools, parks and

public and private golf courses. Golf course uses are further categorized by turf type (e.g.,

fairways versus greens). Consumptive use per acre of crop is computed by:

Dcrop = CU crop Acrop (A.3)

where Dcrop is a total crop consumptive use (m3/s), CU crop is consumptive use factor

(m3/s/m2), and Acrop is crop acreage (m2).

Agricultural land retirement is provided as an option to remove land from production.

All outdoor uses (except lettuce crops) can accept reclaimed water or direct supply from a

source. Therefore, the priorities of which water should be used first or the proportion of

each water source must be user defined. Depending on the proportion selected, total

revenue against total cost varies (Table A.2). Required model parameters for the large

outdoor areas are listed in Table A.3.

Domestic

Domestic use is based on the area’s population and the number of households. Both

indoor and outdoor domestic uses are modeled in reasonable detail. Indoor water use

components are toilets, showers, faucets, evaporative coolers, clothes washers, bathtubs,

and dishwashers. In general, the use by each component is computed by:

D fix = NU fixUSE fix NP (A.4)

 60

where D fix is total fixture use per household per day, NU fix is number of fixture uses per

person per day, NP is number of persons per household, and USE fix is water use per

fixture use.

Typical values for the number of fixture per household and its uses per day are given

in Table A.4. As seen, toilet water use is dependent on installation time. High efficiency

fixtures are assumed to be installed after 1990 while lower efficiency fixtures were

assumed to be installed prior to 1984. Thus, toilets are divided into three age groups,

before 1984, after 1990, and between 1984 and 1990. In the same manner, shower units

have two efficiencies which are low efficiency prior to 1994 and high efficiency after

1994. The total of all fixtures is the summed and scaled by the number of households. A

small percentage of indoor use is assumed to evaporate and the remainder is returned to a

WWTP. Grey water reuse is used first to meet a residence’s outdoor demand.

Outdoor residential water use consists of permanent irrigation water use, fountains,

and swimming pools. Permanent irrigation water use consists of turf watering and drip

irrigation. Outdoor use for irrigation is computed by an equation similar to Eq. A.3.

Swimming pool evaporation is estimated as the evaporation rate times the surface area of

the pool. Outdoor water monthly demand is reduced by the average monthly precipitation

depth. Irrigated areas, average pool size and the percentage of homes with pools and

fountains are user defined. Figure A.2 shows mass balance relationship through four

domestic areas.

Industrial
 61

Industrial use is assumed to be proportional to the number of businesses in the study

area. The total water demand of industrial area is calculated as the sum of indoor and

outdoor water use for general commercial, car washes, and swimming pool uses. The

diversity of industry and their water use makes it difficult to develop more sophisticated

relationships. As defaults, eighty-five percent of industrial use is assumed to be an indoor

use and returned to the WWTP and the remainder considered as outdoor consumptive

use. The default average per business water use was estimated from billing records for

the Upper San Pedro subwatershed (Southern Arizona).

Environmental and recreational uses

Total water demand in environmental and recreational areas is the total consumptive

use of the area’s vegetation. Here, we limit the analysis to three trees (Cottonwood,

Tamarix, and Mesquite), grass and open water. Table A.5 lists the default consumptive

use for each category. The user supplies the areas of the comprising riparian area and the

percent of different trees (Table A.5). The use by each riparian category is computed by:

Dveg = CU veg Aveg (A.5)

where Dveg is total riparian demand (m3/s), CU veg is consumptive use per acre (m3/s/m2),

and Aveg is acreage of vegetation (m2).

Benefits of the riparian area in terms of fishing, float-boating, hiking, and camping

are assumed to be related to the streamflow and the area’s population. Default

relationships were developed by Stewart (personal communication, 2005) based on

compilations of riparian area economic evaluations from around the United States and
 62

regional recreation statistics (Cordell, 1999). These benefits are intended to estimate the

economic value of riparian area but are site specific. Minimum required riparian flows

can be defined or an additional fee to preserve naturalness can be assessed.

3.5 Water Quality Transformations

Biochemical oxygen demand (BOD), total suspended solids (TSS), hardness and

Giardia are modeled as representative water quality indicators. An incremental

deterioration in water quality is assumed during each use. Two water and wastewater

treatment facility representations are available to model treatment of the four pollutants: a

lumped facility with an incremental change or representative removal efficiency and a

series of conventional unit processes with water quality changes based on literature

equations.

Users/Lumped treatment facilities

The lumped incremental improvement or removal efficiency model follows Ocanas

and Mays (1981a and b) in which water quality is improved by a constant increment (Eq.

A.6a) or related to the efficiency of a pollutant removal (Eq. A.6b):

Cout = ∑ (Q C ) + ∆
in in
(A.6a)
∑Q out

(1 − η )∑ Qin Cin
C out = (A.6b)
∑Q out
 63

where ∆ is amount of quality degradation (mg/l) and η is removal efficiency (%). Cin

and Cout are the contaminant concentrations in the treatment plant influent and effluent,

respectively.

Water quality degradation during use can be estimated as the difference between

WTP outflow and WWTP inflow concentrations from historical data and substituted as

∆ in Eq. A.6a. After use, wastewater is returned to a WWTP and simplified water and

wastewater treatment facilities adopt the lumped removal efficiency (Eq. A.6b). Input

parameters for lumped treatment facilities are listed in Table A.6.

Conventional water treatment plants

More complex WTP unit operation algebraic descriptions are available in literature

(Gummerman et al., 1979 and Reynolds, 1982). Each WTP is assumed to contain the

standard WTP unit processes; rapid mixing and chemical addition, flocculation,

disinfection with chlorine, sedimentation, filtration, and sludge handling using drying

beds (Figure A.3).

Capacities of the unit processes are dependent on the design flow (DF, m3/s) that is

assumed to be 1.5 times of inflow at the beginning of the simulation period. For example,

basin volume of rapid mixing (RV, m3) with a constant detention time (DT), 30 second,

is:

RV = DT DF (A.7)

Basin volume of flocculation (FV) is calculated by an assumption of a 30 minute

detention time (FDT) or:

 64

FV = FDT DF (A.8)

The sedimentation basin is designed in a similar manner assuming the basin depth

(DEPTH) is 3.66m (12ft ) or:

5570.4SDT DF
SAREA = (A.9)
DEPTH

where SAREA is the total surface area of the sedimentation basin (m2) and SDT is the

detention time (5 hrs).

The filter surface area (FAREA) depends on the influent flow rate to filtration basin

from the sedimentation basin, FLOW3, or:

694.4 FLOW3 ECF

FAREA = (A.10)
QAVE

where ECF is ratio of design flow and operation flow and QAVE is maximum allowable

average filtration rate (m3/s).

Water quality improvements occur in each unit process except during rapid mixing

that adversely affect TSS. Aluminum hydroxide (AlOH3, mg/l) is formed by injecting

alum in the rapid mixing basin and increases turbidity.

TSS RM = TSSin + ( AlOH 3 )in + 0.26 ALUM (A.11)

where TSSRM and TSSin are TSS concentrations in the effluent from and influent to the

rapid mixing unit, (AlOH3)in is the amount of AlOH3 included in natural inflow, assumed

as 0 mg/l, and ALUM is the injected amount of alum to the rapid mixing unit. The TSS

concentration during flocculation is constant and decreased in sedimentation and

filtration units corresponding to sedimentation tank overflow rate and the filter’s effective

sand size (0.5 mm).

 65

Removal efficiencies of BOD and hardness are assumed to be the same for TSS. For

Giardia, the removal efficiencies during flocculation, disinfection, and filtration are 0.5

log (68.4 %), 0.5 log, and 2.0 log (99.0 %), respectively.

Table A.7 lists the input parameters for the above equations and the default values.

The total water treatment cost is the material and labor costs for construction and

maintenance (Smith, 1986).

Conventional wastewater treatment plant

The WWTP is modeled using the equations collected by Tang et al. (1984 and 1987).

These equations are based on unit operations in a conventional facility: primary settling,

aeration tank, secondary settling, gravity thickening, anaerobic primary digestion,

anaerobic secondary digestion, and vacuum filter (Figure A.4). The unit process sizes are

input to calculate the removal efficiency and cost of each unit. For simplicity, unlike

Tang et al., the recycling rate of water to the primary settling is fixed.

BOD and TSS are simulated using Tang’s equations. However, since relationships are

not available for hardness and Giardia, their removal efficiencies are assumed to be the

same as the lumped system. TSS is the sum of active biomass concentration, volatile

biodegradable suspended solids, fixed suspended solids, and volatile inert suspended

solids. Cost relationships include construction and operations and maintenance. Equation

parameters are listed in Table A.8.

 66

4. APPLICATIONS

To demonstrate how the model can be used, three scenarios are considered on a

hypothetical system for a twenty year planning period (2000-2020). The first scenario

examines the impact of conservation measures aimed at reducing domestic and industrial

demands. The second condition represents the inability to secure additional external

water sources and its impact on long-term system storage. Finally, the cost effectiveness

and impact on water quality of decentralized treatment is analyzed for a disperse supply

system that covers a range of topography.

The hypothetical system is comprised of five sources (precipitation, imported water,

uncontrolled river, regulated river with reservoir, groundwater), twelve users (four

domestic areas, an industrial area, four agricultural areas, two large outdoor turf areas, an

environmental and recreational area), five treatment systems (two water treatment and

three wastewater treatment systems), and a recharge facility. Table A.9 lists the

population input parameters (initial population and growth rates).

Historical precipitation data from Coolidge, AZ was taken from the Arizona

Meteorological Network (AZMET) (http://ag.arizona.edu/azmet/.html) for the period of

1987 to 2004 (Figure A.5). As noted, precipitation provides flow into water sources such

as rivers, reservoirs, and the groundwater aquifer and is assumed to reduce outdoor water

demands for turf and agriculture.

A watershed (5,598 km) of San Pedro River from the Mexican Border through

Benson, AZ was used for calculating runoff. Runoff coefficients for the catchments area

was taken as 0.44 for average cultivated area (Chow et al. 1988. Table 15.1.1). Total soil
 67

loss from catchments area was calculated as a constant rate 476.26 m3/km2/yr (1.0

af/mi2/yr) of drainage area (Branson et al. 1981, Figure 6-24). Only 0.5% of the soil loss

from the basin was assumed to contribute to TSS in inflows to the river and reservoir.

River flows can also be provided from upstream channel reaches. Historical flow data

from the Salt River, AZ (Figure A.6) was taken for this inflow from USGS NWIS web

data (http://waterdata.usgs.gov/nwis) for the period of 1987 to 2004. Inflow to reservoir

and outflow from reservoir was assumed as historical data at Roosevelt Dam in the Salt

River (Figure A.7) that were taken from USGS NWIS web data. Surface water sources

are connected with the underlying aquifer by Eq. A.2. Precipitation also infiltrates to the

aquifer.

Monthly historical data for 1987 to 2004 were converted to seasonal data and to a

time series that had the same statistical characteristics for the period of 2000 to 2020. The

imported water is limited to 1.956 m3/s (50 kafy). Controlled recharge is assumed to

occur through an infiltration basin that was modeled using Eq. A.2.

Table A.10 lists the layout and flow connections between the system components.

Open channel gravity flow transports untreated water when permitted to reduce cost.

However, pumping through pipelines is required in some cases such as extracting

agricultural irrigation water from groundwater. Pipe flow velocities are bounded to less

than 1.5 m/s (5 ft/s) by changing pipe diameters through trial and error by the author

before simulation. The leakage rate from distribution and sewer pipes is assumed to be 10

percent of the flow and evaporation from canals are included. Evaporation is 1.1
 68

m/season (42.5 in/season) for the growing season, and for non-growing season, 0.5

m/season (9.62 in/season) or may be modified as is input.

Default agriculture water use is based upon Pima County, Tucson, AZ data (Table

A.1) and required inputs are the total acreages for each crop/turf area (Table A.2). The

cost and return for crops is computed on an annual basis. Unit yield, prices and costs are

listed in Table A.2. Water costs of new water (direct from a source) and reclaimed water

is $0.024 and $0.02 per m3 ($30 and $25 per acre-ft), respectively. If the flowrate in the

riparian zone/recreation area is less than 95% of the upstream flowrate, a fee of

$35/month per person in the basin is assessed to maintain the natural areas.

Water and wastewater treatment plant can be chosen as lumped or conventional

systems. Table A.9 shows the incremental removal efficiencies/degradations and the

water losses for the lumped treatment systems. Removal efficiencies in conventional

treatment systems are computed using the unit process Eqs. A.8 - A.12 and Tang

equations. Conventional systems were set as the default simulation. Contaminant

concentrations are assumed to be reduced by 30% during flow to the aquifer.

The following results presented in the following sections were developed by either

single model evaluations or extensive trial and error and engineering judgment. They are

not necessarily optimal solutions.

 69

4.1 Scenario 1 – Effectiveness of Conservation Practices

Conservation practices are means to stretch current water supplies. Population growth

controls may also be employed to limit demand increases and to avoid constructing new

infrastructure and/or water supplies. Here, a scenario is posed in which a community

desires to examine the impact of these types of measures on the base condition described

above. Each measure (Table A.11) is initiated independently in 2012.

To reduce indoor use, incentives for (1) faucet, shower head, and toilet replacement

with more efficient fixtures and (2) front loading clothes washer purchases can be

provided. The amount of savings depends upon the age of the existing fixtures/appliance.

An annual government investment must be defined to implement the incentive program.

It is assumed that a $70 incentive will be provided per house and the total annual subsidy

is $100,000.

Generally outdoor use reductions are made by ordinance for new homes. For

example, new homes may be required to have a grey water reuse system in which water

from bathtubs, faucets, clothes washers, and showers is collected for outdoor purposes.

Grey water retrofit construction costs for existing homes are prohibitive. All of the above

measures will reduce demand for pumping but will also reduce the amount of reclaimed

water available for large area irrigation. Other ordinances can include prohibiting

fountains and evaporative coolers in new homes and requiring existing homes to remove

fountains.

Swimming pools are a major consumptive use. In the climate conditions of the

hypothetical community, the average evaporation loss from a swimming pool is about 5
 70

m/yr (19 ft/year). Evaporation losses can be reduced by one or more set of measures. An

incentive program could be introduced for pool cover purchases that would be used in the

off-season. Other ordinances for reducing pool use are reducing pool draining

frequencies, restricting future swimming pools construction, filling existing swimming

pools, or reusing swimming pool drain or back-flush water as grey water.

Other water saving programs that can be evaluated are briefly described below.

During the growing season, outdoor water use restrictions prohibit irrigation on a

particular day or time by ordinance to reduce evaporation volumes. An irrigation

efficiency program will require efficiency gains from 75% to 90%. Landscaping

standards require turf irrigation systems to be replaced by drip irrigation or xeriscaping.

Rainfall harvesting can replace treated water usage to reduce outdoor water use in new or

existing homes. Lastly, a water wasting ordinance imposes fines for wasting or

unreasonable use of water. These losses are assumed to be zero after the ordinance

implement.

Similar conservative measures, such as toilet retrofit, improvement of water drip

irrigation efficiency, outdoor water use restriction, water loss due to violations and

swimming pool savings, can be applied to the industrial sector. Only purchasing pool

covers is supported with a government incentive. All other programs are implemented by

ordinance. In addition, water audits can be completed on these large users. An initial

investment for water audits is $4,000 with annual cost of $2,000 per year thereafter for 10

years and provides an expected 27,137 m3/yr (22 afy) of water savings.
 71

Water demand before and after implementing conservation measures and other basic

input parameters are listed in Table A.12. Incentives have an annual cost for a program

and a cost per house (Table A.13). Table A.14 lists the effects of each of the conservation

measures on the total water supply system’s operation cost and water use savings in

domestic areas for the 20-year simulation period when implemented in year 2012.

Facility construction cost remains the same, while operation cost declines with respect to

water use decrease. Operational cost for pumping and piping is reduced as the amount of

water conveyance decrease. Some conservation measures decrease water use by reducing

demand through high efficient fixtures. Water and wastewater treatment operational cost

also decreases due to the water demand reduction. The other conservation programs

decrease fresh water usage by using reclaimed water through grey water recycling,

rainfall harvesting, or pool discharges, while water demands remain the same.

Removing turf and replacing with drip irrigation systems saves 9.2% of the water

demand (use as well) and total operation cost. Installing grey water reuse systems reduces

the total operation cost by 8.95% and the domestic use by 15.07%. As mentioned in

Table A.11, an incentive program such as for installing grey water reuse system incurs an

additional cost to the water provider. As a result of this expense, the cost savings

percentage is smaller than the water demand reduction percentage. The trends in water

demand at the Domestic Area 1 are shown in Figure A.8 for implementing grey water

reuse and clothes washer retrofit incentive programs.

Some conservation measures, such as reusing pool discharges and restricting outdoor

water use do not have any benefit on water and cost saving. For example, the volume of
 72

pool discharge to wastewater treatment systems is small. Restricting outdoor water use

does not change the volume of water applied since turf water demands are not adjusted

although the irrigation efficiency is improved. Industrial conservation has similar, but

smaller effects (Table A.15) since 85% of the industrial water use is used for indoor

purposes.

When all conservation measures including the installation of rainwater harvesting

system on water supply are simultaneously implemented in 2012, the water use in the

domestic areas fluctuated based on the amount of precipitation and decreased by more

than 70% (Figure A.9 and Table A.16). Total system cost, however, only decreases by

2.6% (Figure A.10). This decrease is not significant to a long-term water management

plan. However, because the primary goal of the plan is ordinarily safe-yield, the positive

effect on groundwater storage is a 0.9% increase over time as a result of increasing

population (Table A.16).

4.2 Scenario 2 – Unavailability of Supply Sources

As water resources become more stressed and climatic variability increases, the

potential for reduced supply is more likely. In addition, environmental systems may also

require additional water. In this scenario, it is assumed that the imported water source is

no longer available and that the existing use of reclaimed water as the riparian zone’s

water source precludes its use for other purposes. Serious repercussions are anticipated in

the water balance and groundwater mining must occur.

 73

The main imported and reclaimed water use is agricultural. Average seasonal

consumptive use of each agricultural area is 0.063 km3/yr (51.30 kafy) and 0.114 km3/yr

(92.36 kafy) for growing and non-growing seasons, respectively. About 16% of

consumptive use during growing season is supplied from imported water (0.010 km3/yr)

and reclaimed water (0.009 km3/yr) under the base scenario that is defined as both

imported and reclaimed water are available. Figure A.11 shows the sources providing

water to Agricultural Area 1. The total amount of water from imported water and

reclaimed water to the four agricultural areas are 0.030 km3/yr and 0.034 km3/yr,

respectively.

Groundwater is used to replace the unavailable supplies for three cases: without

imported water, without reclaimed water, and without imported and reclaimed waters.

Figure A.12 shows the resulting groundwater storage change. Nearly 19% of the

groundwater storage is depleted when imported and reclaimed waters were not available

(Table A.17) demonstrating the need in many communities for these supplies in order to

maintain a sustainable system.

4.3 Scenario 3 – Decentralized Treatment

Decentralized wastewater treatment has become a topic of interest as a cost effective

means to treat and recycle effluent. For small communities, unsewered communities, and

communities covering a range of elevations, cluster wastewater treatment system, also

known as community-wide decentralized wastewater management, may be an appropriate

 74

treatment option. The Minnesota Pollution Control Agency (2000) and Otis (2004)

investigated and described the need of wastewater treatment system in unsewered areas

and the benefits of decentralized wastewater treatment systems. They described a cluster

system as a wastewater collection, treatment and disposal systems that serves a small

number of units.

Here, we consider a distributed wastewater scheme in which multiple satellite

WWTPs are located throughout a community with the ability to treat and distribute

reclaimed water to nearby users (Figure A.13). The configuration is based on a dispersed

urban-suburban community with a maximum population of about 1.2 million. Economies

of scale suggest that, under many conditions, a single large WWTP would be less

expensive than multiple plants. However, when the community covers a range of

elevations and/or a large area, pumping and piping costs for reclaimed water may be

more expensive than construction cost of multiple wastewater treatment plants.

The efficacy of constructing up to two water treatment systems and three wastewater

treatment systems in the hypothetical system is examined in this application. Costs are

computed for treatment plant construction and operation, pumping and piping for water

transfers, system expansion, operation and maintenance of treatment system and energy

needed for pumping. Pipes between sources, treatment systems, and users were assumed

to be laid to cover the least distance and elevation and determined by a trial and error and

engineering judgment.

Cost and treatment capacities

 75

Table A.18 lists the links and treatment facilities for the eight different combinations

compared with the base condition that is defined as two water treatment systems and

three wastewater treatment systems (Table A.10).

Results for one centralized and two decentralized lumped water treatment systems are

shown in Figure A.14. In the two WTP case, water treatment system 1’s capacity is not

increased from its initial size of 0.033 km3/yr while water treatment system 2 is expanded

from a capacity of 0.154 km3/yr to 0.296 km3/yr. When only one centralized WTP is

permitted, its capacity is equal to 0.187 km3/yr to 0.330 km3/yr (conditions 1 and 2 in

Table A.18).

The wastewater treatment system is more complicated as each type of wastewater

treatment system has six alternative wastewater treatment system combinations

(conditions 3 ~ 8 from Table A.18) on each lumped or conventional WWTP

representations. Figure A.15 shows the required simplified treatment plant capacity over

time for one centralized and three decentralized plants (base condition). Unit process

capacities for the one centralized wastewater treatment system alternative are shown in

Figure A.16 as a function of the surface area.

Tables A.19 and A.20 give detailed plant and unit operation capacities for all six

combinations of lumped and conventional system models, respectively. Since these

alternatives do not affect use, the sum of the plant capacities from multiple plant systems

is the same as for the single system.

Figure A.17 shows the total cost change over the planning period for conventional

and lumped treatment system representations, while Table A.21 lists the total cost in
 76

present values, annual cost, and cost differences for the different treatment systems. As

shown in Figure A. 17, total cost jumps every five years because of system construction

and expansion. Lumped construction and operation cost (293.4×106 $/yr, for the base

condition) for treatment systems appear to be overestimated compared to conventional

representations (242.4×106 $/yr, for the base condition). However, the optimal treatment

distribution in this example for both lumped and conventional cases was 2 WTPs and 3

WWTPs. Finally, Table A.22 lists construction, expansion, and operation cost of all

components for all conditions for conventional and simplified treatment plant

representations. Since the transportation cost is dominated by piping/pumping, a remote

system is much more expensive.

The overriding factor in the cost effectiveness of multiple treatment plants is the cost

for transporting water through the system. These costs are determined by the distance and

elevation changes in the region. As seen in Table A.22, the operations and maintenance

costs for pumping when few plants were used was significant compared to treatment

costs. Therefore, construction of two water treatment plants and three wastewater

treatment plants (base condition in Table A.18) is the least expensive treatment system

for the community.

Water quality

Differences in water quality are examined between plant alternatives at key locations

and within the system for one plant configuration. Table A.23 lists the removal

efficiencies for the conventional and lumped WTP and WWTP models. In the unit
 77

operation models for conventional systems, removal efficiencies are dependent upon the

influent water quality. Hence, values in Table A.23 are average removal efficiencies from

the total process.

Figure A.18 shows BOD and TSS level changes in the major water sources. Reservoir

water quality is stable over the period, while the river fluctuates with the discharge. The

TSS level in groundwater improves over the time as a result of continuous recharge of

clean reclaimed water. If credit is received for treatment during recharge, costs can be

reduced by increasing aquifer recharge. Pollutant concentrations in the river and reservoir

(Figures A.18 and A.19) remain relatively constant over time, however groundwater

contamination of all four pollutants are reduced.

Figures A.20, A.21, and A.22 show effluent water quality of BOD and TSS,

hardness, and Giardia from each user. Since water quality deterioration is assumed to be

constant over the time for a given user type, the effluent quality does not change.

Agricultural water demands usually show significant seasonal variations depending on

the crop growing season. Reclaimed water for irrigation may contain high mineral

content and Giardia levels. Giardia levels in the groundwater are maintained below 1/ml:

making it a suitable domestic source.

5. CONCLUSIONS

Given the complexity of water supply systems, decision makers may have difficulty

understanding the impact of water management policies. To understand the impact of

 78

decisions, a modular water system simulation model has been developed. The modular

structure allows a general system model to be constructed with minimal effort. Model

components include detailed domestic indoor and outdoor usage, industrial, agricultural,

and environmental demands, multiple supplies, conveyance between sources and demand

centers, and surface and groundwater storage. Since water quality plays a key role in

water management, it is modeled through the system including water and wastewater

treatment representations. Simulation is performed on seasonal time steps with decisions

at critical periods. Costs are estimated for new infrastructure (conveyance, recharge, or

treatment facilities), water usage, and the implementation of conservation practices.

A series of applications demonstrate the ability to examine the impacts of external

factors and to assist in making a range of decisions. The first application showed the

relative impacts of a set of conservation measures on the water demand and consumptive

use. These results would assist decision makers in determining the utility of

implementing the measures, the general public in understanding their value and both

groups in developing a consensus on accepting the plans. The loss of water sources is

examined in the second application to demonstrate its implications on long-term storage

and system sustainability.

A large spatially distributed system is considered in the final example to evaluate the

potential of distributed water and wastewater treatment. Costs for alternative system

designs were determined using engineering judgment and trial and error. The results

show that the distance and elevation changes between demand centers in the example
 79

system are such that multiple distributed treatment facilities are more cost effective than

single centralized plants.

Three primary extensions of the present level of modeling are recommended. To be a

useful practical tool, the models should be accessible through the internet so the entire

community can perform simulations. Second, in the applications here the resulting

decisions were made based on engineering judgment, these modules should be linked

with optimization routines. Lastly, additional modules should be developed to extend the

model capabilities. These components can be combined in the present structure and

provide more detailed descriptions of unit operation processes.

 80

6. REFERENCES

Ahmad, S., and Simonovic, S. P. (2000). “System dynamics modeling of reservoir

operations for flood management.” Journal of Computing in Civil

Engineering, 14(3), 190-198.

Biddle, S. H. (2001). “Optimizing the TVA reservoir system using RiverWare, bridging

the gap: meeting the world’s water and environmental resources challenges.”

Proceedings of the World Water and Environmental Resources Congress,

ASCE.

Branson, F. A., Gifford, G. F., Renard, K. G., and Hadley, R. F. (1981). Rangeland

hydrology, range science series 1. Society For Range Management, Denver,

CO.

Cai, X., Mckinney, D. C., and Lasdon, L. S. (2001). “Piece-by-Piece approach to solving

large nonlinear water resources management models.” Journal of Water

Resources Planning and management, 127(6), 363-368.

Cai, X., McKinney, D. C., and Lasdon, L. S (2003). “Integrated hydrologic-agronomic-

economic model for river basin management.” Journal of Water Resources

Planning and Management, 129(1), 4-16.

Chow, V. T., Maidment, D. and Mays, L. (1988). Applied hydrology. McGraw-Hill Book

Co. Singapore.
 81

Cohen, D., Shamir, U., and Sinai, G. (2004). “Sensitivity analysis of optimal operation of

irrigation supply systems with water quality considerations.” Irrigation and

Drainage Systems, 18, 227-253.

Cordell, H. Ken (1999). Outdoor recreation in American life. A national assessment of

demand and supply trends. Sagamore Publishing.

Ejeta, M. Z., McGuckin, T., and Mays, L. W. (2004). “Market exchange impact on water

supply planning with water quality.” Journal of Water Resources Planning

and Management, 130(6), 439-449.

Erie, L. L, French, O. F., Bucks, D.A., and Harris, K. (1982). Consumptive use of major

crops in the Southwestern United States. USDA-ARS Conservation Research

Report Number 29.

