UNESCO-IHE

INSTITUTE FOR WATER EDUCATION

Use of hydroinformatics technologies for real time water

quality management and operation of distribution
networks. Case study of Villavicencio, Colombia

José Leonardo Alfonso Segura

MSc Thesis WSE-HI-06-04
March 2006
 Use of hydroinformatics technologies for real time water
quality management and operation of distribution networks.
Case study of Villavicencio, Colombia

Master of Science Thesis

by
J. Leonardo Alfonso S.

Mentor
Andreja Jonoski, PhD (UNESCO-IHE)

Supervisor
Prof. Roland Price, PhD (UNESCO-IHE)

Examination committee
Prof. Roland Price, PhD - Chairman
Anthony Minns, PhD (WL Delft Hydraulics) - External member
Andreja Jonoski, PhD - Member

This research is done for the partial fulfilment of requirements for the Master of Science degree at the
UNESCO-IHE Institute for Water Education, Delft, the Netherlands

Delft
March 2006
The findings, interpretations and conclusions expressed in this study do neither
necessarily reflect the views of the UNESCO-IHE Institute for Water Education, nor of
the individual members of the MSc committee, nor of their respective employers.
 To my Angels,
Sandra, my force, and Valentina, my reason.
Abstract
Most of the water supplies systems in developing countries do not guarantee
the access to safe dinking-water, due to either the intermittency nature of the
systems or the poor operational practices in continuous supplies. These
situations affect directly the efforts that the world is paying with regards to
the achievement of water supply–related UN Millennium Development
Goals.

Within this context, the present research demonstrates the development of a

variety of Hydroinformatics tools developed for real-time management of
different and frequent water quality problems occurring in developing
countries. The developed applications include: a new approach for
switching the intermittent condition towards the pressurised one; posing and
solving the operational responses to contamination events as an optimisation
problem; a demonstration of the utilisation of the Short Messages Services
for real time management of contamination. All the tools have been tested
and applied to a case study of the water supply system of Villavicencio,
Colombia.

As one of the main results, this study demonstrates how the developed tools
combined into one software application (QmagNet-V) can be helpful in the
management of a contamination event, preparing SMS to advise the
operators and to warn the affected population. The direct conclusion of the
research is that the safe-drinking water could be improved in supply
networks by means of the available technology and the hydroinformatics
knowledge. Additionally, the developed application proved to have a real
possibility to be an important complement of the Water Quality Surveillance
Information System, tool developed by the Colombian National Health
Institute.

Keywords: Water Quality, Distribution Networks Management, Short Message Service,

Intermittent Supply, Operational Optimisation

i
Acknowledgements
First of all I would like to give my sincere thanks to UNESCO-IHE and the Watermill
project for providing me the opportunity to follow such a complete, useful and
pioneering Hydroinformatics course.

I want to give my special thankfulness to my supervisor, Dr. Andreja Jonoski, who

provided me his always-clear guidance, as well as his accurate comments and
suggestions during the last months.

My special thanks to Dr. Dimitri Solomatine, who provided me not only his software
knowledge assistance, but also his innovative ideas for my research and to Dr.
Kalanithy Vairavamoorthy, who on several occasions offered me assistance in this
research with his invaluable experience.

I am also very grateful with Dr. Roland K. Price and the entire Hydroinformatics and
Knowledge Management Department staff, not only because of their excellent
educational labour, but also for their human support.

Very special thanks to Germán González, Fernando Ballesteros and Teo Abril, from
Aquadatos Ltda in Bogotá, Colombia, who provided me important amount of
information to finalise this research appropriately.

Extensive thanks also for Dr. Jaime Ortiz from the National Health Institute for
providing me the water quality information collected for SIVICAP, as well as Dr.
Arrieta from the Public Services Superintendence, Colombia.

Finally, I want to offer my sincere thanks to Mrs. Claudia Brakel, my diligent and
attentive fellowship officer, for giving me the faith to make real what twenty months
ago was just a dream.

ii
 Table of Contents

Abstract.......................................................................................................................... i
Acknowledgements ...................................................................................................... ii

1 Introduction .......................................................................................... 8
2 Background of water quality problems in distribution networks . 10
2.1 WHO guidelines for drinking water and the Water Safety Plans................... 10
2.2 Water quality problems in water distribution systems (WDS)....................... 13
2.2.1 Description of quality problems in WDS ............................................... 13
2.3 Water quality problems in intermittent distribution systems (IDS) ............... 15
2.3.1 Description of water quality problems in IDS........................................ 16

3 Objectives of the present research.................................................... 19

4 The supply system of Villavicencio, Colombia ................................ 20
4.1 General description......................................................................................... 20
4.2 Brief Description of the Villavicencio drinking water system ....................... 21
4.2.1 Superficial Sources ................................................................................. 21
4.2.2 Deep Well sources .................................................................................. 21
4.2.3 The network............................................................................................ 22
4.2.4 The existing hydraulic model ................................................................. 23
4.2.5 The water quality .................................................................................... 23
4.3 The intermittency problem in Villavicencio................................................... 24

5 From intermittency to continuous supply: a new approach .......... 28

5.1 Introduction .................................................................................................... 28
5.1.1 Objectives for modelling the intermittency............................................ 28
5.2 Modelling Approaches ................................................................................... 28
5.2.1 Proposed approach.................................................................................. 29
5.2.2 Pilot models tests.................................................................................... 31
5.2.3 Assumptions and limitations using SWMM........................................... 40
5.3 The Villavicencio’s Intermittent Model ......................................................... 41
5.3.1 Description ............................................................................................. 41
5.3.2 Results .................................................................................................... 43

6 Managing water quality in WDS ...................................................... 50

6.1 Introduction .................................................................................................... 50
6.2 The origin: where is the contamination coming from?................................... 50
6.2.1 Database generation................................................................................ 50
6.2.2 Methodology........................................................................................... 51
6.2.3 Assumptions ........................................................................................... 53
6.2.4 Database searching ................................................................................. 53
6.3 The threat: Who is going to be affected?........................................................ 54
6.3.1 Assumptions ........................................................................................... 55
6.4 The actions: what to do?................................................................................. 56
6.4.1 The operation response under emergency: a difficult issue ................... 56
6.4.2 Formulation of the operational response as an optimisation problem.... 56
6.4.3 Tools and methodology followed:.......................................................... 57

iii
 6.4.4 The Genetic Algorithms ......................................................................... 57
6.4.5 COPA module (Changing Operation on Pollutant Affection) ............... 58
6.4.6 The optimisation process ........................................................................ 62
6.4.7 Case study of Villavicencio.................................................................... 67

7 The SMS as a tool for operative and emergency support .............. 71

7.1 Introduction .................................................................................................... 71
7.1.1 Communication process ......................................................................... 71
7.1.2 GSM and SMS........................................................................................ 72
7.1.3 What is SMS? ......................................................................................... 73
7.1.4 SMS for real time warnings and emergency management ..................... 74
7.2 The SMS for managing the water quality in distribution systems ................. 75
7.2.1 The mobile phone for the intermittent supply operation ........................ 75
7.2.2 The mobile phone for the water quality management in WDS .............. 76
7.3 The Villavicencio case.................................................................................... 76
7.3.1 The mobile phones in Villavicencio and Colombia ............................... 76
7.3.2 How the approach should work .............................................................. 78

8 The QMagNet-V application ............................................................. 80

8.1 Description of QMagNet-V ............................................................................ 81
8.2 Required software installations ...................................................................... 82
8.3 User Interface ................................................................................................. 82

9 Discussion ............................................................................................ 92
9.1 General discussion.......................................................................................... 92
9.2 The proposed approach for intermittent supply modelling ............................ 92
9.3 Managing contamination in WDS .................................................................. 94
9.3.1 Finding the contamination sources within the network.......................... 94
9.3.2 Operational responses as an optimization problem ................................ 95
9.4 The SMS for operative support and warning.................................................. 97

10 Conclusions ......................................................................................... 99
10.1 Intermittent supply.......................................................................................... 99
10.2 Pressurized supply .......................................................................................... 99
10.3 The SMS for ensuring safe water ................................................................. 100

11 Recommendations ............................................................................ 101

11.1 The intermittent modelling approach ........................................................... 101
11.2 Managing Water Quality problems in WDS ................................................ 102
11.2.1 Looking at the pollutant sources........................................................... 102
11.2.2 Treating the operational responses as an optimization problem........... 102
11.3 The SMS for operation support and warning ............................................... 103
11.4 The QmagNet-V Application ....................................................................... 103

12 References ......................................................................................... 104

iv
 List of Tables
Table 2.1 Summary of water quality problems associated with water age .................... 14
Table 4.1. Network characterization. Lengths in km per diameter and material
(roughness) ............................................................................................................. 22
Table 5.1 Case A: free outfall......................................................................................... 32
Table 5.2 Case B: Outfall at 2m-height.......................................................................... 33
Table 6.1. Example of records generated for the database............................................. 51
Table 7.1. Communication strategies ............................................................................. 72
Table 8.1 Description of the Pascal units used by QMagNet-V..................................... 81

List of Figures
Figure 2.1 Interrelationship of the chapters of the Guidelines for Drinking Water in
ensuring drinking-water safety ............................................................................... 11
Figure 2.2 Overview of the key steps in developing a water safety plan (WSP) ........... 12
Figure 2.3 Infiltration in IDSs before and after the system resumes.............................. 16
Figure 2.4 Summary of microbial occurrence in Water and Soil Samples .................... 17
Figure 2.5. Examples of risks by backflow contamination ............................................ 18
Figure 4.1. Location of Villavicencio city in Colombia................................................. 20
Figure 4.2 Topography of the city, elevations in meters ................................................ 22
Figure 4.3. Network sectorization recommended by Aquadatos (2000)........................ 23
Figure 4.4. Chloride concentrations for the treated water .............................................. 24
Figure 4.5. Average consumption per capita in Villavicencio ....................................... 25
Figure 4.6. Daily production curve in the treatment plant during April 2000................ 26
Figure 4.7. Comparison of non-revenue water by volume in different Colombian cities
................................................................................................................................ 27
Figure 4.8. Comparison of water losses in different Colombian cities .......................... 27
Figure 5.1 Stability results for case A ............................................................................ 33
Figure 5.2 Stability results for case B............................................................................. 35
Figure 5.3. Relation between time steps and pipe slopes ............................................... 36
Figure 5.4. Flows in the pipes for the SWMM model, in l/s (after the network is
completely filled).................................................................................................... 37
Figure 5.5 Flow in the pipes for EPANET model, in l/s ................................................ 37
Figure 5.6 SWMM node depths (after the network is completely filled)....................... 38
Figure 5.7 EPANET node pressures............................................................................... 38
Figure 5.8. Water reception times at the elements showed in Figure 5.6....................... 39
Figure 5.9. Flow in the link connected to the big storage .............................................. 39
Figure 5.10. Flow and depth comparison ....................................................................... 40
Figure 5.11. Villavicencio’s SWMM model scheme. .................................................... 41
Figure 5.12. Model nodes configuration, where Q: Flow through the outlet; Qd;
household demand flow.......................................................................................... 42
Figure 5.13. Typical demand curve for a pressurized system in terms of daily average
consumption factors................................................................................................ 42
Figure 5.14. SWMM dialog box to configure a pressurized demand pattern ................ 43
Figure 5.15. Link capacities (relation current depth / full depth) at 1 hour ................... 44
Figure 5.16. Filling up process of the Villavicencio network (closing inlet at 1m tank
level) ....................................................................................................................... 45

v
Figure 5.17. Water reception times (closing inlet at 1m tank level) .............................. 46
Figure 5.18. Times of tanks charging, tanks at 1m. ....................................................... 46
Figure 5.19. Link capacities (relation current depth / full depth) at 1 hour ................... 47
Figure 5.20. Water reception times (closing inlet at 0.5m tank level) ........................... 48
Figure 5.21. Times of tanks charging, tanks at 0.5m. .................................................... 48
Figure 5.22. Filling up process of the Villavicencio network (closing inlet at 0.5m tank
level) ....................................................................................................................... 49
Figure 6.1. Flowchart for database generation ............................................................... 52
Figure 6.2 Flowchart to find affected nodes by a given pollution source. ..................... 55
Figure 6.3 Population members for an optimisation problem in WDS using GA.......... 58
Figure 6.4 COPA module flowchart............................................................................... 59
Figure 6.5. Objective function concept for a particular solution (network state)........... 62
Figure 6.6 COPA module and its relationship with GLOBE program........................... 63
Figure 6.7. Discretization option in GLOBE.................................................................. 63
Figure 6.8 Boundaries for the variables ......................................................................... 63
Figure 6.9. Small example. Pollution after 48h simulation. ........................................... 64
Figure 6.10. Best solution for the small example ........................................................... 65
Figure 6.11. Solution found for the small example with GA, after 52 evaluations........ 66
Figure 6.12. Pareto front for the multi-objective optimisation of the small example .... 67
Figure 6.13. Villavicencio - Sector 11............................................................................ 68
Figure 6.14 Affected area of sector 11, Villavicencio after 24 hours of contamination
from J2119.............................................................................................................. 69
Figure 6.15 Optimisation results for sector 11, Villavicencio........................................ 69
Figure 6.16. Pareto front generated by NSGAX for sector 11, Villavicencio................ 70
Figure 7.1. GSM architecture ......................................................................................... 73
Figure 7.2 SMS Network Architecture........................................................................... 74
Figure 7.3 Division for the mobile phone service in Colombia ..................................... 77
Figure 7.4 Evolution of the communication services in Colombia ................................ 77
Figure 7.5. Participation of users per economical status ................................................ 78
Figure 7.6. General visualisation of the messages for managing the intermittent supply
in Villavicencio ...................................................................................................... 78
Figure 7.7 General visualization of the messages for managing the pressurized supply 79
Figure 8.1. QMagNet-V structure .................................................................................. 80
Figure 8.2. QMagNet-V main window........................................................................... 82
Figure 8.3. Dialog box for the Generation Databases module ....................................... 83
Figure 8.4 Dialog box for the possible source searching ............................................... 84
Figure 8.5 Dialog box of the estimator of the affected elements. .................................. 85
Figure 8.6. Dialog box for the operational actions ......................................................... 86
Figure 8.7 Disinfection-by product prediction dialog box ............................................. 88
Figure 8.8. Intermittent supply analysis dialog box ....................................................... 89
Figure 8.9. Maps tool within QmagNet-V...................................................................... 91

vi
List of symbols

GA: Genetic Algorithms

COPA: Changing Operation on Pollutant Affection
DBP: Disinfection -by Products
IDS: Intermittent Distribution Systems
INS: National Health Institute (Spanish initials)
WSP: Water Safety Plan
WDS: Water Distribution Systems
WHO: World Health Organization

vii
1 Introduction

The water distribution networks are the most common way to deliver drinking water to
the people settled in urban areas, because they provide direct access to a large amount of
inhabitants in a safe way. Nevertheless, this is not always true, especially in many cities
in developing countries, where the urban piped water is, if not continuously available,
far away to be safe. This situation is reflected within the World Heath Organization
estimations, which states that 1.1 billion people do not have access to safe and sufficient
drinking/water.

In order to face this situation the United Nations have taken into account within its
Millennium Development Goals, the so-called target 10, “Halve by 2015 the proportion
of people without sustainable access to safe drinking-water and sanitation”. WHO, as
support organization, offers its technical contributions, as the Guidelines for Drinking
Water and the Water Safety Plans, developed even before the MDGs declaration.
Additionally, the MDG target 18 states that “in co-operation with the private sector,
make available the benefits of new technologies, especially information and
communications”. In this regard, the social role that Hydroinformatics, starts to play
becomes clearer in the framework of the Millennium Declaration.

A demonstration of how Hydroinformatics can provide help to the water distribution

systems in developing countries for providing safe-drinking water is presented in this
research, taking the Colombian city of Villavicencio as a case study for analysis. For
this purpose, a collection of different tools was developed for solving different,
common water quality-related problems, which are combined in one software
application. The scope of the research does not include commercial or final user
software, and it is not intended to be a formal Water Safety Plan, but a novel way to
apply some of the proposed tools into the context of these plans.

The research starts with a description of the water quality problems in distribution
networks and how these problems are conceptualised within the WHO Guidelines for
Drinking Water (chapter 2), as a foreword of the objectives that are stated in the chapter
3. Chapter 4 presents a description of the Villavicencio drinking water system, as well
as the problems related with the intermittency of the water supply. After this description,
chapter 5 shows a new approach for switching the intermittent supply towards a
continuous one, and explains why this change would improve the safety of the water
supplied in the city.

Chapter 6 considers different approaches for managing contamination in WDS, by

detecting the source, the affected population and the optimal operational changes that
may be followed when a contaminant is detected in a pressurised, continuous supply
network; the description of different algorithms that were developed for this purpose is
shown in this chapter as well. Chapter 7 starts with a description of the well-known
Short Message Service through mobile phones, and describes how it may be used as a
tool for operative and emergency support; the chapter ends with the demonstration of
how the approach would work for the Villavicencio case. Chapter 8 describes the final
application, named “Water Quality Management in Networks, (QMagNet-V)”, obtained
after the combination of the different tools developed during the research.

8
Chapter 9 discuses each component of the research, including the literature review, the
pros and cons of the proposed approaches, as well as the elements that should be
considered before any of their implementations. After this section, the conclusions of
the investigation are presented in chapter 10 following the general research structure.
The recommendations from this research, which are mainly addressing further
researches to improve and complement the proposed methodologies, are presented in
chapter 11.

9
2 Background of water quality problems in distribution
networks

During the last decades the concern about the access to safe drinking water has been
increased all over the world. In the 1970s, it was part of International Health Care and
the World Water Conference and in the 1980s it was part of the water supply and
sanitation decade. In the new millennium, this issue has been in focus at the
Johannesburg World Summit for Sustainable Development in 2002 and, in the near
future, in the period 2005-2015, with the International Decade for Action, “Water for
Life” (WHO 2004). It is also identified as a critical condition for achieving the
Millennium Declaration Goals adopted by the General Assembly of the United Nations
(UN), in 2000.

The World Health Organization (WHO) has developed a framework with guidelines and
recommendations on how to improve the drinking water quality for protecting the
public health, which are applicable to any water supply system, including the water
distribution networks. As the current research fits in this framework and shows different
approaches for fulfilling some of those recommendations for the supply networks in
developing countries, the WHO framework will be briefly presented.

2.1 WHO guidelines for drinking water and the Water Safety Plans
The Guidelines for Drinking Water is a document developed by the World Health
Organization with the main objective of protecting the public health. The first edition
was released in 1984. In 1995 a plan for updating the document with the most recent
investigations was developed, which lead to the second edition in 1997 and later to the
third edition released in 2004. The current version was developed by 490 experts from
90 developed and developing countries with the experience and research results
collected along eight years (WHO, 2004). The WHO’s premise is simple: “Interventions
to improve the quality of drinking-water provide significant benefits to health”.

The framework followed with the guidelines is presented in Figure 2.1. It is focused on
the following five key points:

1. Health-based targets based on an evaluation of health concerns;

2. System assessment to determine whether the drinking-water supply (from source
through treatment to the point of consumption) as a whole can deliver water that
meets the health-based targets;
3. Operational monitoring of the control measures in the drinking-water supply that
are of particular importance in securing drinking-water safety;
4. Management plans documenting the system assessment, monitoring plans and
describing actions to be taken in normal operation and incident conditions,
including upgrade and improvement, documentation and communication; and
5. A system of independent surveillance that verifies that the above are operating
properly.

