Water Distribution Modelling PDF

ADVANCED WATER
DISTRIBUTION MODELING
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Contributing Authors
Scott Cattran, Rick Hammond, Kevin Laptos, Steven G. Lowry,
Robert F. Mankowski, Stan Plante, John Przybyla, Barbara Schmitz
Peer Review Board

Lee Cesario (Denver Water), Robert M. Clark (U.S. EPA),
Jack Dangermond (ESRI), Allen L. Davis (CH2M Hill),
Paul DeBarry (Borton-Lawson), Frank DeFazio (Franklin G. DeFazio Corp.),
Kevin Finnan (Bristol Babcock), Wayne Hartell (Bentley Systems),
Brian Hoefer (ESRI), Bassam Kassab (Santa Clara Valley Water District),
James W. Male (University of Portland), William M. Richards
(WMR Engineering), Zheng Wu (Bentley Systems ),
and E. Benjamin Wylie (University of Michigan)
Click here to visit the Bentley Institute

Press Web page for more information
C H A P T E R
1
Introduction to Water Distribution
Modeling
Water distribution modeling is the latest technology in a process of advancement that

began two millennia ago when the Minoans constructed the first piped water convey-
ance system. Today, water distribution modeling is a critical part of designing and
operating water distribution systems that are capable of serving communities reliably,
efficiently, and safely, both now and in the future. The availability of increasingly
sophisticated and accessible models allows these goals to be realized more fully than
ever before.
This book is structured to take the engineer through the entire modeling process, from
gathering system data and understanding how a computer model works, through con-
structing and calibrating the model, to implementing the model in system design and
operations. The text is designed to be a first course for the novice modeler or
engineering student, as well as a reference for those more experienced with
distribution system simulations.
This chapter introduces the reader to water distribution modeling by giving an over-
view of the basic distribution system components, defining the nature and purposes of
distribution system simulations, and outlining the basic steps in the modeling process.
The last section of the chapter presents a chronology of advancements in water
distribution.
1.1 ANATOMY OF A WATER DISTRIBUTION SYSTEM

Although the size and complexity of water distribution systems vary dramatically,
they all have the same basic purpose—to deliver water from the source (or treatment
facility) to the customer.
2 Introduction to Water Distribution Modeling Chapter 1
Sources of Potable Water

Untreated water (also called raw water) may come from groundwater sources or sur-
face waters such as lakes, reservoirs, and rivers. The raw water is usually transported
to a water treatment plant, where it is processed to produce treated water (also known
as potable or finished water). The degree to which the raw water is processed to
achieve potability depends on the characteristics of the raw water, relevant drinking
water standards, treatment processes used, and the characteristics of the distribution
system.
Before leaving the plant and entering the water distribution system, treated surface
water usually enters a unit called a clearwell. The clearwell serves three main pur-
poses in water treatment. First, it provides contact time for disinfectants such as
chlorine that are added near the end of the treatment process. Adequate contact time is
required to achieve acceptable levels of disinfection.
Second, the clearwell provides storage that acts as a buffer between the treatment
plant and the distribution system. Distribution systems naturally fluctuate between
periods of high and low water usage, thus the clearwell stores excess treated water
during periods of low demand and delivers it during periods of peak demand. Not
only does this storage make it possible for the treatment plant to operate at a more sta-
ble rate, but it also means that the plant does not need to be designed to handle peak
demands. Rather, it can be built to handle more moderate treatment rates, which
means lower construction and operational costs.
Third, the clearwell can serve as a source for backwash water for cleaning plant filters
that, when needed, is used at a high rate for a short period of time.
In the case of groundwater, many sources offer up consistently high quality water that
could be consumed without disinfection. However, the practice of maintaining a dis-
infectant residual is almost always adhered to for protection against accidental
contamination and microbial regrowth in the distribution system. Disinfection at
groundwater sources differs from sources influenced by surface water in that it is usu-
ally applied at the well itself.
Customers of Potable Water

Customers of a water supply system are easily identified — they are the reason that
the system exists in the first place. Homeowners, factories, hospitals, restaurants, golf
courses, and thousands of other types of customers depend on water systems to pro-
vide everything from safe drinking water to irrigation. As demonstrated throughout
the book, customers and the nature in which they use water are the driving mechanism
behind how a water distribution system behaves. Water use can vary over time both in
the long-term (seasonally) and the short-term (daily), and over space. Good knowledge
of how water use is distributed across the system is critical to accurate modeling.
Transport Facilities
Moving water from the source to the customer requires a network of pipes, pumps,
valves, and other appurtenances. Storing water to accommodate fluctuations in
demand due to varying rates of usage or fire protection needs requires storage facili-
Section 1.1 Anatomy of a Water Distribution System 3
ties such as tanks and reservoirs. Piping, storage, and the supporting infrastructure are
together referred to as the water distribution system (WDS).
Transmission and Distribution Mains. This system of piping is often cate-
gorized into transmission/trunk mains and distribution mains. Transmission mains
consist of components that are designed to convey large amounts of water over great
distances, typically between major facilities within the system. For example, a trans-
mission main may be used to transport water from a treatment facility to storage tanks
throughout several cities and towns. Individual customers are usually not served from
transmission mains.
Distribution mains are an intermediate step toward delivering water to the end cus-
tomers. Distribution mains are smaller in diameter than transmission mains, and
typically follow the general topology and alignment of the city streets. Elbows, tees,
wyes, crosses, and numerous other fittings are used to connect and redirect sections of
pipe. Fire hydrants, isolation valves, control valves, blow-offs, and other maintenance
and operational appurtenances are frequently connected directly to the distribution
mains. Services, also called service lines, transmit the water from the distribution
mains to the end customers.
Homes, businesses, and industries have their own internal plumbing systems to trans-
port water to sinks, washing machines, hose bibbs, and so forth. Typically, the internal
plumbing of a customer is not included in a WDS model; however, in some cases,
such as sprinkler systems, internal plumbing may be modeled.
System Configurations. Transmission and distribution systems can be either
looped or branched, as shown in Figure 1.1. As the name suggests, in looped systems
there may be several different paths that the water can follow to get from the source to
a particular customer. In a branched system, also called a tree or dendritic system, the
water has only one possible path from the source to a customer.
Figure 1.1
Looped and branched
networks
Looped Branched
Looped systems are generally more desirable than branched systems because, coupled
with sufficient valving, they can provide an additional level of reliability. For exam-
ple, consider a main break occurring near the reservoir in each system depicted in
Figure 1.2. In the looped system, that break can be isolated and repaired with little
impact on customers outside of that immediate area. In the branched system, however,
all the customers downstream from the break will have their water service interrupted
until the repairs are finished. Another advantage of a looped configuration is that,
because there is more than one path for water to reach the user, the velocities will be
lower, and system capacity greater.
Figure 1.2
Looped and branched
networks after
network failure
Customers
Without
Service Pipe Break
Customers
Without
Service
Looped Branched
Most water supply systems are a complex combination of loops and branches, with a
trade-off between loops for reliability (redundancy) and branches for infrastructure
cost savings. In systems such as rural distribution networks, the low density of cus-
tomers may make interconnecting the branches of the system prohibitive from both
monetary and logistical standpoints.
1.2 WHAT IS A WATER DISTRIBUTION SYSTEM

SIMULATION?
The term simulation generally refers to the process of imitating the behavior of one
system through the functions of another. In this book, the term simulation refers to the
process of using a mathematical representation of the real system, called a model.
Network simulations, which replicate the dynamics of an existing or proposed sys-
tem, are commonly performed when it is not practical for the real system to be
directly subjected to experimentation, or for the purpose of evaluating a system before
it is actually built. In addition, for situations in which water quality is an issue,
directly testing a system may be costly and a potentially hazardous risk to public
health.
Section 1.2 What Is a Water Distribution System Simulation? 5
Simulations can be used to predict system responses to events under a wide range of
conditions without disrupting the actual system. Using simulations, problems can be
anticipated in proposed or existing systems, and solutions can be evaluated before
time, money, and materials are invested in a real-world project.
For example, a water utility might want to verify that a new subdivision can be pro-
vided with enough water to fight a fire without compromising the level of service to
existing customers. The system could be built and tested directly, but if any problems
were to be discovered, the cost of correction would be enormous. Regardless of
project size, model-based simulation can provide valuable information to assist an
engineer in making well-informed decisions.
Simulations can either be steady-state or extended-period. Steady-state simulations
represent a snapshot in time and are used to determine the operating behavior of a sys-
tem under static conditions. This type of analysis can be useful in determining the
short-term effect of fire flows or average demand conditions on the system. Extended-
period simulations (EPS) are used to evaluate system performance over time. This
type of analysis allows the user to model tanks filling and draining, regulating valves
opening and closing, and pressures and flow rates changing throughout the system in
response to varying demand conditions and automatic control strategies formulated
by the modeler.
Modern simulation software packages use a graphical user interface (GUI) that
makes it easier to create models and visualize the results of simulations. Older-
generation software relied exclusively on tabular input and output. A typical modern
software interface with an annotated model drawing is shown in Figure 1.3.
Figure 1.3
Software interface and
annotated model
drawing
1.3 APPLICATIONS OF WATER DISTRIBUTION MODELS

Most water distribution models (WDMs) can be used to analyze a variety of other
pressure piping systems, such as industrial cooling systems, oil pipelines, or any net-
work carrying an incompressible, single-phase, Newtonian fluid in full pipes. Munic-
ipal water utilities, however, are by far the most common application of these models.
Models are especially important for WDSs due to their complex topology, frequent
growth and change, and sheer size. It is not uncommon for a system to supply hun-
dreds of thousands of people (large networks supply millions); thus, the potential
impact of a utility decision can be tremendous.
Water distribution network simulations are used for a variety of purposes, such as
• Long-range master planning, including both new development and rehabili-
tation
• Fire protection studies
• Water quality investigations
• Energy management
• System design
• Daily operational uses including operator training, emergency response, and
troubleshooting
Long-Range Master Planning

Planners carefully research all aspects of a water distribution system and try to deter-
mine which major capital improvement projects are necessary to ensure the quality of
service for the future. This process, called master planning (also referred to as capital
improvement planning or comprehensive planning), may be used to project system
growth and water usage for the next 5, 10, or 20 years. System growth may occur
because of population growth, annexation, acquisition, or wholesale agreements
between water supply utilities. The capability of the hydraulic network to adequately
serve its customers must be evaluated whenever system growth is anticipated.
Not only can a model be used to identify potential problem areas (such as future low
pressure areas or areas with water quality problems), but it can also be used to size
and locate new transmission mains, pumping stations, and storage facilities to ensure
that the predicted problems never occur. Maintaining a system at an acceptable level of
service is preferable to having to rehabilitate a system that has become problematic.
Rehabilitation
As with all engineered systems, the wear and tear on a water distribution system may
lead to the eventual need to rehabilitate portions of the system such as pipes, pumps,
valves, and reservoirs. Pipes, especially older, unlined, metal pipes, may experience
an internal buildup of deposits due to mineral deposits and chemical reactions within
the water. This buildup can result in loss of carrying capacity, reduced pressures, and
Section 1.3 Applications of Water Distribution Models 7
poorer water quality. To counter these effects of aging, a utility may choose to clean
and reline a pipe. Alternatively, the pipe may be replaced with a new (possibly larger)
pipe, or another pipe may be installed in parallel. Hydraulic simulations can be used
to assess the impacts of such rehabilitation efforts, and to determine the most econom-
ical improvements.
Fire Protection Studies

Water distribution systems are often required to provide water for fire fighting pur-
poses. Designing the system to meet the fire protection requirements is essential and
normally has a large impact on the design of the entire network. The engineer deter-
mines the fire protection requirements and then uses a model to test whether the
system can meet those requirements. If the system cannot provide certain flows and
maintain adequate pressures, the model may also be used for sizing hydraulic
elements (pipes, pumps, etc.) to correct the problem.
Water Quality Investigations

Some models provide water quality modeling in addition to hydraulic simulation
capabilities. Using a water quality model, the user can model water age, source trac-
ing, and constituent concentration analyses throughout a network. For example, chlo-
rine residual maintenance can be studied and planned more effectively, disinfection
by-product formation (DBP) in a network can be analyzed, or the impact of storage
tanks on water quality can be evaluated. Water quality models are also used to study
the modification of hydraulic operations to improve water quality.
Energy Management
Next to infrastructure maintenance and repair costs, energy usage for pumping is the
largest operating expense of many water utilities (Figure 1.4). Hydraulic simulations
can be used to study the operating characteristics and energy usage of pumps, along
with the behavior of the system. By developing and testing different pumping strate-
gies, the effects on energy consumption can be evaluated, and the utility can make an
educated effort to save on energy costs.
Daily Operations
Individuals who operate water distribution systems are generally responsible for mak-
ing sure that system-wide pressures, flows, and tank water levels remain within
acceptable limits. The operator must monitor these indicators and take action when a
value falls outside the acceptable range. By turning on a pump or adjusting a valve,
for example, the operator can adjust the system so that it functions at an appropriate
level of service. A hydraulic simulation can be used in daily operations to determine
the impact of various possible actions, providing the operator with better information
for decision-making.
Figure 1.4
Pumping is one of the
largest operating
expenses of many
utilities
Operator Training. Most water distribution system operators do their jobs very
well. As testimony to this fact, the majority of systems experience very few water out-
ages, and those that do occur are rarely caused by operator error. Many operators,
however, gain experience and confidence in their ability to operate the system only
over a long period of time, and sometimes the most critical experience is gained under
conditions of extreme duress. Hydraulic simulations offer an excellent opportunity to
train system operators in how their system will behave under different loading condi-
tions, with various control strategies, and in emergency situations.
Emergency Response. Emergencies are a very real part of operating a water

distribution system, and operators need to be prepared to handle everything from main
breaks to power failures. Planning ahead for these emergencies by using a model may
prevent service from being compromised, or may at least minimize the extent to
which customers are affected. Modeling is an excellent tool for emergency response
planning and contingency.
System Troubleshooting. When hydraulic or water quality characteristics in
an existing system are not up to standard, a model simulation can be used to identify
probable causes. A series of simulations for a neighborhood that suffers from chronic
low pressure, for example, may point toward the likelihood of a closed valve in the
area. A field crew can then be dispatched to this area to check nearby valves.
1.4 THE MODELING PROCESS

Assembling, calibrating, and using a water distribution system model can seem like a
foreboding task to someone confronted with a new program and stacks of data and
maps of the actual system. As with any large task, the way to complete it is to break it
down into its components and work through each step. Some tasks can be done in par-
allel while others must be done in series. The tasks that make up the modeling process
are illustrated in Figure 1.5. Note that modeling is an iterative process.
Section 1.4 The Modeling Process 9
Figure 1.5
Flowchart of the
modeling process
The first step in undertaking any modeling project is to develop a consensus within
the water utility regarding the need for the model and the purposes for which the
model will be used in both the short- and long-term. It is important to have utility per-
sonnel, from upper management and engineering to operations and maintenance,
commit to the model in terms of human resources, time, and funding. Modeling
should not be viewed as an isolated endeavor by a single modeler, but rather a utility-
wide effort with the modeler as the key worker. After the vision of the model has been
accepted by the utility, decisions on such issues as extent of model skeletonization
and accuracy of calibration will naturally follow.
Figure 1.5 shows that most of the work in modeling must be done before the model
can be used to solve real problems. Therefore, it is important to budget sufficient time
to use the model once it has been developed and calibrated. Too many modeling
projects fall short of their goals for usage because the model-building process takes up
all of the allotted time and resources. There is not enough time left to use the model to
understand the full range of alternative solutions to the problems.
Modeling involves a series of abstractions. First, the real pipes and pumps in the sys-
tem are represented in maps and drawings of those facilities. Then, the maps are
converted to a model that represents the facilities as links and nodes. Another layer of
abstraction is introduced as the behaviors of the links and nodes are described
mathematically. The model equations are then solved, and the solutions are typically
displayed on maps of the system or as tabular output. A model’s value stems from the
usefulness of these abstractions in facilitating efficient design of system improve-
ments or better operation of an existing system.
1.5 A BRIEF HISTORY OF WATER DISTRIBUTION

TECHNOLOGY
The practice of transporting water for human consumption has been around for sev-
eral millennia. From the first pipes in Crete some 3,500 years ago, to today’s complex
hydraulic models, the history of water distribution technology is quite a story. The fol-
lowing highlights some of the key historical events that have shaped the field since its
beginnings.
1500 B.C. — First water distribution pipes used in Crete. The Minoan civilization
flourishes on the island of Crete. The City of Knossos develops an aqueduct system
that uses tubular conduits to convey water. Other ancient civilizations have had sur-
face water canals, but these are probably the first pipes.
250 B.C. — Archimedes principle developed. Archimedes, best known for his dis-
covery of S and for devising exponents, develops one of the earliest laws of fluids
when he notices that any object in water displaces its own volume. Using this princi-
ple, he proves that a crown belonging to King Hiero of Syracuse is not made of gold.
A legend will develop that he discovered this principle while bathing and became so
excited that he ran naked through the streets shouting “Eureka” (I’ve found it).
100 A.D. — Roman aqueducts. The Romans bring water from great distances to
their cities through aqueducts (Figure 1.6). While many of the aqueducts are above-
Section 1.5 A Brief History of Water Distribution Technology 11
ground, there are also enclosed conduits to supply public fountains and baths. Sextus
Julius Frontinus, water commissioner of Rome, writes two books on the Roman water
supply.
Figure 1.6
Roman aqueduct
1455 — First cast iron pipe. Casting of iron for pipe becomes practical, and the first
installation of cast iron pipe, manufactured in Siegerland, Germany, occurs at Dillen-
burg Castle.
1652 — Piped water in Boston. The first water pipes in the U.S. are laid in Boston to
bring water from springs to what is now the Quincy Market area.
1664 — Palace of Versailles. King Louis XIV of France orders the construction of a
15-mile cast iron water main from Marly-on-Seine to the Palace of Versailles. This is
the longest pipeline of its kind at this time, and portions of it remain in service into the
21st century. A section of the line, after being taken out of service, was shipped in the
1960s from France to the United States (Figure 1.7) where it is still on display.
Figure 1.7
King Louis XIV of
France and a section
of the Palace of
Versailles pipeline
Courtesy of the Ductile Iron Pipe Research Association

1732 — Pitot invents a velocity-measuring device. Henri Pitot is tasked with mea-
suring the velocity of water in the Seine River. He finds that by placing an L-shaped
tube into the flow, water rises in the tube proportionally to the velocity squared, and
the Pitot tube is born.
1738 — Bernoulli publishes Hydrodynamica. The Swiss Bernoulli family extends

the early mathematics and physics discoveries of Newton and Leibniz to fluid sys-
tems. Daniel Bernoulli publishes Hydrodynamica while in St. Petersburg and
Strasbourg, but there is a rivalry with his father Johann regarding who actually
developed some of the principles presented in the book. These principles will become
the key to energy principles used in hydraulic models and the basis for numerous
devices such as the Venturi meter and, most notably, the airplane wing. In 1752,
however, it will actually be their colleague, Leonard Euler, who develops the forms of
the energy equations that will live on in years to come.
1754 — First U.S. water systems built. The earliest water distribution systems in the
United States are constructed in Pennsylvania. The Moravian community in Bethle-
hem, Pennsylvania claims to have the first water system, and it is followed quickly by
systems in Schaefferstown and Philadelphia, Pennsylvania. Horses drive the pumps in
the Philadelphia system, and the pipes are made of bored logs. They will later be
replaced with wood stave pipes made with iron hoops to withstand higher pressures.
The first steam driven pumps will be used in Bethlehem ten years later.
1770 — Chezy develops head loss relationship. While previous investigators real-
ized that energy was lost in moving water, it is Antoine Chezy who realizes that V2/
RS is reasonably constant for certain situations. This relationship will serve as the
basis for head loss equations to be used for centuries.
1785 — Bell and spigot joint developed. The Chelsea Water Company in London
begins using the first bell and spigot joints. The joint is first packed with yarn or hemp
and is then sealed with lead. Sir Thomas Simpson is credited with inventing this joint,
which replaced the crude flanged joints used previously.
1839 — Hagen-Poiseuille equation developed. Gotthilf Hagen and Jean Louis Poi-
seuille independently develop the head loss equations for laminar flow in small tubes.
Their work is experimental, and it is not until 1856 that Franz Neuman and Eduard
Hagenbach will theoretically derive the Hagen-Poiseuille equation.
1843 — St. Venant develops equations of motion. Several researchers, including

Louis Navier, George Stokes, Augustin de Cauchy, and Simeon Poisson, work toward
the development of the fundamental differential equations describing the motion of
fluids. They become known as the “Navier-Stokes equations.” Jean-Claude Barre de
Saint Venant develops the most general form of these equations, but the term St.
Venant equations will be used to refer to the vertically and laterally averaged (that is,
one-dimensional flow) form of equations.
1845 — Darcy-Weisbach head loss equation developed. Julius Weisbach publishes

a three-volume set on engineering mechanics that includes the results of his experi-
ments. The Darcy-Weisbach equation comes from this work, which is essentially an
extension of Chezy’s work, as Chezy’s C is related to Darcy-Weisbach’s f by C2=8g/f.
Darcy’s name is also associated with Darcy’s law for flow through porous media,
widely used in groundwater analysis.
1878 — First automatic sprinklers used. The first Parmelee sprinklers are installed.
These are the first automatic sprinklers for fire protection.
1879 — Lamb’s Hydrodynamics published. Sir Horace Lamb publishes his Treatise
on the Mathematical Theory of the Motion of Fluids. Subsequent editions will be pub-
lished under the title Hydrodynamics, with the last edition published in 1932.
1881 — AWWA formed. The 22 original members create the American Water Works
Association. The first president is Jacob Foster from Illinois.
1883 — Laminar/turbulent flow distinction explained. While earlier engineers

such as Hagen observed the differences between laminar and turbulent flow, Osborne
Reynolds is the first to conduct the experiments that clearly define the two flow
regimes. He identifies the dimensionless number, later referred to as the Reynolds
number, for quantifying the conditions under which each type of flow exists. He pub-
lishes “An Experimental Investigation of the Circumstances which Determine
whether the Motion of Water shall be Direct or Sinuous and the Law of Resistance in
Parallel Channels.”
1896 — Cole invents Pitot tube for pressure pipe. Although numerous attempts
were made to extend Henri Pitot’s velocity measuring device to pressure pipes,
Edward Cole develops the first practical apparatus using a Pitot tube with two tips
connected to a manometer. The Cole Pitometer will be widely used for years to come,
and Cole’s company, Pitometer Associates, will perform flow measurement studies
(among many other services) into the 21st century.
1906 — Hazen-Williams equation developed. A. Hazen and G.S. Williams develop

an empirical formula for head loss in water pipes. Although not as general or precise
in rough, turbulent flow as the Darcy-Weisbach equation, the Hazen-Williams equa-
tion proves easy to use and will be widely applied in North America.
1900 – 1930 — Boundary Layer Theory developed. The interactions between flu-
ids and solids are studied extensively by a series of German scientists lead by Ludwig
Prandtl and his students Theodor von Karman, Johan Nikuradse, Heinrich Blasius,
and Thomas Stanton. As a result of their research, they are able to theoretically
explain and experimentally verify the nature of drag between pipe walls and a fluid.
In particular, the experiments of Nikuradse, who glues uniform sand grains inside
pipes and measures head loss, lead to a better understanding of the calculation of the f
coefficient in the Darcy-Weisbach equation. Stanton develops the first graphical rep-
resentation of the relationship between f, pipe roughness, and the Reynolds number,
which later leads to the Moody diagram. This work is summarized in H. Schichting’s
book, Boundary Layer Theory.
1914 — First U.S. drinking water standards established. The U.S. Public Health
Service publishes the first drinking water standards, which will continually evolve.
The U.S. Environmental Protection Agency (U.S. EPA) will eventually assume the
role of setting the water quality standards in the United States.
1920s — Cement-mortar lining of water mains. Cement mortar lining of water

mains is used to minimize corrosion and tuberculation. Procedures for cleaning and
lining existing pipes in place will be developed by the 1930s.
1921 — First Hydraulic Institute Standards published. The first edition of Trade
Standards in the Pump Industry is published as a 19-page pamphlet. These standards
become the primary reference for pump nomenclature, testing, and rating.
1936 — Hardy Cross method developed. Hardy Cross, a structural engineering pro-
fessor at the University of Illinois, publishes the Hardy Cross method for solving head
loss equations in complex networks. This method is widely used for manual calcula-
tions and will serve as the basis for early digital computer programs for pipe network
analysis.
1938 — Colebrook-White equation developed. Cyril Colebrook and Cedric White
of Imperial College in London build upon the work of Prandtl and his students to
develop the Colebrook-White equation for determining the Darcy-Weisbach f in com-
mercial pipes.
1940 — Hunter curves published. During the 1920s and ’30s, Roy Hunter of the
National Bureau of Standards conducts research on water use in a variety of build-
ings. His “fixture unit method” will become the basis for estimating building water
use, even though plumbing fixtures will change over the years. His probabilistic anal-
ysis captured the mathematics of the concept that the more fixtures in a building, the
less likely they are to be used simultaneously.
1944 — Moody diagram published. Lewis Moody of Princeton University publishes
the Moody diagram, which is essentially a graphical representation of the Colebrook-
White equation in the turbulent flow range and the Hagen-Poisseuille equation in the
laminar range. This diagram is especially useful because, at the time, no explicit solu-
tion exists for the Colebrook-White equation. Stanton had developed a similar chart
30 years earlier.
1950 — McIlroy network analyzer developed. The McIlroy network analyzer, an
electrical analog computer, is developed to simulate the behavior of water distribution
systems using electricity instead of water. The analyzer uses special elements called
“fluistors” to reproduce head loss in pipes, because in the Hazen-Williams equation,
head loss varies with flow raised to the 1.85 power, while normal resistors comply
with Ohm’s law, in which voltage drop varies linearly with current.
1950s — Earliest digital computers developed. The Electronic Numerical Integra-
tor and Computer (ENIAC) is assembled at the University of Pennsylvania. It con-
tains approximately 18,000 vacuum tubes and fills a 30 x 50 ft (9 x 15 m) room.
Digital computers such as the ENIAC and Univac show that computers can carry out
numerical calculations quickly, opening the door for programs to solve complex
hydraulic problems.
1956 — Push-on joint developed. The push-on pipe joint using a rubber gasket is
developed. This type of assembly helps speed the construction of piping.
1960s and ’70s — Earliest pipe network digital models created. With the coming
of age of digital computers and the establishment of the FORTRAN programming
language, researchers at universities begin to develop pipe network models and make
them available to practicing engineers. Don Wood at the University of Kentucky, Al
Fowler at the University of British Columbia, Roland Jeppson of Utah State Univer-
sity, Chuck Howard and Uri Shamir at MIT, and Simsek Sarikelle at the University of
Akron all write pipe network models.
Figure 1.8
A computer punch
card
1963 — First U.S. PVC pipe standards. The National Bureau of Standards accepts
CS256-63 “Commercial Standard for PVC Plastic Pipes (SDR-PR and Class T),”
which is the first U.S. standard for polyvinyl chloride water pipe.
1963 — URISA is founded. The Urban and Regional Information Systems Associa-
tion is founded by Dr. Edgar Horwood. URISA becomes the premier organization for
the use and integration of spatial information technology to improve the quality of life
in urban and regional environments.
1960s and ’70s — Water system contamination. Chemicals that can result in health
problems when ingested or inhaled are dumped on the ground or stored in leaky
ponds because of lack of awareness of their environmental impacts. Over the years,
these chemicals will make their way into water distribution systems and lead to
alleged contamination of water systems in places like Woburn, Massachusetts; Phoe-
nix/Scottsdale, Arizona; and Dover Township, New Jersey. Water quality models of
distribution systems will be used to attempt to recreate the dosages of chemicals
received by customers. These situations lead to popular movies like A Civil Action
and Erin Brockovich.
1970s — Early attempts to optimize water distribution design. Dennis Lai and
John Schaake at MIT develop the first approach to optimize water system design.
Numerous papers will follow by researchers such as Arun Deb, Ian Goulter, Uri
Shamir, Downey Brill, Larry Mays, and Kevin Lansey.
1970s — Models become more powerful. Although the earliest pipe network models
could only solve steady-state equations for simple systems, the ’70s bring modeling
features such as pressure regulating valves and extended-period simulations.
1975 — Data files replace input cards. Modelers are able to remotely create data
files on time-share terminals instead of using punched cards.
1975 — AWWA C-900 approved. The AWWA approves its first standard for PVC
water distribution piping. C900 pipe is made to match old cast iron pipe outer diame-
ters.
1976 — Swamee-Jain equation published. Dozens of approximations to the
Colebrook-White equations have been published in an attempt to arrive at an explicit
equation that would give the same results without the need for an iterative solution.
Indian engineers P. K. Swamee and Akalnank Jain publish the most popular form of
these approximations. The use of an explicit equation results in faster numerical solu-
tions of pipe network problems.
1976 — Jeppson publishes Analysis of Flow in Pipe Networks. Roland Jeppson
authors the book Analysis of Flow in Pipe Networks, which presents a summary of the
numerical techniques used to solve network problems.
1980 — Personal computers introduced. Early personal computers make it possible
to move hydraulic analysis to desktop systems. Initially, these desktop models are
slow, but their power will grow exponentially over the next two decades.
Figure 1.9
Time-share
terminal
Early 1980s — Water Quality Modeling First Developed. The concept of modeling
water quality in distribution systems is first developed, and steady state formulations
are proposed by Don Wood at the University of Kentucky and USEPA researchers in
Cincinnati, Ohio.
1985 — “Battle of the Network Models.” A series of sessions is held at the ASCE
Water Resources Planning and Management Division Conference in Buffalo, New
York, where researchers are given a realistic system called “Anytown” and are asked
to optimize the design of that network. Comparison of results shows the strengths and
weaknesses of the various models.
1986 — Introduction of Dynamic Water Quality Models. At the AWWA Distribu-

tion System Symposium, three groups independently introduce dynamic water quality
models of distribution systems.
1988 — Gradient Algorithm. Ezio Todini and S. Pilati publish “A Gradient Algo-
rithm for the Analysis of Pipe Networks,” and R. Salgado, Todini, and P. O'Connell
publish “Comparison of the Gradient Method with some Traditional Methods of the
Analysis of Water Supply Distribution Networks.” The gradient algorithm serves as
the basis for the WaterCAD model.
1989 — AWWA holds specialty conference. AWWA holds the Computers and Auto-
mation in the Water Industry conference. This conference will later grow into the
popular IMTech event (Information Management and Technology).
1990s — Privatization of water utilities. The privatization of water utilities

increases significantly as other utilities experience a greater push toward deregulation.
1991 — Water Quality Modeling in Distribution Systems Conference. The

USEPA and the AWWA Research Foundation bring together researchers from around
the world for a two-day meeting in Cincinnati. This meeting is a milestone in the
establishment of water quality modeling as a recognized tool for investigators.
1991 — GPS technology becomes affordable. The cost of global positioning sys-
tems (GPS) drops to the point where a GPS can be an economical tool for determining
coordinates of points in hydraulic models.
1993 — Introduction of water quality modeling tool. Water quality modeling

comes of age with the development of EPANET by Lewis Rossman of the USEPA.
Intended as a research tool, EPANET provides the basis for several commercial-grade
models.
1990 through present. Several commercial software developers release water distri-
bution modeling packages. Each release brings new enhancements for data manage-
ment and new abilities to interoperate with other existing computer systems.
2001 — Automated calibration. Automated calibration of distribution models

moves from being a research tool to a standard modeling feature with the use of
Genetic Algorithms.
2001 — Security awareness. Water system security increases in importance and util-
ities realize the value of water quality modeling as a tool for protecting a water system.
2002 — Integration with GIS. Water modeling and GIS software become highly
integrated with the release of WaterGEMS, software that combines the functionality
of both tools.
What Next?
Predicting the future is difficult, especially with rapidly changing fields such as the
software industry. However, there are definite trends as data sharing continues to gain
popularity, modeling spreads into operations, and automated design tools add to the
modeler’s arsenal.
The next logical question is, “When will network models eliminate the need for engi-
neers?” The answer is, never. Though a word processor can reduce the number of
spelling and grammar mistakes, it cannot write a best-selling novel. Even as technol-
ogy advances, an essential need still exists for living, breathing, thinking human
beings. A network model is just another tool (albeit a very powerful, multi-purpose
tool) for an experienced engineer or technician. It is still the responsibility of the user
to understand the real system, understand the model, and make decisions based on
sound engineering judgement.
REFERENCES
Mays, L. W. (2000). “Introduction.” Water Distribution System Handbook, Mays, L. W., ed., McGraw Hill,
New York, New York.
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
2
Modeling Theory
Model-based simulation is a method for mathematically approximating the behavior

of real water distribution systems. To effectively utilize the capabilities of distribution
system simulation software and interpret the results produced, the engineer or mod-
eler must understand the mathematical principles involved. This chapter reviews the
principles of hydraulics and water quality analysis that are frequently employed in
water distribution network modeling software.
2.1 FLUID PROPERTIES

Fluids can be categorized as gases or liquids. The most notable differences between
the two states are that liquids are far denser than gases, and gases are highly com-
pressible compared to liquids (liquids are relatively incompressible). The most impor-
tant fluid properties taken into consideration in a water distribution simulation are
specific weight, fluid viscosity, and (to a lesser degree) compressibility.
Density and Specific Weight

The density of a fluid is the mass of the fluid per unit volume. The density of water is
1.94 slugs/ft3 (1000 kg/m3) at standard pressure of 1 atm (1.013 bar) and standard tem-
perature of 32.0oF (0.0oC). A change in temperature or pressure will affect the density,
although the effects of minor changes are generally insignificant for water modeling
purposes.
The property that describes the weight of a fluid per unit volume is called specific
weight and is related to density by gravitational acceleration:
20 Modeling Theory Chapter 2
J = Ug (2.1)
where J = fluid specific weight (M/L2/T2)

U = fluid density (M/L3)
g = gravitational acceleration constant (L/T2)
The specific weight of water, J , at standard pressure and temperature is 62.4 lb/ft3
(9,806 N/m3).
Viscosity
Fluid viscosity is the property that describes the ability of a fluid to resist deformation
due to shear stress. For many fluids, most notably water, viscosity is a proportionality
factor relating the velocity gradient to the shear stress, as described by Newton’s law
of viscosity:
dV
W = P ------ (2.2)
dy
where W = shear stress (M/L/T2)

P = absolute (dynamic) viscosity (M/L/T)
dV
------- = time rate of strain (1/T)
dy
The physical meaning of this equation can be illustrated by considering the two paral-
lel plates shown in Figure 2.1. The space between the plates is filled with a fluid, and
the area of the plates is large enough that edge effects can be neglected. The plates are
separated by a distance y, and the top plate is moving at a constant velocity V relative
to the bottom plate. Liquids exhibit an attribute known as the no-slip condition, mean-
ing that they adhere to surfaces they contact. Therefore, if the magnitude of V and y
are not too large, then the velocity distribution between the two plates is linear.
From Newton’s second law of motion, for an object to move at a constant velocity, the
net external force acting on the object must equal zero. Thus, the fluid must be exert-
ing a force equal and opposite to the force F on the top plate. This force within the
fluid is a result of the shear stress between the fluid and the plate. The velocity at
which these forces balance is a function of the velocity gradient normal to the plate
and the fluid viscosity, as described by Newton’s law of viscosity.
Thick fluids, such as syrup and molasses, have high viscosities. Thin fluids, such as
water and gasoline, have low viscosities. For most fluids, the viscosity remains con-
stant regardless of the magnitude of the shear stress that is applied to it.
Returning to Figure 2.1, as the velocity of the top plate increases, the shear stresses in
the fluid increase at the same rate. Fluids that exhibit this property conform to New-
ton’s law of viscosity and are called Newtonian fluids. Water and air are examples of
Newtonian fluids. Some types of fluids, such as inks and sludge, undergo changes in
viscosity when the shear stress changes. Fluids exhibiting this type of behavior are
called pseudo-plastic fluids.
Section 2.1 Fluid Properties 21
Figure 2.1
Physical interpretation
of Newton’s law of
F viscosity
dy
dV
Relationships between the shear stress and the velocity gradient for typical Newtonian
and non-Newtonian fluids are shown in Figure 2.2. Since most distribution system
models are intended to simulate water, many of the equations used consider Newto-
nian fluids only.
Figure 2.2
Stress versus strain for
plastics and fluids
Elastic Solid
tic
Plas
eal
Id
t = Shear Stress
id
Flu
tic
-p las
udo
Pse
d
Flui
on ian
N ewt
Ideal Fluid
dV/dy
Viscosity is a function of temperature, but this relationship is different for liquids and
gases. In general, viscosity decreases as temperature increases for liquids, and viscos-
ity increases as temperature increases for gases. The temperature variation within
water distribution systems, however, is usually quite small, and thus changes in water
viscosity are considered negligible for this application. Generally, water distribution
system modeling software treats viscosity as a constant [assuming a temperature of
68oF (20oC)].
The viscosity derived in Equation 2.2 is referred to as the absolute viscosity (or
dynamic viscosity). For hydraulic formulas related to fluid motion, the relationship
between fluid viscosity and fluid density is often expressed as a single variable. This
relationship, called the kinematic viscosity, is expressed as follows:
Q = P
--- (2.3)
U
where Q = kinematic viscosity (L2/T)
Just as there are shear stresses between the plate and the fluid in Figure 2.1, there are
shear stresses between the wall of a pipe and the fluid moving through the pipe. The
higher the fluid viscosity, the greater the shear stresses that will develop within the
fluid, and, consequently, the greater the friction losses along the pipe. Distribution
system modeling software packages use fluid viscosity as a factor in estimating the
friction losses along a pipe’s length. Packages that can handle any fluid require the
viscosity and density to be input by the modeler, while models that are developed only
for water usually account for the appropriate value automatically.
Fluid Compressibility
Compressibility is a physical property of fluids that relates the volume occupied by a
fixed mass of fluid to its pressure. In general, gases are much more compressible than
liquids. An air compressor is a simple device that utilizes the compressibility of air to
store energy. The compressor is essentially a pump that intermittently forces air mole-
cules into the fixed volume tank attached to it. Each time the compressor turns on, the
mass of air, and therefore the pressure within the tank, increases. Thus a relationship
exists between fluid mass, volume, and pressure.
This relationship can be simplified by considering a fixed mass of a fluid. Compress-
ibility is then described by defining the fluid’s bulk modulus of elasticity:
dP
E v = – V ------- (2.4)
dV
where Ev = bulk modulus of elasticity (M/L/T2)

P = pressure (M/L/T2)
V = volume of fluid (L3)
All fluids are compressible to some extent. The effects of compression in a water dis-
tribution system are very small, and thus the equations used in hydraulic simulations
are based on the assumption that the liquids involved are incompressible. With a bulk
modulus of elasticity of 410,000 psi (2.83 u 106 kPa) at 68oF (20oC), water can safely
be treated as incompressible. For instance, a pressure change of over 2,000 psi
(1.379 u 104 kPa) results in only a 0.5 percent change in volume.
Section 2.2 Fluid Statics and Dynamics 23
Although the assumption of incompressibility is justifiable under most conditions,

certain hydraulic phenomena are capable of generating pressures high enough that the
compressibility of water becomes important. During field operations, a phenomenon
known as water hammer can develop due to extremely rapid changes in flow (when,
for instance, a valve suddenly closes, or a power failure occurs and pumps stop oper-
ating). The momentum of the moving fluid can generate pressures large enough that
fluid compression and pipe wall expansion can occur, which in turn causes destructive
transient pressure fluctuations to propagate throughout the network. Specialized net-
work simulation software is necessary to analyze these transient pressure effects. For
complete coverage of transient flow, see Chapter 13.
Vapor Pressure
Consider a closed container that is partly filled with water. The pressure in the con-
tainer is measured when the water is first added, and again after some time has
elapsed. These readings show that the pressure in the container increases during this
period. The increase in pressure is due to the evaporation of the water, and the result-
ing increase in vapor pressure above the liquid.
Assuming that temperature remains constant, the pressure will eventually reach a con-
stant value that corresponds to the equilibrium or saturation vapor pressure of water
at that temperature. At this point, the rates of evaporation and condensation are equal.
The saturation vapor pressure increases with increasing temperature. This relationship
demonstrates, for example, why the air in humid climates typically feels moister in
summer than in winter, and why the boiling temperature of water is lower at higher
elevations.
If a sample of water at a pressure of 1 atm and room temperature is heated to 212oF
(100oC), the water will begin to boil since the vapor pressure of water at that tempera-
ture is equal to 1 atm. In a similar vein, if water is held at a temperature of 68oF
(20oC), and the pressure is decreased to 0.023 atm, the water will also boil.
This concept can be applied to water distribution in cases in which the ambient pres-
sure drops very low. Pump cavitation occurs when the fluid being pumped flashes
into a vapor pocket and then quickly collapses. For this to happen, the pressure in the
pipeline must be equal to or less than the vapor pressure of the fluid. When cavitation
occurs, it sounds as if gravel is being pumped, and severe damage to pipe walls and
pump components can result. For complete coverage of cavitation, see Chapter 13.
2.2 FLUID STATICS AND DYNAMICS
Static Pressure
Pressure can be thought of as a force applied normal, or perpendicular, to a body that
is in contact with a fluid. In the English system of units, pressure is expressed in
pounds per square foot (lb/ft2), but the water industry generally uses lb/in2, typically
abbreviated as psi. In the SI system, pressure has units of N/m2, also called a Pascal.
However, because of the magnitude of pressures occurring in distribution systems,

pressure is typically reported in kilo-Pascals (kPa), or 1,000 Pascals.
Pressure varies with depth, as illustrated in Figure 2.3. For fluids at rest, the variation
of pressure over depth is linear and is called the hydrostatic pressure distribution.
P = hJ (2.5)
where P = pressure (M/L/T2)

h = depth of fluid above datum (L)
J = fluid specific weight (M/L2/T2)
Figure 2.3
Static pressure in a
standing water
column
Depth
Pressure = g(Depth)
This equation can be rewritten to find the height of a column of water that can be sup-
ported by a given pressure:
h = P
--- (2.6)
J
The quantity P/ J is called the pressure head, which is the energy resulting from water
pressure. Recognizing that the specific weight of water in English units is 62.4 lb/ft3, a
convenient conversion factor can be established for water as 1 psi = 2.31 ft (1 kPa =
0.102 m) of pressure head.
Example — Pressure Calculation. Consider the storage tank in Figure 2.4 in which the
water surface elevation is 120 ft above a pressure gage. The pressure at the base of the tank is due to
the weight of the column of water directly above it, and can be calculated as follows:
lb
62.4 -----3- 120ft
ft
P = Jh = ----------------------------------
-
2
in
144 2 ------
-
ft
P = 52 psi
Figure 2.4
Storage tank
120 ft
Pbase = 52 psi
Absolute Pressure and Gage Pressure. Pressure at a given point is due to

the weight of the fluid above that point. The weight of the earth’s atmosphere pro-
duces a pressure, referred to as atmospheric pressure. Although the actual atmo-
spheric pressure depends on elevation and weather, standard atmospheric pressure at
sea level is 1 atm (14.7 psi or 101 kPa).
Two types of pressure are commonly used in hydraulics: absolute pressure and gage
pressure. Absolute pressure is the pressure measured with absolute zero (a perfect
vacuum) as its datum, and gage pressure is the pressure measured with atmospheric
pressure as its datum. The two are related to one another as shown in Equation 2.7 and
as illustrated in Figure 2.5. Note that when a pressure gage located at the earth’s sur-
face is open to the atmosphere, it registers zero on its dial. If the gage pressure is neg-
ative (that is, the pressure is below atmospheric), then the negative pressure is called a
vacuum.
P abs = P gage + P atm (2.7)
where Pabs = absolute pressure (M/L/T2)