Fulp, T. and Harkins, J. (2001). “Policy analysis using RiverWare: Colorado river interim

surplus guidelines.” Proceedings of ASCE World Water & Environmental

Resource Congress, Orlando, FL.

Gilmore, A., Magee, T., Fulp, T., and Strezepek, K. (2000). “Multi-objective

optimization of the Colorado river.” Proceedings of the ASCE 2000 Joint

Conference on Water Resources Engineering and Water Resources Planning

and Movement, Minneapolis, MN.

Gummerman, R. C., Russell, L. C, and Sigurd, P. H. (1979). Estimating water treatment

costs. EPA-600/2-79-162a. Cincinnati, OH.

 82

Huang, G. H. and Loucks, D. P. (2000). “An inexact two-stage stochastic programming

model for water resources model for water resources management under

uncertainty.” Civil Engineering and Environmental Systems, 17(2), 95-118.

Loucks, D. P. (1995). “Developing and implementing decision support systems: a

critique and a challenge.” Water Resources Bulletin, 31(4), 571-582.

Magee, T. M. and Goranflo, H., M. (2002). “Optimizing daily reservoir scheduling at

TVA with RiverWare.” Proceedings of the Second Federal Interagency

Hydrologic Modeling Conference, Las Vegas, NV.

Minnesota Pollution Control Agency. (2000). Wastewater Treatment and Collection

Systems.

Nandalal, K. D. W., and Simonovic, S. P. (2003). “Resolving conflicts in water sharing: a

systemic approach.” Water Resources Research, 39(12), 1362-1372.

Ocanas, G. and Mays, L. W. (1981a). “A model for water reuse planning.” Water

Resources Research, 17(1), 25-32.

Ocanas, G. and Mays, L. W. (1981b). “Water reuse planning models: extensions and

applications.” Water Resources Research, 17(5), 1311-1327.

Otis, R. J. (2004). “Unsewered communities: paralyzed by disparate regulatory

programs?” 2004 Onsite Conference X in Sacramento, 21-24 March 2004,

ASAE.

Palmer, R. N, Wright, J. R., Smith, J. A., Cohon, J. L., and ReVelle, C. S. (1980). Policy

analysis of reservoir operation in the Potomac River Basin, Volume I,

executive summary, Johns Hopkins University, Baltimore, MD.

 83

Palmer, R. N, Keyes, A., and Fisher, S. (1993). “Empowering stakeholders through

simulation in water resources planning.” Proceedings of the ASCE Water

Management in the 90s Conference, Seattle, Washington, 451-454

Palmer, R. N., Mohammadi, A., Hahn, M. A., Kessler, D., Dvorak, J. V., and Parkinson,

D. (2000). “Computer assisted decision support system for high level

infrastructure master planning: case of the city of Portland Supply and

Transmission Model (STM).” Proceedings of the ASCE 2000 Joint

Conference on Water Resources Engineering and Water Resources Planning

and Movement, Minneapolis, MN.

Passell, H., Tidwell, V. and Webb, E. (2002). “Cooperative modeling: a tool for

community-based water resource management.” Southwest Hydrology, 1(4),

26.

Reynolds, T. D. (1982). Unit operations and processes in environmental engineering,

PWS Publishers, B/C Engineering Division, Boston, Massachusetts.

Ruth, M. and Pieper, F. (1994). “Modeling spatial dynamics of sea-level rise in a coastal

area.” System Dynamics Review, 10(4), 375-389.

Simonovic, S. P. and Bender, M. J. (1996). “Collaborative planning-support system: an

approach for determining evaluation criteria.” Journal of Hydrology, 177,

237-251.

Simonovic, S. P., Fahmy, H., and El-shorbagy, A. (1997). “The use of object-oriented

modeling for water resources planning in Egypt.” Water Resources

Management, 11, 243-261.

 84

Simonovic, S. P. and Fahmy, H. (1999). “A new modeling approach for water resources

policy analysis.” Water Resources Research, 35(1), 295-304.

Smith, R. (1986). Computer assisted preliminary design for drinking water treatment

process systems. U.S. Environmental Protection Agency, Washington, D.C.,

EPA/600/2-86/007A (NTIS PB86181112).

Sumer, D. Y., Lansey, K. and Richter, H. (2004). “Evaluation of conservation measures

in the Upper San Pedro Basin.” The 2004 World Water and Environmental

Resources Congress, Salt Lake City, UT, ASCE.

Stave, K. A. (2003). “A system dynamics model to facilitate public understanding of

water management options in Las Vegas, Nevada.” Journal of Environmental

Management, 67, 303-313.

Tang, C. C., Brill, E. D., and Pfeffer, J. T. (1984). Mathematical models and optimization

techniques for use in analysis and design of wastewater treatment systems,

Research Report No. 194, Water Resources Center, University of Illinois at

Urbana-Champaign, Urbana, Ill.

Tang, C. C., Brill, E. D., and Pfeffer, J. T. (1987). “Optimization techniques for

secondary wastewater treatment system.” Journal of Environmental

Engineering, 113(5), 935 – 951.

US Army Corps of Engineers (2003). HEC-ResSim User’s Manual, V. 2, September.

(www.hec.usace.army.mil/software/hec-ressim/hecressim-hecressim.htm)

U.S. Bureau of Reclamation (1987). Colorado river simulation system: overview.

Denver, Colorado.
 85

US EPA (1976). Areawide assessment procedures manual. Appendix H. EPA 600/9-76-

014. Cincinnati, OH.

Wurbs, R. A. (1993). “Reservoir system simulation and optimization models.” Journal of

Water Resources Planning and Management, 119(4), 455-472.

Yang, S., Sun, Y., and Yeh, W. W-G. (2000). “Optimization of regional water

distribution system with blending requirements.” Journal of Water Resources

Planning and Management, 126(4), 229-235.

Yen, K. H., and Chen, C. Y. (2001). “Allocation strategy analysis of water resources in

South Taiwan.” Water Resources Management, 15, 283-297.

Zagona, E. A., Fulp, T. J., Shane, R., Magee, T., and Goranflo, H. M. (2001).

“Riverware: a generalized tool for complex reservoir system modeling.”

Journal of the American Water Resources Association, 37(4), 913-929.

 86

7. TABLES

Table A.1. Crop water use for Tucson, AZ (from 2002 Arizona Agricultural
Statistics Bulletin by the Arizona Agricultural Statistics Service)
(cm) Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Total
Alfalfa - - 15.2 15.2 30.5 30.5 30.5 30.5 15.2 15.2 - - 182.9
Upland Cotton 3.0 15.2 12.2 - - 30.5 30.5 15.2 - - - - 106.7
Lettuce - - - - - - 30.5 15.2 30.5 30.5 30.5 - 137.2
Durum Wheat - - 15.2 30.5 15.2 - - - - - - 30.5 91.4
Turf 3.0 6.1 9.1 15.2 15.2 18.3 15.2 15.2 12.2 9.1 6.1 3.0 128.0
(Erie, et al. 1982)

Table A.2. Crop revenues and costs - Pima County, Tucson, AZ

Costs
Price Yield
(without water) Area (km2)
($/ton) (tons/km2)
($/km2)
Alfalfa 100.00 617.6 27,390.8 33.2
Upland Lint 1,322.77 40.7
Cotton
42,186.2 911,344.4
Seed 140.00 71.6
Lettuce 7$/Ct 30,127 Ct 191,292.2 2.7
Durum Wheat 133.00 188.3 19,326.5 55.8

Table A.3. Sub-users and required model parameters for Large outdoor water user
User Required input
School density (number of school per population), average acreage per school,
School
water use per school acreage
Park density (number of parks per population), average acreage per park, water use
Park
per park acreage
Golf course Number of 9 and 18 hole courses per population, average water use per hole
Private golf course Number of 9 and 18 hole private courses per population, average water use per hole
 87

Table A.4. Parameters required for the domestic Area module

Default
Name of model parameters Unit Category
value
Number of households in 2000 (starting year) 11,784 houses
Market penetration rate 50 %
Efficiency 90 %
General
Incentive per home 1,000 USD/houses
Incentive per toilet 100 USD/houses
Incentive per washer 100 USD/houses
Toilet flush frequency 5 flush/p/day
Toilet water use built after 94 1.6 gal/flush
Toilet water use built pre 80 6 gal/flush
Toilet
Toilet water use built between 80 and 94 3.5 gal/flush
Number of households having toilet built pre 1994 10,314 houses
Number of households having toilet built pre 1980 0 houses
Showers per cap per day 0.9 shower/p/day
Shower water use built before 94 5 gal/min
Shower
Shower water use built after 94 2.5 gal/min
Shower time 8 min/shower
Faucet water use 2.5 gal/min
Faucets use per cap per day 4 min/p/day Faucet
Aerator saving 2.94 gal/p
Cooling season 2,500 hr/yr
Coolers per house 1 1/houses
Cooler water use with bleed off 8.1 gal/hr
Evaporative
Cooler water use without bleed off 4 gal/hr
coolers
Percent of cooler bleed off 20 %
Percent of cooler without bleed off 80 %
Percent of houses with evaporative cooler 90 %
Dish cycles per day 0.2 1/p/da Dish-
Water use per cycle of dish washer 10 gal Washer
Water use per bath 32.5 gal
Bathtub
Number of bath per day 0.143 1/p/da
Water use per front load clothes washer 42.3 gal
Clothes-
Water use per top load clothes washer 18.5 gal
washer
Number of cycle of clothes washer 0.3 1/p/da
Fountain filling frequency 4 1/yr
Percent of houses having fountain 1 %
Fountain
Fountain storage 150 gal
Number of fountain per house 1 1/houses
Rainfall collection area 2,000 ft2/houses
Rainwater
Rainfall collection rate 0.6 gal/ft2/in
harvest
Rainfall collection efficiency 50 %
 88

Table A.4. Parameters required for the domestic Area module (Continued)
Default
Name of model parameters Unit Category
value
Evaporation from a pool having a cover 18.79 gal/day/pools
Evaporation from a pool without a cover 38.1 gal/day/pools
Pool volume 16,830 gal/pools
Drain frequency of a pool having a cover 0.1 1/yr
Pool
Drain frequency of a pool without a cover 0.25 1/yr
Percent of drained water of pool reaching aquifer 90 %
Backwash amount of a pool 9.4 gal/day/pools
Percent of houses with swimming pool 9.2 %
2
Turf area per house 600 ft /houses
Base drip area per house 1,200 ft2/houses
Water use of drip irrigation system 0.91 af/acre/yr
Water use of turf area 3.65 af/acre/yr Irrigation
Percent of recharging of outdoor irrigation 0.5 %
Percent of houses with permanent irrigation system 75 %
Irrigation system efficiency 70 %
*gal – gallon, p – person, min – minute, hr – hour, af – acre-ft, USD – US Dollars
 89

Table A.5. Parameters required for riparian area module

Input Parameter Value Unit
Cottonwood 5 mm/day
Tamarix 8 mm/day
Average ET Rate Mesquite 5 mm/day
Grass 2 mm/day
Pan Evapotranspiration 2 mm/day
Default value
Total riparian area 1,610,000 m
Area
Total open channel & sand area 750,000 m2
Total extra-riparian area 20,000,000 m2
Tamarix 20 %
Riparian area
Cottonwood 80 %
Percent of each tree
Mesquite 15 %
Extra-riparian area
Grassland 85 %
 90

Table A.6. Lumped water quality module parameters

Water Removal efficiency (%)
Components loss
BOD TSS Hardness Giardia
(kafy)
Water treatment system 1 1.5 90 90 90 3.0log
Water treatment system 2 1.5 90 90 90 3.0log
Wastewater treatment system 1 1.5 90 90 90 2.0log
Wastewater treatment system 2 1.5 90 90 90 2.0log
Wastewater treatment system 3 4.0 90 90 90 2.0log
Advanced water treatment system 1 1.5 95 95 95 3.5log
Advanced water treatment system 2 1.5 95 95 95 3.5log
Advanced wastewater treatment system 1 4.0 95 95 95 2.5log
Advanced wastewater treatment system 2 1.5 95 95 95 2.5log
Advanced wastewater treatment system 3 1.5 95 95 95 2.5log
Recharge facility - 30 30 30 2.0log
BOD TSS Hardness Giardia
Components
(mg/l) (mg/l) (mg/l as CaCo3) (#/ml)
Initial water quality of precipitation 5 5 2 0
Initial water quality of imported water 30 30 150 100
Initial water quality of groundwater 30 30 250 0
Initial water quality of river 30 30 200 100
Initial water quality of reservoir 30 30 200 20
BOD TSS Hardness Giardia
Components
(mg/l) (mg/l) (mg/l as CaCo3) (#/ml)
Waste quality deterioration of Agricultural area 1 150 130 200 25
Waste quality deterioration of Agricultural area 2 150 130 200 25
Waste quality deterioration of Agricultural area 3 150 130 200 25
Waste quality deterioration of Agricultural area 4 150 130 200 25
Waste quality deterioration of Domestic area 1 200 180 10 50
Waste quality deterioration of Domestic area 2 200 180 10 50
Waste quality deterioration of Domestic area 3 200 180 10 50
Waste quality deterioration of Domestic area 4 200 180 10 50
Waste quality deterioration of Industrial area 250 200 100 10
Waste quality deterioration of Large outdoor area 1 150 130 120 10
Waste quality deterioration of Large outdoor area 2 150 130 120 10
 91

Table A.7. Parameters required for conventional water treatment plant module
Input Value Units
Alum 10 mg/l
Rotation speed for turbine impeller 100/60 rps
Impeller diameter 40 % of basin width ft
Water density 62.4 lb/ft3
Detention time of rapid mixing (DT) 30 sec
Velocity gradient in flocculation 50 1/sec
Detention time of flocculation (FDT) 30 min
Influent temperature 25 C
Dosage of chlorine 100 mg/l
Total organic carbon 20 mg/l
Detention time of sedimentation (SDT) 5 hrs
Depth of sedimentation basin (DEPTH) 12 ft
Percentage of solids in the sludge 75 %
Effective sand size in filtration 0.5 mm
Uniformity coefficient of sand 1.2
Depth of sand in filtration 30/12 ft
Backwash time in filtration 30 min
Time between backwash in filtration 24 hr
Maximum allowable average filtration rate (QAVE) 50 m/day
Influent aluminum hydroxide 10 mg/l
Influent turbidity 10 ntu
Price of dry alum 5 USD/lb
Liquid chlorine cost 20 USD/ton
Backwash material cost for filtration 5,265 USD/yr
Backwash labor cost for filtration 93.7 USD/yr
Pumping efficiency of backwash pumps 90 %
 92

Table A.8. Parameters required for conventional wastewater treatment plant

module
Parameter Value Units
Primary sedimentation:
Constant in Voshel-Sak Model 0.139 -
Constant in Voshel-Sak Model 0.27 -
Constant in Voshel-Sak Model 0.22 -
Sludge settling characteristics:
Thickening constant 24.2 -
Thickening constant 198.7 -
Thickening constant 2.5 -
Thickening constant 2.375 -
Thickening constant 2.803 -
Activated sludge kinetics:
Growth yield coefficient 0.4 g cell/g BOD5
Half-velocity constant 60 g BOD5/m3
Maximum specific utilization coefficient 5 day-1
Endogenous decay coefficient 0.04 day-1
Fraction of cells degradable 0.77 -
Conversion 1.42 g BODL/g VSS
Conversion 1.5 g BODL/g BOD5
Secondary sedimentation:
Constant in Chapman Model 5.69 -
Constant in Chapman Model 0.00403 -
Constant in Chapman Model 11.91 -
Aeration:
Alpha factor in aeration 0.8 -
Beta factor in aeration 0.95 -
DO concentration in aeration tank 1.5 g/m3
DO saturation concentration 9.17 g/m3
Temperature mixed liquor 20 o
C
Oxygen transfer efficiency 0.08 -
Density of air 1.2 Kg/m3
Weight fraction of oxygen in air 0.232 -
Temperature correction constant 1.024 -
Mixing Requirement 28.8 m3 air/m3/d
Gravity thickening:
TSS of thickener supernatant 200 g/m3
Anaerobic digestion:
Primary digestion reaction rate constant 0.632 -
Primary digestion reaction rate constant 3.003 -
Temperature of digester influent 20 o
C
Methane Production 0.35 m3/kg BODL
 93

Table A.8. Parameters required for conventional wastewater treatment plant

module (Continued)
Parameter Value Units
Average ambient temperature 10 o
C
Efficiency of heat exchanger 0.85 -
Heat conduction coefficient 1 W/m2-oC
Outside surface area and volume ration for digester 0.4 -
Worth of digester gas 2.5 $/106 kl
Soluble BOD5 in digester supernatant 500 g/m3
Factor accounting for the effect of rising gas on
thickening in SD 0.25 -
Thickening constant for digested sludge 292.6 -
Thickening constant for digested sludge 2.9 -
TSS of digester supernatant 4,000 g/m3
Height of digester 10 m
Vacuum filtration:
Form time per cycle time 0.33 -
Pressure applied on vacuum filter 83,300 Nt/m2
Viscosity of filtrate 0.00089 Nt-sec/m2
Cycle time 6 min
Specific resistance of sludge 1.00E+12 m/kg
TSS of filtrate 2,000 g/m3
 94

Table A.9. Parameters required for population modules

Initial population Initial households Initial Projected
(person) (houses) growth rate (%) growth rate (%)
Domestic area 1 100,000 36,030 2.5 1.5
Domestic area 2 110,000 41,030 2.5 1.5
Domestic area 3 120,000 51,030 2.5 1.5
Domestic area 4 130,000 66,030 2.5 1.5
Initial number of Water use rate Initial growth Projected growth
business (afy/business) rate (%) rate (%)
Industrial area 931 1.844 2.5 1.5
 95

Table A.10. Network geometry data of hypothetical water supply system

Elevation
Departure/Destination Length (mi) Diameter (in) Conveyance
(ft)
From imported water (2,800 ft)
Water treatment plant 1 2,000 3 216 Canal
Water treatment plant 2 2,100 3 216 Canal
Agricultural area 1 2,000 10 240 Canal
Agricultural area 2 2,000 3 240 Canal
Agricultural area 3 2,400 4 240 Canal
Agricultural area 4 2,400 3 240 Canal
Recharge facility 0 5 158 Canal
From river (2,100 ft)
Water treatment plant 1 2,000 3 12 Canal
Water treatment plant 2 2,100 3 60 Canal
Agricultural area 1 2,000 7 12 Canal
Agricultural area 2 2,000 10 24 Canal
Agricultural area 3 2,400 10 12 Canal
Agricultural area 4 2,400 10 12 Canal
From reservoir (5 ft)
Water treatment plant 2 2,100 3 12 Canal
Agricultural area 1 2,000 7 12 Canal
Agricultural area 2 2,000 7 24 Canal
Agricultural area 3 2,400 7 24 Canal
Agricultural area 4 2,400 7 24 Canal
From groundwater (300 ft)
Water treatment plant 1 2,000 0.02 60 Canal
Water treatment plant 2 2,100 0.02 60 Canal
Agricultural area 1 2,000 Pumping
Agricultural area 2 2,000 Pumping
Agricultural area 3 2,400 Pumping
Agricultural area 4 2,400 Pumping
Large outdoor area 1 1,700 0.02 60 Canal
Large outdoor area 2 1,700 0.02 60 Canal
Domestic area 1 1,800 0.02 60 Canal
Domestic area 2 2,600 0.02 60 Canal
Domestic area 3 2,200 0.02 60 Canal
Domestic area 4 2,600 0.02 60 Canal
Industrial 1,800 0.02 60 Canal
From water treatment plant 1 (2,000 ft)
Agricultural area 1 2,000 5 60 Pipe
Agricultural area 2 2,000 20 Decision Alternative flow
Agricultural area 3 2,400 20 Decision Alternative flow
Agricultural area 4 2,400 20 Decision Alternative flow
Large outdoor area 1 1,700 5 60 Pipe
Large outdoor area 2 1,700 20 Decision Alternative flow
Domestic area 1 1,800 20 Decision Alternative flow
Domestic area 2 2,600 20 Decision Alternative flow
Domestic area 3 2,200 20 Decision Alternative flow
Domestic area 4 2,600 20 Decision Alternative flow
Industrial 1,800 4 60 Pipe
 96

Table A.10. Network geometry data of hypothetical water supply system (Continued)
Elevation
Departure/Destination Length (mi) Diameter (in) Conveyance
(ft)
From water treatment plant 2 (2,100 ft)
Agricultural area 1 2,000 20 Decision Alternative flow
Agricultural area 2 2,000 5 60 Pipe
Agricultural area 3 2,400 5 60 Pipe
Agricultural area 4 2,400 5 60 Pipe
Large outdoor area 1 1,700 20 Decision Alternative flow
Large outdoor area 2 1,700 5 60 Pipe
Domestic area 1 1,800 5 60 Pipe
Domestic area 2 2,600 5 60 Pipe
Domestic area 3 2,200 5 60 Pipe
Domestic area 4 2,600 5 60 Pipe
Industrial 1,800 20 60 Pipe
From agricultural area (2,000 ft)
Riparian area 0 5 12 Canal
From agricultural area 2 (2,000 ft)
Riparian area 0 5 12 Canal
From agricultural area 3 (2,400 ft)
Riparian area 0 5 12 Canal
From agricultural area 4 (2,400 ft)
Riparian area 0 5 12 Canal
From large outdoor area 1 (1,700 ft)
Riparian area 0 3 12 Canal
From large outdoor area 2 (1,700 ft)
Riparian area 0 3 12 Canal
From domestic area 1 (1,800 ft)
Wastewater treatment plant 1 2,100 20 Decision Alternative flow
Wastewater treatment plant 2 1,700 2 60 Pipe
Wastewater treatment plant 3 2,500 20 Decision Alternative flow
From domestic area 2 (2,600 ft)
Wastewater treatment plant 1 2,100 20 Decision Alternative flow
Wastewater treatment plant 2 1,700 5 60 Pipe
Wastewater treatment plant 3 2,500 5 60 Pipe
From domestic area 3 (2,200 ft)
Wastewater treatment plant 1 2,100 5 60 Pipe
Wastewater treatment plant 2 1,700 20 Decision Alternative flow
Wastewater treatment plant 3 2,500 5 60 Pipe
From domestic area 4 (2,600 ft)
Wastewater treatment plant 1 2,100 5 60 Pipe
Wastewater treatment plant 2 1,700 20 Decision Alternative flow
Wastewater treatment plant 3 2,500 8 60 Pipe
From industrial area (1,800 ft)
Wastewater treatment plant 1 2,100 3 60 Pipe
Wastewater treatment plant 2 1,700 20 Decision Alternative flow
Wastewater treatment plant 3 2,500 20 Decision Alternative flow
 97

Table A.10. Network geometry data of hypothetical water supply system (Continued)
Elevation
Departure/Destination Length (mi) Diameter (in) Conveyance
(ft)
From wastewater treatment plant 1 (2,100 ft)
Riparian area 0 9 12 Canal
Agricultural area 1 2,000 5 60 Pipe
Agricultural area 2 2,000 5 60 Pipe
Agricultural area 3 2,400 20 Decision Alternative flow
Agricultural area 4 2,400 20 Decision Alternative flow
Large outdoor area 1 1,700 5 60 Pipe
Large outdoor area 2 1,700 20 Decision Alternative flow
Recharge facility 0 8 12 Pipe
From wastewater treatment plant 2 (1,700 ft)
Riparian area 0 9 12 Canal
Agricultural area 1 2,000 20 Decision Alternative flow
Agricultural area 2 2,000 5 60 Pipe
Agricultural area 3 2,400 20 Decision Alternative flow
Agricultural area 4 2,400 20 Decision Alternative flow
Large outdoor area 1 1,700 20 Decision Alternative flow
Large outdoor area 2 1,700 5 60 Pipe
Recharge facility 0 8 12 Pipe
From wastewater treatment plant 3 (2,500 ft)
Riparian area 0 9 12 Canal
Agricultural area 1 2,000 5 60 Pipe
Agricultural area 2 2,000 5 60 Pipe
Agricultural area 3 2,400 5 60 Pipe
Agricultural area 4 2,400 5 60 Pipe
Large outdoor area 1 1,700 20 Decision Alternative flow
Large outdoor area 2 1,700 20 Decision Alternative flow
Recharge facility 0 8 12 Pipe
 98

Table A.11. Conservation measures for domestic/industrial use

Users Programs Objective systems
Domestic Incentive Front load washing machine for existing and new houses
indoor programs Faucet, Shower and Toilet for houses built before 1994
Evaporation cooler and fountain for existing and new houses
Incentive Incentives to purchase pool covers for existing and new houses
programs Water irrigation efficiency increasing for existing and new houses
Grey water reuse system for new houses
Reduced swimming pool use – Public education
Restrict future swimming pool development
Eliminate existing swimming pools
Domestic
Discharge pool water for eventual reuse for existing and new swimming
outdoor
pools
Ordinances Outdoor water use restriction for existing and new houses
Landscaping standards and regulation for existing houses
Landscaping standards and regulation for new houses
Rainwater harvesting for new houses
Rainwater harvesting for existing houses
Water loss due to violation
Incentive
Incentives to purchase pool covers for existing and new houses
program
Toilet for existing and new businesses
Reduced swimming pool use – Public education
Restrict future swimming pool development
Industrial Eliminate existing swimming pools
Ordinances Discharge pool water for eventual reuse for existing and new swimming
pools
Outdoor water use restriction for existing and new houses
Water irrigation efficiency increasing for existing and new houses
Large water user audits for existing and new businesses
Population Growth restriction
 99

Table A.12. Parameter values before and after a conservation measure

implementation for domestic/industrial uses
Before After
Alternatives Other parameters
implementation implementation
Population growth rate 2.5 % 1.5 %
Faucet 10 gal/day/p 7.06 gal/day/p
Clothes washers 42.3 gal 18.5 gal 0.3 cycle/p/day
8 min/shower
Shower 5 gal/min 2.5 gal/min
0.9 shower/p/day
Toilet 3.5 gal/flush 1.6 gal/flush 5 flush/p/day
Evaporative cooler 2,500 hr cooling season/yr
4 gal/hr 1.2 gal/hr
(without bleed off) 1 cooler/house
90% cooler/houses
Evaporative cooler 80% no bleed off/coolers
8.1 gal/hr 2.43 gal/hr
(with bleed off) 20% bleed off/coolers
4 fills/yr, 1 fountain/house
Fountains 150 gal 45 gal/hr
1% fountain/houses
Swimming pool evaporation with Swimming pools with
38.1 gal/day/pools 18.79 gal/day/pools
cover and without cover cover: 50%
Reduced swimming pool use –
50 % pools with cover 80% pools with cover
Public education
Houses percentage with swimming
9.2% 0%
pool
Discharge swimming pools to Recharge rate to
0% 100 %
French drain groundwater: 80 %
Water drip irrigation efficiency 75 % 90 %
Water demand for irrigation Turf: 3.3 afy/acre Drip: 0.91 afy/acre
Landscaping standards and No more than 10 % of
No regulation
regulation irrigated turf
2,000 ft2/house
Rainfall harvesting No harvesting Efficiency: 50 %
0.6 gal/ft2/in
Grey water reuse system Reuse system cost = $500/houses
Large water user audits 0 af/yr 22 af/yr
21,400 gal/violations
Water loss due to violation – 1 violations/house/yr 0 violation
Domestic outdoor
2603
21,400 gal/violations
Violations in industrial area 4 violations/business/yr 0 violation
144
<<Commercial area>> Water use for commercial car = 10 gal/car
Number of washes of commercial car = 0 /day
Car wash
<<Domestic area>>water use for a car = 15 gal/car
Number of washes of commercial car = 60 /day
 100