10
 Figure 2.1 Interrelationship of the chapters of the Guidelines for Drinking Water in ensuring
drinking-water safety
(Source: adopted from WHO 2004).

The Guidelines are supported by several, separated documents that provide background
information substantiating the derivation of the guidelines and providing guidance on
good practice towards effective implementation. One of them is the so-called Water
Safety Plan (WSP), which details the recommendations for the system assessment,
monitoring and management and communication procedures. The primary objectives of
a Water Safety Plan, according to WHO (2004), are:

ƒ Development of an understanding of the specific system and its capability to

supply water that meets health-based targets;
ƒ Identification of potential sources of contamination and how they can be
controlled; validation of control measures employed to control hazards;
ƒ Implementation of a system for monitoring the control measures within the
water system;
ƒ Timely corrective actions to ensure that safe water is consistently supplied; and
ƒ Undertaking verification of drinking-water quality to ensure that the WSP is
being implemented correctly and is achieving the performance required to meet

11
 relevant national, regional and local water quality standards or objectives.

The steps to build a Water Safety Plan are shown in the Figure 2.2:

Figure 2.2 Overview of the key steps in developing a water safety plan (WSP)
The scope of the present research includes some of the elements considered in the last
two steps, which are the management procedures for corrective actions, and the
communication procedures.

12
2.2 Water quality problems in water distribution systems (WDS)
Regardless of the uncertainty associated with information on water-related deceases
around world (coming from variety of the procedures used for monitoring and
reporting), the best estimates from public literature state that the range of deaths
associated with this problem falls between 2 and 5 million every year. In addition,
missed working days, educational opportunities, health attention costs and family
resources reduction are also consequences of a non-safe access to drinking water
(Hassan 2004).

The public health issue is not only related to the contamination of the superficial water
bodies, but also with the poor quality into the water distribution networks themselves,
and it is usually the deprived people who are most affected. In fact, it has been
documented that the poor people are those who are exposed to a higher risk from
infectious diarrhoeal decease from poor drinking water (Payment et al 2001). Godfrey
and Howard (2004) state that when the piped water supplies are contaminated, the poor
are more likely to use it either because they cannot afford a home treatment system or
because they do not have storage facilities to take water from as a secondary source.
Additionally, the bottled water solution is also too expensive to be used for the poor in
these situations.

Furthermore, the problem is related to the property of piped systems to spread deceases.
Despite the fact that drinking water may not be as important transmission route for
epidemic deceases as others (Cairncross 1990), it has been documented (Ford et al
1996), that the water may widely spread an outbreak even when it is not contaminated
(Godfrey et al 2004). In other words, the water in the distribution system may act as a
transport device for the deceases.

In addition to the mentioned problems, the methodologies for testing the water quality
are not helping too much for reducing the threats. The traditional method of “end-
product-testing” for monitoring the drinking water quality is useless for preventive
action, due to two basic reasons (Godfrey et al 2004, Medema et al 2003, OECD–WHO
2003): Firstly, it works with a small amount of water quality data compared with the
amount of water produced, and therefore the conclusions about the water quality are too
compromised. Secondly, the water quality analyses take so much time to show results
that by the moment of its availability the water has been already delivered to the
consumers.

As will be shown, this research is making a contribution for reducing the water quality
threats in distribution systems by making use of specific computer applications. These
applications can be seen as building components for a full-fledged Early Warning
System (Brosnan 1999). Such a system consists of water distribution model, water
quality model integrated with tools for operational guidance and for rapid
communication of warnings to all concerned parties - operational managers, as well as
consumers.

2.2.1 Description of quality problems in WDS

The causes for water quality problems can be of very diverse nature. Although Brosnan
(1999) concluded that the water supply systems are primarily vulnerable to pollution
either by intentional actions or accidental events, Ainsworth (2004) has established that

13
harmless bacteria may already exist into a distribution system even in presence of
residual disinfectant. The same author states that in spite of the inoffensive nature of
these bacteria, it can contribute to the growing of other life forms. This approach is also
supported by Maier (1999), who asserts that even the most advanced treatment may
allow passing small amounts of bacteria into the distribution system.

Under these considerations, it is possible to observe that it is not necessary to wait for a
special event to obtain contaminated water. Intentional actions (such as terrorist attacks)
are less frequent problems in developing countries. Accidental events can sometimes be
with serious consequences, and need to be treated as special cases. However, most of
the causes of water quality problems occurring on a more regular basis in developing
countries are not of such nature. The set of water quality problems that are subject of
this thesis, are in fact all associated with ‘normal’ operation of the network, as presented
in the following sections.

2.2.1.1 Water Age

According to AWWA (2002), the water age is one of the most important factors for
water quality deterioration within the distribution systems, and it is affected by two
main processes: the interactions between the pipe wall and the water and the reactions
into the mass of water itself. These processes affect the water quality during the
distribution leading to physical, chemical and aesthetic transformations.

Chemical transformations, include disinfections by-product formation and corrosion-

related problems, while biological transformations include disinfections by-product
biodegradation, nitrification and microbial re-growth. The physical issues, which do not
have potential impacts on public health, are temperature, sediment deposition and
colour. (Table 2.1)
Table 2.1 Summary of water quality problems associated with water age
Chemical issues Biological Issues Physical Issues
*Disinfection by-
*Disinfection by-product
product Temperature increases
Formation
Biodegradation
Disinfectant decay *Nitrification Sediment Deposition
*Corrosion control *Microbial regrowth /
Colour
effectiveness recovery / shielding
Taste and odour Taste and odour
* Denotes water quality problem with direct potential public health impact.
(Source: adopted from AWWA-EES (2002).

2.2.1.2 Inadequate water treatment in extreme events

Periodic excesses of sediment load from the source may overwhelm some treatment
processes, such as filtration, that may lead to the pathogens entering into the system.
These can be so important, that they can affect the served population with detectable
gastrointestinal illness (OECD–WHO 2003).

2.2.1.3 Integrity of the water distribution system

When the water leaving the treatment plant is deemed to be safe, several cross

14
connections may re-contaminate the water. The following are four types of such
connections (OECD–WHO 2003):
1. Infiltration
There are three conditions that must exist together for water contamination through
infiltration: first, contaminated groundwater surrounding the network, because of sewer
systems leakage, for instance. Second, low or nil pressures into the pipes, situation that
may occur when a service interruption is required or when deficient operation of the
network is taking place. And finally, the deficient state of the pipes, such as non-
hermetic joints, or pinholes presence due to corrosion.
2. Back siphonage
This situation occurs when negative pressures in the system may suck in contaminated
water, though submerged supply pipes into domestic tanks. It is clear that this problem
is due to bad plumbing practices at housing levels.
3. Open drinking water storage reservoirs
Storage reservoir contamination is likely to occur when it is opened. Faecal
contamination may produce because of birds’ excreta. It can also occur that the birds
enter into the reservoirs through non-protected roof vents.
4. Line construction and repair
Every time that a main pipe has to be repaired, it is necessary to fulfil strict procedures
to avoid the introduction of contaminated soil or debris with the consequent entering of
pathogens into the network distribution.

2.2.1.4 Microbial re-growth in the distribution system

As was stated before, various bacterial species might be present in the network even
when faecal indicators in the treatment plant are not detected, and they can grow under
favourable conditions in the pipe walls, which may lead to the formation of the so-
called biofilm. (Howard 2002). Although he also states that the biofilm may deteriorate
the water quality in the distribution networks, Payment et al (2001) report that the
biofilm formation has not current relation with any outbreak of waterborne decease.

This circumstance is recurrent in several literature sources (Ainsworth 2004, Havelaar et

al 1992, WHO 2004), and it proves that the current knowledge about the implications of
the biofilms in human health has not been completely understood.

2.3 Water quality problems in intermittent distribution systems (IDS)

When a water distribution system does not provide water 24h per day to their customers,
but they are supplied by turns during some hours, such system is known as intermittent.
The intermittency of the water supply in distribution networks is a very common
situation in cities and towns of developing countries, where the continuous supply is
sometimes considered as a ‘pipe dream’ (World Bank 2004). These systems may have
zones of the network with completely empty pipes, while some other zones receive
water with low pressures.

The causes of the intermittency in water supply systems are derived most of the times
from a water scarcity in the region or from poor operation and maintenance practices.
(Totsuka et al 2004). Nevertheless, water consumption culture of the users of the system
may also be an important cause for discontinuous service, as well as the general belief
that an intermittent system is least costly than a 24h-piped supply.

15
The consequences of having a system working under intermittency are, among others,
the following, (Totsuka et al 2004):
• Inequitable water distribution among the population.
• Higher costs for the costumers, due to the necessity for the system to have additional
facilities, such tanks, pumps or alternative supplies.
• Wasting water as a result of the generalized mistrust on the system, which leads to
the people to storage as much water as they can.
• Higher costs for the water supplier, in the sense that the system needs more
operational efforts to work.
• Availability of water at inconvenient hours for the people.
• Malfunctioning of the households’ water meters due to the presence of air, which
leads to an inaccurate reading, with the consequent negative economical impact over
the population and the supplier.
• Water contamination and health hazard.

Furthermore, according to the Mumbai Conference in Intermittent Water Supply held

on 2000, managers have the wrong perception of advantages of the intermittent supply,
such as reduction of the water leakage, equity in the distribution and enough time for
reparations. These beliefs may lead to refusing a possible change from intermittent
towards continuous supply.

2.3.1 Description of water quality problems in IDS

The water quality problems associated with the intermittency are discussed with more
detail, since it is the main issue for this research. Two processes can be distinguished as
source of contamination in intermittent systems: infiltration and backflow.

2.3.1.1 Infiltration

The infiltration occurs when non-treated water enters into the piped system. Where this
occurs, a concentrated “slug” of polluted water flowing through the network can
expected when the charging of the system is resumed, increasing the risk for the users
(Figure 2.3).

Deficient pipe condition b

Pipe with no pressure

Contaminated groundwater

Mixture treated – contaminated water

Figure 2.3 Infiltration in IDSs before and after the system resumes

As mentioned earlier, the three conditions that must exist together to have infiltration
are:

16
 • Contaminated groundwater surrounding the network.
• Nil or negative pressures within the network.
• Defects in the pipes.
1. Contaminated groundwater surrounding the network.

The presence of pollution in urban groundwater may be high. LeChevallier et al (2002),

carried out a study over eight utilities in the United States, by collecting 66 samples of
both water and soil immediately next to pipes and making a microbial analysis on all of
them. Total and faecal coliforms, as well as viruses and microorganisms where
measured, and the results are shown in Figure 2.4.

Figure 2.4 Summary of microbial occurrence in Water and Soil Samples

Source: LeChevallier et al (2002)

2. Low or nil pressures into the pipes

The situation of nil or negative pressures is associated more for fully pressurized
systems than intermittent supplies, since it is related to pressure transients, which are
caused by abrupt changes of the velocity of water, due to valve closure, pump stoppage
or pipe fracture (LeChevallier et al, 2002). Obviously, in intermittent supplies, the pipes
are emptying regularly and therefore the pressure in them is equal to the atmospheric
pressure.

3. Pipe defects

The pipes in a system may have different defects, due to deficient construction,
deficient maintenance or poor quality of its materials. Non-hermetic joints, or pinholes
presence due to corrosion are common problems.

2.3.1.2 Backflow

On the other hand, the backflow occurs when negative pressures in the pipes suck in the
untreated water from domestic storages, which are wrongly connected to the supply
network. In intermittent supplies this may happen when the system is emptying, in such
a way that a wedge of water that travels downstream sucks up the air behind it,

17
generating negative pressures. Some examples of situation when the supply water is at
risk of backflow contamination are shown in Figure 2.5

Figure 2.5. Examples of risks by backflow contamination

Source: Gleick 1998

18
3 Objectives of the present research

The primary objective of the present research is to gather and implement a suitable
collection of hydroinformatics tools for dealing with several different water-quality
problems frequently occurring in distribution networks that operate in developing
countries. Such tools can then be combined into one software application, which can
serve for overall improvement of water quality real-time management and operation of
these systems.

The case study of Villavicencio, Colombia, is utilized to demonstrate how the

application, and therefore these collected tools, could help in the management of such
systems, and how the water-related health threats could be mitigated. In its objectives
this research directly addresses the MDG target of halving, by 2015, the proportion of
people without sustainable access to safe drinking water and basic sanitation (part of
MDG 8: ‘Ensure environmental sustainability’).

The specific objectives of this research are associated with a subset of water quality
problems for which suitable approaches and tools will be developed. They are listed as
follows:
• To propose a new approach for switching an intermittent supply system towards
a continuous, pressurized supply system, by using an innovative modelling
methodology of these systems.
• To pose the operational responses to contamination events as an optimisation
problem, which can then be solved by applying optimisation methods based on
randomised search.
• To demonstrate how the mobile technology, particularly short message services
(SMS), can be used in real time management.
• Integrating of the developed tools for these three specific objectives into one
application, which can serve as a demonstrator of possible ways for facilitating
the access of water quality information for customers (to ensure safe drinking
water) and for operators (to have optimal operational advices during an
emergency situation).

The following chapter will first present the characteristics of the water supply system of
the town of Villavicencio, Colombia. The water supply system serves as a test case for
all tools and approaches developed in this thesis.

In the following chapters the approaches taken for achieving each of the four specific
objectives will be described. These chapters are organised in a similar fashion. They
first introduce short theoretical background and literature review, and then they
introduce the proposed methodologies on simple examples. Then, the developed
approaches are tested for the case of Villavicencio.

19
4 The supply system of Villavicencio, Colombia

4.1 General description

Villavicencio is the capital of Meta department in the central Colombia, Figure 4.1. It is
situated about 135km at the southeast of Bogotá, in the eastern foothills of the Andes, in
the right hand side of the Meta River. It is the chief urban centre of the llanos and
forests of Eastern Colombia. Coffee, bananas, rice, beer, and soap are the main products
of the region, and it is considered the entry door to the eastern plains of Colombia,
which occupy 40% of the national territory.

Figure 4.1. Location of Villavicencio city in Colombia

Villavicencio has about 318.000 inhabitants in the urban area. The drinking water
consumption is about 65 m3/day per subscriber, or about 450 L/day per capita, being
one of highest of Colombia. La Esmeralda is the treatment plant that supplies the city,
with a total capacity of 1.6 m3/s in average and it has the basic traditional processes,
from screening to filtration. Disinfection and pH stabilisation exist as well.

The situation of the excessive consumption is in part due to a 17% of the users do not
have meter and most of the houses have storage tanks with deficient overflow control.
Nearly 84% of the water consumption is residential, and half of this correspond to
medium-low income population.

The unemployment rate of 14%, combined with the fact that this is one of the most
expensive cities of Colombia (Banco de la República, 2004), is an indication of the

20
social difficulties that this city is facing. On the other hand, according to Arrieta (2002),
during the period 2001-2002 Villavicencio collected only the 67% of tests required for
water quality analysis by the decree 475-1998, Colombian law.

The combination of these situations leads to an interesting opportunity to select

Villavicencio for the application of the present research.

4.2 Brief Description of the Villavicencio drinking water system

4.2.1 Superficial Sources
The city of Villavicencio has five different superficial sources for supplying. Three of
them bring water to La Esmeralda Treatment Plant, whereas the other two provide water
to the Fuentes Altas system, which is not treated at the present time.

• Quebrada Honda: The main source of the city, since it provides 1600 l/s to the
system. It has worked since 1999, after four years of construction of the intake
with sand removal and the 15.8km-length and 825mm-diameter transmission
line that ends in La Esmeralda Treatment Plant. The river often destroys it when
the water arises until its foundation level.
• Guatiquía River: It has two different inlets, known as Puente Abadía and
Bavaria. Both intakes are utilized as an alternative support for the main source,
when the water destroys that pipe. The first source comes into La Esmeralda
Treatment Plant through a 9.5km-length and 610mm-diameter pipe, whereas the
second one arrives to the same plant through a 1.6km-length and 600mm-
diameter pipe. These sources can supply 600 l/s each.
• Caño Parrado: This source is located at 5km from La Esmeralda Treatment Plant.
It can provide 30 l/s in dry season and 100 l/s in rain season, discharges that are
transported through a 300mm-length pipe.
• Caño Buque Alto: This source is part of the so-called Fuentes Altas gravity
system, and supplies directly into the network system a discharge of 150 l/s
through a pipe of 3.5km-length and 200mm-diameter. It has not any treatment
process because this water was considered appropriate for drinking many years
ago. Nevertheless, nowadays, it is being built a new treatment plant for it.
• Caño Maizaro: This source is also part of the Fuentes Altas gravity system. It
provides approximately 160 l/s through a pipe of 450-diameter. Both sources of
Fuentes Altas are mixed in a common reservoir, from where two pipes of
200mm-diameter supply a specific area of the city.

4.2.2 Deep Well sources

There are eleven deep wells in different areas of Villavicencio. These wells are utilized
when the superficial sources have insufficient water in the dry seasons or when the
network becomes depressurised for excess in the demand. Two out of those wells, are
supplying neighbourhoods that are completely isolated from the municipal network
system. Six out of the other nine wells are working at this moment, and the other three
are not, because those are too old or supply insignificant discharge. In total, the
discharge that is provided by the wells is about 80 l/s.

21
4.2.3 The network
The supply network of Villavicencio is formed by 350 km of pipes with diameters
between 10 and 84cm; detailed information can be obtained from Table 4.1. The
treatment plant is located at the north – west part of the city at the highest elevations, as
shown in Figure 4.2.
Table 4.1. Network characterization. Lengths in km per diameter and material (roughness)
Hazen Williams roughness coefficient
Diameter
130 150 112 120 140 91 135 100 Grand Total
mm
25 0,173 0,173
51 0,492 5,485 2,939 0,25 4,33 13,496
64 0,314 0,314
76 169,441 29,534 18,498 217,473
102 14,49 1,411 0,354 2,054 0,373 18,682
152 0,538 24,194 10,673 0,723 0,155 0,03 36,313
203 12,374 5,5 2,713 20,587
254 3,205 5,253 1,458 0,606 10,522
305 0,927 0,162 3,597 4,686
356 1,489 6,435 2,493 4,166 14,583
406 0,356 0,622 2,053 0,181 3,212
457 0,057 0,057
508 3,873 3,873
610 3,994 3,994
686 1,879 1,879
838 0,054 0,054
Grand Total 1,03 232,45 62,53 0,35 28,19 4,86 20,28 0,21 349,90

La Esmeralda
Treatment plant

Fuentes Altas
Altas
Source

Figure 4.2 Topography of the city, elevations in meters

Aquadatos (2000) designed the hydraulic sectorization of the network, in order to

improve the operative management of the system (Figure 4.3). The system was divided
in eleven independent sectors, which are separated by permanently closed valves.

22
 1 2

3
8
4
7

5 6 11

10
9

Figure 4.3. Network sectorization recommended by Aquadatos (2000)

4.2.4 The existing hydraulic model

In order to optimise the network operation, Aquadatos (2000) built a model of the water
supply system in WaterCad 5.0, with the following characteristics:
ƒ No pipes below 10cm were included (1405 nodes and 1953 pipes)
ƒ Consumption per capita 300 L/day.
ƒ Factor for wasting water of 50%.
ƒ Pipe roughness reduced according with their age and material
ƒ The valve status were kept as observed in the field.

For the purpose of this research, the model was rebuilt with the following characteristics:
ƒ EPANET-based.
ƒ Pipes from 5cm were considered. (2781 nodes and 4030 pipes. From the
hydraulics point of view, the pressure and discharges may not change, but the
water quality does, i.e. water age, which accuracy may change with a network
skeletonization, (AWWA 2002).
ƒ Consumption per capita 180 L/day without factor for water wasted.
ƒ Pipe roughness reduced according with their age and material.