Pgage = gage pressure (M/L/T2)
Patm = atmospheric pressure (M/L/T2)
In most hydraulic applications, including water distribution systems analysis, gage
pressure is used. Using absolute pressure has little value, since doing so would simply
result in all the gage pressures being incremented by atmospheric pressure. Addition-
ally, gage pressure is often more intuitive because people do not typically consider
atmospheric effects when thinking about pressure.
Figure 2.5
Gage versus absolute
pressure
Pgage = 0 psi
Pabs = 14.6 psi
H = 20 ft
Water
Pgage = 8.7 psi

Pabs = 23.3 psi
Velocity and Flow Regime

The velocity profile of a fluid as it flows through a pipe is not constant across the
diameter. Rather, the velocity of a fluid particle depends on where the fluid particle is
located with respect to the pipe wall. In most cases, hydraulic models deal with the
average velocity in a cross-section of pipeline, which can be found by using the fol-
lowing formula:
Q
V = ---- (2.8)
A
where V = average fluid velocity (L/T)

Q = pipeline flow rate (L3/T)
A = cross-sectional area of pipeline (L2)
The cross-sectional area of a circular pipe can be directly computed from the diameter
D, so the velocity equation can be rewritten as:
4Q
V = ---------- (2.9)
2
SD
where D = diameter (L)

For water distribution systems in which diameter is measured in inches and flow is
measured in gallons per minute, the equation simplifies to:
Q
V = 0.41 ------2 (2.10)
D
where V = average fluid velocity (ft/s)

Q = pipeline flow rate (gpm)
D = diameter (in.)
Reynolds Number. In the late 1800s, an English scientist named Osborne Rey-
nolds conducted experiments on fluid passing through a glass tube. His experimental
setup looked much like the one in Figure 2.6 (Streeter, Wylie, and Bedford, 1998).
The experimental apparatus was designed to establish the flow rate through a long
glass tube (meant to simulate a pipeline) and to allow dye (from a smaller tank) to
flow into the liquid. He noticed that at very low flow rates, the dye stream remained
intact with a distinct interface between the dye stream and the fluid surrounding it.
Reynolds referred to this condition as laminar flow. At slightly higher flow rates, the
dye stream began to waver a bit, and there was some blurring between the dye stream
and the surrounding fluid. He called this condition transitional flow. At even higher
flows, the dye stream was completely broken up, and the dye mixed completely with
the surrounding fluid. Reynolds referred to this regime as turbulent flow.
When Reynolds conducted the same experiment using different fluids, he noticed that
the condition under which the dye stream remained intact not only varied with the
flow rate through the tube, but also with the fluid density, fluid viscosity, and the
diameter of the tube.
Figure 2.6
Experimental
apparatus used to
determine Reynolds
number
Based on experimental evidence gathered by Reynolds and dimensional analysis, a

dimensionless number can be computed and used to characterize flow regime. Con-
ceptually, the Reynolds number can be thought of as the ratio between inertial and vis-
cous forces in a fluid. The Reynolds number for full flowing circular pipes can be
found using the following equation:
Re = VDU
------------ = VD
-------- (2.11)
P Q
where Re = Reynolds number

D = pipeline diameter (L)
U = fluid density (M/L3)
P = absolute viscosity (M/L/T)
Q = kinematic viscosity (L2/T)
The ranges of the Reynolds number that define the three flow regimes are shown in
Table 2.1. The flow of water through municipal water systems is almost always turbu-
lent, except in the periphery where water demand is low and intermittent, and may
result in laminar and stagnant flow conditions.
Table 2.1 Reynolds number for various flow regimes

Flow Regime Reynolds Number
Laminar < 2000
Transitional 2000–4000
Turbulent > 4000
Velocity Profiles. Due to the shear stresses along the walls of a pipe, the velocity
in a pipeline is not uniform over the pipe diameter. Rather, the fluid velocity is zero at
the pipe wall. Fluid velocity increases with distance from the pipe wall, with the max-
imum occurring along the centerline of the pipe. Figure 2.7 illustrates the variation of
fluid velocity within a pipe, also called the velocity profile.
The shape of the velocity profile will vary depending on whether the flow regime is
laminar or turbulent. In laminar flow, the fluid particles travel in parallel layers or
lamina, producing very strong shear stresses between adjacent layers, and causing the
dye streak in Reynolds’ experiment to remain intact. Mathematically, the velocity
profile in laminar flow is shaped like a parabola as shown in Figure 2.7. In laminar
flow, the head loss through a pipe segment is primarily a function of the fluid viscos-
ity, not the internal pipe roughness.
Turbulent flow is characterized by eddies that produce random variations in the veloc-
ity profiles. Although the velocity profile of turbulent flow is more erratic than that of
laminar flow, the mean velocity profile actually exhibits less variation across the pipe.
The velocity profiles for both turbulent and laminar flows are shown in Figure 2.7.
Section 2.3 Energy Concepts 29
Figure 2.7
Velocity profiles for
v v different flow regimes
Uniform Velocity Profile Laminar Profile
Turbulent Profile
2.3 ENERGY CONCEPTS

Fluids possess energy in three forms. The amount of energy depends on the fluid’s
movement (kinetic energy), elevation (potential energy), and pressure (pressure
energy). In a hydraulic system, a fluid can have all three types of energy associated
with it simultaneously. The total energy associated with a fluid per unit weight of the
fluid is called head. The kinetic energy is called velocity head (V2/2g), the potential
energy is called elevation head (Z), and the internal pressure energy is called pressure
head (P/ J ). While typical units for energy are foot-pounds (Joules), the units of total
head are feet (meters).
2
H = Z+P V-
--- + ----- (2.12)
J 2g
where H = total head (L)

Z = elevation above datum (L)
V = velocity (L/T)
Each point in the system has a unique head associated with it. A line plotted of total
head versus distance through a system is called the energy grade line (EGL). The sum
of the elevation head and pressure head yields the hydraulic grade line (HGL), which
corresponds to the height that water will rise vertically in a tube attached to the pipe
and open to the atmosphere. Figure 2.8 shows the EGL and HGL for a simple pipe-
line.
In most water distribution applications, the elevation and pressure head terms are
much greater than the velocity head term. For this reason, velocity head is often
ignored, and modelers work in terms of hydraulic grades rather than energy grades.
Therefore, given a datum elevation and a hydraulic grade line, the pressure can be
determined as
P = J HGL – Z (2.13)
Figure 2.8
Energy and hydraulic
grade lines Head Loss, hL
2
V
Velocity Head, 2g
EGL
HGL
P
Pressure Head,
Flow
Elevation Head, Z
Datum
Energy Losses
Energy losses, also called head losses, are generally the result of two mechanisms:
• Friction along the pipe walls
• Turbulence due to changes in streamlines through fittings and appurtenances

Head losses along the pipe wall are called friction losses or head losses due to fric-
tion, while losses due to turbulence within the bulk fluid are called minor losses.
2.4 FRICTION LOSSES

When a liquid flows through a pipeline, shear stresses develop between the liquid and
the pipe wall. This shear stress is a result of friction, and its magnitude is dependent
on the properties of the fluid that is passing through the pipe, the speed at which it is
moving, the internal roughness of the pipe, and the length and diameter of the pipe.
Consider, for example, the pipe segment shown in Figure 2.9. A force balance on the
fluid element contained within a pipe section can be used to form a general expression
describing the head loss due to friction. Note the forces in action:
• Pressure difference between sections 1 and 2
• The weight of the fluid volume contained between sections 1 and 2
• The shear at the pipe walls between sections 1 and 2

Assuming the flow in the pipeline has a constant velocity (that is, acceleration is equal
to zero), the system can be balanced based on the pressure difference, gravitational
forces, and shear forces.
Section 2.4 Friction Losses 31
P 1 A 1 – P 2 A 2 – ALJ sin D – W o NL = 0 (2.14)
where P1 = pressure at section 1(M/L/T2)

A1 = cross-sectional area of section 1(L2)
P2 = pressure at section 2 (M/L/T2)
A2 = cross-sectional area of section 2 (L2)
A = average area between section 1 and section 2 (L2)
L = distance between section 1 and section 2 (L)
D = angle of the pipe to horizontal
W o = shear stress along pipe wall (M/L/T2)
N = perimeter of pipeline cross-section (L)
Figure 2.9
Free body diagram of
water flowing in an
inclined pipe
The last term on the left side of Equation 2.14 represents the friction losses along the
pipe wall between the two sections. By recognizing that sin( D ) = (Z2-Z1)/L, the equa-
tion for head loss due to friction can be rewritten to obtain the following equation.
(Note that the velocity head is not considered in this case because the pipe diameters,
and therefore the velocity heads, are the same.)
NL P P
h L = W o ------- = §© -----1- + Z 1·¹ – §© -----2- + Z 2·¹ (2.15)
JA J J
where hL = head loss due to friction (L)

Z1 = elevation of centroid of section 1 (L)
Z2 = elevation of centroid of section 2 (L)
Recall that the shear stresses in a fluid can be found analytically for laminar flow
using Newton’s law of viscosity. Shear stress is a function of the viscosity and veloc-
ity gradient of the fluid, the fluid specific weight (or density), and the diameter of the
pipeline. The roughness of the pipe wall is also a factor (that is, the rougher the pipe
wall, the larger the shear stress). Combining all these factors, it can be seen that
W o = F U P V D H (2.16)
where U = fluid density (M/L3)

P = absolute viscosity (M/L/T)
V = average fluid velocity (L/T)
D = diameter (L)
H = index of internal pipe roughness (L)
Darcy-Weisbach Formula
Using dimensional analysis, the Darcy-Weisbach formula was developed. The for-
mula is an equation for head loss expressed in terms of the variables listed in Equation
2.16, as follows (note that head loss is expressed with units of length):
2 2
LV 8fLQ
h L = f ----------- = ---------------- (2.17)
D2g gD S
5 2
where f = Darcy-Weisbach friction factor

The Darcy-Weisbach friction factor, f, is a function of the same variables as wall shear
stress (Equation 2.16). Again using dimensional analysis, a functional relationship for
the friction factor can be developed:
VDU H H
f = F § ------------ ----· = F § Re ----· (2.18)
© P D¹ © D¹

The Darcy-Weisbach friction factor is dependent on the velocity, density, and viscos-
ity of the fluid; the size of the pipe in which the fluid is flowing; and the internal
roughness of the pipe. The fluid velocity, density, viscosity, and pipe size are
expressed in terms of the Reynolds number. The internal roughness is expressed in
terms of a variable called the relative roughness, which is the internal pipe roughness
( H ) divided by the pipe diameter (D).
In the early 1930s, the German researcher Nikuradse performed an experiment that
would become fundamental in head loss determination (Nikuradse, 1932). He glued
uniformly sized sand grains to the insides of three pipes of different sizes. His experi-
ments showed that the curve of f versus Re is smooth for the same values of H /D.
Partly because of Nikuradse’s sand grain experiments, the quantity H is called the
equivalent sand grain roughness of the pipe. Table 2.2 provides values of H for
various materials.
Other researchers conducted experiments on artificially roughened pipes to generate

data describing pipe friction factors for a wide range of relative roughness values.
Table 2.2 Equivalent sand grain roughness for various pipe materials
Equivalent Sand Roughness, H

Material (ft) (mm)
Copper, brass 1x10-4 – 3x10-3 3.05x10-2 – 0.9
Wrought iron, steel 1.5x10-4 – 8x10-3 4.6x10-2 – 2.4
Asphalted cast iron 4x10-4 – 7x10-3 0.1 – 2.1
Galvanized iron -4
3.3x10 – 1.5x10 -2
0.102 – 4.6
Cast iron -4
8x10 – 1.8x10 -2
0.2 – 5.5
Concrete 10-3 – 10-2 0.3 – 3.0
Uncoated cast iron 7.4x10 -4
0.226
Coated cast iron 3.3x10 -4
0.102
Coated spun iron 1.8x10-4 5.6x10-2
Cement 1.3x10-3 – 4x10-3 0.4 – 1.2
Wrought iron 1.7x10 -4
5x10-2
Uncoated steel 9.2x10 -5
2.8x10-2
Coated steel 1.8x10-4 5.8x10-2
Wood stave 6x10-4 – 3x10-3 0.2 – 0.9
PVC 5x10 -6
1.5x10-3
Compiled from Lamont (1981), Moody (1944), and Mays (1999)
Colebrook-White Equation and the Moody Diagram. Numerous for-

mulas exist that relate the friction factor to the Reynolds number and relative rough-
ness. One of the earliest and most popular of these formulas is the Colebrook-
White equation:
1 = – 0.86 ln § ----------- 2.51 ·

H - +------------
----- (2.19)
f © 3.7D Re f ¹
The difficulty with using the Colebrook-White equation is that it is an implicit func-
tion of the friction factor (f is found on both sides of the equation). Typically, the
equation is solved by iterating through assumed values of f until both sides are equal.
The Moody diagram, shown in Figure 2.10, was developed from the Colebrook-White
equation as a graphical solution for the Darcy-Weisbach friction factor.
It is interesting to note that for laminar flow (low Re) the friction factor is a linear
function of the Reynolds number, while in the fully turbulent range (high H /D and
high Re) the friction factor is only a function of the relative roughness. This difference
occurs because the effect of roughness is negligible for laminar flow, while for very
turbulent flow the viscous forces become negligible.
Swamee-Jain Formula. Much easier to solve than the iterative Colebrook-
White formula, the formula developed by Swamee and Jain (1976) also approximates
the Darcy-Weisbach friction factor. This equation is an explicit function of the Rey-
nolds number and the relative roughness, and is accurate to within about one percent
of the Colebrook-White equation over a range of
3 8
4 u 10 d Re d 1 u 10 and
–6 –2
1 u 10 d H e D d 1 u 10
1.325
f = -------------------------------------------------- (2.20)
H 5.74 2
ln §© ------------ + ------------·¹
3.7D Re 0.9
Because of its relative simplicity and reasonable accuracy, most water distribution
system modeling software packages use the Swamee-Jain formula to compute the
friction factor.
Hazen-Williams
Another frequently used head loss expression, particularly in North America, is the
Hazen-Williams formula (Williams and Hazen, 1920; ASCE, 1992):
Cf L 1.852
h L = ---------------------------
1.852 4.87
Q (2.21)
C D
where hL = head loss due to friction (ft, m)

L = distance between sections 1 and 2 (ft, m)
C = Hazen-Williams C-factor
D = diameter (ft, m)
Q = pipeline flow rate (cfs, m3/s)
Cf = unit conversion factor (4.73 English, 10.7 SI)
The Hazen-Williams formula uses many of the same variables as Darcy-Weisbach,
but instead of using a friction factor, the Hazen-Williams formula uses a pipe carrying
capacity factor, C. Higher C-factors represent smoother pipes (with higher carrying
capacities) and lower C-factors describe rougher pipes. Table 2.3 shows typical C-
factors for various pipe materials, based on Lamont (1981).
Values of ( VD ) for water at 60 F (diameter in inches, velocity in ft./sec.)
0.1 0.2 0.4 0.6 1 2 4 6 10 20 40 60 100 200 400 600 1000 2000 4000 10,000
0.10
Section 2.4
64
0.09 Laminar flow, f = Re
0.08
0.05
0.07
0.06
0.03
0.05 0.02
2
0.04
hL
0.01
L V
D 2g
0.006
0.03
0.004
0.025
0.002
Relative roughness, / D
0.020 0.001
Friction factor, f =
0.0006
ft. mm 0.0004
0.015 Riveted steel 0.003 - 0.03 0.9 - 9
Concrete 0.001- 0.01 0.3 - 3 0.0002
Smooth
Wood stave 0.0006 - 0.003 0.18 - 0.9
Cast iron 0.00085 0.25 pipes
Galvanized iron 0.0005 0.15
Asphalted cast iron 0.0004 0.12
Steel or wrought iron 0.00015 0.045 0.00005
0.010 Drawn Tubing 0.000005 0.0015
0.009
0.008 8
10 3 2(103) 4 6 8 104 2(104) 4 6 8 105 2(105) 4 6 8 106 2(106) 4 6 8 107 2(107) 4 6 8 10
VD
Reynolds number, Re = v
From L. F. Moody, “Friction Factors for Pipe Flow,” Trans. A.S.M.E., Vol. 66, 1944, used with permission.
Friction Losses
Moody diagram
Figure 2.10
35
Lamont found that it was not possible to develop a single correlation between pipe
age and C-factor and that, instead, the decrease in C-factor also depended heavily on
the corrosiveness of the water being carried. He developed four separate “trends” in
carrying capacity loss depending on the “attack” of the water on the pipe. Trend 1,
slight attack, corresponded to water that was only mildly corrosive. Trend 4, severe
attack, corresponded to water that would rapidly attack cast iron pipe. As can be seen
from Table 2.3, the extent of attack can significantly affect the C-factor. Testing pipes
to determine the loss of carrying capacity is discussed further on page 196.
Table 2.3 C-factors for various pipe materials

C-factor Values for Discrete Pipe Diameters
1.0 in. 3.0 in. 6.0 in. 12 in. 24 in. 48 in.
Type of Pipe
(2.5 cm) (7.6 cm) (15.2 cm) (30 cm) (61 cm) (122 cm)
Uncoated cast iron - smooth and 121 125 130 132 134
new
Coated cast iron - smooth and 129 133 138 140 141
new
30 years old
Trend 1 - slight attack 100 106 112 117 120
Trend 2 - moderate attack 83 90 97 102 107
Trend 3 - appreciable 59 70 78 83 89
attack
Trend 4 - severe attack 41 50 58 66 73
60 years old
attack
100 years old
attack
Miscellaneous
Newly scraped mains 109 116 121 125 127
Newly brushed mains 97 104 108 112 115
Coated spun iron - smooth and 137 142 145 148 148
new
Old - take as coated cast iron
of same age
Galvanized iron - smooth and 120 129 133
new
Wrought iron - smooth and new 129 137 142
Coated steel - smooth and new 129 137 142 145 148 148
Uncoated steel - smooth and 134 142 145 147 150 150
new
Table 2.3 (cont.) C-factors for various pipe materials

C-factor Values for Discrete Pipe Diameters
1.0 in. 3.0 in. 6.0 in. 12 in. 24 in. 48 in.
Type of Pipe
(2.5 cm) (7.6 cm) (15.2 cm) (30 cm) (61 cm) (122 cm)
Coated asbestos cement - clean 147 149 150 152
Uncoated asbestos cement - 142 145 147 150
clean
Spun cement-lined and spun 147 149 150 152 153
bitumen- lined - clean
Smooth pipe (including lead, 140 147 149 150 152 153
brass, copper, polyethylene,
and PVC) - clean
PVC wavy - clean 134 142 145 147 150 150
Concrete - Scobey
Class 1 - Cs = 0.27; clean 69 79 84 90 95
Class 2 - Cs = 0.31; clean 95 102 106 110 113
Class 3 - Cs = 0.345; clean 109 116 121 125 127
Class 4 - Cs = 0.37; clean 121 125 130 132 134
Best - Cs = 0.40; clean 129 133 138 140 141
Tate relined pipes - clean 109 116 121 125 127
Prestressed concrete pipes - 147 150 150
clean
Lamont (1981)
From a purely theoretical standpoint, the C-factor of a pipe should vary with the flow
velocity under turbulent conditions. Equation 2.22 can be used to adjust the C-factor
for different velocities, but the effects of this correction are usually minimal. A two-
fold increase in the flow velocity correlates to an apparent five percent decrease in the
roughness factor. This difference is usually within the error range for the roughness
estimate in the first place, so most engineers assume the C-factor remains constant
regardless of flow (Walski, 1984). However, if C-factor tests are done at very high
velocities (i.e., >10 ft/s), then a significant error can result when the resulting C-
factors are used to predict head loss at low velocities.
V o 0.081
C = C o § ------· (2.22)
© V¹
where C = velocity adjusted C-factor

Co = reference C-factor
Vo = reference value of velocity at which C0 was determined (L/T)
Manning Equation
Another head loss expression more typically associated with open channel flow is the
Manning equation:
2
C f L nQ
h L = -----------------------
- (2.23)
5.33
D
where n = Manning roughness coefficient

As with the previous head loss expressions, the head loss computed using Manning
equation is dependent on the pipe length and diameter, the discharge or flow through
the pipe, and a roughness coefficient. In this case, a higher value of n represents a
higher internal pipe roughness. Table 2.4 provides typical Manning’s roughness coef-
ficients for commonly used pipe materials.
Table 2.4 Manning’s roughness values

Manning Manning
Material Material
Coefficient Coefficient
Asbestos cement .011 Corrugated metal .022
Brass .011 Galvanized iron .016
Brick .015 Lead .011
Cast iron, new .012 Plastic .009
Concrete Steel
Steel forms .011 Coal-tar enamel .010
Wooden forms .015 New unlined .011
Centrifugally spun .013 Riveted .019
Copper .011 Wood stave .012
Comparison of Friction Loss Methods

Most hydraulic models have features that allow the user to select from the Darcy-
Weisbach, Hazen-Williams, or Manning head loss formulas, depending on the nature
of the problem and the user’s preferences.
The Darcy-Weisbach formula is a more physically-based equation, derived from the
basic governing equations of Newton’s Second Law. With appropriate fluid viscosi-
ties and densities, Darcy-Weisbach can be used to find the head loss in a pipe for any
Newtonian fluid in any flow regime.
The Hazen-Williams and Manning formulas, on the other hand, are empirically-based
expressions (meaning that they were developed from experimental data), and gener-
ally only apply to water under turbulent flow conditions.
The Hazen-Williams formula is the predominant equation used in the United States,
and Darcy-Weisbach is predominant in Europe. The Manning formula is not typically
used for water distribution modeling; however, it is sometimes used in Australia.
Table 2.5 presents these three equations in several common unit configurations. These
equations solve for the friction slope (Sf), which is the head loss per unit length of
pipe.
Section 2.5 Minor Losses 39
Table 2.5 Friction loss equations in typical units

Equation Q (m3/s); D (m) Q (cfs); D (ft) Q (gpm); D (in.)
Darcy-Weisbach 2 2 2
S f = 0.083f Q S f = 0.025f Q S f = 0.031f Q
---------------------- ---------------------- ----------------------
5 5 5
D D D
Hazen-Williams 10.7- § Q 1.852 4.73- § Q 1.852 10.5- § Q 1.852
S f = ----------- ----· S f = ----------- ----· S f = ----------- ----·
4.87 © C¹ 4.87 © C¹ 4.87 © C¹
D D D
Manning 2 2 2
S f = 10.3 nQ S f = 4.66 nQ S f = 13.2 nQ
------------------------- ------------------------- -------------------------
5.33 5.33 5.33
D D D
Compiled from ASCE (1975) and ASCE/WEF (1982)
2.5 MINOR LOSSES

Head losses also occur at valves, tees, bends (see Figure 2.11), reducers, and other
appurtenances within the piping system. These losses, called minor losses, are due to
turbulence within the bulk flow as it moves through fittings and bends. Figure 2.12
illustrates the turbulent eddies that develop within the bulk flow as it travels through a
valve and a 90-degree bend.
Figure 2.11
48-in. elbow fitting
Head loss due to minor losses can be computed by multiplying a minor loss coeffi-
cient by the velocity head, as shown in Equation 2.24.
Figure 2.12
Valve and bend cross-
sections generating
minor losses
2 2
V Q
h m = K L ------ = K L ------------2 (2.24)
2g 2gA
where hm = head loss due to minor losses (L)

KL = minor loss coefficient
V = velocity (L/T)
A = cross-sectional area (L2)
Q = flow rate (L3/T)
Minor loss coefficients are found experimentally, and data are available for many dif-
ferent types of fittings and appurtenances. Table 2.6 provides a list of minor loss coef-
ficients associated with several of the most commonly used fittings. More thorough
treatments of minor loss coefficients can be found in Crane (1972), Miller (1978), and
Idelchik (1999).
For water distribution systems, minor losses are generally much smaller than the head
losses due to friction (hence the term "minor" loss). For this reason, many modelers
frequently choose to neglect minor losses. In some cases, however, such as at pump
stations or valve manifolds where there may be more fittings and higher velocities,
minor losses can play a significant role in the piping system under consideration.
Like pipe roughness coefficients, minor head loss coefficients will vary somewhat
with velocity. For most practical network problems, however, the minor loss coeffi-
cient is treated as constant.
Valve Coefficient
Most valve manufacturers can provide a chart of percent opening versus valve coeffi-
cent (Cv), which can be related to the minor loss (KL) by using Equation 2.25.
Section 2.5 Minor Losses 41
Table 2.6 Minor loss coefficients

Fitting KL Fitting KL
Pipe entrance o
90 smooth bend
Bellmouth 0.03-0.05 Bend radius/D = 4 0.16-0.18
Rounded 0.12-0.25 Bend radius/D = 2 0.19-0.25
Sharp-edged 0.50 Bend radius/D = 1 0.35-0.40
Projecting 0.78 Mitered bend
Contraction – sudden T = 15o 0.05
D2/D1=0.80 0.18 T = 30o 0.10
D2/D1=0.50 0.37 T = 45o 0.20
D2/D1=0.20 0.49 T = 60o 0.35
Contraction – conical T = 90o 0.80
D2/D1=0.80 0.05 Tee

D2/D1=0.50 0.07 Line flow 0.30-0.40
D2/D1=0.20 0.08 Branch flow 0.75-1.80
Expansion – sudden Tapping T Branch
D2/D1=0.80 0.16 d = tapping hole diameter 1.97/(d/D)4
D = main line diameter
D2/D1=0.50 0.57 Cross
D2/D1=0.20 0.92 Line flow 0.50
Expansion – conical Branch flow 0.75
D2/D1=0.80 0.03 o
45 Wye
D2/D1=0.50 0.08 Line flow 0.30
D2/D1=0.20 0.13 Branch flow 0.50
Gate valve – open 0.39 Check valve – conventional 4.0
3/4 open 1.10 Check valve – clearway 1.5
1/2 open 4.8 Check valve – ball 4.5
1/4 open 27 Cock – straight through 0.5
Globe valve – open 10 Foot valve – hinged 2.2
Angle valve – open 4.3 Foot valve – poppet 12.5
Butterfly valve – open 1.2
Walski (1984)
4 2
KL = Cf D e Cv (2.25)
where D = diameter (in., m)

Cv = valve coefficient [gpm/(psi)0.5, (m3/s)/(kPa)0.5]
Cf = unit conversion factor (880 English, 1.22 SI)
Equivalent Pipe Length

Rather than including minor loss coefficients directly, a modeler may choose to adjust
the modeled pipe length to account for minor losses by adding an equivalent length of
pipe for each minor loss. Given the minor loss coefficient for a valve or fitting, the
equivalent length of pipe to give the same head loss can be calculated as:
KL D
L e = ----------
- (2.26)
f
where Le = equivalent length of pipe (L)

D = diameter of equivalent pipe (L)
f = Darcy-Weisbach friction factor
The practice of assigning equivalent pipe lengths was typically used when hand calcu-
lations were more common because it could save time in the overall analysis of a
pipeline. With modern computer modeling techniques, this is no longer a widespread
practice. Because it is now so easy to use minor loss coefficients directly within a
hydraulic model, the process of determining equivalent lengths is actually less effi-
cient. In addition, use of equivalent pipe lengths can unfavorably affect the travel time
predictions that are important in many water quality calculations.
2.6 RESISTANCE COEFFICIENTS

Many related expressions for head loss have been developed. They can be mathemati-
cally generalized with the introduction of a variable referred to as a resistance coeffi-
cient. This format allows the equation to remain essentially the same regardless of
which friction method is used, making it ideal for hydraulic modeling.
z
hL = KP Q (2.27)
where hL = head loss due to friction (L)

KP = pipe resistance coefficient (Tz/L3z - 1)
z = exponent on flow term
Equations for computing KP with the various head loss methods are given next.
Darcy-Weisbach
L -
K P = f ----------------
z
(2.28)
2gA D

L = length of pipe (L)
D = pipe diameter (L)
A = cross-sectional area of pipeline (L2)
z = 2
Section 2.6 Resistance Coefficients 43
Hazen-Williams
Cf L
K P = ------------------
- (2.29)
z 4.87
CD
where KP = pipe resistance coefficient (sz/ft3z-1, sz/m3z-1)

L = length of pipe (ft, m)
C = C-factor with velocity adjustment
z = 1.852
D = pipe diameter (ft, m)
Manning
z
C f Ln
K P = -------------
- (2.30)
5.33
D
where n = Manning’s roughness coefficient

z =2
Cf = unit conversion factor [4.64 English, 10.3 SI (ASCE/WEF, 1982)]
Minor Losses
A resistance coefficient can also be defined for minor losses, as shown in the equation
below. Like the pipe resistance coefficient, the resistance coefficient for minor losses
is a function of the physical characteristics of the fitting or appurtenance and
the discharge.
2
hm = KM Q (2.31)
where hm = head loss due to minor losses (L)

KM = minor loss resistance coefficient (T2/L5)
Solving for the minor loss resistance coefficient by substituting Equation 2.24
results in:
¦ KL
K M = -------------- (2.32)
2
2gA
where ¦ KL = sum of individual minor loss coefficients

2.7 ENERGY GAINS – PUMPS

On many occasions, energy needs to be added to a hydraulic system to overcome ele-
vation differences, friction losses, and minor losses. A pump is a device to which
mechanical energy is applied and transferred to the water as total head. The head
added is called pump head and is a function of the flow rate through the pump. The
following discussion is oriented toward centrifugal pumps because they are the most
frequently used pumps in water distribution systems. Additional information about
pumps can be found in Bosserman (2000), Hydraulic Institute Standards (2000),
Karassik (1976), and Sanks (1998).
Pump Head-Discharge Relationship

The relationship between pump head and pump discharge is given in the form of a
head versus discharge curve (also called a head characteristic curve) similar to the
one shown in Figure 2.13. This curve defines the relationship between the head that
the pump adds and the amount of flow that the pump passes. The pump head versus
discharge relationship is nonlinear, and as one would expect, the more water the pump
passes, the less head it can add. The head that is plotted in the head characteristic
curve is the head difference across the pump, called the total dynamic head (TDH).
This curve must be described as a mathematical function to be used in a hydraulic
simulation. Some models fit a polynomial curve to selected data points, but a more
common approach is to describe the curve by using a power function in the following
form:
m
hP = ho – c QP (2.33)
where hP = pump head (L)

ho = cutoff (shutoff) head (pump head at zero flow) (L)
QP = pump discharge (L3/T)
c, m = coefficients describing pump curve shape
More information on pump performance testing is available in Chapter 5 (see page
197).
Affinity Laws for Variable-Speed Pumps. A centrifugal pump’s charac-
teristic curve is fixed for a given motor speed and impeller diameter, but it can be
determined for any speed and any diameter by applying relationships called the affin-
ity laws. For variable-speed pumps, these affinity laws are presented as follows:
Q P1 e Q P2 = n 1 e n 2 (2.34)
2
h P1 e h P2 = n 1 e n 2 (2.35)
where QP1 = pump flow at speed 1 (L3/T)

n1 = pump speed 1 (1/T)
hP1 = pump head at speed 1 (L)
Section 2.7 Energy Gains – Pumps 45
Figure 2.13
Pump head
200.0 characteristic curve
Shutoff Head
150.0
Design Point
Head ,ft
100.0
Maximum Flow
50.0
0.0
0.0 250 500 750 1000 1250 1500 1750
Flow, gpm
Thus, pump discharge rate is directly proportional to pump speed, and pump dis-
charge head is proportional to the square of the speed. Using this relationship, once
the pump curve at any one speed is known, then the curve at another speed can be pre-
dicted. Figure 2.14 illustrates the affinity laws for variable-speed pumps where the
line through the pump head characteristic curves represents the locus of best effi-
ciency points.
Inserting Equations 2.34 and 2.35 into Equation 2.33 and solving for h gives a general
equation for adjusting pump head curves for speed:
2 2–m m
h P2 = n h o – cn Q P2 (2.36)
where n = n2/n1
System Head Curves

The purpose of a pump is to overcome elevation differences and head losses due to
pipe friction and fittings. The amount of head the pump must add to overcome eleva-
tion differences is dependent on system characteristics and topology (and independent
of the pump discharge rate), and is referred to as static head or static lift. Friction and
minor losses, however, are highly dependent on the rate of discharge through the
pump. When these losses are added to the static head for a series of discharge rates,
the resulting plot is called a system head curve (see Figure 2.15).
Figure 2.14
Relative speed factors
for variable-speed
pumps
Figure 2.15
A family of system
head curves
Section 2.7 Energy Gains – Pumps 47
The pump characteristic curve is a function of the pump and independent of the sys-
tem, while the system head curve is dependent on the system and is independent of
the pump. Unlike the pump curve, which is fixed for a given pump at a given speed,
the system head curve is continually sliding up and down as tank water levels change
and demands change. Rather than there being a unique system head curve, a family of
system head curves forms a band on the graph.
For the case of a single pipeline between two points, the system head curve can be
described in equation form as follows:
z 2
H = hl + ¦ KP Q + ¦ KM Q (2.37)
where H = total head (L)

hl = static lift (L)
KP = pipe resistance coefficient (Tz/L3z - 1)
Q = pipe discharge (L3/T)
z = coefficient
KM = minor loss resistance coefficient (T2/L5)
Thus, the head losses and minor losses associated with each segment of pipe are
summed along the total length of the pipeline. When the system is more complex, the
interdependencies of the hydraulic network make it impossible to write a single equa-
tion to describe a point on the system curve. In these cases, hydraulic analysis using a
hydraulic model may be needed. It is helpful to visualize the hydraulic grade line as
increasing abruptly at a pump and sloping downward as the water flows through pipes
and valves (see Figure 2.16).
Figure 2.16
Schematic of
hydraulic grade line
for a pumped system
Pump Operating Point

When the pump head discharge curve and the system head curve are plotted on the
same axes (as shown in Figure 2.17), only one point lies on both the pump character-
istic curve and the system head curve. This intersection defines the pump operating
point, which represents the discharge that will pass through the pump and the head
that the pump will add. This head is equal to the head needed to overcome the static
head and other losses in the system.
Figure 2.17
System operating
point
Pump Head
Curve
Pump Operating Point

Head, ft
Head Losses
Curve
m Head
Syste
Static Lift
Flow, gpm
Other Uses of Pump Curves

In addition to the pump head-discharge curve, other curves representing pump behav-
ior describe power, water horsepower, and efficiency (see Figure 2.18), and are dis-
cussed further in Chapter 3 (see page 95) and Chapter 5 (see page 197). Since utilities
want to minimize the amount of energy necessary for system operation, the engineer
should select pumps that run as efficiently as possible. Pump operating costs are dis-
cussed further in Chapter 10 (see page 436).
Another issue when designing a pump is the net positive suction head (NPSH)
required (see page 324). NPSH is the head that is present at the suction side of the
pump. Each pump requires that the available NPSH exceed the required NPSH to
ensure that local pressures within the pump do not drop below the vapor pressure of
the fluid, causing cavitation. As discussed on page 23, cavitation is essentially a boil-
ing of the liquid within the pump, and it can cause tremendous damage. The NPSH
required is unique for each pump model, and is a function of flow rate. The use of a
calibrated hydraulic model in determining available net positive suction head is dis-
cussed further on page 324.
Section 2.8 Network Hydraulics 49
Figure 2.18
Pump efficiency curve
2.8 NETWORK HYDRAULICS

In networks of interconnected hydraulic elements, every element is influenced by
each of its neighbors; the entire system is interrelated in such a way that the condition
of one element must be consistent with the condition of all other elements. Two con-
cepts define these interconnections:
• Conservation of mass
• Conservation of energy
Conservation of Mass
The principle of conservation of mass (shown in Figure 2.19) dictates that the fluid
mass entering any pipe will be equal to the mass leaving the pipe (since fluid is typi-
cally neither created nor destroyed in hydraulic systems). In network modeling, all
outflows are lumped at the nodes or junctions.
¦ Qi – U = 0 (2.38)
pipes
where Qi = inflow to node in i-th pipe (L3/T)

U = water used at node (L3/T)
Figure 2.19
Conservation of mass
principle U
Q1 Q3
Q2
Note that for pipe outflows from the node, the value of Q is negative.
When extended-period simulations are considered, water can be stored and
withdrawn from tanks, thus a term is needed to describe the accumulation of water at
certain nodes:
¦ Q i – U – dS
------ = 0 (2.39)
dt
pipes
dS
where ------ = change in storage (L3/T)
dt
The conservation of mass equation is applied to all junction nodes and tanks in a net-
work, and one equation is written for each of them.
Conservation of Energy
The principle of conservation of energy dictates that the difference in energy between
two points must be the same regardless of the path that is taken (Bernoulli, 1738). For
convenience within a hydraulic analysis, the equation is written in terms of head as
follows:
2 2
P1 V1 P2 V2
1 + ------ + ------ + ¦ h P = Z 2 + ------ + ------ + ¦ h L + ¦ h m (2.40)
J 2g J 2g
where Z = elevation (L)

V = velocity (L/T)
Section 2.8 Network Hydraulics 51

hP = head added at pumps (L)
hL = head loss in pipes (L)
hm = head loss due to minor losses (L)
Thus the difference in energy at any two points connected in a network is equal to the
energy gains from pumps and energy losses in pipes and fittings that occur in the path
between them. This equation can be written for any open path between any two
points. Of particular interest are paths between reservoirs or tanks (where the differ-
ence in head is known), or paths around loops because the changes in energy must
sum to zero, as illustrated in Figure 2.20.
Figure 2.20
The sum of head
2’ Loss
losses around a pipe
loop is equal to zero
A B
3’
ss
Lo
Lo
ss
1’
A to B to C to A = 0
+ 2’ + 1’ - 3’ = 0
Solving Network Problems

Real water distribution systems do not consist of a single pipe and cannot be
described by a single set of continuity and energy equations. Instead, one continuity
equation must be developed for each node in the system, and one energy equation
must be developed for each pipe (or loop), depending on the method used. For real
systems, these equations can number in the thousands.
The first systematic approach for solving these equations was developed by Hardy
Cross (1936). The invention of digital computers, however, allowed more powerful
numerical techniques to be developed. These techniques set up and solve the system
of equations describing the hydraulics of the network in matrix form. Because the
energy equations are nonlinear in terms of flow and head, they cannot be solved
directly. Instead, these techniques estimate a solution and then iteratively improve it
until the difference between solutions falls within a specified tolerance. At this point,
the hydraulic equations are considered solved.
Some of the methods used in network analysis are described in Bhave (1991); Lansey
and Mays (2000); Larock, Jeppson, and Watters (1999); and Todini and Pilati (1987).
2.9 WATER QUALITY MODELING