Table A.13. Government subsidy for conservation incentives

Government Cost
Alternatives
investment ($/yr) ($/house)
Front load washing machine for existing and new houses 100,000 70
Shower, toilet, and faucet for houses built before 1994 100,000 100
Grey water reuse system for new houses - 200
Evaporation cooler and fountain for existing and new houses –
100,000 70
domestic and industrial area
Incentives to purchase pool covers 100,000 70

Table A.14. Cost and water savings next 8 years (2012-2020) resulting when
individual conservation measures are implemented in year 2012 in domestic area 4
Fresh
Operation
water Operation Fresh
cost in
use in cost water use
Alternatives domestic
domestic reduction reduction
area
area (%) (%)
($/yr x 105)
(km3/yr)
Base condition 0.054 18.93
Front load clothes washer for existing houses 0.052 18.32 3.25 3.59
Shower, toilet, and faucet for existing houses 0.054 18.79 0.76 1.10
Evaporative cooler and fountain for existing houses 0.052 18.27 3.48 3.82
Grey water reuse system for new houses 0.046 17.24 8.95 15.07
Reduced swimming pool use for existing and new
0.054 18.87 0.33 0.33
houses
Reduced incentives to purchase pool covers and
0.054 18.83 0.55 0.55
splash recovery system for existing and new houses
Reduced restrict future swimming pool development
0.054 18.81 0.64 0.64
for existing and new houses
Reduced eliminate exiting swimming pools 0.054 18.91 0.09 0.09
Reduced discharge pool water for eventual reuse 0.054 18.93 0.00 0.00
Water drip irrigation efficiency for existing and new
0.054 18.92 0.06 0.41
houses
Landscaping standards and regulations for
0.049 17.18 9.23 9.23
existing houses
Landscaping standards and regulations for new users
- It is turned on if the same regulation for existing 0.053 18.34 3.11 3.11
houses is on.
Outdoor water use restriction 0.054 18.93 0.00 0.00
Water loss due to violations 0.054 18.91 0.11 0.11
Rainwater harvesting for new houses 0.054 18.77 0.86 0.86
Rainwater harvesting for existing houses 0.053 18.45 2.55 2.55
 101

Table A.15. Cost and water savings next 8 years (2012-2020) resulting when
individual conservation measures are implemented in year 2012 in industrial area
Fresh
Operation Fresh
water use Operation
cost in water
in cost
Conservation measure industrial use
industrial Savings
area Savings
area (%)
($/yr x 10 5) (%)
(kafy)
Base condition 1.99 4.63
Toilet for existing and new houses 1.98 4.62 0.32 0.32
Reduced swimming pool use
1.99 4.63 0.09 0.09
for existing and new houses
Reduced incentives to purchase pool covers
and splash recovery system 1.99 4.63 0.15 0.15
for existing and new houses
Reduced restrict future swimming pool
1.99 4.63 0.00 0.00
development for existing and new houses
Reduced eliminate exiting swimming pools 1.98 4.62 0.38 0.38
Reduced discharge pool water for eventual reuse 1.99 4.63 0.00 0.00
Water drip irrigation efficiency
1.98 4.61 0.49 0.49
for existing and new houses
Outdoor water use restriction 1.98 4.61 0.49 0.49
Large water user audits 1.98 4.61 0.37 0.48

Table A.16. Consumptive use in the domestic areas and groundwater storage
Domestic area/ No conservation All conservation
Difference (%)
groundwater storage measures measures implementing
Domestic area 1 (km3/yr) 0.013 0.002 -84.03
Domestic area 2 (km3/yr) 0.016 0.003 -80.81
Domestic area 3 (km3/yr) 0.018 0.004 -77.05
Domestic area 4 (km3/yr) 0.024 0.007 -70.76
Groundwater storage (km3) 27.44 27.69 0.89
 102

Table A.17. Groundwater storage after 20-year simulation resulting under

alternative scenarios of water availability
Storage after 20-year Difference
Alternatives
(kaf) (%)
Base condition 6,299
No imported water 5,823 7.56
No reclaimed water 5,773 8.35
No imported water and reclaimed water 5,122. 18.68

Table A.18. Water and wastewater constructed under different conditions

Number of treatment
Condition Water plant built Wastewater plant built
plant
Base condition 1, 2 1, 2, 3 5
Condition 1 2 1, 2, 3 4
Condition 2 1 1, 2, 3 4
Condition 3 1, 2 2, 3 4
Condition 4 1, 2 1, 3 4
Condition 5 1, 2 1, 2 4
Condition 6 1, 2 3 3
Condition 7 1, 2 1 3
Condition 8 1, 2 2 3
 103

Table A.19. Capacity over time of wastewater treatment plants using lumped
representation
Number First period Second period Third period Fourth period
Condition
of Plants (km3/yr) (km3/yr) (km3/yr) (km3/yr)
Base 0.026, 0.032, 0.032, 0.039, 0.040, 0.048, 0.050, 0.060,
1,2,3
condition 0.031 0.038 0.047 0.058
Condition
0.089 0.109 0.135 0.168
3, 4, 5
1,2 0.044, 0.045 0.054, 0.055 0.067, 0.067 0.084, 0.084
Condition
2,3 0.032, 0.057 0.039, 0.070 0.048, 0.086 0.060, 0.108
6, 7, 8 1,3 0.044, 0.045 0.054, 0.055 0.067, 0.067 0.084, 0.084

Table A.20. Unit process capacity in conventional wastewater treatment system

representation for each five year period
Time Three treatment system Two treatment system Single
perio (base condition) Condition 3 Condition 4 Condition 5 treatment
d Plant 1 Plant 2 Plant 3 Plant 2 Plant 3 Plant 1 Plant 3 Plant 1 Plant 2 system
Surface area of primary clarifier Ap (m2)
1 684 835 802 835 1,490 1,156 1,169 1,155 1,169 2,328
2 1,205 1,460 1,407 1,460 2,616 2,029 2,047 2,032 2,044 4,081
3 1,496 1,801 1,740 1,801 3,241 2,512 2,530 2,519 2,523 5,047
4 1,871 2,239 2,167 2,239 4,043 3,134 3,148 3,146 3,136 6,287
Volume of aeration tank V (m3)
1 9,844 12,018 11,549 12,018 21,445 16,638 16,825 16,632 16,831 33,516
2 17,339 21,011 20,247 21,011 37,661 29,206 29,466 29,246 29,426 58,747
3 21,531 25,928 25,046 25,928 46,652 36,164 36,417 36,265 36,316 72,655
4 26,934 32,224 31,189 32,224 58,198 45,107 45,314 45,286 45,135 90,496
Surface area of secondary clarifier Af (m2)
1 1,435 1,752 1,683 1,752 3,127 2,426 2,453 2,425 2,454 4,888
2 2,527 3,063 2,952 3,063 5,492 4,259 4,296 4,264 4,291 8,568
3 3,139 3,780 3,652 3,780 6,804 5,274 5,310 5,288 5,296 10,597
4 3,927 4,699 4,548 4,699 8,488 6,578 6,608 6,604 6,582 13,199
Surface area of thickener Ag (m2)
1 1,237 1,506 1,448 1,506 2,670 2,076 2,099 2,075 2,100 4,160
2 2,172 2,625 2,531 2,625 4,681 3,637 3,669 3,642 3,664 7,284
3 2,690 3,232 3,124 3,232 5,791 4,496 4,527 4,509 4,515 9,001
4 3,357 4,010 3,882 4,010 7,216 5,600 5,626 5,622 5,604 11,204
Volume of primary digester Vd (m3)
1 3,725 4,533 4,358 4,533 8,032 6,247 6,317 6,245 6,319 12,512
2 6,538 7,901 7,618 7,901 14,081 10,943 11,039 10,958 11,025 21,908
3 8,094 9,726 9,399 9,726 17,419 13,526 13,619 13,563 13,582 27,070
4 10,100 12,063 11,679 12,063 21,704 16,845 16,922 16,911 16,856 33,692
Surface area of secondary digester Ad (m2)
1 79 99 94 99 181 139 141 139 141 287
2 142 174 168 174 321 246 249 247 248 506
3 179 218 210 218 400 308 310 308 309 628
4 226 273 264 273 501 386 388 388 386 785
Surface area of vacuum filter (m2)
1 163 198 191 198 349 272 275 272 275 542
2 286 345 332 345 611 476 480 476 479 948
3 353 423 409 423 755 587 591 589 590 1,171
4 439 524 508 524 940 730 734 733 731 1,457
 104

Table A.21. Total cost in the present value and annual cost in decentralized
treatment alternatives
Lumped treatment plant representation
Total cost in
Total cost in the Differences
Alternatives annual cost
present value (x 106$) (%)
(x 106$/yr)
Base condition
4,266 278.8 -
(2 water and 3 wastewater system)
Condition 1 5,492 329.0 11.79
Condition 2 7,325 478.8 24.39
Condition 3 4,932 323.5 10.25
Condition 4 4,539 312.8 6.61
Condition 5 4,379 324.4 10.55
Condition 6 4,717 314.6 7.22
Condition 7 5,009 334.0 13.83
Condition 8 5,341 357.9 21.98
Conventional treatment system representation
Total cost in the Total cost in
Differences
Alternatives present value annual cost
(%)
(x 106$) (x 106$/yr)
Base condition
3,707 242.4 -
(2 water and 3 wastewater system)
Condition 1 4,973 325.1 34.15
Condition 2 6,618 432.6 78.51
Condition 3 4,449 290.8 20.00
Condition 4 4,050 264.8 9.27
Condition 5 3,894 254.5 5.03
Condition 6 4,330 283.0 16.80
Condition 7 4,626 302.4 24.77
Condition 8 4,955 323.9 33.65
 105

Table A.22. Annual construction, expansion, and O&M cost for each conveyance
and treatment system (× 106$/yr)
Pipe Pump
Alternatives
Construction Expansion O&M Construction Expansion O&M
Base condition 10.77 24.85 65.24 3.62 5.03 9.50
Condition 1 19.29 44.52 116.81 3.65 5.15 10.16
Condition 2 33.26 76.77 201.44 3.62 5.05 10.53
Condition 3 17.64 40.71 94.37 3.59 4.99 9.68
Condition 4 13.96 32.23 78.34 3.58 4.95 9.45
Condition 5 13.59 31.37 72.98 3.63 4.78 9.19
Condition 6 17.85 41.20 88.01 3.55 4.91 9.72
Condition 7 20.92 48.28 97.21 3.64 4.79 9.43
Condition 8 24.06 55.54 106.89 3.69 4.80 9.49
Canal Recharge Facility
Alternatives
Construction Expansion O&M Construction Expansion O&M
Base condition 0.39 1.11 1.59 0.01 0.18 0.03
Condition 1 0.36 1.02 1.53 0.01 0.20 0.03
Condition 2 0.36 1.01 1.53 0.01 0.20 0.03
Condition 3 0.39 1.11 1.51 0.01 0.23 0.04
Condition 4 0.39 1.11 1.50 0.01 0.21 0.03
Condition 5 0.39 1.11 1.50 0.01 0.28 0.04
Condition 6 0.39 1.11 1.42 0.01 0.26 0.04
Condition 7 0.39 1.11 1.41 0.01 0.31 0.05
Condition 8 0.39 1.11 1.42 0.01 0.30 0.05
Lumped system Conventional treatment system
Water treatment system 1 Water treatment system 1
Alternatives
Construction Expansion O&M Construction Expansion O&M
Base condition 0.72 0.00 0.58 0.01 0.00 2.08
Condition 1 0.00 0.00 0.00 0.00 0.00 0.00
Condition 2 3.76 9.02 3.69 0.01 0.03 10.43
Condition 3 0.72 0.00 0.58 0.01 0.00 2.08
Condition 4 0.72 0.00 0.58 0.01 0.00 2.08
Condition 5 0.72 0.00 0.58 0.01 0.00 2.08
Condition 6 0.72 0.00 0.58 0.01 0.00 2.08
Condition 7 0.72 0.00 0.58 0.01 0.00 2.08
Condition 8 0.72 0.00 0.58 0.01 0.00 2.08
Lumped system Conventional treatment system
Water treatment system 2 Water treatment system 2
alternatives
Construction Expansion O&M Construction Expansion O&M
Base condition 2.14 5.53 2.14 0.01 0.03 9.07
Condition 1 2.76 5.53 2.60 0.01 0.03 10.43
Condition 2 0.00 0.00 0.00 0.00 0.00 0.00
Condition 3 2.14 5.53 2.14 0.01 0.03 9.07
Condition 4 2.14 5.53 2.14 0.01 0.03 9.07
Condition 5 2.14 5.53 2.14 0.01 0.03 9.07
Condition 6 2.14 5.53 2.14 0.01 0.03 9.07
Condition 7 2.14 5.53 2.14 0.01 0.03 9.07
Condition 8 2.14 5.53 2.14 0.01 0.03 9.07
 106

Table A.22. Annual construction, expansion, and O&M cost for each conveyance
and treatment system (× 106$/yr) (Continued)
Lumped system Conventional treatment system
Wastewater treatment system 1 Wastewater treatment system 1
alternatives
Construction Expansion O&M Construction Expansion O&M
Base condition 2.62 2.59 7.13 0.29 0.74 0.34
Condition 1 2.62 2.59 7.13 0.29 0.74 0.34
Condition 2 2.62 2.59 7.13 0.29 0.74 0.34
Condition 3 0.00 0.00 0.00 0.00 0.00 0.00
Condition 4 3.69 3.64 9.95 0.41 1.05 0.47
Condition 5 3.69 3.65 10.00 0.41 1.05 0.47
Condition 6 0.00 0.00 0.00 0.00 0.00 0.00
Condition 7 5.85 5.77 15.68 0.68 1.69 0.78
Condition 8 0.00 0.00 0.00 0.00 0.00 0.00
Lumped system Conventional treatment system
Wastewater treatment system 2 Wastewater treatment system 2
alternatives
Construction Expansion O&M Construction Expansion O&M
Base condition 2.98 2.93 7.91 0.33 0.83 0.38
Condition 1 2.98 2.93 7.91 0.33 0.83 0.38
Condition 2 2.98 2.93 7.91 0.33 0.83 0.38
Condition 3 2.98 2.93 7.91 0.33 0.83 0.38
Condition 4 0.00 0.00 0.00 0.00 0.00 0.00
Condition 5 3.72 3.66 9.87 0.42 1.04 0.47
Condition 6 0.00 0.00 0.00 0.00 0.00 0.00
Condition 7 0.00 0.00 0.00 0.00 0.00 0.00
Condition 8 5.85 5.77 15.68 0.68 1.69 0.78
Lumped system Conventional treatment system
Wastewater treatment system 3 Wastewater treatment system 3
alternatives
Construction Expansion O&M Construction Expansion O&M
Base condition 2.91 2.86 7.78 0.32 0.81 0.37
Condition 1 2.91 2.86 7.78 0.32 0.81 0.37
Condition 2 2.91 2.86 7.78 0.32 0.82 0.37
Condition 3 4.36 4.31 11.77 0.50 1.25 0.57
Condition 4 3.72 3.66 9.93 0.42 1.05 0.48
Condition 5 0.00 0.00 0.00 0.00 0.00 0.00
Condition 6 5.85 5.77 15.68 0.68 1.69 0.78
Condition 7 0.00 0.00 0.00 0.00 0.00 0.00
Condition 8 0.00 0.00 0.00 0.00 0.00 0.00
 107

Table A.23. Removal efficiency (%) of different water and wastewater treatment
system representations
Conventional treatment plant system representation
Plants BOD TSS Hardness Giardia
Water treatment plant 1 90.0 99.3 90.0 99.9
Water treatment plant 2 90.0 99.2 90.0 99.9
Wastewater treatment plant 1 92.1 94.8 90.0 99.0
Wastewater treatment plant 2 91.8 94.7 90.0 99.0
Wastewater treatment plant 3 91.8 94.7 90.0 99.0
Lumped treatment system representation
Plants BOD TSS Hardness Giardia
Water treatment plant 1 90.0 90.0 90.0 99.9
Water treatment plant 2 90.0 90.0 90.0 99.9
Wastewater treatment plant 1 90.0 90.0 90.0 99.0
Wastewater treatment plant 2 90.0 90.0 90.0 99.0
Wastewater treatment plant 3 90.0 90.0 90.0 99.0
 108

8. FIGURES

PRECIPITATION

Consumptive On-site
Domestic area Industrial area
Use treatment plant
Large outdoor area

Advanced water
Wastewater
treatment plant
treatment plant
IMPORTED WATER
Agricultural area

Advanced
Water treatment plant wastewater
treatment plant

Recharge Facility

GROUNDWATER
RESERVOIR

RIVER
Environmental/
Recreational area

Figure A.1. Schematic of a generic water supply system. Solid lines indicate supply
flows and dashed lines indicate effluent flows. All components (uses and supplies)
have a consumptive use loss.
 109

Figure A.2. Water balance for four domestic areas

Figure A.3. Powersim representation of a conventional water treatment system

 110

Figure A.4. Powersim representation of a conventional wastewater treatment system

 111

5
Rainfall (inches)

Jan-87 Jan-89 Jan-91 Jan-93 Jan-95 Jan-97 Jan-99 Jan-01 Jan-03 Jan-05

Figure A.5 Monthly rainfall for Coolidge, AZ (Jan. 1987 – Dec. 2004) taken from the
Arizona Meteorological Network (AZMET) (http://ag.arizona.edu/azmet/.html)

16000

12000
Natural streamflow (cfs)

8000

4000

Jan-87 Jan-89 Jan-91 Jan-93 Jan-95 Jan-97 Jan-99 Jan-01 Jan-03

Figure A.6. Salt River, AZ at USGS 09498500 monthly streamflows (Jan. 1987 – Sep.
2004) from NWISWeb data (http://waterdata.usgs.gov/nwis)
 112

25000
Reservoir inflow
Reservoir outflow

20000
flowrate (cfs)

15000

10000

5000

Jan-87 Jan-89 Jan-91 Jan-93 Jan-95 Jan-97 Jan-99 Jan-01 Jan-03

Figure A.7. Salt River, AZ monthly inflows and outflows at Roosevelt Dam in the
site of USGS 09497500 and USGS 09502000 (Jan. 1987-Sep. 2004) from NWISWeb
(http://waterdata.usgs.gov/nwis)
 113

35.0

base condition
30.0 gray w ater reuse system
clothes w asher
25.0
Water Demand (kafy)

20.0

15.0

10.0

5.0

0.0
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020

Year

Figure A.8. Water use in domestic Area 1 with no conservation measures, with
implementation of a grey water reuse and a fixture replacement incentive programs
 114

50

DO1_base
45 DO2_base
DO3_base
40 DO4_base
DO1_on all programs
Water demand (kafy)

DO2_on all programs

35
DO3_on all programs
DO4_on all programs
30

25

20

15

10
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Year

Figure A.9. Comparison of water use in four domestic areas for no conservation
measures (base) and after implementing all conservation measures (DO – Domestic
Area) in 2012. The variability in use after implementation is related to the climatic
conditions.
 115

4.0E+09

3.5E+09

3.0E+09 Base condition

Conservation measures

2.5E+09
Total cost ($)

2.0E+09

1.5E+09

1.0E+09

5.0E+08

0.0E+00
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Year

Figure A.10. Cost for constructing and operating entire system over time for no
conservation measures (base) and after implementing all conservation measures in
year 2012

Figure A.11. Inflows from various sources to agricultural area 1 during growing
season for meeting consumptive use with reclaimed water and imported water
 116

9000.0

8000.0

7000.0
Groundwater storage (kaf)

6000.0

5000.0
base
no IW
4000.0
no RW
3000.0 no IW&RW

2000.0

1000.0

0.0
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020

Year

Figure A.12. Groundwater storage change with different water sources being
unavailable (IW and RW – imported water and reclaimed water, respectively)

Figure A.13. Powersim representation for the water supply system with distributed
wastewater treatment plants
 117

300
CWT1
Water treatment system capacity (kafy)

CWT2
250 Single Capacity

200

150

100

50

0
2000-2005 2006-2010 2011-2015 2016-2020
Year

Figure A.14. Water treatment plant capacities for alternative configurations (CWT1
and CWT2 - central water treatment plant 1 and 2, respectively.)
 118

Watewater treatment system capacity (kafy) 160.0

CWWT1
140.0
CWWT2
120.0 CWWT3
Single Capacity
100.0

80.0

60.0

40.0

20.0

0.0
2000-2005 2006-2010 2011-2015 2016-2020
Year

Figure A.15. Capacity of wastewater treatment plants for plant configurations using
lumped treatment plant representation (CWWT1, CWWT2, CWWT3 are
wastewater treatment plant 1, 2, and 3, respectively.)
 119

35,000.00
Primary Clarifier
capacity - surface area of unit operations (m )
Conventional wastewater treatment system 2
2
Aeration Tank
30,000.00 Secondary Clarifier
Thickener
25,000.00 Primary Digester
Secondary Digester
20,000.00 Vacuum Filter

15,000.00

10,000.00

5,000.00

-
2000-2005 2006-2010 2011-2015 2016-2020
Year

Figure A.16. Unit operation capacities over time for a single centralized
conventional wastewater treatment plant
 120

7.0E+09
base condition
CWT1 only
6.0E+09 CWT2 only
CWWT1 only
CWWT2 only
5.0E+09
CWWT3 only
CWWT1 and CWWT2
Total cost ($)

4.0E+09 CWWT2 and CWWT3

CWWT1 and CWWT3

3.0E+09

2.0E+09

1.0E+09

0.0E+00
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Year

(a)
8.0E+09
base condition
CWT1 only
7.0E+09 CWT2 only
CWWT1 only
6.0E+09 CWWT2 only
CWWT3 only
5.0E+09 CWWT1 and CWWT2
Total cost ($)

CWWT2 and CWWT3

4.0E+09 CWWT1 and CWWT3

3.0E+09

2.0E+09

1.0E+09

0.0E+00
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Year

(b)
Figure A.17. Cost of water supply system over time for different treatment plant
configurations ((a) lumped representation (b) conventional water treatment system)
 121

35.00

33.00

31.00

29.00
Water quality (mg/l)

27.00

25.00
TSS in river
23.00 TSS in reservoir
TSS in groundwater
BOD in river
21.00
BOD in reservoir
BOD in groundwater
19.00

17.00

15.00
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Year

Figure A.18. Water quality (BOD and TSS) over time in the river, aquifer and
surface reservoir
 122

300.00

250.00
Water quality (mg/l or #/ml)

200.00

150.00

100.00

50.00

0.00
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Year
hardness in river hardness in reservoir hardness in groundwater
Giardia in river Giardia in reservoir Giardia in groundwater

Figure A.19. Water quality (hardness and Giardia) over time in the river, aquifer
and surface reservoir
 123

300.0

280.0

260.0

240.0
Water quality (mg/l)

220.0

200.0

180.0

160.0

140.0

120.0
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Year

TSS in agricultural area TSS in large outdoor area TSS in domestic area TSS in industrial area
BOD in agricultural area BOD in large outdoor area BOD in domestic area BOD in industrial area

Figure A.20. Water quality (BOD and TSS) over time supplied to various users
 124

450.000

400.000

350.000
Water quality (mg/l as CaCo3)

300.000

250.000

200.000

150.000

100.000
hardness in agricultural area
hardness in large outdoor area
50.000
hardness in domestic area
hardness in industrial area
0.000
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Year

Figure A.21. Water quality (hardness) over time supplied to various users
 125

90.00

80.00

70.00

60.00
Water quality (#/ml)

50.00

40.00

Giardia in agricultural area

30.00
Giardia in large outdoor area
Giardia in domestic area
20.00
Giardia in industrial area

10.00

0.00
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Year

Figure A.22. Water quality (Giardia) over time supplied to various users
 126

APPENDIX B. APPLICATION OF THE SHUFFLED FROG

LEAPING ALGORITHM FOR THE OPTIMIZATION OF A

GENERAL LARGE-SCALE WATER SUPPLY SYSTEM

B. Graph
 127

Application of the Shuffled Frog Leaping Algorithm for the Optimization

of a General Large-Scale Water Supply System

G. Chung1 and K. Lansey2

ABSTRACT

A water supply system is a complicated network of pipes, canals and storage and

treatment facilities that collects, treats, stores, and distributes water to consumers.

Increasing population and its associated demands requires systems to be expanded and

adapted over time to provide a sustainable water supply. Comprehensive design tools are

needed to assist managers determine how to plan for future growth. In this study, a

general large-scale water supply system model was developed to minimize the total

system cost by integrating a mathematical supply system representation and applying the

shuffled frog leaping algorithm optimization scheme (SFLA).

The developed model was applied to two hypothetical water communities. The

capacities for the system components including water transport and treatment facilities

are model decision variables. An explicit representation of energy consumption cost for

the transporting water in the model assists in determining the efficacy of satellite

wastewater treatment facilities.

1
Graduate Student, Department of Civil Engineering and Engineering Mechanics, The University of Arizona, Tucson,
AZ 85721, USA (Tel: 1-520-360-9554, E-mail: gunhui@email.arizona.edu)
2
Professor, Department of Civil Engineering and Engineering Mechanics, The University of Arizona, Tucson, AZ
85721, USA (Tel: 1-520-621-2512, Fax: 1-520-621-2550, E-mail: lansey@engr.arizona.edu)
 128

Although the water supply systems studied contained highly nonlinear terms in the

formulation as well as several hundred decisions variables, the stochastic search

algorithm, SFLA, successfully found solutions that satisfied all the constraints for the

studied networks.
 129

1. INTRODUCTION AND BACKGROUND

A water supply system is a collection of water transport structures, pumping stations,

and water treatment and storage facilities that are managed to supply the desired amount

of water with the desired quality to consumers. With increasing water demand from

domestic and industrial areas, sustainable water supply becomes more important and the

development of a long-term water supply plan is challenging because of complexity of

the system and uncertainties in the future.

In the southwest Unites States and many arid and semi-arid regions, groundwater is a

major water source but, oftentimes, it has been mined to meet the increased water

demands. As a result, ground water table level have fallen requiring better management

plans or identifying alternative water sources. Reclaimed and imported water are

potential alternatives that may replace groundwater for agricultural and other purposes. In

the future as new supplies become more limited, water reclaimed, after a high level of

treatment, may be necessary to meet potable demands.

Water supply planning requires considering current demand, future growth, and

available supplies. Supply costs include capital for construction, operation and

maintenance. As reclaimed water becomes a more integral part of the water supply

system, the cost of transporting water from a treatment facility to users becomes more

critical in decision making. This introduces another complexity to the planning process in

that the cost of distribution may exceed the gains in economies of scale that are obtained
 130

if wastewater treatment is centralized. Hence, smaller distributed wastewater treatment

facilities may be a cost effective alternative over a single central plant.

Little research has been conducted on water supply system planning optimization.

Ocanas and Mays (1981a and b) formulated and solved a water reuse planning

optimization model using non-linear programming under steady and dynamic conditions.

The steady state model consisted of a nonlinear objective function, linear and nonlinear

constraints for a single period. A large-scale generalized reduced gradient technique was

used to solve this optimization problem (Ocanas and Mays 1981a). In the follow-up

paper, the same technique with successive linear programming methods was applied to a

dynamic water reuse planning model with single and multiple periods (Ocanas and Mays

1981b). Water quality was considered in both papers. In the dynamic model, the capacity

expansion of treatment facilities was considered at the beginning of the period and

operation costs were included in the objective function. This model provided a basis of

the optimization structure for a water supply management system. Conveyance systems

were considered as lumped units without detailed representations of energy loss and

capacity. Later, Ejeta et al. (2004) applied a general approach to the studies in Rio

Grande in New Mexico and Texas including a total suspended solids (TSS) constraint.

The objective of this study was to maximize total net benefit.