4.2.5 The water quality

Different sources for the water quality have been consulted for the present research, but
unfortunately, the reports shows contradictory information. Certainly, the information
that the supplier reports to the Public Service Superintendence on February 2004 shows
that the quality measured in twelve different points in the network indicates that the
water was adequate for consumption. On the other hand, the National Health Institute
through the so-called “Information System for the Water Quality Surveillance,

23
SIVICAP” (Ortiz et al 2004), shows that 66% out of 672 measurements made in
different points of the network yield unacceptable value for total coliforms in 2001, as
well as 85% out of 13 measurements in 2002, 100% out of 3 in 2003 and 64% out of
407 in 2004.

In spite of these bad results, it has been found that the quality of the treated water, just
at the moment when the water leaves the treatment plant, has a good condition for
drinking, at least for the period when the information is available. Colour, pH and
conductivity have correct values, as well as the total and residual chloride for
disinfection, as is shown in Figure 4.4. This could be an indicator of that the treated
water is re-polluted in the network, and the most probable reason is the intermittency of
the service.
Chloride concentrations of treated water
leaving the plant
0,9
0,8
Concentration mg/L

0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
3- 04

4- 04

5- 04

6- 04

7- 04

8- 04

9- 04

-2 4

-2 4

4
-2 4

-2 4
-2 4

-2 4
10 0 0

15 00

00
13 00

14 00
11 00

12 00
20
20

20

20

20

20

20

-2

-2

-2
-2

-2

-2
2-

2-

2-

2-
2-

2-

2-

2-
2-

Date
Chloride Residual Chloride Total

Figure 4.4. Chloride concentrations for the treated water

4.3 The intermittency problem in Villavicencio.

As noted before, the users of an intermittent supply system tries to store as much water
as they can; they leave their storages opened because they do not know when the water
will come to their homes. In Villavicencio, Colombia, the circumstance is not too
different.

First of all, more than 70% of the population has domestic tanks with capacities
between 7 and 9 cubic meters, and all of them are practically under the houses
(Aquadatos 2000). Different measures carried out in different zones of the city showed
that these tanks receive water with discharges about 0.5 l/s when the pressure is 20
meters water column.

Secondly, these storages cause huge water demands. This can be proved according to
the data collected by Aquadatos (2000), who installed new micrometers in several
homes where the instrument did not exit. Those micrometers were monitored during one
week, without previous users’ awareness and with no effect on their billing, in order to

24
avoid any change of the behaviour consumption. A summary of the conclusions is
shown in the following graph (Figure 4.5), and listed further:
Average consumption per capita in
Villavicencio

1000
Consumption (L/day)
800
636 Billing
600 Theoretical cons.
Micrometers
400 251 Measurements
180
200 112

Figure 4.5. Average consumption per capita in Villavicencio

Source: Aquadatos (2000)
ƒ The consumption that is billed, in an average basis, is 33% lower than the
theoretical consumption for a city with the characteristics of Villavicencio.
ƒ The average consumption of one user without micrometer is 88% more than the
average consumption of one user with micrometer.
ƒ 57% of the users are consuming between 0 and 20 m3/day, meanwhile 32% is
consuming between 20 and 40 m3/day.
ƒ If the consumption from the micrometers were taken, then the total discharge for
the city would be 506.87 l/s.
ƒ The measurements were performed at the entrance of particular sectors. The
paramount difference between the micrometers value and the other
consumptions proves the exaggerated use of residential tanks.

Furthermore, the water production from the Treatment Plant was also measured. For
April of 2000, the following water production curve was obtained (Figure 4.6):

25
 Average consumption curve for the city of Villavicencio
for the month of April 2000
1,850

1,800

1,750

1,700
Discharge (lps)

1,650

1,600

1,550

1,500

1,450

1,400

1,350
09:00

10:00

11:00

12:00

13:00

14:00

15:00

16:00

17:00

18:00

19:00

20:00

21:00

22:00

23:00

00:00

01:00

02:00

03:00

04:00

05:00

06:00

07:00
Hour of day

Figure 4.6. Daily production curve in the treatment plant during April 2000
Source: Aquadatos (2000)
It can be observed that the production for the city is about 1,6 cubic meter per second,
for 318,000 inhabitants. A simple relation leads to the fact that each person has an
equivalent demand of about 435 l/s in average. It can also be observed that the
consumption curve is completely different to a curve of a pressurized system. Two facts
reflect the intermittency impact: first, the peak at 10:00 pm, and second, the high water
production at nocturnal hours. To have an idea if the intermittency of the water supply
in Villavicencio, 12 turns per week in one of its sectors was found by Aquadatos (2000).

Although one of the possible reasons of the excessive water production showed in
Figure 4.6 is the leakage in pipes, Aquadatos (2000) found that only 16% of the leaks
repaired during that year were in the piped system, while the rest were in the household
connections.

In summary, the city is producing three times more water than what it really needs, but
in contrast, the inhabitants are actually consuming a normal volume of water.
Additionally, they are being charged with a tariff lower than the real consumption a
according to the characteristics of the city. This means that the treated water is being
wasted for filling up an important number of huge tanks.

The non-revenue of water by volume, the water losses and other performance indicators
were compared among the most important Colombian cities and documented by
Umbarila and Alfonso (2003). The Figure 4.7 and Figure 4.8 show clearly the situation
for Villavicencio.

26
 Figure 4.7. Comparison of non-revenue water by volume in different Colombian cities
Source: Adopted from Umbarila and Alfonso (2003)

Figure 4.8. Comparison of water losses in different Colombian cities

Source: Adopted from Umbarila and Alfonso (2003)

27
5 From intermittency to continuous supply: a new approach

5.1 Introduction
As stated in section 2.3, the intermittency is not an ideal condition to operate a network
since it leads, not only to a deficient water supply in terms of quality and quantity, but
also to higher costs for both operators and customers. The present chapter looks for an
approach that allows starting a sort of metamorphosis on these systems, to pass from
intermittent to pressurized condition, working on the water consumption culture. The
main idea is to develop a model to be used for predicting the water reception times to
each household and the time when their domestic tanks might be completely filled. This
information would be sent to the costumers in order for them to plan their water use,
preventing the unnecessary big water storage, and consequently the huge wasting water
that characterize these systems. It has to be noted, however, that this will be only
possible in regions where the water resources are available during the whole year and
where the problem is a lack of good management.

5.1.1 Objectives for modelling the intermittency

Wasting water may be considered as part of the water consumption culture, so it is both
a cause and a consequence of having intermittent systems in those settlements where the
water resource is available during the year. This vicious circle is, therefore, a very
difficult issue. A way to convert the intermittency state to the continuous one is, then, to
work in the culture consumption.

The main objective for modelling the intermittency is to have information about who is
getting water and at what time, when the systems resumes, in order to develop criteria to
operate the system under this condition. Other objectives are:

• To know the water reception times of the users after the system restarts.
• To know how long time is required to fill up all the tanks of the city.
• To know the time when the tanks are completely filled.
• To establish how long time is needed for the network to be completely pressurized
after the system restarts.
• To find out the best time for closing the domestic taps in order to help the system to
become pressurized.

5.2 Modelling Approaches

An extensive literature review has shown that the problem of modelling an intermittent
system been studied by Srinivasa -Reddy and Elango (1989), Vairavamoorthy(1994,
2000), Akinpelu (2001) an others. The main consideration here is that intermittent
systems have no demand pattern; the storage tanks of the customers will fill up
whenever the systems provides water, until they are full and the float valve closes.
Therefore, the customer demands are pressure-dependent, since the tank inlets behave
as uncontrolled free flowing outlets.

Macke et al (2000) developed a software interaction using EPANET and ARC-VIEW in

28
order to simulate the conditions of the intermittent supply, by using storage tanks
instead of pressure junctions. Nevertheless, that model has two main disadvantages,
such as the disability of modelling the network when the pipes are empty or at free flow
and the fact that the tanks have not their own household demand.

Godfrey and Howard (2004) developed a charging model to simulate the charging up
process of the water network after supply resumes. It takes into account the assumptions
made by Liou and Hunt (1986), which are numerated as follows:
• The pipe that is being charged has a water column with a well-defined front.
• The pressure is atmospheric at the front of such column.
• The water is incompressible.
• Frictional flow-resistance is used for steady flow.

Unfortunately, it was not possible to obtain that model during the time of this research.

Other researchers, (Gomez et al 2001; Vasconcelos et al 2003a – 2003b), which were

not specifically dedicated to the study of the intermittent supply, has produced different
models for the phase transition from free surface flow to pressurized flow for drainage
and sewer systems. The second authors stated that the rigid column model is able to
predict the fill box behaviour fairly well, although it is not able to give a detailed
pressure distribution along a pipeline, and that the Preissmann Slot based models are
able to predict that detailed description, both during the pipe filling process and when
closed pipe transients are produced. (Vasconcelos et al 2003c)

5.2.1 Proposed approach

An approach, firstly studied by Akinpelu E. and Macdonalds in a MSc thesis in
Cambridge in 1997 (unfortunately, it was not possible to obtain any related document),
which uses the concepts of the free surface flow is suggested again. For this purpose,
the numerical model named Storm Water Management Model SWMM 5.0 (Rossman
2005a) is used.

SWMM is a dynamic rainfall-runoff simulation model used for single event or long-
term (continuous) simulation of runoff quantity and quality from primarily urban areas.
It was first developed in 1971, and different researchers have improved it afterwards. In
the latest version (5.0), the code was completely rewritten with modern software
engineering techniques. Special emphasis in the numerical sub-model known in
previous versions as Extran, was made (Rossman 2005b), which includes a method for
the solution of the full dynamic wave flow routing that allows the simulation of
surcharge conditions, backwater flow and pressurized flow.

SWMM solves the Saint Venant equations, which for flow along an individual conduit
can be expressed as:

29
where x is distance along the conduit, t is time, A is cross-sectional area, Q is flow rate,
H is the hydraulic head of water in the conduit (elevation head plus any possible
pressure head), Sf is the friction slope (head loss per unit length), hL is the local energy
loss per unit length of conduit, and g is the acceleration of gravity. The area A is
function of flow depth y, which in turn can be obtained from the head H. Thus the
dependent variables in these equations are flow rate Q and head H, which are functions
of distance x and time t. Besides,

where n is the Manning roughness coefficient, V is the flow velocity (equal to the flow
rate Q divided by the cross-sectional area A), R is the hydraulic radius of the flow’s
cross-section, and k = 1.49 for US units or 1.0 for metric units.

An additional continuity relationship is needed for the nodes where the pipes are
connected, which assumes a continuous surface area between a pipe and the node where
is connected.

where Astore is the surface area of the node itself, ΣAs is the surface area contributed
by the conduits connected to the node, and ΣQ is the net flow into the node (inflow –
outflow) contributed by all conduits connected to the node. SWMM solves the dynamic
wave flow routing model with a method of successive approximations under relaxation,
to check the head values at the nodes. This method is similar to the iterative method of
previous versions. Additionally, a variable routing time step that tries to adjust the
Courant number criteria when the simulation unfolds is used.

Nevertheless, when surcharge condition is faced, the surface area of the conduit
becomes zero and the previous equation is no longer applicable. SWMM 5.0 uses then
an additional equation, which corresponds to the continuity equation for a node in a
looped, pressurized system and states that the flow that enters a node must be the same
as the flow that comes out from it. In this case, no under relaxation is performed over
the resulting heads.

The stability criterion of the numerical solution is given by the Courant number, which
comes from the restriction for the celerity to be less than the velocity of the pipe, and is
defined from the SWMM numerical method as:

L = tV
(1 − Fr )
Fr

where L is the conduit length, t is the time step, v is the velocity in the conduit and Fr is
the Froude Number. Substituting Fr by its definition, yields,

30
 L=t ( gy + V )
This is a particular case of the stability expression Courant-Friedrich-Levy, which states
that the maximum time step of the model must be:

In which T is the narrowest free surface width at the given channel cross section, and
that corresponds to the narrow width of the Preismann slot when surcharged conditions
take place.

The idea behind the utilisation of SWMM for modelling the intermittency in a supply
network is that the flow within a pipe might have initially a free surface behaviour when
the system resumes, and then it becomes totally full. The well-defined front with a
vertical column is not obtained anymore, but the obtained front depends on the
hydraulics within the pipe. This hypothesis could be particularly true in sloping
networks, where the lowest part should be filled up first.

5.2.1.1 General assumptions

The assumptions for modelling a supply network with a tool conceived for modelling a
drainage system are the following:
• An infinite-capacity storage tank with an initial level is set to simulate the source of
water (treatment plant), at constant head.
• The pressure junctions are simulated as manholes, which are treated as piezometers
during the analysis. The cross sectional area of each piezometer is set to about
0.00001 m2, in order to avoid additional volume for water storage during the
simulation.
• The height of each piezometer is set at the same level of the initial water level of the
huge storage tank that simulates the treatment plant, in order to avoid overflows at
the nodes.
• A storage tank is connected to each piezometer by means of a conduit that allows a
head-dependant flow through it and that starts flowing when the head in the
piezometer is about 2m. This configuration simulates the vertical distance between
the household connection and the outlet to the tank.

5.2.2 Pilot models tests

In order to observe the behaviour of SWMM as a tool for simulating the charging
process, two pilot models were carried out. The first model corresponds to one single
pipe and the second one uses a simple looped network.

5.2.2.1 Single pipe model

It is expected that the nodes are filled as soon as the water fills up its previous pipe
completely, and the normal behaviour of a pressurized pipe (i.e. steep piezometric line
due to friction head losses) is achieved at the end of the simulation. The analysed
system consists in a 0.2m-diameter pipeline, connected at the upstream side to a storage

31
 tank that is big enough, to ensure a constant head of 4m during the analysis. A
horizontal pipeline is considered, as well as nine computational points (manholes) along
the pipe. Furthermore, a controlled orifice is placed at the storage outflow, which is
initially closed, but switches to the opened state after 10 seconds of simulation.

The following scenarios were addressed:

• Case A: With the downstream end of the pipe as free outfall.

• Case B: With the downstream end of the pipe as a 2m-high outfall (weir).

For each case, the following outputs are observed:

o The effect of different time step / pipe lengths relationships.
o The effect of the manholes’ storage area.
o The effect of the variable time step option.

o The effect of different time step / pipe lengths relationships.

o The effect of the manholes’ storage area.
o The effect of the variable time step option.

CASE A: FREE OUTFALL

Table 5.1 Case A: free outfall

Analysis Time Variable Manhole
Length Max V Max Q
time step Time storage Instabilities Comments
(m) Pipe 2 Pipe 2
(min) (s) step area m2
60 10 100 No 0.00001 3.4 87 Medium
60 50 100 No 0.00001 2.3 72 No Ok
60 100 100 No 0.00001 23 714 High
60 10 100 Yes 0.00001 3.4 87 High
60 50 100 Yes 0.00001 2.3 71 No Ok
60 100 100 Yes 0.00001 3.2 101 Starting Rest Ok
60 10 100 No 1 3.2 84 Medium
60 50 100 No 1 2.3 72 No Ok
60 100 100 No 1 21 649 Very High
60 10 100 Yes 1 3.2 84 Medium
60 50 100 Yes 1 2.3 71 No Ok
60 100 100 Yes 1 3.2 101 Starting Rest Ok

32
 Best profile: Water Elevation Profile: Node 1 - 11

10

11
1

9
1 2 3 4 5 6 7 8 9 10
5

Elevation (m)
3

1
11

0 100 200 300 400 500 600 700 800 900

Distance (m)
01/31/2006 00:05:00

Example Water Elevation Profile: Node 1 - 11

instability at

10

11
1

9
1 2 3 4 5 6 7 8 9 10
the 5

beginning: 4
Elevation (m)

1
11

0 100 200 300 400 500 600 700 800 900

Distance (m)
01/31/2006 00:01:40

Example of Water Elevation Profile: Node 1 - 11

high

10

11
1

9
1 2 3 4 5 6 7 8 9 10
instability 5

4
Elevation (m)

1
11

0 100 200 300 400 500 600 700 800 900

Distance (m)
01/31/2006 00:23:20

Figure 5.1 Stability results for case A

CASE B: 2m HEIGHT OUTFALL

Table 5.2 Case B: Outfall at 2m-height

Analysis Time Variable Manhole
Length Max V Max Q
time step Time storage Instabilities Comments
(m) Pipe 2 Pipe 2
(min) (s) step area m2
60 10 100 No 0.00001 3.4 87 High
60 50 100 No 0.00001 2.3 72 Low Ok
60 100 100 No 0.00001 22.7 700 Very High
60 10 100 Yes 0.00001 3.4 87 High
60 50 100 Yes 0.00001 2.3 71 Ok
60 100 100 Yes 0.00001 3.2 101 Starting Rest Ok
60 10 100 No 1 3.2 84 Medium

33
60 50 100 No 1 2.3 72 Ending Ok rest
60 100 100 No 1 21 649 Very High
60 10 100 Yes 1 3.2 84 Low Ok
60 50 100 Yes 1 2.3 71 Ok
60 100 100 Yes 1 3.2 101 Starting Ok rest

Best Water Elevation Profile: Node 1 - 11

profile:

10

11
1

9
1 2 3 4 5 6 7 8 9 10
5

4
Elevation (m)

3
11
2

0 50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850
Distance (m)
01/31/2006 00:10:00

Example Water Elevation Profile: Node 1 - 11

instability

10

11
1

9
1 2 3 4 5 6 7 8 9 10
at the 5

beginning 4
Elevation (m)

3
11
2

0 50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850
Distance (m)
01/31/2006 00:01:40

Example Water Elevation Profile: Node 1 - 11

of

10

11
1

1 2 3 4 5 6 7 8 9 10
instability 5

at the end 4
Elevation (m)

3
11
2

0 50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850
Distance (m)
01/31/2006 00:30:00

34
Example Water Elevation Profile: Node 1 - 11
of high

10

11
1

9
1 2 3 4 5 6 7 8 9 10
instability 5

Elevation (m)
3
11
2

0 50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850
Distance (m)
01/31/2006 00:25:00

Figure 5.2 Stability results for case B

Comments:
• For time step 10s, the profile advances with pulses (oscillations), whenever a node is
reached. Anyway, these solutions are always stable, this is, the initial errors remain
small when iteration progresses (Popescu, 2004).
• The best result is when the time step is 50s, and it is not affected by the variable
time step parameter, neither by the value of the manhole surface area.
• For time step of 100s, the instability is very high, unless the automatic time step
option is used. Nevertheless, still some fluctuation can be observed in the first steps.
• The best result gives a maximum discharge, which occurs at the beginning of the
simulation, of 72 l/s and a velocity of 2.3 m/s.
• The limit time step, just before instability is observed, is around 60s. (Best result
means that the discharge curve has neither oscillations nor peaks.)

A similar analysis was performed for a pipe with different slopes. The conclusions are
the following:

Slope: -1%
• The best time step (stability condition and no oscillations) is 24s. For 25s, the
solution becomes instable for both options of Normal Flow Limit.
• When the water surface slope smaller than conduit slope is used as Normal Flow
Limit, the best time step is 15s. Dramatic instability is observed for time step of
16s, and important oscillations are observed for time steps less than 15.
• When “Froude number bigger than 1” is used as Normal Flow Limit option, the
instability is impossible to avoid, even if the variable time step option is used.
• The sensibility to the Normal Flow Limit options are due to the fact that for this
slope, the flow is within the critical range, this is, Froude number between 0.9
and 1.1. The flow slope is then changed between the supercritical and subcritical
states and the calculation of the heads at the nodes results quite different in
successive time steps, leading to high instability.