Water quality modeling is a direct extension of hydraulic network modeling and can
be used to perform many useful analyses. Developers of hydraulic network simulation
models recognized the potential for water quality analysis and began adding water
quality calculation features to their models in the mid 1980s. Transport, mixing, and
decay are the fundamental physical and chemical processes typically represented in
water quality models. Water quality simulations also use the network hydraulic solu-
tion as part of their computations. Flow rates in pipes and the flow paths that define
how water travels through the network are used to determine mixing, residence times,
and other hydraulic characteristics affecting disinfectant transport and decay. The
results of an extended period hydraulic simulation can be used as a starting point in
performing a water quality analysis.
The equations describing transport through pipes, mixing at nodes, chemical forma-
tion and decay reactions, and storage and mixing in tanks are adapted from Grayman,
Rossman, and Geldreich (2000). Additional information on water quality models can
be found in Clark and Grayman (1998).
Transport in Pipes
Most water quality models make use of one-dimensional advective-reactive transport
to predict the changes in constituent concentrations due to transport through a pipe,
and to account for formation and decay reactions. Equation 2.41 shows concentration
within a pipe i as a function of distance along its length (x) and time (t).
wC i Q i wC i
-------- = ----- -------- + T C i i = 1...P (2.41)
wt A i wx
where Ci = concentration in pipe i (M/L3)

Qi = flow rate in pipe i (L3/T)
Ai = cross-sectional area of pipe i (L2)
T C i = reaction term (M/L3/T)
Equation 2.41 must be combined with two boundary condition equations (concentra-
tion at x = 0 and t = 0) to obtain a solution. Solution methods are described later in this
section.
The equation for advective transport is a function of the flow rate in the pipe divided
by the cross-sectional area, which is equal to the mean velocity of the fluid. Thus, the
bulk fluid is transported down the length of the pipe with a velocity that is directly
proportional to the average flow rate. The equation is based on the assumption that
longitudinal dispersion in pipes is negligible and that the bulk fluid is completely
mixed (a valid assumption under turbulent conditions). Furthermore, the equation can
also account for the formation or decay of a substance during transport with the sub-
stitution of a suitable equation into the reaction term. Such an equation will be devel-
oped later. First, however, the nodal mixing equation is presented.
Section 2.9 Water Quality Modeling 53
Mixing at Nodes
Water quality simulation uses a nodal mixing equation to combine concentrations
from individual pipes described by the advective transport equation, and to define the
boundary conditions for each pipe as mentioned previously. The equation is written
by performing a mass balance on concentrations entering a junction node.
§ ·
¨ ¦ Q i C i n + U j¸
i
¨ i IN j ¸
¨ C OUT j = ------------------------------------------
-¸ (2.42)
¨ ¸
¨ ¦ Qi ¸
© i OUT j ¹
where COUTj = concentration leaving the junction node j (M/L3)

OUTj = set of pipes leaving node j
INj = set of pipes entering node j
Qi = flow rate entering the junction node from pipe i (L3/T)
C i n = concentration entering junction node from pipe i (M/L3)
i
Uj = concentration source at junction node j (M/T)

The nodal mixing equation describes the concentration leaving a network node (either
by advective transport into an adjoining pipe or by removal from the network as a
demand) as a function of the concentrations that enter it. The equation describes the
flow-weighted average of the incoming concentrations. If a source is located at a junc-
tion, constituent mass can also be added and combined in the mixing equation with
the incoming concentrations. Figure 2.21 illustrates how the nodal mixing equation is
used at a pipe junction. Concentrations enter the node with pipe flows. The incoming
concentrations are mixed according to Equation 2.42, and the resulting concentration
is transported through the outgoing pipes modeled as demand leaving the system. The
nodal mixing equation assumes that incoming flows are completely and instanta-
neously mixed. The basis for the assumption is that turbulence occurs at the junction
node, which is usually sufficient for good mixing.
Mixing in Tanks
Pipes are sometimes connected to reservoirs and tanks as opposed to junction nodes.
Again, a mass balance of concentrations entering or leaving the tank or reservoir can
be performed.
dC Qi
--------k- = ----- C i np t – C k + T C k (2.43)
dt Vk
where Ck = concentration within tank or reservoir k (M/L3)

Qi = flow entering the tank or reservoir from pipe i (L3/T)
Vk = volume in tank or reservoir k (L3)
T (Ck) = reaction term (M/L3/T)
Figure 2.21
Nodal mixing
Demand
Equation 2.43 applies when a tank is filling. During a hydraulic time step in which the
tank is filling, the water entering from upstream pipes mixes with water that is already
in storage. If the concentrations are different, blending occurs. The tank mixing equa-
tion accounts for blending and any reactions that occur within the tank volume during
the hydraulic step. During a hydraulic step in which draining occurs, terms can be
dropped and the equation simplified.
dC
--------k- = T C k (2.44)
dt
Specifically, the dilution term can be dropped because it does not occur. Thus, the
concentration within the volume is subject only to chemical reactions. Furthermore,
the concentration draining from the tank becomes a boundary condition for the advec-
tive transport equation written for the pipe connected to it.
Equations 2.43 and 2.44 assume that concentrations within the tank or reservoir are
completely and instantaneously mixed. This assumption is frequently applied in water
quality models. There are, however, other useful mixing models for simulating flow
processes in tanks and reservoirs (Grayman et al., 1996). For example, contact basins
or clearwells designed to provide sufficient contact time for disinfectants are fre-
quently represented as simple plug-flow reactors using a “first in first out” (FIFO)
model. In a FIFO model, the first volume of water to enter the tank as inflow is the
first to leave as outflow.
If severe short-circuiting is occurring within the tank, a “last in first out” (LIFO)
model may be applied, in which the first volume entering the tank during filling is the
last to leave while draining. More complex tank mixing behavior can be captured
using more generalized “compartment” models. Compartment models have the ability
to represent mixing processes and time delays within tanks more accurately.
Many water distribution models offer a simple two-compartment model, as shown in
Figure 2.22 (Rossman, 2000). In this type of model, water enters or exits the tank
through a completely mixed inlet-outlet compartment, and if the first compartment is
completely full, the overflow is exchanged with a completely mixed second main
compartment. The inlet-outlet compartment can represent short-circuiting with the
last flow in becoming the first out (LIFO). The main compartment can represent a
stagnant or dead zone that will contain older water than the first compartment. The
only parameter for this model is the fraction of the total tank volume in the first
compartment. Selection of an appropriate value for this fraction is generally done by
comparing model results to field measurements of a tracer or chlorine residual.
Figure 2.22
Two-compartment
mixing model
Main Zone
Inlet -
Outlet
Figure 2.23 illustrates a more complex, three-compartment model for a tank with a
single pipe for filling and draining. This example illustrates a tank that is stratified.
New (good quality) water entering the tank occupies the first compartment and is then
transferred to a mixing compartment containing older water, and finally, to a third
dead-zone compartment that contains much older, poorer quality water. The model
simulates the exchange of water between different compartments, and in doing so,
mimics complex tank mixing dynamics. CompTank, a model that can be used to sim-
ulate the three-compartment model, as well as the other models described, is available
as part of an AWWA Research Foundation report (Grayman et al., 2000).
All the models mentioned in this section can be used to simulate a non-reactive (con-
servative) constituent, as well as decay or formation reactions for substances that react
over time. The models can also be used to represent tanks that either operate in fill
and draw mode or operate with simultaneous inflow and outflow.
Chemical Reaction Terms

Equations 2.42, 2.43, and 2.44 compose the linked system of first-order differential
equations solved by typical water quality simulation algorithms. This set of equations
and the algorithms for solving them can be used to model different chemical reactions
known to impact water quality in distribution systems. Chemical reaction terms are
present in Equations 2.43 and 2.44. Concentrations within pipes, storage tanks, and
reservoirs are a function of these reaction terms. After water leaves the treatment plant
and enters the distribution system, it is subject to many complex physical and chemi-
cal processes, some of which are poorly understood, and most of which are not mod-
eled. Three chemical processes that are frequently modeled, however, are bulk fluid
reactions, reactions that occur on a surface (typically the pipe wall), and formation
reactions involving a limiting reactant. First, an expression for bulk fluid reactions is
presented, and then a reaction expression that incorporates both bulk and pipe wall
reactions is developed.
Figure 2.23
Three-compartment
tank mixing model Air
Poor Quality
Poor Quality
Mixing Zone
Mixed
Good Quality
Good Quality
Fill Drain
Bulk Reactions. Bulk fluid reactions occur within the fluid volume and are a
function of constituent concentrations, reaction rate and order, and concentrations of
the formation products. A generalized expression for nth order bulk fluid reactions is
developed in Equation 2.45 (Rossman, 2000).
n
T C = r kC (2.45)
where T C = reaction term (M/L3/T)

k = reaction rate coefficient [(L3/M)n-1/T]
C = concentration (M/L3)
n = reaction rate order constant
Equation 2.45 is the generalized bulk reaction term most frequently used in water
quality simulation models. The rate expression accounts for only a single reactant
concentration, tacitly assuming that any other reactants (if they participate in the reac-
tion) are available in excess of the concentration necessary to sustain the reaction. The
sign of the reaction rate coefficient, k, signifies that a formation reaction (positive) or
a decay reaction (negative) is occurring. The units of the reaction rate coefficient
depend on the order of the reaction. The order of the reaction depends on the compo-
sition of the reactants and products that are involved in the reaction. The reaction rate
order is frequently determined experimentally.
Zero-, first-, and second-order decay reactions are commonly used to model chemical
processes that occur in distribution systems. Figure 2.24 is a conceptual illustration
showing the change in concentration versus time for these three most common reac-
tion rate orders. Using the generalized expression in Equation 2.45, these reactions
can be modeled by allowing n to equal 0, 1, or 2 and then performing a regression
analysis to experimentally determine the rate coefficient.
The most commonly used reaction model is the first order decay model. This has been
applied to chlorine decay, radon decay, and other decay processes. A first order decay
is equivalent to an exponential decay, represented by Equation 2.46.
– kt
Ct = Co e (2.46)
where Ct = concentration at time t (M/L3)

Co = initial concentration (at time zero)
k = reaction rate (1/T)
For first order reactions, the units of k are (1/T) with values generally expressed in 1/
days or 1/hours. Another way of expressing the speed of the reaction is the concept of
half-life that is frequently used when describing the decay rate for radioactive materi-
als. The half-life is the time it takes for the concentration of a substance to decrease to
50 percent of its original concentration. For example, the half-life of radon is approx-
imately 3.8 days, and the half-life of chlorine can vary from hours to many days. The
relationship between the decay rate, k, and half-life is easily calculated by solving
Equation 2.45 for the time t when Ct/Co is equal to a value of 0.5. This results in
Equation 2.47.
0.693
T = – ------------- (2.47)
k
For example, if the decay rate k is –1.0, the half-life is 0.693 days.
Figure 2.24
Conceptual
Conservative
illustration of
concentration versus
time for zero, first-,
and second-order
decay reactions
Concentration
0 Order
1st Order
2nd Order
Time
Bulk and Wall Reactions. Disinfectants are the most frequently modeled con-
stituents in water distribution systems. Upon leaving the plant and entering the distri-
bution system, disinfectants are subject to a poorly characterized set of potential
chemical reactions. Figure 2.25 illustrates the flow of water through a pipe and the
types of chemical reactions with disinfectants that can occur along its length. Chlo-
rine (the most common disinfectant) is shown reacting in the bulk fluid with natural
organic matter (NOM), and at the pipe wall, where oxidation reactions with biofilms
and the pipe material (a cause of corrosion) can occur.
Many disinfectant decay models have been developed to account for these reactions.
The first-order decay model has been shown to be sufficiently accurate for most distri-
bution system modeling applications and is well established. Rossman, Clark, and
Grayman (1994) proposed a mathematical framework for combining the complex
reactions occurring within distribution system pipes. This framework accounts for the
physical transport of the disinfectant from the bulk fluid to the pipe wall (mass trans-
fer effects) and the chemical reactions occurring there.
Figure 2.25
Disinfectant reactions
occurring within a
typical distribution
Cl Cl Cl Biofilm
system pipe
Cl
N.O.M. Cl
Qi N.O.M.
N.O.M.
Cl
Cl
Cl Cl Corrosion
T C = r KC (2.48)
where K = overall reaction rate constant (1/T)

Equation 2.48 is a simple first-order reaction (n = 1). The reaction rate coefficient K,
however, is now a function of the bulk reaction coefficient and the wall reaction coef-
ficient, as indicated in the following equation.
kw kf
K = kb + ----------------------------
- (2.49)
RH kw + kf
where kb = bulk reaction coefficient (1/T)

kw = wall reaction coefficient (L/T)
kf = mass transfer coefficient, bulk fluid to pipe wall (L/T)
RH = hydraulic radius of pipeline (L)
The rate that disinfectant decays at the pipe wall depends on how quickly disinfectant
is transported to the pipe wall and the speed of the reaction once it is there. The mass
transfer coefficient is used to determine the rate at which disinfectant is transported
using the dimensionless Sherwood number, along with the molecular diffusivity coef-
ficient (of the constituent in water) and the pipeline diameter.
SH d
k f = --------- (2.50)
D
where SH = Sherwood number

d = molecular diffusivity of constituent in bulk fluid (L2/T)
D = pipeline diameter (L)
For stagnant flow conditions (Re < 1), the Sherwood number, SH, is equal to 2.0. For
turbulent flow (Re > 2,300), the Sherwood number is computed using Equation 2.51.
0.83 § Q· 0.333
S H = 0.023Re --- (2.51)
© d¹

Q = kinematic viscosity of fluid (L2/T)
For laminar flow conditions (1 < Re < 2,300), the average Sherwood number along
the length of the pipe can be used. To have laminar flow in a 6-in. (150-mm) pipe, the
flow would need to be less than 5 gpm (0.3 l/s) with a velocity of 0.056 ft/s (0.017 m/
s). At such flows, head loss would be negligible.
D Q
0.0668 § ----· Re § ---·
© L¹ © d¹
SH = 3.65 + ------------------------------------------------------------ (2.52)
D Q 2e3
1 + 0.04 §© ----·¹ Re §© ---·¹
L d
where L = pipe length (L)

Using the first-order reaction framework developed immediately above, both bulk
fluid and pipe wall disinfectant decay reactions can be accounted for. Bulk decay
coefficients can be determined experimentally. Wall decay coefficients, however, are
more difficult to measure and are frequently estimated using disinfectant concentra-
tion field measurements and water quality simulation results.
Formation Reactions. One shortcoming of the first-order reaction model is that

it accounts for the concentration of only one reactant. This model is sufficient if only
one reactant is being considered. For example, when chlorine residual concentrations
are modeled, chlorine is assumed to be the limiting reactant and the other reactants—
material at the pipe walls and natural organic matter (NOM)—are assumed to be
present in excess. The behavior of some disinfection by-product (DBP) formation
reactions, however, differs from this assumption. NOM, not chlorine, is frequently the
limiting reactant. DBP formation is just one example of a generalized class of reac-
tions that can be modeled using a limiting reactant. The reaction term for this class of
formation and decay reactions as proposed by Rossman (2000) is shown in Equation
2.53.
n–1
T C = r k C lim – C C (2.53)
where Clim = limiting concentration of the reaction (M/L3)
A first-order growth rate to a limiting value has been used to represent the formation
of trihalomethanes, a common form of DBP, in distribution systems (Vasconcelos et
al., 1996). Mathematically this is represented by Equation 2.54 and is shown graphi-
cally in Figure 2.26.
– kt
THM t = C o + > FP – C o @ > 1 – e @ (2.54)
where THM(t) = THM concentration at time t

Co = initial THM concentration
FP = formation potential (concentration)
k = reaction rate (a positive value)
Figure 2.26
First-order growth
rate to a limiting value
Other Types of Water Quality Simulations

Although the water quality features of individual software packages vary, the most
common types of water quality simulations, in addition to the constituent analysis
already described, are source trace and water age analyses. The solution methods used
in both of these simulations are actually specific applications of the method used in
constituent analysis.
Source Trace Analysis. For the sake of reliability, or to simply provide suffi-
cient quantities of water to customers, a utility often uses more than one water supply
source. Suppose, for instance, that two treatment plants serve the same distribution
system. One plant draws water from a surface source, and the other pulls from an
underground aquifer. The raw water qualities from these sources are likely to differ
significantly, resulting in quality differences in the finished water as well.
Using a source trace analysis, the areas within the distribution system influenced by a
particular source can be determined, and, more important, areas where mixing of
water from different sources has occurred can be identified. The significance of
source mixing depends on the quality characteristics of the waters. Sometimes, mix-
ing can reduce the aesthetic qualities of the water (for example, creating cloudiness as
solids precipitate, or causing taste and odor problems to develop), and can contribute
to disinfectant residual maintenance problems. Source trace analyses are also useful
in tracking water quality problems related to storage tanks by tracing water from stor-
age as it is transported through the network.
A source trace analysis is a useful tool for better management of these situations. Spe-
cifically, it can be used to determine the percentage of water originating from a partic-
ular source for each junction node, tank, and reservoir in the distribution system
model. The procedure the software uses for this calculation is a special case of
constituent analysis in which the trace originates from the source as a conservative
constituent with an output concentration of 100 units. The constituent transport and
mixing equations introduced in the beginning of this section are then used to simulate
the transport pathways through the network and the influence of transport delays and
dilution on the trace constituent concentration. The values computed by the
simulation are then read directly as the percentage of water arriving from the source
location.
Water Age Analysis. The chemical processes that can affect distribution sys-
tem water quality are a function of water chemistry and the physical characteristics of
the distribution system itself (for example, pipe material and age). More generally,
however, these processes occur over time, making residence time in the distribution
system a critical factor influencing water quality. The cumulative residence time of
water in the system, or water age, has come to be regarded as a reliable surrogate for
water quality. Water age is of particular concern when quantifying the effect of stor-
age tank turnover on water quality. It is also beneficial for evaluating the loss of disin-
fectant residual and the formation of disinfection by-products in distribution systems.
The chief advantage of a water age analysis when compared to a constituent analysis
is that once the hydraulic model has been calibrated, no additional water quality cali-
bration procedures are required. The water age analysis, however, will not be as pre-
cise as a constituent analysis in determining water quality; nevertheless, it is an easy
way to leverage the information embedded in the calibrated hydraulic model. Con-
sider a project in which a utility is analyzing mixing in a tank and its effect on water
quality in an area of a network experiencing water quality problems. If a hydraulic
model has been developed and adequately calibrated, it can immediately be used to
evaluate water age. The water age analysis may indicate that excessively long resi-
dence times within the tank are contributing to water quality degradation. Using this
information, a more precise analysis can be planned (such as an evaluation of tank
hydraulic dynamics and mixing characteristics, or a constituent analysis to determine
the impact on disinfectant residuals), and preliminary changes in design or operation
can be evaluated.
The water age analysis reports the cumulative residence time for each parcel of water
moving through the network. Again, the algorithm the software uses to perform the
analysis is a specialized case of constituent analysis. Water entering a network from a
source is considered to have an age of zero. The constituent analysis is performed
assuming a zero-order reaction with a k value equal to +1 [(mg/l)/s]. Thus, constituent
concentration growth is directly proportional to time, and the cumulative residence
time along the transport pathways in the network is numerically summed.
Using the descriptions of water quality transport and reaction dynamics provided
here, and the different types of water quality-related simulations available in modern
software packages, water quality in the distribution system can be accurately pre-
dicted. Water quality modeling can be used to help improve the performance of distri-
bution system modifications meant to reduce hydraulic residence times, and as a tool
for improving the management of disinfectant residuals and other water quality-
related operations. Continuing advancements in technology combined with more
stringent regulations on quality at the customer’s tap are motivating an increasing
number of utilities to begin using the powerful water quality modeling capabilities
already available to them.
Solution Methods
The earliest water quality models of distribution systems were steady-state models
(Wood, 1980 and Males, Clark, Wehrman, and Gates, 1985). These models used
simultaneous equation or “marching out” solution methods to determine the steady-
state water quality concentrations throughout the distribution systems. However, it
quickly became apparent that steady-state water quality models were of limited use in
representing actual systems due to the temporal variability in distribution system
operation, the impacts of tanks on water quality, and temporal changes in source con-
centrations. This led to the development of several dynamic water quality models dur-
ing the mid to late 1980s (Clark, Grayman, Males, and Coyle, 1986; Hart, Meader,
and Chiang, 1986; Liou and Kroon, 1987; and Grayman, Clark, and Males, 1988).
Two methods are available to solve the dynamic water quality equations used in water
quality models. One method is based on an Eulerian approach that divides each sepa-
rate pipe into a series of equal length sub-links. The other method is a Lagrangian
approach that tracks parcels of water of homogeneous water quality concentrations as
they move through the pipe system. These solution methods are shown graphically in
Figures 2.27 and 2.28 and explained in detail in the paragraphs that follow. In both
solution methods, a hydraulic model must first be applied in extended-period simula-
tion (EPS) mode to determine the flow, flow direction, and velocity in each pipe at all
times during the simulation.
The Eulerian Approach. With an Eulerian approach (illustrated in Figure

2.27), an observer located at a fixed location watches water as it flows by. Grayman,
Clark, and Males (1988) developed an Eulerian solution method for water quality
modeling in distribution systems, and Rossman, Boulos, and Altman (1993) formal-
ized this method and named it the Discrete Volume Method (DVM).
In DVM, for each time period, a pipe is divided into a series of sub-links with the sub-
link length selected so that the time of travel through each sub-link is equal to a user-
selected water quality time step that remains constant throughout the simulation. As a
result, water moves from one sub-link to the next adjacent downstream sub-link in
one water quality time step.
In order to meet this constraint, the sub-link length varies from pipe to pipe and within
a pipe as flow changes. If the constituent being simulated is reactive, then the water
quality concentration is adjusted according to the appropriate reaction method during
each water quality time step.
At a junction at the downstream end of one or more pipes, the water quality concen-
tration in the junction is calculated by taking a flow-weighted average of incoming
inflows, as described previously by Equation 2.42. Water moves instantaneously
through pumps and valves without a change in water quality. At the end of a hydraulic
time step, if there is a change in flow or direction, then the sub-link gridding is
changed, and the water quality concentrations at the end of the previous time step are
used to define the initial water quality in each of the new sub-links. There are special
numerical assumptions made to accommodate “problem” situations such as very short
pipes (with travel time less than the water quality time step) and very long pipes (with
a very large number of sub-links).
Figure 2.27
Eulerian solution
method
The Lagrangian Approach. In the Lagrangian approach (illustrated in Figure

2.28), rather than observing the flow from the “sidelines,” the observer moves with
the flow. Additionally, rather than having a fixed grid, parcels of water with homoge-
neous concentrations are tracked through the pipe. New parcels are added when water
quality changes occur due to changes in source quality or when parcels are combined
at junctions. In order to reduce the number of parcels, algorithms have been devel-
oped for combining adjacent parcels when the difference in concentrations are less
than a user-defined tolerance. Liou and Kroon (1987) and Hart, Meader, and Chiang
(1986) developed Lagrangian solution methods for water distribution system water
quality models.
Lagrangian solutions can be either time-driven or event-driven. In a time-driven
method, conditions are updated at a fixed time step. In an event-driven model, condi-
tions are updated when the source water quality changes or when the front of a parcel
References 65
reaches a junction. In comparing the methods, Rossman and Boulos (1996) found that
the Lagrangian time-driven method is the most efficient and versatile of the solution
methods for water distribution system quality models.
Figure 2.28
Lagrangian solution
method
REFERENCES
ASCE. (1975). Pressure Pipeline Design for Water and Wastewater. ASCE, New York, New York.
ASCE/WEF. (1982). Gravity Sanitary Sewer Design and Construction. ASCE, Reston, Virginia.
ASCE Committee on Pipeline Planning. (1992). Pressure Pipeline Design for Water and Wastewater.
ASCE, Reston, Virginia.
Bernoulli, D. (1738). Hydrodynamica. Argentorati.
Bhave, P. R. (1991). Analysis of Flow in Water Distribution Networks. Technomics, Lancaster, Pennsylva-
nia.
Bosserman, B. E., (2000). “Pump System Hydraulic Design.” Water Distribution System Handbook, Mays,
L. W., ed., McGraw-Hill, New York, New York.
Clark, R. M., and Grayman, W. M. (1998). Modeling Water Quality in Distribution Systems. AWWA, Den-
ver, Colorado.
Clark, R. M., Grayman, W. M., Males, R. M., and Coyle, J. A. (1986). “Predicting Water Quality in Distri-
bution Systems.” Proceedings of the AWWA Distribution System Symposium, American Water Works
Association, Denver, Colorado.
Crane Company (1972). Flow of Fluids through Valves and Fittings. Crane Co., New York, New York.
Cross, H. (1936). “Analysis of Flow in Networks of Conduits or Conductors.” University of Illinois Experi-
ment Station Bulletin No. 286, Department of Civil Engineering, University of Illinois, Champaign
Urbana, Illinois.
Grayman, W. M., Clark, R. M., and Males, R. M. (1988). “Modeling Distribution System Water Quality:
Dynamic Approach.” Journal of Water Resources Planning and Management, ASCE, 114(3).
Grayman, W. M., Deininger, R. A., Green, A., Boulos, P. F., Bowcock, R. W., and Godwin, C. C. (1996).
“Water Quality and Mixing Models for Tanks and Reservoirs.” Journal of the American Water Works
Association, 88(7).
Grayman, W. M., Rossman, L. A., and Geldreich, E. E. (2000). “Water Quality.” Water Distribution Sys-
tems Handbook, Mays, L. W., ed., McGraw-Hill, New York, New York.
Grayman, W. M., Rossman, L. A., Arnold, C., Deininger, R. A., Smith, C., Smith, J. F., and Schnipke, R.
(2000). Water Quality Modeling of Distribution System Storage Facilities. AWWA.
Hart, F. F., Meader, J. L., and Chiang, S. N. (1986). “CLNET—A Simulation Model for Tracing Chlorine
Residuals in a Potable Water Distribution Network.” Proceedings of the AWWA Distribution System
Symposium, American Water Works Association, Denver, Colorado.
Hydraulic Institute (2000). Pump Standards. Parsippany, New Jersey.
Idelchik, I. E. (1999). Handbook of Hydraulic Resistance. 3rd edition, Begell House, New York, New York.
Lamont, P. A. (1981). “Common Pipe Flow Formulas Compared with the Theory of Roughness.” Journal
of the American Water Works Association, 73(5), 274.
Lansey, K., and Mays, L. W. (2000). “Hydraulics of Water Distribution Systems.” Water Distribution Sys-
tems Handbook, Mays, L. W., ed., McGraw-Hill, New York, New York.
Larock, B. E., Jeppson, R. W., and Watters, G. Z. (1999). Handbook of Pipeline Systems. CRC Press, Boca
Raton, Florida.
Liou, C. P. and Kroon, J. R. (1987). “Modeling the Propagation of Waterborne Substances in Distribution
Networks.” Journal of the American Water Works Association, 79(11), 54.
Karassik, I. J., ed. (1976). Pump Handbook. McGraw-Hill, New York, New York.
Males, R. M., Clark, R. M., Wehrman, P. J., and Gates, W. E. (1985). “An Algorithm for Mixing Problems
in Water Systems.” Journal of Hydraulic Engineering, ASCE, 111(2).
Mays, L. W., ed. (1999). Hydraulic Design Handbook. McGraw-Hill, New York, New York.
Miller, D. S. (1978). Internal Flow Systems. BHRA Fluid Engineering, Bedford, United Kingdom.
Moody, L. F. (1944). “Friction Factors for Pipe Flow.” Transactions of the American Society of Mechanical
Engineers, Vol. 66.
Nikuradse (1932). “Gestezmassigkeiten der Turbulenten Stromung in Glatten Rohren.” VDI-Forschungsh,
No. 356 (in German).
Rossman, L.A. (2000). EPANET Users Manual. Risk Reduction Engineering Laboratory, U.S. Environ-
mental Protection Agency, Cincinnati, Ohio.
Rossman, L. A., and Boulos, P. F. (1996). “Numerical Methods for Modeling Water Quality in Distribution
Systems: A Comparison.” Journal of Water Resources Planning and Management, ASCE, 122(2), 137.
Rossman, L. A., Boulos, P. F., and Altman, T. (1993). “Discrete Volume-Element Method for Network
Water-Quality Models.” Journal of Water Resources Planning and Management, ASCE, 119(5), 505.
Rossman, L. A., Clark, R. M., and Grayman, W. M. (1994). “Modeling Chlorine Residuals in Drinking
Water Distribution Systems.” Journal of Environmental Engineering, ASCE, 1210(4), 803.
Sanks, R. L., ed. (1998). Pumping Station Design. 2nd edition, Butterworth, London, UK.
Streeter, V. L., Wylie, B. E., and Bedford, K. W. (1998). Fluid Mechanics. 9th edition, WCB/McGraw-Hill,
Boston, Massachusetts.
Swamee, P. K., and Jain, A. K. (1976). “Explicit Equations for Pipe Flow Problems.” Journal of Hydraulic
Engineering, ASCE, 102(5), 657.
Todini, E., and Pilati, S. (1987). “A Gradient Method for the Analysis of Pipe Networks.” Proceedings of
the International Conference on Computer Applications for Water Supply and Distribution, Leicester
Polytechnic, UK.
Vasconcelos, J. J., Boulos, P. F., Grayman, W. M., Kiene, L., Wable, O., Biswas, P., Bhari, A., Rossman, L.,
Clark, R., and Goodrich, J. (1996). Characterization and Modeling of Chlorine Decay in Distribution
Systems. AWWA, Denver, Colorado.
Walski, T. M. (1984). Analysis of Water Distribution Systems. Van Nostrand Reinhold, New York, New
York.
Williams, G. S., and Hazen, A. (1920). Hydraulic Tables. John Wiley & Sons, New York, New York.
Wood, D. J. (1980). “Slurry Flow in Pipe Networks.” Journal of Hydraulics, ASCE, 106(1), 57.
Discussion Topics and Problems 67
DISCUSSION TOPICS AND PROBLEMS
Read the chapter and complete the problems. Submit your work to Haestad Methods
and earn up to 11.0 CEUs. See Continuing Education Units on page xxix for more
information, or visit www.haestad.com/awdm-ceus/.
2.1 Find the viscosity of the fluid contained between the two square plates shown in the figure. The top
plate is moving at a velocity of 3 ft/s.
F = 50 lb
t = 0.5 in.
6 ft
2.2 Find the force P required to pull the 150 mm circular shaft shown in the figure through the sleeve at
a velocity of 1.5 m/s. The fluid between the shaft and the sleeve is water at a temperature of 15 oC.
75 mm
P
150 mm
2 mm
2.3 Find the pressure at the base of a container of water having a depth of 15 m.
2.4 How high is the water level from the base of an elevated storage tank if the pressure at the base of the
tank is 45 psi?
2.5 Water having a temperature of 65 F is flowing through a 6-in. ductile iron main at a rate of 300 gpm.
o
Is the flow laminar, turbulent, or transitional?
2.6 What type of flow do you think normally exists in water distribution systems: laminar, turbulent, or
transitional? Justify your selection with sound reasoning.
2.7 What is the total head at point A in the system shown in the figure if the flow through the pipeline is
1,000 gpm? What is the head loss in feet between point A and point B?
A B
PA= 62 psi PB= 48 psi
8 in. Q
550 ft 550 ft
Datum
2.8 For the piping system shown in Problem 2.7, what would the elevation at point B have to be in order
for the reading on the two pressure gages to be the same?
2.9 Assuming that there are no head losses through the Venturi meter shown in the figure, what is the
pressure reading in the throat section of the Venturi? Assume that the discharge through the meter is
158 l/s.
P = 497 kPa
P = ? kPa
400 mm 150 mm
2.10 What is the head loss through a 10-in. diameter concrete water main 2,500 ft in length if water at
60oF is flowing through the line at a rate of 1,250 gpm? Solve using the Darcy-Weisbach formula.
2.11 For Problem 2.10, what is the flow through the line if the head loss is 32 ft? Solve using the Darcy-
Weisbach formula.
2.12 Find the length of a pipeline that has the following characteristics: Q=41 l/s, D=150 mm, Hazen-
Williams C=110, HL=7.6 m.
2.13 For the pipeline shown in Problem 2.7, what is the Hazen-Williams C-factor if the distance between
the two pressure gages is 725 ft and the flow is 1,000 gpm?
2.14 English Units: Compute the pipe resistance coefficient, K , for the following pipelines.
P
Length Diameter Hazen-Williams Pipe Resistance

(ft) (in.) C-factor Coefficient (Kp)
1,200 12 120
500 4 90
75 3 75
3,500 10 110
1,750 8 105
SI Units: Compute the pipe resistance coefficient, KP , for the following pipelines.
Length Diameter Hazen-Williams Pipe Resistance

(m) (mm) C-factor Coefficient (Kp)
366 305 120
152 102 90
23 76 75
1067 254 110
533 203 105
2.15 English Units: Compute the minor loss term, K , for the fittings shown in the table below.
M
Minor Loss Pipe Size Minor Loss Term

Type of Fitting/Flow Condition
Coefficient (in.) (KM)
Gate Valve - 50% Open 4.8 8
Tee - Line Flow 0.4 12
90o Mitered Bend 0.8 10
Fire Hydrant 4.5 6
SI Units: Compute the minor loss term, KM, for the fittings shown in the table below.
Minor Loss Pipe Size Minor Loss Term

Type of Fitting/Flow Condition
Coefficient (mm) (KM)
Gate Valve - 50% Open 4.8 200
Tee - Line Flow 0.4 300
90o Mitered Bend 0.8 250
Fire Hydrant 4.5 150
2.16 English Units: Determine the pressures at the following locations in a water distribution system,
assuming that the HGL and ground elevations at the locations are known.
HGL Elevation Pressure

Node Label
(ft) (ft) (psi)
J-1 550.6 423.5
J-6 485.3 300.5
J-23 532.6 500.0
J-5 521.5 423.3
J-12 515.0 284.0
SI Units: Determine the pressures at the following locations in a water distribution system, assuming
that the HGL and ground elevations at the locations are known.
HGL Elevation Pressure
Node Label
(m) (m) (kPa)
J-1 167.8 129.1
J-6 147.9 91.6
J-23 162.3 152.4
J-5 159.0 129.0
J-12 157.0 86.6
2.17 Using the concept of conservation of mass, is continuity maintained at the junction node shown in
the following figure?
QP-10 = 38 gpm P -10
P -8 J -10 P -9
QP-8 = 55 gpm QP-9 = 72 gpm
DJ-10 = 45 gpm
QP-12 = 120 gpm

P -12
2.18 Find the magnitude and direction of the flow through pipe P-9 so that continuity is maintained at
node J-10 in the following figure.
QP-10= 3.0 L/s P -10
P -8 J -10 P -9
Q P-9 = ? L/s
QP-8 = 1.4 L/s
DJ-10 = 0.9 L/s
QP-12= 1.9 L/s P -12

2.19 Does conservation of energy around a loop apply to the loop shown in the following figure? Why or
why not? The total head loss (sum of friction losses and minor losses) in each pipe and the direction
of flow are shown in the figure.
P -23
HL,P-23 = 7.73 ft
P -32 HL,P-32 = 3.76 ft

HL,P-25 = 1.87 ft P -25
HL,P-27 = 2.10 ft
P -27
2.20 Does Conservation of Energy apply to the system shown in the figure? Data describing the physical
characteristics of each pipe are presented in the table below. Assume that there are no minor losses in
this loop.
P -23
QP-23 = 22.7 L/s
P -32 QP-32 = 4.7 L/s QP-25 = 7.6 L/s P -25
QP-27 = 27.1 L/s
P -27
Length Diameter Hazen-Williams

Pipe Label
(m) (mm) C-factor
P-23 381.0 305 120
P-25 228.6 203 115
P-27 342.9 254 120
P-32 253.0 152 105
2.21 Find the discharge through the system shown in the figure. Compute friction loss using the Hazen-
Williams equation.
225 ft
R -2
Pipe 2:
L=5,000 ft
D=4 in.
C=85
125 ft
R -1
2.22 Find the pump head needed to deliver water from reservoir R-1 to reservoir R-2 in the following fig-
ure at a rate of 70.8 l/s. Compute friction losses using the Hazen-Williams equation.
68.6 m
R -2
Pipe 2:
L=1524 m
D=102 mm
C=85
38.1 m
R -1 Pipe 1:
L=152 m
D=305 mm
C=120
2.23 Compute the age of water at the end of a 12-in. pipe that is 1,500 ft in length and has a flow of 900
gpm. The age of the water when it enters the pipeline is 7.2 hours.
2.24 Suppose that a 102-mm pipe is used to serve a small cluster of homes at the end of a long street. If
the length of the pipe is 975 m, what is the age of water leaving it if the water had an age of 6.3 hours
when entering the line? Assume that the water use is 1.6 l/s.
2.25 Given the data in the tables below, what is the average age of the water leaving junction node J-4
shown in the figure? What is the flow rate through pipe P-4? What is the average age of the water
arriving at node J-5 through pipe P-4? Fill in your answers in the tables provided.
J -2
P -2
J -1 P -1 J -4 P -4 J -5
35 gpm
P -3
J -3
Flow Length Diameter

Pipe Label
(gpm) (ft) (in.)
P-1 75 1,650 10
P-2 18 755 8
P-3 23 820 6
P-4 2,340 10
Average Age
Node Label
(hours)
J-1 5.2
J-2 24.3
J-3 12.5
J-4
J-5
2.26 What will be the concentration of chlorine in water samples taken from a swimming pool after 7
days if the initial chlorine concentration in the pool was 1.5 mg/l? Bottle tests performed on the pool
water indicate that the first-order reaction rate is -0.134 day-1.
2.27 Do you think that the actual reaction rate coefficient for water in the swimming pool described above
(i.e., the water being considered remains in the pool, and is not being stored under laboratory condi-
tions) would be equal to -0.134 day-1? Suggest some factors that might cause the actual reaction rate
to differ. Would these factors most likely cause the actual reaction rate to be greater than or less than
-0.134 day-1?
2.28 For the system presented in Problem 2.25, what is the concentration of a constituent leaving node J-
4 (assume it is a conservative constituent)? The constituent concentration is 0.85 mg/L in pipe P-1,
0.50 mg/l in pipe P-2, and 1.2 mg/l in pipe P-3.
2.29 What is the fluoride concentration at the end of a 152-mm diameter pipeline 762 m in length if the
fluoride concentration at the start of the line is 1.3 mg/l? Fluoride is a conservative species; that is, it
does not decay over time. Ignoring dispersion, if there is initially no fluoride in the pipe and it is
introduced at the upstream end at a 2.0 mg/l concentration, when will this concentration be reached
at the end of the line if the flow through the pipe is 15.8 l/s? Assume that there are no other junction
nodes along the length of this pipe.
2.30 A community has found high radon levels in one of their wells. Because radon decays relatively
quickly, they are exploring the use of a baffled clearwell at the well in order to provide some time
delay before the water enters the distribution system. Their goal is to reduce the radon levels by 80
percent in the clearwell. The half-life for radon is 3.8 days. What minimum residence time is
required in the clearwell to meet this goal? Is the use of the clearwell as a means of meeting their
goal reasonable?
2.31 Trihalomethane levels leaving a treatment plant are 20 ug/l. Based on bottle tests, the formation was
found to follow a first-order reaction with a growth rate of 2 l/day and an ultimate formation poten-
tial of 100 ug/l. What would the THM concentration be after 1 day? How long would it take for the
THM concentration to reach 0.99 ug/l?
2.32 A city is building a 6-in. diameter, 5000-ft pipeline to serve a small community with an average con-
sumption of 20 gpm. Based on tests, they estimate that the bulk decay rate for chlorine is –0.5 1/day
and the wall decay rate will be –1 feet per day. If the chlorine residual at the start of the pipe is
1 mg/l, what will the chlorine residual be in the water delivered to the community? If you only con-
sidered the effects of bulk decay, what would the chlorine residual be? If you only considered wall
decay, what would the chlorine residual be? (Hint: Using a distribution system model, build a net-
work model composed of a reservoir, a single pipe, and a junction representing the community.)
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
3
Assembling a Model
As Chapter 1 discusses, a water distribution model is a mathematical description of a

real-world system. Before building a model, it is necessary to gather information
describing the network. In this chapter, we introduce and discuss sources of data used
in constructing models.
The latter part of the chapter covers model skeletonization. Skeletonization is the pro-
cess of simplifying the real system for model representation, and it involves making
decisions about the level of detail to be included.
3.1 MAPS AND RECORDS