In this paper, the water reuse planning system formulated by Ocanas and Mays

(1981a and b) is extended to consider component hydraulic capacities and improve

scalability. Decision variables include the capacities of water transport facilities - such as

pipes, pumps, and canals - as well as the capacity of treatment facilities. This step extends
 131

previous work in this area by optimizing overall system costs with an explicit

representation of energy consumption costs and evaluating the tradeoff of multiple

satellite wastewater treatment facilities. The problem considered in this study is highly

nonlinear and deals with a large-scale water supply system that involves several hundred

decisions. A stochastic search algorithm is successfully applied to determine the optimal

water supply plans for two hypothetical communities.

2. PROBLEM DESCRIPTION

The overall planning goal is to minimize the total costs of construction, operation and

maintenance of the water supply system. The problem considered in this paper extends

earlier work by considering operations costs as a function of flow rates and selected

component sizes. The supply system may have multiple sources and users, and contain

one or more water and wastewater treatment facilities. Unlike previous models,

conveyance system hydraulics are directly embedded in the model to more realistically

estimate operation expenses. One or more planning periods can be represented to allow

delaying expansion investments or to take advantage of economies of scale by

constructing excess capacity in early decision periods.

Figures B.1 and B.2 show system schematics for two communities. Potential flow

paths are shown between sources, sinks and users that are denoted as nodes. Water uses

are agricultural, domestic, industrial and large outdoor irrigation (golf courses, school,

and parks). Available water supply sources can include surface reservoirs or groundwater
 132

aquifers that have storage capabilities and rivers that cannot store water over time. Any

flow provided to domestic and industrial users from surface water sources must be treated

at a water treatment facility (WT) while surface water provided to agricultural and

outdoor uses does not need to be treated. Wastewater return flows from domestic and

industrial uses must be treated at a wastewater treatment plant (WW). After wastewater

treatment, reclaimed water may be supplied to agricultural or large irrigation areas,

discharged to the river, or recharged to the aquifer through infiltration basins.

Groundwater banking can also be achieved by recharging imported waters or surface

supplies.

The primary system constraints are to satisfy conservation of mass at all locations and

components in the system. In addition, each user has defined water quality and demand

requirements, while all supplies are limited by flow capacity or storage volume.

Conveyance system conditions related to canal capacity and pipeline/pump sizes are

formulated to ensure proper sizing and energy consumption. The mathematical form of

the problem is given in the next section.

3. PROBLEM FORMULATION

A general system can be represented by a set of N nodes and A arcs. Arcs denote

water transmission systems while nodes are locations where water is collected from or

split between a set of arcs. The set of nodes is comprised of several subsets representing

sources (IS) and users (IR). Sources are further divided into storage sources (ISS) that
 133

have storage carried over in time (groundwater aquifers and surface reservoirs) and non-

storage sources (INS) that cannot store water over time (rivers and imported waters (IIW)).

Note that a river is represented as upstream (IRU) and downstream nodes (IRD) that are

connected by a river arc. Water is withdrawn by users from upstream nodes and returned

from treatment plants to the downstream river. Water user nodes represent domestic,

industrial, and irrigational purposes and water and wastewater treatment plants (IWT, and

IWWT, respectively).

An arc transmits flow from a node i to a node j. Arcs represent the sets of canals (IC),

pipe connections (IP), pump connections (IU). Pipe connections may have pump stations

at their upstream source depending on the elevation difference and energy losses between

the two connected nodes. Recharge basin arcs (IB) can be used to transmit imported or

treated water to the aquifer to bank or recharge water in the aquifer. In addition, rivers

and users may recharge the aquifer through seepage or infiltration that is represented by

infiltration arcs (II).

The capacities of the water transmission structures such as pipes, pumps, and canals

and treatment plants are described as structural variables. Flow allocations over the water

supply network are operational variables.

The objective function consists of the function of construction and expansion, and

operations and maintenance (O&M) costs for all components (pipes, canals, pumps, and

treatment facilities) or:

Minimize Z * = f1 (κ ijt ) + f 2 (κ ijt ) + f 3 ( χ ijt , H ijt ) + f 4 ( wit ) + f 5 (qijt , wit ) + f 6 (B.1)

Each term in Eq. B.1 is non-linear relationship with respect to the decision variables.
 134

Pipe construction and expansion costs are given by (Clark et al. 2002):

f1 (κ ijt )
= ∑i∈I ,t =T ∑ j∈I P ,t =T
{
xijt 57.198 + 0.35κ ijt + 0.62κ ijt
1.54
+ 0.0018κ ijt
1.9
+ 0.0062κ ijt
1.83

}L
P

0.73 1.8 0.93 0.71

− 0.062κ ijt + 0.02κ ijt + 0.23κ ijt + 0.0022κ ijt ij

(B.1a)

Canal construction and expansion costs are (US Army Corp of Engineers 1980):

f 2 (κ ijt )
⎧ t
t 2 ENR CITY t ENR CITY ⎫
t (B.1b)
= ∑i∈I ∑ ⎨ 0.39κ ij + 55.30 L κ
ij ij ⎬
C ,t = T j∈I C , t = T
⎩ 2877 2877 ⎭

Pump construction and expansion costs are given by (Walski et al. 1987):

f 3 ( χ ijt , H ijt )
= ∑i∈I
P U I U ,t = T ∑ j∈I P U I U , t = T
(500χ t 0.7
ij H ijt
0.4
) (B.1c)

Water and wastewater treatment facility construction and expansion are approximated by

(Tang et al. 1987):

f 4 (qijt , wit )
= ∑i∈I
WT , t = T
(
yit 2897.13 wit + 35987 + ∑i∈I ) WWT , t = T
(
yit 10811.92 wit + 5454228 )
(B.1d)

Operation and maintenance of pipes, canals, pumps, and treatment facilities are given by

(Clark et al. 2002; US Army Corp of Engineers 1980; Walski et al. 1987; Tang et al.

1987):
 135

f 5 (qijt , wit )

∑
⎡ 1
o∈O ⎢
[∑ ∑ x (27.7 + 0.3qijo ) Lij
t

⎣ (1 + I )
o i∈I P ,{t |t ≤ o} j∈I P ,{t |t ≤ o} ij

⎛ ENR o ⎞
+ ∑i∈I ∑ j∈I C ,{t |t ≤ o}⎜
⎜ 0. 0254 L q o 0.572
+ ( 0. 078 + 0.0135 q o
) L ⎟
C , {t | t ≤ o } 1850 ⎟⎠
ij ij ij ij
⎝
t ⎧ o 0.935 ENR CITY ⎫
o
+ ∑i∈I U I ,{t |t ≤ o} ∑ j∈I U I ,{t |t ≤ o} µij ⎨79.47∆ ij qij + 4560qij + 320qij
o o o 0.58
⎬
P U P U
⎩ 2877 ⎭
( )
+ ∑i∈I ,{t |t ≤ o} yit 28.97 wit + 360 + ∑i∈I ,{t |t ≤ o} yit 108.12 wit + 54542
WT WWT
( )] ]
(B.1e)

where IP, IU, IC, IWT, and IWWT are the set of pipe, pump, and canal arcs and water and

wastewater treatment plant nodes, respectively. The superscript t indicates the

construction and expansion time while o denotes the operation period used for evaluating

O&M costs. T and O represent the time sets for construction and expansion, and

operation and maintenance, respectively.

In terms of construction decisions in the above equations, a binary decision variable x

for a pipe identifies whether or not pipe i in the set of links, IP, will be installed with

diameter κ over its length of L. Similarly, µ is a binary decision variable representing

whether or not a pump will be installed with design pump discharge and head, χ and H,

respectively. Note that pumps can be installed at the beginning of defined pipelines or

stand alone as pump connections. The binary decision variable, yi, indicates whether a

water or wastewater treatment plant, will be built with design capacity, w. If only a single

plant capacity is to be determined no binary variable is needed. No discrete variables are

needed for the other components as their decision variables are permitted to go to zero.
 136

Canal variables are represented by the continuous canal depth κ for the given canal length

L.

Operation and maintenance costs (Eq. B.1e) are calculated for each component and

summed over the planning period, O. O&M costs are functions of the component

capacity and flow, q, that is also a model decision variable. The double summations for

the pipe, canal and pump are used to indicate the corresponding connection from a node, i,

to a node, j.

The objective also includes several system defined parameters. ∆ in the pump terms

corresponds to the elevation difference between two nodes, i and j. CITY is a

construction cost factor that varies by location. The ENR cost factor at year t is used to

consider inflation rate in the estimation of construction costs that is given by:

ENR t = -7.7 × 109 + 15.7t − 12.0 × 103 t 2 + 4.1t 3 − 0.5 × 10 −3 t 4 (B.2)

3.1 Simple Bound Constraints on System Flows and Component Sizes

Bounds on the decision variable are:

κ ij , χ ij , H ij ≥ 0 ∀ IC U IP U IU (B.3)

wi ≥ 0 ∀ I WT U I WWT (B.4)

xij ∈ {0, 1} ∀ I P (B.5)

κ ij ≤ M ij xij ∀ IP (B.6)

µ ij = { 0, 1 } ∀ IP U IU (B.7)
 137

χ ij ≤ M ij µij ∀ IP U IU (B.8)

H ij ≤ M ij µij ∀ I P U I U (B.9)

yi ∈ {0, 1} ∀ I WT U I WWT (B.10)

wi ≤ M i yi ∀ I WT U I WWT (B.11)

qij ≥ 0 ∀A (B.12)

qij ≤ M ij xij ∀ IP (B.13)

qij ≤ M ij µij ∀ IU (B.14)

The terms Mij or Mi are assigned a large value or to the upper bound of the

corresponding component size. From these constraints, if the binary variable

corresponding to a design component is set to zero, the design variable is constrained to

be equal to zero. Otherwise, the component can be added to the model.

Lower bounds are given for pipe diameters, canal depths and pump capacities in the

set of all construction variables, IP, IC, and IU (Eq. B.3). Eq. B.4 defines lower bounds on

water and wastewater treatment plant capacities in the sets IWT and IWWT. As noted,

binary variables, x, (Eq. B.5) are introduced to denote the non-existence/existence of a

pipe connection. If the connection is included, i.e., x =1, pipe diameters are bounded by

the maximum diameter (Eq. B.6). The second set of binary variables, µij, (Eq. B.7) is

associated with pump installation and Eqs. B.8 and B.9 define the upper bounds of pump

capacity and discharge, respectively. The last set of binary variable, yi, represents water

and wastewater treatment plants construction (Eq. B.10) and is bounded by the maximum
 138

construction capacity (Eq. B.11). Finally, the operational pipe and pump flows have a

lower bound of zero and an upper bound equal to their capacities (Eqs. B.12 - B.14).

3.2 Node Constraints

In addition to simple bounds on system flows, q, mass balances at system nodes,

demands and flow availability and requirements also limit the range of the flow rates. By

conservation of mass, imported water and water and wastewater treatment plants, must

balance for each operation period, O, or:

∑ j
q oji − ∑ j qijo = 0 i ∈ (I NS \ I RD ) U I WT U I WWT ∀O (B.15)

Similar mass balances at user nodes (the set IR) are expanded to include consumptive

use ( CU oj ) and seepage losses to the aquifer ( LOSS j ) terms or:

∑ j
q oji − ∑ j qijo = CU oj + LOSS j i ∈ I R , ∀O (B.16)

Mass balances for storage nodes (surface reservoirs and groundwater aquifers) are

written over time in terms of water elevation, WEL by:

WEL =WEL o o −1
+
∑ q −∑ q
j
o
ji j
o
ij
(B.17)
i i
AREAi

where AREAi is a surface area of storage nodes.

In addition to mass balance, other physical limits may be imposed at nodes. For non-

storage nodes, the total flows supplied to a treatment facility must be less than the plant

capacity, wi, for each operation period, O, or:

 139

∑ j
q oji ≤ wit ∀i ∈ I WT U I WWT , ∀t ∈ T ≤ ∀o ∈ O (B.18)

Similarly, the release from downstream river nodes must exceed the minimum flow

requirement in the river, RQ or:

∑ j
q oji − ∑ j qijo ≥ RQi i ∈ I RD ∈ I NS , ∀O (B.19)

To meet demands, the sum of the flows entering a user node j must equal or exceed

the nodal demand , Do, or:

∑q i
o
ij ≥ D io ∀ j ∈ I R ∀O (B.20)

where i is the set of nodes supplying demand node j. When minimizing costs, the net total

flow should be equality the demand since, exceeding it, will incur additional costs.

For storage nodes, the water level must be maintained above the required water

elevation, REL or:

WELoi ≥ RELi i ∈ I SS , ∀O (B.21)

3.3 Arc Related Flow Constraints

A significant extension in this formulation is the direct consideration of energy and

flow relationships for pipe/pump systems (the sets IP and IU). Conservation of energy is

applied to these arcs as:

⎡ 4 H ijt qijo
2
8 f Lij o 2 ⎤
⎢ ( H ij −
t
2
) − q
5 ij
− ∆o
ij
⎥ ≥ H min , j − 10000(1 − µijt )
⎢⎣ 3 3χ ij t
g π κ ij
2 t
⎥⎦
i

∀( i , j ) ∈ I P , ∀t ∈ T ≤ ∀o ∈ O (B.22)
 140

where the terms correspond to the pump head added (in parenthesis), the pipe friction

head loss (the Darcy-Weisbach equation with a constant friction factor), and the elevation

difference, ∆, between the upstream and downstream nodes. The downstream head at

node j must satisfy a defined minimum requirement, Hmin,j. When necessary to meet this

requirement, a pump station may be constructed by setting µ equal to 1 that allows the

χ and H to be non-zero in Eqs. B.8 and B.9. Note the pump relationship is developed

assuming a standard quadratic form (USEPA, 2000) in which the cutoff head (maximum

head at zero flow) equals 4/3H and the maximum flow equals twice the design flow (2χ).

For flows to domestic, industrial, agricultural and large outdoor demand nodes that

may be pumped directly to the user from the aquifer or a nearby location (the set IU). A

similar energy equation is written for a pump station without the pipe as:

t o 2
4 t H ij qij
( H ij − − ∆oij ) ≥ H min , j − 10000(1 − µijt ) ∀( i , j ) ∈ I U , ∀t ∈ T ≤ ∀o ∈ O (B.23)
3 t2
3χ ij

To maintain hydraulic efficiency, pump connection flows are required to be between

50% and 150% of the pump flow capacity for all pumps (the sets IP and IU):

qijo ≥ 0.5 χ ijt ∀(i, j ) ∈ I P U I U , ∀t ∈ T ≤ ∀o ∈ O (B.24)

qijo ≤ 1.5 χ ijt ∀(i, j ) ∈ I P U I U , ∀t ∈ T ≤ ∀o ∈ O (B.25)

Canal flows are limited by the canal capacity. Maximum canal flows are computed

using Manning’s open channel flow equation that relates the channel characteristics

(slope (S), roughness (n), and geometry), the channel depth that is a decision variable (κ),

and the channel flow rate ( qijo ) for all channels IC as:
 141

8
1.49 ⎛ 1
2 ⎜ 1 + z ij − z ij ⎞⎟κ ijt 3 S ij 2 ∀( i , j ) ∈ I C , ∀t ∈ T ≤ ∀o ∈ O
2
qijo ≤ (B.26)
nij ⎝ ⎠

Note the canal shape is assumed as the most efficient trapezoidal channel.

Infiltration ( qijo , i is river and j is groundwater) through a river bed to an aquifer is

calculated by Darcy’s equation:

⎛ WELoRI − WELoGW ⎞
qijo = ⎜⎜ COND ⎟⎟( BOT ) LRI ∀(i, j ) ∈ I I (B.27)
⎝ ELRI − WELGW
o
⎠

and WEL o
=
∑ j
q oji − ∑ j qijo
if a node i is river (B.28)
RI
(BOT )VRI

where WELoRI is water elevation in the river in the year o, WELoGW is the groundwater

elevation in the year o, ELRI is the riverbed elevation, and LRI , V, BOT and COND are

the river length, velocity, bottom width and channel hydraulic conductivity, respectively.

Groundwater flows from recharge basins is computed as the product of the rate of

recharge depth times the basin area or:

qijo = AreabasinV (B.29)

where Areabasin is groundwater basin area and V is the recharge rate.

3.4 Water Quality Constraints

Water quality constraints ensure that water supplied to users or returned to sources

satisfy their water quality requirements. Flows entering a node are assumed to be

completely mixed and the downstream water quality is the flow weighted average of the
 142

incoming water concentrations, c. This value must be below a defined minimum value,

WQR. When water is used or treated, it undergoes a change in quality. For a treatment

facility or a river, the qualities leaving the nodes are assumed to be improved by removal

efficiency (WQRE) or:

co
=
∑cj
o
ji q oji (1 − WQREi )
≤ WQRi ∀ i ∈ I NS U I WT U I WWT ∀O (B.30)
∑q
i o
j ij

where WQRE is the removal efficiency of the contaminant in the river or treatment

facility.

Consumers detrimentally affect water quality that is represented by an incremental

addition to the contaminant concentration, WQ∆. The water quality leaving the node is

then computed by:

c o
=
∑cq
j
o
ji
o
ji
+ WQ∆ i ≤ WQRi ∀ i ∈ I R , ∀O (B.31)
∑q
i o
j ij

Water quality is assumed unchanged during transport in canals and pipes.

If the node represents storage components, all stored water is assumed to completely

mix with the inflow water and the average contaminant concentration is computed by:

⎧ci0 ≤ WQRi at year = 0

⎪ o −1 o −1
⎨ o ci (WELi AREAi ) + ∑
c o q o (1 − WQREi )
j ji ji i ∈ I SS , ∀O
⎪ci = ≤ WQRi otherwise
⎩ (WELoi AREAi )

(B.32)
 143

3.5 Objective Function Penalty Term

Random search techniques can be useful in solving a large nonlinear problem.

However, these techniques are only applicable to unconstrained optimization problems.

To insure the constraints are satisfied, constraint violations are generally embedded as a

penalty term in the objective function (Eq. B.1). Here, a simple penalty term, f6, that

scales the absolute value of constraint violations by a large user defined number is

applied for twelve of the above constraints (Eqs. B.18 – B.32 (excluding Eqs. B.17, B.27,

B.28, and B.29)). Constraints B.3 - B.14 are direct functions of a single independent

decision variable and can be restricted during the random search.

Total penalty term, f6, is the sum of penalties from each of the 12 constraints or:

f 6 = fn20 + fn21 + fn22 + fn23 + fn24 + fn25

(B.33)
+ fn26 + fn27 + fn28 + fn32 + fn33 + fn34

where fni is the penalty term for Eq. i.

Since the objective is to minimize cost, the positive penalty term will be minimized to

satisfy the constraints and optimize the solution. The penalty term is zero when the

constraint is satisfied. A scaled violation term by a user defined value M ' is added if the

constraint is not met. For example, the term for Eq. B.26 is given by:

⎧ 1.49 ⎛ 2
8
⎞⎟κ t 3 S 2
1

⎪ 0 , if q o
≤ 2 ⎜ 1 + z − z
⎪
ij
nij ⎝ ij ij
⎠ ij ij

fn 28 = ⎨
⎛ 8 1 ⎞
⎪ M ' ⎜ q o − 1.49 2 ⎛⎜ 1 + z 2 − z ⎞⎟κ t 3 S 2 ⎟ , otherwise
⎪ ⎜⎝ ij nij ⎝ ij ij
⎠ ij ij ⎟⎠
⎩
∀( i , j ) ∈ I C , ∀t ∈ T ≤ ∀o ∈ O (B.34)
 144

A single M ' value is applied for all constraints because it is difficult to predict which

constraint will have significant effect on the objective function. For simplicity, this large

factor is not changed during the optimization process.

3.6 Summary of Formulation

The set of equations (Eqs. B.1 - B.32) forms the basis for the water supply

optimization problem. The formulation is general and can consider multiple demand

centers and supply sources, one or more treatment facilities and linkages between users.

Distances and system topography can also be represented. One or more planning periods

can be evaluated with multiple years within each planning period. The decision variables

are a mixture of continuous and binary variables. The equations are highly nonlinear. A

heuristic stochastic search algorithm known as the shuffled frog leaping algorithm

(SFLA) is applied to solve the problem.

4. SHUFFLED FROG LEAPING ALGORITHM (SFLA)

The posed water supply problem is non-linear and non-convex. The degree of

nonlinearity causes difficulties in solving the problem using non-linear programming.

Nonlinear formulation was constructed in MINOS, however, no improvement was

achieved from the initial solutions. Therefore, stochastic search technique, the shuffled

frog leaping algorithm (SFLA) (Eusuff et al., 2006), was applied.

 145

SFLA is applicable to problems with continuous and discrete decision variables. It is

a descent based stochastic search method that begins with an initial population of frogs

whose characteristics, known as memes, represent the decision variables.

Memes are units of knowledge that, like an idea, improve as more information is

gained. In a physical domain, memes represent a frog’s position that is improved by

comparing its objective function value with other frogs and changes in a frog’s

characteristics (memes) are made by moving (leaping) in the direction of lower objective

function value in case of minimum problem. In a frog’s domain, he is moving to

locations with higher food concentrations. This movement is similar to the evolution of

an idea that changes as each individual learns something from others.

In SFLA, the total population is partitioned into groups (memeplexes) that search

independently. After a number of iterations, the groups are combined and reformed to

pass information between individuals, similar to the cross-fertilization of ideas between

different organizations. This goal of the overall process is to determine global optimal

solutions. As such, a random component is also included to generate new points that

maintain a diverse search. The process of local group searches and population shuffles

continues until convergence to an optimum is reached or a user-defined number of frogs

have been evaluated.

The global exploration and local search algorithms are presented in the following

steps. This algorithm is slightly modified from Eusuff et al. (2006) in the local search that

extends the potential step size to be beyond the current frog locations. In Eusuff et al, a

new point was limited to fall between the existing frogs.

 146

4.1 Global Exploration

Step 0: Initialize: Select m and N, where m = number of memeplexes and N = number of

frogs in each memeplex. Therefore, the total population size in the swamp, F = mN.

Step 1: Generate a virtual population: Sample F virtual frogs U(1), U(2), …, U(F) in the

feasible space Ω ⊂ ℜ d where d is the number of decision variables (i.e., number of

menotype(s) in a meme carried by a frog). The ‘ith’ frog is represented as a vector of

decision variable values U(i) = (U i1 ,U i2 ,...,U id ) , i = 1, …, F.

Compute the performance value, f(i), for each frog U(i).

Step 2: Rank frogs: Sort the F frogs in order of decreasing performance value in case of

minimum problem. Store them in array X = {U(i), f(i), I = 1…F}, so that i = 1 represents

the frog with the best performance value. Record the best frog’s position in the entire

population (F frogs), PX, (where PX = U(1)).

Step 3: Partition frogs into memeplexes: Partition array X into m memeplexes Y1, Y2, …,

Ym, each containing n frogs, such that:

Y k = [U ( j ) k , f ( j ) k | U ( j ) k = U (k + m( j − 1)), f ( j ) k = f (k + m( j − 1)), j = 1,..., n], k = 1,..., m

For example, for m = 3, rank 1 goes to memeplex 1, rank 2 goes to memeplex 2, rank 3

goes to memeplex 3, rank 4 goes to memeplex 1 again, and so on.

Step 4: Memetic evolution within each memeplex: Evolve each memeplex Yk, k = 1, …,

m according to the frog leaping algorithm (FLA) outlined below.

 147

Step 5: Shuffle memeplexes: After a defined number of memetic evolutionary steps

within each memeplex, replace Y1, …, Ym into X, such that X = {Yk, k = 1, …, m}. Sort X

in order of decreasing performance value. Update the population best frog’s position, PX.

Step 6: Check convergence: If the convergence criteria are satisfied, stop. Otherwise,

return to Step 3. Typically, the decision on when to stop is made by a pre-specified

number of consecutive time loops when at least one frog carries the “best memetic

pattern” without change. Alternatively, a maximum total number of function evaluations

can be defined.

4.2 Local Exploration: Frog Leaping Algorithm (FLA)

Step 0: Set im (iteration count) and iN (shuffle count) equal to zero. The number of

iterations and shuffles are limited to user-defined values ic and is, respectively. Form an

initial random set of frogs and evaluate each frogs objective function value.

Step 1: Set im = im + 1.

Step 2: Set iN = iN + 1.

Step 3: Construct a sub-memeplex: The frogs’ goal is to move towards the optimal ideas

by improving their memes. An individual frog is updated using the presently available

information. The new frog is returned to the memeplex and another frog is updated. This

strategy is consistent with evolution of ideas since the best information available is used

unlike a genetic algorithm in which the entire population is updated prior to using any

information gained. To complete this process, a subset of the memeplex, (Yk, k = 1, …,

 148

m), called a sub-memeplex, (Ziq, iq = 1, …, q) is considered. Figure B.3 presents the

structure of population, memeplex, and sub-memeplex. The sub-memeplex selection

strategy from a memeplex (Yk) having n frogs is to give higher weights to frogs that have

higher performance values and less weight to those with lower performance values. The

weights are assigned with a triangular probability distribution, i.e.,

pj = 2(n+1-j)/(n(n+1)) , j = 1, ... , n

such that, within a memeplex, (Yk, k = 1, …, m), the frog with best performance has the

highest probability of being selected for the sub-memeplex, p1 = 2/(n+1) and the frog

with worst performance has the lowest probability, pn = 2/(n(n+1)).

Here, q distinct frogs are selected randomly from n frogs in each memeplex (Yk, k =

1, …, m) to form the sub-memeplex array (Ziq, iq = 1, …, q). The sub-memeplex is sorted

so that frogs are arranged in order of decreasing performance. Record the best (iq = 1; iq

= 1, ..., q) and worst (iq = q; iq = 1,…, q) frog’s position in the sub-memeplex as vectors

PB and PW, respectively.

Step 4: Improve the worst frog’s position: When improving a frog’s position, it can adapt

their ideas from the best frog within the memeplex (group), PB, or from the global

(population) best, PX. The direction, step size and new position are first computed for the

frog with worst performance in the sub-memeplex (Ziq, iq = 1, …, q). The computation

includes identifying the direction of improvement (gradient) and the magnitude of change

(step length) in that direction. The direction of change (positive or negative) is defined by

the movement toward the sub-memeplex best or (PB - PW) where P represents the location

vector. This change involves both in magnitude as well as direction of the decisions.
 149

The magnitude of the step size is randomly selected as a proportion of change

direction. It is limited by the maximum step size, Smax. The present version of the

algorithm considers a pre-specified fraction of the bound of the variable value as Smax.

Mathematically, the step size is defined as:

Step size, |s| = MIN[2(rand (PB - PW)), Smax ]

where rand is a random number in the range [0,1]. The new position is then computed by:

Z(iq = q) = PW + s (B.35)

Note that the multiplier of 2 in the step size calculation allows the frog’s new position to

be between the two frogs or move beyond the better frog’s location depending on rand.

Then, if the new position is beyond feasible boundary for any decision variable, it is

forced to be the boundary value. This modification improves upon convergence seen in

Eusuff et al (2006).

Compute the new performance value f(iq = q). If the new f(iq = q) is better than the old

f(iq = q), i.e., if the move produces a benefit, then replace the old Z(iq = q) with the new one

and go to Step 7. Otherwise go to Step 5.

Step 5: If Step 4 cannot produce a better result then the step and new position are

computed for that frog by:

Step size, |s| = MIN[2(rand(PX - PW)), Smax ]

and the new position is computed by Eq. B.35.

Compute the new performance value f(iq = q) for point Z(iq = q). If the new f(iq = q) is better

than the old f(iq = q), i.e., if the evolution produces a benefit, then replace the old Z(iq = q)

with new one and go to Step 7. Otherwise go to Step 6.