Slope -4%,
• The best time step is 10s, but dramatic instability is observed for greater time
steps. The type of Normal Flow Limit used does not affect this result. Lesser
time steps imply oscillations in the initial part of the simulation.

Similar analyses for different slopes were carried out without using the variable time
step option. Figure 5.3 shows the relationship between the maximum time step and limit

35
for which the solution remains stable, for both negative slopes (in the same direction of
the flow) and positive slopes (in the opposite direction of the flow). It can be observed
that, for negative slopes, the time steps need to be smaller as the slope pipe is steeper; in
contrast, for positive slopes, the time steps can be larger as the pipe is steeper.

Relation between time steps and pipe slopes

manholes separation: 100m)

120
100
Limit Time Step (s)

80
60

40

20
0
-4 -3 -2 -1 0 1 2
Slope (% )

Figure 5.3. Relation between time steps and pipe slopes

The graph shown in the Figure 5.3 illustrates that the best option for modelling with
SWMM is considering the smallest time step as possible, in order to avoid instabilities
due to critical elements such as short or steep pipes included in the system.

5.2.2.2 Small network model

For looped, pressurized systems, SWMM 5.0 is also useful, since the Saint Venant
equations are solved with the Preismann slot approach, and the continuity equations at
the nodes correspond to the input-output flow balance. A comparison is made between
EPANET and SWMM for an identical system.

The conditions for both models are the following:

• Infinite source of water with a constant head of 50m.
• The network nodes have the same elevation, so the system is flat.
• All the pipes are 1000m-length, 0.2m-diameter and 0.01-manning roughness.
• Constant demand of 0.5 l/s in each node.

For the SWMM model, the following assumptions were made:

• The manholes (treated further as piezometers) are 52m-high, and have a surface area
of 0.00001m2.
• Each piezometer is connected to a storage, which invert elevation is one meter
below the piezometer invert elevation.
• The connection between the piezometers and the storages are made trough an outlet,
which allows flowing 0.5 l/s with no dependency on the piezometer’s height.
• Each storage has an output flow of 0.5 l/s, which is entered as a negative inflow in
the model; this condition makes the storages to remain emptying at the same rate as
they are receiving water.

36
The following figures (5.4 to 5.7 ) show some of the results:
12/16/2005 09:59:57

3.09
0.50 0.50
12.50 6.00 0.50 0.50 0.50
1.64 0.75

6.00 -2.41 -0.95 -0.40 -0.25

2.41 2.16
0.50 0.50 0.50 0.50 0.50
1.40 0.68
3.09

-2.16 -1.20 -0.62 -0.42

0.95
0.50
0.50 0.50 0.50 0.50
1.20 0.95 0.50
1.64

-1.40 -0.95 -0.57 -0.43

0.32
0.50 0.50 0.50 0.50
0.40 0.62 0.57 0.50
0.75

-0.68 -0.50 -0.32 -0.25

0.50 0.50 0.50 0.50

0.25 0.42 0.43 0.25 0.50

Figure 5.4. Flows in the pipes for the SWMM model, in l/s (after the network is completely filled)
Day 1, 12:00 AM

3.09
12.50 6.00
1.64 0.75

6.00 -2.41
-0.95 -0.40 -0.25
2.41
2.16

1.40 0.68
3.09

-2.16 -1.20 -0.62 -0.42

0.95

1.20 0.95 0.50

1.64

-1.40 -0.95 -0.57 -0.43

0.40 0.62 0.57 0.32

0.75

-0.68 -0.50 -0.32 -0.25

0.25 0.42 0.43 0.25

Figure 5.5 Flow in the pipes for EPANET model, in l/s

37
 12/16/2005 09:59:57

o2 0.00 0.00 0.00 0.00 0.00

50.00 49.74 49.54 49.49 49.47 49.47

0.00 o8
0.00 0.00 0.00
49.54 0.00
49.51 49.48 49.47
49.47

0.00
0.00 o12 0.00 0.00
49.49 0.00
49.48 49.48 49.47 49.47

0.00
o14 0.00
0.00 0.00
0.00
49.47 49.47 49.47 49.47 49.47

0.00 0.00 0.00 0.00 o16

0.00
49.47 49.47 49.47 49.47 49.47

Figure 5.6 SWMM node depths (after the network is completely filled)
Day 1, 12:00 AM

0.00 49.74 49.54 49.49 49.48 49.47

49.54
49.51 49.49 49.47
49.48

49.49
49.49 49.48 49.47 49.47

49.48 49.48 49.47 49.47 49.47

49.47 49.47 49.47 49.47 49.47

Figure 5.7 EPANET node pressures

It can be observed that both flows and pressures (piezometric depths) are practically the
same, which confirm the validity of SWMM for simulating the pressurized flow.
Nevertheless, it is not the pressurized condition that is going to be analysed, but the
filling up process. For this purpose, the water reception times are observed for different
storages, and this is found by looking at the flow time series for different points (Figure
5.8).

38
 Figure 5.8. Water reception times at the elements showed in Figure 5.6

In spite of the use of different ranges for the time step, oscillations during the first
moments of the simulation, where the pipes changed from free surface to full state,
could not be avoided. Figure 5.9 shows the velocity value for the pipe that is connected
to the big storage:

Flow in the link connected to the big storage

200.0

180.0

160.0

140.0

120.0
Flow (LPS)

100.0

80.0

60.0

40.0

20.0

0.0
1 2 3 4 5 6 7 8 9
Elapsed Time (hours)

Figure 5.9. Flow in the link connected to the big storage

The relationship between this oscillations and the change on the pipe state from free
surface to full is evident if both flow and depth of such pipe is analysed for the first
instant of the simulation, for which the following Figure 5.10 is used:

39
 Flow and Depth comparison in the link connected to the big storage

Flow (LPS) Depth (m)

200 0.3
180
0.25
160
140
0.2

Depth in pipe (m)

120
Flow (lps)

100 0.15
80
0.1
60
40
0.05
20
0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45
Elapsed time (hr)

Figure 5.10. Flow and depth comparison

It can be observed that both depth and velocity start increasing until a point when the
pipe is completely filled. Then a sudden drop of the velocity takes place and this occurs
exactly when the water reaches the downstream end of the pipe, which becomes
surcharged. The second oscillation wave corresponds to the surcharge of the next pipe
connected at the downstream side of the current one. These oscillations could not be
avoided by using different time steps.

After 3.5h of simulation, the last node receives water, and then the pressure wave
suddenly travels backward, allowing the whole nodes reaching the maximum depth
possible, which corresponds to the pressure head in the pressurized system.
Unfortunately, no access to data in a real network or laboratory experiment to prove this
behaviour was found.

5.2.3 Assumptions and limitations using SWMM

• No intermediate computational points along a conduit are available, thus, only
the upstream and downstream are used. This may lead to misconceptions when
the results are being analysed.
• The status report prepared by SWMM after a simulation only shows the top-five
junctions with continuity problems, while, in some cases, the complete list is
needed for a better understanding of the model.
• The SWMM 5.0 user cannot set a convergence tolerance on the maximum
fractional difference for flows found between iterations, neither the maximum
number of iterations allowed. This can only be done by changing the source
code and recompiling it.
• When visualizing a pipe profile, no possibility of zoom exists.
• Although the open SWMM code was carefully explored, no clue of the
Preissmann slot width was found, parameter that could explain some of the
instabilities of the numerical method in particular cases.

40
5.3 The Villavicencio’s Intermittent Model
5.3.1 Description
A 2839 nodes and 4131 pipes model of the Villavicencio city was build to be used in
SWMM 5.0.

La Esmeralda treatment plant

Fuentes Altas

Figure 5.11. Villavicencio’s SWMM model scheme.

The following considerations were made, additionally to those mentioned previously:

• All of the valves in the system are opened when the system resumes.
• A completely empty network before the system resumes is assumed.
• Each storage tank represents a group of real household tanks, and they are
located at the centroid of the consumers that are connected to it.
• The capacity of each tank is estimated taking into account an average of 8m3 per
household and a constant depth of 1m for all of them. The surface area does not
change with the tank depth.
• The invert elevation of each storage tank (the bottom of the storage) is set to one
meter below the invert elevation of the node from which is connected. This
simulates the subterranean condition of those tanks in reality.
• The Villavicencio’s secondary source known as Fuentes Altas was not taken into
account.
• A flow routing time step of 30 seconds with the variable time step option (to
reduce instabilities) was used.
• The configuration of the nodes is as shown in Figure 5.12. The outlet element
allows having pressure dependant flow Q = f(h), where h is the water depth in
the node.

41
Figure 5.12. Model nodes configuration, where Q: Flow through the outlet; Qd; household
demand flow

• Aquadatos (2000) measured the discharge in some of the Villavincecio’s tanks

and found and average flow of 0.5 l/s for 20m pressure. This value was taken
into account for defining the relationship Q outlet = 0.05*h^0.5.
• A normal daily consumption curve (i.e. the obtained from a fully pressurized
system) is set as an outflow discharge time dependant-curve from each storage
tank. A typical consumption curve was taken from measurements made in
Roldanillo, Valle, a city with similar weather as Villavicencio, (Aquadatos,
2002), Figure 5.13.

low med hig ave

Daily average consumption

3.5
3
2.5
factor

2
1.5
1
0.5
0
0 2 4 6 8 10 12 14 16 18 20 22 24
Hour of day (starting from midnight)

Figure 5.13. Typical demand curve for a pressurized system in terms of daily average
consumption factors

• In the SWMM model, this outflow is referred as a negative value for Flow
Constituent in the Dry Weather option, with an hourly time pattern defined by
multipliers, Figure 5.14. Furthermore, the average value shown in the same
figure is calculated as a proportion of the area, which is served by a particular
node. These areas were estimated by means of the Thiessen polygons.
• After several model tests, the total period of simulation chosen was of seven
days. As it will be shown, longer period of simulation is needed to fill up
completely the whole tanks. Nevertheless, the computational effort is so time-
consuming (simulating a period of 1 hour requires a modelling time of more
than 1 hour), that this simulation time was chosen as a proper time to fulfil the
requirements of this research.

42
 Figure 5.14. SWMM dialog box to configure a pressurized demand pattern

The adopted methodology is described as follows:

Two different scenarios related with the domestic storage closure were considered:
• The user keeps the storage tap open until it is completely filled (1m).
• The user closes his/her storage until half of its depth is reached (0.5m).

The simulations are done using the following control rules, which were set for each
node:
• If the flow in a particular outlet is bigger than 0, the outlet flow is multiplied by
a factor of 1. This means that the supplied water has arrived to the domestic tank
that is connected to such outlet. The multiplication of the flow by the unity is
made for reporting purposes.
• If the depth of a particular domestic storage is bigger than 1, then the outlet from
which it is receiving water is closed.

In order to achieve the objectives stated before, the water reception times to the users
after the system restarts, the discharge time series for each outlet connected to a
domestic storage were analysed. Similar time series can be used to know the time when
these tanks become full and the required time for achieving this. This information is
stored by SWMM into an output file, which prints, among other data, the control results.

The SWMM 5.0 interface can only show the time series for up to six different elements.
In order to get these results for every node, a simple application that reads the output
file was developed (further details in chapter 7).

5.3.2 Results
The results for the two considered scenarios are presented as follows:

43
5.3.2.1 Closing the tanks when its water level is 1m.
04/02/2000 01:00:

Link
Capacity
0.00
0.33
0.66
0.98

Figure 5.15. Link capacities (relation current depth / full depth) at 1 hour
One hour after starting the filling of the system is observed that the almost all pipes are
completely full, and only a small neighbourhood located at the south part of the city still
has pipes with free surface flow (Figure 5.15).

Nevertheless, at this time quite few domestic tanks are receiving water, including some
houses located at the northeast of the city; It can be observed in the Figure 5.16 that
after 6 hours about 50% of the city starts receiving water into their tanks and after 24
hours, in the lowest part of the city, some tanks are completely filled up. It can be
considered that the situation becomes static after 2 days, in the sense that no significant
changes are observed if compared with the situation at 4 days, except for the fact that
some tanks of the lowest part, where the water level starts decreasing because of the
household consumption, and because those tanks are already closed, since the level of
1m was achieved. The following legend applies for the Figure 5.16

44
 1 hour 6 hours

12 hours 24 hours

2 days 4 days
Figure 5.16. Filling up process of the Villavicencio network (closing inlet at 1m tank level)
The water reception times are shown in Figure 5.17. It can be confirmed that practically
the whole city starts receiving water during the first hour, when the pipes are already
pressurized, as shown in Figure 5.15. After the first 15 minutes, and during the

45
following 30 minutes, about 70% of the city has received water corresponding mainly to
the households located in the lowest part of the city.
Water reception times hystogram
(taps closed when tanks water level is 1m)

350 100%
90%
300
80%
Number of housholders

250 70%

200 60%
50%
150 40%
100 30%
20%
50
10%
0 0%
0

00
:0

:0

:0

:0

:0

:0
0:

0:

5:

5:

0:

0:

5:

0:

5:

5:
00

05

15

30

45

10
:1

:2

:2

:3

:4

:5

:5

:0

:0

:1
:

:
00

00

00

00

00

00

00

00

00

00

00

00

01

01

01

01
Time (starting when system resumes)

Frequency % accumulated

Figure 5.17. Water reception times (closing inlet at 1m tank level)

Times when tanks are completely full

(time when tank should be closed)
Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Day 7
60 100%
90%
50
80%
Number of housholders

70%
40
60%
30 50%
40%
20
30%
20%
10
10%
0 0%
4

2
0

8
16

24

32

40
48

56

64

72

80

88

96
0
8

10

11

12

12

13

14

15

16

16

Hours (starting when system resumes)

Frequency % accumulated

Figure 5.18. Times of tanks charging, tanks at 1m.

In spite of the relatively fast network charging, the situation for the households’ tanks
charge is quite slow. The Figure 5.18 shows that about a 35% of the tanks that received
water during the first seven days, are completely filled after two days of system
restarting, but, in contrast, nearly 15% of those tanks have not received water after the
seventh day.

It has to be mentioned that 1060 out of the 2720 domestic tanks (about 60%) were still
not completely filled up after 7 days of simulation. The reason for this is that the water

46
depth in the nodes where these tanks are connected is not high enough to allow larger
flows entering to them.

5.3.2.2 Closing the tanks when its water level is 0.5m.

04/02/2000 01:00:00

Link
Capacity
0.00
0.33
0.66
0.99

Figure 5.19. Link capacities (relation current depth / full depth) at 1 hour
Practically the same result compared with Figure 5.15 is observed, in the sense that
almost all the pipes are completely full with exception of the south part. The same
exercise done for the first scenario yields imperceptible differences in the accumulated
percentages, as can be noted by the comparison between Figure 5.17 and Figure 5.20.
This was expected, since the closure of the tanks has not taken place in the first hour.

It was also expected to obtain reduction of required times to fill the tanks completely.
Figure 5.21 shows that about 56% of the tanks that receives water during the first seven
days are completely filled during the first 2 days. Additionally, while in the first case
the maximum number of households that filled up their tanks was slightly more than 50
in the third day, this value was now increased up to 90 in the first day. The total number
of completely filled tanks in this scenario is 1960 in seven days, which is 53% more
than the obtained value in the first scenario.

If same exercise were carried out again for closing tanks at minor levels, larger number
of completely filled tanks would be obtained in shorter times.

47
 Water reception times hystogram
(taps closed when tanks water level is 0.5m)

350 100%
90%
300
80%
Number of housholders

250 70%

200 60%
50%
150 40%
100 30%
20%
50
10%
0 0%
0

00

00

00

00

00

00

00
:0

:0

:0

:0

:0
0:

0:

5:

5:

5:

0:

5:

0:

5:

0:

5:
00

05

15

30

40
:1

:2

:2

:3

:4

:5

:5

:0

:0

:1

:1
:

:
00

00

00

00

00

00

00

00

00

00

00

00

01

01

01

01
Time (starting when system resumes)

Frequency % accumulated

Figure 5.20. Water reception times (closing inlet at 0.5m tank level)

Times when tanks are completely full

(time when tank should be closed)
Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Day 7
100 100%
90 90%
80 80%
Number of housholders

70 70%
60 60%
50 50%
40 40%
30 30%
20 20%
10 10%
0 0%
80

88

96
64

72
16

24

32

40

48

56

8
2

0
0

10

11

12

12

13

14

15

16

16

Hours (starting when system resumes)

Frequency % accumulated

Figure 5.21. Times of tanks charging, tanks at 0.5m.

48
 1 hour 6 hours

12 hours 24 hours

2 days 4 days
Figure 5.22. Filling up process of the Villavicencio network (closing inlet at 0.5m tank level)
The discussion is followed in the section 9.2

49
6 Managing water quality in WDS

6.1 Introduction
The water supply management in distribution networks in medium and small cities of
developing countries is generally carried out based on empirical practices. This leads
not only to financial problems for the supplier, but also, and more importantly, to
population health problems.

Even in industrialized countries, the management of water quality pollution in piped,

pressurized systems is not completely mastered, and good examples are the Missouri
and Milwaukee outbreaks that occurred in the early 90s in the US (Walski 2000, Hebert
2001).

The present chapter presents a new approach for managing the contamination in water
distribution systems, by using a combination between the current knowledge for
modelling and the communication technology available nowadays.

In few words, the main idea is that a contaminant hazard that is predicted with a model
can be communicated on time to the right group of people, which could be the managers
of the supply systems or the consumers under direct risk.

6.2 The origin: where is the contamination coming from?

When a water distribution network becomes contaminated is necessary to know, not
only where the contaminated water is flowing to and who is going to be affected from it.
In order to take respective corrective actions it is also important to identify where the
contamination is originating from.

Finding the source of the pollution that contaminates the treated water in a piped
network is one of the main difficulties for the supply managers, due to the complexity
of those systems. Thus, the Water Safety Plans (WHO 2004) recommend undertaking a
hazard assessment and risk characterization to identify and understand how hazards
enter into the water supply.

An approach to find the possible contamination sources in pressurized systems is

presented. First of all, it is proposed to generate a database of with records of times of
pollution arrival in certain nodes, based on different scenarios of the contaminant source
location. Such a database can be generated offline with the hydrodynamic and water
quality EPANET model. Secondly, when a contaminant is detected in a monitored node
a search is performed inside that database for the possible sources.

6.2.1 Database generation

The main idea is to run several contaminant scenarios of pollution, through a water
quality model on top of the hydrodynamic network model, in an offline fashion, in order
to generate a database of polluted nodes each hour for different location of
contamination sources.

50
The database includes 4 fields (Table 6.1):
1. The tracer value at a monitored node.
2. The node where the pollutant source is currently considered.
3. The time when the value in (1) occurs, in seconds.
4. The current monitored node.

Table 6.1. Example of records generated for the database

1 2 3 4
34,6 J17 65100 J7
30,9 J17 64800 J7
14,7 J17 64500 J7

6.2.2 Methodology
An algorithm created in Borland Delphi 7 was used to set a contaminant in each node of
the EPANET model, run it and store the results in a text file. As input, the console
application asks for the model file name, and the target tracer, parameter that will be
discussed further. The respective flowchart is presented in the Figure 6.1.