Many potential sources are available for obtaining the data required to generate a
water distribution model, and the availability of these sources varies dramatically
from utility to utility. The following sections discuss some of the most commonly
used resources, including system maps, as-built drawings, and electronic data files.
System Maps
System maps are typically the most useful documents for gaining an overall under-
standing of a water distribution system because they illustrate a wide variety of
valuable system characteristics. System maps may include such information as
• Pipe alignment, connectivity, material, diameter, and so on
• The locations of other system components, such as tanks and valves
• Pressure zone boundaries
• Elevations
• Miscellaneous notes or references for tank characteristics
• Background information, such as the locations of roadways, streams, plan-
ning zones, and so on
• Other utilities
76 Assembling a Model Chapter 3
Topographic Maps
A topographic map uses sets of lines called contours to indicate elevations of the
ground surface. Contour lines represent a contiguous set of points that are at the same
elevation and can be thought of as the outline of a horizontal “slice” of the ground sur-
face. Figure 3.1 illustrates the cross-sectional and topographic views of a sphere, and
Figure 3.2 shows a portion of an actual topographic map. Topographic maps are often
referred to by the contour interval that they present, such as a 20-foot topographic
map or a 1-meter contour map.
By superimposing a topographic map on a map of the network model, it is possible to
interpolate the ground elevations at junction nodes and other locations throughout the
system. Of course, the smaller the contour interval, the more precisely the elevations
can be estimated. If available topographic maps cannot provide the level of precision
needed, other sources of elevation data need to be considered.
Topographic maps are also available in the form of Digital Elevation Models (DEMs),
which can be used to electronically interpolate elevations. The results of the DEM are
only as accurate as the underlying topographic data on which they are based; thus, it is
possible to calculate elevations to a large display precision but with no additional
accuracy.
Figure 3.1
Topographic Looking Down
representation of a Generates
hemisphere Plan View
400
300
200
100
Looking from
0
the Side
40
Profile Generates
0
Profile View
300
20
0
10
0
Plan
As-Built Drawings
Site restrictions and on-the-fly changes often result in differences between original
design plans and the actual constructed system. As a result, most utilities perform
post-construction surveys and generate a set of as-built or record drawings for the
purpose of documenting the system exactly as it was built. In some cases, an inspec-
tor's notes may even be used as a supplemental form of documentation. As-built
drawings can be especially helpful in areas where a fine level of precision is required
for pipe lengths, fitting types and locations, elevations, and so forth.
Section 3.1 Maps and Records 77
Figure 3.2
Typical topographic
map
As-built drawings can also provide reliable descriptions of other system components
such as storage tanks and pumping stations. There may be a complete set of drawings
for a single tank, or the tank plans could be included as part of a larger construction
project.
Electronic Maps and Records

Many water distribution utilities have some form of electronic representation of their
systems in formats that may vary from a nongraphical database, to a graphics-only
Computer-Aided Drafting (CAD) drawing, to a Geographic Information System (GIS)
that combines graphics and data.
Nongraphical Data. It is common to find at least some electronic data in non-
graphical formats, such as a tracking and inventory database, or even a legacy text-
based model. These sources of data can be quite helpful in expediting the process of
model construction. Even so, care needs to be taken to ensure that the network topol-
ogy is correct, because a simple typographic error in a nongraphical network can be
difficult to detect.
Computer-Aided Drafting. The rise of computer technology has led to many
improvements in all aspects of managing a water distribution utility, and mapping is
no exception. CAD systems make it much easier to plug in survey data, combine data
from different sources, and otherwise maintain and update maps faster and more reli-
ably than ever before.
Even for systems having only paper maps, many utilities digitize those maps to con-
vert them to an electronic drawing format. Traditionally, digitizing has been a process
of tracing over paper maps with special computer peripherals, called a digitizing tab-
let and puck (see Figure 3.3). A paper map is attached to the tablet, and the drafts-
person uses crosshairs on the puck to point at locations on the paper. Through
magnetic or optical techniques, the tablet creates an equivalent point at the
appropriate location in the CAD drawing. As long as the tablet is calibrated correctly,
it will automatically account for rotation, skew, and scale.
Figure 3.3
A typical digitizing
tablet
Another form of digitizing is called heads-up digitizing (see Figure 3.4). This method
involves scanning a paper map into a raster electronic format (such as a bitmap),
bringing it into the background of a CAD system, and electronically tracing over it on
a different layer. The term heads-up is used because the draftsperson remains focused
on the computer screen rather than on a digitizing tablet.
Geographic Information Systems. A Geographic information system
(GIS) is a computer-based tool for mapping and analyzing objects and events that
happen on earth. GIS technology integrates common database operations such as
query and statistical analysis with the unique visualization and geographic analysis
benefits offered by maps (ESRI, 2001). Because a GIS stores data on thematic layers
linked together geographically, disparate data sources can be combined to determine
relationships between data and to synthesize new information.
Section 3.2 Model Representation 79
Figure 3.4
Network model
overlaid on an aerial
photograph
GIS can be used for tasks such as proximity analysis (identifying customers within a
certain distance of a particular node), overlay analysis (determining all junctions that
are completely within a particular zoning area), network analysis (identifying all
households impacted by a water-main break), and visualization (displaying and com-
municating master plans graphically). With a hydraulic model that links closely to a
GIS, the benefits can extend well beyond just the process of building the model and
can include skeletonization, demand generalization, and numerous other operations.
3.2 MODEL REPRESENTATION

The concept of a network is fundamental to a water distribution model. The network
contains all of the various components of the system, and defines how those elements
are interconnected. Networks are comprised of nodes, which represent features at spe-
cific locations within the system, and links, which define relationships between nodes.
Network Elements
Water distribution models have many types of nodal elements, including junction
nodes where pipes connect, storage tank and reservoir nodes, pump nodes, and con-
trol valve nodes. Models use link elements to describe the pipes connecting these
nodes. Also, elements such as valves and pumps are sometimes classified as links
rather than nodes. Table 3.1 lists each model element, the type of element used to rep-
resent it in the model, and the primary modeling purpose.
Table 3.1 Common network modeling elements

Element Type Primary Modeling Purpose
Reservoir Node Provides water to the system
Tank Node Stores excess water within the system and releases that water
at times of high usage
Junction Node Removes (demand) or adds (inflow) water from/to
the system
Pipe Link Conveys water from one node to another
Pump Node Raises the hydraulic grade to overcome elevation differences
or link and friction losses
Control Node Controls flow or pressure in the system based on specified
Valve or link criteria
Naming Conventions (Element Labels). Because models may contain tens

of thousands of elements, naming conventions are an important consideration in mak-
ing the relationship between real-world components and model elements as obvious
as possible (see Figure 3.5). Some models allow only numeric numbering of ele-
ments, but most modern models support at least some level of alphanumeric labeling
(for example, “J-1,” “Tank 5,” or “West Side Pump A”).
Figure 3.5
Schematic junction
with naming
convention “115” = Sequential Number
Description
J5 - 115 - Elm Street

“5” = Zone 5
“J” = Junction Node
Naming conventions should mirror the way the modeler thinks about the particular
network by using a mixture of prefixes, suffixes, numbers, and descriptive text. In
general, labels should be as short as possible to avoid cluttering a drawing or report,
but they should include enough information to identify the element. For example, a
naming convention might include a prefix for the element type, another prefix to indi-
cate the pressure zone or map sheet, a sequential number, and a descriptive suffix.
Of course, modelers can choose to use some creativity, but it is important to realize
that a name that seems obvious today may be baffling to future users. Intelligent use
of element labeling can make it much easier for users to query tabular displays of
model data with filtering and sorting commands. In some cases, such as automated
calibration, it may be very helpful to group pipes with like characteristics to make cal-
ibration easier. If pipe labels have been set up such that like pipes have similar labels,
this grouping becomes easy.
Rather than starting pipe labeling at a random node, it is best to start from the water
source and number outward along each pipeline. In addition, just as pipe elements
were not laid randomly, a pipe labeling scheme should be developed to reflect that.
For example, consider the pipes in Figure 3.6 (Network A), which shows that the
pipes were laid in four separate projects in four different years. By labeling the pipes
as shown in Figure 3.6 (Network B), the user will be able to more rationally group,
filter, and sort pipes. For example, pipes laid during the 1974 construction project
were labeled P-21, P-22, and so on so that those pipes could be grouped together. This
can have major time-saving benefits in working with a large system.
Figure 3.6
Logical element
labeling schemes
Boundary Nodes. A boundary node is a network element used to represent

locations with known hydraulic grade elevations. A boundary condition imposes a
requirement within the network that simulated flows entering or exiting the system
agree with that hydraulic grade. Reservoirs (also called fixed grade nodes) and tanks
are common examples of boundary nodes.
Every model must have at least one boundary node so that there is a reference point
for the hydraulic grade. In addition, every node must maintain at least one path back
to a boundary node so that its hydraulic grade can be calculated. When a node
becomes disconnected from a boundary (as when pipes and valves are closed), it can
result in an error condition that needs to be addressed by the modeler.
Network Topology
The most fundamental data requirement is to have an accurate representation of the
network topology, which details what the elements are and how they are intercon-
nected. If a model does not faithfully duplicate real-world layout (for example, the
model pipe connects two nodes that are not really connected), then the model will
never accurately depict real-world performance, regardless of the quality of the
remaining data.
System maps are generally good sources of topological information, typically includ-
ing data on pipe diameters, lengths, materials, and connections with other pipes.
There are situations in which the modeler must use caution, however, because maps
may be imperfect or unclear.
False Intersections. Just because mains appear to cross on a map does not nec-
essarily mean that a hydraulic connection exists at that location. As illustrated in Fig-
ure 3.7, it is possible for one main to pass over the other (called a crossover).
Modeling this location as an intersecting junction node would be incorrect, and could
result in serious model inaccuracies. Note that some GISs automatically assign nodes
where pipes cross, which may not be hydraulically correct.
When pipes are connected in the field via a bypass (as illustrated in Figure 3.7), the
junction node should only be included in the model if the bypass line is open. Since
the choice to include or omit a junction in the model based on the open or closed sta-
tus of a bypass in the field is somewhat difficult to control, it is recommended that the
bypass itself be included in the model. As a result, the modeler can more easily open
or close the bypass in accordance with the real system.
Figure 3.7
Pipe crossover and
crossover with bypass
Cross Crossover Crossover w/

Bypass Line
Converting CAD Drawings into Models. Although paper maps can some-
times falsely make it appear as though there is a pipe intersection, CAD maps can
have the opposite problem. CAD drawings are often not created with a hydraulic
model in mind; thus, lines representing pipes may visually appear to be connected on
a large-scale plot, but upon closer inspection of the CAD drawing, the lines are not
actually touching. Consider Figure 3.8, which demonstrates three distinct conditions
that may result in a misinterpretation of the topology:
• T-intersections: Are there supposed to be three intersecting pipes or two

non-intersecting pipes? The drawing indicates that there is no intersection,
but this could easily be a drafting error.
• Crossing pipes: Are there supposed to be four intersecting pipes or two non-
intersecting pipes?
• Nearly connecting line endpoints: Are the two pipes truly non-intersecting?
Automated conversion from CAD drawing elements to model elements can save time,
but (as with any automated process) the modeler needs to be aware of the potential
pitfalls involved and should review the end result. Some models assist in the review
process by highlighting areas with potential connectivity errors. The possibility of
difficult-to-detect errors still remains, however, persuading some modelers to trace
over CAD drawings when creating model elements.
Figure 3.8
Common CAD
conversion errors
3.3 RESERVOIRS
The term reservoir has a specific meaning with regard to water distribution system
modeling that may differ slightly from the use of the word in normal water distribu-
tion construction and operation. A reservoir represents a boundary node in a model
that can supply or accept water with such a large capacity that the hydraulic grade of
the reservoir is unaffected and remains constant. It is an infinite source, which means
that it can theoretically handle any inflow or outflow rate, for any length of time,
without running dry or overflowing. In reality, there is no such thing as a true infinite
source. For modeling purposes, however, there are situations where inflows and out-
flows have little or no effect on the hydraulic grade at a node.
Reservoirs are used to model any source of water where the hydraulic grade is con-
trolled by factors other than the water usage rate. Lakes, groundwater wells, and
clearwells at water treatment plants are often represented as reservoirs in water distri-
bution models. For modeling purposes, a municipal system that purchases water from
a bulk water vendor may model the connection to the vendor’s supply as a reservoir
(most current simulation software includes this functionality).
For a reservoir, the two pieces of information required are the hydraulic grade line
(water surface elevation) and the water quality. By model definition, storage is not a
concern for reservoirs, so no volumetric storage data is needed.
3.4 TANKS
A storage tank (see Figure 3.9) is also a boundary node, but unlike a reservoir, the
hydraulic grade line of a tank fluctuates according to the inflow and outflow of water.
Tanks have a finite storage volume, and it is possible to completely fill or completely
exhaust that storage (although most real systems are designed and operated to avoid
such occurrences). Storage tanks are present in most real-world distribution systems,
and the relationship between an actual tank and its model counterpart is typically
straightforward.
Section 3.4 Tanks 85
Figure 3.9
Storage tanks
For steady-state runs, the tank is viewed as a known hydraulic grade elevation, and the
model calculates how fast water is flowing into or out of the tank given that HGL.
Given the same HGL setting, the tank is hydraulically identical to a reservoir for a
steady-state run. In extended-period simulation (EPS) models, the water level in the
tank is allowed to vary over time. To track how a tank’s HGL changes, the relationship
between water surface elevation and storage volume must be defined. Figure 3.10 illus-
trates this relationship for various tank shapes. For cylindrical tanks, developing this
relationship is a simple matter of identifying the diameter of the tank, but for non-
cylindrical tanks it can be more challenging to express the tank’s characteristics.
Some models do not support noncylindrical tanks, forcing the modeler to approximate
the tank by determining an equivalent diameter based on the tank’s height and capac-
ity. This approximation, of course, has the potential to introduce significant errors in
hydraulic grade. Fortunately, most models do support non-cylindrical tanks, although
the exact set of data required varies from model to model.
Regardless of the shape of the tank, several elevations are important for modeling pur-
poses. The maximum elevation represents the highest fill level of the tank, and is
usually determined by the setting of the altitude valve if the tank is equipped with one.
The overflow elevation, the elevation at which the tank begins to overflow, is slightly
higher. Similarly, the minimum elevation is the lowest the water level in the tank
should ever be. A base or reference elevation is a datum from which tank levels are
measured.
The HGL in a tank can be referred to as an absolute elevation or a relative level,
depending on the datum used. For example, a modeler working near the “Mile High”
city of Denver, Colorado, could specify a tank’s base elevation as the datum, and then
work with HGLs that are relative to that datum. Alternatively, the modeler could work
with absolute elevations that are in the thousands of feet. The choice of whether to use
absolute elevations or relative tank levels is a matter of personal preference. Figure
3.11 illustrates these important tank elevation conventions for modeling tanks. Notice
that when using relative tank levels, it is possible to have different values for the same
level, depending on the datum selected.
Figure 3.10
Volume versus level
curves for various 1.0
tank shapes
0.9
0.8
0.7
0.6
Volume Ratio
0.5
0.4
Spherical
Cylindrical/
0.3
Rectangular
45 Deg. Cone
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Depth Ratio
Water storage tanks can be classified by construction material (welded steel, bolted
steel, reinforced concrete, prestressed concrete), shape (cylindrical, spherical, torroi-
dal, rectangular), style (elevated, standpipe, ground, buried), and ownership (utility,
private) (Walski, 2000). However, for pipe network modeling, the most important
classification is whether or not the tank “floats on the system.” A tank is said to float
on the system if the hydraulic grade elevation inside the tank is the same as the HGL
in the water distribution system immediately outside of the tank. With tanks, there are
really three situations that a modeler can encounter:
1. Tank that floats on the system with a free surface
2. Pressure (hydropneumatic) tank that floats on the system
3. Pumped storage in which water must be pumped from a tank

Section 3.4 Tanks 87
Figure 3.11
Important tank
elevations
Figure 3.12 shows that elevated tanks, standpipes, and hydropneumatic tanks float on
the system because their HGL is the same as that of the system. Ground tanks and
buried tanks may or may not float on the system, depending on their elevation. If the
HGL in one of these tanks is below the HGL in the system, water must be pumped
from the tank, resulting in pumped storage.
A tank with a free surface floating on the system is the simplest and most common
type of tank. The pumped storage tank needs a pump to deliver water from the tank to
the distribution system and a control valve (usually modeled as a pressure sustaining
valve) to gradually fill the tank without seriously affecting pressure in the surrounding
system.
Figure 3.12
Relationship between
HGL floating, pressurized,
and pumped tanks
Ground
Buried
//=//=
//=//=
//=//=
Ground
Hydro- Elevated Standpipe
Buried Pneumatic
Pumped Storage Floating on System
Hydropneumatic Tanks. In most tanks, the water surface elevation in the tank
equals the HGL in the tank. In the case of a pressure tank, however, the HGL is higher
than the tank’s water surface. Pressure tanks, also called hydropneumatic tanks, are
partly full of compressed air. Because the water in the tank is pressurized, the HGL is
higher than the water surface elevation, as reflected in Equation 3.1.
HGL = C f P + Z (3.1)
where HGL = HGL of water in tank (ft, m)

P = pressure recorded at tank (psi, kPa)
Z = elevation of pressure gage (ft, m)
In steady-state models, a hydropneumatic tank can be represented by a tank or reser-
voir having this HGL. In EPS models, the tank must be represented by an equivalent
free-surface tank floating on the system. Because of the air in the tank, a hydropneu-
matic tank has an effective volume that is less than 30 to 50 percent of the total
volume of the tank. Modeling the tank involves first determining the minimum and
maximum pressures occurring in the tank and converting them to HGL values using
Equation 3.1. The cross-sectional area (or diameter) of this equivalent tank can be
determined by using Equation 3.2.
V eff
A eq = -----------------------------------------------
- (3.2)
HGL max – HGL min
where Aeq = area of equivalent tank (ft2, m2)

Veff = effective volume of tank (ft3, m3)
HGLmax = maximum HGL in tank (ft, m)
HGLmin = minimum HGL in tank (ft, m)
The relationship between the actual hydropneumatic tank and the model tank is
shown in Figure 3.13.
Using this technique, the EPS model of the tank will track HGL at the tank and vol-
ume of water in the tank, but not the actual water level.
3.5 JUNCTIONS
As the term implies, one of the primary uses of a junction node is to provide a loca-
tion for two or more pipes to meet. Junctions, however, do not need to be elemental
intersections, as a junction node may exist at the end of a single pipe (typically
referred to as a dead-end). The other chief role of a junction node is to provide a loca-
tion to withdraw water demanded from the system or inject inflows (sometimes
referred to as negative demands) into the system.
Junction nodes typically do not directly relate to real-world distribution components,
since pipes are usually joined with fittings, and flows are extracted from the system at
any number of customer connections along a pipe. From a modeling standpoint, the
Section 3.5 Junctions 89
importance of these distinctions varies, as discussed in the section on skeletonization

on page 112. Most water users have such a small individual impact that their with-
drawals can be assigned to nearby nodes without adversely affecting a model.
Figure 3.13
Pump off
a hydropneumatic
tank and a model tank
Veff
Equivalent
Model
Tank
Pressure Tank Pump on
HGLmax
Veff
HGLmin
Pump
Junction Elevation
Generally, the only physical characteristic defined at a junction node is its elevation.
This attribute may seem simple to define, but there are some considerations that need
to be taken into account before assigning elevations to junction nodes. Because pres-
sure is determined by the difference between calculated hydraulic grade and eleva-
tion, the most important consideration is, at what elevation is the pressure most
important?
Selecting an Elevation. Figure 3.14 represents a typical junction node, illus-

trating that at least four possible choices for elevation exist that can be used in the
model. The elevation could be taken as point A, the centerline of the pipe. Alterna-
tively, the ground elevation above the pipe (point B), or the elevation of the hydrant
(point C), may be selected. As a final option, the ground elevation at the highest ser-
vice point, point D, could be used. Each of these possibilities has associated benefits,
so the determination of which elevation to use needs to be made on a case-by-case
basis. Regardless of which elevation is selected, it is good practice to be consistent
within a given model to avoid confusion.
Figure 3.14
Elevation choices for
a junction node
High Service - D (650’) //=//= //=//=//=//=
//=
//=
//=
//=
/=
Hydrant //=/
Elevation - C (635’) =//=
/= //
//=/ Service
Ground - B (630’)
Line
Pipe - A (622’)
The elevation of the centerline of the pipe may be useful for determining pressure for
leakage studies, or it may be appropriate when modeling above-ground piping sys-
tems (such as systems used in chemical processing). Ground elevations may be the
easiest data to obtain and will also overlay more easily onto mapping systems that use
ground elevations. They are frequently used for models of municipal water distribu-
tion systems. Both methods, however, have the potential to overlook poor service
pressures because the model could incorrectly indicate acceptable pressures for a cus-
tomer who is notably higher than the ground or pipe centerline. In such cases, it may
be more appropriate to select the elevation based on the highest service elevation
required.
In the process of model calibration (see Chapter 7 for more about calibration), accu-
rate node elevations are crucial. If the elevation chosen for the modeled junction is not
the same as the elevation associated with recorded field measurements, then direct
pressure comparisons are meaningless. Methods for obtaining good node elevation
data are described in Walski (1999).
3.6 PIPES
A pipe conveys flow as it moves from one junction node to another in a network. In
the real world, individual pipes are usually manufactured in lengths of around 18 or
20 feet (6 meters), which are then assembled in series as a pipeline. Real-world pipe-
lines may also have various fittings, such as elbows, to handle abrupt changes in
direction, or isolation valves to close off flow through a particular section of pipe.
Figure 3.15 shows ductile iron pipe sections.
For modeling purposes, individual segments of pipe and associated fittings can all be
combined into a single pipe element. A model pipe should have the same characteris-
tics (size, material, etc.) throughout its length.
Section 3.6 Pipes 91
Figure 3.15
Ductile iron pipe
sections
Length
The length assigned to a pipe should represent the full distance that water flows from
one node to the next, not necessarily the straight-line distance between the end nodes
of the pipe.
Scaled versus Schematic. Most simulation software enables the user to indi-
cate either a scaled length or a user-defined length for pipes. Scaled lengths are auto-
matically determined by the software, or scaled from the alignment along an
electronic background map. User-defined lengths, applied when scaled electronic
maps are not available, require the user to manually enter pipe lengths based on some
other measurement method, such as use of a map wheel (see Figure 3.16). A model
using user-defined lengths is a schematic model. The overall connectivity of a sche-
matic model should be identical to that of a scaled model, but the quality of the plani-
metric representation is more similar to a caricature than a photograph.
Even in some scaled models, there may be areas where there are simply too many
nodes in close proximity to work with them easily at the model scale (such as at a
pump station). In these cases, the modeler may want to selectively depict that portion
of the system schematically, as shown in Figure 3.17.
Diameter
As with junction elevations, determining a pipe’s diameter is not as straightforward as
it might seem. A pipe’s nominal diameter refers to its common name, such as a 16-in.
(400-mm) pipe. The pipe’s internal diameter, the distance from one inner wall of the
pipe to the opposite wall, may differ from the nominal diameter because of manufac-
turing standards. Most new pipes have internal diameters that are actually larger than
the nominal diameters, although the exact measurements depend on the class (pres-
sure rating) of pipe.
Figure 3.16
Use of a map
measuring wheel for
measuring pipe
lengths
Figure 3.17
Scaled system with a
schematic of a pump
station
Scaled System
Pump Station Schematic

(not to scale)
For example, Figure 3.18 depicts a new ductile iron pipe with a 16-in. nominal diam-
eter (ND) and a 250-psi pressure rating that has an outside diameter (OD) of 17.40 in.
and a wall thickness (Th) of 0.30 in., resulting in an internal diameter (ID) of 16.80 in.
(AWWA, 1996).
To add to the confusion, the ID may change over time as corrosion, tuberculation, and
scaling occur within the pipe (see Figure 3.19). Corrosion and tuberculation are
related in iron pipes. As corrosion reactions occur on the inner surface of the pipe, the
reaction by-products expand to form an uneven pattern of lumps (or tubercules) in a
process called tuberculation. Scaling is a chemical deposition process that forms a
material build-up along the pipe walls due to chemical conditions in the water. For
Section 3.6 Pipes 93
example, lime scaling is caused by the precipitation of calcium carbonate. Scaling can
actually be used to control corrosion, but when it occurs in an uncontrolled manner it
can significantly reduce the ID of the pipe.
Figure 3.18
Cross-section of a
16-in. pipe
Figure 3.19
Pipe corrosion and
tuberculation
Courtesy of Donald V. Chase, Department of Civil Engineering, University of Dayton
Of course, no one is going to refer to a pipe as a 16.80-in. (426.72-mm) pipe, and

because of the process just described, it is difficult to measure a pipe’s actual internal
diameter. As a result, a pipe’s nominal diameter is commonly used in modeling, in
combination with a roughness value that accounts for the diameter discrepancy. How-
ever, using nominal rather than actual diameters can cause significant differences
Red Water
Distribution systems with unlined iron or steel A positive saturation index indicates that the pipe
pipes can be subject to water quality problems should be protected, provided that sufficient alka-
related to corrosion, referred to as red water. Red linity is present.
water is treated water containing a colloidal sus- More recently, corrosion inhibitors such as zinc
pension of very small, oxidized iron particles that orthophosphate and hexametaphosphate have
originated from the surface of the pipe wall. Over become popular in red water prevention (Ben-
a long period of time, this form of corrosion weak- jamin, Reiber, Ferguson, Vanderwerff, and Miller,
ens the pipe wall and leads to the formation of 1990; Mullen and Ritter, 1974; Volk, Dundore,
tubercles. The most obvious and immediate Schiermann, and LeChevallier, 2000). Several
impact, however, is that the oxidized iron particles theories exist concerning the predominant mecha-
give the water a murky, reddish-brown color. This nism by which these inhibitors prevent corrosion.
reduction in the aesthetic quality of the water
The effectiveness of corrosion control measures
prompts numerous customer complaints.
can be dependent on the hydraulic flow regime
Several alternative methods are available to con- occurring in the pipe. Several researchers have
trol the pipe corrosion that causes red water. The reported that corrosion inhibitors and carbonate
most traditional approach is to produce water that films do not work well in pipes with low velocities
is slightly supersaturated with calcium carbonate. (Maddison and Gagnon, 1999; McNeil and
When the water enters the distribution system, the Edwards, 2000). Water distribution models pro-
dissolved calcium carbonate slowly precipitates vide a way to identify pipes with chronic low veloc-
on the pipe walls, forming a thin, protective scale ities, and therefore more potential for red water
(Caldwell and Lawrence, 1953; Merrill and Sanks, problems. The effect of field operations meant to
1978). The Langelier Index (an index of the corro- control red water (for example, flushing and blow-
sive potential of water) can be used as an indica- offs) can also be investigated using hydraulic
tion of the potential of the water to precipitate model simulations.
calcium carbonate, allowing better management
of the precipitation rate (Langelier, 1936).
when water quality modeling is performed. Because flow velocity is related to flow
rate by the internal diameter of a pipe, the transport characteristics of a pipe are
affected. Chapter 7 discusses these calibration issues further (see page 255). Typical
roughness values can be found in Section 2.4.
Minor Losses
Including separate modeling elements to represent every fitting and appurtenance
present in a real-world system would be an unnecessarily tedious task. Instead, the
minor losses caused by those fittings are typically associated with pipes (that is, minor
losses are assigned as a pipe property).
In many hydraulic simulations, minor losses are ignored because they do not contrib-
ute substantially to the overall head loss throughout the system. In some cases,
however, flow velocities within a pipe and the configuration of fittings can cause
minor losses to be considerable (for example, at a pump station). The term “minor” is
relative, so the impact of these losses varies for different situations.
Section 3.7 Pumps 95
Composite Minor Losses. At any instant in time, velocity in the model is con-
stant throughout the length of a particular pipe. Since individual minor losses are
related to a coefficient multiplied by a velocity term, the overall head loss from sev-
eral minor losses is mathematically equivalent to having a single composite minor
loss coefficient. This composite coefficient is equal to the simple sum of the individual
coefficients.
3.7 PUMPS
A pump is an element that adds energy to the system in the form of an increased
hydraulic grade. Since water flows “downhill” (that is, from higher energy to lower
energy), pumps are used to boost the head at desired locations to overcome piping
head losses and physical elevation differences. Unless a system is entirely operated by
gravity, pumps are an integral part of the distribution system.
In water distribution systems, the most frequently used type of pump is the centrifugal
pump. A centrifugal pump has a motor that spins a piece within the pump called an
impeller. The mechanical energy of the rotating impeller is imparted to the water,
resulting in an increase in head. Figure 3.20 illustrates a cross-section of a centrifugal
pump and the flow path water takes through it. Water from the intake pipe enters the
pump through the eye of the spinning impeller (1) where it is then thrown outward
between vanes and into the discharge piping (2).
Figure 3.20
Cross-section of a
centrifugal pump
Frank M. White, Fluid Mechanics, 1994, McGraw-Hill, Inc. Reproduced by permission of the McGraw-Hill Companies.
Pump Characteristic Curves

With centrifugal pumps, pump performance is a function of flow rate. The perfor-
mance is described by the following four parameters, which are plotted versus
discharge.
• Head: Total dynamic head added by pump in units of length (see page 44)
• Efficiency: Overall pump efficiency (wire-to-water efficiency) in units of
percent (see pages 199 and 442)
• Brake horsepower: Power needed to turn pump (in power units)

• Net positive suction head (NPSH) required: Head above vacuum (in units
of length) required to prevent cavitation (see page 48)
Only the head curve is an energy equation necessary for solving pipe network prob-
lems. The other curves are used once the network has been solved to identify power
consumption (energy), motor requirements (brake horsepower), and suction piping
(NPSH).
Fixed-Speed and Variable-Speed Pumps. A pump characteristic curve is

related to the speed at which the pump motor is operating. With fixed-speed pumps,
the motor remains at a constant speed regardless of other factors. Variable-speed
pumps, on the other hand, have a motor or other device that can change the pump
speed in response to the system conditions.
A variable-speed pump is not really a special type of pump, but rather a pump con-
nected to a variable-speed drive or controller. The most common type of variable-
speed drive controls the flow of electricity to the pump motor, and therefore controls
the rate at which the pump rotates. The difference in pump speed, in turn, produces
different head and discharge characteristics. Variable-speed pumps are useful in appli-
cations requiring operational flexibility, such as when flow rates change rapidly, but
the desired pressure remains constant. An example of such a situation would be a net-
work with little or no storage available.
Power and Efficiency. The term power may have one of several meanings
when dealing with a pump. These possible meanings are listed below:
• Input power: The amount of power that is delivered to the motor, usually in
electric form
• Brake power: The amount of power that is delivered to the pump from the
motor
• Water power: The amount of power that is delivered to the water from the
pump
Of course, there are losses as energy is converted from one form to another (electricity
to motor, motor to pump, pump to water), and every transfer has an efficiency associ-
ated with it. The efficiencies associated with these transfers may be expressed either
as percentages (100 percent is perfectly efficient) or as decimal values (1.00 is per-
fectly efficient), and are typically defined as follows:
• Motor efficiency: The ratio of brake power to input power

• Pump efficiency: The ratio of water power to brake power
• Wire-to-water (overall) efficiency: The ratio of water power to input power
Pump efficiency tends to vary significantly with flow, while motor efficiency remains
relatively constant over the range of loads imposed by most pumps. Note that there
may also be an additional efficiency associated with a variable-speed drive. Some

engineers refer to the combination of the motor and any speed controls as the driver.
Figure 3.21 shows input power and wire-to-water efficiency curves overlaid on a typi-
cal pump head curve. Notice that the input power increases as discharge increases,
and head decreases as discharge increases. For each impeller size, there is a flow rate
corresponding to maximum efficiency. At higher or lower flows, the efficiency
decreases. This maximum point on the efficiency curve is called the best efficiency
point (BEP).
Figure 3.21
Pump curves with
efficiency, NPSH, and
horsepower overlays
Courtesy of Peerless Pumps
Obtaining Pump Data. Ideally, a water utility will have pump operating
curves on file for every pump in the system. These are usually furnished to the utility
with the shop drawings of the pump stations or as part of the manufacturer’s submit-
tals when replacing pumps. If the pump curve cannot be located, a copy of the curve
can usually be obtained from the manufacturer (provided the model and serial num-
bers for the pump are available).
To perform energy cost calculations, pump efficiency curves should also be obtained.
Note that the various power and efficiency definitions can be confusing, and it is
important to distinguish which terms are being referred to in any particular document.
Every pump differs slightly from its catalog model, and normal wear and tear will
cause a pump's performance to change over time. Thus, pumps should be checked to
verify that the characteristic curves on record are in agreement with field perfor-
mance. If an operating point does not agree with a characteristic curve, a new curve
can be developed to reflect the actual behavior. More information is available on this
subject in Chapter 5 (see page 199).
Positive Displacement Pumps

Virtually all water distribution system pumps are The suction side node would have a demand set
centrifugal pumps. However, pipe network models equal to the pump flow, while the discharge node
are used in other applications—such as chemical would have an inflow set equal to this flow. The
feeds, low-pressure sanitary sewer collection sys- model will then give the suction and discharge
tems, and sludge pumping—in which positive dis- HGLs and pressures at the nodes. (Custom
placement pumps (for example, diaphragm, extended curve options can also be used.)
piston, plunger, lobe, and progressive cavity Because the suction and discharge systems are
pumps) are used. Unlike centrifugal pumps, these separated, it is important for the modeler to
pumps produce a constant flow, regardless of the include a tank or reservoir on both the suction and
head supplied, up to a very high pressure. discharge sides of the pump. Otherwise, the
The standard approximations to pump curves model will not be able to satisfy the law of conser-
used in most models do not adequately address vation of mass. For example, if the demands on
positive displacement pumps because the head the discharge side do not equal the inflow to the
characteristic curve for such pumps consists of a discharge side, the model may not give a valid
virtually straight, vertical line. Depending on the solution. Because most models assume demands
model, forcing a pump curve to fit this shape usu- as independent of pressure, inflows must equal
ally results in warning messages. system demands, plus or minus any storage
An easy way to approximate a positive displace- effects. If no storage is present, the model cannot
ment pump in a model is to not include a pump at solve unless inflows and demands are equal.
all but rather to use two nodes—a suction node
and a discharge node—that are not connected.
Even though a pump curve on record may not perfectly match the actual pump char-
acteristics, many utilities accept that the catalogued values for the pump curve are
sufficiently accurate for the purposes of the model, and forgo any performance testing
or field verification. This decision is dependent on the specific situation.
Model Representation
In order to model a pump’s behavior, some mathematical expression describing its
pump head curve must be defined. Different models support different definitions, but
most are centered on the same basic concept, furnishing the model with sufficient
sample points to define the characteristic head curve.
Selecting Representative Points. As discussed previously, the relationship

between pump head and discharge is nonlinear. For most pumps, three points along
the curve are usually enough to represent the normal operating range of the pump.
These three points include
• The zero-discharge point, also known as the cutoff or shutoff point
• The normal operating point, which should typically be close to the best effi-
ciency point of the pump
• The point at the maximum expected discharge value
It is also possible to provide some models with additional points along the pump
curve, but not all models treat these additional data points in the same way. Some
models perform linear interpolation between points, some fit a polynomial curve
between points, and others determine an overall polynomial or exponential curve that
fits the entire data set.
Constant Power Pumps. Many models also support the concept of a constant
power pump. With this type of pump, the water power produced by the pump remains
constant, regardless of how little or how much flow the pump passes.
Water power is a product of discharge and head, which means that a curve depicting
constant water power is asymptotic to both the discharge and head axes, as shown in
Figure 3.22.
Figure 3.22
Characteristic pump
curve for a constant
Equivalent power pump
pump curve
Actual
Head
pump
curve
Discharge
Some modelers use a constant power pump definition to define a curve simply
because it is easier than providing several points from the characteristic curve, or
because the characteristic curve is not available. The results generated using this defi-
nition, however, can be unreliable and sometimes counter-intuitive. As shown in Fig-
ure 3.22, the constant power approximation will be accurate for a specific range of
flows, but not at very high or low flows. For very preliminary studies when all the
modeler knows is the approximate size of the pump, this approximation can be used
to get into pipe sizing quickly. However, it should not be used for pump selection.
The modeler must remember that the power entered for the constant power pump is
not the rated power of the motor but the water power added. For example, a 50 hp
motor that is 90 percent efficient, running at 80 percent of its rated power, and con-
nected to a pump that is operating at 70 percent efficiency will result in a water power
of roughly 25 hp (that is, 50 u 0.9 u 0.8 u 0.7 ). The value 25 hp, not 50 hp, should be
entered into the model.
Node versus Link Representation. A pump can be represented as a node or
a link element, depending on the software package. In software that symbolizes
pumps as links, the pump connects upstream and downstream nodes in a system the
same way a pipe would. A link symbolization more closely reflects the internal math-
ematical representation of the pump, but it can introduce inaccuracies. For example,
Figure 3.23 illustrates how the pump intake and discharge piping may be ignored and
the head losses occurring in them neglected.
Figure 3.23
Comparison of an
actual pump and a
pump modeled as a
link element
Other models represent pumps as nodes, typically with special connectivity rules (for
example, only allowing a single downstream pipe). This nodal representation is less
error-prone, more realistic, and easier for the modeler to implement. Nodal represen-
tation may also be more intuitive, since a real-world pump is usually thought of as
being in a single location with two distinct hydraulic grades (one on the intake side
and one on the discharge side). Figure 3.24 illustrates a nodal representation of a
pump.
Figure 3.24
Comparison of an
actual pump and a
pump modeled as a HGL HGL
node element
Real World Model (Node)
3.8 VALVES
A valve is an element that can be opened and closed to different extents (called throt-
tling) to vary its resistance to flow, thereby controlling the movement of water
through a pipeline (see Figure 3.25). Valves can be classified into the following five
general categories:
• Isolation valves
• Directional valves
• Altitude valves
• Air release and vacuum breaking valves
• Control valves
Section 3.8 Valves 101
Figure 3.25
Different valve types
Check Valve Gate Valve Butterfly Valve
Courtesy of Crane Co. All Rights Reserved.
Some valves are intended to automatically restrict the flow of water based on pres-
sures or flows, and others are operated manually and used to completely turn off por-
tions of the system. The behaviors of different valve types vary significantly
depending on the software used. This section provides an introduction to some of the
most common valve types and applications.
Isolation Valves
Perhaps the most common type of valve in water distribution systems is the isolation
valve, which can be manually closed to block the flow of water. As the term “isola-
tion” implies, the primary purpose of these valves is to provide a field crew with a
means of turning off a portion of the system to, for example, replace a broken pipe or
a leaky joint. Well-designed water distribution systems have isolation valves through-
out the network, so that maintenance and emergencies affect as few customers as pos-
sible. In some systems, isolation valves may be intentionally kept in a closed position
to control pressure zone boundaries, for example.
There are several types of isolation valves that may be used, including gate valves (the
most popular type), butterfly valves, globe valves, and plug valves.
In most hydraulic models, the inclusion of each and every isolation valve would be an
unnecessary level of detail. Instead, the intended behavior of the isolation valve
(minor loss, the ability to open and close, and so on) can be defined as part of a pipe.
A common question in constructing a model is whether to explicitly include minor

losses due to open gate valves, or to account for the effect of such losses in the Hazen-
Williams C-factor. If the C-factor for the pipe with no minor losses is known, an
equivalent C-factor that accounts for the minor losses is given by:
§ · 0.54
¨ L ¸
C e = C ¨ ------------------------------¸ (3.3)
¨ § 6K L-· ¸
© L + D © --------- f ¹¹
where Ce = equivalent Hazen-Williams C-factor accounting for minor losses

L = length of pipe segment (ft, m)
6 KL = sum of minor loss coefficients in pipe
For example, consider a 400-ft (122-m) segment of 6-in. (152-mm) pipe with a C-
factor of 120 and an f of 0.02. From Equation 3.3, the equivalent C-factor for the pipe
including a single open gate valve (KL = 0.39) is 118.4. For two open gate valves, the
equivalent C-factor is 116.9. Given that C-factors are seldom known to within plus or
minus 5, these differences are generally negligible. Note that if a model is calibrated
without explicitly accounting for many minor losses, then the C-factor resulting from
the calibration is the equivalent C-factor, and no further adjustment is needed.
Directional Valves
Directional valves, also called check valves, are used to ensure that water can flow in
one direction through the pipeline, but cannot flow in the opposite direction (back-
flow). Any water flowing backwards through the valve causes it to close, and it
remains closed until the flow once again begins to go through the valve in the forward
direction.
Simple check valves commonly use a hinged disk or flap to prevent flow from travel-
ing in the undesired direction. For example, the discharge piping from a pump may
include a check valve to prevent flow from passing through the pump backwards
(which could damage the pump). Most models automatically assume that every pump
has a built-in check valve, so there is no need to explicitly include one (see Figure
3.26). If a pump does not have a check valve on its discharge side, water can flow
backwards through the pump when the power is off. This situation can be modeled
with a pipe parallel to the pump that only opens when the pump is off. The pipe must
have an equivalent length and minor loss coefficient that will generate the same head
loss as the pump running backwards.
Figure 3.26
A check valve HGL
operating at a pump
HGL
Demand Demand
Check Valve Check Valve
Pump Off Pump On
Mechanically, some check valves require a certain differential in head before they will
seat fully and seal off any backflow. They may allow small amounts of reverse flow,
which may or may not have noteworthy consequences. When potable water systems
are hydraulically connected to nonpotable water uses, a reversal of flow could be
disastrous. These situations, called cross-connections, are a serious danger for water
distributors, and the possibility of such occurrences warrants the use of higher quality
check valves. Figure 3.27 illustrates a seemingly harmless situation that is a potential
cross-connection. A device called a backflow preventer is designed to be highly sensi-
tive to flow reversal, and frequently incorporates one or more check valves in series to
prevent backflow.
Figure 3.27
A potential cross-
connection
As far as most modeling software is concerned, there is no difference in sensitivity

between different types of check valves (all are assumed to close completely even for
the smallest of attempted reverse flows). As long as the check valve can be repre-
sented using a minor loss coefficient, the majority of software packages allow them to
be modeled as an attribute associated with a pipe, instead of requiring that a separate
valve element be created.
Altitude Valves
Many water utilities employ devices called altitude valves at the point where a pipe-
line enters a tank (see Figure 3.28). When the tank level rises to a specified upper
limit, the valve closes to prevent any further flow from entering, thus eliminating
overflow. When the flow trend reverses, the valve reopens and allows the tank to drain
to supply the usage demands of the system.
Most software packages, in one form or another, automatically incorporate the behav-
ior of altitude valves at both the minimum and maximum tank levels and do not
require explicit inclusion of them. If, however, an altitude valve does not exist at a
tank, tank overflow is possible, and steps must be taken to include this behavior in the
model.
Figure 3.28
Altitude valve
controlling the Max. Fill Level
maximum fill level of (Upper Limit)
a tank
Filling/Draining
//=//= //=//= //=//= //=//= //=//= //=//= //=//= //=//=