 150

Step 6: Censorship: If the new position is either infeasible or worse than the old position,

the spread of the defective meme is stopped by randomly generating a new frog ‘r’ at a

feasible location to replace the frog whose new position was not favorable to progress.

Compute f(r) and set Z(iq = q) = r and f(iq = q) = f(r).

Step 7: Upgrade the memeplex: After the memetic change for the worst frog in the sub-

memeplex, replace Ziq in their original locations in Yk. Sort Yk in order of decreasing

performance value.

Step 8: If iN < is , go to Step 2.

Step 9: If im < ic , go to Step 1. Otherwise return to global search to shuffle

memeplexes.

Steps 4 and 5 of the FLA are similar in philosophy to particle swarm optimization. A

descent direction is identified for a particular frog and the frog is moved in that direction.

Here, however, since the global search is also introduced in the shuffle operation, only

the local minimum is used rather than the complete population best (Step 4) unless no

improvement is made (Step 5). Since a descent direction is implicitly applied, it may be

fruitful to perform a line search rather than a random step but the simpler approach is

taken here.
 151

5. APPLICATIONS

The optimization problem (Eqs. B.1 - B.32) has been formulated for the water supply

systems shown in Figures B.1 and B.2 for 20-year planning periods. New structural

component construction is permitted at the outset (year 1) and new components or

existing component expansion may be added after 10 years. Biochemical Oxygen

Demand (BOD) is used as the representative water quality parameter.

5.1 Single Wastewater Treatment Plant System

The first system to be optimized (Figure B.1) consists single water and wastewater

plants, multiple sources (imported water, groundwater aquifer, and surface water) and

two demands centers (domestic and agricultural). Three types of water transport

structures are used depending upon the connection: canal, pipe and/or pump. All canal

flows for the conveyance of imported and raw water sources are driven by gravity.

Agricultural areas and water treatment plant may directly pump groundwater from

aquifers available near their location so do not require a pipe link. Other flows are

transported through pipes that may require a pump station to supply the energy necessary

to pass flow through the pipeline and satisfy the minimum pressure head requirement at

the outlet (14.0 m of water = 137.9 kPa = 20 psi). Groundwater replenishment through

recharge basins is assumed to occur at a constant rate of 9.1 m/yr (30 ft/yr). Seepage

losses from users to the aquifer are assumed as 0.1% of total user demand.
 152

Input parameters for the single wastewater treatment plant system are shown in

Tables B.1 and B.2. The system contains the total of eighteen arcs (Figure B.1). Origin,

destination nodes and arc lengths are listed in Table B.3. As seen in Figure B.1, the

network consists of six canal depth construction decisions (6) and seven pump/pipeline

arcs with their three design decisions (pipe diameter and pump flow capacity and head)

for a total of 21 decisions. The network also includes two pump links for which the pump

design flow and head (four total design decisions) must be selected and two treatment

facilities with plant capacity as decision variables (total of 2 decisions). Thus, 33 design

decision variables are to be determined for each of the two planning periods or a total of

66 design decisions.

Ten of the 18 arc flows in each period are independent control decision variables that

are also selected by the optimization model. The remaining 8 are dependent variables that

are computed from the mass balance constraints defined in Eqs. B.15 and B.16.

Therefore, the final optimization problem contains a total 86 of decision variables for the

two design periods (66 design and 20 control decision variables).

The following SFLA parameters were selected from experience and preliminary

testing: the total number of population (F = 3000), memeplex (m = 10), frogs in each

memeplex (N = 300), evolutionary steps (iN = 300), and frogs in a sub-memeplex (iq =

300) are established and applied in the single wastewater treatment plant system. The

problem was run on a Dell Inspiron with a Centrino Duo T2300 1.6GHz and 1GB of

RAM and was solved in 5.5 min. after 131 thousand function evaluations. Figure B.4

shows the progress of solution with respect to the number of function evaluations. The
 153

penalty term that accounts for constraint violations fell dramatically in early iterations

after which the total system cost gradually decreased. Total construction and operation

cost for the single treatment plant system for the 20-year period is $ 771 million (present

value for year 0) or an annual cost of $47 million.

Table B.4 lists the optimal component designs and the optimal network solution is

depicted graphically in Figure B.5. Water and wastewater treatment plant capacities are

0.05 and 0.05 km3/yr during the first design period, respectively. The water and

wastewater treatment plants are not expanded at year 10 suggesting that economies of

scale made oversizing in the first period to be more desirable than future expansion.

Increased transport capacity was required to/from the domestic area in year 10 to

convey the increased demands. Although the flow allocation from the aquifer to the farms

remains constant over time, the groundwater pump capacity was expanded to overcome

the required lift to overcome the drop in the aquifer water level (Table B.5).

The system has abundant amount of water in downstream river and subsurface (Table

B.5) to preserve sustainability. Therefore, the domestic area is supplied mostly from river

through water treatment plant, and pumped water from the aquifer is the main source for

agricultural area.

BOD concentrations in the aquifer and the downstream river remain steady and below

their 30 mg/l water quality requirements (Table B.5). Influents to domestic and

agricultural area also have better quality than the required, 5 mg/l and 30 mg/l,

respectively.
 154

As shown in Table B.7, pipe construction is the dominant cost for this system and

economies of scale compel some pipe installations to be constructed in the initial period.

Pipe and pump connections require significant operation costs as compared with the

operation of canal and treatment plants.

5.2 Multiple Wastewater Treatment Plant System

As shown in Figure B.2, multiple water users and wastewater treatment plants are

included in the multiple wastewater treatment plant system in order to investigate a more

generalized system. This network is consists of six users - three domestic areas, one

industrial, one agricultural, and one large outdoor area – and three wastewater treatment

plants. In general, input parameters used for the multiple wastewater treatment plant

system are the same as for the single wastewater treatment plant system except for the

initial population at the domestic areas (Table B.8). Table B.9 summarizes the multiple

wastewater treatment plant system nodal parameters. This network has 44 arc

connections and their lengths are given in Table B.10. The structural design include 6

canals (6 parameters for canal depth), 29 pipes of which pump station could be built

depending on energy relationship (29 parameters for each pipe diameter, pump design

capacity, and pump head), 2 pumps (2 parameters for each pump design capacity and

head), and 4 treatment plants (4 parameters for capacities). The structures can be built or

expanded in year 1 and 10. Total structural design variables are 101 (= 6 + 3 × 29 + 2 × 2

+ 4) for each design period and a total 202 for whole operational period.
 155

Flow allocations through twenty-three arcs out of forty four are defined as operation

decision variables while the remaining 21 arcs are dependent variables that are computed

by mass balance equations. In total, the final problem consists of 248 decision variables

(2 design periods × 124 decision variables (101 and 23 for design and operation

variables, respectively)).

The problem was solved using the computer system cited in the previous section in

about 70 minutes and nearly 582 thousands function evaluations. The optimal cost for the

system was $837 million as the present value in the starting year of the planning period

and the estimated annual cost was $51 million. Final optimal solution is depicted in

Figure B. 6. Although the optimal solution found may not be the global optimal solution

due to high discrete nonconvexity associated with the study system, the optimization

process demonstrates the improvement in overall system cost and reduction in the penalty

term (Figure B.7).

Since the system has sufficient local water to meet user demands, imported water is

not purchased. Domestic areas 1 and 2 are supplied from the aquifer which has good

enough water quality to supply domestic demands, and domestic area 3 and industrial

area are supplied from upstream river through water treatment plant.

Pipe, pump, and canal capacities are given in Table B.11. Table B.12 summarizes the

capacities for water and wastewater treatment plants. Canals carry water from the

upstream river to agricultural and large outdoor areas. Large outdoor turf use is partially

supplied with treated wastewater. As for the single treatment plant case, most pipe

construction occurred at the outset due to economies of scale. Pipe construction cost
 156

dominates the total system cost (Table B.13). Water quality parameters for influent and

effluent are satisfactory and listed in Table B.14. Water elevation, source discharge,

water demand and population are summarized in Tables B.15 and B.16, respectively.
 157

6. CONCLUSIONS

As water demands grow, water supply system capacities must be increased to provide

the desired water. A poorly designed system can waste money and energy. Optimization

of the system can assist decision makers make good decisions to respond to long term

changes. In this study, a large-scale general water supply system optimization model is

developed using deterministic nonlinear programming. The approach was applied to two

moderate and larger hypothetical water networks.

Difficulties in solving the problem using gradient based NLP methods led to the

application of a heuristic stochastic search algorithm, the Shuffled Frog Leaping

Algorithm (SFLA). As the example water communities, the single (7 nodes) and multiple

wastewater treatment plant system (13 nodes) have been optimized in terms of the total

system cost. Total number of decision variables in the single and multiple wastewater

treatment plant system application are 86 and 248, respectively. The resulting annual

minimum costs were $47 million and $51 million, respectively.

The developed single and multiple wastewater treatment plant applications have

sufficient local water to supply user demand, so no external water is purchased.

Economical scale suggests construction of enough treatment and transportation facilities

at the outset in many connections. Pipe construction cost and pipe and pump operation

cost dominates total operation cost. Supply water to user has better quality than the

required and water source keep enough and better quality water to preserve environment.

Further research efforts are needed to develop more detail water supply system, for
 158

example, by introducing discrete pipe diameter or uncertainties in the parameters.

Different kind of random search technique such as Shuffled Complex Evolution (SCE)

can be applied and compare the results to find a proper algorithm for the water supply

system.
 159

7. NOMENCLATURE

Indices and Sets

N a set of nodes in a network (sources, users, and treatment plants)

A a set of arcs (i, j) from a node i to a node j in a network, ∀i, j ∈ N

W a set of pollutants

T a set of design period t, ∀t ∈ T = {1, 6}

O a set of operation period t, ∀o ∈ O = {1, 6}

Subsets,

I C a set of canal connections, I C ⊆ A

I P a set of pipe connections, I P ⊆ A

I U a set of pump connections, I U ⊆ A

I B a set of connections through a recharge basin to an aquifer, I B ⊆ A

I I a set of seepage from users to an aquifer and riverbed infiltration, I I ⊆ A

I R a set of users, I R ⊆ N

I S a set of sources, I S ⊆ N

I SS a set of storage sources, I SS ⊆ I S ⊆ N

I NS a set of non-storage sources, I NS ⊆ I S ⊆ N

I IW imported water, I IW ⊆ I NS ⊆ I S ⊆ N

I RU river upstream node, I RU ⊆ I NS ⊆ I S ⊆ N

 160

I RD river downstream node, I RD ⊆ I NS ⊆ I S ⊆ N

I WT a set of water treatment plants, I WT ⊆ N

I WWT a set of wastewater treatment plants, I WWT ⊆ N

Data

f Darcy Weisbach coefficient

nij Manning coefficient of pipes and canals from i to j ∀I P U I C

zij Channel side slope from i to j ∀I C

COND hydraulic conductivity

I interest rate

CITY city multiplier

IWA imported water available

o
POPi population at an operation year o at i i ∈ IR , o ∈ O

0
At year 0, POPi initial population at a node i

POPi o = POP 0 (1 + POPGR)o

PP o precipitation at an operation year o o∈O

AREAi area of a node i if a node i is a storage source i ∈ IS

CU i consumptive use at a node i ∀N

LOSS i seepage loss to an aquifer at a node i ∀N

ELi elevation at a node i ∀N

Lij length for an arcs (i, j) ∀A

 161

RELi required water elevation at a node i if a node i is a storage source i ∈ I S

RQi required discharge at a node i if a node i is a non-storage source i ∈ I S

WQRi water quality requirement at a node i ∀I R U I S

WQ∆ i water quality increasing at a node i ∀I R

WQREi water quality removal efficiency at a node i ∀I S U I T

cik0 water quality of pollutant k in a node i in year 0 if a node i is a storage source

i ∈ I SS

BOT river bottom width

LRI river length

VRI average velocity in river

Hmin,j minimum pressure requirement at the end of pipe and pump connections

Areabasin groundwater basin area

V recharge rate

Decision variables

xijt takes value 1 if an arcs (i, j) is built at a design period t

and 0 otherwise ∀I P U I C , t ∈ T

µijt takes value 1 if a pump in an arcs (i, j) is built at a design period t

and 0 otherwise ∀I P U I U , t ∈ T

yit takes value 1 if a water and wastewater treatment plant (i) is built at a design period t

and 0 otherwise ∀I WT U I WWT , t ∈ T

 162

κ ijt pipe diameter [L] or canal depth [L] for an arcs (i, j) at a design period t

∀I P U I C , t ∈ T

χ ijt capacity of pump for an arcs (i, j) [L3/T] at a design period t

∀I P U I U , t ∈ T

H ijt Design head of pump for an arcs (i, j) [L] at a design period t

∀I P U I U , t ∈ T

qijo flow allocation for an arcs (i, j) at an operation year o [L3/T] ∀A ,

o∈O

wit capacity at a node i at a design period t [L3] ∀I WT U I WWT , t ∈ T

c oijk water quality concentration of pollutant k for an arcs (i, j) in an operation year o

WELoi water table elevation at a node i in an operation year o ∀I S , o ∈ O

∆oij elevation differences for the pump installation at an arcs (i, j) in an operation year o

⎧⎪WELoj if j ∈ I S ⎧⎪WELoi if i ∈ I S
=⎨ −⎨
⎪⎩ EL j otherwise ⎪⎩ ELi otherwise

⎛ ∆oij ⎞
Sijo Channel bottom slope for an arcs (i, j) in an operation year o ⎜= ⎟ ∀I C ,
⎜ L ⎟
⎝ ij ⎠

o∈O

D io demand at a node i in an operation year o ∀I R , o ∈ O

Indices and sets related to SFLA

U(i) frogs in entire population, i = 1, …, F

 163

Yi frogs in a memeplex, i = 1, …, m

Zi frogs in a sub-memeplex, i = 1, …, iq

Data related to SFLA

F total population

M number of memeplexes

N number of frogs in each memeplex

d number of decision variables

im iteration count in frog leaping algorithm

iN shuffle count

Smax maximum step size

ic maximum number of iterations

is maximum number of shuffles

Variables related to SFLA

F(i) performance value of a frog U(i)

PX best frog in entire population F

PB best frog in a sub-memeplex

PW worst frog in a sub-memeplex

s step size
 164

8. REFERENCES

Clark, R. M., Sivaganesan, M., Selvakumar, A., and Sethi, V. (2002). “Cost models for

water supply distribution systems.” Journal of Water Resources Planning and

Management, 128(5), 312-321.

Ejeta, M. Z., McGuckin , J. T., and Mays, L. W. (2004). “Market exchange impact on

water supply planning with water quality.” Journal of Water Resources

Planning and Management, 130(6), 439–449.

Eusuff, M., Lansey, K., and Pasha, F. (2006). “Shuffled frog-leaping algorithm: a

memetic meta-heuristic for discrete optimization.” Engineering Optimization,

38(2), 129-154.

Ocanas, G. and Mays, L. W. (1981a). “A model for water reuse planning.” Water

Resources Research, 17(1), 25–32.

Ocanas, G. and Mays, L. W. (1981b). “Water reuse planning models: extensions and

applications.” Water Resources Research, 17(5), 1311-1327.

Tang, C. C., Brill, E. D., and Pfeffer, J. T. (1987). “Optimization techniques for

secondary wastewater treatment system.” Journal of Environmental

Engineering, 113(5), 935-951.

US. Army Corps of Engineers, (1980). Methodology for areawide planning studies.

Engineer Technical. Letter No. 1110-2-502, Washington, D.C.

United States Environmental Protection Agency, (2000). EPANET2 Users Manual.

EPA/600/R-00/057, Cincinnati, OH.

 165

Walski, T. M., Brill, E. D., Gessler, J., Goulter, I. C., Jeppson, R. M., Lansey, K., Lee,

H., Liebman, J. C., Mays, L., Morgan, D. R., and Ormsbee, L. (1987). “Battle

of the network models: epilogue.” Journal of Water Resources Planning and

Management, 113(2), 191-203.

 166

9. TABLES

Table B.1. Input parameters for the single wastewater treatment plant system
application
Parameter Value Unit
Darcy-Weisbach coefficient, f 0.02
Manning's coefficient, n 0.014
Canal side slope, S 2
Hydraulic conductivity, COND 9.144 m/yr
Imported water, IWA 0.062 km3/yr
Initial population, POP0 300,000
Population growth rate, POPGR 2.7 %/yr
Interest rate, I 2.0 %/yr
City multiplier, CITY 1
Annual precipitation, PP 69.8 mm/yr
Basin area, AREA 132,771 km2
Required water elevation of groundwater, REL 397 m
Required discharge of downstream river, RQ 2.83 m3/s
Initial water quality of groundwater, cik0 , i = GW 0.04 mg/l
Water quality of imported water, cik0 , i = IW 30.0 mg/l
0
Precipitation water quality, c , i = PP
ik 1.0 mg/l
Upstream river bottom width, BOT 3.05 m
Upstream river length, LRI 4,023 m
Average velocity of upstream river, VRI 1.52 m/s
Agricultural consumptive use, CU 0.26 km3/yr
* GW – groundwater, IW – imported water, PP = precipitation
 167

Table B.2. Nodal input parameters for the single wastewater treatment plant system
application
Water quality Quality Pollutant
Area Elevation
Nodes requirement degradation removal
(km2) (m)
(mg/l) (mg/l) efficiency (%)
Imported water 0 671 30 0 0
Groundwater 13277 518 30 0 0.8
Upstream river 0 641 30 0 0.2
Downstream
0 653 30 0 0.2
river
Water
0 610 3 0 0.99
treatment plant
Domestic area 0 579 5 200 0
Agricultural
133 610 30 100 0
area
Wastewater
0 640 30 0 0.99
treatment plant
* Initial water surface elevation.
 168

Table B.3. Arc lengths for the single wastewater treatment plant system application
Arcs Origin Destination Length (km)
1 IW** AG 16.1
2 IW WT 4.8
3 RIU WT 4.8
4 RIU AG 11.3
5 WT DO 8.0
6 WT AG 160.9
7 GW DO -
8 WW AG 14.5
9 GW WT -
10 GW AG -
11 IW GW 8.0
12 WT RID 56.3
13 DO GW -
14 DO WW 8.0
15 WW RID 160.9
16 AG GW -
17 AG RID 160.9
18 RIU GW -
** IW - Imported water, GW - Groundwater, RIU – Upstream river, RID – Downstream river,
WT - Water treatment plant, DO - Domestic area, AG - Agricultural area, WW - Wastewater
treatment plant
 169

Table B.4. Optimal solution for the single wastewater treatment plant system
application
Design Design
Flow allocation Canal Depth
Pipe diameter (mm) pump discharge pump head
Leaving Entering (m3/s) (m)
(m3/s) (m)
node node
First Second First Second First Second First Second First Second
period periods period periods period periods period periods period periods
IW AG 0.00 0.00 0 0
IW WT 0.00 0.00 0 0
RIU WT 0.34 2.07 1 1
RIU AG 0.59 0.59 1 1
590 1417
WT DO 0.70 2.88 0.00 0.00 0 0
(600) (1500)
WT AG 0.00 0.00 0 0 0 0 0 0
520 737
GW DO 1.03 1.03 0.68 0.69 61 74
(600) (900)
144 144
WW AG 0.02 0.02 0.00 0.00 0 0
(200) (200)
GW WT 0.36 0.80 0.24 0.55 145 145
GW AG 7.67 7.67 5.13 8.87 87 87
IW GW 0.39 0.39 1 1
WT RID 0.00 0.00 0 0 0 0 0 0
DO GW 0.00 0.00
832 832
DO WW 1.11 1.42 0.74 1.03 107 107
(900) (900)
1172 1172
WW RID 1.08 1.40 0.74 0.93 152 183
(1200) (1200)
AG GW 0.14 0.14
AG RID 0.00 0.00 0 0
RIU GW 0.00 0.00
*the values in () indicate commercial pipe diameter available.
 170

Table B.5. Storage water sources and water demands for the single wastewater
treatment plant system application

Groundwater Downstream river Domestic Agricultural

Year elevation Discharge (m3/s) Population demand demand
(m) (km3/yr) (km3/yr)

2001 518.2 59.3 300,000 0.044 0.26

2002 517.5 59.4 308,100 0.045 0.26
2003 516.9 59.4 316,419 0.046 0.26
2004 516.3 59.4 324,962 0.047 0.26
2005 515.6 59.4 333,736 0.048 0.26
2006 515.0 59.5 342,747 0.049 0.26
2007 514.4 59.5 352,001 0.051 0.26
2008 513.7 59.5 361,505 0.052 0.26
2009 513.1 59.5 371,266 0.053 0.26
2010 512.4 59.6 381,290 0.055 0.26
2011 511.9 59.1 391,585 0.056 0.26
2012 511.4 59.1 402,157 0.057 0.26
2013 510.8 59.2 413,016 0.059 0.26
2014 510.3 59.2 424,167 0.060 0.26
2015 509.8 59.2 435,620 0.062 0.26
2016 509.2 59.3 447,381 0.063 0.26
2017 508.7 59.3 459,461 0.065 0.26
2018 508.2 59.3 471,866 0.067 0.26
2019 507.6 59.4 484,607 0.068 0.26
2020 507.1 59.4 497,691 0.070 0.26
 171

Table B.6. Water quality change (BOD) (mg/l) for the single wastewater treatment
plant system application
Downstream
Year Groundwater WT effluent DO influent AG influent WW effluent
River
2001 0.0* 0.8 0.0 0.0 0.7 2.0
2002 0.0 0.8 0.0 0.0 0.8 2.0
2003 0.0 0.8 0.0 0.0 0.8 2.0
2004 0.0 0.8 0.0 0.0 0.8 2.0
2005 0.0 0.8 0.0 0.0 0.8 2.0
2006 0.0 0.8 0.0 0.0 0.8 2.0
2007 0.0 0.8 0.0 0.0 0.8 2.0
2008 0.0 0.8 0.0 0.0 0.8 2.0
2009 0.0 0.8 0.0 0.0 0.8 2.0
2010 0.0 0.8 0.0 0.0 0.8 2.0
2011 0.0 0.8 0.0 0.0 0.8 2.0
2012 0.0 0.8 0.0 0.0 0.8 2.0
2013 0.0 0.8 0.0 0.0 0.8 2.0
2014 0.0 0.8 0.0 0.0 0.8 2.0
2015 0.0 0.8 0.0 0.0 0.8 2.0
2016 0.0 0.8 0.0 0.0 0.8 2.0
2017 0.0 0.8 0.0 0.0 0.8 2.0
2018 0.0 0.8 0.0 0.0 0.8 2.0
2019 0.0 0.8 0.0 0.0 0.8 2.0
2020 0.0 0.8 0.0 0.0 0.8 2.0
* 0.0 means almost close to 0.0 mg/l.

Table B.7. Construction and operation cost for the single wastewater treatment
plant system application
Cost Pipes Pumps Canals Treatment plants
6
Construction (× 10 $) 190.79 26.07 33.59 6.08
Operation at year 1
(× 106 $/yr)
21.22 5.39 0.07 0.06
Expansion (× 106 $) 5.86 2.77 0.00 0.15
Operation at year 10
(× 106 $/yr)
22.72 6.22 0.13 0.06
 172

Table B.8. Additional input parameters for the multiple wastewater treatment plant
system application
Parameter Value
Initial population for domestic area 1 300,000
Initial population for domestic area 2 400,000
Initial population for domestic area 3 600,000

Table B.9. Nodal input parameters for the multiple wastewater treatment plant
system application
Water quality Quality Pollutant
Area Elevation
Nodes requirement degradation removal
(km2) (m)
(mg/l) (mg/l) efficiency (%)
Imported water 0 671 30 0 0
Groundwater 13,277 518* 30 0 0.8
Upstream river 0 640 30 0 0.2
Downstream
0 549 30 0 0.2
river
Water treatment
0 610 3 0 0.99
plant
Domestic area 1 258.0 579 5 200 0
Domestic area 2 344.0 610 5 200 0
Domestic area 3 516.0 613 5 200 0
Industrial area 2063.9 579 5 200 0
Agricultural area 244.9 610 30 100 0
Large outdoor
173.4 610 30 100 0
area
Wastewater
0 610 30 0 0.99
treatment plant 1
Wastewater
0 579 30 0 0.99
treatment plant 2
Wastewater
0 640 30 0 0.99
treatment plant 3
* Initial water surface elevation.
 173

Table B.10. Arc Lengths for the multiple wastewater treatment plant system
application
Leaving Entering Leaving Entering Length
Length (km)
point point point point (km)
1 RIU WT 5 23 DO3 WW3 13
2 IW WT 5 24 ID WW1 5
3 IW AG 16 25 ID WW2 32
4 IW GW 8 26 ID WW3 32
5 RIU AG 11 27 WW1 AG 8
6 RIU LO 5 28 WW1 RID 14
7 WT DO1 10 29 WW1 LO 24
8 WT DO2 10 30 WW2 AG 32
9 WT DO3 10 31 WW2 RID 14
10 WT ID 6 32 WW2 LO 8
11 GW DO1 0 33 WW3 AG 8
12 GW DO2 0 34 WW3 RID 14
13 GW DO3 0 35 WW3 LO 24
14 GW ID 0 36 GW AG -
15 DO1 WW1 32 37 GW LO -
16 DO1 WW2 3 38 DO1 GW -
17 DO1 WW3 32 39 DO2 GW -
18 DO2 WW1 32 40 DO3 GW -
19 DO2 WW2 8 41 ID GW -
20 DO2 WW3 8 42 RIU GW -
21 DO3 WW1 8 43 AG GW -
22 DO3 WW2 32 44 LO GW -
**
IW - Imported water, GW - Groundwater, RIU – Upstream river, RID – Downstream river, WT
- Water treatment plant, DO - Domestic area, AG - Agricultural area, WW - Wastewater
treatment plant
 174

Table B.11. Optimal solution for the multiple wastewater treatment plant system
application
Design Design
Flow allocation Pipe diameter Canal Depth
pump discharge pump head
Leaving Enterin (m3/s) (mm) (m)
(m3/s) (m)
node g node
First Second First Second First Second First Second First Second
period period period period period period period period period period
RIU WT 4.0 4.5 5 5
IW WT 0.0 0.0 0 0
IW AG 0.0 0.0 0 0
IW GW 0.0 0.0 0 0
RIU AG 5.8 8.6 10 10
RIU LO 0.6 1.1 3 3
WT DO1 0.0 0.0 0 0 0.0 0.0 0.0 0.0
WT DO2 0.0 0.0 0 0 0.0 0.0 0.0 0.0
WT DO3 2.4 3.0 999 999 2.4 2.8 104.4 104.4
WT ID 1.4 1.6 768 768 1.2 2.0 152.4 243.8
GW DO1 1.2 1.5 0 0 1.7 2.0 149.1 149.1
GW DO2 1.6 2.9 0 0 1.1 1.7 82.8 90.9
GW DO3 0.0 0.0 0 0 0.0 0.0 0.0 0.0
GW ID 0.6 1.1 0 0 0.6 2.1 67.4 88.4
DO1 WW1 0.0 0.0 0 0 0.0 0.0 0.0 0.0
DO1 WW2 0.0 0.0 0 0 0.0 0.0 0.0 0.0
DO1 WW3 1.2 1.5 989 989 0.8 2.0 151.9 151.9
DO2 WW1 0.0 0.0 0 0 0.0 0.0 0.0 0.0
DO2 WW2 0.0 0.0 0 0 0.0 0.0 0.0 0.0
DO2 WW3 1.5 2.0 985 985 2.5 2.5 152.4 152.4
DO3 WW1 0.6 1.8 661 1140 1.3 2.1 41.6 41.6
DO3 WW2 0.0 0.0 0 0 0.0 0.0 0.0 0.0
DO3 WW3 1.6 1.1 852 852 1.2 1.2 152.1 152.1
ID WW1 1.5 1.9 757 757 1.6 2.1 152.4 165.9
ID WW2 0.0 0.0 0 0 0.0 0.0 0.0 0.0
ID WW3 0.5 0.7 777 777 0.6 0.8 152.4 154.8
WW1 AG 0.0 0.0 0 0 0.0 0.0 0.0 0.0
WW1 RID 1.2 2.3 183 183 0.0 0.0 0.0 0.0
WW1 LO 0.9 1.4 788 788 0.7 1.0 128.6 217.5
WW2 AG 0.0 0.0 0 0 0.0 0.0 0.0 0.0
WW2 RID 0.0 0.0 0 0 0.0 0.0 0.0 0.0
WW2 LO 0.0 0.0 0 0 0.0 0.0 0.0 0.0
WW3 AG 0.0 0.0 0 0 0.0 0.0 0.0 0.0
WW3 RID 4.8 5.3 113 113 0.0 0.0 0.0 0.0
WW3 LO 0.0 0.0 0 0 0.0 0.0 0.0 0.0
GW AG 0.0 0.0 0.0 0.0 0.0 0.0
GW LO 0.0 0.0 0.0 0.0 0.0 0.0
DO1 GW 0.0 0.0
DO2 GW 0.0 0.0
DO3 GW 0.0 0.0
ID GW 0.0 0.0
RIU GW 0.0 0.0
AG GW 0.0 0.0
LO GW 0.0 0.0
 175