51
 Start

Ptg: Pollution j:=0

target.
Qm: Quality in a i := 0
monitored node.
Qk: Quality in a Set pollutant in node i
normal node.
T: Period of t := 0
simulation.
nNodes: Number Read Qm
of nodes.
nMonit: Number
Qm > Ptg Inc(t)
of monitored
nodes.
k := 0
Inc(j)
Read Qk
Inc(i)
Inc(k)
n
Qk = 100
y
Write Qm, Node i, t, Node j

n
k = nNodes
y
n
t=T
y
n
i = nNodes
y
n
j = nMonit

End
Figure 6.1. Flowchart for database generation
As it can be noted, each source node may pollute several nodes at different times, and
therefore the number of records may be considerable. For example, a model with 25
nodes, 15 minutes time step for a simulation period of 24 hours, a tracer target of 5%
and 3 monitored nodes to analyse, yields a database with 15 458 records.

Since the number of monitored stations should not be changed, there are three
remaining parameters that may change the size of the database, which are the number of
potential source nodes, the reporting time step of the model and the tracer value
(pollution target) for considering a node as polluted. Nevertheless, it is clear that the
ideal contamination target is zero, so the minimal presence of the tracer should be
reported, and then the database size is increased. On the other hand, the quality time
step should be consistent with the time resolution available in the quality monitoring

52
facilities installed in the network and for this reason this could be difficult to reduce.

6.2.3 Assumptions
The following assumptions were made:
ƒ Only one node at time is contaminated for each run of the model.
ƒ All nodes were considered as potential pollutant sources.
ƒ The contaminant is treated as a tracer, so there is not any distinction on the
particular chemical, biological or radioactive characteristics, and no decay rate is
used.
ƒ The contaminant is always applied at 0:00 hours into the system.

6.2.4 Database searching

The aim of having the database mentioned before is to look for the possible nodes
where the contamination is originated. The idea is that, given a positive signal of
contamination from a known monitored node, to perform a backward search through the
database to find the most probable sources. The user enters the monitored node that
detects the contaminant and the tracer value measured. As output, a list of possible
sources (the nodes with 100% of contaminant) and the times when such contamination
took place in each source is displayed.

6.2.4.1 Methodology

In this methodology the following procedure is followed:

ƒ A monitoring facility available in a given node is measuring a pollutant in every
given time step.
ƒ The pollutant is introduced in a particular node at time ti.
ƒ The pollutant is detected downstream, in the monitored node, at a time tm, with
an established tracer value.
ƒ At time tm, at least one upstream node must have 100% of tracer, and this is the
possible source of pollution.

In order to perform the searching, several methods were attempted, which includes the
conversion of the generated text file into an Access database to apply filters, the
development of a Delphi code to read the text file, and the use of algorithms
implemented in MATLAB. The first method was discarded because the Access
conversion implied huge size of the mdb files, several tens of times bigger than the
simple text file; the second method implied the invention of an optimum algorithm, but
its development cannot be included in the scope of this research.

The advantage of using MATLAB is that the variables are stored as matrices, and the
commands are designed for managing them optimally. The “find” function fulfilled the
requirement of speed of searching with an acceptable text file size. The steps are as
follows:
1. Read the text file, so each column is stored in a matrix and each column has its
own index.
2. Read the monitored node entered by the user.
3. Filter the indexes of the fourth column that correspond to such monitored node
and make a new indexation of the filtered records.
4. Read the tracer detected at the monitored node.
5. Filter the indexes of the first column from the indexes generated in point 3 that

53
 correspond to such tracer value.
6. At this stage, only the “responsible” nodes of the pollution remain in the list.

6.2.4.2 Assumptions

Some of the assumptions made for the pollutant sources searching are related with the
other fields mentioned in the WHO’s Water Safety Plans (WHO 2004), and that are out
of the scope of this research.
• It is assumed that a proper monitoring system to measure the pollution with a
given time step resolution, is available in different points in the network.
• The causes of the contamination are not studied.
• The pollutant is modelled as a tracer, therefore, neither concentrations nor decay
values are considered;
• The monitored nodes are able to measure a contaminant in terms of tracer
percentage.

6.3 The threat: Who is going to be affected?

Once the source of contamination is identified it is important to know how and where it
is going through the piped system. At this stage, a water quality model, on top of a
hydrodynamic one, becomes highly useful to predict the behaviour of the flow.

Nevertheless, it is not that easy to play with a model when the contamination is taking
place and the population is under risk, even for an expert modeller, so an apriori easy
modelling procedure might be helpful in these cases.

For this situation, it becomes useful the use of a tool like EPANET 2.0 (Rossman 2000),
which is a free solver engine for pressurized networks developed by the US’
Environmental Protection Agency. The big advantage of this software, additionally to
the null cost, is that the source code is opened for developers and that a toolkit for
programmers is also available. The toolkit provides a series of functions that allow the
customisation use of EPANET's hydraulic and water quality solution engine to external
applications. For the present research, the Toolkit version 2.00.07, released on January
2001, was utilised.

In order to find which nodes are affected by a given pollutant introduced within a pipe
network, a Delphi application, which uses the capabilities of the EPANET´s toolkit for
programmers, was developed. The main idea is that, given a contaminant value at a
specific node, the application set up the EPANET model properly, runs it and captures
the output results. The flowchart is presented in Figure 6.2.

The advantage of using this application is that the user does not have to manipulate the
model.

Additional capabilities were introduced into the flowchart shown, in order to get the
information for the hydrants and valves as well. This was done by entering additional
input parameters as text files, which contains a list of nodes that can be considered as
hydrants, and a list of pipes that can be considered as valves. This information is used
extensively for the operational advices module, which is part of the scope of this
research and that is discussed in the following sections.

54
6.3.1 Assumptions
A pre-built, reliable EPANET model is required, since the application asks for the INP
file that contains it. This model should be calibrated in order to expect reliable results.

The application considers the pollutant as a tracer, and therefore the node where it is
introduced is treated as a constant source of a non-reacting constituent with a
concentration of 100.

Start

Read source node: S

Read quality value of source: Qs

Read quality target: qt

Open EPANET model

T: Period of
nNodes := Get Number of Nodes simulation.

Set Qs in node s

i := 0

t := 0

Run hydraulic model

Run quality model

inc(t) inc(i)
Qi := Get quality at node i
y
n
Qi > qt

Print Qi, t
y
t=T n
y
n
i = nNodes

End
Figure 6.2 Flowchart to find affected nodes by a given pollution source.

55
6.4 The actions: what to do?
Two types of operational actions can be distinguished: preventive and corrective. The
first are focused on attacking the causes of potential problems before they appear, and,
clearly this is recommendable, even if it is often a very complex task. Nevertheless,
several recommendations for preventive actions exist under the name of preventive
maintenance, which include actions on storages, pipes and fittings (Ainsworth 2004).
The corrective actions are focused on solving the problem that already exists, through
recommended actions stated in plans that previously have been developed.

In this sense, it is clear that the proposed research is helping to the development and
well functioning of the corrective actions. When a quality problem is detected it is
necessary to start an action plan to mitigate it. This plan must contain, among other
activities, the operational actions that need to be executed directly on the system, to
reduce the damaging concentrations of the contaminant.

The actions to execute when a water quality emergency occurs may be similar to the
actions of preventive maintenance in tanks, reservoirs, fittings and pipelines, which
include cleaning, flushing and valve manoeuvres, depending on each specific problem.

6.4.1 The operation response under emergency: a difficult issue

When an event of contamination takes place in a distribution network, the reaction must
be quick (to avoid contamination propagation) and accurate (to avoid mistakes).
However under emergency conditions, the process to make decisions is very difficult
and stressing. From one side, the decision on which operation actions to take is not easy,
even if a hydraulic and quality model of the system are available; from another side,
such decision must be taken in a restricted period of time, with the associated concern to
manage the situation under pressure.

A common practice to solve the problem is to try to isolate the contaminant by

operating valves, to reduce the affected area of the system by the pollutant. After that,
open accessories like hydrants, vents or drains are used to flush out the contaminated
water. The selection of the appropriate set of elements to be operated is a main issue,
because the behaviour of the hydraulics and the water quality in looped systems is quite
complex (Hart et al 1992), and because a wrong selection could make the situation
worse as happened in one of the Missouri outbreaks (Walski 2001).

On the other hand, the restricted time for reaction, and the stress associated with the
public health concern, makes inappropriate to play with a model, at that particular
moment in order to find out a solution. It has to be pointed out that, even with enough
time for analysis and with a reliable model, an appropriate solution may not come up
easily, simply for the fact that the number of elements for changing might be up to
several tens.

In the present chapter this problem is addressed by utilizing the currently available
optimisation methods together with the hydraulic and quality solvers for pressurized
networks.

6.4.2 Formulation of the operational response as an optimisation problem

The problem can be stated as follows: minimize the number of affected nodes by the

56
pollution using as few operational changes as possible. Any modification made in
valves, hydrants or pumps with respect to the original network state is understood as
operational change. In order to solve this problem the EPANET model is coupled with
an optimisation procedure. The following assumptions are taken:

1. The source of the contamination is known at a particular model node.

2. The pollutant is treated as a tracer, which can be adopted as a conservative one,
therefore pure transport through the pipes is considered.
3. The initial concentration (tracer) of the pollutant is 100%.
4. A node with a tracer percentage above 10% is considered affected by the
pollution.
5. The simulation time is 24 hours, that is, the solution found must solve the
problem in less than one day.
6. A hydraulic and water quality model built in EPANET is required, as well as a
list of the junctions that can be used as hydrants, a list of pipes that can be closed
like valves and a list of pumps that could be switched. The initial status of these
elements is also needed.
7. The opening of a hydrant is simulated as a demand of 6 l/s, which is the fire
flow by design. This discharge has a constant pattern.

6.4.3 Tools and methodology followed:

In order to face the problem, the following tools were used:
• EPANET 2.0 programmer toolkit.
• Borland Delphi 7.0
• GLOBE - Global and evolutionary optimisation tool.
• NSGA – Pareto Front Builder for multi-objective optimisation.

The adopted methodology is described as follows:

• Two optimisation approaches were studied: single objective and multi-objective.
• For both approaches, the Genetic Algorithm method is used. Nevertheless, for the
first approach the GLOBE program (Solomatine 1999) is used, meanwhile the
NSGA program is used for the second approach.
• Both programs use the developed application COPA (Changing Operation in
Pollutant Affection), which is explained later.
• A small network was used to prove the validity of the optimisation in both cases.
• The analysis and comments were stated.

6.4.4 The Genetic Algorithms

For the optimisation process the Genetic Algorithms (GA) approach is used. It is a
directed search algorithm based on the mechanics of biological evolution, developed by
John Holland during the 1970’s. It follows the Darwin’s theory of evolution, which uses
natural selection as driving force. This concept is applied to the distribution networks by
considering one network as a chromosome, a group of networks as a population, and a
set of different populations as a generation (Figure 6.3). By performing the GA
operations (selection, crossover, mutation), the best chromosomes among a population
of a particular generation (according to a certain evaluation parameter previously
defined), are selected to engender a new population for the next generation. The process
is repeated until a specific number of generations are evaluated, yielding as result the
best chromosome (network configuration) that hopefully represents the optimal solution.

57
 0 0 1 0
1 1

1 1 1 0

1 0
0 0 1 0
Chromosome

1 1 0 1
0 1

0 1 0 1

0 0
1 1 1 0
0: Closed
1: Opened
Population

Generation
Figure 6.3 Population members for an optimisation problem in WDS using GA

6.4.5 COPA module (Changing Operation on Pollutant Affection)

6.4.5.1 Description

This application performs the coupling of the EPANET model with optimisation
procedures. The optimisation procedures require many model executions. This is
achieved by using the automation options of the EPANET programmer toolkit. It was
developed as a console application in Borland Delphi 7 and it has the following
flowchart (Figure 6.4):

58
 Start Run model (t) Evaluate
Penalty
Read Files Read Qi
n Print
Change model Qi > Qg Num changes
with new status y
Inc (AffNodes) Print
Calculate number MaxAffNodes
n
of changes i = NN
y Print
keep MaxAffNodes Penalty

Qi: quality value in node i Inc (t) End

Qg: quality target
NN: Number of nodes n
t>T
t: Current time
T: Total time y

Figure 6.4 COPA module flowchart

The COPA module needs different text files to work properly. A description of each one
is explained as follows:

File: Parameters
Type: Text
Contents: Water quality parameters for simulation
Line 1: Node source of pollution
Line 2: Initial quality value in the source (tracer or chemical)
Line 3: Quality value target.
Line 4: Type of analysis: C = Chemical, T = Tracer
Line 5: Time of simulation in hours.
Samples: J9 J9
1 100
0.2 10
C T
24 48

Notes: 1. The node source of pollution must have the same

name in the EPANET model.
2. The quality values have either units of concentration
(mg/L) if the type of analysis is chemical or units of
tracer (%) if the type of analysis is tracer.

File: Elements

59
Type: Text
Contents: List of junctions, pipes and pumps whose status can change.
Sample: P2
P6
P10
P14
PMP18
J17

Notes: 1. The valves are represented as pipes that can be

switched on or off, and must start with the letter ‘P’.
2. The pump elements must start with the sequence
‘PMP’.
3. The hydrants are represented as junctions which
demand can change to 6 l/s. They must start with the
letter ‘J’.
4. The names must be the same as are used in the
EPANET model with the stated specifications.

File: Initial status

Type: Text
Contents: List of zeros and ones that represent the initial status of the
elements in the network, correspondingly with the elements
file.
Sample: 1
1
1
1
0
0

Notes: 1. For pipes (valves), ‘0’ means closed and ‘1’ means
opened.
2. For pumps, ‘0’ means ON and ‘1’ means OFF.
3. For junctions (hydrants), ‘0’ means normal demand
(i.e. daily average in l/s with an associated
consumption), and ‘1’ means fire demand of 6 l/s, also
associated with the pattern defined in the EPANET
model.
4. The numbers can be also real numbers between 0 and
1, which internally are rounded to the nearest integer,
either 0 or 1.
5. The number of records must be the same as the
number of the elements’ file.

File: G.pin
Type: Text

60
 Contents: List of zeros and ones that represent the status of the elements
in the network, which are going to be tested. It must be
correspondent with the elements file.
Sample: 0
0
0
0
1
1

Notes: 6. For pipes (valves), ‘0’ means closed and ‘1’ means
opened.
7. For pumps, ‘0’ means ON and ‘1’ means OFF.
8. For junctions (hydrants), ‘0’ means normal demand
(i.e. daily average in l/s with an associated
consumption), and ‘1’ means fire demand of 6 l/s, also
associated with the pattern defined in the EPANET
model.
9. The numbers can be also real numbers between 0 and
1, which internally are rounded to the nearest integer,
either 0 or 1.
10. The number of records must be the same as the
number of the elements’ file.

Additional note: The EPANET model must have defined two patterns of consumption
called ‘1’ and ‘2’. The first one is used as the pattern for the hydrant flow, while the
second one is used as the pattern for the normal domestic consumption.

After the algorithm showed in COPA is completed, the file G.RSP is created; it contains
the value of the objective function for the network status defined in G.PIN

6.4.5.2 The Objective Function

6.4.5.2.1 Single objective

The objective function used for the optimisation algorithms was initially defined as:

f (c1 , c2 ) = c1 + c2
2 2

where c1 is associated to the fact of having contaminated nodes and the c2 is associated
to the operational effort required to set the network to a desirable condition. As first
approximation, the simple value of the number of contaminated nodes for c1 and the
number of changes in the network with respect to the initial status for c2 are used.

This approach for the objective function was selected because it represents a distance
between a desirable condition (having no contaminated nodes with no network changes,
which means function value of zero), and the situation of having some number of
contaminated nodes and network changes. Naturally, the objective is to minimize such
distance, thus, a single optimisation objective is considered.

61
 c2
Affected nodes
(cost 2)

f (c1 , c2 ) = c1 + c2
2 2

c1
Changes in the network
(cost 1)
Figure 6.5. Objective function concept for a particular solution (network state)
It can be noted that the objective function is chastising both the affected nodes and the
changes in the network with the same weight. On the other hand, no restrictions on their
values have been added.

6.4.5.2.2 Multi-objective

The multi-objective approach tries to minimize several objectives independently. In the

stated problem, the costs c1 and c2 mentioned above are considered as separated
objectives to be minimized.

6.4.6 The optimisation process

6.4.6.1 Single objective

In order to perform the optimisation process, the GLOBE software (Solomatine 1999)
was used firstly. In GLOBE several different global optimisation algorithms are
implemented and the user can choose any of them. For the present research, the Genetic
Algorithms approach is used, method that was briefly described in section 6.4.4.

The role of GLOBE is to generate an input vector (network status, in this case),
supplying it to COPA module via the file G.PIN, which contains real numbers. For the
problem under consideration, each element status is considered as one variable, which
means that there are as many variables as elements (valves, pumps or hydrants) to
change. The Figure 6.6 shows the conceptual model of COPA and its interaction with
GLOBE.

It must be noted that the real numbers that GLOBE creates in the G.PIN file, have a
particular characteristic: they can only be either zeros or ones. To deal with this problem,
two different steps were made: firstly, the parameter “discretization”, in the GLOBE’s
options tab (Figure 6.7), is set as 1; secondly, the boundaries of each variable were set
between 0 and 1, (Figure 6.8).

62
 Figure 6.6 COPA module and its relationship with GLOBE program

Figure 6.7. Discretization option in GLOBE

Figure 6.8 Boundaries for the variables

In order to test the method, a small test example was developed (Figure 6.9). It consists
of a system of 45 pipes with same diameter, length and Manning roughness (0.20m,
1000m and 0.01), and 25 junctions with zero elevation and 0.5 l/s of constant demand.
The reservoir provides a constant head of 50m to the system. The valves, originally

63
opened, are V2, V3, V4, V5, and V7; the hydrants, originally closed, are H1 and H2,
and the source node of contamination is shown.

V7

Trace J9
0.00 V5
0.00 Contam
0.00
50.00
percent
V4 H1

Trace J9
0.00
0.00
H2 V3
0.00
50.00
percent
V2

Figure 6.9. Small example. Pollution after 48h simulation.

The problem consists on looking for the best configuration of the network that avoids
the pollutant propagation (tracer up to 50% in a node is not allowed) with the least
number of network changes in 48h simulation. In this case there are 7 variables, that can
be either 0 or 1, and the network has an initial status: 1111100. If the Genetic
Algorithms’ optimisation algorithm is used, considering an initial population of 20 and
4 generations, the result is obtained in less than 20 seconds and utilizing only 52
evaluations. The value of the penalty function is 3.60, with 3 affected nodes and 2
changes. The binary numbers are the status of the elements V7, V5, V4, V3, V2, H2, H1,
correspondingly. (Figure 6.10)

64
 The small example: optimisation results

12
11
Number of affected junctions

10
9
8
7
6
5
4
3
2 0111110
1
0
0 1 2 3 4 5 6 7 8
Number of changes in the network
Figure 6.10. Best solution for the small example

The solution, in terms of operational actions, is (Figure 6.11):

• Change V7 to closed.
• Change H2 to opened.

In spite of this surprisingly simple solution, an intuitive evaluation by inspection

without model perhaps would never brought it up, since the concept of network
isolation is not intuitive. Certainly, V7 was never contaminated during the initial
conditions of the network, so possibly it would never be selected for changing; when V7
is closed, it pushes the clean water to flow against the contamination, and forces to flush
it out through H2. On the other hand, the rest of the valves, which are polluted if no
actions are taken, were not changed at all in the optimal solution. This result is quite
unimaginable.