Vault Vault
No Flow
Altitude Valve Open Altitude Valve Closed
Air Release Valves and Vacuum Breaking Valves

Most systems include special air release valves to release trapped air during system
operation, and air/vacuum valves that discharge air upon system start-up and admit
air into the system in response to negative gage pressures (see Figure 3.29). These
types of valves are often found at system high points, where trapped air settles, and at
changes in grade, where pressures are most likely to drop below ambient or atmo-
spheric conditions. Combination air valves that perform the functions of both valve
types are often used as well.
Air release and air/vacuum valves are typically not included in standard water distri-
bution system modeling. The importance of such elements is significant, however, for
advanced studies such as transient analyses.
Figure 3.29
Air release and air/
vacuum valves
Air Release Valve Vacuum Breaking Valve
Courtesy of Val-Matic Valve and Manufacturing Corporation, Elmhurst, Illinois.
Control Valves
For any control valve, also called regulating valve, the setting is of primary impor-
tance. For a flow control valve, this setting refers to the flow setting, and for a throttle
control valve, it refers to a minor loss coefficient. For pressure-based controls, how-
ever, the setting may be either the hydraulic grade or the pressure that the valve tries
to maintain. Models are driven by hydraulic grade, so if a pressure setting is used, it is
critically important to have not only the correct pressure setting, but also the correct
valve elevation.
Given the setting for the valve, the model calculates the flow through the valve and
the inlet and outlet HGL (and pressures). A control valve is complicated in that,
unlike a pump, which is either on or off, it can be in any one of the several states
described in the following list. Note that the terminology may vary slightly between
models.
• Active: Automatically controlling flow
- Open: Opened fully
- Closed (1): Closed fully
- Throttling: Throttling flow and pressure
• Closed (2): Manually shut, as when an isolating valve located at the control
valve is closed
• Inactive: ignored
Because of the many possible control valve states, valves are often points where
model convergence problems exist.
Pressure Reducing Valves (PRVs). Pressure reducing valves (PRVs) throttle

automatically to prevent the downstream hydraulic grade from exceeding a set value,
and are used in situations where high downstream pressures could cause damage. For
example, Figure 3.30 illustrates a connection between pressure zones. Without a PRV,
the hydraulic grade in the upper zone could cause pressures in the lower zone to be
high enough to burst pipes or cause relief valves to open.
Figure 3.30
Schematic network
HGL
illustrating the use of
a pressure reducing
valve
Without PRV
Tank
Target Maximum Grade on
With PRV Hill
Higher
Service
Area
PRV
Lower
Service
Area
Unlike the isolation valves discussed earlier, PRVs are not associated with a pipe but
are explicitly represented within a hydraulic model. A PRV is characterized in a
model by the downstream hydraulic grade that it attempts to maintain, its controlling
status, and its minor loss coefficient. Because the valve intentionally introduces losses
to meet the required grade, a PRV's minor loss coefficient is really only a concern
when the valve is wide open (not throttling).
Like pumps, PRVs connect two pressure zones and have two associated hydraulic
grades, so some models represent them as links and some represent them as nodes.
The pitfalls of link characterization of PRVs are the same as those described previ-
ously for pumps (see page 99).
Pressure Sustaining Valves (PSVs). A pressure sustaining valve (PSV)
throttles the flow automatically to prevent the upstream hydraulic grade from drop-
ping below a set value. This type of valve can be used in situations in which unregu-
lated flow would result in inadequate pressures for the upstream portion of the system
(see Figure 3.31). They are frequently used to model pressure relief valves (see page
313).
Like PRVs, a PSV is typically represented explicitly within a hydraulic model and is
characterized by the upstream pressure it tries to maintain, its status, and its minor
loss coefficient.
Figure 3.31
Schematic network HGL
illustrating the use of Tank
a pressure sustaining Target Minimum Grade on
valve Hill
Without PSV
HGL
Higher
Service
With PSV Area
PSV
Lower
Service
Area
Flow Control Valves (FCVs). Flow control valves (FCVs) automatically throt-
tle to limit the rate of flow passing through the valve to a user-specified value. This
type of valve can be employed anywhere that flow-based regulation is appropriate,
such as when a water distributor has an agreement with a customer regarding maxi-
mum usage rates. FCVs do not guarantee that the flow will not be less than the setting
value, only that the flow will not exceed the setting value. If the flow does not equal
the setting, modeling packages will typically indicate so with a warning.
Similar to PRVs and PSVs, most models directly support FCVs, which are character-
ized by their maximum flow setting, status, and minor loss coefficient.
Section 3.9 Controls (Switches) 107
Throttle Control Valves (TCVs). Unlike an FCV where the flow is specified
directly, a throttle control valve (TCV) throttles to adjust its minor loss coefficient
based on the value of some other attribute of the system (such as the pressure at a crit-
ical node or a tank water level). Often the throttling effect of a particular valve posi-
tion is known, but the minor loss coefficients as a function of position are unknown.
This relationship can frequently be provided by the manufacturer.
Valve Books
Many water utilities maintain valve books, which are sets of records that provide
details pertaining to the location, type, and status of isolation valves and other fittings
throughout a system. From a modeling perspective, valve books can provide valuable
insight into the pipe connectivity at hydraulically complex intersections, especially in
areas where system maps may not show all of the details.
3.9 CONTROLS (SWITCHES)

Operational controls, such as pressure switches, are used to automatically change the
status or setting of an element based on the time of day, or in response to conditions
within the network. For example, a switch may be set to turn on a pump when pres-
sures within the system drop below a desired value. Or a pump may be programmed
to turn on and refill a tank in the early hours of the morning.
Without operational controls, conditions would have to be monitored and controlled
manually. This type of operation would be expensive, mistake-prone, and sometimes
impractical. Automated controls enable operators to take a more supervisory role,
focusing on issues larger than the everyday process of turning on a pump at a given
time or changing a control valve setting to accommodate changes in demand. Conse-
quently, the system can be run more affordably, predictably, and practically.
Models can represent controls in different ways. Some consider controls to be sepa-
rate modeling elements, and others consider them to be an attribute of the pipe, pump,
or valve being controlled.
Pipe Controls
For a pipe, the only status that can really change is whether the pipe (or, more accu-
rately, an isolation valve associated with the pipe) is open or closed. Most pipes will
always be open, but some pipes may be opened or closed to model a valve that auto-
matically or manually changes based on the state of the system. If a valve in the pipe
is being throttled, it should be handled either through the use of a minor loss directly
applied to the pipe or by inserting a throttle control valve in the pipe and adjusting it.
Pump Controls
The simplest type of pump control turns a pump on or off. For variable-speed pumps,
controls can also be used to adjust the pump’s relative speed factor to raise or lower
the pressures and flow rates that it delivers. For more information about pump relative
speed factors, see Chapter 2 (page 44).
The most common way to control a pump is by tank water level. Pumps are classified
as either “lead” pumps, which are the first to turn on, or “lag” pumps, the second to
turn on. Lead pumps are set to activate when tanks drain to a specified minimum level
and to shut off when tanks refill to a specified maximum level, usually just below the
tank overflow point. Lag pumps turn on only when the tank continues to drain below
the minimum level, even with the lead pump still running. They turn off when the tank
fills to a point below the shut off level for the lead pump. Controls get much more
complicated when there are other considerations such as time of day control rules or
parallel pumps that are not identical.
Regulating Valve Controls
Similar to a pump, a control valve can change both its status (open, closed, or active)
and its setting. For example, an operator may want a flow control valve to restrict flow
more when upstream pressures are poor, or a pressure reducing valve to open com-
pletely to accommodate high flow demands during a fire event.
Indicators of Control Settings
If a pressure switch setting is unknown, tank level charts and pumping logs may pro-
vide a clue. As shown in Figure 3.32, pressure switch settings can be determined by
looking at tank level charts and correlating them to the times when pumps are placed
into or taken out of service. Operations staff can also be helpful in the process of
determining pressure switch settings.
Figure 3.32
Correlation between
tank levels and pump
Pump Flow Rate
operation
Time
Tank Level
Time
Section 3.10 Types of Simulations 109
3.10 TYPES OF SIMULATIONS

After the basic elements and the network topology are defined, further refinement of
the model can be done depending on its intended purpose. There are various types of
simulations that a model may perform, depending on what the modeler is trying to
observe or predict. The two most basic types are
• Steady-state simulation: Computes the state of the system (flows, pres-
sures, pump operating attributes, valve position, and so on) assuming that
hydraulic demands and boundary conditions do not change with respect to
time.
• Extended-period simulation (EPS): Determines the quasi-dynamic behav-
ior of a system over a period of time, computing the state of the system as a
series of steady-state simulations in which hydraulic demands and boundary
conditions do change with respect to time.
Steady-State Simulation
As the term implies, steady-state refers to a state of a system that is unchanging in
time, essentially the long-term behavior of a system that has achieved equilibrium.
Tank and reservoir levels, hydraulic demands, and pump and valve operation remain
constant and define the boundary conditions of the simulation. A steady-state simula-
tion provides information regarding the equilibrium flows, pressures, and other vari-
ables defining the state of the network for a unique set of hydraulic demands and
boundary conditions.
Real water distribution systems are seldom in a true steady state. Therefore, the
notion of a steady state is a mathematical construct. Demands and tank water levels
are continuously changing, and pumps are routinely cycling on and off. A steady-state
hydraulic model is more like a blurred photograph of a moving object than a sharp
photo of a still one. However, by enabling designers to predict the response to a
unique set of hydraulic conditions (for example, peak hour demands or a fire at a par-
ticular node), the mathematical construct of a steady state can be a very useful tool.
Steady-state simulations are the building blocks for other types of simulations. Once
the steady-state concept is mastered, it is easier to understand more advanced topics
such as extended-period simulation, water quality analysis, and fire protection studies
(these topics are discussed in later chapters).
Steady-state models are generally used to analyze specific worst-case conditions such
as peak demand times, fire protection usage, and system component failures in which
the effects of time are not particularly significant.
Extended-Period Simulation
The results provided by a steady-state analysis can be extremely useful for a wide
range of applications in hydraulic modeling. There are many cases, however, for
which assumptions of a steady-state simulation are not valid, or a simulation is
required that allows the system to change over time. For example, to understand the
effects of changing water usage over time, fill and drain cycles of tanks, or the
response of pumps and valves to system changes, an extended-period simulation
(EPS) is needed.
It is important to note that there are many inputs required for an extended-period sim-
ulation. Due to the volume of data and the number of possible actions that a modeler
can take during calibration, analysis, and design, it is highly recommended that a
model be examined under steady-state situations prior to working with extended-
period simulations. Once satisfactory steady-state performance is achieved, it is much
easier to proceed into EPSs.
EPS Calculation Process. Similar to the way a film projector flashes a series of
still images in sequence to create a moving picture, the hydraulic time steps of an
extended-period simulation are actually steady-state simulations that are strung
together in sequence. After each steady-state step, the system boundary conditions are
reevaluated and updated to reflect changes in junction demands, tank levels, pump
operations, and so on. Then, another hydraulic time step is taken, and the process con-
tinues until the end of the simulation.
Simulation Duration. An extended-period simulation can be run for any length

of time, depending on the purpose of the analysis. The most common simulation dura-
tion is typically a multiple of 24 hours, because the most recognizable pattern for
demands and operations is a daily one. When modeling emergencies or disruptions
that occur over the short-term, however, it may be desirable to model only a few hours
into the future to predict immediate changes in tank level and system pressures. For
water quality applications, it may be more appropriate to model a duration of several
days in order for quality levels to stabilize.
Even with established daily patterns, a modeler may want to look at a simulation
duration of a week or more. For example, consider a storage tank with inadequate
capacity operating within a system. The water level in the tank may be only slightly
less at the end of each day than it was at the end of the previous day, which may go
unnoticed when reviewing model results. If a duration of one or two weeks is used,
the trend of the tank level dropping more and more each day will be more evident.
Even in systems that have adequate storage capacity, a simulation duration of 48
hours or longer can be helpful in better determining the tank draining and filling char-
acteristics.
Hydraulic Time Step. An important decision when running an extended-period

simulation is the selection of the hydraulic time step. The time step is the length of
time for one steady-state portion of an EPS, and it should be selected such that
changes in system hydraulics from one increment to the next are gradual. A time step
that is too large may cause abrupt hydraulic changes to occur, making it difficult for
the model to give good results.
For any given system, predicting how small the time increment should be is difficult,
although experience is certainly beneficial in this area. Typically, modelers begin by
assuming one-hour time steps, unless there are considerations that point to the need
for a different time step.
Section 3.10 Types of Simulations 111
Why Use a Scenario Manager?

When water distribution models were first created, For example, a scenario may consist of the peak
data were input into the computer program by hour demands in 2020 paired with infrastructure
using punch cards, which were submitted and data that includes a proposed tank on Washington
processed as a batch run. In this type of run, a Hill and a new 16-in. (400 mm) pipe along North
separate set of input data was required to gener- Street. This idea of building model runs from alter-
ate each set of results. Because a typical model- native data sets created by the user is more intui-
ing project requires analysis of many alternative tive than the batch run concept, and is consistent
situations, large amounts of time were spent cre- with the object-oriented paradigm found in mod-
ating and debugging multiple sets of input cards. ern programs. Further, descriptive naming of sce-
When data files replaced punch cards, the batch narios and alternative data sets provides internal
approach to data entry was carried over. The documentation of the user’s actions.
modeler could now edit and copy input files more Because alternative plans in water modeling tend
easily, but there was still the problem of trying to to grow out of previous alternatives, a good sce-
manage a large number of model runs. Working nario manager will use the concept of inheritance
with many data files or a single data file with doz- to create new child alternatives from existing par-
ens of edits was confusing, inefficient, and error- ent alternatives. Combining this idea of inherit-
prone. ance with construction of scenarios from
The solution to this problem is to keep alternative alternative data sets gives the model user a self-
data sets within a single model data file. For documenting way to quickly create new and better
example, data for current average day demands, solutions based on the results of previous model
maximum day demands with a fire flow at node runs.
37, and peak hour demands in 2020 can be cre- A user accustomed to performing batch runs may
ated, managed, and stored in a central database. find some of the terminology and concepts
Once this structure is in place, the user can then employed in scenario management a bit of a chal-
create many runs, or scenarios, by piecing lenge at first. But, with a little practice, it becomes
together alternative data sets. difficult to imagine building or maintaining a model
without this versatile feature.
When junction demands and tank inflow/outflow rates are highly variable, decreasing
the time step can improve the accuracy of the simulation. The sensitivity of a model to
time increment changes can be explored by comparing the results of the same analysis
using different increments. This sensitivity can also be evaluated during the calibra-
tion process. Ultimately, finding the correct balance between calculation time and
accuracy is up to the modeler.
Intermediate Changes. Of course, changes within a system don’t always

occur at even time increments. When it is determined that an element’s status changes
between time steps (such as a tank completely filling or draining, or a control condi-
tion being triggered), many models will automatically report a status change and
results at that intermediate point in time. The model then steps ahead in time to the
next even increment until another intermediate time step is required. If calculations
are frequently required at intermediate times, the modeler should consider decreasing
the time increment.
Other Types of Simulations

Using the fundamental concepts of steady-state and extended-period simulations,
more advanced simulations can be built. Water quality simulations are used to ascer-
tain chemical or biological constituent levels within a system or to determine the age
or source of water (see page 61). Automated fire flow analyses establish the suitability
of a system for fire protection needs. Cost analyses are used for looking at the mone-
tary impact of operations and improvements. Transient analyses are used to investi-
gate the short-term fluctuations in flow and pressure due to sudden changes in the
status of pumps or valves (see page 573).
With every advance in computer technology and each improvement in software meth-
ods, hydraulic models become a more integral part of designing and operating safe
and reliable water distribution systems.
3.11 SKELETONIZATION
Skeletonization is the process of selecting for inclusion in the model only the parts of
the hydraulic network that have a significant impact on the behavior of the system.
Attempting to include each individual service connection, gate valve, and every other
component of a large system in a model could be a huge undertaking without a signif-
icant impact on the model results. Capturing every feature of a system would also
result in tremendous amounts of data; enough to make managing, using, and trouble-
shooting the model an overwhelming and error-prone task. Skeletonization is a more
practical approach to modeling that allows the modeler to produce reliable, accurate
results without investing unnecessary time and money.
Eggener and Polkowski (1976) did the first study of skeletonization when they sys-
tematically removed pipes from a model of Menomonie, Wisconsin, to test the sensi-
tivity of model results. They found that under normal demands, they could remove a
large number of pipes and still not affect pressure significantly. Shamir and Hamberg
(1988a, 1988b) investigated rigorous rules for reducing the size of models.
Skeletonization should not be confused with the omission of data. The portions of the
system that are not modeled during the skeletonization process are not discarded; rather,
their effects are accounted for within parts of the system that are included in the model.
Skeletonization Example
Consider the following proposed subdivision, which is tied into an existing water sys-
tem model. Figures 3.33, 3.34, 3.35, and 3.36 show how demands can be aggregated
from individual customers to nodes with larger and larger nodal service areas.
Although a modeler would almost never include the individual connections as shown
in Figure 3.33, this example, which can be extrapolated to much larger networks,
shows the steps that are followed to achieve various levels of skeletonization.
As depicted in the network segment in Figure 3.33, it is possible to not skeletonize at
all. In this case, there is a junction at each service tap, with a pipe and junction at each
house. There are also junctions at the main intersections, resulting in a total of nearly
50 junctions (not including those required for fire hydrants).
Section 3.11 Skeletonization 113
Figure 3.33
An all-link network
0.3 0.2 0.4 0.3 0.2 0.2 0.3
0.4
0.3 0.4
0.3 0.2 0.3 0.3 0.4
0.3
0.3 0.3 0.2 0.2 0.3
0.3
0.6
0.3
0.3 0.3 0.2 0.2 0.4
Demand in gpm
The same subdivision could be modeled again, but slightly more skeletonized. Instead
of explicitly including each household, only the tie-ins and main intersections are
included. This level of detail results in a junction count of less than 20 (Figure 3.34).
Note that in this level of skeletonization, hydraulic results for the customer service
lines would not be available since they were not included in the model. If results for
service lines are not important, then the skeletal model shown in Figure 3.34 repre-
sents an adequate level of detail.
Figure 3.34
Minimal
skeletonization
The system can be skeletonized even more, modeling only the ends of the main piping
and the major intersections (Figure 3.35). Attributing the demands to the junctions
becomes a little trickier since a junction is not being modeled at each tap location.
The demands for this model are attributed to the junction nearest to the service (fol-
lowing the pipeline). The dashed boundary areas indicate the contributing area for
each model junction. For example, the junction in the upper right will be assigned the
demand for eight houses, while the lower right junction has demands for ten houses,
and so on.
Figure 3.35
Moderate
skeletonization 3.5 gpm
2.5 gpm
2.7 gpm
An even greater level of skeletonization can be achieved using just a single junction
node where the subdivision feeds from the existing system. The piping within the
entire subdivision has been removed, with all demands being attributed to the remain-
ing junction (see Figure 3.36). In this case, the model will indicate the impact of the
demands associated with the subdivision on the overall hydraulic network. However,
the modeler will not be able to determine how pressures and flows vary within the
subdivision.
Figure 3.36
Maximum
skeletonization
8.7 gpm
An even broader level of skeletonization is possible in which even the junction node
where the subdivision piping ties into the main line is excluded. The subdivision
demands would simply be added to a nearby junction, where other effects may be
combined with those from several other subdivisions that also have not been included
in detail. As this example demonstrates, the extent of skeletonization depends on the
intended use of the model and, to a large degree, is subject to the modeler’s discretion.
Skeletonization Guidelines
There are no absolute criteria for determining whether a pipe should be included in
the model, but it is safe to say that all models are most likely skeletonized to some
degree. Water distribution networks vary drastically from one system to another, and
modeling judgment plays a large role in the creation of a solution. For a small-
diameter system, such as household plumbing or a fire sprinkler system, small
differences in estimated flow rate may have perceptible effects on the system head
losses. For a large city system, however, the effects of water demanded by an entire
subdivision may be insignificant for the large-transmission main system.
Opposing Philosophies. There are definitely opposing philosophies regarding
skeletonization that stem from different modeling perspectives. Some modelers assert
that a model should never be bigger than a few hundred elements, because no one can
possibly digest all of the data that pours out of a larger model. Others contend that a
model should include all the pipes, so that data-entry can be done by less skilled per-
sonnel, who will not need to exercise judgment about whether or not an element
should be included. Followers of this approach then use database queries, automated
consolidation algorithms, and demand allocation procedures (see page 136) to gener-
ate skeletonized models for individual applications.
Somewhere in the Middle. Most network models, however, fall somewhere
between the two extremes. The level of skeletonization used depends on the intended
use of the model. At one extreme, energy operation studies require minimal detail,
while determining available fire flow at individual hydrants requires the most. For
master planning or regional water studies, a broader level of skeletonization will typi-
cally suffice. For detailed design work or water quality studies, however, much more
of the system needs to be included to accurately model the real-world system.
The responsibility really comes back to the modeler, who must have a good under-
standing of the model’s intended use and must select a level of detail appropriate for
that purpose. Most modelers choose to develop their own skeletonization guidelines.
Elements of High Importance

Any elements that are important to the system or can potentially influence system
behavior should be included in the model. For most models this criterion includes
• Large water consumers
• Points of known conditions, such as sampling points
• Critical points with unknown conditions
• Large-diameter pipes
• Pipes that complete important loops
• Pumps, control valves, tanks, and other controlling elements
Elements of Unknown Importance

If the modeler is unsure what the effects of including or excluding specific elements
may be, there is a very simple method that can be used to find out exactly what the
effects are on the system. Run the model and see what happens.
A base skeleton can be created using experience and judgment, with pipes of ques-
tionable importance included. The model should be run over a range of study condi-
tions and the results noted. One or more questionable pipes can then be closed
(preventing them from conveying water) and the model run again. If the modeler
determines that the results from the two analyses are essentially the same, then the
pipes apparently did not have a significant effect on the system and can be removed
from the skeleton.
If a pipe’s level of significance cannot be determined or is questionable, it is usually
better to leave the pipe in the model. With older, nongraphical interfaces, it was often
desirable to limit the number of pipes as much as possible to prevent becoming lost in
the data. With the advanced computers and easy-to-use software tools of today, how-
ever, there are fewer reasons to exclude pipes from the model.
Automated Skeletonization
An increasing number of water utilities are linking their models to GIS systems and
even creating models from scratch by importing data from their GIS. However, there
are generally far more GIS elements than the user would want pipes in the model pre-
senting an obstacle for a smooth data conversion process. For example, Figure 3.37
shows how a single pipe link from a model can correspond to a large number of GIS
elements. The number of pipes in the GIS is even greater when each hydrant lateral
and service line is included in the GIS. Of course, the modeler can manually eliminate
pipes from the model, but this task can be extremely tedious and error prone, espe-
cially if it must be repeated for several time periods or planning scenarios. Thus, it is
highly desirable and clearly more efficient to automate the process of model skeleton-
ization.
Figure 3.37
GIS pipes versus
model pipes
Simply removing pipes and nodes from a model based on a rule, such as pipe size, is a
straightforward process. The process becomes complicated, however, when it is nec-
essary to also keep track of the demands (and associated demand patterns) and emitter
coefficients that were assigned to the nodes being removed, and it becomes even more
complicated when one tries to account for the hydraulic capacity of the pipes being
removed.
Skeletonization is not a single process but several different low-level element removal
processes that must be applied in series to ensure that the demands are logically
brought back to their source of supply. The skeletonization process also involves
developing rules for pumps, tanks, and valves, and deciding which pipes and nodes
should be identified as nonremovable.
As with manual skeletonization, the degree to which a system is skeletonized depends
on the type of raw data and the ultimate purpose of the model. If the raw data are a
complete GIS of the system including service lines and hydrant laterals and the model
is going to be used to set up pump controls or study energy costs, it may be possible to
remove the overwhelming majority of the pipes. On the other hand, if the model was
built manually from distribution maps and the model is to be used to determine avail-
able fire flow at every hydrant, then there may be little room for skeletonization.
The individual processes involved with skeletonization are discussed in the subsec-
tions that follow.
Simple Pipe Removal. The simplest type of pipe removal is when pipes are
simply removed from the system based on size or other criteria without any consider-
ation of their effects on demand loading or hydraulic capacity. This can be useful
when importing data from a GIS if the dataset contains service lines and hydrant later-
als. This type of pipe removal is usually practiced before demands are assigned to
model nodes (as a preprocessing step), although that is not always the case. Some
models that claim to perform automated skeletonization only perform this type of
skeletonization process.
Removing Branch Pipes. The next simplest type of skeletonization consists of
removing dead-end branches that do not contain tanks at the end. This process is
referred to as branch trimming, or branch collapsing, and the user needs to determine
whether some finite number of branches should be trimmed or if the network should
be trimmed back to a pipe that is part of a loop. Figure 3.38 shows how a branch is
trimmed back to a node that is part of a loop. When dead-end branch pipes are
removed, the removal has no effect on the carrying capacity of the remainder of the
system.
Figure 3.38
Branch pipe removal
J-13
(branch trimming) J-12
Q=8
Q=4
J-11
Q=5
J-10
Before Trimming Q=10
J-10
After Trimming Q=27
Removing Pipes in Series (with no other pipes connected to the

common node). In most cases, removing pipes in series (sometimes called pipe
merging) has a negligible effect on model performance. For example, in Figure 3.39,
pipes P-121 and P-122 can be combined to form a new P-121. In this example, the
demand (Q) at J-12 is split evenly between the two nodes at the ends of the resulting
pipe. Depending on the situation, however, other rules regarding demands can be
applied. For example, either of the nodes could receive all the demand, or the demand
could be split according to user-specified rules.
Figure 3.39
Series pipe removal –
similar attributes
J-11 J-12 J-13
Q=5 Q=8 Q=5
P-121 P-122
L=350 L=250
D=8 D=8
C=120 C=120
Before Series Pipe Removal
J-11 J-13
Q=9 Q=9
P-121
L=600
D=8
C=120
After Series Pipe Removal
If the node between two pipes in series has a large demand, removing it may
adversely impact the model results. To prevent such situations, the modeler may con-
sider setting a limit on flows such that nodes that exceed the limit cannot be elimi-
nated.
A key issue in combining two pipes into one lies in determining the attributes of the
resulting pipe. In Figure 3.39, the length of the resulting pipe is equal to the sum of
the lengths of the two pipes being combined and because the two pipes have the same
diameter and C-factor, the resulting pipe also has the same diameter and C-factor.
The problem becomes more complicated when the two pipes have different attributes,
as shown in Figure 3.40. In this case, the length is still the sum of the length of the two
pipes, but now there are an infinite number of combinations of diameter and C-factor
that would produce the same head loss through the pipe. As an option, the modeler can
choose to use the diameter and C-factor of one of the pipes as the attributes for the
resulting pipe. Or, the modeler can pick either the C-factor or the diameter for the result-
ing pipe and then calculate the other property. For example, if the modeler specifies
the diameter, then the C-factor of the resulting pipe can be given by Equation 3.4.
§ L r · 0.54 § Li · – 0.54
C r = ¨ -----------
-¸ ¨ ¦ ------------------------
- ¸ (3.4)
© D 4.87
r
¹ © i D 4.87
i Ci ¹
1.85
where L = length (ft, m)

D = diameter (in., m)
r = subscript referring to resulting pipe
i = subscript referring to the i-th pipe being combined
The mathematics are considerably more complicated when using the Darcy-Weisbach
equation.
In Figure 3.40, the length of the resulting pipe is 600 ft, so that if an 8-in. diameter
pipe was used, the Hazen-Williams C-factor of that pipe would be 55, and if a 6-in.
diameter was used, the C-factor would be 118. Either of these values will give the cor-
rect head loss. Minor loss coefficients and check valves can then be assigned to the
resulting pipe if needed.
Figure 3.40
Series pipe removal –
different attributes
Removing Parallel Pipes. Another way to skeletonize a system is to remove

parallel pipes. (Two pipes are considered to be in parallel if they have the same begin-
ning and ending nodes.) When removing parallel pipes, one of the pipes is considered
to be the dominant pipe and the length and either the diameter or C-factor from that
pipe is used for the new equivalent pipe. Depending on whether the diameter or C-
factor is used from the dominant pipe, the other parameter is calculated using equiva-
lent pipe formulas. For example, if the diameter of the dominant pipe is used, then the
C-factor is given by the following equation:
0.54 2.63
Lr Ci Di
C r = -----------
-
2.63 ¦
------------------ (3.5)
0.54
Dr i Li
In Figure 3.41, the length and diameter of P-40 are kept, but to account for the
removal of P-41, the capacity of P-40 is increased by increasing the C-factor.
Other factors to consider when removing parallel pipes are check valves and minor
losses. If both pipes have check valves, then the resulting pipe should also have a
check valve. Accurately assigning minor loss coefficients when determining equiva-
lent pipes, however, can be more difficult. In most cases, assigning some average
value does not cause a significant error.
Figure 3.41
Removing parallel
pipes
Removing Pipes to Break Loops. The types of pipe removal described in

the preceding sections can reduce the complexity of a model somewhat, but to dra-
matically reduce system size for typical water distribution systems, it is necessary to
actually break loops. Although two parallel pipes are considered a loop, they can be
handled with the basic action described in the previous section, and the hydraulic
capacity can be accounted for using Equation 3.5. This section applies to loops with
more than two attachments to the remainder of the system.
Consider the three-pipe loop in Figure 3.42 made up of pipes P-31, P-32, and P-33.
Removing any pipe in the loop can possibly result in a branch system that can be fur-
ther skeletonized using the methods described previously. However, in contrast to the
unique solutions that result from the pipe removal operations described in the preced-
ing sections, the results of breaking this loop by removing a pipe are different depend-
ing on which pipe is removed because the removal can have an impact on the carrying
capacity of the rest of the system.
Figure 3.42
Removing pipes in
loop
Therefore, there needs to be a rule for determining which pipe should be removed
first. Usually, it is best to remove the pipe with the least carrying capacity, which may
be defined as the smallest or the pipe with the minimum value of the quantity
2.63
CD
----------------- (3.6)
0.54
L
In Figure 3.42, pipe P-33 has the lowest carrying capacity so its removal should have
the least adverse impact to the carrying capacity of the system.
It is important to note that removing the pipe with the least carrying capacity does not
always do the least harm to the overall accuracy of the model. In some cases, a pipe
with very little capacity may be very important in some scenario and may need to be
kept in spite of its low carrying capacity.
Summary of Basic Pipe Removals. The results of the possible pipe

removal actions can be summarized as shown in Table 3.2. The first three actions are
fairly simple in that the system will end up with the correct flows and head loss. With
the fourth removal action, however, some carrying capacity is lost and removing one
pipe from a loop will give a different carrying capacity than removal of a different
pipe.
Table 3.2 Summary of pipe removal actions

Action Effect on Node Loss of System Capacity
Remove branch pipe Removes node No
Remove pipe in series Removes node No
Remove pipe in parallel No nodes removed No
Remove pipe from loop No nodes removed Yes
Removing Nonpipe Elements. Removing link-type elements other than fully

open pipes can be problematic, thus special rules must be developed for handling the
skeletonization of other network elements including pumps, tanks, closed pipes, and
valves.
A closed pipe or a pump that is not running has essentially already been skeletonized
out of the system and any effort to skeletonize it is trivial. If the element may be open,
however, then it should be treated as being open during the removal process.
Some other rules regarding the skeletonization of other network elements include the
following:
• When loads are being aggregated from removed nodes, they cannot be
passed through pumps, control valves, check valves, or closed valves.
• Pumps, control valves, and check valves in a branch can be trimmed and rep-
resented as an outflow from the remaining upstream system.
• Pumps, control valves, and check valves can be removed from series, paral-
lel, or looped systems only if their effect can be accounted for, which is usu-
ally difficult.
• If there is a check valve on a pipe in series, the resulting pipe must also have
the check valve.
• Tanks are usually too important to be removed during skeletonization and no

pipes connected to tanks should be removed.
Complex Skeletonization. Skeletonizing a real system involves applying the

basic removal actions in a sequence. In general, it is best to perform the skeletonizing
actions in the order given in Table 3.2. First, remove all dead-ends or branch pipes,
then remove series pipes, then combine parallel pipes, and finally, remove loops.
After each action, it is necessary to review the network because the previous action
may have created a dead-end or a parallel pipe that did not exist previously.
Figure 3.43, which shows a network being reduced, illustrates these actions. The net-
work looks like a dead-end branch and if one were doing the skeletonization manu-
ally, a modeler would simply add together the demands and place them on node J-10.
However, it is difficult for a computer to recognize that this is a branch, and it must
first eliminate series pipes and loops to identify the branch.
Figure 3.43
Steps for applying
automatic
skeletonization to P-53
complex pipe systems P-52 P-53 P-52
P-54
P-51 P-51
J-10 J-10
Original System Remove Pipe P-54 in Series
P-52
P-51
J-10 J-10
Remove Parallel Pipe P-53 Remove Branch Pipes P-51 and P-52
Stopping Criteria. Using the basic steps described in the preceding sections,
automated skeletonization can reduce any network to a handful of tanks and pumps.
In most cases, however, a user would not want this much reduction. The key to stop-
ping the skeletonization lies in defining criteria for links and nodes not to be skeleton-
ized.
Usually, the user will specify that all pipes with a certain diameter or larger will not be
skeletonized. This preserves the larger pipes in the system. The user can also specify
that certain pipes, especially those that close loops, are not to be removed (or that the
basic action of removing pipes from loops will not be carried out at all). The user can
also specify that if a pipe removal removes a node with greater than a specified
demand, then that removal action will not be carried out.
After these limits are set, the skeletonization process continues until it results in a sys-
tem that is skeletonized to the level specified by the user.
Skeletonization Conclusions
No hard and fast rules exist regarding skeletonization. It all depends on perspective
and the intended use of the model. For a utility that operates large transmission mains
and sells water to community networks, a model may be skeletonized to include only
Section 3.12 Model Maintenance 125
the source and large-diameter pipes. For a community that receives water from that
utility, the opposite may be true. Although most planning and analysis activities can
be performed successfully with a moderately skeletonized model, local fire flow eval-
uations and water quality analyses call for little to no skeletonization.
3.12 MODEL MAINTENANCE

Once a water distribution model is constructed and calibrated, it can be modified to
simulate and predict system behavior under a range of conditions. The model repre-
sents a significant investment on the part of the utility, and that investment should be
maximized by carefully maintaining the model for use well into the future.
Good record-keeping that documents model runs and history is necessary to ensure
that the model is used correctly by others or at a later date, and that time is not wasted
in deciphering and reconstructing what was done previously. There should be notes in
the model files or paper records indicating the state of the system in the various model
versions. These explanations will help subsequent users determine the best model run
to use as a starting point in future analyses.
Although the initial calibrated model reflects conditions in the current system, the
model is frequently used to test future conditions and alternative piping systems. The
scenario manager features in modeling software (see page 111) enable the user to
maintain the original model while keeping track of numerous proposed changes to the
system, some of which are never constructed. Eventually, a model file may contain
many “proposed” facilities and demands that fall into the following categories:
• Installed
• Under design or construction
• To be installed later
• Never to be installed
The user needs to periodically update the model file so that installed piping is accu-
rately distinguished from proposed facilities, and that facilities that will most likely
never be installed are removed from the model. The modeler also needs to be in regu-
lar contact with operations personnel to determine when new piping is placed into ser-
vice. Note that there may be a substantial lag between the time that a pipe or other
facility is placed into service, and the time that facility shows up in the system map or
GIS.
Once a master plan or comprehensive planning study is completed, model use typi-
cally becomes sporadic, though the model will still be used to respond to developer
inquiries, address operations problems, and verify project designs. Each of these spe-
cial studies involves creating and running additional scenarios. A single model even-
tually becomes cluttered with extraneous data on alternatives not selected.
A good practice in addressing these special studies is to start from the existing model
and create a new data file that will be used to study alternative plans. Once the project
design is complete, the facilities and demands associated with the selected plan
should be placed into the main model file as future facilities and demands. The ver-
sion of the model used for operational studies should not be updated until the facilities
are actually placed into service.
REFERENCES
American Water Works Association (1996). “Ductile Iron Pipe and Fitting.” AWWA Manual M-41, Denver,
Colorado.
Benjamin, M. M., Reiber, S. H., Ferguson, J. F., Vanderwerff, E. A., and Miller, M. W. (1990). Chemistry of
corrosion inhibitors in potable water. AWWARF, Denver, Colorado.
Caldwell, D. H., and Lawrence, W. B. (1953). “Water Softening and Conditioning Problems.” Industrial
Engineering Chemistry, 45(3), 535.
Eggener, C. L., and Polkowski, L. (1976). “Network Modeling and the Impact of Modeling Assumptions.
Journal of the American Water Works Association, 68(4), 189.
ESRI. (2001). “What is a GIS?” http://www.esri.com/library/gis/abtgis/what_gis.html.
Langelier, W. F. (1936). “The Analytical Control of Anti-Corrosion in Water Treatment.” Journal of the
American Water Works Association, 28(10), 1500.
Maddison, L. A., and Gagnon, G. A. (1999). “Evaluating Corrosion Control Strategies for a Pilot-Scale
Distribution System.” Proceedings of the Water Quality Technology Conference, American Water
Works Association, Denver, Colorado.
McNeil, L. S., and Edwards, M. (2000). “Phosphate Inhibitors and Red Water in Stagnant Iron Pipes.”
Journal of Environmental Engineering, ASCE, 126(12), 1096.
Merrill, D. T. and Sanks, R. L. (1978). Corrosion Control by Deposition of CaCO3 Films. AWWA, Denver,
Colorado.
Mullen, E. D., and Ritter, J. A. (1974). “Potable-Water Corrosion Control.” Journal of the American Water
Works Association, 66(8), 473.
Shamir, U. and Hamberg, D. (1988a). “Schematic Models for Distribution Systems Design I: Combination
Concept.” Journal of Water Resources Planning and Management, ASCE, 114(2), 129.
Shamir, U. and Hamberg, D. (1988b). “Schematic Models for Distribution Systems Design II: Continuum
Approach.” Journal of Water Resources Planning and Management, ASCE, 114(2), 141.
Volk, C., Dundore, E., Schiermann, J., LeChevallier, M. (2000). “Practical Evaluation of Iron Corrosion
Control in a Drinking Water Distribution System.” Water Research, 34(6), 1967.
Walski, T. M. (1999). “Importance and Accuracy of Node Elevation Data.” Essential Hydraulics and
Hydrology, Haestad Press, Waterbury, Connecticut.
Walski, T. M. (2000). “Hydraulic Design of Water Distribution Storage Tanks.” Water Distribution System
Handbook, Mays L. W., ed., McGraw Hill, New York, New York.
3.1 Manually find the flow rate through the system shown in the figure and compute the pressure at node
J-1. Also, find the suction and discharge pressures of the pump if it is at an elevation of 115 ft. Use
the Hazen-Williams equation to compute friction losses. Assume hP is in ft and Q is in cfs.
300 ft
R -B
Pipe 3:
L=1,000 ft
D=12 in.
C=120
Pipe 2:
L=2,200 ft J-1
hP = 225 - 10Q
1.50
D=12 in. Elev = 150 ft
C=120
125 ft
R-A
Pipe 1:
L=220 ft
D=16 in.
C=120
3.2 Manually find the flow in each pipeline and the pressure at node J-1 for the system shown in the fig-
ure. Assume that hP is in m and Q is in m3/s and note the demand at junction J-1 of 21.2 l/s. Use the
Hazen-Williams equation to compute friction losses.
Hint: Express the flow in Pipe 3 in terms of the flow in Pipe 1 or Pipe 2.
91.4 m
R-B
Pipe 3:
L=304.8 m
D=305 mm
C=120
Pipe 2: J -1
L=670.6 m Elev = 45.7 m
hP = 68.58 - 639.66Q
1.50
D=305 mm
C=120
Q = 21.2 l/s
38.1 m
R-A Pipe 1:
L=67.1 m
D=406 mm
C=120
3.3 English Units: Manually find the discharge through each pipeline and the pressure at each junction
node of the rural water system shown in the figure. Physical data for this system are given in the
tables that follow. Fill in the tables at the end of the problem.
P -12
J-12 J-11 J-10
P -11 P -10
P -8 P -9
J-8 J-7 J-9
J-5
J-6
P -7
P -5
P -4 P -6
J-3
J-4
P -3
R -1
P -1 P -2
(Not To Scale) J-1 J-2


Pipe Label
(ft) (in.) C-factor
P-1 500 10 120
P-2 1,200 6 120
P-3 4,200 10 120
P-4 600 6 110
P-5 250 4 110
P-6 500 4 100
P-7 5,200 8 120
P-8 4,500 4 100
P-9 5,500 3 90
P-10 3,000 6 75
P-11 570 6 120
P-12 550 4 80
Elevation Demand
Node Label
(ft) (gpm)
R-1 1050 N/A
J-1 860 40
J-2 865 15
J-3 870 30
J-4 875 25
J-5 880 5
J-6 885 12
J-7 880 75
J-8 850 25
J-9 860 0
J-10 860 18
J-11 850 15
J-12 845 10
Flow Head loss

Pipe Label
(gpm) (ft)
P-1
P-2
P-3
P-4
P-5
P-6
P-7
P-8
P-9
P-10
P-11
P-12
HGL Pressure
Node Label
(ft) (psi)
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
J-11
J-12
SI Units: Manually find the discharge through each pipeline and the pressure at each junction node
of the rural water system shown in the figure. Physical data for this system are given in the tables
that follow. Fill in the tables at the end of the problem.
Pipe Label
(m) (mm) C-factor
P-1 152.4 254 120
P-2 365.8 152 120
P-3 1,280.2 254 120
P-4 182.9 152 110
P-5 76.2 102 110
P-6 152.5 102 100
P-7 1,585.0 203 120
P-8 1,371.6 102 100
P-9 1,676.4 76 90
P-10 914.4 152 75
P-11 173.7 152 120
P-12 167.6 102 80
Elevation Demand
Node Label
(m) (l/s)
R-1 320.0 N/A
J-1 262.1 2.5
J-2 263.7 0.9
J-3 265.2 1.9
J-4 266.7 1.6
J-5 268.2 0.3
J-6 269.7 0.8
J-7 268.2 4.7
J-8 259.1 1.6
J-9 262.1 0
J-10 262.1 1.1
J-11 259.1 0.9
J-12 257.6 0.6
Flow Head loss

Pipe Label
(l/s) (m)
P-1
P-2
P-3
P-4
P-5
P-6
P-7
P-8
P-9
P-10
P-11
P-12
HGL Pressure
Node Label
(m) (kPa)
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
J-11
J-12
3.4 Determine the effect of placing demands at points along a pipe rather than at the end node (point D)
for the 300-m long pipe segment A-D shown in the figure. The pipe has a diameter of 150 mm and a
roughness height of 0.0001 m, and the kinematic viscosity of water at the temperature of interest is
1x10-6 m2/s. The total head at Point A is 200 m, and the ground elevation along the pipe is 120 m.
The flow past point A is 9 l/s. Points A, B, C, and D are equidistant from each other.
Upstream Intermediate Intermediate End

Point A Point B Point C Point D
a) Assume that there is no water use along the pipe (that is, flow is 9 l/s in all segments). Determine
the head loss in each segment and the pressure head (in meters) at points B, C, and D.
b) Assume that a small amount of water is used at points B and C (typical of a pipe in a residential
neighborhood), such that the flow in the second and third segments decreases to 8 and 7 l/s,
respectively. Determine the pressures at points B, C, and D.
c) Assume that the water is withdrawn evenly along the pipe, such that the flows in the second and
third segments are 6 and 3 l/s, respectively. Find the pressures at points B, C, and D.
d) At these flows, do the pressures in the pipe vary significantly when the water use is lumped at the
endpoint versus being accounted for along the length of the pipe? Would you expect a similar
outcome at much higher flows?
Pressure in meters of water
Point Part (a) Part (b) Part (c)

B
C
D
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
4
Water Consumption
The consumption or use of water, also known as water demand, is the driving force
behind the hydraulic dynamics occurring in water distribution systems. Anywhere
that water can leave the system represents a point of consumption, including a cus-
tomer’s faucet, a leaky main, or an open fire hydrant.
Three questions related to water consumption must be answered when building a
hydraulic model: (1) How much water is being used? (2) Where are the points of con-
sumption located? and (3) How does the usage change as a function of time? This
chapter addresses these questions for each of the three basic demand types described
below.
• Customer demand is the water required to meet the non-emergency needs of
users in the system. This demand type typically represents the metered por-
tion of the total water consumption.
• Unaccounted-for water (UFW) is the portion of total consumption that is
“lost” due to system leakage, theft, unmetered services, or other causes.
• Fire flow demand is a computed system capacity requirement for ensuring
adequate protection is provided during fire emergencies.
Determining demands is not a straightforward process like collecting data on the
physical characteristics of a system. Some data, such as billing and production
records, can be collected directly from the utility but are usually not in a form that can
be directly entered into the model. For example, metering data are not grouped by
node. Once this information has been collected, establishing consumption rates is a
process requiring study of past and present usage trends and, in some cases, the pro-
jection of future ones.
After consumption rates are determined, the water use is spatially distributed as
demands, or loads, assigned to model nodes. This process is referred to as loading the
model. Loading is usually a multistep process that may vary depending on the prob-
lem being considered. The following steps outline a typical example of the process
the modeler might follow.
134 Water Consumption Chapter 4
1. Allocate average-day demands to nodes.