Table B.12. Treatment plant capacities for the multiple wastewater treatment plant
system application
Water Wastewater Wastewater
Wastewater
Capacity (km3/yr) treatment treatment treatment
treatment plant 3
plant plant 1 plant 2
1st design period 0.25 0.13 0.00 0.25
2nd design period 0.25 0.13 0.00 0.25

Table B.13. Construction and operation cost for the multiple wastewater treatment
plant system application
Cost Pipes Canals Pumps Treatment plants
Construction (× 106 $) 116.81 85.40 31.42 20.29
Operation in the 1st period
22.45 0.40 2.67 0.20
(× 106 $/yr)
Expansion (× 106 $) 3.42 0.17 3.97 0.04
Operation in the 2nd period
26.66 0.64 1.92 0.20
(× 106 $/yr)
 176

Table B.14. Water quality change (BOD) (mg/l) for the multiple wastewater
treatment plant system application
Ground Downstream WT DO1 DO2 DO3 ID AG LO
Year
water River effluent influent influent influent influent influent influent
2001 0.040 1.703 0.008 0.040 0.040 0.008 0.011 0.800 1.548
2002 0.008 1.709 0.008 0.008 0.008 0.008 0.008 0.800 1.548
2003 0.002 1.716 0.008 0.002 0.002 0.008 0.007 0.800 1.548
2004 0.001 1.723 0.008 0.001 0.001 0.008 0.007 0.800 1.548
2005 0.000 1.729 0.008 0.000 0.000 0.008 0.007 0.800 1.548
2006 0.000 1.735 0.008 0.000 0.000 0.008 0.006 0.800 1.548
2007 0.000 1.741 0.008 0.000 0.000 0.008 0.006 0.800 1.548
2008 0.000 1.748 0.008 0.000 0.000 0.008 0.006 0.800 1.548
2009 0.000 1.753 0.008 0.000 0.000 0.008 0.006 0.800 1.548
2010 0.000 1.759 0.008 0.000 0.000 0.008 0.006 0.800 1.548
2011 0.000 2.846 0.008 0.000 0.008 0.000 0.006 0.800 1.467
2012 0.000 2.802 0.008 0.000 0.008 0.000 0.006 0.800 1.467
2013 0.000 2.761 0.008 0.000 0.008 0.000 0.006 0.800 1.467
2014 0.000 2.722 0.008 0.000 0.008 0.000 0.006 0.800 1.467
2015 0.000 2.687 0.008 0.000 0.008 0.000 0.005 0.800 1.467
2016 0.000 2.654 0.008 0.000 0.008 0.000 0.005 0.800 1.467
2017 0.000 2.624 0.008 0.000 0.008 0.000 0.005 0.800 1.467
2018 0.000 2.595 0.008 0.000 0.008 0.000 0.005 0.800 1.467
2019 0.000 2.568 0.008 0.000 0.008 0.000 0.005 0.800 1.467
2020 0.000 2.543 0.008 0.000 0.008 0.000 0.005 0.800 1.467
 177

Table B.15. Source storage, river discharge and water quality from wastewater
treatment plants in the multiple wastewater treatment plant system application
Groundwater Downstream river WW1 WW2 WW3
Year elevation Discharge (m3/s) effluent effluent effluent
(m) (mg/l) (mg/l) (mg/l)
2001 520.68 11.88 2.00 0.00 2.00
2002 537.72 12.02 2.00 0.00 2.00
2003 554.04 12.17 2.00 0.00 2.00
2004 569.60 12.32 2.00 0.00 2.00
2005 584.40 12.47 2.00 0.00 2.00
2006 598.41 12.63 2.00 0.00 2.00
2007 611.60 12.79 2.00 0.00 2.00
2008 623.96 12.96 2.00 0.00 2.00
2009 635.46 13.13 2.00 0.00 2.00
2010 646.08 13.31 2.00 0.00 2.00
2011 650.78 11.95 2.00 0.00 2.00
2012 654.38 12.13 2.00 0.00 2.00
2013 656.87 12.32 2.00 0.00 2.00
2014 658.21 12.52 2.00 0.00 2.00
2015 658.37 12.72 2.00 0.00 2.00
2016 657.31 12.93 2.00 0.00 2.00
2017 655.02 13.14 2.00 0.00 2.00
2018 651.44 13.36 2.00 0.00 2.00
2019 646.56 13.58 2.00 0.00 2.00
2020 640.32 13.81 2.00 0.00 2.00
 178

Table B.16. Population in the multiple wastewater treatment plant system

application
Population Population Population
Year
Domestic area 1 Domestic area 2 Domestic area 3
2001 300,000 400,000 600,000
2002 308,100 410,800 616,200
2003 316,419 421,892 632,837
2004 324,962 433,283 649,924
2005 333,736 444,981 667,472
2006 342,747 456,996 685,494
2007 352,001 469,335 704,002
2008 361,505 482,007 723,010
2009 371,266 495,021 742,531
2010 381,290 508,386 762,580
2011 391,585 522,113 783,169
2012 402,157 536,210 804,315
2013 413,016 550,688 826,031
2014 424,167 565,556 848,334
2015 435,620 580,826 871,239
2016 447,381 596,509 894,763
2017 459,461 612,614 918,921
2018 471,866 629,155 943,732
2019 484,607 646,142 969,213
2020 497,691 663,588 995,382
 179

Table B.17. Water demands in the multiple wastewater treatment plant system
application (m3/s)
Domestic area Domestic area Domestic area
Year Industrial area Large outdoor
1 2 3
2001 1.38 1.81 2.66 1.60 2.54
2002 1.42 1.85 2.73 1.64 2.61
2003 1.45 1.90 2.80 1.69 2.68
2004 1.49 1.95 2.87 1.73 2.75
2005 1.53 2.00 2.94 1.78 2.83
2006 1.56 2.05 3.02 1.83 2.90
2007 1.60 2.10 3.10 1.87 2.98
2008 1.64 2.16 3.18 1.93 3.06
2009 1.68 2.21 3.26 1.98 3.15
2010 1.73 2.27 3.35 2.03 3.23
2011 1.77 2.33 3.43 2.09 3.32
2012 1.82 2.39 3.52 2.14 3.41
2013 1.86 2.45 3.62 2.20 3.50
2014 1.91 2.51 3.71 2.26 3.59
2015 1.96 2.58 3.81 2.32 3.69
2016 2.01 2.64 3.91 2.38 3.79
2017 2.06 2.71 4.01 2.45 3.89
2018 2.11 2.78 4.12 2.51 4.00
2019 2.17 2.85 4.23 2.58 4.11
2020 2.22 2.93 4.34 2.65 4.22
 180

10. FIGURES

1
Canal
Pipe/Pump
Canal
Imported water 200 Domestic Area

5
Seepage
Canal

Water treatment plant Pipe/Pump

3 13
7 14
Pipe/Pump
Pump
Pipe/Pump
Canal

Pipe/Pump
6 Groundwater
12
River 16
Upstream Pipe/Pump Wastewater
Seepage
treatment plant
11
Canal 8
400 Agricultural Area
1700
Canal Pump
9
River
DownstreamCanal
1500
Pipe/Pump
Infiltration

18

10
Groundwater

Figure B.1. Single wastewater treatment plant supply system schematic. Bold arcs
represent the 10 decision variables and thin arcs are dependent flows that are
computed from mass balance constraints (Note the number in arcs are correspond
those in Table B.3).
 181

Figure B.2. Multi-wastewater plant system schematic (WT – water treatment plant,
DO1 , DO2, DO3 – the first, second and third domestic areas, respectively, ID –
industrial area, WW1, WW2, WW3 – the first, second and third potential
wastewater treatment plants, respectively, AG – agricultural area, LO – large
outdoor area) Note the number in arcs are correspond those in Table B.10.
 182

Population

Memeplex 1
Sub-memeplex
Sub-memeplex
Sub-memeplex

Memeplex 2
Sub-memeplex
Sub-memeplex
Sub-memeplex

Memeplex 3
Sub-memeplex
Sub-memeplex
Sub-memeplex

Figure B.3. Representation of population, memeplexes, and sub-memeplexes in

SFLA
 183

Figure B.4. Best solution and penalty term change with the number of function
evaluations in the single WWT plant system
 184

Pipe/Pump
D = 600 / 1500 mm
Imported water
Q = 0.70 / 2.88 cms
Water Treatment Plant
3 (0.05 / 0.1 km3/yr) 5
Canal
W=1m
Canal
W=1m
Q = 0.34 / 2.07 cms Pump 9 Groundwater
Qp = 0.24 / 0.55 cms
Q = 0.39 cms Hp = 145 m Pump
Q = 0.36 / 0.80 cms Qp = 5.13 / 8.77 cms
Hp = 87 m
Q = 7.67 cms 10
River 4
Agricultural Area
Upstream Canal
W=1m
Pump
11 Q = 0.59 cms
Pump D = 600 / 900 mm
D = 200 / 200 mm Qp = 0.68 / 0.69 cms 7
800 Hp = 61 / 74 m
Q = 0.02 cms
Q = 1.03 cms

River Wastewater Treatment Plant

Downstream
1500 (0.1 km3/yr)
Pipe/Pump
Recharge

D = 1200 mm Pipe/Pump
Infiltration Qp = 0.74 / 0.93 cms D = 900 mm
0.1 kafy 18 Hp = 152 / 183 m Qp = 0.74 / 1.03 cms
Q = 1.08 / 1.40 cms 14 Hp = 107 m
Q = 1.11 / 1.42 cms

Groundwater Domestic Area

Figure B.5. Optimal solution of single wastewater plant system application showing
the flow allocations and arc capacities/design variables – results from first design
period / second design period, respectively. (Note the single value indicates no
expansion in the capacity or flow allocations)
 185

Pump
Qp = 1.7 / 2.0 cms Pipe/Pump
Hp = 149.1 m D = 989 mm
Pipe/Pump
10 Q = 1.2 / 1.5 cms Qp = 0.8 / 1.2 cms
D = 788 mm
DO1 Hp = 151.9 m
Qp = 0.7 / 1.0 cms
Q = 1.2 / 1.5 cms
Pipe/Pump 40 Hp = 128.6 / 217.5 m
15 D = 661 / 1140 mm Q = 0.9 / 1.4 cms
WW1
Qp = 1.3 / 2.1 cms
Hp = 41.6 m
Pump Pipe/Pump Q = 0.6 / 1.8 cms
27
Qp = 0.6 / 2.1 cms D = 985 mm
Hp = 67.4 / 88.4 m Qp = 2.5 cms Pipe/Pump
Q = 0.6 / 1.1 cms Hp = 152.4 m D = 757 mm
11 DO2 Q = 1.5 / 2.0 cms Qp = 1.6 / 2.1 cms
Hp = 152.4 / 165.9 m
18 Q = 1.5 / 1.9 cms
Pipe/Pump
D = 999 mm
Qp = 2.4 / 2.8 cms 20
WT Hp = 104.4 m WW2
8 Q = 2.4 / 3.0 cms DO3

22

WW3
Pipe/Pump
D = 852 mm Pipe/Pump
Pipe/Pump D = 183 mm
Qp = 1.2 cms 23
D = 768 mm
Qp = 1.2 / 2.0 cms
Hp = 152.1 m
24 42 Q = 1.2 / 2.3 cms
Q = 1.6 / 1.1 cms
Hp = 152.4 / 243.8 m Pipe/Pump Pipe/Pump
9 Q = 1.4 / 1.6 cms D = 777 mm D = 113 mm
ID
Qp = 0.6 / 0.8 cms Q = 4.8 / 5.3 cms Canal
Hp = 152.4 / 154.8 m W=3m
1 Q = 0.5 / 0.7 cms Q = 0.6 / 1.1 cms
LO
Canal
4 5
W=5m AG
Q = 4.0 / 4.5 cms Canal
W = 10 m
Q = 5.8 / 8.6 cms
River upstream River downstream

48 Recharge from DO1, DO2, DO3, ID, AG, and LO

Recharge
Pumping Groundwater Pumping

Figure B.6. Optimal solution of multiple wastewater plant system application

showing the flow allocations and arc capacities/design variables – results from first
design period / second design period, respectively. (Note the single value indicates no
expansion in the capacity or flow allocations)
 186

Figure B.7. Optimal system cost and penalty term changes with the number of
function evaluations for the multiple WWT plant system.
 187

APPENDIX C: RELIABLE WATER SUPPLY SYSTEM

DESIGN UNDER UNCERTAINTY

C. Graph
 188

Reliable Water Supply System Design under Uncertainty

G. Chung1, K. Lansey2, and G. Bayraksan3

ABSTRACT

Long term reliability is the most important design factor for water supply systems. Water

supply systems are particularly impacted by uncertain future conditions. Many research

efforts have attempted to account for data uncertainty while simultaneously improving

economical feasibility and attaining system reliability. However, the large problem size

and the correlated uncertainties make solving the problem difficult to solve. To consider

correlated uncertainties in water demand and supply, this study applies the robust

optimization approach of Bertsimas and Sim to a water supply system design problem.

Robust optimization aims to find a solution that remains feasible under data

uncertainty. For instance, a water supply system will be “robust” so that it can meet

demand under extreme drought conditions. However, such a system can be too

conservative and costly. It is possible to vary the degree of conservatism to allow for a

decision maker to understand the trade-off between system reliability and economical

feasibility/cost.

In this study, the uncertainty factors are controlled by the degree of conservatism such

that the system stability is guaranteed under uncertain conditions. The degree of

1
Graduate Student, Department of Civil Engineering and Engineering Mechanics, The University of Arizona, Tucson,
AZ 85721, USA (Tel: 1-520-360-9554, E-mail: gunhui@email.arizona.edu)
2
Professor, Department of Civil Engineering and Engineering Mechanics, The University of Arizona, Tucson, AZ
85721, USA (Tel: 1-520-621-2512, Fax: 1-520-621-2550, E-mail: lansey@engr.arizona.edu)
3
Assistant Professor, Department of Systems Industrial Engineering, The University of Arizona, Tucson, AZ 85721,
USA (Tel: 1-520-621-2605, E-mail: guzinb@sie.arizona.edu )
 189

conservatism is presented as a form of the probability bound of constraint violation. As a

result, the total cost increases as the degree of conservatism is increased, i.e., the

probability bound of constraint violation is decreased. A trade-off appears to exist

between the level of conservatism and imported water purchase, i.e., cost increase. It was

found that the robust optimization approach can be a useful tool to find a solution that

prevents system failure at a certain level of risk within the available budget.
 190

1. INTRODUCTION

A municipal water supply system is defined as the physical infrastructure to treat,

deliver water to and collect water from users. The capacities of alternative components

are based upon predictions of future population and climatic conditions. Uncertainty in

predicting these conditions is inherent in all water supply systems. Thus, to reduce the

risk of failure during future operations, it is desirable to consider these uncertainties

during the planning process. A decision made with a deterministic model may result in

two consequences; lower net benefits than expected (i.e., it is more costly to provide the

desired water) or some probability of system failure, where failure is defined as not

meeting a given demand or other system constraint (Watkins and McKinney 1997).

These consequences may be rectified in real-time operations at some cost but flexibility

must be built into the system during the design process to allow for those adjustments.

Deterministic optimization that is based on satisfying demand/supply conditions without

consideration of uncertainty removes this flexibility. Thus, a reliability-based design tool

is needed that can assist decision makers plan a long-term water supply scheme to cope

with the future changes in water demands and supplies.

The complexity of the system and the correlated uncertainties make incorporating

uncertainty a challenging exercise. A number of stochastic optimization approaches have

been applied to water supply system design and operation. Most works have adopted

multi-stage linear or nonlinear optimization techniques. The main objectives of these

studies were to minimize expected total cost for water transfer to spot-markets (Lund and
 191

Israel 1995); to develop long- as well as short-term water supply management strategies

(Wilchfort and Lund 1997); to manage water supply capacity under water shortage

conditions (Jenkins and Lund 2000); and to design and operate a water supply system

(Elshorbagy et al. 1997).

Some water supply optimization studies have considered the aspect of system failure

risk. For example, Fiering and Matalas (1990) investigated the robustness of water supply

planning with respect to global climate change for regions where water storage capacity

is limited. Watkins and McKinney (1997) considered uncertainty factors by introducing

the standard deviation of the objective function as a constraint into a two-stage stochastic

model by Lund and Israel (1995). This is embedded in the robust optimization framework

of Mulvey et al. (1995).

Chance constraint modeling intends to limit decisions more directly by considering

uncertainty in model input. For instance, chance constraint model may explicitly limit the

probability of not being able to meet a constraint. Chance constraint models, while

intuitively easy to model, are usually non-convex causing difficulties in optimization and

the approach requires numerical integration of the probability distribution or, if an

invertible probability distribution is assumed to hold, has difficulty considering parameter

correlations.

In this paper, the robust optimization framework of Bertsimas and Sim (2004) is used

to develop a reliable water supply system design. A robust solution can be defined as one

that remains feasible under uncertainty. This type of robust optimization was first

introduced by Soyster (1973) to solve linear programming problems. Soyster’s model,

 192

which is linear, significantly constrains the objective function to assure robustness; thus

conservative solutions are found that may be practically unrealistic. Ben-Tal and

Nemirovski (1999 and 2000), El-Ghaoui and Lebret (1997), and El-Ghaoui et al. (1998)

extended the Soyster model. These extensions, however, introduced a higher degree of

non-linearity. Since real systems themselves are likely to be nonlinear, these approaches

make the problem more complicated and difficult to find a solution. The approach of

Bertsimas and Sim controls the degree of conservatism for the system reliability without

increasing the difficulty in solving the original problem.

2. ROBUST OPTIMIZATION FRAMEWORK

The classical assumption in deterministic mathematical programming is that all

parameters (input data) are known precisely and can be represented by some nominal

values. This is rarely the case in real applications since many parameters contain

uncertainties such as in measurement and/or uncertainties due to future. One way to deal

with uncertainty is to design a system that is “robust” to changes in the parameters. That

is, the system remains feasible and operates in a near-optimal fashion for a variety of

values that the uncertain parameters can take. Soyster (1973) formulated the following

deterministic linear programming model to find a solution that is feasible for all uncertain

data belonging to a convex set:

maximize cx

n
subject to ∑ a~ x ij j ≤ bi , a~ij ∈ K j , j = 1,..., n , ∀i (C.1)
j =1
 193

x j ≥ 0 , ∀j

where K j is a nonempty convex set and it considers “columnwise” uncertainty,

A • j ∈ K j ; i.e., A • j = {a~1 j , a~2 j ,..., a~mj }, i = 1,..., m .

Here, without loss of generality only the coefficients of the inequality constraints, a~ j ,

that are a subset of Kj contain uncertainty. This model introduces hard constraints (i.e.,

ones that must be satisfied) for all subsets of Kj. As a result, the optimal solution may

sacrifice a significant portion of the optimality of a nominal problem (i.e., deterministic

problem with mean parameter values, a j ) to guarantee robustness. Thus, problem

solutions can be quite conservative.

Hard constraints are very important and must be met in some engineering problems,

such as dam or embankment designs. In these cases, structural failure can cause

significant damages. Hence, the stability of the structure is the primary concern and it is

written as a hard requirement. Conservatism to guarantee system reliability will, however,

increase the total cost for the construction and operation. In contrast, some flexibility can

be incorporated into some planning models to find the best economical options where

rigidness of the model may not be appropriate.

To relax the problem, Ben-Tal and Nemirovski (1999 and 2000) deals with the

uncertainty associated with the constraints in a different manner. For uncorrelated

variables, they introduce data uncertainty in the form of random perturbations as follows:

a~ij = aij + εη~ij aij = aij + η~ij aˆij (C.2)

 194

where a~ij is the mean value, ε > 0 is a given uncertainty level and η~ij are random

variables that are symmetrically distributed within the interval of [-1, 1]. A norm of

multiplication of the nominal value, aij , and the uncertainty level, ε, is represented by âij

that is assumed to be a bounded, symmetric (but not necessarily uniform) random

variable. The range of the uncertain parameter a~ij is [aij − aˆij , aij + aˆij ] . Random

perturbations affecting the uncertain data entries j of a particular inequality constraint i

are assumed identically and independently distributed (iid).

Ben-Tal and Nemirovski (2000) then modified Soyster’s model by introducing

additional variables, y and z, as follows:

maximize cx

subject to ∑ a x + ∑ aˆ
j
ij j
j∈J i
ij yij + Ω i ∑ aˆ
j∈J i
z ≤ bi
2 2
ij ij ∀i (C.3)

− yij ≤ x j − zij ≤ yij ∀i, j ∈ J i

l≤x≤u

y ≥ 0,

where Ji is the set of indices of the uncertain data elements in constraint i and Ω i is a

user-defined positive conservatism control parameter for each constraint i. Ben-Tal and

Nemirovski (2000) showed that the problem is feasible in (x, y, z) with the ith constraint
2
being violated with a probability of at most e ( − Ωi / 2 ) . Although the degree of conservatism

can be controlled, this approach results in a conic quadratic problem that is

computationally harder to solve than the original LP.

 195

To overcome this computational difficulty, Bertsimas and Sim (2004) develop a new

approach that retains the linearity of Soyster’s model while controlling the level of

conservatism. Consider the following stochastic optimization problem:

maximize cx

n
subject to ∑ x a~
j =1
j ij ≤ bi ∀i, j ∈ J i (C.4)

l≤x≤u

Uncertainty is modeled in the same way as Ben-Tal and Nemirovski. That is, for the

ith constraint, J i represents the set of indices that correspond to uncertain a~ij . a~ij , j ∈ J i ,

are assumed to be independent, symmetric and bounded random variables as given in Eq.

C.2.

To control the degree of conservatism, Bertisimas and Sim introduce an additional

parameter, Γi , that can take any real value within the range of [0, J i ] , in a manner that

the most significant coefficients up to the ⎣Γi ⎦ th order is fully allowed to vary within their

uncertainty intervals and the ( ⎣Γi ⎦ +1)th order significant coefficient, ait is bounded by

(Γi − ⎣Γi ⎦)âit , while the remaining coefficients are fixed at their nominal values. Then,

Eq. C.4 is reformulated in a robust form to improve the system reliability:

maximize cx

subject to

⎧ ⎫
∑a x ij j + max
{ Si U{ti }|Si ⊆ J i , Si = ⎣Γi ⎦,ti∈J i \ Si }
⎨ ∑ aˆij y j + (Γi − ⎣Γi ⎦)aˆiti yt ⎬ ≤ bi ∀i (C.5)
j ⎩ j∈Si ⎭
 196

− yj ≤ xj ≤ yj ∀j ∈ J i

l≤x≤u

y≥0

At optimality, y j = x*j for all j.

The nominal (deterministic) problem would have constraint ∑a xij j ≤ bi from the

inequality constraint in Eq. C.4. In Bertismas and Sim’s robust formulation, however, an

additional max-term keeps the system feasible. The degree of conservatism (i.e., the

degree of uncertainty) is represented by Γi in the max-term. Note that when Γi = 0, the ith

constraint is equivalent to that of the nominal problem, ∑a x

ij j ≤ bi , and when Γi = J i ,

the ith constraint is completely protected against uncertainty. As the max-term in the

constraint increases, the nominal-term must be reduced to satisfy the constraint bound, bi.

In other words, system conservatism will have an increasing effect on the max-term as Γi

value is increased and accordingly the capacity of system component, x j , will be reduced

to satisfy the constraint. In this manner, the tradeoff between the degree of conservatism

and corresponding system capacity, i.e., system economic feasibility, can be evaluated.

The above formulation assumes that the uncertain coefficients are independent. This

assumption is unlikely to be valid in many systems so an extended uncertainty model for

correlated random variables was proposed by Bertsimas and Sim (2004). Suppose a

number of different sources of uncertainty affect the system, the randomness in the ith

constraint can then be represented as:

 197

a~ij = aij + ∑η~ik g kj , ∀j ∈ J i (C.6)

k∈Ki

where η~ik are independently and symmetrically distributed random variables in the range

[-1, 1], K i is number of uncertainty sources that affect the coefficients in the ith

constraint, aij is the nominal value of a~ij , and as before J i is the set of indices of

uncertain parameters in the ith constraint that is subject to uncertainty. If K i = 2, all

uncertain coefficients in the ith constraint are affected by two bounded random variables

by the nominal values of g kj , k ∈ K i . Thus, each uncertain parameter, aij , j ∈ J i in the

ith constraint is affected by g kj , k ∈ K i .

The robust model with correlated uncertain coefficients can then be rewritten as

(Bertsimas and Sim, 2004):

maximize cx

subject to

⎧⎪ ⎫⎪
∑a x
j
ij j + max
{ Si U{ti }|Si ⊆ K i , Si = ⎣Γi ⎦,ti∈Ki \ Si }
⎨ ∑ ∑
⎪⎩k∈Si j∈J i
g kj x j + (Γi − ⎣Γi ⎦) ∑ g ti j x j ⎬ ≤ bi
j∈J i ⎪⎭
∀i (C.7)

l≤x≤u

The problem solution is affected in a similar manner as Eq. C.5 with the objective

function value decreasing with increasing Γi . Correlation effects will increase the max-

term to further reduce the objective function value. With this formulation, the max-term

can be represented as a linear program and thus, the robust solutions of linear problems

can be obtained by solving LP problems without an increased complexity. The degree of

 198

conservatism can be controlled using the robust optimization approach (Eqs. C.5 or C.7)

by varying Γi . The water supply study herein includes nonlinear constraint functions but

only linear terms in the nonlinear constraints were considered as being uncertain.