An intuitive solution would be to close V2, V3, V4, V5 –in order to isolate the
contaminated area, and to open H2 to flush the contaminant out; nevertheless, this
solution requires 5 changes and the total affected nodes is eight, yielding a function
value of 9.43, which, evidently, is far away of being an optimal solution. This proves
the difficulty of choosing the best configuration even in a simple network, with enough
time to analyse it.

65
 closed
V7

Contam V5

Trace J9
0.00
0.00
0.00 V4 H1
50.00
percent

Trace J9
0.00 opened
0.00 H2 V3
0.00
50.00
percent

V2

Figure 6.11. Solution found for the small example with GA, after 52 evaluations

6.4.6.2 Multi-objective

As mentioned before, the multi-objective problem was defined as the minimization of

the number of affected nodes and the number of changes in the network as separate
objectives. The optimisation process was performed with NSGAX (Barreto et al, 2006)
software, still under development, that uses a GA code developed by Kalyanmoi Deb.
The NSGAX provides the best options for each variable independently, which are
minimised at the same time.

The NSGAX was applied for the small problem. After several runs, it was clear that
more generations than for the single objective are required to have the previous result.
In Figure 6.12 the solution set of the Pareto front is presented, where the binary
numbers are the status of the elements V7, V5, V4, V3, V2, H2, H1, correspondingly.

66
 The small example: optimisation results

12
11 1111100

Number of affected junctions

10
9
8
7
6
5 1111110
4
3 0111110
2
1
0
0 1 2 3
Number of changes in the network

Figure 6.12. Pareto front for the multi-objective optimisation of the small example

Due to the discrete nature of the problem, the resulting Pareto front is in reality a set of
few points; the first one, with no network changes, which may be a solution in certain
cases, and eleven affected nodes (situation shown in Figure 6.9); the second solution,
with 5 affected nodes and one change (open H2) appears like a possible decision action.
The third solution is the same as obtained for the single objective optimisation problem.

6.4.7 Case study of Villavicencio

A case study in Villavicencio is considered. The aim is to apply the optimisation
methodologies shown before, using the COPA module. A source of pollution is
supposed to be identified in a node of the hydraulic sector number 11 of the
Villavicencio’s supply network (Figure 6.13), and the best options for flushing has to be
found.

The sector 11 has been modelled by means of 341 junctions and 487 pipes, with eight
hydrants and 89 are valves. This means that 97 elements are possible to be changed in
the optimisation processes, and therefore this is a problem with 97 variables. For this
special case a modified version of GLOBE was used, which allows up to 100 variables.

For the demonstration, the following parameters are used:

• The source of the pollution is in the node J2119, with a tracer value of 100%.
• The tracer target is 5%, which means, nodes with higher values are not allowed.
• The simulation time is 24 hours.

67
 J2119

Hydrant Valve Pipe diam (mm)

Figure 6.13. Villavicencio - Sector 11.

6.4.7.1 Assumptions for the model

ƒ Normal consumption per capita assumed (180 L/day), and normal

diurnal demand curve.
ƒ The recommended network improvements by Aquadatos (2002), which
include the implementation of the network sectorization and the
installation of micrometers to control the household demands, were
totally taken into account.
ƒ The model is calibrated properly.

If no changes in the network were made, after 24 hours 29 junctions would be impacted,
as shown in Figure 6.14.

68
 Trace J2119
0.00
0.00
0.00
5.00
percent

Trace J2119
0.00
0.00
0.00
5.00
percent

Figure 6.14 Affected area of sector 11, Villavicencio after 24 hours of contamination from J2119

6.4.7.2 Single objective approach

Due to the huge amount of possible solutions, and in order to accelerate the optimisation
process, a restriction was included, which consist on neglecting the solutions that
require more than 30 operational movements. The results are shown in Figure 6.15.
Sector 11: Optimization results

60

50
Number of affected junctions

40

30

20

10

0
0 5 10 15 20 25 30 35
Number of changes in the network

Figure 6.15 Optimisation results for sector 11, Villavicencio

The point with where the objective function is minimal is marked by a square. It
requires 15 movements and affects 18 junctions, which yields a function value of 22.67.

69
6.4.7.3 Multi-objective approach

Pareto front for optimisation problem

Sector 11 Villavicencio

45
40
35
Affected junctions

30
25
20
15
10
5
0
0 5 10 15 20 25 30 35 40 45 50
Number of movements

Figure 6.16. Pareto front generated by NSGAX for sector 11, Villavicencio.
The Pareto front can be used for the decision makers to choose between a range of
possibilities. The big point at the left of the Figure 6.16 shows the situation for which no
changes are made. In this case, in one hand it can be seen that after 35 movements is
possible to have the lower amount of contaminated nodes (only 4); on the other hand,
fewer movements, generally, will affect more people.

In this particular case is important to note that it may not be possible to manage so high
number of operational changes, due to the limited resources for it, but anyway, the
overall picture of the situation results clear for the decision maker.

Further discussion is followed in the section 9.3.2.

70
7 The SMS as a tool for operative and emergency support

7.1 Introduction
One of the steps of the Water Safety Plans recommended by WHO (2005) are the
communication strategies, which should include:
ƒ Procedures for promptly advising of any significant incidents within the
drinking-water supply, including notification of the public health authority;
ƒ Summary information to be made available to consumers – for example, through
annual reports and on the Internet; and
ƒ Establishment of mechanisms to receive and actively address community
complaints in a timely fashion.

7.1.1 Communication process

According to Brosnan (1999), the factors that must be considered to establish a
communication trigger are the nature and magnitude of the adverse event, the type of
actions that will be taken as a response, and an estimate of the possible effects of the
contaminant. The analysis of these factors may result in communicating advice like
“boil water”.

For the proposed methodology, once the water quality model results are evaluated and it
is established that a message is required, the automatic application will prepare it
according to the type of problem. The idea is to send such messages only to the people
affected by the water quality problem.

For this purpose, a database with the mobile phone users will be connected in a
geographic basis with the water distribution model. In this manner, only the population
included in the areas highlighted by the model as risky areas, will be warned.
Nevertheless, in extreme events, in which the population health is in high risk, other
ways of communications can be used for complementing the function of the mobiles,
but care must be taken to avoid false alarms to the population that is not directly
affected.

The points to analyse in this stage are:

ƒ When is absolutely necessary to send a warning?

ƒ Who will be warned?
ƒ When they will be warned?
ƒ For how long?

A plain definition of the concentration baseline for each contaminant must be set, as
well as its deviation, from which a response is triggered. This trigger should be
evaluated according to the difference between the baseline and the deviation, and its
strictness depends on the consequences of the adverse event in the community and also
on the cost of the response activities (Brosnan 1999).

71
 In case of detecting those deviations the warning messages must be prepared, following
predefined communication criteria. An example by Godfrey and Howard (2004a)
establishes the communication strategies that should be included within a Water Safety
Plan, which are summarized in Table 7.1.
Table 7.1. Communication strategies
Severity of
Operations Senior Regulator/
hazard Risk manager Consumers
manager management health body
event
Informed Informed
Must be Must be Advised immediately on
Catastrophic immediately immediately
informed informed necessary action approved
by risk by risk
immediately immediately by risk manager
manager manager
Must only be
Must be Informed only Informed only Advised immediately on
informed if
Minor informed if problem if problem necessary action approved
problem
immediately persists persists by risk manager
persists
Source: Godfrey and Howard (2004a)

This study considers the methodology of sending the short messages through mobile
phones. Although the preparation and the final location to send the messages will be
performed by the QmagNet-V, the final decision to spread them will be only executed
by humans, as well as the use of other ways of communication, depending on each
particular case.

7.1.2 GSM and SMS

GSM is the abbreviation for Global System for mobile communication, where the
mobile concept refers to the fact that the communication between two terminals can be
performed even though these are in movement. The GSM is a digital cellular
communication network, which provides, among others, the following services:

It manages several income calls simultaneously

ƒ Waiting calls
ƒ Conference calls
ƒ Calls identification
ƒ SMS
ƒ Send and receive electronic mail

The GSM network can be divided into four main parts (Figure 7.1):

ƒ The Mobile Station (MS), which is carried by the subscriber

ƒ The Base Station Subsystem (BSS), which controls the radio link with the MS
by means of the Base Transceiver Stations (BTS). It includes also the Base
Station Controller (BSC), which is responsible for RR (Radio Resource)
allocation to the MS, the frequency administration and handover between BTSs.
ƒ The Network and Switching Subsystem (NSS), the main part of which is the
Mobile services Switching Center (MSC) and which performs the switching of
calls between the mobile and other fixed or mobile network users, as well as
management of mobile services, such as authentication The Real time Transport
Control Protocol (RTCP) enables the receiver to detect if there is any packet loss
and to compensate for any delay jitter.
ƒ The Operation and Support Subsystem, which oversees the proper operation and

72
 setup of the network.

Other elements within the GSM systems are:

ƒ The Home Location Register (HLR), a database within the HPLMN (Home
Public Land Mobile Network). It provides routing information for MT calls
and SMS and is also responsible for the maintenance of user subscription
information.
ƒ The VLR (Visitor Location Register) or SGSN (Serving GPRS Support
Node), which receives the subscription information.
ƒ The AuC, which is associated with the HLR (Home Location Register) in a
GSM network.

Figure 7.1. GSM architecture

7.1.3 What is SMS?

The Short Message Service (SMS) is the ability to send and receive text messages to
and from mobile telephones, typically 160 characters in length, and was first introduced
into the GSM services in 1995. The text can comprise of words or numbers or an
alphanumeric combination.

73
Some of the benefits of the SMS are:
ƒ Delivery of notifications and alerts
ƒ Guaranteed message delivery
ƒ Reliable, low-cost communication mechanism for concise information
ƒ Ability to screen messages and return calls in a selective way
ƒ Increased subscriber productivity

The integration with other data and Internet-based applications and the delivery of
messages to multiple subscribers at a time are sophisticated SMS functionalities that are
exploited by the present research.

SMS can do all the things that alphanumeric pagers do, and unlike many existing paging
services, the network can acknowledge receipt of messages, so there is no question
about whether the recipient received the message. Messages are sent via a message
centre owned by the network provider. If the message is undeliverable, either because
the recipient is out of coverage or has switched off the phone, the network can store the
message until it becomes possible to transmit. Messages can be sent from a mobile
phone via the keypad or the attached computer. It is also possible to send the message to
the message centre via the Public Switched Telephone Network (PSTN), the Integrated
Services Digital Network (ISDN), Packet-Switched Network (PSN), or the Internet.

Figure 7.2 SMS Network Architecture

7.1.4 SMS for real time warnings and emergency management

Recent news stories have highlighted the use of SMS in international relief efforts and
emergency communications, since it has proven to be extremely valuable in conveying
concise information quickly, and reliably. Lately, the SMS have been starting to be used
as a reliable and fast communication way in emergencies and disasters. Recent
researches have taken this service into account for flood management in urban drainage
systems (Nayeema 2006), and flood plains (Sisay 2003).

A big example was shown by the Pakistani government, which recently granted free
mobile phone use during the aftermath of the October 8th 2005 earthquake in Kashmir -
an area where previously mobile phone service was scant at best.

Additionally, Senate Subcommittee on Disaster Prevention and Prediction, US, has been
discussing during 2005 about expanding the Emergency Alert System beyond
televisions and radios so that they can use SMS to automatically notify cell phone users
about disasters or terror attacks. The Federal Emergency Management Agency is about

74
to start testing a new Geo-Targeted Alerting System that would provide warnings only
to potentially affected households and businesses (rather than just blasting warnings out
to everyone). (McCullough 2005).

7.2 The SMS for managing the water quality in distribution systems
Within the previous chapters different approaches for modelling different situations in
supply systems have been presented. The condition of the intermittent supply and the
way to switch it to the continuous state, as well as a new way to manage the
contamination detected in pressurized systems were the discussed topics.

These approaches are complemented with the introduction of the SMS to be used as an
efficient way of communication of the information generated by such approaches.

7.2.1 The mobile phone for the intermittent supply operation

The approach proposed in the chapter 5 requires the communication of the following
aspects to the customers:
ƒ The water reception times for each householder after the system resumes.
ƒ The times when the tanks are completely filled.

The first point allows the users to plan their water use, which leads to the reduction of
wasted water, and therefore, more water is available for filling up the network.
Consequently, the network is pressurized and the supply becomes continuous.

The second point allows the managers to know the times when each tank is filled, and
therefore, to control the water consumption, by asking the users to close their taps, to
avoid overflowing.

The messages can be either for fully operational purposes or for generating awareness
among the customers from the point of view of the consumption culture.

Examples for fully operative purposes:

ƒ “Your house will receive water at 2:00pm; wait at least 1 hour to use it for
drinking”
ƒ “Boil the water you are receiving now before use it for drinking”
ƒ “The water you are receiving is safe. Do not fill your tank”
ƒ “Please close your tank at 4:00pm and take water directly from the network”

Examples for generating awareness:

ƒ “Check if your tank is being overflowed. If so, please close it”
ƒ “With your help we can get continuous service. Do not waste water!”
ƒ “You do not need to use your domestic tank anymore. Take water from the
network. Is safe!

The messages for generating awareness should be sent in a periodic basis, but they
should not be so frequent that result hackneyed for people. This is important because
those messages can affect the attention needed by the costumers to the operative
messages.

For especial Institutions, Public Authorities, etc, other messages that should be

75
considered in certain cases are listed below:
ƒ “Possible outbreak for contaminated water in sector 3. Alert level: 3”
ƒ “Contaminant: ‘_____’ detected in zone 3. Start action plan”
ƒ “Contaminated water: hospitals, schools and big centres must be informed
immediately”

7.2.2 The mobile phone for the water quality management in WDS
The approach discussed along the chapter 6 states that the text messages have two main
receptors:

ƒ The operators
ƒ The affected people

The operators will receive messages to support the operation in a pollution event, to
know the contamination sources, the affected area of the city and the recommended
actions to mitigate the problem. These messages should be centralized, in such a way
that the last decision of something related with operation, or even related to warn people
in last instance, be led by the manager.

The affected people will receive basically warning messages, indicating what kind of
action must be taken, and, if applicable, for how long. The customers may also receive
informative messages about the possibilities of provisional supply or alternative water
sources.

Examples of messages for the operators are:

ƒ “The following valves must be closed before 1:30pm: V1, V2…, keep them
closed until 5:00pm”
ƒ “The following hydrants must be opened between 1:30pm and 4:00pm: H1,
H2…”
ƒ “About 500 people affected in zone 3 during 4 hours. 2 truck-tanks or 1500
bottles of litter might be needed”

Examples of messages for the affected people would include:

ƒ “Emergency: water could be polluted. Do not use it for drinking or cooking.

Pass this message to your neighbours.”
ƒ “Bottled water will be available in the coming hours”
ƒ “Water is now safe for drinking, but it is advisable to boil it before use in the
coming 2 days”

7.3 The Villavicencio case

7.3.1 The mobile phones in Villavicencio and Colombia
Only until 1993 Colombia created the law for the regulation of the mobile phones, the
142/94, and in 1994 was officially opened the competition between different candidates
to be the mobile service providers for the country. The law divided the country in three
zones, Atlantic, East (where Villavicencio is located) and West, as showed in Figure 7.3.
It was established also that each of these zones had two different companies, one private

76
and one mixed. Initially, the three private companies were Celumovil (later on under the
name of Bellsouth), Comcel and Caribe Celular, and nowadays, after important political
and financial movements, the companies that provide the service are Movistar, Comcel
and Colombia Movil Ola.

Atlantic zone

West zone
East zone

Figure 7.3 Division for the mobile phone service in Colombia

Evolution of communication services in Colombia

20

15

10

0
1998 1999 2000 2001 2002 2003

Fixed/100 hab Mobile/100 hab Internet/100 hab

Figure 7.4 Evolution of the communication services in Colombia

Source: Mincomunicaciones (2004)
Although the penetration of the communication services in Colombia are still below the
world average and even below some Latin-American countries, the evolution is tending
to increase within the years, especially for the mobile phone service, as can be noted in
Figure 7.4. More recent information states that the mobile phone users in Colombia
increased a 63% between 2003 and 2004, and during 2005 it increased in a record
percentage of 109%, which means that 21.9 million are currently using the mobile
phone service (SIC 2006).

The situation for the poor is relatively good, since the people with low economic status
(1, 2 and 3), are using the mobile service in a considerable bigger proportion than the
people of high economic status (4, 5 and 6), Figure 7.5. This is explained for the

77
convenience of the prepaid service conditions, which do not require legal contracts and
therefore no financial obligations are taken.

Status 1, 2, 3 (low) Status 4, 5, 6 (high)

Figure 7.5. Participation of users per economical status

Source: Mincomunicaciones (2004)

7.3.2 How the approach should work

The Comcel Company is nowadays the one that have 100% of coverage in
Villavicencio, so the messages could be sent properly. The Figure 7.6 shows how the
approach should work for the intermittent case.
“Your house will receive
“The water you are water at 9:00am; wait at
receiving is safe. Do not least 1 hour to use it for
drinking”.
fill your tank with it”

“With your help we can

get continuous service. Do
not waste water!”

“Please close your tank at

4:00pm and take water
directly from the
network”.
“Check if your tank is
being overflowed. If so,
please close it”.

Figure 7.6. General visualisation of the messages for managing the intermittent supply in
Villavicencio
For the pressurized case, situation could be visualized as shown in Figure 7.7

78
 “Emergency: water could
be polluted. Do not use it
for drinking or cooking.
Pass this message to your “Bottled water will be
neighbours.” available in the coming
hours”

“Water is now safe for

drinking, but it is
advisable to boil it before
use in the coming 2 days”

“About 500 people “The following hydrants

affected in zone 3 during 4 must be opened between
hours. 2 truck-tanks or 1:30pm and 4:00pm: H1,
1500 litter-bottles might H2…”
be needed”
“The following valves
must be closed before
1:30pm: V1, V2…, keep
them closed until 5:00pm”

Operator
Figure 7.7 General visualization of the messages for managing the pressurized supply
Further discussion is followed in section 9.4.

79
8 The QMagNet-V application

Up to now, several individual applications for pressurized systems, such as the water
quality database generation, the pollution source searching and the operational advices
using Genetic Algorithms have been developed according to what has been discussed in
the previous chapters. Additional applications include the reading of SWMM report for
intermittent systems and the management of the short messages services.

The aim of this chapter is to show how all of these applications were combined into a
general application, which is called Water Quality Management in Networks
(QMagNet-V). The application was coded in Delphi 7.0 with the structure shown in
Figure 8.1.

QMagNet-V

Pressurized systems Intermittent systems

Offline analysis SWMM report reading

Quality Database Generator

Short Message Service
Online analysis Preparation

Pollution source locator

Pollution impact estimator

Operational advisor

Short Message Service

Additional tool
Sending

DBP prediction

Figure 8.1. QMagNet-V structure

Firstly, the user chooses either the pressurised system or the intermittent system to be
analysed. For the first choice, there are two possibilities: offline or online analysis. The
possibility to generate the quality database is considered to be performed offline due to
the long time required for it. Once the database is generated, there is no need to generate
it again if the operational conditions are not changed, and the online analysis can be
executed. Alternatively, within the online procedures is possible to perform the where –
who – what analysis when a pollutant is detected. Each one of these steps (locator,
estimator, advisor, respectively), have the option to prepare a message, either for the
supply managers or, if required, for the impacted population.

80
On the other hand, the choice of intermittent water supply requires that the SWMM
model is already ran, and that the report file is available. This option will return a list
with the water reception times to each node and also a list with the times at which the
household tanks should be completely filled.