2. Develop peaking factors for steady-state runs (page 153) or diurnal curves for
EPS runs (page 155).
3. Estimate fire and other special demands.
4. Project demands under future conditions for planning and design.
This chapter presents some of the methods to follow when undertaking the process of
loading a water distribution system model.
4.1 BASELINE DEMANDS

Most modelers start by determining baseline demands to which a variety of peaking
factors and demand multipliers can be applied, or to which new land developments
and customers can be added. Baseline demands typically include both customer
demands and unaccounted-for water. Usually, the average day demand in the current
year is the baseline from which other demand distributions are built.
Data Sources
Pre-Existing Compiled Data. The first step in finding demand information
for a specific utility should always be to research the utility’s existing data. Previous
studies, and possibly even existing models, may have a wealth of background infor-
mation that can save many hours of investigation.
However, many utilities do not have existing studies or models, or may have only lim-
ited resources to collect this type of information. Likewise, models that do exist may
be outdated and may not reflect recent expansion and growth.
System Operational Records. Various types of operational records are avail-
able that can offer insight into the demand characteristics of a given system. Treat-
ment facility logs may provide data regarding long-term usage trends such as seasonal
pattern changes or general growth indications. Pumping logs and tank level charts
(such as the one shown in Figure 4.1) contain data on daily system usage, as well as
the changing pattern of demand and storage levels over time.
Water distribution systems may measure and record water usage in a variety of forms,
including
• Flow information, such as the rate of production of a treatment or well
facility
• Volumetric information, such as the quantity of water consumed by a
customer
• Hydraulic grade information, such as the water level within a tank
The data described above are frequently collected in differing formats and require
conversion before they can be used. For example, tank physical characteristics can be
used to convert tank level data to volumes. If data describing the temporal changes in
tank levels are incorporated, volumes can be directly related to flow rates.
Section 4.1 Baseline Demands 135
Figure 4.1
Tank level chart
Courtesy of the City of Waterbury, CT Bureau of Water
Customer Meters and Billing Records. If meters are employed throughout

a system, they can be the best source of data for determining customer demands. Cus-
tomers are typically billed based on a volumetric measure of usage, with meter read-
ings taken on a monthly or quarterly cycle. Using these periodically recorded usage
volumes, customers’ average usage rates can be computed. Billing records, therefore,
provide enough information to determine a customer’s baseline demand, but not
enough to determine fluctuations in demand on a finer time scale such as that required
for extended-period simulations.
Ideally, the process of loading demand data into a model from another source would
be relatively automatic. Cesario and Lee (1980) describe an early approach to auto-
mate model loading. Coote and Johnson (1995) developed a system in Valparaiso,
Indiana in which each customer account was tied to a node in their hydraulic model.
With the increasing popularity of geographic information systems (GIS) among water
utilities, more modelers are turning to GIS to store and manipulate demand data to be
imported into the model. Stern (1995) described how Cybernet data were loaded from
a GIS in Los Angeles, and Basford and Sevier (1995) and Buyens, Bizier, and
Combee (1996) describe similar applications in Newport News, Virginia, and Lake-
land, Florida, respectively. As GIS usage becomes more widespread, more utilities
will construct automated links between customer data, GIS, and hydraulic models.
Spatial Allocation of Demands

Although water utilities make a large number of flow measurements, such as those at
customer meters for billing and at treatment plants and wells for production monitor-
ing, data are usually not compiled on the node-by-node basis needed for modeling.
The modeler is thus faced with the task of spatially aggregating data in a useful way
and assigning the appropriate usage to model nodes.
The most common method of allocating baseline demands is a simple unit loading
method. This method involves counting the number of customers [or acres (hectares)
of a given land use, number of fixture units, or number of equivalent dwelling units]
that contribute to the demand at a certain node, and then multiplying that number by
the unit demand [for instance, number of gallons (liters) per capita per day] for the
applicable load classification. For example, if a junction node represents a population
of 200, and the average usage is 100 gal/day/person (380 l/day/person), the total base-
line demand for the node would be 20,000 gal/day (75,710 l/day).
In applying unit demands, the user must be careful to understand what is accounted
for by that measure. Equations 4.1, 4.2, and 4.3 show three different unit demands that
can be determined for a utility (Male and Walski, 1990).
Demands in the United Kingdom

Not all water systems are universally metered as Demand patterns in the UK are similar to most
is customary in North America. For example, in other developed nations, and the patterns can be
the United Kingdom, only roughly 10 percent of established by DMA or groups of DMAs. Data log-
the domestic customers are metered. ging is used to determine individual demand pat-
Instead of metering individual customers, distribu- terns only for the largest users.
tion systems in the UK are divided into smaller Because most residences are not metered, unac-
zones, called District Metered Areas (DMAs), counted-for water in the UK is large, but most of
which are isolated by valving and are fed through this water is delivered to legitimate users and can
a smaller number of inlet and outlet meters (WRc, be estimated fairly reasonably. The amount of
1985). The number of properties in a DMA is actual leakage depends on pressure, burst fre-
known fairly precisely, and usually varies from 500 quency, leakage control policy, and age of pipes.
to 5,000 properties but can go as high as 10,000. Despite the differences in metering practices
The flows are recorded using data-logging tech- between the UK and North America, loading of
nology or telemetered to a central location. the model still involves many of the same steps,
Per capita consumption at the unmetered resi- and the system metering data collected in the UK
dences is estimated to be on the order of 150 can make calibration easier than in locations with-
liters per capita per day, although there is consid- out pervasive distribution metering.
erable variation (Ofwat, 1998). Some of the
variation is attributed to different socioeconomic
classes as accounted for by ACORN (A Classifi-
cation of Residential Neighborhoods), which
classifies properties in England and Wales into
categories such as “modern family housing with
higher income” to “poorest council estates.”
system-wide use = (production) / (domestic customers) (4.1)
non-industrial use = (production – industrial use) / (domestic customers) (4.2)
domestic use = (domestic metered consumption) / (domestic customers) (4.3)
All three unit demands can be determined on a per capita or per account basis.
Although all three can be referred to as unit demands, each yields a different result,
and it is important that the modeler understand which unit demand is being used. The
first unit demand includes all nonemergency uses and is the largest numerical value;
the second excludes industrial uses; the third excludes industrial use and
unaccounted-for water and is the smallest. If the third unit demand is used, then
unaccounted-for water and industrial use must be handled separately from the unit
demands. This approach may be advantageous where industrial use is concentrated in
one portion of the network.
Another approach to determining the baseline demand for individual customers
involves the use of billing records. However, rarely does a system have enough
recorded information to directly define all aspects of customer usage. Even in cases
where both production records and full billing records are available, disagreements
between the two may exist that need to be resolved.
Two basic approaches exist for filling in the data gaps between water production and
computed customer usage: top-down and bottom-up. Both of these methods are based
on general mass-balance concepts and are shown schematically in Figure 4.2.
Figure 4.2
Approaches to model
Start: Production Data Production
loading
Large Users Nodes / Service Area
Goal: Nodal Demands Nodes Nodes Nodes Nodes Nodes Nodes
Meter
Routes
Start: Meter Records Billing Records Unaccounted-For
Top-down demand determination involves starting from the water sources (at the
“top”) and working down to the nodal demands. With knowledge about the produc-
tion of water and any large individual water customers, the remainder of the demand
is disaggregated among the rest of the customers. Bottom-up demand determination is
exactly the opposite, starting with individual customer billing records and summing
their influences using meter routes as an intermediate level of aggregation to deter-
mine the nodal demands.
Most methods for loading models are some variation or combination of the top-down
and bottom-up approaches, and tend to be system-specific depending on the availabil-
ity of data, the resources for data-entry, and the need for accuracy in demands. For
some systems, the decision to use top-down or bottom-up methods can be made on a
zone-by-zone basis.
Cesario (1995) uses the terms estimated consumption method and actual consumption
method to describe these two approaches. However, both methods involve a certain
level of estimation. An intermediate level of detail can be achieved by applying the
top-down approach with usage data on a meter-route-by-meter-route basis (AWWA,
1989).
Most design decisions, especially for smaller pipes, are controlled by fire flows, so
modest errors in loading have little impact on pipe sizing. The case in which loading
becomes critical is in the tracking of water quality constituents through a system,
because fire flows are not typically considered in such cases.
Example — Top-Down Demand Determination. Consider a system that serves a

community of 1,000 people and a single factory, which is metered. Over the course of a year, the total
production of potable water is 30,000,000 gallons (114,000 m3). The factory meter registered a usage
of 10,000,000 gallons (38,000 m3). Determining the average per capita residential usage in this case is
straightforward:
Total volume of residential usage = (Total usage) – (Non-residential usage)

= 30,000,000 gallons – 10,000,000 gallons
= 20,000,000 gallons
Residential volume usage per capita = (Total volume of residential usage) / (no. of residents)
= 20,000,000 gallons / 1,000 capita
= 20,000 gallons/capita
Residential usage rate per capita (given that prior volume calculations were for a period of one year)
= (Residential volume usage per capita) / time
= (20,000 gallons/capita/year) u (1 year / 365 days)
= 55 gallons/capita/day = 210 liters/capita/day
Models usually require demands in gallons per minute or liters per second, which gives
0.038 gpm/capita = 0.0024 l/s/capita
Next, the approximate number of people (or houses) per node (e.g., 25 houses with 2.5 residents per
house = 62.5 residents/node) is determined to give average nodal demand of
2.37 gpm/node =0.15 l/s/node
These average residential nodal demands can be adjusted for different parts of town based on popula-
tion density, amount of lawn irrigation, and other factors.
Example — Bottom-Up Demand Determination. Each customer account is assigned

an x-y coordinate in a GIS. Then, each account can be assigned to a node in the model based on poly-
gons around each node in the GIS. (If a GIS is not available, customer accounts can be directly
assigned to a node in the customer service information system used for billing purposes.) Then, each
account in the customer information database records can be assigned to a model node. By querying
the customer information database, the average demand at each node for any billing period can be
determined.
The billing data must now be corrected for unaccounted-for water. Consider a user who decides to
allocate unaccounted-for water uniformly to each node. The daily production is 82,000 gpd, and
metered sales are 65,000 gpd. For each node, the demand must be corrected for unaccounted-for
water. One approach is to assign unaccounted-for water in proportion to the demand at a node using:
Corrected demand = (Node consumption) u [(Production) / (Metered Sales)]
For a node with a consumption of 4.2 gpm, the corrected demand is:
Corrected demand = (4.2 gpm) u (82,000/65,000) = 5.3 gpm = 0.33 l/s
As can be seen in the preceding examples, bottom-up demand allocation requires a

great deal of initial effort to set up links between accounts and nodes, but after this
work is done, the loads can be recalculated easily. Of course, the corrections due to
unaccounted-for water and the fact that instantaneous demands are most likely not
equal to average demands suggest that both approaches are subject to error.
Example — Demand Allocation. In a detailed demand allocation, a key step is determin-

ing the customers assigned to each node. Figure 4.3 demonstrates the allocation of customer demands
to modeled junction nodes. The dashed lines represent the boundaries between junction associations.
For example, the junction labeled J-1 should have demands that represent nine homes and two com-
mercial establishments. Likewise, J-4 represents the school, six homes, and one commercial building.
Figure 4.3
Allocating demands Homes
to network junctions
J-1 J-2
Commercial
Establishments
School
J-5 J-4 J-3
Node Service
Area Boundary (typ.)
Following demand allocation, the modeler must ensure that demands have been assigned to junction
nodes in such a way that (1) the sums of the nodal demands system-wide and in each pressure zone
are in agreement with production records, and (2) the spatial allocation of demands closely approxi-
mates actual demands.
When working with high-quality GIS data, the modeler can much more precisely
assign demands to nodes. Nodal demands can be loaded using several GIS-related
methodologies, ranging from a simple inverse-pipe-diameter allocation model to a
comprehensive polygon overlay. The inverse-pipe-diameter approach assumes that
demands are associated with small-diameter pipes, whereas large-diameter pipes are
mainly used for transmission and thus should have less “weight” associated with
them. More detailed methods make use of extensive statistical data analysis and GIS
processing by combining layers of data that account for variables such as population
changes over time, land use, seasonal changes, planning, and future development
rates. Davis and Brawn (2000) describe an approach they employed to allocate
demands using a GIS.
Using GIS for Demand Allocation

As discussed previously, an integral part of creating a water distribution model is the
accurate allocation of demands to the node elements within the model. The spatial
analysis capabilities of GIS make it a logical tool for the automation of the demand
allocation process.
The following sections provide descriptions of some of the automated allocation strat-
egies that have been used successfully.
Meter Assignment. This allocation strategy uses the spatial analysis capabili-
ties of GIS to assign geocoded (possessing coordinate data based on physical location,
such as an x-y coordinate) customer meters to the nearest demand node. Therefore,
this type of model loading is a point-to-point demand allocation technique, meaning
that known point demands (customer meters) are assigned to network demand points
(demand nodes). Meter assignment is the simplest technique in terms of required data,
because there is no need for service polygons to be applied (see Figure 4.4).
However, meter assignment can prove less accurate than the more complex allocation
strategies because “nearest” is determined by straight-line proximity between the
demand node and the consumption meter. Piping routes are not considered, so the
nearest demand node may not be the location from which the meter actually receives
its flow. In addition, the actual location of the service meter may not be known. Ide-
ally, these meter points should be placed at the location of the tap, but the centroid of
the building or land parcel may be all that is known about a customer account.
Figure 4.4
Meter assignment
Meter Aggregation. Meter aggregation is the technique of assigning all meters

within a service polygon to a specified demand node (see Figure 4.5). Service poly-
gons define the service area for each of the demand junctions.
Meter aggregation is a polygon-to-point allocation technique because the service
areas are contained in a GIS polygon layer and the demand junctions are contained in
a point layer. The demands associated with each of the service-area polygons are
assigned to the respective demand node points.
Because of the need for service polygons, the initial setup for this approach is more
involved than for the simpler meter assignment strategy, with the tradeoff being
greater control over the assignment of meters to demand nodes. Automated construc-
tion of the service polygons may not produce the desired results, so it may be neces-
sary to manually adjust the polygon boundaries, especially at the edges of the
drawing.
Figure 4.5
Meter aggregation
Flow Distribution. This strategy involves distributing a lump-sum demand

among a number of service polygons (service areas) and, by extension, their associ-
ated demand nodes. The lump-sum area is a polygon for which the total (lump-sum)
demand of all of the service areas (and their demand nodes) is known (metered), but
the distribution of the total demand among the individual nodes is not. Lump-sum
areas can be based on pressure zones, meter routes, or other criteria.
The known demand within the lump-sum area is divided among the service polygons
within the area using one of two techniques: equal distribution or proportional distri-
bution. The equal distribution option simply divides the known demand evenly
between the demand nodes. For example, in Figure 4.6, the total demand in meter
route A may be 55 gpm (3.48 l/s), and the total demand in meter route B may be 72
gpm (4.55 l/s). Since there are 11 nodes in meter route A and 8 nodes in meter route
B, the demand at each node will be 5 gpm (0.32 l/s) and 9 gpm (0.57 l/s), respectively.
Figure 4.6
Equal flow
distribution
The proportional distribution option divides the lump-sum demand among the service
polygons based on one of two attributes of the service polygons: the area or the popu-
lation. That is, the greater the percentage of the lump-sum area or population that a
service polygon contains, the greater the percentage of total demand that will be
assigned to that service polygon.
Each service polygon has an associated demand node, and the demand that is calcu-
lated for each service polygon is assigned to this demand node. For example, if a ser-
vice polygon makes up 50 percent of the lump-sum polygon’s area, then 50 percent of
the demands associated with the lump-sum polygon will be assigned to the demand
node associated with that service polygon.
Flow distribution strategies require the definition of lump-sum area or population

polygons, service polygons, and their related demand nodes.
Sometimes, a combination of demand allocation methods is recommended. One case
where this technique is particularly helpful is in accounting for unaccounted-for
water. A meter assignment or meter aggregation method can be used to distribute the
normal demands, and a flow distribution technique can be used in addition to assign
the unaccounted-for water.
Point Demand Assignment. A point demand assignment technique is used to
directly assign a demand to a demand node. This strategy is primarily a manual oper-
ation, and is used to assign large (generally industrial or commercial) water users to
the demand node that serves the consumer in question. This technique is unnecessary
if all demands are accounted for by using one of the other allocation strategies.
Projection of Future Demands. Automated techniques have also been devel-
oped to assist in the assignment of future demands to nodes. These are similar to flow
distribution allocation except that the type of base layer that is used to intersect with
the service layer may contain information other than average-day demands.
Demand projection relies on a polygon layer that contains data regarding expected
future conditions. Some data types that can be used for this include future land use
and projected population, in combination with a demand density (for example, gal-
lons per capita per day), with the polygons based on traffic analysis zones, census
tracts, planning districts, or another classification. Many of these data types do not
include demand information, so demand density is required to translate the informa-
tion contained in the future-condition polygons into projected demand values.
Methods of using water-use data based on population or land use involve overlaying
those polygons on node service-area polygons and are described in more detail in
Chapter 12.
Categorizing Demands
Sometimes water users at a single node fall into several categories, and the modeler
would like to keep track of these categories within the model. Composite demands
enable the modeler to do this type of tracking. The modeler can selectively search for
all demands of a certain type (for example, residential or industrial) and make adjust-
ments. The modeler can also make changes to the characteristics of an entire category,
and all of the customers of that type will automatically be modified.
Composite Demands. Whether a unit-loading-based or a billing-record-based
method is used to generate the baseline demand, the user may need to convert it into a
composite demand at a particular node. This conversion is necessary since a junction
node does not always supply a single customer type. When more than one demand
type is served by a particular junction, the demand is said to be a composite. Deter-
mining the total rate of consumption for a junction node with a composite demand is a
simple matter of summing the individual components. Composite demands are also a
way of keeping track of unaccounted-for water independent of the other demands at a
node.
When temporal patterns are applied to composite demands, the total demand for a
junction at any given time is equal to the sum of each baseline demand times its
respective pattern multiplier. It is also possible with most software packages to assign
a different pattern to the different components of the composite demand.
Q i t = ¦ Bi j Pi j t (4.4)

j
where Qi, t = total demand at junction i at time t (cfs, m3/s)

Bi, j = baseline demand for demand type j at junction i (cfs, m3/s)
Pi, j, t = pattern multiplier for demand type j at junction i at time t
Nomenclature. Depending on the scale of the model, the demand type may con-
sist of such broad categories as “residential,” “commercial,” and “industrial,” or be
broken down into a finer level of detail with categories such as “school,” “restaurant,”
“multifamily dwelling,” and so on.
An issue that arises when discussing demands is that each utility classifies customers
differently. For example, an apartment may be a “residential” account at one utility, a
“commercial” account at another, and a “multifamily residential” account at yet
another. Schools may be classified as “institutional,” “commercial,” “public,” or sim-
ply “schools.” A modeler working for a utility can easily adapt to the naming conven-
tions, but a consultant who works with many utilities may have a difficult time
keeping track of the nomenclature when moving from one system to another.
Mass Balance Technique

Regardless of whether a modeler is studying the entire system, one particular pressure
zone, or an individual customer, mass balance techniques are useful for determining
changes in demand occurring on a finer time scale than a monthly billing cycle. For a
water distribution system, a mass balance simply indicates that what goes into the sys-
tem must be equal to what comes out of the system or zone (accounting for changes in
storage). In equation form, this can be stated as follows:
Q demand = Q inflow – Q outflow + 'V storage e 't (4.5)
where Qinflow = average rate of production (cfs, m3/s)

Qdemand = average rate of demand (cfs, m3/s)
Qoutflow = average outflow rate (cfs, m3/s)
'V storage = change in storage within the system (ft3, m3)
't = time between volume measurements (s)
Note that the rates of production and demand in the above equation are representative
of the average flow rates over the time period. The change in storage, however, is
found by taking the difference between storage volumes at the beginning and end of
the time period for each tank, as follows:
'V storage = ¦ Vi t + 't – Vi t (4.6)

i
where V i t + 't = storage volume of tank i at time t+ ' t (ft3, m3)

Vi, t = storage volume of tank i at time t (ft3, m3)
When calculating volume changes in storage, a sign convention must apply. If the vol-
ume in storage decreased during the time interval, then that volume is added to the
inflows, and if it increased over the time period, then it is subtracted from inflows.
For upright cylindrical tanks (or any tank with vertical sides), the change in storage
can be determined directly from the change in tank level, as follows:
'V storage = ¦ Hi t + 't – Hi t Ai t (4.7)

i
where H i t + 't = water level at beginning of times step t+ ' t at tank i (ft, m)
Hi,t = water level at beginning of times step t at tank i (ft, m)
Ai,t = surface area of tank i during time step t (ft2, m2)
Example — Mass Balance. Consider a pressure zone with a single cylindrical tank having
a diameter of 40 feet. At the beginning of a daily monitoring interval, the water level is at 28.3 feet,
and at the beginning of the next day it is 29.1 feet. During that time, the total flow into that zone is
determined to be 455 gallons per minute, and there is no outflow to other zones. What is the total
average daily demand within the zone?
Knowing the tank’s diameter, its area is found to be
2 2
S 40 - = 1256 2
SD - = ----------------
A = --------- ft
4 4
The change in storage is then found as
'V = A(Hi+1 - Hi) = 1256 ft2 (29.1 ft – 28.3 ft) u 7.48 gal/ft3 = 7,516 gal
Storage in the tank increased over the monitoring period, thus the sign convention dictates that the
flows to storage be subtracted from the total inflow. With the change in storage and the average
inflow, the average zone demand occurring over the hourly monitoring period is
7516 - = 449.8 gpm

Q = 455 – -----------------
60 u 24
This answer makes sense because the average zone demand must be smaller than the average inflow
for the tank to fill during the monitoring period.
Using Unit Demands

In the case of new water customers, flows can usually be estimated based on similar
customers in the community. Numerous investigators have compiled typical water
consumption for different types of facilities. To use this data, the modeler needs to
determine the number of units (for example, number of rooms in a hotel or number of
seats in a restaurant) and multiply by the typical unit flow to determine the average
daily flow from that establishment.
Table 4.1 provides typical unit loads for a number of different types of users. Ranges
are given because there is considerable variation between establishments within a
given category.
Table 4.1 Typical rates of water use for various establishments

Range of Flow
User (l/person or unit/day) (gal/person or unit/day)
Airport, per passenger 10–20 3–5
Assembly hall, per seat 6–10 2–3
Bowling alley, per alley 60–100 16–26
Camp
Pioneer type 80–120 21–32
Children’s, central toilet and bath 160–200 42–53
Day, no meals 40–70 11–18
Luxury, private bath 300–400 79–106
Labor 140–200 37–53
Trailer with private toilet and bath, 500–600 132–159
per unit (2 1/2 persons)
Country clubs
Resident type 300–600 79–159
Transient type serving meals 60–100 16–26
Dwelling unit, residential
Apartment house on individual well 300–400 79–106
Apartment house on public water supply, 300–500 79–132
unmetered
Boardinghouse 150–220 40–58
Hotel 200–400 53–106
Lodging house and tourist home 120–200 32–53
Motel 400–600 106–159
Private dwelling on individual well or 200–600 53–159
metered supply
Private dwelling on public water 400–800 106–211
supply, unmetered
Factory, sanitary wastes, per shift 40–100 11–26
Table extracted from Ysuni, 2000 based on Metcalf and Eddy, 1979
Table 4.1 (cont.) Typical rates of water use for various establishments
Range of Flow
User (l/person or unit/day) (gal/person or unit/day)
Fairground (based on daily attendance) 2–6 1–2
Institution
Average type 400–600 106–159
Hospital 700–1200 185–317
Office 40–60 11–16
Picnic park, with flush toilets 20–40 5–11
Restaurant (including toilet)
Average 25–40 7–11
Kitchen wastes only 10–20 3–5
Short order 10–20 3–5
Short order, paper service 4–8 1–2
Bar and cocktail lounge 8–12 2–3
Average type, per seat 120–180 32–48
Average type, 24 h, per seat 160–220 42–58
Tavern, per seat 60–100 16–26
Service area, per counter seat (toll road) 1000–1600 264–423
Service area, per table seat (toll road) 600–800 159–211
School
Day, with cafeteria or lunchroom 40–60 11–16
Day, with cafeteria and showers 60–80 16–21
Boarding 200–400 53–106
Self-service laundry, per machine 1000–3000 264–793
Store
First 7.5 m (25 ft) of frontage 1600–2000 423–528
Each additional 7.5 m of frontage 1400–1600 370–423
Swimming pool and beach, toilet and shower 40–60 11–16
Theater
Indoor, per seat, two showings per day 10–20 3–5
Outdoor, including food stand, per car 10–20 3–5
(3 1/3 persons)
Table extracted from Ysuni, 2000 based on Metcalf and Eddy, 1979
Other investigators have linked water use in nonresidential facilities to the Standard
Industrial Classification (SIC) codes for each industry as shown in Table 4.2. To use
this table, the modeler determines the number of employees in an industry and multi-
plies the number by the use rate given in the table. As is the case with Table 4.1, there
will be considerable variation about the typical values given Table 4.2.
Table 4.2 Average rates of nonresidential water use from establishment-level data
Use Rate Sample
Category SIC Code
(gal/employee/day) Size
Construction 31 246
General building contractors 15 118 66
Heavy construction 16 20 30
Special trade contractors 17 25 150
Manufacturing 164 2790
Food and kindred products 20 469 252
Textile mill products 22 784 20
Apparel and other textile products 23 26 91
Lumber and wood products 24 49 62
Furniture and fixtures 25 36 83
Paper and allied products 26 2614 93
Printing and publishing 27 37 174
Chemicals and allied products 28 267 211
Petroleum and coal products 29 1045 23
Rubber and miscellaneous plastics products 30 119 116
Leather and leather products 31 148 10
Stone, clay, and glass products 32 202 83
Primary metal industries 33 178 80
Fabricated metal products 34 194 395
Industrial machinery and equipment 35 68 304
Electronic and other electrical equipment 36 95 409
Transportation equipment 37 84 182
Instruments and related products 38 66 147
Miscellaneous manufacturing industries 39 36 55
Transportation and public utilities 50 226
Railroad transportation 40 68 3
Local and interurban passenger transit 41 26 32
Trucking and warehousing 42 85 100
U.S. Postal Service 43 5 1
Water transportation 44 353 10
Transportation by air 45 171 17
Transportation services 47 40 13
Communications 48 55 31
Electric, gas, and sanitary services 49 51 19
Wholesale trade 53 751
Wholesale trade–durable goods 50 46 518
Wholesale trade–nondurable goods 51 87 233
Table from Dziegielweski, Opitz, and Maidment, 1996
Table 4.2 (cont.) Average rates of nonresidential water use from establishment-level data
Use Rate Sample
Category SIC Code
Retail trade 93 1044
Building materials and garden supplies 52 35 56
General merchandise stores 53 45 50
Food stores 54 100 90
Automotive dealers and service stations 55 49 498
Apparel and accessory stores 56 68 48
Furniture and home furnishings stores 57 42 100
Eating and drinking places 58 156 341
Miscellaneous retail 59 132 161
Finance, insurance, and real estate 192 238
Depository institutions 60 62 77
Nondepository institutions 61 361 36
Security and commodity brokers 62 1240 2
Insurance carriers 63 136 9
Insurance agents, brokers, and service 64 89 24
Real estate 65 609 84
Holding and other investment offices 67 290 5
Services 137 1878
Hotels and other lodging places 70 230 197
Personal services 72 462 300
Business services 73 73 243
Auto repair, services, and parking 75 217 108
Miscellaneous repair services 76 69 42
Motion pictures 78 110 40
Amusement and recreation services 79 429 105
Health services 80 91 353
Legal services 81 821 15
Educational services 82 110 300
Social service 83 106 55
Museums, botanical, zoological gardens 84 208 9
Membership organizations 86 212 45
Engineering and management services 87 58 5
Services, NEC 89 73 60
Public administration 106 25
Executive, legislative, and general 91 155 2
Justice, public order, and safety 92 18 4
Administration of human resources 94 87 6
Table 4.2 (cont.) Average rates of nonresidential water use from establishment-level data
Use Rate Sample
Category SIC Code
Environmental quality and housing 95 101 6
Administration of economic programs 96 274 5
National security and international affairs 97 445 2
Unaccounted-For Water
Ideally, if individual meter readings are taken for every customer, they should exactly
equal the amount of water that is measured leaving the treatment facility. In practice,
however, this is not the case. Although inflow does indeed equal outflow, not all of the
outflows are metered. These “lost” flows are referred to as unaccounted-for water
(UFW).
There are many possible reasons why the sum of all metered customer usage may be
less than the total amount of water produced by the utility. The most common reasons
for discrepancies are leakage, errors in measurement, and unmetered usage. Ideally,
customer demands and unaccounted-for water should be estimated separately. In this
way, a utility can analyze the benefits of reducing unaccounted-for water.
Unaccounted-for water must be loaded into the model just like any other demand.
However, the fact that it is unaccounted-for means that the user does not know where
to place it. Usually, the user simply calculates total unaccounted-for water and divides
that quantity equally among all nodes. If the modeler knows that one portion of a sys-
tem has a greater likelihood of leakage because of age, then more unaccounted-for
water can be placed within that section.
Leakage. Leakage is frequently the largest component of UFW and includes dis-
tribution losses from supply pipes, distribution and trunk mains, services up to the
meter, and tanks. The amount of leakage varies from system to system, but there is a
general correlation between the age of a system and the amount of UFW. Newer sys-
tems may have as little as 5 percent leakage, while older systems may have 40 percent
leakage or higher. Leakage tends to increase over time unless a leak detection and
repair program is in place. Use of acoustic detection equipment to listen for leaks is
Other factors affecting leakage include system pressure (the higher the pressure, the
more leakage), burst frequencies of mains and service pipes, and leakage detection
and control policies. These factors make leakage very difficult to estimate, even with-
out the complexity of approximating other UFW causes. If better information is not
available, UFW is usually assigned uniformly around the system.
Figure 4.7
Use of leak detection
equipment
Meter Under Registration. Flow measurement errors also contribute to UFW.

Flow measurements are not always exact, and thus metered customer usage may con-
tain inaccuracies. Some flow meters under-register usage at low flow rates, especially
as they get older.
Unmetered Usage. Systems may have illegal connections or other types of
unmetered usage. Not all unmetered usage is indicative of water theft. Fire hydrants,
blow-offs, and other maintenance appurtenances are typically not metered.
4.2 DEMAND MULTIPLIERS

By definition, baseline demands during a steady-state simulation do not change over
time. However, in reality, water demand varies continuously over time according to
several time scales:
• Daily. Water use varies with activities over the course of a day.
• Weekly. Weekend patterns are different from weekdays.
• Seasonal. Depending on the extent of outdoor water use or seasonal
changes, such as tourism, consumption can vary significantly from one sea-
son to another.
Section 4.2 Demand Multipliers 153
• Long-term. Demands can grow due to increases in population and the

industrial base, changes in unaccounted-for water, annexation of areas previ-
ously without service, and regionalization of neighboring water systems.
The modeler needs to be cognizant of the impacts of temporal changes on all of these
scales. These time-varying demands are handled in the model by either
• Steady-state runs for a particular condition, or
• Extended-period model runs
For extended-period simulations, the model requires both baseline demand data and
information on how demands vary over time. Modeling of these temporal variations is
described in the next section.
In steady-state runs, the user can build on the baseline demand by using multipliers
and/or assigning different demands to specific nodes. Fortunately, the entire demand
allocation need not be redone.
The following are some examples of demand events frequently considered:
• Average-day demand: The average rate of demand for an average day (past,
present, or future)
• Maximum-day demand: The average rate of use on the maximum usage
day (past, present, or future)
• Peak-hour demand: The average rate of usage during the maximum hour of
usage (past, present, or future)
• Maximum day of record: The highest average rate of demand for the his-
torical record
Peaking Factors
For some consumption conditions (especially predicted consumption conditions),
demands can be determined by applying a multiplication factor or a peaking factor.
For example, a modeler might determine that future maximum day demands will be
double the average-day demands for a particular system. The peaking factor is calcu-
lated as the ratio of discharges for the various conditions. For example, the peaking
factor applied to average-day demands to obtain maximum day demands can be found
by using Equation 4.8.
PF = Q max e Q avg (4.8)
where PF = peaking factor between maximum day and average-day demands

Qmax = maximum day demands (cfs, m3/s)
Qavg = average-day demands (cfs, m3/s)
Determining system-wide peaking factors is fairly easy because most utilities keep
good records on production and tank levels. However, peaking factors for different
types of demands applied at individual nodes are more difficult to determine, because
individual nodes do not necessarily follow the same demand pattern as the system as a
whole.
Peaking factors from average day to maximum day tend to range from 1.2 to 3.0, and
factors from average day to peak hour are typically between 3.0 and 6.0. Of course,
these values are system-specific, so they must be determined based on the demand
characteristics of the system at hand.
Fire flows represent a special type of peaking condition, and they are described on
page 165. Fire flows are usually added to maximum day flows when evaluating the
capacity of the system for fire fighting.
Demands in Systems with High Unaccounted-For Water. Using glo-
bal demand multipliers for projections in systems with high unaccounted-for water is
based on the assumption that the relative amount of unaccounted-for water will
remain constant in the future. Unaccounted-for water can also be treated as one of the
parts of a composite demand, as discussed on page 144. If unaccounted-for water is
reduced, then the utility will see higher peaking factors because unaccounted-for
water tends to flatten out the diurnal demand curve. Walski (1999) describes a method
for correcting demand multipliers for systems where leakage is expected to change
over time.
§M -----·
M © A ¹ c Qc + L
----- = ----------------------------- (4.9)
A Qc + L
where M/A = corrected multiplier

(M/A)c = multiplier for consumptive users only
Qc = water use through customer meters in future (cfs, m3/s)
L = leakage in future (cfs, m3/s)
Example — Peaking Factors. If the multiplier for metered customers (M/A)c is 2.1, and
the metered demand (Qc) is projected to be 2.4 MGD in a future condition, then the overall multiplier
can be determined based on estimated future leakage as shown in the following table.
Leakage
M/A
(MGD)
0.0 2.1
0.5 1.9
1.0 1.8
Because leakage contributes the same to average and peak demands, the peak demand multipliers
increase as leakage decreases. The numerical value of (M/A)c can be calculated using current year
data and Equation 4.10.
M
----- Q c + L – L
§M A
-----· = -----------------------------------
- (4.10)
© A¹ c Q c
Section 4.3 Time-Varying Demands 155
The L and Q values are based on current year actual values. For example, in this problem, say that the
current year overall multiplier is 1.8, the metered demand is 1.5 MGD, and the leakage is 0.6 MGD.
The multiplier for metered consumption is then
§M 1.8 1.5 + 0.6 – 0.6- = 2.1

-----· = ----------------------------------------------
© A¹ c 1.5
Commercial Building Demands. A means of estimating design demands for

proposed commercial buildings is called the Fixture Unit Method. If the nature of the
customer/building is known, and the number and types of water fixtures (toilets, dish-
washers, drinking fountains, and so on) can be calculated, then the peak design flow
can be determined. The fixture unit method accounts for the fact that it is very
unlikely that all of the fixtures in a building will be operated simultaneously. Chapter
9 contains more information on using this method (see page 399).
4.3 TIME-VARYING DEMANDS

Water usage in municipal water distribution systems is inherently unsteady due to
continuously varying demands. In order for an extended period simulation to accu-
rately reflect the dynamics of the real system, these demand fluctuations must be
incorporated into the model.
The temporal variations in water usage for municipal water systems typically follow a
24-hour cycle called a diurnal demand pattern. However, system flows experience
changes not only on a daily basis, but also weekly and annually. As one might expect,
weekend usage patterns often differ from weekday patterns. Seasonal differences in
water usage have been related to climatic variables such as temperature and precipita-
tion, and also to the changing habits of customers, such as outdoor recreational and
agricultural activities occurring in the summer months.
Diurnal Curves
Each city has its own unique level of usage that is a function of recent climatic condi-
tions and the time of day. (Economic growth also influences demands, but its effect
occurs over periods longer than the typical modeling time horizon, and it is accounted
for using future demand projections.) Figure 4.8 illustrates a typical diurnal curve for
a residential area. There is relatively low usage at night when most people sleep,
increased usage during the early morning hours as people wake up and prepare for the
day, decreased usage during the middle of the day, and finally, increased usage again
in the early evening as people return home.
For other water utilities and other types of demands, the usage pattern may be very
different. For example, in some areas, residential irrigation occurs overnight to mini-
mize evaporation, which may cause peak usage to occur during the predawn hours.
For small towns that are highly influenced by a single industry, the diurnal pattern
may be much more pronounced because the majority of the population follows a sim-
ilar schedule. For example, if a large water-using industry runs 24 hours per day, the
overall demand pattern for the system may appear relatively flat because the steady
industrial usage is much larger than peaks in the residential patterns.
Figure 4.8
A typical diurnal
curve
Developing System-Wide Diurnal Curves