The level of conservatism, Γi , is a useful tool to investigate system robustness against

failure. If system failure can be presented as a probability, it would give better

understanding of system safety. Bertismas and Sim (2004) relate Γi to a probability level

and show various probability bounds of constraint violation. Below, we review a bound

(Eq. 18, Bertismas and Sim (2004)) that is easy to compute and that can also be relatively

tight. Let x* be an optimal solution to Eq. C.5, then the probability that the ith constraint

is violated satisfies:

⎛ ⎞
Pr ⎜⎜ ∑ a~ij x *j > bi ⎟⎟ ≤ B(n, Γi ) (C.8)
⎝ j ⎠

with:

n
B(n, Γi ) ≤ (1 − µ )C (n, ⎣v ⎦) + ∑ C (n, l ) (C.9)
l = ⎣v ⎦+1

Γi + n
where n = K i , ν = , µ = ν − ⎣ν ⎦ and
2

⎧1
⎪ 2 n , if l = 0 or l = n,
⎪
C (n, l ) = ⎨ (C.10)
⎛ ⎛ n ⎞ ⎛ n − l ⎞ ⎞⎟
⎪ 1 n
exp⎜⎜ n log⎜⎜ ⎟⎟ + l log⎜ ⎟ , otherwise.
⎪⎩ 2π (n − 1)l ⎝ ⎝ 2(n − l ) ⎠ ⎝ l ⎠ ⎟⎠

For the model with correlated data (Eq. C.7) and n = K i , the bound is calculated in a

similar fashion.
 199

To compute Γi values, the probability level in Eq. C.8 and B(n, Γi ) is defined for

each of the i uncertain constraints. Assuming Eq. C.9 is tight, Γi can be directly

computed using Eq. C.10. Each uncertain constraint is considered independently to

determine its corresponding Γi . With the set of Γi , the optimal solution for the defined

probability level is determined by solving problem C.5 or C.7. A range of probability

levels can be evaluated to provide the decision maker the tradeoff between robustness

and cost.

3. APPLICATION TO WATER SUPPLY SYSTEM

3.1 Problem Statement

The robust optimization methodology is applied to a realistic hypothetical water

supply system (Figure C.1). The problem objective is to minimize the total cost for the

construction, expansion, and operation of the system. The appendix includes a complete

list of indices, sets, input parameters, and variables. To model the water supply system,

we have a graph G = (N, A), where N is the set of nodes and A is the set of arcs (Figure

C.1). N consists of 8 nodes: sources (NS - imported water, aquifer, upstream river, and

downstream river), users (NU - domestic and irrigation), and water and wastewater

treatment plants (NWT and NWWT, respectively). Note a river is represented by upstream

(NRU) and downstream (NRD) nodes and a connecting arc. Sources are divided into two

subsets: storage sources (NSS) and non-storage sources (NNS) depending on the capability
 200

of storing water from the previous time step. For instance, a groundwater aquifer is a

storage source while imported water and upstream and downstream rivers are non-storage

sources.

The system can be constructed or expanded in year t and operational time is

represented by o. T and O are the set of construction/expansion and operational times. A

15-year planning period is evaluated and new facilities may only be constructed in years

1 and 6. Existing infrastructure is in place at year 0. Operation variables ( qijo ) over arcs (i,

j) during each operation period, o, are determined each year ( ∆o = 1 ) for the first 5 years

(first design period) and every other year ( ∆o = 2 ) for the last 10 years (second design

period).

In out model, A consists of 18 arcs (i, j) that connects nodes i and j. Arcs represent

canals (AC − arcs 1, 2, 3, 4 and 5 in Figure C.1), pipe lines (AP −arcs 8, 9, 10 and 11),

pump stations (AU − arcs 6 and 7), rainfall or mountain frontal recharge (AR – arcs 12, 13,

14 and 15), and seepage or infiltration (AI – arcs 16, 17 and 18). Pump stations are

located in arcs to overcome friction losses and elevation differences through pipe

connections. To permit water banking, arc 4 represents a canal to carry imported water to

recharge basins.

Before providing the mathematical model details, an overview of model equations is

presented in verbal terms. The objective function and system constraints are:

Minimize

f1 (κ ijt ) - construction cost of pipe arcs (i, j) in expansion year t

 201

+ f 2 (d ijt ) - construction cost of canal arcs (i, j) in expansion year t

+ f 3 ( χ ijt , H ijt ) - construction cost of pump on arcs (i, j) in expansion year t

+ f 4 ( wit ) - construction cost of treatment plant i in expansion year t

+ f 5 (qijo , wio ) - operation and maintenance cost of the system in operation year o

+ f 6 ( IW o ) - water purchased cost from outside of basin in operation year o

Subject to

1 Meet water demand for water users (domestic and agricultural area)
○

2 Do not exceed water and wastewater treatment plant capacity

○

3 Restrict the amount of imported water to the external water availability

○

4 Satisfy conservation of mass in non-storage nodes, users, and treatment plants.

○

5 Meet required river discharge at downstream river node

○

6 Meet required groundwater storage to maintain a sustainable system

○

7 Satisfy conservation of mass in storage nodes

○

8 Limit canal flow by maximum canal capacity

○

9 Maintain pump operating efficiency

○

10 Meet minimum pressure requirement at the end of pipe arcs

○

11 Meet minimum pressure requirement at the end of pump arcs

○

Design decisions are the arc or node capacities, such as pipe sizes, pump design head

and flow rate, canal depth, and water and wastewater treatment plant capacities. Flow

allocations over the water supply network are operational control variables.
 202

3.2 Objective Function

The objective function is to minimize the total cost for construction and operations

and maintenance (O&M) for the system components (pipes, canals, pumps, and treatment

facilities):

Minimize Z * = f1 (κ ijt ) + f 2 (d ijt ) + f 3 ( χ ijt , H ijt ) + f 4 ( wit ) + f 5 (qijo , wio ) + f 6 IW o ( ) (C.11)

Next, we will explain how we determined the terms of the objective function in more

detail. For the first term, the pipe construction cost, we use the construction cost function

developed by Clark et al. (2002) in terms of pipe diameters that connect nodes i and j at

year t ( κ ijt ). Then, the pipe construction cost is given by:

f1 (κ ijt )
= ∑t∈T ∑i∈A ∑ j∈A P
{
xijt 57.198 + 0.35κ ijt + 0.62κ ijt
1.54
+ 0.0018κ ijt
1.9
+ 0.0062κ ijt
1.83

}L
P

0.73 1.8 0.93 0.71

− 0.062κ ijt + 0.02κ ijt + 0.23κ ijt + 0.0022κ ijt ij

(C.11a)

The product of the pipe length, Lij , and the constant term in the brackets in Eq. C.11a

gives a positive cost even when the pipe diameter, κ, is zero (i.e., no connection is

desired). Therefore, a binary variable ( xijt ) is added to the model to indicate the decision

of construction, thus existence, of a pipe from node i to j at time t.

Canal flows are driven by gravity and canal construction cost is a function of the

channel depth ( d ijt ) (US. Army Corps of Engineers 1980):

f 2 (d ijt ) = ∑t∈T ∑i∈A ∑ j∈I C

cij d ijt
C

⎧ t ENR CITY ⎫
t (C.11b)
= ∑t∈T ∑i∈A ∑ j∈AC
1.45⎨ 55.30 L d
ij ij ⎬
C
⎩ 2877 ⎭
 203

The rated pump head ( H ijt ) and discharge ( χ ijt ), that define the most efficient pump

operation point, are the pump design variables. The pump construction cost function

given in Walski et al. (1987) is adopted in this study and the total cost was summed over

the set of pumps ( A P U A U ):

f 3 ( χ ijt , H ijt ) = ∑t∈T ∑i∈A

P UAU
∑ j∈A P U A U
(500χ t 0.7
ij H ijt
0.4
) (C.11c)

Water and wastewater treatment facility construction and expansion costs are

computed from their capacities ( wit ) (Tang et al. 1987):

f 4 ( wit ) = ∑t∈T ∑i∈N

WT
(2897.13 w t
i )
+ 35987 + ∑t∈T ∑i∈N
WWT
(10811.92 w
t
i + 5454228 )
(C.11d)

All costs are converted to the equivalent present values for year 0 by applying the

present worth factor of 1/(1+I)yr-1 where I is the interest rate. Operations and maintenance

(O&M) costs are calculated for each operation period, o, by:

10
f 5 (qijo , wio ) = ∑ ⎢
⎡ 1
o =1 ⎣ (1 + I )
[
o −1 ∑i∈A P ∑ j∈A P ij
x o (27.7 + 0.3qijo ) Lij

⎛ ENR o ⎞
+ ∑i∈A ∑ j∈A ⎜⎜ 0.0254 Lij qij o 0.572
+ (0.078 + 0.0135qij ) Lij
o
⎟
C C
⎝ 1850 ⎟⎠ (C.11e)
⎧ 0.935 ENR CITY ⎫
o
+ ∑i∈A UA ∑ j∈A UA ⎨79.47∆ ij qijo + 4560qijo + 320qijo
0.58
⎬
P U P U
⎩ 2877 ⎭
+ ∑i∈N
WT
(28.97 w o
i )
+ 360 + ∑i∈N
WWT
( 108.12 w o
i + 54542 ∆o o )]
The five terms in Eq. C.11e correspond to the O&M cost terms for the pipes, canals,

pumps, water treatment and wastewater treatment plants, respectively. The parameters

∆ ij in the pump term are the elevation differences between the pipe endpoints.
 204

CITY and ENR are parameters that account for local cost variations and the inflation

rate, respectively. The ENR factor for year t is computed by:

ENR t = -7.7 × 109 + 15.7t − 12.0 × 103 t 2 + 4.1t 3 − 0.5 × 10 −3 t 4

Water ( IW o ) can be purchased and imported to the basin at a unit cost of C IW . A

time step factor (∆oo) accounts for the variable decision period durations. The imported

water cost then becomes:

10
⎡ 1 ⎤
f 6 ( IW o ) = ∑ ⎢ C IW ∑ j = IW q oji ⎥∆o o (C.11f)
o =1 ⎣ (1 + I )
o −1
⎦

3.3 Simple Decision Bounds

Simple decision variable bounds and the pipe length constraints are:

κ ijt ≥ ε ∀ A P , ∀T (C.12)

H ijt ≥ 0 ∀ A P U A U , ∀T (C.13)

d ijt ≥ 0 ∀ AC (C.14)

χ ijt ≥ ε ∀ A P U A U , ∀T (C.15)

wit ≥ 0 ∀ N WT U N WWT , ∀T (C.16)

qijo ≥ 0 ∀A , ∀O (C.17)

qijo ≤ M ijo xijo ∀ AP (C.18)

qijo ≤ M ijo µijo ∀ AU (C.19)

Eqs. C.12, C.13, C.14, C.15, C.16 and C.17 represent the lower bounds of the system

variables for pipe diameter, pump design head, canal depth, pump design capacity, water
 205

and wastewater treatment plant capacity, and the flow allocations (control variable),

respectively. The lower bound on pipe diameter and pump capacities (Eq. C.14) is

assigned a small value (10-5) instead of zero to avoid numerical error because Eqs. C.32

and C.33 cannot be divided by zero and keep reasonable objective function values. In

addition to simple bounds on decision variables, constraints on flows at nodes and

through arcs are also necessary. The terms, M ijo , in Eqs. C.18 and C.19 are assigned large

values or upper bounds on the corresponding flows. If the binary variables ( xijo and µijo )

are set to zero, the control variable for the corresponding arc must be zero. Otherwise,

flowrates can be allocated to those arcs.

3.4 Flow Constraints through Nodes

~
Water demands for each operation period o ( Dio ), must be satisfied for each demand

center i (agriculture and domestic areas) by supply from upstream sources, j ∈ N S :

~
∑ j∈NS
q oji ≥ Dio , ∀i ∈ N U , ∀O (C.20)

Similarly, the total flow through a water or wastewater treatment plant cannot exceed

the plant capacity, wit or:

∑ j∈N
q oji ≤ wit , ∀i ∈ N WT U N WWT , ∀t ∈ T ≤ ∀o ∈ O (C.21)

The amount of imported water inflow to the system is limited by external water
~
availability ( IW o ) and is computed as the sum of the outflows from the imported water

node (IW) or:

 206

~
o
qIWj ≤ IW o , ∀j ∈ N U U N S , ∀O (C.22)

~
Natural runoff resulting from precipitation ( P ) on the upstream watershed that has

area, Ab, contributes to the river flow and aquifer. Assuming 60% of the precipitation is

lost and evaporates, 30% of the rainfall is assumed to be an inflow to the upstream river

node and 10% of rainfall recharges the aquifer. The volumes of streamflow and aquifer

recharge are the product of the precipitation and the contributing area or:
~
qijo = 0.3P o Ab , i = precipitation, j = upstream river ∀O (C.23)

~
qijo = 0.1P o Ab , i = precipitation, j = groundwater ∀O (C.24)

By conservation of mass, inflows and outflows at some non-storage nodes must

balance. These equality constraints are written for users (NU), water treatment plants

(NWT), and wastewater treatment plants (NWWT) or:

∑ j
q oji − ∑ j qijo = 0 , i ∈ (N NS \ N RD ) U N WT U N WWT ∀O (C.25)

To account for changing stream conditions and the location of inflows and outflows

along a river, rivers are modeled with an upstream and downstream node. Both are non-

storage nodes. Flow through upstream nodes must balance therefore they are included in

Eq. C.23 constraints. Both upstream and downstream nodes must supply a minimum

downstream flow, RQ, to satisfy environmental requirements or:

∑ j
q oji − ∑ j qijo ≥ RQi , i ∈ N NS , ∀O (C.26)

Storage nodes (groundwater aquifers and surface reservoir (not shown in this

application)) retain water over time. To maintain a sustainable system, water storage, WS,

must exceed a required volume, RS, for all storage sources, Nss and times:
 207

WSio ≥ RSi , i ∈ N SS , ∀O (C.27)

where WSio is computed for every operation time step ( ∆o )for the storage node by

conservation of mass:

WSio = WSio−1 + (∑ q j
o
ji )
− ∑ j qijo ∆o o , i ∈ N SS , ∀O (C.28)

By modifying the precipitation coefficient in Eq. C.28, this constraint can be also

written for a surface storage reservoir.

3.5 Flow Constraints through Arcs

Arc flows are based on hydraulic relationships that introduce a series of constraints

on the flows. A canal’s capacity is estimated using Manning’s open channel flow

equation, the defined channel hydraulic characteristics (slope (S), roughness (n), canal

side slope (z) and geometry), and channel depth ( d ijo ) that is a decision variable. Flow in

each canal in the set AC during each time period must be less than its capacity or:

8
1.49 ⎛ 1
2 ⎜ 1 + zij − zij ⎞⎟d ijo 3 Sij 2 , ∀(i, j ) ∈ A C , ∀O
2
qijo ≤ (C.29)
nij ⎝ ⎠

To maintain a reasonable pump operating efficiency, the pump discharge flow rates

must be maintained between 50 and 150% of the pump design capacity ( χ ijt ) for all pump

arcs and for all operational times or:

qijo ≥ 0.5 χ ijt , ∀(i, j ) ∈ A U , ∀t ∈ T ≤ ∀o ∈ O (C.30)

qijo ≤ 1.5 χ ijt , ∀(i, j ) ∈ A U , ∀t ∈ T ≤ ∀o ∈ O (C.31)

 208

Pipelines carry all potable supplies and whenever flow is required to move up-

gradient. A complete distribution system is too complex to model in this formulation so

the arc to domestic users is intended to be representative of that system. Pipeline arcs

must provide water to the downstream node with a minimum energy. Here, a minimum

pressure head ( H min , j ) of 14.0 m (equivalent to 137.9 kPa or 20 psi) was required. If the

upstream elevation head is insufficient to overcome friction and provide that downstream

pressure, a pump station may be installed.

Conservation of energy is written for the pipeline arc or:

⎡ 4 H ijt qijo
2
8 f Lij o 2 ⎤
⎢ ( H ij −
t
2
)− q − ∆ ij ⎥ ≥ H min , j − 10000(1 − µijt )
5 ij
o

⎢⎣ 3 3χ ij t
g π κ ij
2 t
⎥⎦
i

∀(i, j ) ∈ A P , ∀t ∈ T ≤ ∀o ∈ O (C.32)

The upstream node is assumed to be a free surface (no pressure or velocity head) and

flow is carried to the downstream location at an elevation difference of ∆oij . Pipe friction

losses are computed using the Darcy-Weisbach equation assuming fully turbulent flow

and a constant friction factor ( f ). Finally, the energy added by the pump is given by the

first two terms on the left hand side of Eq. C.32. This representative pump curve

equation (Walski et al. 1987) is a function of the design variables; pump head and

discharge ( H ijt and χ ijt ). When necessary to meet this requirement, a pump station may

be constructed by setting µ equal to 1 that allows flow allocation to this arc to be non-

zero (Eq. C.19).

 209

Flow may also be pumped directly from the aquifer to a user (e.g., agricultural users).

In this case, the pipeline friction loss terms are omitted from Eq. C.32 resulting in:

t o 2
4 t H ij qij
( H ij − 2
− ∆oij ) ≥ H min , j − 10000(1 − µijt ) ∀(i, j ) ∈ A U , ∀t ∈ T ≤ ∀o ∈ O (C.33)
3 3χ ij
t

3.6 Data Uncertainty and Robust Formulation

Predictions of future conditions inherently involve uncertainty. The most significant

uncertainties for a water supply system are the water demands and supplies that arise

from the predictions of future population and precipitation, respectively. The

uncertainties are complicated by correlation between the variables. Consumptive use and

imported water are dependent on the amount of precipitation. It is likely that during

drought conditions, water demand, particularly consumptive use, will increase while less

water will be available. Precipitation appears directly in the relationships for estimating

streamflow and groundwater storage (Eqs. C.26 and C.27). In this study, uncertainties in
~ ~
parameters of precipitation ( P o ), water demand ( Dio ) and imported water availability

~
( IW o ), and their correlation with precipitation are considered.

All random variables are assumed to be bounded symmetric distributions. For

instance, water demand for agricultural area has a lower bound and an upper bound and is

a random variable between these boundaries. Note that non-symmetric distribution could

also be modeled. As an independent random variable, the precipitation is expressed as:

~
P o = P o + η~1o Pˆ o , ∀O (C.34)
 210

where P o is the nominal precipitation in operational period o, P̂ o is half of the

precipitation interval that is assumed to be 10% of nominal precipitation, and η~1o is a

random variable in the interval [-1, 1]. Therefore, the range of precipitation is

[P o
]
− Pˆ o , P o + Pˆ o .

User demand can be expressed in a similar manner. Domestic and agricultural area

demand takes values according to a bounded symmetric distribution with mean equal to

the nominal value of Dio , its half interval, D̂io , and its correlation with precipitation,

− ρP̂ o Ai . Assuming that water demands are random variables and are correlated with

precipitation, we can write:

~
Dio = Dio + η~i Dˆ io − η~1o ρPˆ o Ai ∀i ∈ N U , ∀O (C.35)

In particular,
~
D5o = D5o + η~2 Dˆ 5o − η~1o ρPˆ o ADO _ OUT domestic area (C.35a)

~
D6o = D6o + η~3 Dˆ 6o − η~1o ρPˆ o AAG agricultural area (C.35b)

where ADO _ OUT and AAG are the outdoor land areas in domestic and agricultural irrigation.

~ ~
D5o and D6o are total water demand in domestic and agricultural area, respectively. Water

demand in domestic area is calculated based on population at time o. Since population is

assumed to increase, domestic water demand rises and the range of random parameter is

increased as well because the half interval of random parameter is set as 10% of nominal

value. It is a realistic assumption because future prediction should have more uncertainty

than the present value. Correlation effect from precipitation to user demand is considered
 211

in the third term in Eqs. C.35a and C.35b. Since rainfall reduces domestic outdoor water

demand and irrigation amount in agricultural area, outdoor acreage in domestic area and

agricultural area are included in the correlation effect terms. The outdoor area for the

domestic user is assumed to be a fraction of the total domestic area and η~2 and η~3 are

random variables in the interval [-1, 1].

Finally, the imported water availability has random perturbation and correlation (with

precipitation) components:
~
IW o = IW o + η~4 IWˆ o + η~1o Pˆ o Ab' , ∀O (C.36)

where Ab' is the area of the basin producing the external water supply. The bar term

( IW o ) denotes the nominal imported water while the hat variable ( IWˆ o ) is the half of

the range of the variable perturbation that is assumed to be 10% of nominal values. The

random variables, η~4 , are generated in the interval of [-1,1].

As indicated by the sign on the precipitation term in Eq. C.35, water demands have a

negative correlation with precipitation while the positive sign in Eq. C.36 denotes the

positive correlation between imported water availability and precipitation. That is, during

periods of high precipitation within the basin, user demand decrease. Similarly

precipitation on the basin contributing to the external water source, the available

imported water volume will increase.

Two questions arise when introducing uncertainty into a problem: what level of

uncertainty needs to be considered and how reliable the system needs to be. These

questions are contradictory since more uncertainty reduces the system reliability. Thus,
 212

the level of uncertainty (or conversely, the level of conservatism) needs to be controlled

to attain a certain degree of system reliability. Here, the degree of conservatism, Γi,

introduced by Bertisimas and Sim (2004) controls the level of system reliability as

described previously in Section 2.

Following Bertisimas and Sim (2004), the constraints that contain the uncertain

precipitation, water demand and imported water availability (Eqs. C.20, C.22, C.26, and

C.27) are rewritten in the form of Eq. C.7.

Precipitation within the basin flows cause uncertainty in the groundwater systems and

in the upstream river inflows. The minimum groundwater storage constraint (Eq. C.27)

and can be reformulated as robust constraint by accounting for the uncertainty in input

over time as:

o o
IGS + ∑ (q12j − q25j − q26j ) + 0.1Ab ∑ ( P j + η~1 j Pˆ j ) ≥ RS 2 , ∀i ∈ N SS , ∀O (C.37)
j =1 j =1

Rearranging terms gives:

o o ⎛ o ⎞
− IGS − 0.1Ab ∑ P j x j − ∑ (q12j − q25j − q26j ) + RS 2 + ⎜⎜ − 0.1Ab ∑η~1 j Pˆ j ⎟⎟ ≤ 0 (C.38)
j =1 j =1 ⎝ j =1 ⎠

and introducing the first set of Γ j (= Γ1...Γ10 ) values corresponding to the ten operation

periods results in the final constraint form:

o o o
− IGS − 0.1Ab ∑ P j x j − ∑ (q12j − q25j − q26j ) + RS 2 + − 0.1Ab ∑ Γ j Pˆ j ≤ 0 (C.39)
j =1 j =1 j =1

Γ j ∈ [0, o] , j ∈ {1, 2, ..., 10}

 213

The minimum river flow constraint (Eq. C.26) can also be rewritten in robust form in

terms of the inflow due to precipitation (Eq. C.23) or:

0.3 Ab ( P o + η~1o Pˆ o ) + q78

o
− q34
o
− q36
o
≥ RQ8 i ∈ N RD ⊆ N NS , ∀O (C.40)

o
Ö q78 − q34
o
− q36
o
+ 0.3 Ab P o + η~1o 0.3 Ab Pˆ o − RQ8 ≥ 0 (C.41)

Multiplying each part by (-1), so that it is in equivalent form to Bertismas and Sim

(2004):

− q78
o
+ q34
o
+ q36
o
( )
− 0.3 Ab P o + η~1o − 0.3 Ab Pˆ o + RQ8 ≤ 0 (C.42)

Introducing Γ11 to for the robustness constraint gives:

− q78
o
+ q34
o
+ q36
o
− 0.3 Ab P o + Γ11 − 0.3 Ab Pˆ o + RQ8 ≤ 0 (C.43)

Γ11 ∈ [0, 1]

The uncertainty in user water demand including its correlation with precipitation can be

included in a general form in Eq. C.20 as:

∑q j
o
ji ≥ Dio + η~i Dˆ io − η~1o ρPˆ o Ai ∀i ∈ N U , ∀O (C.44)

Writing Eq. C.20 specifically for domestic and agricultural users gives:

o
q45 + q25
o
+ ADO _ OUT P o ≥ D5o domestic area

o
Ö q45 + q25
o
+ ADO _ OUT ( P o + η~1o Pˆ o ) ≥ D5o + η~2 Dˆ 5o − η~1o ρPˆ o ADO _ OUT (C.45)

and

q16o + q26
o
+ q36
o
+ q76
o
+ AAG P o ≥ D6o agricultural area

Ö q16o + q26
o
+ q36
o
+ q76
o
+ AAG ( P o + η~1o Pˆ o ) ≥ D6o + η~3 Dˆ 6o − η~1o ρPˆ o AAG (C.46)

Eq. C.45 is rearranged to the form of Bertismas and Sim or:

 214

− q45
o
− q25
o
( )
+ D5o − ADO _ OUT P o + η~2 Dˆ 5o + η~1o − ADO _ OUT Pˆ o (1 + ρ ) ≤ 0 (C.47)

The randomness defined by the fifth and sixth terms in Eq. C.47 should be maximized to

ensure the constraint against failure. Since only two random parameters appear in Eq.

C.47, the robust formulation can be written explicitly when Γ12 is greater than 1:

− q45
o
− q25
o
+ D5o x1 − ADO _ OUT P o x2
( ) ( )
+ max{Dˆ 5o + (Γ12 − 1) − ADO _ OUT Pˆ o (1 + ρ ) ; (Γ12 − 1) Dˆ 5o + − ADO _ OUT Pˆ o (1 + ρ ) } ≤ 0
(C.48)

Similarly, agricultural area water demand constraint can be rewritten as the form of

Bertismas and Sim when Γ13 is greater than 1 or:

− q16o − q26
o
− q36
o
− q76
o
+ D6o − AAG P o + η~2 Dˆ 6o − η~1o AAG Pˆ o (1 + ρ ) ≤ 0 (C.49)

− q16o − q26
o
− q36
o
− q76
o
+ D6o − AAG P o
(C.50)
+ max{Dˆ 6o + (Γ13 − 1) AAG Pˆ o (1 + ρ ) ; (Γ13 − 1) Dˆ 6o + AAG Pˆ o (1 + ρ ) } ≤ 0

Finally, Eq. C.22 that requires that external water volume use must be less than

imported water availability can also be rewritten in the form of Bertismas and Sim or:
~
∑ j = IW
q oji ≤ IW o i ∈ N U U N S , ∀O

Ö q12o + q14o + q16o ≤ IW o + η~4 IWˆ o + η~1o ρPˆ o Ab' (C.51)

Ö q12o + q14o + q16o − IW o − η~4 IWˆ o − η~1o Ab' Pˆ o ρ ≤ 0 (C.52)

Introducing the uncertain level parameter, Γ14 ∈ {0,2} this constraint is converted to

an explicit robust form in the same manner as the water demand constraints when Γ14 > 1

or:
 215

q12o + q14o + q16o − IW o

(C.53)
+ max{ IWˆ o + (Γ14 − 1) ρPˆ o Ab' ; (Γ14 − 1) IWˆ o + ρPˆ o Ab' } ≤ 0

3.7 Probability Bounds

In the robust formulation, Eqs. C.39, C.43, C.48, C.50, and C.54 replace their

deterministic forms (Eqs. C.20, C.22, C.26, and C.27). Fourteen additional constants, Γi ,

with uncertain parameters are included to the control conservatism of the system in these

fourteen constraints (Table C.1). To consider conditions that adversely affect the system,

the minimum values of Γi are set to 0 ( i ∈ [1, 14] ). The values of K i (i.e., number of

uncertain parameters in constraint i) are the upper bounds of the parameter Γi . With Γi

equal to 0, the max-terms equal zero, the mean parameter values are used in the

optimization model and no uncertainty is considered.

For a defined value of Γi , the robust form of the problem is deterministic. To

examine the effect of uncertainties, the problem is solved for alternative values of the

conservatism parameters and the decision maker can then judge the tradeoff between the

conservatism and total cost. As described in Section 2, the values of Γi are calculated

using Eq. C.9 for a given allowable constraint violation probability and listed in Table

C.2. For the same probability level, the Γi values vary due to the different number of

random variables appears in problem constraints. Figure C.2 shows probability bounds

for n = K i = 10.
 216

4. RESULTS AND DISCUSSION

In some semi-arid regions, wastewater effluent is discharged to a normally dry or low

flow channel. Over time, a downstream riparian habitat develops that the effluent flow

sustains. If communities move to using reclaimed wastewater effluent for nonpotable and,

potentially, potable uses, this water would no longer be released to the riparian area. Thus,

communities face depletion of both surface and subsurface water sources and the decision

to maintain environmental flows.