An additional tool was also added, which is the Disinfection-by Products prediction in
the network, according to the research made by Sohn et al (2004). Nevertheless, it was
not connected to the message service due to time restrictions.

8.1 Description of QMagNet-V

The QMagNet-V utilizes the following separated units, which are described in the
following table.

Table 8.1 Description of the Pascal units used by QMagNet-V

Name Description
Epanet2 Contains the functions of the EPANET’s Toolkit for
programmers
GenDatabases Contains the algorithm to generate the databases as
described in section 6.2.1
General Contains the design of the main window. Requires
sendingSMS.exe, which is a VB application that uses
SMS_COMAPI 1.0 Type Library.
LookingUpSources Contains the algorithm described in section 6.2.4, to look
for the possible contamination sources. Requires the
MATLAB libraries ComObj and ComCtrls and the file
TestSearchMatlab.asv.
Quality Contains the implementation of the models for DBP
prediction
RunEpanet Contains the code to perform the calculations and to display
the hydraulic results and the DBP predictions using the unit
Quality.
RunGlobe Contains the algorithms to perform the optimization
process using the COPA module and GLOBE (installation
required), as mentioned in section 6.4.6
SMSpreparation Contains the algorithms to prepare and store the messages
generated by each module, as mentioned.
SolEXEC2 Contains functions for running windows-based programs
from Delphi applications; developed by D. Solomatine
TheMaps Contains code for displaying, read only, the Villavicencio
map and the EPANET and SWMM models. Requires the
ActiveX object Whip!, by Autodesk. Edition is not
possible.
Unit1ReadSWMM Contains the algorithm for searching through the report file
generated by SWMM, the water reception times for each
household.
WhosAffected Contains the algorithms to find the affected nodes running
EPANET, as described in section 6.3

81
8.2 Required software installations
QmagNet-V needs to have installed the following software to run successfully:
ƒ GLOBE 2.0, software to solve optimization problems.
ƒ Whip! 4.0, ActiveX component by Autodesk, for visualizing CAD drawings.
ƒ MATLAB 7.0, software for advanced engineering programming.
ƒ Clickatell COM-API, COM object to prepare SMS messaging.

8.3 User Interface

The following figures illustrate the interface of the QMagNet-V application.

a f

g
b

e
h i

Figure 8.2. QMagNet-V main window.

82
 a. Generate Quality Database
The dialog box is shown in Figure 8.3. Details of the algorithm used can be found in
section 6.2.1

Figure 8.3. Dialog box for the Generation Databases module

83
 b. Where is the source?
Details of the algorithm used can be found in section 6.2.4

Figure 8.4 Dialog box for the possible source searching

84
 c. Who is affected and when?

The dialog box is shown in Figure 8.5. Details of the algorithm used can be found in
section 6.3

Figure 8.5 Dialog box of the estimator of the affected elements.

85
 d. What to do?

Only the single objective approach has been implemented due to time restrictions.
Details of the algorithm used can be found in section 6.4, and the dialog box is shown in
Figure 8.6.

Figure 8.6. Dialog box for the operational actions

86
 e. Disinfection-by Product prediction
As an additional part of this research, the model developed by Gary Amy and
documented by Sohn et al (2004) has been implemented within QMagNet-V application.
The model consists on empirical equations for predicting the Disinfection-by Products
(DBP) when the water in the treatment plant is coagulated with alum or iron, which
generates the so-called Trihalomethanes (THM) and the Haloacetic acids (HAA).
According to USEPA people who drink water containing THM and HAA in excess over
many years increase the risk of getting cancer and cardiovascular and cerebral-vascular
diseases.

These parameters are function of the dissolved organic carbon (DOC), the chlorine, the
Bromide, the water temperature, the pH and the age of the water since it is treated. The
models are the following, for pH of 7.5 and temperature of 20ºC:

THM = 3.296(DOC ) (Cl 2 )0.261 (Br − )0.223 (time )0.264

0.801

HAA = 5.220(DOC ) (Cl 2 ) (Br − ) (time )

0.585 0.565 0.031 0.153

These models have to be corrected with for different pH and water temperature with the
following equations:

( )
THM = THM @ pH =7.5,Temp = 20o C *1.156 ( pH −7.5 ) *1.0263(Temp − 20 )
HAA = (HAA @ pH = 7.5,Temp = 20o C
)* 0.932( pH − 7.5 )
*1.021(Temp − 20 )

The Figure 8.7 shows the developed dialog box for this module.

87
 Figure 8.7 Disinfection-by product prediction dialog box

1. Text box to input the EPANET model file to be analysed.

2. Performs the hydraulic and quality run of the EPANET model.
3. Displays the hydraulic results (pressure in H20) and the quality results (Water
age in hours).
4. Section to enter the quality parameters used for treatment in the treatment plant,
and to perform the DBP prediction in the network.
5. Displays in a tree all the information for the nodes at different times, including
the pressure, age, residual chlorine, trihalomethanes and haloacetic acids.

88
 f. Read SWMM report for Intermittent Supply
The Figure 8.8 shows the developed dialog box for this module. Detailed description of
the algorithms used and the concepts are shown in section 5.

Figure 8.8. Intermittent supply analysis dialog box

g. Send messages in tray

The Send Messages in Tray command of QMagNet-V reads and sends the message
stored in the file “OnlyOneMSMs.txt”, which can have the following formats:

1. "S" (means that a single message is going to be sent)

Format: {"phone" "message"}.
Example:
S
12345678910' 'This is a message'
2. "B" (means that a batch of messages is going to be sent).
Format: {"message" "comma-separated phone numbers"}
Example:
B
‘This is a message’
12345678910,12345678911,12345678912,12345678913

89
 1

Before the messages are sent, the registration procedure has to be carried out in order to
use the SMS_COMAPI 1.0 object of Clickatell provider. It consists on a username, a
password and the API ID, information that is supplied by Clickatell SMS Gateway.

Once as the registration is done and the button “Send Messages” is pressed, the external
file “sendSMS.exe”, which is a VB application that uses SMS_COMAPI 1.0 Type
Library, is executed. This sends the message and returns an error code that is logged at
the “SMSreport.txt” file, and printed up in the memo 1. The generated report may have
the following formats:

---------MESSAGES SENDING REPORT----------

The message has not sent
Error number : 1
Error description : Send Message Failed.

---------MESSAGES SENDING REPORT----------

Message successfully sent!
Error number : 0
Error description :
Message Status :

h. Maps
The button “Maps” allows visualizing the Villavicencio map in a separate form (

Figure 8.9). It uses the features provided by Whip!, such as zooming, panning, and layer
management. The map characteristic is useful for the manager to locate sectors, pipes,
hydrants, valves, etc, but it does not allow having direct interaction with the drawing
elements. In other words, the tool does not pretend to replace a GIS system, which is
highly recommended for WDS managing, but to provide some geographical orientation.

90
Both Villavicencio map and the EPANET model can be viewed with this tool.

Figure 8.9. Maps tool within QmagNet-V

i. Close
Closes the application.

91
9 Discussion

9.1 General discussion

• As part of the Water Safety Plan, which should be completely developed, the
need of a Hazard Assessment and a Risk Characterization certainly cannot be
avoided. This plan will support not only any suggestion that comes from the
proposed methodologies, but also the decision making process.

• The reliability of the models used in this research must be evaluated before
implementation. On one side, the intermittent model must be validated with
other methods and should be calibrated with field information, which is very
demanding since it requires the information of tanks levels at different times; On
the other side, the pressurized network model is just a supposition of what it is
expected in the future, after the consumption will be reduced.

9.2 The proposed approach for intermittent supply modelling

• It was demonstrated that faster closure of the tanks when the water arrives to
each house, results in faster achievement of completely pressurized network. A
rapid calculation shows that the network has a volume of storage of about
8000m3 within its pipes, and that the water required for filling up completely
those tanks is about 240.000m3. This means that with only 3.3% of the volume
required by the domestic tanks, the network could be completely filled.
Therefore, if the users are asked to close their tanks as soon as they receive
water, the system will pressurize faster and smaller water quantities will need to
be produced in the treatment plant, with the consequent financial and
environmental costs reduction.

• The showed charging up process of a pipe or network does not take into account
the effect of the air pressurization inside the pipes, which is a very important
issue for the operational methods of supply systems. For this research,
appropriate venting elements are considering to be available in the network and
to work properly.

• The oscillatory behaviour of the velocity within the pipes was observed in both
single pipes and looped networks, and it was observed that choosing an
appropriate time step for calculation, these oscillations are smoothed. This
means that they might be due to implemented numerical method, instead of
being related to physical situation; nevertheless, oscillations in piezometers exist
when sudden pipe filling is done. Vanconcelos et al (2004) observed physical
oscillations in the pressure during the filling up process in places where the pipe
was changed to the full state in a laboratory experiment.

• From the point of view of the SWMM numerical scheme, additional junctions
are needed to have enough computational points along a pipe, situation that

92
 affects the continuity of the system for surcharged systems.

• As mentioned before, the closed valves in the Villavicencio model were not
taken into account and this leads to a completely different charging process and
water reception times. The condition of completely open valves for the whole
system was taken because their status is not well documented when the system
restarts. Generally, the main valves are completely opened before the system
resumes, and then they are partially opened, by a proportion that depends on the
expertise of the current operator, but still with few technical criteria and without
any proper registration of its movements.

• The assumption to have a completely empty network just before the system
restarts is of course not true. Even if this was desired, this condition could hardly
be achieved in reality, since water bodies may remain static in local, lower parts
of the system. Interesting experiences from UNESCO-IHE staff experts such as
Prof. Vairavamoorthy, show that this particular problem is quite difficult to face
from the modelling point of view. The hydraulic behaviour of the pipe flow
when a static mass in a siphon is reached has not been completely understood,
and this is an issue still under research.

• The assumption of not considering the secondary supply source of the city,
Fuentes Altas, has a relation with the fact of the valves status. Certainly, if all
valves are opened, the system becomes completely interconnected in such a way
that both, Esmeralda treatment plant and Fuentes Altas are connected to each
other. Under this condition, the Esmeralda treatment plant would be flooded
after the systems fills up, since it is in a lower elevation than Fuentes Altas.

• As stated before, a typical consumption curve from a pressurized system of

similar weather conditions was assumed, because this information is not
available for Villavicencio. It is expected that different consumption factors do
not affect excessively the water reception times, because the water that enters
into those tanks is much more than the water that is consumed.

• The closure of the tanks should be performed by means of a float. According to

Aquadatos (2000), about 70% of the domestic tanks did not have such control
device in Villavicencio. Nevertheless, strong effort has been made from the
water supply company to make aware the users to keep their tanks with floats.
Unfortunately, at the present time the condition of those controls is unknown, as
well as the results of the social awareness campaign.

• The extraordinary computational time required to model the Villavicencio city

with a routing time step of 30 seconds does not allow its utilisation for real-time
management purposes, but bigger time steps may lead to higher instabilities.
Nevertheless, under the assumption that the initial conditions of the system
when it resumes will not be significantly different, the same offline results can
be used for management. Of course, big variations of these conditions may
produce totally different results.

• Unfortunately, there was no possibility to calibrate the model, since it requires

an exigent and unavailable real time monitoring data, such as the initial state of

93
 the levels in the households’ tanks, the discharge of main lines, the piezometric
changes in many points, the water reception time in each tank, as well as the
monitoring of its water level changes. Nevertheless, a demonstration for
modelling the intermittent process was shown, and it is expected that future
researchers can adopt, use and test it for systems where such information will be
available.

9.3 Managing contamination in WDS

The proposed research did not consider the pollutant as a chemical component, due
mainly to the lack of good water quality information in Villavicencio, which
wasincomplete and in some cases contradictory.

9.3.1 Finding the contamination sources within the network

• The assumption of having reliable and rapid water quality monitoring points in
the network is enough for the scope of this research. It is important to note that
WHO (2004) have an entire chapter dedicated to the monitoring issue, and that
its real-time performance is still under development.

• In this respect, many investigations have been developed, with special eagerness
after the terrorist attacks from the 9/11 event, such as the optimal location of
monitoring stations (Watson et al 2004, Tania et al 2004) and the different
technological approaches for early chemical, microbial and radioactive detection.
Brosnan (1999) states that, from the microbial point of view, the accuracy,
sensitivity and reliability are necessary for detecting small quality changes in
distribution systems, and that the monitoring systems should be able to detect
events such system failures, bad cross-connections and the presence of common
microorganisms. This author also warned that, although the measurements of
turbidity, chlorine residual and conductivity are rapid, they frequently indicate
hazards that do not exists, and therefore that new methods must be developed.

• Although the causes of contamination were not considered, they have to be

treated in complementary researches before implementing the project in any city.
The Guidelines for Drinking Water (WHO 2005) deal with this issue following a
preventive fashion within the Water Safety Plan, since often is more efficient to
invest in preventive processes than to invest in infrastructure to manage a hazard.
These processes include:

ƒ Distribution system maintenance.

ƒ Availability of backup systems (power supply).
ƒ Maintaining an adequate disinfectant residual.
ƒ Cross connection and backflow prevention devices implemented.
ƒ Fully enclosed distribution system and storages.
ƒ Appropriate repair procedures including subsequent disinfection of water
mains.
ƒ Maintaining adequate system pressure.
ƒ Maintaining security to prevent sabotage, illegal tapping and tampering.

94
 • Although these control measures are just mentioned into the WHO’s Water
Safety Plans, some other documents, such as the Safe Piped Water (WHO and
IWA 2004), offers much more detailed description, and it is advisable to go
through it to complement similar projects.

9.3.1.1 Database generation discussion

• The assumption of having only one contaminated node in each run is enough for
the present research scope. It is necessary to be aware, nevertheless, that the
pollution in a network may be produced in several nodes at a time, particularly
in areas susceptible to very low or very high pressures, or zones with high pipe
failure ratio.

• The consideration of having a tracer pollutant within the water quality model is a
simple but useful approach for the current study, since the main issue to look at
is the probable time of pollution occurrence in a particular node. It would be
even better, naturally, if decay models of pollutant concentration were
considered. In this case, comparing concentrations and estimating decaying
times for a particular chemical could find the time and, with the velocities within
the pipes, the location of the source of pollution.

• The database size should not be a restriction for choosing the time step of the
model, and, even more, it should not be a restriction for setting up the pollutant
target. Firstly, because the current computational technology allows managing
huge amount of information, and secondly because the contamination objective
must be taken from technical criteria, including the water quality and health
hazard analysis.

• The database generation should also consider different times of pollution,

instead of just applying it at 0:00 hours. This can largely affect the times of
pollutant arrival to the nodes, because the velocity within pipes increases as the
peak demands take place. In other words, if a pollution event occurs at the hour
of maximum demand, more nodes will be polluted in a minor period of time.

9.3.1.2 Searching database discussion

From the point of view of the database searching, it is clear that more than one sources
might be returned, and the operation manager has two choices: either searching for the
most probable source node (according with a network characterization and risk analysis,
not included in this project), or looking at the group of nodes as an areal problem. With
the first choice, he/she could be interested to know the best operational advice to
mitigate the pollutant effect, and at this point the COPA module, section 6.4.5, might be
useful.

9.3.2 Operational responses as an optimization problem

• The assumption of knowing the source of contamination is an ideal situation that
has been discussed in detail previously, and it may be the most important for the
reliability of the subsequent procedures.

• The assumption of the node affection could be weak if the concentration of a

95
 real chemical pollutant is considered instead. Certainly, a node affected by a
smaller percentage of a tracer may also be affected by a particular chemical or
biological contaminant with important public health impact. The developed
application, however, has implemented a parallel routine with the option of
using a chemical concentration. Again, the main issue here is appropriate
monitoring.

• The objective function used and described in section 6.4.5.2 considers both
number of affected nodes and the number of operational changes, with equal
weight within its conception. Nevertheless it is clear that the affected nodes
could have much more impact on the function than the number of network
changes, and therefore a different approach could be used, for example,
including these parameters with different factors or exponents, in terms of costs.
This situation should improve the convergence of the solution, since the function
will produce bigger values if more affected nodes are found.

• The optimisation problem should take into account the residual pressure in the
neighbouring nodes where the operational changes take place. Certainly, several
hydrants opened at the same time may cause, not only the discontinuity in the
service, but also potential back-siphonage effects due to the presence of negative
pressures, with the associated water quality problems. If the proposed approach
is used, the residual pressure in the higher zones must be verified to avoid
negative values during the opening of the hydrants.

• The use of pumping stations for flushing the pollutant out the network is a
solution that should be considered carefully. Indeed, a quality characterisation of
the groundwater must be performed first in order to make it capable for flushing
operation. For the Villavicencio case, an important number of old wells exist,
but they have been gradually abandoned, as the current network system has
expanded. The problem of their use has been discussed for the fact that, when
the supply service has been supported by those wells, customers’ complaints
related with bad odour and colour were often received at the supplier company.
In any case, the COPA module has been designed to take into account the
possibility of switching the existing well pumps either on or off.

• The optimisation problem should also take into consideration restrictions over
the involved costs. For instance, a limited number of operational changes in the
network may be possible, for the reason that the human and technological
resources to carry out such task might be limited.

• The optimisation process shown only considers the possibility to have one
contaminated node as a source, which is coming from the database; nevertheless,
several nodes could be potential sources of contamination at the same time and
therefore this should be complemented in both methodologies.

• If the optimisation problem has smaller size (only few elements to change), an
exhaustive search over the entire space of solutions can easily solve it. As the
variables can only take either the value 0 or 1, the total possibilities are 2v,
where v is the number of variables, and no optimisation process would be
needed for few of them. Nevertheless, for more complex problems, as the real

96
 ones, this method would demand much more computational effort than any
optimisation method.

• The Pareto front is used for multi-objective optimisation where commonly the
objectives have close, inverse relationship among them, in such a way that an
increase of one objective produces the decreases of other objective. In the
flushing problem this is not always the case, since it is not true that when more
elements are changed the number of affected nodes will be smaller.

• For big optimisation problems as shown in the Villavicencio’s case, the

computational effort is high. The case with 97 variables took about 15 minutes
to be completed, using an initial population of 100 and 8 generations. This time
is still very long if an emergency situation is considered. Some solutions for this
situation are considered in the recommendations.

• For the present research the Genetic Algorithms method was used for the
optimisation process and it proves to work well for water distribution problems,
as it has been demonstrated by other researches during the last decade
(Vairavamoorthy 2004).

9.4 The SMS for operative support and warning

• The right of consumers to health-related information on the water supplied to
them for domestic purposes is fundamental. However, in many communities, the
simple right of access to information will not ensure that individuals are aware
of the quality of the water supplied to them; furthermore, the probability of
consuming unsafe water may be relatively high. The Guidelines for Drinking
Water recommends at this respect that the agencies responsible for monitoring
should develop strategies for disseminating and explaining the significance of
health-related information.

• The messages have to be reliable and must avoid either false positive or false
negative messages. This could be achieved by following the Early Warning
System concepts (Brosnan 1999, Hasan et al 2004), which demand clear data
quality objectives that should include specificity (what to measure), sensitivity
(how results could be affected by different factors), accuracy (the quality of the
result), precision (an indication of the uniformity of the measurements) and
recovery (how easy is to get the results). As can be noted, this is a very complex
problem that has to be solved before implementing any warning system.

• One of the proposed messages for the intermittent supply assumed that one-hour
time is enough for the treated water to clean the pipes after systems resuming,
and therefore enough time to provide safe water to the consumers. It is, naturally,
a rough estimation, and consequently a serious research in this aspect must be
done before adapting such message.