A system-wide diurnal curve can be constructed using the same mass balance tech-
niques discussed earlier in this chapter. The only elaboration is that the mass balance
is performed as a series of calculations, one for each hydraulic step of an EPS simula-
tion.
Time Increments. The amount of time between measurements has a direct corre-
lation to the resolution and precision of the constructed diurnal curve. If measure-
ments are only available once per day, then only a daily average can be calculated.
Likewise, if measurements are available in hourly increments, then hourly averages
can be used to define the pattern over the entire day.
If the modeler tries to use a time step that is too small, small errors in tank water level
can lead to large errors in water-use calculations. This type of error is explained fur-
ther in Walski, Lowry, and Rhee (2000). Modeling of hydraulic time steps smaller
than one hour is usually only justified in situations in which tank water levels change
rapidly. Even if facility operations (such as pump cycling) occur frequently, it may
still be acceptable for the demand pattern time interval to be longer than the hydraulic
time step.
The modeler should be aware that incremental measurements can still overlook a peak
event. For example, consider something as simple as determining the peak-hour
demand (the highest average demand over any continuous one-hour period). If mea-
surements are taken every hour on the hour, then the determination of the computed
peak hour will only be accurate if the actual peak begins and ends right on an even
hour increment (such as a peak occurring from 7:00 to 8:00 a.m.). The modeler will
underestimate peak hour usage if the true peak occurs, for example, from 7:15 to 8:15
a.m. The diurnal demand curve in Figure 4.9 illustrates this point. As can also be seen
in Figure 4.9, as time increments become smaller, peak flows become higher (for
instance, the 15-minute peak is higher than the one-hour peak).
Figure 4.9
Diurnal Curve Peak Missed Missed peak on a
diurnal curve due to
model time step
Usage
Hourly Measurements
12am 5am 6am 7am 8am 9am 10am 11am 11pm

Time
Developing Customer Diurnal Curves

Frequently, developing a diurnal curve for a specific customer requires more informa-
tion than can be extracted from typical billing records. In these situations, more inten-
sive data collection methods are needed to portray the time-variant nature of the
demands.
Data Logging for Customer Usage. Manually reading a customer’s water
meter at frequent intervals would obviously be a tedious and expensive undertaking.
The process of data logging refers to the automated gathering of raw data in the field.
These data are later compiled and analyzed for a variety of purposes, among them the
creation of diurnal demand curves. Various applications of data logging are described
in papers by Brainard (1994); Rhoades (1995); and DeOreo, Heaney, and Mayer
(1996).
There have been many recent advances in data-logging technology, making it a reli-
able and fairly inexpensive way to record customer water usage. Figure 4.10 illus-
trates a typical meter/data logger setup. Utilities can now easily place a data logger on
a customer’s meter and determine that customer’s consumption pattern. While it
would be nice to have such a detailed level of data on all customers, the cost of obtain-
ing the information is only justifiable for large customers and a sampling of smaller
ones.
Figure 4.10
A typical meter/data
logger setup
Meter-Master Model 100EL Flow Recorder manufactured by F.S. Brainard & Company
Representative Customers. Although it is possible to study a few customers

in detail and extend the conclusions of that study to the rest of the system, this type of
data extrapolation has some inherent dangers. The probability of selecting the “per-
fect” average customer is small, and any deviation from the norm or error in measure-
ment will be compounded when it is applied to an entire community. As with all
statistical data collection methods, the smaller the sample size, the less confidence
there can be in the results.
There are also applications in which use of a representative customer is inappropriate
under any circumstances. With large industries, for example, there may be no rela-
tionship at all between the volumes and patterns of usage even though they share a
similar zoning classification. Therefore, demands for large consumers (industries,
hospitals, hotels, and so on) and their diurnal variations should be individually deter-
mined.
Even if data logging cannot be applied to all customers, studying the demands of large
consumers and applying the top-down demand determination concept to the smaller
consumers can still yield reasonable demand calculations. The large customer data is
subtracted from the overall system or zone usage, and the difference in demand is
attributable to the smaller customers.
It is impossible to know with absolute certainty when water will be used or how much
water is used in a short period of time, even though usage per billing period is known
exactly. Bowen, Harp, Baxter, and Shull (1993) collected data from single and multi-
family residential customers in several U.S. cities. These demand patterns can be used
as a starting point for assigning demand patterns to residential nodes.
Buchberger and Wu (1995) and Buchberger and Wells (1996) developed a stochastic
model for residential water demands and verified it by collecting extensive data on
individual residential customers. The model is particularly useful for evaluating the
hydraulics of dead-ends and looped systems in the periphery of distribution networks.
The researchers found that the demand at an individual house cannot simply be multi-
plied by the number of houses to determine the demand in a larger area. The methods
that they developed provide a way of combining the individual stochastic demands
from individual customers who are brushing their teeth or running their washing
machines, dishwashers, and so on into the aggregate for use in a larger area over a
longer time interval.
In general, hotels and apartments have demand patterns similar to those of residential
customers, office buildings have demand patterns corresponding to 8 a.m. to 5 p.m.
operations, and retail area demand patterns reflect 9 a.m. to 9 p.m. operations. Every
large industry that uses more than a few percentage points of total system production
should have an individual demand pattern developed for it.
Defining Usage Patterns within a Model

Usage could be defined directly by describing a series of actual flow versus time
points for each junction in the system. One shortcoming of this type of definition is
that it does not offer much data reuse for nodes with similar usage patterns. Conse-
quently, most hydraulic models express demands by using a constant baseline demand
multiplied by a dimensionless demand pattern factor at each time increment.
A demand multiplier is defined as
Mult i = Q i e Q base (4.11)
where Multii = demand multiplier at the ith time step

Qi = demand in ith time step (gpm, m3/s)
Qbase = base demand (gpm, m3/s)
The series of demand pattern multipliers models the diurnal variation in demand and
can be reused at nodes with similar usage characteristics. The baseline demand is
often chosen to be the average daily demand (although peak day demand or some
other value can be used). Assuming a baseline demand of 200 gpm, Table 4.3 illus-
trates how nodal demands are computed using a base demand and pattern multipliers.
Table 4.3 Calculation of nodal demands using pattern multipliers

Pattern
Time Demand
Multiplier
0:00 0.7 200 gpm u 0.7 = 140 gpm
Table 4.3 Calculation of nodal demands using pattern multipliers

Pattern
Time Demand
Multiplier
1:00 1.1 200 gpm u 1.1 = 220 gpm
2:00 1.8 200 gpm u 1.8 = 360 gpm
As one can imagine, usage patterns are as diverse as the customers themselves. Figure
4.11 illustrates just how different diurnal demand curves for various classifications
can be. A broad zoning classification, such as commercial, may contain differences
significant enough to warrant the further definition of subcategories for the different
types of businesses being served. For instance, a hotel may have a demand pattern that
resembles that of a residential customer. A dinner restaurant may have its peak usage
during the late afternoon and evening. A clothing store may use very little water,
regardless of the time of day. Water usage in an office setting may coincide with cof-
fee breaks and lunch hours.
Figure 4.11
Diurnal curve for Businesses Single Family
different user
Demand Multiplier
categories Demand Multiplier
Time Time
Factory Restaurant
Demand Multiplier
Demand Multiplier
Time Time
There will sometimes be customers within a demand classification whose individual

demand patterns differ significantly from the typical demand pattern assigned to the
classification as a whole. For most types of customers, the impact such differences
have on the model is insignificant. For other customers, such as industrial users, errors
in the usage pattern may have a large impact on the model. In general, the larger the
individual usage of a customer, the more important it is to ensure the accuracy of the
consumption data.
Stepwise and Continuous Patterns. In a stepwise demand pattern, demand

multipliers are assumed to remain constant over the duration of the pattern time step.
A continuous pattern, on the other hand, refers to a pattern that is defined indepen-
dently of the pattern time step. Interpolation methods are used to compute multiplier
values at intermediate time steps. If the pattern time step is reset to a smaller or larger
value, the pattern multipliers are automatically recalculated. The pattern multiplier
value is updated by linearly interpolating between values occurring along the continu-
ous curve at the new time step interval. The result is a more precise curve fit that is
independent of the time step specified, as shown in Figure 4.12.
Figure 4.12
Stepwise and
continuous pattern
variation
Continuous
1.5
Stepwise
Pattern Multiplier
Average
1.0
0.5
0.0
0 8 12 18 24
Time of Day
For example, the pattern from Table 4.3 can be extended to show how a typical model
might determine multipliers after the time step had been changed from 1 hour to 15
minutes over the time period 0:00 to 1:00. As Table 4.4 shows, a pattern multiplier for
an intermediate time increment in a continuous pattern can differ significantly from
its stepwise pattern counterpart.
Table 4.4 Interpolated stepwise and continuous pattern multipliers

Time Pattern Multiplier Stepwise Multiplier Continuous Multiplier
0:00 0.7 0.7 0.7
0:15 0.7 0.7 0.8
0:30 0.7 0.7 0.9
0:45 0.7 0.7 1.0
1:00 1.1 1.1 1.1
Pattern Start Time and Repetition. When defining and working with pat-
terns, it is important to understand how the pattern start time is referenced. Does pat-
tern hour 2 refer to 2:00 a.m., or does it refer to the second hour from the beginning of
a simulation? If a model simulation begins at midnight, then there is no difference

between military time and time step number. If the model is intended to start at some
other time (such as 6:00 a.m., when many systems have refilled all their tanks), then
the patterns may need to be adjusted, advancing or retarding them in time accordingly.
Most modelers accept that demand patterns repeat every 24 hours with only negligible
differences, and are willing to use the same pattern each day in such a way that hours
25 and 49 use the same demands as the first hour. For a factory with three shifts, a pat-
tern may repeat every eight hours. Other patterns may not repeat at all. Each software
package handles pattern repetition in its own way; thus, some research and experi-
mentation may be required to produce the desired behavior for a particular applica-
tion.
4.4 PROJECTING FUTURE DEMANDS

Water distribution models are created not only to solve the problems of today, but also
to prevent problems in the future. With almost any endeavor, the future holds a lot of
uncertainty, and demand projection is no exception. Long-range planning may
include the analysis of a system for 5-, 10-, and 20-year time frames. When perform-
ing long-term planning analyses, estimating future demands is an important factor
influencing the quality of information provided by the model.
Section 4.4 Projecting Future Demands 163
The uncertainty of this process puts the modeler in the difficult position of trying to
predict the future. The complexity of such analyses, however, can be reduced to some
extent with software that supports the creation and comparison of a series of possible
alternative futures. Testing alternative future projections provides a way for the mod-
eler to understand the sensitivity of decisions regarding demand projections. Scenario
management tools in models help make this process easier. Even the most compre-
hensive scenario management, however, is just another tool that needs to be applied
intelligently to obtain reasonable results.
Historical Trends
Since the growth of cities and industries is hard to predict, it follows that it is also dif-
ficult to predict future water demands. Demand projections are only as accurate as the
assumptions made and the methods used to extrapolate development. Some cities
have relatively stagnant demands, but others experience volatile growth that chal-
lenges engineers designing water systems.
How will the economy affect local industries? Will growth rates continue at their cur-
rent rate, or will they level off? Will regulations requiring low-flow fixtures actually
result in a drop in water usage? What will be the combined result of increased popula-
tion and greater interest in water conservation? These questions are all difficult to
answer, and no method exists that can answer them with absolute certainty.
In general, the decision about which alternative future projection should be used is not
so much a modeling decision as a utility-wide planning decision. The modeler alone
should not try to predict the future, but rather facilitate the utility decision-makers’
process of coming to a consensus on likely future demands.
Figure 4.13 illustrates some possible alternative futures given a historical demand pat-
tern. In spite of its shortcomings, the most commonly used method for predicting
demands is to examine historical demand trends and to extrapolate them into the
future under the assumption that they will continue.
Spatial Allocation of Future Demands

Planning departments and other groups may provide population projections for future
years and associate these population estimates with census tracts, traffic analysis
zones, planning districts, or other areas. The data must then be manipulated to deter-
mine the spatial allocation of nodal demands for the water model.
This manipulation requires a good deal of judgment on the modeler’s part, reflecting
the uncertainty of the process. Predictions concerning the future, by their nature, con-
tain varying degrees of uncertainty. If the significant factors affecting community
growth have been identified, the modeler can usually save time by making good judg-
ments about how the current baseline demand allocation can be modified and reused
for planning purposes. It is also important for the modeler to consider the future fire
protection requirements. Because fire protection demands are often much larger than
baseline demands, they are usually a major factor in future pipe-sizing decisions.
Figure 4.13
Several methods for
projecting future 5.0
demands
4.5
Constant Percent Growth
Growth to Buildout
4.0
Peak Day Demand, MGD
3.5 Linear Growth
3.0
Economic Downturn
2.5
Annual Demand Data
2.0
1.5
1960 1970 1980 1990 2000 2010 2020 2030 2040
Time, year
Disaggregated Projections
Rather than basing projections on extrapolation of flow rate data, it is somewhat more
rational to examine the causes of demand changes and then project that data into the
future. This technique is called disaggregated projection. Instead of predicting
demands, the user predicts such things as industrial production, number of hotel
rooms, and cost of water, and then uses a forecasting model to predict demand.
The simplest type of disaggregated demand projection involves projecting population
and per capita demand separately. In this way, the modeler can, for example, separate
the effects of population growth from the effects of a decrease in per capita consump-
tion due to low-volume fixtures and other water conservation measures.
These types of approaches attempt to account for many variables that influence future
demands, including population projections, water pricing, land use, industrial growth,
and the effects of water conservation (Vickers, 1991; and Macy, 1991). The IWR-
Main model (Opitz et al., 1998; Dziegielewski and Boland, 1989) is a sophisticated
model that uses highly disaggregated projections to forecast demands.
The most difficult factor to predict when performing a projection is drastic change in
the economy of an area (for example, a military base closure or the construction of a
factory). Using disaggregated projections, population projections can be modified
Section 4.5 Fire Protection Demands 165
more rationally than can flow projections when developing demand forecasts that
reflect these types of events.
Population Estimates. Planning commissions often have population studies
and estimates that predict the future growth of a city or town. Though population esti-
mates usually contain uncertainties, they can be used as a common starting point for
any model requiring future estimates, such as water distribution models, sewer plans,
and traffic models.
Starting with current per capita usage rates or projections of per capita usage trends,
future demands can be estimated by taking the product of the future population and
the future per capita usage. In areas that are already densely populated, the growth
may be only slightly positive, or even negative.
The United States Geological Survey (USGS) publishes per capita water consumption
rates for each state, but these values include nonmunicipal uses such as power genera-
tion and agriculture. A per capita consumption rate developed in this manner cannot
be widely applied because there are large differences in water consumption among
customers in different areas within a particular state.
Land Use. Sometimes, water demands can be estimated based on land use desig-
nations such as single-family residential, high-density residential, commercial, light
industrial, heavy industrial, and so on. Information regarding a representative water
usage rate based on land use can then aid in planning for other areas that are in the
same category.
As with population estimates, using land use designation requires some level of pre-
diction regarding future growth in every area from residential land use to industrial
and commercial operations. For example, the loss or gain of a single large industry
can have a tremendous effect on the overall consumption in the system.
4.5 FIRE PROTECTION DEMANDS

When a fire is in progress, fire protection demands can represent a huge fraction of
the total demand for the system. The effects of fire demands are difficult to derive pre-
cisely since fires occur with random frequency in different areas, with each area hav-
ing unique fire protection requirements. Generally, the amount of water needed to
adequately fight a fire depends on the size of the burning structure, its construction
materials, the combustibility of its contents, and the proximity of adjacent buildings.
For some systems, fire protection is a lower priority than water quality or construction
costs. To reduce costs in situations in which customers are very spread out, such as in
rural areas, the network may not be designed to provide fire protection. Instead, the
fire departments rely on water tanker trucks or other sources for water to combat fires
(for example, ponds constructed specifically for that purpose).
One of the primary benefits of providing water for fire protection is a reduction in the
insurance rates of residents and businesses in the community. In the United States,
community fire protection infrastructure (the fire-fighting capabilities of the fire
department and the capacity of the water distribution network) is audited and rated by
the Insurance Services Office (ISO) using the Fire Protection Rating System (ISO,
1998). In Canada, the Insurers Advisory Organization (IAO) evaluates water supply
systems using the Grading Schedule for Municipal Fire Protection (IAO, 1974). The
ISO evaluation process is summarized in AWWA M-31 (1998).
In Europe, no fire prevention standards exist that apply to all European countries;
therefore, each country must develop or adopt its own fire flow requirements. For
example, the flow rates that the UK Fire Services ideally require to fight fires are
based on the national guidance document on the provision of water for fire fighting
(Water UK and LGA, 1998). Similarly, the German standards (DVGW, 1978), the
French standards (Circulaire, 1951, 1957, and 1967), the Russian standards (SNIP,
1985), and others, are based on fire risk categories that assign the level of risk accord-
ing to the type of premises to be protected, fire-spread risk, installed fire proofing, or
any combination of these factors.
Because systems will be evaluated using ISO methods, engineers in the United States
usually base design of fire protection systems on the ISO rating system, which
includes determining fire flow demands according to the ISO approach. Although the
actual water needed to fight a fire depends on the structure and the fire itself, the ISO
method yields a Needed Fire Flow (NFF) that can be used for design and evaluation
of the system. Different calculation methods are used for different building types,
such as residential, commercial, or industrial.
For one- and two-family residences, the needed fire flow is determined based on the
distance between structures, as shown in Table 4.5.
Table 4.5 Needed fire flow for residences two stories and less
Distance Between Buildings Fire Flow
(ft) (gpm)
More than 100 500
31-100 750
11-30 1,000
Less than 11 1,500
For commercial and industrial structures, the needed fire flow is based on building
area, construction class (that is, frame or masonry construction), occupancy (such as
a department store or chemical manufacturing plant), exposure (distance to and type
of nearest building), and communication (types and locations of doors and walls). The
formula can be summarized as:
0.5
NFF = 18FA OX + P (4.12)
where NFF = needed fire flow (gpm)

F = class of construction coefficient
A = effective area (ft2)
O = occupancy factor
X = exposure factor
P = communication factor
References 167
The procedure for determining NFF is documented in the Fire Protection Rating Sys-
tem (1998) and AWWA M-31 (1998). The minimum needed fire flow is not less than
500 gpm (32 l/s), and the maximum is no more than 12,000 gpm (757 l/s). Most fre-
quently, the procedure produces values less than 3,500 gpm (221 l/s). Values are
rounded to the nearest 250 gpm (16 l/s) for NFFs less than 2,500 gpm (158 l/s), and to
the nearest 500 gpm (32 l/s) for values greater than 2,500 gpm (158 l/s). Values are
also adjusted if a building is equipped with sprinklers.
In addition to a flow rate requirement, a requirement exists for the duration over
which the flow can be supplied. According to ISO (1998), fires requiring 3,500 gpm
(221 l/s) or less are referred to as receiving “Public Fire Suppression,” and those
requiring greater than 3,500 gpm (221 l/s) are classified as receiving “Individual
Property Fire Suppression.” For fires requiring 2,500 gpm (158 l/s) or less, a two-hour
duration is sufficient; for fires needing 3,000 to 3,500 gpm (190 to 221 l/s), a three-
hour duration is used; and for fires needing more than 3,500 gpm (221 l/s), a four-
hour duration is used along with slightly different rules for evaluation.
Methods for estimating sprinkler demands are based on the area covered and a flow
density in gpm/ft2 as described in NFPA 13 (1999) for commercial and industrial
structures, and in NFPA 13D (1999) for single- and two-family residential dwellings.
For residences, the sprinklers shall provide at least 18 gpm (1.14 l/s) when one sprin-
kler operates and no less than 13 gpm (0.82 l/s) per sprinkler when more than one
operates. For commercial and industrial buildings, the flow density can vary from
0.05 to 0.35 gpm/ft2 (2 to 14 l/min/m2), depending on the hazard class associated with
the building and the floor area. Sprinkler design is covered in detail on page 401.
NFPA 13 provides a chart for determining flow density based on whether occupancy
is light, ordinary hazard, or extra hazard. A hose stream requirement exists as well for
water used to supplement the sprinkler flows. These values range from 100 to 1000
gpm (6.3 to 63 l/s), depending on the hazard classification.
REFERENCES
American Water Works Association (1989). “Distribution Network Analysis for Water Utilities.” AWWA
Manual M-32, Denver, Colorado.
American Water Works Association (1998). “Distribution System Requirements for Fire Protection.”
AWWA Manual M-31, Denver, Colorado.
Basford, C., and Sevier, C. (1995). “Automating the Maintenance of Hydraulic Network Model Demand
Database Utilizing GIS and Customer Billing Records.” Proceedings of the AWWA Computer Confer-
ence, American Water Works Association, Norfolk, Virginia.
Bowen, P. T., Harp, J., Baxter, J., and Shull, R. (1993). Residential Water Use Patterns. AWWARF, Denver,
Colorado.
Brainard, B. (1994). “Using Electronic Rate of Flow Recorders.” Proceedings of the AWWA Distribution
System Symposium, American Water Works Association, Omaha, Nebraska.
Buchberger, S. G., and Wu, L. (1995). “A Model for Instantaneous Residential Water Demands.” Journal of
Hydraulic Engineering, ASCE, 121(3), 232.
Buchberger, S. G., and Wells, G. J. (1996). “Intensity, Duration, and Frequency of Residential Water
Demands.” Journal of Water Resources Planning and Management, ASCE, 122(1), 11.
Buyens, D. J., Bizier, P. A., and Combee, C. W. (1996). “Using a Geographical Information System to
Determine Water Distribution Model Demands.” Proceedings of the AWWA Annual Conference, Ameri-
can Water Works Association, Toronto, Canada.
Cesario, A. L., and Lee T. K. (1980). “A Computer Method for Loading Model Networks.” Journal of the
Cesario, A. L. (1995). Modeling, Analysis, and Design of Water Distribution Systems. American Water
Works Association, Denver, Colorado.
Circulaire des Ministreres de l’Intériur et de l’Agriculture du Février (1957). Protection contre l’incendie
dans les communes rurales. Paris, France.
Circulaire du Ministrere de l’Agriculture du Auout (1967). Réserve d’eau potable. Protection contre
l’incendie dans les communes rurales. Paris, France.
Circulaire Interministérielle du Décembre (1951). Alimentation des communes en eau potable - Lutte con-
tre l’incendie. Paris, France.
Coote, P. A., and Johnson, T. J. (1995). “Hydraulic Model for the Mid-Size Utility.” Proceedings of the
AWWA Computer Conference, American Water Works Association, Norfolk, Virginia.
Davis, A. L., and Brawn, R. C. (2000). “General Purpose Demand Allocator (DALLOC).” Proceedings of
the Environmental and Water Resources Institute Conference, American Society of Civil Engineers,
Minneapolis, Minnesota.
DeOreo, W. B., Heaney, J. P., and Mayer, P. W. (1996). “Flow Trace Analysis to Assess Water Use.” Jour-
nal of the American Water Works Association, 88(1), 79.
DVGW. (1978). “DVGW W405 Bereitstellung von Löschwasser durch die Öffentliche Trinkwasserversor-
gung.” Deutscher Verein des Gas — und Wasserfaches, Franfurt, Germany.
Dziegielewski, B., and Boland J. J. (1989). “Forecasting Urban Water Use: the IWR-MAIN Model.” Water
Resource Bulletin, 25(1), 101 – 119.
Dziegielewski, B., Opitz, E. M., and Maidment, D. (1996). “Water Demand Analysis.” Water Resources
Handbook, Mays, L. W., ed., McGraw-Hill, New York, New York.
Insurance Advisory Organization (IAO) (1974). Grading Schedule for Municipal Fire Protection. Toronto,
Canada.
Insurance Services Office (ISO) (1998). Fire Suppression Rating Schedule. New York, New York.
Macy, P. P. (1991). “Integrating Construction and Water Master Planning.” Journal of the American Water
Works Association, 83(10), 44 – 47.
Male, J. W., and Walski, T. M. (1990). Water Distribution: A Troubleshooting Manual. Lewis Publishers,
Chelsea, Florida.
Metcalf & Eddy, Inc. (1979). Water Resources and Environmental Engineering. 2nd Edition, McGraw-Hill,
New York, New York.
National Fire Protection Association (NFPA) (1999). “Sprinkler Systems in One- and Two-Family Dwell-
ings and Manufactured Homes.” NFPA 13D, Quincy, Massachusetts.
National Fire Protection Association (NFPA) (1999). “Standard for Installation of Sprinkler Systems.”
NFPA 13, Quincy, Massachusetts.
Office of Water Services (Ofwat) (1998). 1997-98 Report on Leakage and Water Efficiency. http://
www.open.gov.uk/ofwat/leak97.pdf, United Kingdom.
Opitz, E. M., et al. (1998). “Forecasting Urban Water Use: Models and Application.” Urban Water Demand
Management and Planning, Baumann D., Boland, J. and Hanemann, W. H., eds., McGraw Hill. New
York, New York, 350.
Rhoades, S. D. (1995). “Hourly Monitoring of Single-Family Residential Areas.” Journal of the American
Water Works Association, 87(8), 43.
References 169
SNIP (1985). Water Supply Standards (in Russian). 2.04.02-84, Moscow, Russia.
Stern, C. T. (1995). “The Los Angeles Department of Water and Power Hydraulic Modeling Project.” Pro-
ceedings of the AWWA Computer Conference, American Water Works Association, Norfolk, Virginia.
Vickers, A. L. (1991). “The Emerging Demand Side Era in Water Conservation.” Journal of the American
Walski, T. M. (1999). “Peaking Factors for Systems with Leakage.” Essential Hydraulics and Hydrology,
Haestad Press, Waterbury, Connecticut.
Walski, T. M., Lowry, S. G., and Rhee, H. (2000). “Pitfalls in Calibrating an EPS Model.” Proceedings of
the Environmental and Water Resource Institute Conference, American Society of Civil Engineers, Min-
neapolis, Minnesota.
Water Research Centre (WRc) (1985). District Metering, Part I - System Design and Installation. Report
ER180E, United Kingdom.
Water UK and Local Government Association (1998). National Guidance Document on the Provision of
Water for Fire Fighting. London, United Kingdom.
Ysuni, M. A. (2000). “System Design: An Overview.” Water Distribution Systems Handbook, Mays, L. W.,
ed., McGraw-Hill, New York, New York.
4.1 Develop a steady-state model of the water distribution system shown in the figure. Data describing
the system and average daily demands are provided in the tables that follow.
P -2
R-1
J -1 P -3 J -2 P -4
J -3
P -1
P -5 P -6 P -7 P -8
J -4 P -9 J -5
Hazen-
Length Diameter Minor Loss
Pipe Label Williams
(ft) (in.) Coefficient
C-factor
P-1 500 12 120 10
P-2 2,600 10 120 0
P-3 860 8 120 0
P-4 840 8 120 5
P-5 710 6 120 0
P-6 1,110 4 120 0
P-7 1,110 4 120 0
P-8 710 6 120 0
P-9 1,700 6 120 0
Elevation Demand
Node Label
(ft) (gpm)
R-1 750 N/A
J-1 550 250
J-2 520 75
J-3 580 125
J-4 590 50
J-5 595 0
a) Fill in the tables below with the pipe and junction node results.
Flow Hydraulic Gradient
Pipe Label
(gpm) (ft/1000 ft)
P-1
P-2
P-3
P-4
P-5
P-6
P-7
P-8
P-9
Hydraulic Grade Pressure

Node Label
(ft) (psi)
J-1
J-2
J-3
J-4
J-5
b) Complete the tables below assuming that all demands are increased to 225 percent of average-
day demands.

Pipe Label
(gpm) (ft/1000 ft)
P-1
P-2
P-3
P-4
P-5
P-6
P-7
P-8
P-9

Node Label
(ft) (psi)
J-1
J-2
J-3
J-4
J-5
c) Complete the tables below assuming that, in addition to average-day demands, there is a fire flow
demand of 1,850 gpm added at node J-3.
Pipe Label
(gpm) (ft/1000 ft)
P-1
P-2
P-3
P-4
P-5
P-6
P-7
P-8
P-9

Node Label
(ft) (psi)
J-1
J-2
J-3
J-4
J-5
4.2 English Units: Perform a 24-hour extended-period simulation with a one-hour time step for the sys-
tem shown in the figure. Data necessary to conduct the simulation are provided in the tables that fol-
low. Alternatively, the pipe and junction node data has already been entered into Prob4-02.wcd. Use
a stepwise format for the diurnal demand pattern. Answer the questions presented at the end of this
problem.
J-9 J-10
Crystal Lake P -15
P -16 P -14
Suction
J-1 J-8 J-7
Discharge
P -12 P -13
West Carrolton
Tank
P -1 P -11 P -10
J-2 J-3 J-6
P -2 P -9 P -8
P -3
J-4
Miamisburg
Tank P -4
P -5
J-5
P -6 P -7
(Not To Scale)

Pipe Label
(ft) (in.) C-factor
Suction 25 24 120
Discharge 220 21 120
P-1 1,250 6 110
P-2 835 6 110
P-3 550 8 130
P-4 1,010 6 110
P-5 425 8 130
P-6 990 8 125
P-7 2,100 8 105
P-8 560 6 110
P-9 745 8 100
P-10 1,100 10 115
P-11 1,330 8 110
P-12 890 10 115
P-13 825 10 115
P-14 450 6 120
P-15 690 6 120
P-16 500 6 120
Elevation Demand
Node Label
(ft) (gpm)
Crystal Lake 320 N/A
J-1 390 120
J-2 420 75
J-3 425 35
J-4 430 50
J-5 450 0
J-6 445 155
J-7 420 65
J-8 415 0
J-9 420 55
J-10 420 20
Pump Curve Data

Head Flow
(ft) (gpm)
Shutoff 245 0
Design 230 1,100
Max Operating 210 1,600
Elevated Tank Information
Miamisburg West Carrolton

Tank Tank
Base Elevation (ft) 0 0
Minimum Elevation (ft) 535 525
Initial Elevation (ft) 550 545
Maximum Elevation (ft) 570 565
Tank Diameter (ft) 49.3 35.7
Diurnal Demand Pattern
Time of Day Multiplication Factor

Midnight 1.00
6:00 am 0.75
Noon 1.00
6:00 pm 1.20
Midnight 1.00
a) Produce a plot of the HGL in the Miamisburg and West Carrolton tanks as a function of time.
b) Produce a plot of the pressures at node J-3 versus time.

SI Units: Perform a 24-hour extended-period simulation with a one-hour time step for the system
shown in the figure. Data necessary to conduct the simulation are provided in the tables below. Alter-
natively, the pipe and junction node data has already been entered into Prob4-02m.wcd. Use a step-
wise format for the diurnal demand pattern. Answer the questions presented at the end of this
problem.
Pipe Label
(m) (mm) C-factor
Suction 7.6 610 120
Discharge 67.1 533 120
P-1 381.0 152 110
P-2 254.5 152 110
P-3 167.6 203 130
P-4 307.8 152 110
P-5 129.5 203 130
P-6 301.8 203 125
P-7 640.1 203 105
P-8 170.7 152 110
P-9 227.1 203 100
P-10 335.3 254 115
P-11 405.4 203 110
P-12 271.3 254 115
P-13 251.5 254 115
P-14 137.2 152 120
P-15 210.3 152 120
P-16 152.4 152 120
Elevation Demand
Node Label
(m) (l/s)
Crystal Lake 97.5 N/A
J-1 118.9 7.6
J-2 128.0 4.7
J-3 129.5 2.2
J-4 131.1 3.2
J-5 137.2 0
J-6 135.6 9.8
J-7 128.0 4.1
J-8 126.5 0
J-9 128.0 3.5
J-10 128.0 1.3
Pump Curve Data
Head Flow
(m) (l/s)
Shutoff 74.6 0
Design 70.1 69
Max Operating 64.0 101
Elevated Tank Information
Miamisburg West Carrolton

Tank Tank
Base Elevation (m) 0 0
Minimum Elevation (m) 163.1 160.0
Initial Elevation (m) 167.6 166.1
Maximum Elevation (m) 173.7 172.2
Tank Diameter (m) 15.0 10.9
Time of Multiplication
Day Factor
Midnight 1.00
6:00 a.m. 0.75
Noon 1.00
6:00 p.m. 1.20
Midnight 1.00
a) Produce a plot of the HGL in the Miamisburg and West Carrolton tanks as a function of time.
b) Produce a plot of the pressures at node J-3 versus time.
4.3 Develop a steady-state model for the system shown in the figure and answer the questions that fol-
low. Data necessary to conduct the simulation are provided in the following tables. Alternatively, the
pipe and junction node data has already been entered into Prob4-03.wcd. Note that there are no
minor losses in this system. The PRV setting is 74 psi.
Lower Pressure Zone Newtown Res.

J-10 P-16 J-9 P-15 J-8
P-1
PRV-1 P-14
P-17 P-19
PMP-1
P-2
J-11 P-18 J-3 P-5 P-4 J-2 P-3
J-1
P-6
Central Tank P-13
J-4 J-7 P-9

Pressure Zone Boundary P-20
P-7
P-12
J-5 P-8 J-6

PMP-2
P-11
Higher Pressure Zone
P-10
High Field Res.


Pipe Label
(ft) (in.) C-factor
P-1 120 24 120
P-2 435 16 120
P-3 2,300 12 120
P-4 600 10 110
P-5 550 10 110
P-6 1,250 12 110
P-7 850 12 110
P-8 4,250 12 120
P-9 2,100 12 120
P-10 50 24 105
P-11 250 16 105
P-12 1,650 10 115
P-13 835 8 110
P-14 800 8 100
P-15 1,300 6 95
P-16 1,230 6 95
P-17 750 6 95
P-18 1,225 8 95
P-19 725 6 100
P-20 155 4 75
Elevation Demand
Node Label
(ft) (gpm)
High Field Reservoir 1,230 N/A
Newtown Reservoir 1,050 N/A
Central Tank 1,525 N/A
J-1 1,230 0
J-2 1,275 0
J-3 1,235 120
J-4 1,250 35
J-5 1,300 55
J-6 1,250 325
J-7 1,260 0
J-8 1,220 100
J-9 1,210 25
J-10 1,210 30
J-11 1,220 45
PRV-1 1,180 N/A
PMP-1 1,045 N/A
PMP-2 1,225 N/A
Pump Curve Data
PMP-1 PMP-2
Head Flow Head Flow
(ft) (gpm) (ft) (gpm)
Shutoff 550 0 320 0
Design 525 750 305 1,250
Max Operating 480 1,650 275 2,600
a) Fill in the tables for pipe and junction node results.

Pipe Label
(gpm) (ft/1000 ft)
P-1
P-2
P-3
P-4
P-5
P-6
P-7
P-8
P-9
P-10
P-11
P-12
P-13
P-14
P-15
P-16
P-17
P-18
P-19
P-20

Node Label
(ft) (psi)
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
J-11
Analyze the following demand conditions for this system by using the average-day demands as your
base demands.
b) Increase all demands to 150 percent of average-day demands. What are the pressures at nodes J-2
and J-10?
c) Add a fire flow demand of 1,200 gpm to node J-4. What is the discharge from the Newtown
pump station? What is the pressure at node J-4?
d) Replace the demand of 120 gpm at node J-3 with a demand of 225 gpm. How does the pressure at
node J-3 change between the two demand cases?
e) Replace the existing demands at nodes J-3, J-9, J-10, and J-11 with 200 gpm, 50 gpm, 90 gpm,
and 75 gpm, respectively. Is Central Tank filling or draining? How does the tank condition com-
pare with the original simulation before demands were changed?
4.4 Perform an extended-period simulation on the system from part (a) of Problem 4.3. However, first
add a PRV to pipe P-6 and close pipe P-14. Note that pipe P-6 must split into two pipes when the
PRV is inserted. Specify the elevation of the PRV as 1,180 ft and the setting as 74 psi.
The simulation duration is 24 hours and starts at midnight. The hydraulic time step is 1 hour. The
capacity and geometry of the elevated storage tank and the diurnal demand pattern are provided
below. Assume that the diurnal demand pattern applies to each junction node and that the demand
pattern follows a continuous format. Assume that the High Field pump station does not operate.
Central Tank Information
Base Elevation (ft) 1,260

Minimum Elevation (ft) 1,505
Initial Elevation (ft) 1,525
Maximum Elevation (ft) 1,545
Tank Diameter (ft) 46.1
Time of Day Multiplication Factor

Midnight 0.60
3:00 a.m. 0.75
6:00 a.m. 1.20
9:00 a.m. 1.10
Noon 1.15
3:00 p.m. 1.20
6:00 p.m. 1.33
9:00 p.m. 0.80
Midnight 0.60
a) Produce a plot of HGL versus time for Central Tank.
b) Produce a plot of the discharge from the Newtown pump station versus time.
c) Produce a plot of the pressure at node J-3 versus time.
d) Does Central Tank fill completely? If so, at what time does the tank completely fill? What hap-
pens to a tank when it becomes completely full or completely empty?
e) Why does the discharge from the Newtown pump station increase between midnight and 6:00
a.m.? Why does the discharge from the pump station decrease, particularly after 3:00 p.m.?
f) Does the pressure at node J-3 vary significantly over time?
4.5 Given a pressure zone with one pump station pumping into it and a smaller one pumping out of it,
and a single 40-ft diameter cylindrical tank, develop a diurnal demand pattern. The pumping rates
and tank water levels are given in the table below. The pumping rates are the average rates during the
hour, and the tank levels are the values at the beginning of the hour.
What is the average use in this pressure zone? What is the average flow to the higher pressure zone?
Pump In Pump Out Tank Level

Hour
(gpm) (gpm) (ft)
0 650 0 35.2
1 645 210 38.5
2 645 255 40.4
3 652 255 42.1
4 310 255 43.5
5 0 255 42.8
6 0 255 39.6
7 0 0 36.0
8 0 0 33.4
9 225 0 30.3
10 650 0 28.9
11 650 0 30.5
12 650 0 32.1
13 650 0 33.8
14 650 45 35.8
15 645 265 37.5
16 645 260 37.5
17 645 260 37.2
18 645 260 36.5
19 645 260 36.4
20 645 255 36.7
21 645 150 37.2
22 115 0 38.7
23 0 0 38.3
24 0 0 37.1
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
5
Testing Water Distribution Systems
Verifying that a water distribution model replicates field conditions requires an inti-
mate knowledge of how the system performs over a wide range of operating condi-
tions. For example, can the model reproduce the flow patterns and pressures that
occur during periods of peak summertime usage, or can the model accurately simulate
chlorine decay? Collecting water distribution system data in the field provides valu-
able insight into system performance and is an essential part of calibration.
Data collection, the first step in the model calibration process, is discussed in depth
within this chapter. The chapter begins with a brief discussion of system testing,
including descriptions of some simple tests for measuring flow and pressure, as well
as some of the pitfalls that may be encountered. The details of performing fire hydrant
flow tests, head loss tests, pump performance tests, and water quality tests are dis-
cussed as well. The chapter concludes with a discussion of the importance of data
quality, particularly when automated calibration methods are used.
5.1 TESTING FUNDAMENTALS
Pressure Measurement
Pressures are measured throughout the water distribution system to monitor the level
of service and to collect data for use in model calibration. Pressure readings are com-
monly taken at fire hydrants (see Figure 5.1) but can also be read at hose bibs (also
called spigots); home faucets; pump stations (both suction and discharge sides);
tanks; reservoirs; and blow-off, air release, and other types of valves.
If the measurements are taken at a location other than a direct connection to a water
main (for example, at a house hose bib), the head loss between the supply main and
the site where pressure is measured must be considered. Of course, the best solution is
to have no flow (and hence no head loss) between the main and the gage. To check if
flow into the building is occurring, listen at the hose bib for the sound of rushing water.
182 Testing Water Distribution Systems Chapter 5
Figure 5.1
Pressure gages on a
fire hydrant
When measuring pressure, slight fluctuations may be seen on the gage due to chang-
ing flows in the system. Devices such as pressure snubbers and liquid-filled pressure
gages can be used to dampen the pressure fluctuations, unless the fluctuations them-
selves are a source of interest.
Pressure gages are most accurate when measuring pressures within 50 to 75 percent
of the maximum value on the scale. Using several pressure gages of varying pressure
ranges is advisable when working with a water distribution system. A pressure gage
with a range of 0 to 100 psi (690 kPa) is commonly used; however, a pressure gage
that can read up to 200 psi (1,380 kPa) may be necessary for measurements taken at a
pump discharge or at a low elevation. If pressure measurements are taken on the suc-
tion side of a pump, then a pressure gage capable of reading negative pressures, called
a pressure-vacuum gage, may be required. Remember that it is the elevation of the
gage, not the elevation of the node, that is used in calculating the elevation of the HGL
(see page 252).
Flow Measurement
Flow is measured at key locations throughout a system to provide insight into flow
patterns and system performance, develop consumption data, and determine flow
rates for calibration.
Many of the tests described in this chapter require measuring flow in pipes. A variety
of flow meters are available for this purpose, including Venturi meters, magnetic flow-
meters, and ultrasonic meters. Pressure and flow metering and recording equipment
should be calibrated regularly and undergo routine performance checks to ensure that
it is in good working order. Furthermore, even if a flow meter is accurate and cali-
brated, the monitoring station may use an analog gage or dial readout that has a coarse
level of precision, which limits the overall precision.
Section 5.1 Testing Fundamentals 183
The extent of flow measurement employed varies from system to system. Usually,
flow is measured continuously at only a few key locations in the distribution system
such as treatment plants and pump stations. Data from these sites should be used to
the greatest extent possible in system calibration. Flow from higher to lower pressure
zones can also be measured at pressure zone boundaries using combination pressure
reducing valve/flow meters (Walski, Gangemi, Kaufman, and Malos, 2001). More
rarely, systems employ in-line flow meters at key points throughout the network and
transmit the flow rates back to a control center using Supervisory Control and Data
Acquisition (SCADA) systems and telemetry (See Chapter 6). This type of comprehen-
sive flow monitoring is not typically done in the United States; however, more utility
managers and operators are starting to see the value of in-line flow information.
Temporary flow metering may be a cost-effective option to check pump discharges or
to see if in-line flow measurements are required throughout the system. Field mea-
surement using a Pitot rod is shown in Figure 5.2. The rod is inserted into the pipe to
measure total head and pressure head, which can then be converted into velocity
(Walski, 1984a). The Pitot rod should not be confused with the Pitot gage, which
measures velocity head only. Clamp-on or insertion electromagnetic or ultrasonic
meters may also be used.
Placement of the flow-measuring device is important. To be sure that disturbances
caused by any bends or obstructions do not influence the readings, the device should
be placed far enough downstream of the disturbance (usually at a distance of approxi-
mately 10 times the pipe diameter) that the effects will have completely dissipated.
In certain cases it may be desirable to isolate one end of the pipe such that all of the
flow through the pipe is diverted through a hydrant for measurement. The hydrant
flow can then be measured with a hydrant Pitot gage as described in Section 5.2.
Net flow in and out of a tank during a time period can be measured by monitoring
water level in the tank and then calculating the flow based on cross-sectional area in
the tank.
Figure 5.2
Tip of Pitot rod
inserted into clear
pipe
Potential Pitfalls in System Measurements