The hypothetical community posed here requires developing new water supply

structures and sources while preserving environmental flow in the river stream and to the

aquifer. An external water source can be imported at a cost per unit volume plus the cost

of the conveyance to transport the water to the community.

The robust optimization method was applied to minimize the total cost of

construction, expansion, and operations and maintenance of a hypothetical water supply

system. The system includes subsurface (aquifer), surface, and imported water sources,

domestic and agricultural irrigation users, and water and wastewater treatment plants.

Input parameters for the system, nodes, and arcs are summarized in Tables C.3, C.4, and

C.5, respectively. As described above, a 15 year planning period that was divided into

two design periods and 10 operation periods was considered. Available water supply

system is listed in Table C.6. Figure C.3 shows flows and water supply structures at year

0. Groundwater is the main source to supply water demand of domestic and agricultural
 217

areas at year 0. As the result, groundwater is depleted and water sustainability becomes

an issue in the hypothetical community. Groundwater storage at year 0 is 9.50 km3 that is

below the defined minimum storage of 9.93 km3. The optimized water supply plan must

increase groundwater storage to 9.93 km3 in the next fifteen years. To ensure riparian

health in this application, a minimum downstream river flow requirement is defined as

11.4 m3/s.

The water system’s arcs consist of five canals, four pipelines, and two pump stations.

The associated design decision at each design epoch are the canal depths, d, (5 canals),

pipe diameters, κ, (4 pipelines), pump design discharges, χ, (4 pump/pipelines + 2

pumps), pump design heads, H, (4 pump/pipelines + 2 pumps), and water and wastewater

treatment plant capacities, w, (2 plants). Thus, the total number of design decision

variables is 46 (23 × 2 design periods). In addition, the flows on 11 independent arcs

must be determined for each of the 10 operation periods. Lastly, the number of binary

variables for pipe flowrate, x, (4 pipelines × 10 operation periods) and pump flowrate, µ,

{(4 pump/pipelines + 2 pumps) × 10 operation periods} is 100. Thus, the optimization

problem contains a total number of 256 decision variables including 100 binary variables.

The continuous mixed-integer nonlinear problem were solved using GAMS/BARON

global optimization solver with the relative termination tolerance of 0.05 (Sahinidis and

Tawarmalani, 2005).

To demonstrate the overall formulation, this system is optimized for violation

probabilities ranging from 0.1 (the most conservative) to 1.0 (nominal). Figures C.4 and

C.5 show flows and design variables at year 1 in nominal (i.e., all Γ = 0) and robust
 218

problems (P = 0.1). When constraint violation probability equals 0.1, the solution ensures

that all of the constraints remain feasible at least 90% of the time. The set of fourteen Γi

in this case (P = 0.1) is {1.00, 2.00, 3.00, 3.72, 4.20, 4.34, 4.61, 4.91, 4.93, 5.29, 1.00,

2.00, 2.00, 2.00} (Table C.2).

Groundwater storage requirement constraints (Eq. C. 39) have more uncertain

parameters ( K i ) in later operation periods. Uncertainty in yearly precipitation is

generated independently and total uncertainty increases over time. Domestic demands

increase with larger populations (Table C.12).

Optimal arc flows for the nominal and robust problems are listed in Tables C.7 and

C.8, respectively. Most noticeably, to meet the water demands imported water at year 1 in

the nominal problem (7.17 m3/s) is increased to 10.02 m3/s when P = 0.1. Due to the

expense of imported water ($0.81/m3), both alternatives replenish the aquifer with that

source in year 1 and no additional imported water is purchased. Both cases use reclaimed

water from wastewater treatment plant as the primary agricultural purpose. The robust

solution requires other sources while the nominal solution only uses a small amount

beyond reclaimed water.

Tables C.9 and C.10 lists the optimal design decisions for the two problems.

Domestic area demand increases over time as the population grows while agricultural

demand decreases after the 5th operation year. Therefore, few components are expanded

in the second design epoch (year 5). Capacities and heads of pumps that provide water

to and from domestic areas are expanded in nominal condition to meet the higher demand.

Increased domestic demand is supplied from the upstream river through the water
 219

treatment plant that reduces water supply to the agricultural area from the upstream river.

Reclaimed water from wastewater treatment plant replaces this flow after year 5 and

requires the pump capacity to be expanded (Table C.9).

When the constraint violation probability is 0.1, only the pump capacity from

wastewater treatment plant to agricultural area is expanded; from 5.78 m3/s to 7.46 m3/s

because of increasing uncertainty in precipitation (Table C.10). Table C.11 lists

capacities of water and wastewater treatment plants in nominal problem and when

constraint violation probability is 0.1. A smaller water treatment plant is constructed in

the robust solution because supplying water to the domestic area from the aquifer is more

economical than expanding the water treatment plant. In this case, the aquifer is

recharged with imported water. The wastewater treatment plant capacity, however,

increases when constraint violation probability is 0.1 because effluent from domestic area

increases as demand rises.

As seen in Figure C.6, the optimal total cost depends upon the degree of

conservatism. As conservatism is raised (and constraint violation probability is

decreased), the total cost increases to insure system reliability by enlarging components

sizes or purchasing more imported water. In particular, the cost increases dramatically

between constraint violation probabilities of 0.7 and 0.5. At lower allowable constraint

violation probabilities, the cost is relatively flat. Up to about P = 0.5, the existing

expanded water treatment plant supplies water to domestic users. The flatness of the

curve demonstrates the economies of scale in meeting needs through the treatment

facility.
 220

As the robustness requirement is increased, the strategy to meet demands changes.

The water treatment plant is not expanded rather its capacity remains at the initial size

(Table C.11). Given the demand, developing or expanding the conveyance system (canal

and pumping system) to import water and deliver it to domestic users becomes viable. As

seen in Figure C.7, the amount of imported water parallels the increase in cost. Above

the level of P=0.7, economies of scale again dominate and the incremental cost and

supply result in a relatively flat response.

5. CONCLUSIONS

In this study, a robust optimization approach was applied in a hypothetical water

supply system to minimize the total system cost. Considering that data in real systems

inherently involve uncertainties, it is important to consider these uncertainties during the

design process to improve the system reliability and robustness. This study found that the

robust optimization approach of Bertsimas and Sim could be a useful tool in the design of

water supply systems without introducing additional complexity into the optimization

problem.

Uncertainty was considered in water demand, water availability, and correlations

from precipitation to water demand and water availability, 14 of Γi parameters were

added to the model, the probability of violating a constraint was related to those

parameters, and the problem was solved for a range of Γi . The result was an NLP similar

in structure to the original NLP.

 221

The hypothetical water supply system included subsurface, surface, imported water

sources, domestic and agricultural irrigational users, and water and wastewater treatment

plants as the system components. Initially, the system has available infrastructure that

was built to primarily provide water from surface and subsurface water sources.

Overall 15 years of the planning period was applied for the system evaluation. Two

design periods and 10 operation periods were evaluated to determine the values of total

256 decision variables with 100 binary decision variables. The problem was solved using

the GAMS/BARON software by a mixed branch and bound – (which NLP method)

method.

The effects of the degree of conservatism and the available water supply on total

system cost are investigated using probability bounds. Probability bounds are tied the

constraint violation as at most the probability of violation. From nominal problem to

conservative case (probability of violation is 0.1), optimal total costs are calculated and

the solutions are compared.

Water demand is driven by population growth rate, only a few expansion of structure

is suggested as an economic solution. Because of high cost to purchasing imported water,

external water source is introduced once at year 1 and recharges aquifer. Compared to the

nominal solution, conservative solutions import more water to maintain system reliability

and preserve environmental flows. As the result, total construction and operation cost

increase in the reliable water supply design. Given the small application system, water

treatment plant exits at year 0, expansion of the treatment plant is cheaper than building

other transportation facility to supply domestic area. After probability of constraint

 222

violation is 0.5, economic solution is to construct larger canal and pumping station to

supply domestic demand. In this transaction period, total cost increases rapidly due to the

construction and importing water, then become relatively slow after then. The higher cost,

thus, is largely attributed to external water purchases.

Future work should move toward better representing the decisions including

discretized variables for pipe diameters and adding additional uncertain parameters, such

as temporal correlation over the operational time span. Also, adding more system

components such as commercial and industrial areas, public outdoor area and reservoirs

and distributed water and wastewater treatment plants can be examined to investigate

tradeoff between construction and piping cost. Additional water quality parameters would

improve the realism of the system representation. Finally, the system should be tested in

real supply network.

 223

6. NOMENCLATURE

Indices and Sets

N a set of nodes in a network (sources, users, and treatment plants)

A a set of arcs (i, j) from a node i to a node j in a network, ∀i, j ∈ N

T a set of design periods t, ∀t ∈ T

O a set of operation periods t, ∀o ∈ O

Subsets,

A C a set of canal connections, AC ⊆ A

A P a set of pipe connections, AP ⊆ A

A U a set of pump connections, AU ⊆ A

A R a set of flows from rainfall, AR ⊂ A

A I a set of seepage from users to an aquifer and riverbed infiltration, A I ⊆ A

N U a set of users, NU ⊆ N

N S a set of sources, NS ⊆ N

N SS a set of storage sources, N SS ⊆ N S ⊆ N

N NS a set of non-storage sources, N NS ⊆ N S ⊆ N

N IW imported water, N IW ⊆ N NS ⊆ N S ⊆ N

N RU river upstream node, N RU ⊆ N NS ⊆ N S ⊆ N

N RD river downstream node, N RD ⊆ N NS ⊆ N S ⊆ N

 224

N WT a set of water treatment plants, N WT ⊆ N

N WWT a set of wastewater treatment plants, N WWT ⊆ N

Data

f Darcy Weisbach coefficient

nij Manning coefficient of pipes and canals from i to j ∀A P U A C

zij channel side slope from i to j ∀A C

COND hydraulic conductivity

I interest rate

CITY city multiplier

o
POPi population at an operation year o at i i ∈ NU , o ∈ O

0
At year 0, POPi initial population at a node i

POPi o = POP 0 (1 + POPGR) o

POPGR population growth rate

∆o o length of an operation period o

Ab basin area

Ab' basin area contributing to imported water

Lij length for an arc (i, j) ∀A

RQio required discharge at a node i in an operation year o ∀i ∈ N RD , o ∈ O

WSio water storage at a node i in an operation year o i ∈ N SS , o ∈ O

 225

RSio required storage at a node i in an operation year o ∀i ∈ N SS , o ∈ O

ELi elevation at a node i ∀N

Hmin,j minimum pressure requirement at the end of pipe and pump connections

CIW unit cost of purchasing imported water

∆ ij elevation differences at an arc (i, j) ( = ELj – ELi)

⎛∆ ⎞
Sij channel bottom slope for an arc (i, j) = ⎜ ij ⎟
⎜L ⎟
⎝ ij ⎠

Stochastic data
~
Po precipitation at an operation year o o∈O
~
IW o imported water available at period o o∈O
~
Dio demand at a node i in an operation year o ∀N U , o ∈ O

Decision variables

qijo operation flow rate for an arc (i, j) at an operation year o [L3/T] ∀A , o ∈ O

κ ijt pipe diameter [L] for an arc (i, j) at a design period t ∀A P , t ∈ T

d ijt canal depth [L] for an arc (i, j) at a design period t ∀A C , t ∈ T

χ ijt capacity of pump for an arc (i, j) [L3/T] at a design period t ∀A P U A U , t ∈ T

H ijt pump design head for an arc (i, j) [L] at a design period t ∀A P U A U , t ∈ T

wit capacity at a node i at a design period t [L3] ∀A WT U A WWT , t ∈ T

xijt takes value 1 if an arcs (i, j) is built at a design period t

 226

and 0 otherwise ∀A P , t ∈ T

µijt takes value 1 if a pump in an arcs (i, j) is built at a design period t

and 0 otherwise ∀A U , t ∈ T

7. REFERENCES

Ben-Tal ,A. and Nemirovski, A. (1999). “Robust solutions of uncertain linear programs.”

OR Letters, 25, 1-13.

Ben-Tal, A. and Nemirovski, A. (2000). “Robust solutions of Linear Programming

problems contaminated with uncertain data.” Mathematical Programming, 88, 411-

424.

Bertsimas, D. and Sim, M. (2004). “The price of robustness.” Operations Research,

52(1), 35 – 53.

Clark, R. M, Sivaganesan, M., Selvakumar, A., and Sethi, V. (2002). “Cost models for

water supply distribution systems.” Journal of Water Resources Planning and

Management, 128(5), 312-321.

El-Ghaoui, L. and Lebret, H. (1997). “Robust solutions to least-square problems to

uncertain data matrices.” SIAM Journal on Matrix Analysis and Applications, 18,

1035-1064.

El-Ghaoui, L., Oustry, F. and Lebret, H. (1998). “Robust solutions to uncertain

semidefinite programs.” SIAM Journal on Optimization, 9, 33-52.

 227

Elshorbagy, W., Yakowitz, D., and Lansey, K. (1997). “Design of engineering systems

using a stochastic decomposition approach.” Engineering Optimization, 27(4), 279-

302.

Fiering, M. B. and Matalas, N. C. (1990). Decision making under uncertainty, climate

change and U.S. water resources, P. E. Waggoner, ed., John Wiley & Sons, Inc.,

New York, N.Y.

Jenkins, M. W. and Lund, J. R. (2000). “Integrating yield and shortage management

under multiple uncertainties.” Journal of Water Resources Planning and

Management, 126(5), 288-297.

Lund, G. R. and Israel, M. (1995). “Optimization of transfer in urban water supply

planning.” Journal of Water Resources Planning And Management, 121(1), 41-48.

Mulvey, J. M., Vanderbei, R. J., and Zenios, S. A. (1995). “Robust optimization of large-

scale systems.” Operations Research, 43(2), 264-281.

Sahinidis, N. and Tawarmalani, M. (2005). GAMS/BARON Solver Manual

Soyster, A. L. (1973). “Convex programming with set-inclusive constraints and

applications to inexact linear programming.” Operations Research, 21, 1154-1157.

Tang, C. C., Brill, E. D., and Pfeffer, J. T. (1987). “Optimization techniques for

secondary wastewater treatment system.” Journal of Environmental Engineering,

113(5), 935-951.

US. Amy Corps of Engineers, (1980). Methodology for areawide planning studies.

Engineer Technical. Letter No. 1110-2-502, Washington, D.C.

 228

Watkins, D. W. and McKinney, D. C. (1997). “Finding robust solutions to water

resources problems.” Journal of Water Resources Planning and Management, 123(1),

49 – 58.

Wilchfort, G. and Lund, J. R. (1997). “Shortage management modeling for urban water

supply systems.” Journal of Water Resources Planning and Management, 123(4),

250-258.

Walski, T. M., Brill, E. D., Gessler, J., Goulter, I. C., Jeppson, R. M., Lansey, K., Lee,

H., Liebman, J. C., Mays, L., Morgan, D. R., and Ormsbee, L. (1987). “Battle of the

network models: epilogue.” Journal of Water Resources Planning and Management,

113(2), 191-203.
 229

8. TABLES

Table C.1. Gammas for robust formulation

Variables Description Range
Γ1 Flow from precipitation to groundwater in operation period 1 [0-1]
Γ2 Flow from precipitation to groundwater in operation period 2 [0-2]
Γ3 Flow from precipitation to groundwater in operation period 3 [0-3]
Γ4 Flow from precipitation to groundwater in operation period 4 [0-4]
Γ5 Flow from precipitation to groundwater in operation period 5 [0-5]
Γ6 Flow from precipitation to groundwater in operation period 6 [0-6]
Γ7 Flow from precipitation to groundwater in operation period 7 [0-7]
Γ8 Flow from precipitation to groundwater in operation period 8 [0-8]
Γ9 Flow from precipitation to groundwater in operation period 9 [0-9]
Γ10 Flow from precipitation to groundwater in operation period 10 [0-10]
Γ11 Flow from precipitation to river [0-1]
Γ12 Domestic area demand satisfaction [0-2]
Γ13 Agricultural area demand satisfaction [0-2]
Γ14 Imported water availability [0-2]
 230

Table C.2. Choice of Γi as a function of the maximum probability of constraint

violation

Probability of constraint violation

Γi
0.01 0.02 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Γ1 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.60 0.20 0.00 0.00 0.00
Γ2 2.00 2.00 2.00 2.00 1.82 1.47 1.11 0.76 0.40 0.05 0.00 0.00
Γ3 3.00 3.00 3.00 2.64 2.15 1.67 1.18 0.70 0.21 0.00 0.00 0.00
Γ4 4.00 4.00 3.72 2.99 2.26 1.68 1.18 0.67 0.17 0.00 0.00 0.00
Γ5 5.00 5.00 4.20 3.01 2.40 1.79 1.19 0.58 0.00 0.00 0.00 0.00
Γ6 6.00 5.91 4.34 3.33 2.51 1.77 1.16 0.54 0.00 0.00 0.00 0.00
Γ7 6.92 6.58 4.61 3.45 2.57 1.86 1.16 0.45 0.00 0.00 0.00 0.00
Γ8 7.64 7.05 4.91 3.57 2.69 1.84 1.12 0.43 0.00 0.00 0.00 0.00
Γ9 8.10 7.10 4.93 3.75 2.71 1.91 1.12 0.32 0.00 0.00 0.00 0.00
Γ10 8.20 7.63 5.29 3.79 2.85 1.91 1.12 0.32 0.00 0.00 0.00 0.00
Γ11 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.60 0.20 0.00 0.00 0.00
Γ12 2.00 2.00 2.00 2.00 1.82 1.47 1.11 0.76 0.40 0.05 0.00 0.00
Γ13 2.00 2.00 2.00 2.00 1.82 1.47 1.11 0.76 0.40 0.05 0.00 0.00
Γ14 2.00 2.00 2.00 2.00 1.82 1.47 1.11 0.76 0.40 0.05 0.00 0.00
 231

Table C.3. Input parameters used for the hypothetical water supply system
Parameter Value Unit
Darcy-Weisbach coefficient, f 0.02
Manning's coefficient, n 0.014
Canal side slope, z 2
Hydraulic conductivity, COND 9.14 m/yr
Imported water availability, IW 19.6 m3/s
Initial population, POP0 1,200,000
Population growth rate, POPGR 2.7 %/yr
Interest rate, I 2.0 %/yr
City multiplier, CITY 1
Annual precipitation, P 533.4 mm/yr
Basin area, Ab 12,645 km2
Basin area contributing to imported water, Ab' 13,909 km2
Required groundwater storage, RS2 9.93 km3
Required downstream river flow, RQ3 11.4 m3/s
Unit cost of purchasing imported water, CIW 0.81 $/m3
o
Agricultural consumptive use (1 – 5 periods), DAG 12.5 m3/s
o
Agricultural consumptive use (6 – 10 periods), DAG 11.3 m3/s

Table C.4. Node characteristics for hypothetical water supply system

Nodes Area (km2) Loss (m3/s)
Agricultural area 1,214 0
Domestic area 2,974 0.0002
Imported water 0 0
Water treatment plant 0 0.0002
Wastewater treatment plant 0 0.0002
 232

Table C.5. Arc lengths, type and elevation differences of arcs in the hypothetical
water supply system
Links Origin Destination Length (m) Elevation difference (m) Connection type
1 RIU* WT 4,506 -30 Canal
2 RIU AG 45,062 -91 Canal
3 IW WT 16,093 -61 Canal
4 IW GW 77,249 -152 Canal
5 IW AG 61,155 -122 Canal
6 GW DO 0 244 Pump
7 GW AG 0 152 Pump
8 WT DO 4,506 -30 Pipe/Pump
9 DO WW 16,093 -15 Pipe/Pump
10 WW AG 16,093 -3 Pipe/Pump
11 WW RID 45,062 -46 Pipe/Pump
* IW - Imported water, GW - Groundwater, RIU – Upstream river, RID – Downstream
river, WT - Water treatment plant, DO - Domestic area, AG - Agricultural area, WW -
Wastewater treatment plant

Table C.6. Infrastructure in the system at year 0

Canal depth Flowrate Pipe diameter Pump head
Links Pump flow (m3/s)
(m) (m3/s) (mm) (m)
1 1.9 3.9 0 0.0 0.0
2 0.9 0.8 0 0.0 0.0
3 0.0 0.0 0 0.0 0.0
4 0.0 0.0 0 0.0 0.0
5 0.0 0.0 0 0.0 0.0
6 0.0 2.8 0 4.3 125.0
7 0.0 5.5 0 6.6 103.6
8 0.0 3.9 1829 4.7 30.5
9 0.0 3.9 1524 4.7 45.7
10 0.0 0.0 0 0.0 0.0
11 0.0 3.9 1829 4.7 30.5
 233

Table C.7. Flow allocations along operational periods for nominal problem
Operational
1 2 3 4 5 6 7 8 9 10
Period
FRIuTWT* 6.93 7.12 7.31 7.51 7.71 8.12 6.97 8.12 7.11 8.12
FRIuTAG 0.84 0.84 0.81 0.61 0.41 0.00 0.84 0.00 0.84 0.00
FIWTWT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
FIWTGW 7.17 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
FIWTAG 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
FGWTDO 0.00 0.00 0.00 0.00 0.00 0.01 1.61 0.93 2.43 1.95
FGWTAG 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
FWTTDO 6.93 7.12 7.31 7.51 7.71 8.12 6.97 8.12 7.11 8.12
FDOTWW 6.93 7.12 7.31 7.51 7.71 8.13 8.58 9.04 9.54 10.06
FWWTAG 6.93 6.10 7.31 6.42 7.71 8.13 8.58 9.04 9.54 9.54
FWWTRID 0.00 1.01 0.00 1.09 0.00 0.00 0.00 0.00 0.00 0.52

* FRIUTWT - flow allocation from upstream river to water treatment plant, FRIUTAG –
flow allocation from upstream river to agricultural area, FIWTWT - flow allocation from
imported water to water treatment plant, FIWTGW - flow allocation from imported
water to groundwater, FIWTAG - flow allocation from imported water to agricultural
area, FGWTDO - flow allocation from groundwater to domestic area, FGWTAG - flow
allocation from groundwater to agricultural area, FWTTDO - flow allocation from water
treatment plant to domestic area, FDOTWW - flow allocation from domestic area to
wastewater treatment plant, FWWTAG - flow allocation from wastewater treatment
plant to agricultural area, FWWTRID - flow allocation from wastewater treatment plant
to downstream river

Table C.8. Flow allocations along operational periods when probability of violation
is 0.1
Operational
1 2 3 4 5 6 7 8 9 10
Period
FRIuTWT 6.05 6.16 6.16 6.16 6.16 6.16 6.16 6.16 6.16 6.16
FRIuTAG 0.11 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
FIWTWT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
FIWTGW 10.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
FIWTAG 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
FGWTDO 2.19 2.27 2.48 2.71 2.94 3.47 3.86 4.39 4.95 5.54
FGWTAG 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
FWTTDO 6.05 6.16 6.16 6.16 6.16 6.16 6.16 6.16 6.16 6.16
FDOTWW 7.83 8.02 8.23 8.45 8.67 9.19 9.58 10.09 10.63 11.20
FWWTAG 7.83 8.02 8.23 8.45 8.67 9.19 9.58 10.09 10.63 11.20
FWWTRID 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
 234

Table C.9. Design decisions of nominal problem

Canal depth Pipe diameter Pump design Cap Pump design head
(m) (mm) (m3/s) (m)
1 period 2 period 1 period 2 period 1 period 2 period 1 period 2 period
FRIUTWT 2.4 2.4 FGWTDO 4.26 4.26 152.4 152.4
FRIUTAG 0.9 0.9 FGWTAG 6.57 6.57 103.6 103.6
FIWTWT 0.0 0.0 FWTTDO 152 152 5.14 5.41 30.5 30.5
FIWTGW 2.0 2.0 FDOTWW 148 148 5.14 11.11 45.7 45.7
FIWTAG 0.0 0.0 FWWTAG 146 146 5.14 6.36 8.3 8.3
FWWTRID 152 152 4.69 4.69 30.5 30.5

Table C.10. Design decisions when probability of violation is 0.1

Canal depth Pipe diameter Pump design Cap Pump design head
(m) (mm) (m3/s) (m)
1 period 2 period 1 period 2 period 1 period 2 period 1 period 2 period
FRIUTWT 2.2 2.2 FGWTDO 4.26 4.26 152.4 152.4
FRIUTAG 0.9 0.9 FGWTAG 6.57 6.57 103.6 103.6
FIWTWT 0.0 0.0 FWTTDO 152 152 4.69 4.69 145.4 145.4
FIWTGW 2.3 2.3 FDOTWW 182 182 7.46 7.46 45.7 45.7
FIWTAG 0.0 0.0 FWWTAG 138 138 5.78 7.46 8.3 8.3
FWWTRID 152 152 4.69 4.69 30.5 30.5
 235

Table C.11. Treatment facility capacities for different probability violation (P)
Water treatment plant Wastewater treatment plant
(m3/s) (m3/s)
1 period 2 period 1 period 2 period
Nominal condition 7.71 8.12 7.71 10.06
P = 0.1 6.16 6.16 8.67 11.20

Table C.12. User demand in operational periods for the nominal problem
Domestic area demand Agricultural area consumptive use
Period dt*
(m3/s) (m3/s)
1 1 9.78 12.52
2 1 10.04 12.52
3 1 10.31 12.52
4 1 10.59 12.52
5 1 10.88 12.52
6 2 11.17 12.52
7 2 11.78 11.26
8 2 12.43 11.26
9 2 13.11 11.26
10 2 13.83 11.26
*dt – time interval of an operational period
 236

9. FIGURES

Figure C.1. Water supply system network schematic. Bold arcs represent 14
conveyance structures to be sized and thin arcs represent infiltration from users and
sources to the aquifer.
 237

Figure C.2. Probability bounds for n= K i =10

 238

1-Canal
q = 3.9 cms
d =1.9 m 4 - Water treatment
plant

3 - Upstream 8-Pipe/Pump
River q = 3.9 cms
K = 1829 mm
2-Canal 2 - Groundwater X = 4.7 cms
q = 0.8 cms 6-Pump
H = 30.5 m
d = 0.9 m q = 2.8 cms
X = 4.3 cms
7-Pump H = 125 m
q = 5.5 cms 5 - Domestic Area
X = 6.6 cms
H = 103.6 m 9-Pipe/Pump
q = 3.9 cms
K = 1524 mm
6 - Agricultural Area X = 4.7 cms
11-Pipe/Pump H = 45.7 m
q = 3.9 cms
K = 1829 mm
7 - Wastewater
X = 4.7 cms treatment plant
8 - Downstream H = 30.5 m
River

Figure C.3. Infrastructure in water supply system network before optimization.

schematic. (q – flowrate; d = canal depth; K = pipe diameter; X = pump capacity; H
= pump head)
 239

Figure C.4. Optimal water supply system operation of nominal problem at year 1
(precipitation – arcs 12, 13, 14, and 15; infiltration – arcs 16, 17, and 18; q –
flowrate; d = canal depth; K = pipe diameter; X = pump capacity; H = pump head)
 240

Figure C.5. Optimal water supply system operation at year 1 when probability
violation is 0.1 (precipitation – arcs 12, 13, 14, and 15; infiltration – arcs 16, 17, and
18; q – flowrate; d = canal depth; K = pipe diameter; X = pump capacity; H = pump
head)
 241

Figure C.6. Optimal total cost of the water supply system as a function of the
probability bound of constraint violation
 242

Figure C.7. Total amount of imported water purchased as a function of the

probability bound of constraint violation