• As stated by Hebbert (2001), at present time the technology for faster and
reliable pollution detection methods is limited, very costly and subject to
numerous false positives, and therefore more investigations for improving this

97
 issue are necessary.

• In general, the reliability of the process of sending the mobile phone messages is
higher, at this stage of the project, than the reliability of the proposed processes
that generate the messages, and this is the reason to recommend their
improvement before any implementation takes place.

• The proposed methodology for pressurized systems should consider also

messages to be sent to institutional authorities and health bodies when the water
becomes polluted. For example, educational buildings such as schools and
universities should be specially warned, in such a way that action plans can be
executed properly. Hospitals would have to be prepared for potential outbreaks
in case that a positive harmful contamination that cannot be mitigated, takes
place in the network. In this case the proper alarm level must be set for the
competent authorities.

• The messages for the affected people may not come to some of them, due to the
lack of mobile phones. At this respect, it is advisable that the warned person also
notifies its neighbours; furthermore, special bodies or community groups may
receive the messages and then spread it out to their neighbours again. The
messages could also be sent to the radio stations and local television service, but
with particular prudence in the sense that only some sectors of the city might
become contaminated, and consequently a message without enough
specifications may result in a clear false alarm for many people.

98
10 Conclusions

• The present research showed a different way to manage the water quality in
distribution networks, firstly acting against the intermittency and secondly by
using hydroinformatics tools such as models, software development and
communication technology.

• The WHO’s Guidelines for Drinking Water and the Water Safety Plan are
valuable documents that should be followed by every supply system in
developing countries. Their implementation would be a clear step toward the
Millennium Development Goals’ achievement. Explicitly, goal 7, point b -
“Reduce by half the proportion of people without sustainable access to safe
drinking water” would be touched, in addition to the implicit relationship with
the others goals.

• The National Health Institute (INS), Colombia, has developed a tool for the
water quality surveillance at national level, the SIVICAP, but the public health
is still under risk because the collected data is not useful for warning. Since the
methods and tools generated during this research are focused in warning, both
instruments could be used for complementing each other.

• The QMagNet-V application is a demonstration of what is possible to achieve

by the combination of tools for ensuring safe water, but still it needs to be
complemented with other processes such as the water quality monitoring and to
be proved and validated before a real implementation in a water supply system.

10.1 Intermittent supply

• The water produced by the treatment plant of Villavicencio fulfils the
requirements of quality, but unfortunately it is re-contaminated during the
distribution. The intermittency of the service is the main reason for this, which
leads to the intrusion of pollution within the pipes when they are empty.

• The intermittent condition in Villavicencio may be converted towards a

continuous supply by asking the people to use less their domestic tanks.

• The approach of using the free surface flow to model the network charging when
the system resumes needs to be compared with other developed charging models
in order to verify its validity.

10.2 Pressurized supply

• Due to the complexity of the flushing problem of pollutant in a distribution
network, intuitive solutions could be useless and even risky for the public health.

99
 • Elements not affected by a pollutant in time forward can still play a crucial role
for flushing it out.

10.3 The SMS for ensuring safe water

• Villavicencio has the required mobile phone service coverage to apply the tool
of the SMS for management and operation of the distribution system.

• The poor people can get the water quality information through the mobiles, since
a large proportion of these groups of people have access to this technology, and
therefore they can by directly benefited.

• Two-way communication implementation would improve importantly the

performance and usefulness of the proposed methodology, since some water
quality information may come indirectly from user complaints.

100
11 Recommendations

• A Geographic Information System can complement the proposed methodology.

That tool allows visualizing the water quality problem in a geographical basis, as
well as helping to understand the possible solutions that the QMagNet-V may
advise.

• QmagNet-V is not at a level of a final-user application, and consequently some

bugs may appear during its execution. Further researches could improve the
algorithms and the user interface.

• As stated in the precedent chapters, the groundwater is one of the possible

sources of contamination of the piped systems. For this reason, an additional
complement for the methodology would be the integration of groundwater
modelling and sewer leakage modelling, from the hydrodynamics and
contamination points of view. These models could be vertically integrated with
the charging process, in order to understand how the processes are occurring
between such systems.

11.1 The intermittent modelling approach

• A laboratory research to observe the filling up process in a network is needed in
order to have a better understanding of it, and certainly, to validate the results
observed with SWMM. Many researchers have studied this process
experimentally in single pipes for storm drainage projects, and therefore, they
have not taken into account additional behaviour of water in looped systems
with demand in the nodes.

• Similar analysis as presented should be carried out using other storm modelling
packages, such as MOUSE, in order to know the differences, similarities,
advantages and disadvantages of these packages for modelling the filling up
process in a distribution network.

• More research is needed to explain the oscillatory behaviour of the water depth
in a SWMM junction (piezometer), which at the same time will explain the
velocity oscillations in a pipe when it is being filled up.

• Advanced mathematical models have been developed (Vasconcelos, 2003c) for

the rapid filling pipe problem for sewer and drainage systems, which take into
account the effect of the pressurized air. An interesting research to be carried out
further, could be the interaction of these models for studying the specific
problem of the intermittency in supply networks.

• For further research with the Villavicencio’s charging model, the closed valves
must be taken into account in the process.

101
 • Due to the enormous computational time required for modelling a system like
Villavicencio, and also that the main purpose of it is the real time management
of the network during the charging process, different methodologies can later be
investigated. For instance, the use of neural networks to emulate the SWMM
model under different conditions could be an interesting and useful research.

11.2 Managing Water Quality problems in WDS

11.2.1 Looking at the pollutant sources
• A complementary project may consider a database generation for multiple
contaminated nodes at the same time. In this case it is probably not convenient
to consider the totality of the model nodes as potential sources of contamination
into the algorithm, but instead those could be selected from a hazard assessment
and risk characterization studies.

• Applying the pollution for different daily times should complement the proposed
approach for generating databases. Certainly, it should consider the effect of the
demand changes along the day, which has a direct impact over the velocities in
the system and therefore in the reception time of the pollutant to the nodes.

11.2.2 Treating the operational responses as an optimization problem

• The objective function could be modified to consider other practical situations,
such as the travel time of the operators from the water supply station to the place
where the network changes are required and the time needed by the operators for
closing the valves, which will depend on the valve diameter.

• The COPA module could be easily adapted to accept more that two different
source nodes. Naturally, the most important issue here is to have the proper set
of monitoring facilities that allow having the information of these nodes in time
to be used for COPA.

• As a complement to the best configuration network for flushing a pollutant,

future researches could take into consideration to provide the optimal path that
must be taken by the operators to change the network. For this, the elements
with more impact in the network should be the first to be switched; moreover,
the geographical location of each element and the time required for operating
them should be considered.

• It is necessary to develop a methodology for selecting the elements that are

possible to change in large systems, since this will determine the size of the
optimisation problem. As mentioned before, the number of possibilities is 2var,
thus any additional element exponentially increases the solution domain.
Nevertheless, this is a very difficult task, due to the fact that any element could
be crucial for the solution, as demonstrated in the small example shown in
section 6.4.6.

• As stated before, the optimisation process is taking a considerable amount of

time for big systems. One of the reasons for this situation is that every vector

102
 released by GLOBE, which is constituted by real numbers, are rounded to the
closest integer number 0 or 1 by the COPA module. This process, therefore,
implies that one identical operational status of the network is being evaluated a
considerable number of times. For this reason, a new algorithm that avoids the
unnecessary repetition of a certain solution that was already evaluated is
necessary to complement the proposed routine.

• Another cause for the required time to find an optimal operational advice is the
manipulation of the text files between COPA and GLOBE, and also the text files
required by COPA itself, which need to be opened and closed every time a
solution is investigated. A best strategy is to manage the information from the
Random Access Memory of the computer instead of managing it from its Read
Only Memory.

• Only GA was used in this research, but efficient, newer methodologies are
implemented in GLOBE and their performance should be evaluated for the
stated problem. It is possible that different algorithms provide better
performance in terms of time.

11.3 The SMS for operation support and warning

• Further research must be carried out for studying the process that takes place
when treated water collides with a mass of polluted water that has been intruded
into a pipe and to establish the time required by the treated water for flushing out
the dirty water completely. This time, then, would be included into the messages
to advice people the best hour to use safe drinking water after the system
resumes.

• New related projects should consider the two-way communication instead of

only one-way, as is proposed in the present research. Certainly, the user should
be able to send complaints about the water quality in order for the operators to
act properly. The two-way idea, in addition, could be adapted within the
framework of the supply management for requests and complaints, which is a
legal right for the consumers of any service in Colombia.

• The amazing growth and development of the mobile technology allows

nowadays having multimedia interface with the users. Therefore, the proposed
methodology for real time advice and warning could also be complemented with
the use of digital maps that could be sent additionally to the text messages to the
mobile phones as an informative complement of the emergency situation.

11.4 The QmagNet-V Application

The QmagNet-V needs to be complemented with the multi-objective approach, since the
current version only provides operational advices for the single-objective approach.

Sophisticated software engineering may improve the computational times for some of
the proposed methodologies.

103
12 References

Ainsworth, R. (ed.) (2004). Safe piped water : managing microbial water quality in
piped distribution systems. Geneva, Switzerland, World Health Organization and
London, UK, IWA Publishing.

Aquadatos ltda. (2000) Diseño e Implementación del Programa de Agua no

Contabilizada en Villavicencio, Meta. [Spanish][Design and Implementation of the
Unaccounted-For Water program for Villavicencio, Meta], Bogotá, Colombia.

Aquadatos ltda. (2002) Diseño e Implementación del Programa de Agua no

Contabilizada en 16 municipios del Departamento del Valle. [Spanish][Design and
Implementation of the Unaccounted-For Water program for 16 cities of Valle
Department], Bogotá, Colombia.

Arrieta G. (2002). Calidad del Agua en Colombia [Spanish][Water Quality in

Colombia]. Superintentencia de Servicios Públicos. Bogotá.

AWWA – EES (2002), Effects of Water Age on Distribution System Water Quality.
Prepared by AWWA with assistance from Economic and Engineering Services, Inc.
EPA White Paper.

Banco de la República (2004). Notas económicas regionales, region Llanos Orientales

[Spanish][Regional economic notes. Easter Plane Region]. Banco de la República,
Subgerencia de Estudios Económicos, Centro Regional de Estudios Económicos –
CREE, Villavicencio.

Barreto W, Solomatine D. (2006) NSGAX software for multi-objective optimisation

problems. (Under development) UNESCO-IHE, Delft, The Netherlands.

Brosnan T. M. (1999) Early Warning Monitoring to Detect Hazardous Events in Water

Supplies, An ILSI Risk Science Institute Workshop Report, ISBN: 1-57881-075-2.

Cairncross A (1990) Health impacts in developing countries: new evidences and new
prospects. Journal of the Institution of Water and Environmental Management. 4(6)
571-577.

Ford T., Colwell R. (1996) A global decline in microbiological safety of water: A call
for action. A report from The American Academy of Microbiology.

Godfrey S., Howard G. (2004) Improved risk assessment and management in urban
piped water supplies, Department for International Development – DFID, London.

Godfrey, S. and Howard, G., (2004a). Water Safety Plans (WSP) for Urban Piped
Water Supplies in Developing Countries, WEDC, Loughborough University, UK.

Gleick Andrews G. (1998) Backflow protections for private water systems. Oregon

104
State University Extension Service.

Gleick, Peter H. (2002) Dirty Water: Estimated Deaths from Water-Related Diseases
2000-2020. Pacific Institute for Studies in Development, Environment, and Security.
Pacific Institute Research Report.

Gomez M. Achiaga V. (2001) Mixed flow modelling produced by pressure fronts from
upstream and downstram extremes. Dep. Hydraulics, Maritime and Environmental
Engineering. UPC, Barcelona, Spain.

Hart, F. Wheeler W (1992), Modelling Chlorine Residuals in Water Distribution

Systems. Presented at the NEWWA Computer Symposium.

Hassan J., States S and Deininger R. (2004) Safeguarding the Security of Public Water
Supplies Using Early Warning Systems: A Brief Review. Universities Council on Water
Resources. Journal of Contemporary Water Research and Education, 129: 27-33

Hebert R. (2001) A brief discussion of water security issues following the September 11,
2001 terrorist attacks. ECO Resources, Inc. (Sugar Land, Texas) and The National
Council for Public-Private Partnerships (Washington, DC), US.

Howard, A.G. (2002), Water Supply Surveillance: A reference manual, WEDC,

Loughborough University, UK.

Liou C, Hunt W (1986) Filling of pipes with ondulating elevation profiles, J. Hydr.
Engrg, 122(10) 534-539 Cited in Godfrey S., Howard G. (2004) Improved risk
assessment and management in urban piped water supplies, Department for
International Development – DFID, London.

LeChevallier M, Gullick R, Kari M. (2002) The Potential for Health Risks from
Intrusion of Contaminants into the Distribution System from Pressure Transients.
American Water Works Service Company, Inc., Voorhees, NJ.

Maier S.H. (1999), Modelling water quality for water distribution systems. PhD Thesis.
Brunel University, Uxbridge, United Kingdom.

Macke S. Battermann (2001) A Strategy to Reduce Technical Water Losses for

Intermittent Water Supply [Online] http://sdteffen.de/diplom/web/ [Accessed 9 Nov
2005].

McCullough L. (2005) Emergency Alerts: Coming to Your Cell Phone? [8 paragraphs]

PC World [Online]
http://www.pcworld.com/resource/article/0,aid,122025,pg,1,RSS,RSS,00.asp [Accessed
10 Mar 2006].

Medema, G. J. Payment, P. Dufour, A. Robertson, W. Waite, M. Hunter, P. Kirby, R.

and Anderson, Y. (2003) Safe drinking water: an ongoing challenge. Assessing
Microbial Safety of Drinking Water - Improving approaches and methods, in Fewtrell,
L., (ed)., IWA, London.

Mincomunicaciones (2004) Análisis del sector de las Comunicaciones en Colombia [in

105
Spanish], Bogotá, Colombia.
Nayeema N (2006), Urban Flood Warning System with wireless technology [under
development], MSc Thesis, UNESCO-IHE, Delft, Netherlands.

OECD–WHO (2003), Assessing microbial safety of drinking water. Improving

Approaches & Methods. IWA Publishing, London.

Ortiz J, Bonilla H, Arrieta G, Florez L, Nava G (2004) Sistema de información para

vigilancia de calidad de agua potable –SIVICAP [In Spanish] [Information System for
the Water Quality Surveillance]. Instituto Nacional de Salud (INS), Bogotá, Colombia

Payment P, Hunter P. (2001) Endemic and epidemic infectious intestinal decease and its
relationship in drinking water. In: Fewtrell L. and Bartram J. Water Quality: guidelines,
standards and health, IWA, London, pp 61-68.

Popescu I. (2004) Numerical Methods for Differential Equations, Lecture Notes,

HH521/04/1, UNESCO-IHE, Delft, The Netherlands.

Price R. (2001), Mathematical Modelling. Lecture notes HH357/01/1, UNESCO-IHE

Delft, The Netherlands.

Rossman L (2000) Epanet Users Manual. EPA, USA.

Rossman L. (2005b) Storm water management model SWMM, quality assurance report:
Dynamic Wave Flow Routing, Water Supply and Water Resources Division, EPA,
Cincinnati, OH, US.

Sisay, S (2003) Real-Time Decision Support System for Flood Management, IHE, Delft,
The Netherlands.

Srinivasa Reddy, L.S., Elango, K. (1989). “Analysis of water distribution networks with
head-dependent outlets,” Civil Engineering Systems, Vol 6(3), pp: 102-110. Cited in:
Totsuka N., Trifunovic N, Vairavamoorthy K. (2004) Intermittent urban water supply
under water starving situations. 30th WEDC International Conference, Vientiane, Lao
PDR.

Sohn J, Amy G, Cho J, Lee Y, Yoon Y (2004) Disinfectant decay and disinfection by-
products formation model development: chlorination and ozonation by-products. Water
research 38, 2461 – 2478.

Solomatine D. (1992) Software Engineering. An introduction. - Lecture Notes. - Delft,

The Netherlands: International Institute for Infrastructural, Hydraulic and
Environmental Engineering (IHE), 1991, 2000.

Solomatine, D.P. (1999). Two strategies of adaptive cluster covering with descent and
their comparison to other algorithms. Journal of Global Optimization, 1999, vol. 14, No.
1, pp. 55-78.

SIC Superintendencia Industria y Comercio (2006), Boletín de Prensa Enero 12 2006

[In Spanish] Bulletin Press Jan 12th 2006 [Online]
http://www.sic.gov.co/Articulos_Pagina_Principal/Boletines_Prensa/2006/Incremento_

106
109,76_Numero.php [Accessed 14 March 2006]

Totsuka N., Trifunovic N, Vairavamoorthy K. (2004) Intermittent urban water supply

under water starving situations. 30th WEDC International Conference, Vientiane, Lao
PDR.

Umbarila P, Alfonso L (2003) Methodology for the assessment and monitoring of

management of supply networks. Second International Conference. Efficient 2003,
Tenerife, Spain.

Vairavamoorthy, K. (1994). Water Distribution Networks: Design and Control for

Intermittent Supply. PhD Thesis, Imperial college: London. Cited in Totsuka N.,
Trifunovic N, Vairavamoorthy K. (2004) Intermittent urban water supply under water
starving situations. 30th WEDC International Conference, Vientiane, Lao PDR.

Vairavamoorthy, Ali, Akinpelu, Lin. (2000) An Appropriate Design Tool For

Intermittent Water Supply Systems. Technical report, South Bank University, London.
Cited in Macke S. Battermann (2001) A Strategy to Reduce Technical Water Losses for
Intermittent Water Supply [Online] http://sdteffen.de/diplom/web/ [Accessed 9 Nov
2005].

Vairavamoorthy K. (2004) Special Topics in Water Transport and Distribution Part II.
Lecture notes EE517/04/1, UNESCO-IHE, Delft, The Netherlands.

Vasconcelos, J.G. and Wright, S.J. (2003a), Laboratory Investigation of Surges Formed
During Rapid Filling of Stormwater Storage Tunnel. Proceedings of 30th IAHR
Congress, Thessaloniki, Greece.

Vasconcelos, J.G. and Wright, S.J. (2003b), Surges Associated with Air Expulsion in
Near-Horizontal Pipelines. Proceedings 4th Joint ASME-JSME Fluids Engineering
Conference, Honolulu, Hawaii.

Vasconcelos, J.G. and Wright, S.J. (2003c), Numerical Modelling of the Transition
Between Free Surface and Pressurized Flow in Storm Sewers, University of
Michigan, Ann Arbor, USA

Vasconcelos, J.G. and Wright, S.J. (2004), Applications and Limitations of Single-
Phase Models to the Description of the Rapid Filling Pipe Problem, University of
Michigan, Ann Arbor, USA

Walski T. (2001), Water Distribution Models for Emergency Response, Water Security
Summit, Hartford, Connecticut, US.

Watson J.P., Greenbergz H., Hart W. (2004), A Multiple-Objective Analysis of Sensor

Placement Optimization in Water Networks. Department of Computer Science,
University of New Mexico, Albuquerque NM, US.

WHO (2004) Guidelines for drinking water quality. Volume 1. Recommendations.

WHO, Geneva. 3 ed.

WHO/UNICEF (2004) Meeting the MDG Drinking Water and Sanitation Target, a

107
Mid-Term Assessment of Progress. World Health Organization and United Nations
Children’s Fund.

World Bank (WB). (2004). World Development Report 2004. World Bank:
Washington D.C.

108