Flow measurement tests can be beneficial, but there are potential drawbacks to keep
in mind. Testing may result in disruption of service to some customers. For example,
fire flow tests typically cause lower than normal pressures and higher than normal
velocities, particularly in residential areas. Higher velocities can entrain sediments in
pipes or shear against tuberculation on pipe walls, causing customers to experience
discolored water.
Customers may, either by accident or necessity, be disconnected from the system
when valves are operated to facilitate flow tests. As described in the following sec-
tions, head loss tests require the operation of system valves to isolate sections of water
main. Valve operation needs to be carefully planned when conducting such tests to
avoid inadvertently disconnecting customers from the system. To avoid surprises, cus-
tomers should be notified prior to the tests.
5.2 FIRE HYDRANT FLOW TESTS

Obtaining data for a wide range of operating conditions, including peak (high)
demand periods, would be difficult without fire hydrant flow tests. These tests can be
used to simulate high flow conditions (see page 218) and allow the system behavior to
be analyzed under extreme conditions. Fire hydrant flow tests are primarily used to
measure the fire flow capacity of the system. They also provide data on pressures
within the system under static conditions (no hydrants flowing) and stressed condi-
tions (high flows occurring at the hydrants) and can be used in conjunction with the
hydraulic model to calibrate parameters such as pipe roughness (Walski, 1988). Pro-
cedures for conducting fire hydrant flow tests are described in AWWA (1989) and
ISO (1963).
Section 5.2 Fire Hydrant Flow Tests 185
Two or more hydrants are required to perform a fire hydrant flow test, as illustrated in
Figure 5.3. One hydrant is identified as the residual hydrant(s), where all pressure
measurements are taken, and the other is identified as the flowed hydrant(s), where all
flow measurements are taken. When the flowed hydrant(s) is closed, referred to as
static conditions, the pressure at the residual hydrant is called the static pressure.
When one or more of the flowed hydrants are open, referred to as flowed conditions,
the pressure at the residual hydrant is called the residual pressure.
Figure 5.3
Hydrant flow test
Residual Flowed Hydrants
Hydrant
Q1 Q2 Q3
//=//= //=//= //=//=

Pressure
Gage
Conducting a fire hydrant flow test is a simple procedure, and a number of these tests
can be conducted throughout the system in a day’s time. Although not essential, many
utilities have a policy requiring that the residual hydrant be opened and allowed to
flow prior to connecting the pressure gage. This precaution helps remove any particles
that have accumulated in the hydrant lateral and barrel since it was last exercised.
After that, a pressure gage is connected to the residual hydrant and a static pressure
reading taken.
Next, the first of the flowed hydrants is opened and flowed. Once the readings stabi-
lize, a reading is taken at the flowed hydrant using a hand-held or clamp-on Pitot gage
(shown in Figure 5.4) or a Pitot diffuser (shown in Figure 5.5). Meanwhile, another
pressure reading is taken at the residual hydrant. Once the residual pressure is taken
and the discharge rate of the flowed hydrant is recorded, the same procedure can be
repeated for additional hydrants if needed.
The number of hydrants that should be flowed during a test is determined by the pres-
sure drop observed at the residual hydrant. Usually, a drop of at least 10 psi (70 kPa)
is needed to give good results. In a 6- to 8-in. pipe (150 to 200 mm), flowing a single
hydrant is sufficient. For larger pipes, more hydrants may need to be flowed.
Pitot Gages and Diffusers

Because a Pitot gage (shown in Figure 5.4) converts virtually all of the velocity head
associated with the flow stream to pressure head, the Pitot gage pressure reading can
be converted to a hydrant discharge rate using the orifice relationship in Equation 5.1.
2
Q = Cf Cd D P (5.1)
Figure 5.4
Hand-held Pitot gage
where Q = hydrant discharge (gpm, l/s)

Cd = discharge coefficient
D = outlet diameter (in., cm)
P = pressure reading from Pitot gage (psi, kPa)
For a typical 2.5-in. (64 mm) outlet with a discharge coefficient of 0.9, Equation 5.1
can be reduced to:
Q = 167 P
The discharge coefficient in Equation 5.1 accounts for the decrease in the diameter of
flow that occurs between the hydrant opening and the end of the Pitot gage, as well as
the head losses through the opening. The coefficient depends on the geometry of the
inside of the hydrant opening and can be determined by feeling the inside of the
hydrant nozzle (see Figure 5.6).
The Pitot diffuser is similar to a Pitot gage except that it incorporates a nozzle that
redirects the flow from the hydrant, reducing its momentum and thus the potential for
erosion. Because the velocity head sensor is measuring inside the diffuser at a point
where the pressure is not equal to zero, a slightly modified formula is required to
compute flow. This formula varies with the manufacturer of the diffuser (Walski and
Lutes, 1990; and Morin and Rajaratnam, 2000). For example, for the Pitot diffuser
shown in Figure 5.5, the coefficient of 167 given previously reduces to 140.
Figure 5.5
Pitot diffuser
Figure 5.6
Discharge coefficients
at hydrant openings
Rounded Square and Sharp Projecting

Cd = 0.9 Cd = 0.8 Cd = 0.7
To briefly review, the procedure for conducting a fire hydrant flow test is as follows:
1. Place a pressure gage on the residual hydrant and record static pressure.
2. Take the 2 ½-in. (64 mm) cap off of the flowed hydrant.
3. Feel the inside of the hydrant opening to determine its geometry.
4. Slowly start the flow.
5. Once readings stabilize, take a Pitot gage reading at the flowed hydrant(s).
6. Simultaneously measure the residual pressure(s) at the residual hydrant(s).
7. Slowly close the hydrants.
8. Assign the discharge coefficient according to the geometry of the hydrant open-
ing.
9. Determine the hydrant discharge rate by using Equation 5.1 or the equation pro-
vided by the Pitot diffuser manufacturer.
Once all of the data have been collected, a table similar to Table 5.1 can be con-
structed to present the results of the fire hydrant flow test.
Table 5.1 Results of fire hydrant flow test

Residual Hydrant #1 Hydrant #2 Hydrant #3 Total
Number of
Pressure Discharge Discharge Discharge Discharge
Hydrants Flowing
(psi) (gpm) (gpm) (gpm) (gpm)
0 78 N/A N/A N/A 0
1 72 1,360 N/A N/A 1360
2 64 1,150 975 N/A 2125
3 49 850 745 600 2195
When sufficient resources are available, additional residual pressure measurements

can be taken during the fire hydrant flow test at various locations throughout the sys-
tem. Taking these additional pressure readings will provide more information on how
the hydraulic grade changes across the system. Depending on the nature of the water
distribution system, the pressure drop may be localized to the vicinity of the flowed
hydrants.
If the hydrant flow test is conducted to provide data for model calibration, it is
extremely important to note the boundary conditions at the time of the test. Recall that
boundary conditions reflect the water levels in tanks and reservoirs, as well as the
operational status of any high-service pumps, booster pumps, or control valves (for
example, pressure reducing valves) for both static and flowed conditions. As will be
discussed in Chapter 7 (see page 261), these boundary conditions must also be
defined in the hydraulic model.
In addition, system demands in place at the time of the test need to be replicated in the
model. It is important to note the time of day and the weather conditions when the test
was performed to assist in establishing the demands and boundary conditions.
Potential Problems with Fire Flow Tests

Fire hydrant flow tests are a useful tool. They do, however, present some areas of con-
cern. Because the discharges from fire hydrants can be quite large, the following sug-
gestions can reduce potential problems associated with these flows.
1. Minimize the period of time over which hydrants are flowed to limit flooding
potential. (In some locations it may be necessary to dechlorinate water before it
can be discharged into receiving waters.)
2. Direct the flow through the 2 ½-in. (64 mm) nozzle opening instead of the 4 ½-in.
(115 mm) opening. This will help to reduce street flooding while still producing
flow velocities sufficient for calibration.
Evaluating Distribution Capacity

with Hydrant Tests
The results of hydrant flow tests described in this 3. All pumps and regulating valves that would
chapter are used primarily to evaluate the distribu- open during an actual fire are open and operat-
tion system’s capacity to provide water for fighting ing during the test.
fires. The standard formula for converting the test 4. There is sufficient water quantity to supply the
flow to the distribution capacity at some desired fire throughout the duration of the fire event.
residual pressure—usually 20 psi (135 kPa)—was
5. Tank level is at normal day low level.
developed by the Insurance Services Office
(1963), and is given in AWWA M-17 (1989) as: 6. The residual and flowed hydrants are close to
one another (Walski, 1984b).
P s – P r 0.54
Q r = Q t §© -----------------·¹ Water system models can explicitly account for
Ps – Pt these factors and are a more accurate and flexible
where Qr = fire flow at residual pressure Pr (gpm, l/s) way of assessing available fire flow at a given
Qt = hydrant discharge during test (gpm, l/s)
residual pressure. However, this equation is still
widely used.
Ps = static pressure (psi, kPa)
Pr = desired residual pressure (psi, kPa) The previous equation can also be rearranged to
Pt = residual pressure during test (psi, kPa) provide a rough estimate of residual pressure for
some future flow, given hydrant flow test results,
The value of Qr is referred to as the distribution
according to
main capacity in that location, and is used in eval-
uation of water systems for insurance purposes. Q r 1.85
P r = P s – P s – P t § ------·
Assumptions made when using the above equa- © Qt¹
tion are as follows:
In this case, Qr is the estimated flow, and Pr is the
1. Head loss is negligible during static conditions.
pressure that will exist at that flow rate, given that
2. Demands correspond to maximum day all other conditions remain the same.
demands.
3. Use hydrant diffusors to reduce the high velocity of the hydrant stream. This will
help to avoid erosion problems and damage to vegetation.
4. Conduct fire hydrant flow tests during warm weather to avoid ice problems.
5. Notify customers who may be impacted by the test beforehand. In some systems,
hydrant flow tests can stir up sediments and rust, causing temporary water quality
problems.
6. Make sure to open and close the hydrants gradually, as sudden changes in flow
can induce dangerous pressure surges in the system.
7. Make sure that the residual and flowed hydrants are hydraulically close to one
another. It is possible to have two hydrants that are near each other at the street
but are fed by different mains that may not be hydraulically connected for several
blocks. Ideally, the flowed and residual hydrants would be located side-by-side
on the same pipeline, but because this will almost never be the case, accuracy can
instead be improved by minimizing the flow between the hydrants. (The flow can
often be reduced by bracketing the residual hydrant between two flowed
hydrants.)
Using Fire Flow Tests for Calibration

In addition to measuring the fire protection capacity of the network, fire hydrant flow
tests can provide valuable data for hydraulic model calibration. To use the results of a
test, a demand equivalent to the hydrant discharge should be assigned to the junction
node in the model that corresponds to the flowed hydrant. When the hydraulic simula-
tion is conducted, the HGL at the junction node representing the residual hydrant
should agree with the HGL measured in the field. Note that comparisons between
field measurements and model results should be done in terms of HGL, not pressure
(see page 252). Although this section refers to pressure comparisons, remember that
in practice, the field pressures should be converted to the equivalent HGL before com-
paring them to the model results.
Consider the system shown in Figure 5.7 and the results of the fire hydrant flow test
presented in Table 5.1. The top half of the figure illustrates the model representation
of a series of hydrants where nodes J-23, J-24, and J-25 correspond to Hydrants 1, 2,
and 3 respectively; and J-22 corresponds to the residual hydrant. The hydrant flow test
results outlined in Table 5.1 can be described in four unique scenarios:
• Static conditions where none of the hydrants are flowing
• Hydrant 1 is flowing
• Hydrants 1 and 2 are flowing simultaneously
• Hydrants 1, 2, and 3 are flowing simultaneously
Figure 5.7
Field measured flows
are modeled as Q1 Q2 Q3
demands in a network
simulation
J-22 J-23 J-24 J-25
Residual
Hydrant Hydrant 1 Hydrant 2 Hydrant 3
Q1 Q2 Q3
//=//= //=//= //=//=

Pressure
Gage
The scenario in which only Hydrant 1 is flowing results in a discharge of 1,360 gpm
(0.086 m3/s) and a residual pressure of 72 psi (497 kPa). Therefore, a demand of 1,360
gpm will be placed at model node J-23, and when the hydraulic simulation is con-
ducted, the pressure computed at node J-22 will be compared to the residual pressure
of 72 psi measured in the field. If the pressure at J-22 is close to that figure, the model
will be nearly calibrated (at least for this one condition).
On the other hand, if the pressure at J-22 is not close to the measured pressure, adjust-
ments need to be made to the model to bring it into better agreement. Identifying the
actual adjustments that need to be made depends on the cause of the discrepancy.
Section 5.3 Head Loss Tests 191
Chapter 7 has more information regarding reasons why differences might occur as
well as details on modeling the results of flow tests. The procedure described previ-
ously is repeated for each of the flowed conditions, and the parameters are changed as
necessary to obtain a suitable match between observed and computed pressures.
It is critical that the modeling nodes used to represent the hydrants are placed in
exactly the same location as the hydrants in the field. Accurate placement is particu-
larly important for calibration purposes, as illustrated in the following example.
In Figure 5.8a, the pressure measurements are taken at the residual hydrant, and the
model representation of the hydrant is at J-35 (Figure 5.8b), a few hundred feet away.
The modeler may have justified this simplification by reasoning that the locations of
the residual hydrant and node J-35 are relatively close together, and that the pressures
should be similar because the elevations are approximately the same. During calibra-
tion, the modeler then (mistakenly) compares the field-measured pressure at the resid-
ual hydrant to the modeled pressure at J-35 and adjusts the model to achieve an
acceptable match.
Figure 5.8
Residual Importance of node
Hydrant location
J-35
Q Q
6" Water Main
8" Water Main
(a) (b)
What the modeler has failed to consider in this situation is the head loss between the
two points (J-35 and the actual hydrant location) during the fire hydrant flow test. If
the head loss is significant, the computed pressure at node J-35 would be higher than
the computed pressure at the residual hydrant. By trying to match pressures at differ-
ent locations, the modeler could introduce inaccuracies into the model. The subject of
model calibration and the use of fire hydrant flow tests for that purpose are treated in
greater detail in Chapter 7.
5.3 HEAD LOSS TESTS

The purpose of a head loss test is to directly measure the head loss and discharge
through a length of pipe—information that can then be used to compute the pipe
roughness. Head loss tests can be performed using either the two-gage or the parallel-
pipe method. The two-gage method uses pressure readings from two standard pres-
sure gages to determine the head loss over the pipe length, and the parallel-pipe
method uses a single pressure differential gage to find the head loss.
The length of water main being tested is typically located between two fire hydrants.
During a head loss test, valves are closed downstream of the length of test pipe to
hydraulically isolate the test section. Thus, all flow through the section is directed to
the downstream fire hydrant for measurement. Assuming that the internal pipe diame-
ter is known, head loss, pipe length, and flow rate are then measured between the two
points and used to compute the internal pipe roughness using the expressions for the
Hazen-Williams C-factor and the Darcy-Weisbach friction factor (Equations 5.2 and
5.3).
1.852 1 e 1.852
§ C f LQ ·
C = ¨ -----------------------
4.87
-¸ (5.2)
© hL D ¹
where C = Hazen-Williams C-factor

L = length of test section (ft, m)
Q = flow through test section (cfs, m3/s)
hL = head loss due to friction (ft, m)
D = diameter of test section (ft, m)
D2g
f = h L ----------2- (5.3)
LV

g = gravitational acceleration constant (32.2 ft/s2, 9.81 m/s2)
V = velocity through test section (ft/s, m/s)
The velocity is determined from the flow and diameter by using Equation 2.9:
4Q-
V = ---------
2
SD
To apply the friction factor to other pipes, it is necessary to convert f to absolute

roughness. Equation 5.4 is the Colebrook-White formula solved for roughness.
1 · 2.51
H- = 3.7 exp § -------------------
--- – ------------ (5.4)
D © – 0.86 f¹ Re f
where H = absolute roughness

Re = Reynolds number
For smooth pipes, the above equation can occasionally yield negative numbers, which
should be converted to zero roughness (that is, hydraulically smooth pipe).
Two-Gage Test
For the two-gage test (shown in Figure 5.9), the test section is located between two
fire hydrants and is isolated by closing the downstream valves. The pressures at both
of the fire hydrants are measured using standard pressure gages, and these pressures
are then converted to HGLs. The head loss over the test section is then computed as
the difference between the HGLs at the two fire hydrants, as shown in Equation 5.5.
McEnroe, Chase, and Sharp (1989) found that to overcome uncertainties in measuring
length, diameter, and flow, a pressure drop of 15–20 psi (100 - 140 kPa) should be
attained.
Figure 5.9
The two-gage head
Flowed
loss test
Hydrant 1 Hydrant 2 Hydrant
//=//= //=//= //=//=

Pressure
Gage
Test Section Closed Valve
h L = HGL U – HGL D (5.5)
where HGLU = hydraulic grade at upstream fire hydrant (ft, m)

HGLD = hydraulic grade at downstream fire hydrant (ft, m)
Realizing that the HGL can be more generally described using the difference in pres-
sure and elevation between the upstream and downstream hydrants, Equation 5.5 can
be rearranged to yield
hL = Cf PU – PD + ZU – ZD (5.6)
where PU = pressure at upstream fire hydrant (psi, kPa)

PD = pressure at downstream fire hydrant (psi, kPa)
ZU = elevation at upstream fire hydrant (ft, m)
ZD = elevation at downstream fire hydrant (ft, m)
Head loss occurs only when there is a flow; therefore, if no flow is passing through
the test section, the HGL values at the upstream and downstream hydrants will be the
same. Even so, the pressures at the upstream and downstream hydrants may be differ-
ent as a result of the elevation difference between them. Assuming a no-flow condi-
tion, the head loss in Equation 5.6 is set to zero and the elevation difference can be
expressed through the use of pressures, as shown in the following equation.
Z U – Z D = – C f P US – P DS (5.7)
where PUS = pressure at upstream hydrant, static conditions (psi, kPa)

PDS = pressure at downstream hydrant, static conditions (psi, kPa)
Substituting Equation 5.7 into 5.6 provides a new expression for determining the head
loss between two hydrants. This expression eliminates the need to obtain the elevation
of the pressure gages by using two sets of pressure readings: static and flowed.
h L = C f > P UT – P DT – P US – P DS @ (5.8)
where PUT = pressure at upstream hydrant, flowed conditions (psi, kPa)

PDT = pressure at downstream hydrant, flowed conditions (psi, kPa)
In some situations, the test section may be located near a permanent system meter,
such as at the discharge of a pump station, and thus the flow meters at the pump sta-
tion can be used instead of a hydrant. A pressure gage located on the pipe just before
it leaves the pump station can give the upstream pressure. The downstream pressure
must be measured sufficiently far away such that the head loss will be much greater
than the error associated with measuring it. It may be necessary to close valves at tees
and crosses along the pipeline to obtain this long run of pipe with constant flow. Wal-
ski and O’Farrell (1994) described how head loss testing equipment can be installed
with important transmission mains to assist routine head loss testing.
Parallel-Pipe Test
Figure 5.10 illustrates the concept of the parallel-pipe head loss test. As with the two-
gage test, a test section is isolated between two hydrants by closing the downstream
valves. Then a hose equipped with a differential pressure gage is connected between
the two hydrants in parallel with the pipe test section. Because there is no flow, and
consequently no head loss, through the hose or gage, the hydraulic grade on each side
of the gage is equal to the hydraulic grade of the hydrant on that same side. Therefore,
the measured pressure differential can be used in the following expression to calculate
the head loss through the pipe.
h L = C f u 'P (5.9)
where 'P = differential pressure reading (psi, kPa)

The head loss, or pressure head difference, over the test section can be found for any
fluid by dividing the differential pressure ( 'P ) by the specific weight of the fluid ( J ).
Because the pressure readings are taken at one location (at the pressure differential
gage), there is no need to consider the elevation of either hydrant. However, if water
in the parallel hose is allowed to change temperature from the water in the pipes,
errors can occur (Walski, 1985). Accordingly, water in the hose should be kept mov-
ing whenever a reading is not being taken. This can be accomplished by opening a
small valve (pit-cock) at the differential pressure gage.
Figure 5.10
Parallel-pipe head loss
Small Differential
test
Diameter Pressure
Hose Gage Flowed
//=//= //=//= //=//=
Test Section Closed Valve
The procedure for finding the discharge through the test section is similar to the one
used for fire hydrant flow tests. A Pitot gage is used to measure the velocity head at
the flowed hydrant, assuming the flow out of the hydrant equals the flow through the
test section. The orifice formula (Equation 5.1) is then used to convert the Pitot gage
reading into the discharge from the hydrant (McEnroe, Chase, and Sharp, 1989).
McEnroe, Chase, and Sharp (1989) found that to overcome uncertainties in measuring
length, diameter, and flow, a pressure drop of 2–3 psi (14 - 21 kPa) for the parallel-
pipe method should be attained.
Potential Problems with Head Loss Tests

Regardless of the method used for measuring head loss, all flow that passes through
the test section is directed out of a flowed hydrant by closing the valve downstream of
the flowed hydrant. When working with a looped system, isolation valves on some
side mains may also be closed, as shown in Figure 5.11. To ensure that no customers
are taken out of service when closing valves, the utility should examine system maps
to verify that alternate flow paths (loops) are available within the system. As a check,
have one individual watch the pressure gage as the valve is being closed and be ready
to give a signal if the pressure drops to zero.
Frequently, there will be customers connected to the test section between the two
hydrants. To obtain accurate results, the customer water usage during the head loss
tests should be negligible compared to the amount of water discharged through the
flowed hydrant. Recall from Equations 5.2 and 5.3 that the discharge is assumed to
reflect the total amount of water that passes through the test section. Therefore, if the
amount of water that passes through the test section is significantly different from the
measured discharge due to withdrawals at other points in the system, a correction

must be made.
Figure 5.11
Use of isolation
Side
valves during a head
Main
loss test Flowed
Closed Valves
Closed Valve
Test Section
Side
Main
Using Head Loss Test Results for Calibration

The process of using the results of a head loss test is fairly straightforward. Head loss
tests provide information on the internal roughness of a pipe; therefore, once the head
loss tests are complete, the calculated roughness values can simply be supplied to the
computer model. The extent of head loss testing, however, is dependent on the project
budget. Some projects are planned such that a sample of mains that are representative
of the system are selected for testing. Then the results are extrapolated to the rest of
the system.
One way to limit the amount of head loss testing that must be done to get valid data is
to perform head loss tests for a wide variety of pipe sizes, types, and ages. These data
can then be placed in chart form, showing pipe roughness values as a function of pipe
and size (Ormsbee and Lingireddy, 1997). Roughness values are then selected based
on the age and size of the selected main.
Systems in which pipe roughness varies over a wide range usually contain a signifi-
cant amount of unlined cast iron pipe. Sharp and Walski (1988) showed that equiva-
lent sand grain roughness heights in unlined, commercial, cast iron pipe increased
linearly with time. Therefore, in terms of absolute roughness:
H = H o + at (5.10)
where H = roughness height at age t (in., mm)

H o = roughness height when pipe was new (t=0) (in., mm)
a = rate of change in roughness height (in./year, mm/year)
t = age of pipe (years)
Roughness height for new cast iron pipe is usually on the order of 0.008 in. (0.18 mm).
Section 5.4 Pump Performance Tests 197
By measuring the roughness height for a few pipes in a head loss test, it is possible to
determine the coefficient in Equation 5.10 (the rate of change of roughness, a) and use
the value for other pipes of that type, provided that the corrosive characteristics of the
water have not changed significantly. Walski, Edwards, and Hearne (1989) developed
a method for adjusting values when water quality had changed during the life of a
pipe.
For those using Hazen-Williams C-factor instead of equivalent sand grain roughness
height, the relationship between C and age is related to the base 10 log of the rough-
ness height and diameter.
H o + at
C = 18.0 – 37.2 log §© ----------------·¹ (5.11)
D
where D = diameter (in., mm)
Using data from Lamont (1981) and Hudson (1966), Sharp and Walski (1988) per-
formed a regression analysis using Equation 5.10, relating the corrosivity of the water
using the Langelier Index, shown in Table 5.2. It should be noted that values for any
water system are specific to that system.
Table 5.2 Correlation between Langelier Index and the roughness growth rate
a a Langelier
Description
(in./year) (mm/year) Index
Slight attack 0.00098 0.025 0.0
Moderate attack 0.003 0.076 -1.3
Appreciable attack 0.0098 0.25 -2.6
Severe attack 0.030 0.76 -3.9
5.4 PUMP PERFORMANCE TESTS

There are four types of pump characteristic curves: head, brake horsepower, effi-
ciency, and NPSH (see page 49). Although modelers can usually rely on pump char-
acteristic curves that are provided by the manufacturer, it is good practice to check
these curves against pump performance data collected in the field. The following sec-
tion discusses how to determine points for the head characteristic curve. It is followed
by a discussion of measuring efficiency, which is needed for pump energy analysis.
Head Characteristic Curve

As presented in Chapter 2 (see page 44), the head characteristic curve gives total
dynamic head as a function of discharge through the pump. Consider the pump shown
in Figure 5.12. If the energy equation is applied between the discharge (section 2) and
suction (section 1) sides of the pump, the following expression is obtained.
2 2
V dis V suc
h dis + ---------
- = h suc + h P – h L – h m + ---------- (5.12)
2g 2g
where hdis = pump discharge head (ft, m)

Vdis = velocity at point where discharge head is measured (ft/s, m/s)
g = gravitational acceleration constant (32.2 ft/sec2, 9.81 m/sec2)
hsuc = pump suction head (m, ft)
hP = head added at pump (m, ft)
hL = head loss due to friction (m, ft)
hm = minor head losses due to fittings and appurtenances (m, ft)
Vsuc = velocity at point where suction head is measured (ft/s, m/s)
Figure 5.12
Pump performance
test
Clearwell 1 Pump 2 Valve

Q
Suction Discharge
Because sections 1 and 2 are close together, any head losses due to friction, hL, will
usually be negligible. In addition, the minor losses that occur within the pump as a
result of changing streamlines are not directly considered through the hm term.
Accordingly, the head loss terms are usually set to zero, and the minor losses within
the pump are addressed through the pump head term, hP.
Assuming that sections 1 and 2 have the same elevation, Equation 5.12 can be rewrit-
ten as shown:
2 2
P dis P suc· § V dis V suc·
h P = § ---------
© J - – ----------
+ ¨ ---------- – ----------¸ + h L + h m (5.13)
J ¹ © 2g 2g ¹
where Pdis = discharge pressure (psi, kPa)

Psuc = suction pressure (psi, kPa)
A pump head characteristic curve is a plot of hP versus flow. As shown in Equation
5.13, suction and discharge pressures and the suction and discharge velocity heads are
needed to develop the curve. The velocity heads can be calculated based on the flow
through the pump (most pump stations are equipped with flow meters) and the suction
and discharge pipe diameters. Because the suction and discharge pipe diameters are
usually not significantly different for water distribution pumps, the difference
between velocity head terms is often negligible.
If the pump is equipped with pressure gages on the suction and discharge lines, the
pressure information can also be easily collected. In some instances, however, a pump
Section 5.4 Pump Performance Tests 199
will have a pressure gage on the discharge side only. In this case, the suction head can
be found by applying the energy equation to the suction side of the pump, making
sure that all head losses between a hydraulic boundary condition and the pump are
accounted for. If the pump does not have a discharge pressure gage, then the energy
equation can be applied between the pump discharge and a point of known head (a
boundary condition). Again, all head losses between the two points must be consid-
ered.
The pump head characteristic curve is developed by finding the pump heads for a
series of corresponding pump flows. To do so, the operator varies pump flows through
the use of a valve on the discharge side of the pump. With the discharge valve wide
open, the pump is turned on and allowed to arrive at full speed. Next, the suction pres-
sure, discharge pressure, and pump flow are measured. The result of substituting these
measured values into Equation 5.13 is a point on the pump curve. Then the valve is
adjusted slightly, and another set of pressure and flow data is collected. This process
is repeated, closing the valve a little more each time, until the desired number of data
points have been obtained. The key to developing a useful curve is to vary the dis-
charge over the entire range, from shutoff head to maximum flow. In some cases, it
may be necessary to operate hydrants or blow-offs to get sufficiently high flows.
Pump Efficiency Testing

Typically, only the head characteristic curve is needed for modeling; however, some
models determine energy usage at pump stations as well as flow and head. To deter-
mine energy usage, the model must convert the water power produced by the pump
into electric power used by the pump. This conversion is done using the efficiency
relationships summarized below.
ep = (water powerout ) / (pump powerin ) (5.14)
em = (pump powerin ) / (electric powerin ) (5.15)
where ep = pump efficiency (%)

em = motor efficiency (%)
Pump power refers to the brake horsepower on the pump shaft, and it is difficult to
measure in the field. Therefore, all that can be calculated is the overall (wire-to-
water) efficiency.
ew-w = e p u e m = (water powerout ) / (electric powerin ) (5.16)
where ew-w = wire-to-water efficiency (%)

Although efficiency is expressed as a decimal in the above equations and in most cal-
culations, it is generally discussed in terms of percentages. Water power is computed
from the following relationship:
WP = C f Qh P J (5.17)
where WP = water power (hp, Watts)

Q = flow rate (gpm, l/s)

hP = head added at pump (ft, m)
J = specific weight of water (lb/ft3, N/m3)
Cf = unit conversion factor (4.058 u 10-6 English, 0.001 SI)
The measurement of electric power depends on the instrumentation available at the
pump station. Large stations may have a direct readout of kilowatts, thus the wire-to-
water efficiency can be easily computed by converting the water power and electric
power to the same units. In other cases, it may be possible to measure the current
drawn in amps. Knowing the voltage, power factor, and number of phases, the electric
power drawn can be determined as
EP = VI N PF (5.18)
where EP = electrical power (watts)

V = voltage (volts)
I = current averaged over all legs (amps)
N = number of phases
PF = power factor
Except for the motors driving the smallest pumps, pump motors are generally three-
phase. The power factor is a function of the motor size and the load for three-phase
motors. Additional information can be found in WEF (1997).
At some pump stations, there may be no instrumentation available for measuring elec-
tric power, and it may be difficult for electricians to directly determine amperage. In
these situations, it is necessary to measure the energy usage at the building power
meter and divide the energy use by time to determine power. If the meter is being read
directly, be sure to account for other sources of power consumption.
Similar to the head characteristic curve, the efficiency curve can be developed by set-
ting a flow rate, measuring the necessary parameters, and then adjusting the flow until
sufficient points to form a curve are determined.
Potential Problems with Pump Performance Tests

A key piece of information needed for the model representation of the pump is the
shutoff head (the head at zero flow). To find this point, the discharge valve is closed
and measurements are taken while the pump is operating. It is important to note that if
the pump operates with the valve closed for an extended period of time, the water in
the pump may begin to heat, potentially damaging the pump and seals. Thus, the mea-
surements must be taken quickly.
Another potential area of concern is electricity billing rates. Some water utilities
include an electricity demand charge in their billing structure that is typically based
on the highest 15- or 30-minute peak power usage period for the pump station. This
demand charge, which can be quite high (US$14/kW for example), is applied to all of
the current billing period, and may be applied to subsequent billing periods for up to a
year. It is important to note that pump testing may require large amounts of energy,
Section 5.5 Extended-Period Simulation Data 201
and care should be taken that a new and expensive demand charge is not set for the
utility.
Using Pump Performance Test Data for Calibration

The data obtained from a pump performance test are used to generate the pump head
versus discharge and efficiency curves, which are used to mathematically model the
performance of the pumps. The pump test data collected are input into the model,
which then uses curve-fitting techniques to create the relationships describing the
pump efficiency and head curves.
5.5 EXTENDED-PERIOD SIMULATION DATA

Most of the testing described in Sections 5.1 to 5.4 results in static measurements of
the distribution system—that is, measurements taken at a single point in time under a
single set of conditions. This information is useful for estimating various parameters
used in steady-state and EPS models. When an extended-period simulation model is
developed, it is necessary to supplement the static field testing with field measure-
ments taken over a period of several days. This information can be used for calibrat-
ing an EPS model (see Chapter 7) and validating that an existing EPS model
adequately represents the behavior of the distribution system over time.
Two types of data that are useful for calibrating and validating an extended-period
simulation model are
• Time-varying measurements of flow, pressure, and tank water levels in the
distribution system
• Concentrations of a conservative tracer over time throughout the system
The following sections discuss these two types of data.
Distribution System Time-Series Data

Flows, pressures, tank water levels, and other characteristics vary throughout the dis-
tribution system both temporally and spatially. Seasonal variations, variations by day
of the week, diurnal variations, and small time scale stochastic variations typically
occur. If an extended-period model of the distribution system has been properly con-
structed and calibrated, the model results should approximately mimic the behavior of
the system over a period of time. Such temporally and spatially varying data are fre-
quently collected for use in calibrating an EPS model (see Chapter 7).
Frequently, time-series data is available automatically through remote telemetry that
is part of a SCADA (Supervisory Control and Data Acquisition) system (see Chapter
6). This information can usually be easily downloaded or converted to a format for
use in the calibration process. Though information may be transmitted at very fre-
quent intervals, for most calibration purposes, measurements every 15 to 60 minutes
are generally adequate. SCADA data can be supplemented by that from flow meters
or pressure gages and data loggers installed for short-term data collection. Section 5.1
describes different types of flow meters.
Conducting a Tracer Test

In a tracer test, a conservative substance is added to the water in a distribution system
over a period of time, and the movement of the tracer through the system is deter-
mined by measuring its concentration over time at stations located at key points
within the system (Grayman, 2001). The resulting data may be used in conjunction
with an EPS hydraulic model and a water quality model in the calibration process (see
Chapter 7 for details on the use of these data for calibration).
The steps in conducting a tracer test are as follows:
1. A conservative tracer is identified for a distribution system. The tracer can be a
chemical that is added to the flow at an appropriate location taking into account
the study objective and location specific details or, for the situation where there
are multiple sources of water, a naturally occurring difference in the water
sources, such as hardness. Chemicals that are typically used include fluoride, cal-
cium chloride, sodium chloride, and lithium chloride. Selection of the tracer gen-
erally depends on government regulations (for example, some localities will not
allow the use of fluoride), the availability and cost of the chemicals, the methods
for adding the chemical to the system, and the measuring and analysis devices.
For example, a tracer chemical may be selected because it is inexpensive and can
Section 5.6 Water Quality Sampling 203
be added using a water plant’s existing dry chemical feed system. The amount of
tracer that is added depends on the measurement methods (that is, it must be great
enough so that differences in concentration can be measured) and on regulations
(resulting concentrations should not exceed allowable levels).
2. Before beginning the tests, it is recommended to simulate the test with the model
to determine the likely results. Determining the results ahead of time assists in
identifying the best sampling locations and times, and identifies the most likely
concentrations to make sure that they are in the range of the measuring equip-
ment.
3. A controlled field experiment is performed in which either: (1) the conservative
tracer is injected into the system for a prescribed period of time; (2) a conserva-
tive substance that is normally added, such as fluoride, is shut off for a prescribed
period; or (3) a naturally occurring substance that differs between sources is
traced.
4. During the field experiment, the concentration of the tracer is measured at
selected locations in the distribution system. It is desirable to have quick feed-
back on the movement of the tracer through the system so that adjustments can be
made in the sampling schedule. For example, if the tracer takes more time than
expected to reach a station, sampling needs to be extended beyond the originally
planned period. With some tracers, such as fluoride, more accurate measurements
can be made in the laboratory. To satisfy the need for quick feedback and accu-
rate measurements, a quick method, such as use of a Hach handheld digital meter,
can be employed for immediate feedback in conjunction with samples taken in
bottles for later analysis in the lab. Measurements should continue until the tracer
has reached the areas of the distribution system with the oldest water. In a system
that contains a tank or multiple tanks, that may require a tracer test lasting many
days (Clark, Grayman, Goodrich, Deininger, and Hess, 1991).
5. During the tracer test, other parameters that are required by a hydraulic model,
such as tank water levels, pump operations, flows, and so on, should be collected
at frequent intervals as well. This information is needed as part of the model cali-
bration/validation process.
6. Frequently, a tracer test is conducted in conjunction with a water quality study
(see Section 5.6) so that other constituents, such as chlorine residual, may be
measured at the same time that tracer measurements are made.
5.6 WATER QUALITY SAMPLING

When extending a calibrated hydraulic model to include water quality, various physi-
cal and chemical parameters must be determined. Some tests require bench scale
analyses that can easily be conducted in a modestly outfitted water quality laboratory.
Other measurements can be made directly in the field. The sections that follow
describe the types of tests that are performed in order to support the development of a
water quality model.
A calibrated, extended-period simulation hydraulic model provides a starting point
for water quality modeling. Steady-state hydraulic analysis is not adequate because it
does represent operational characteristics that vary temporally and does not account
for the effect of storage and mixing in tanks and reservoirs, a factor known to contrib-
ute to the degradation of water quality. As described in Chapter 2, transport, mixing,
and chemical reactions depend on the pipe flows, transport pathways, and residence
times of water in the network (all are network characteristics determined by the
hydraulic simulation). Therefore, a calibrated extended-period hydraulic model is a
prerequisite for any water quality modeling project. After a hydraulic model is pre-
pared, some types of water quality modeling analyses (especially water age and
source tracing) can be conducted with little additional effort whereas modeling reac-
tive constituents requires additional information on reaction rate coefficients.
Disinfectant residuals (chlorine, chloramines) decay due to reactions in the bulk water
and reactions that take place at the pipe wall. Disinfection by-products (DBP) grow
over time in the distribution system, and so a formation reaction rate is required by a
model. Bulk reaction coefficients are required for all nonconservative substances, and
wall reaction coefficients are required for disinfectant modeling. Boundary conditions
and initial conditions are needed for all substances. Determining bulk and wall reac-
tion coefficients involves laboratory analysis and field studies, as discussed in the fol-
lowing section. The determination of boundary and initial conditions is simpler and is
addressed in Section 7.5.
Laboratory Testing
For constituent analysis, reaction dynamics can be specified using bulk and wall reac-
tion coefficients. Bulk reaction coefficients can be associated with individual pipes
and storage tanks or applied globally. Wall reaction coefficients can be associated
with individual pipes, applied globally, or assigned to groups of pipes with similar
characteristics. Unlike bulk reaction coefficients, which can be determined through
laboratory testing, wall reaction coefficients must be measured using field tests or
determined as part of the calibration process, as discussed later in this section and in
Section 7.5.
Bulk Reaction Coefficients. Recall that the parameter used to express the rate
of the reaction occurring within the bulk fluid is called the bulk reaction coefficient.
Bulk reaction coefficients can be determined using a simple experimental procedure
called a bottle test. A bottle test allows the bulk reactions to be separated from other
processes that affect water quality, and thus the bulk reaction can be evaluated solely
as a function of time. Conceptually, the volume of water in a bottle can be thought of
as a water parcel being transported down a pipe (see page 52). A bottle test allows for
the evaluation of the impact of transport time on water quality and for an experimental
determination of the parameters necessary to model this process accurately.
Determining the length of the bottle test and the frequency of sampling is the first and
most critical decision. The duration and frequency of sampling will influence the
error associated with the experimental determination of the rate coefficient. The dura-
tion of the experiment should reflect the transport times occurring in the network. If,
for example, a water age analysis using the calibrated hydraulic model indicates that
residence times range from 5 to 7 days, conducting a 7-day test would provide bulk
reaction data over the entire range.
Bottle Test Procedure

1. Preparation 3. Sample Testing
-Plan the length of the experiment. -Store samples in complete darkness with the
-Collect materials needed for the experiment. temperature held constant (a water bath or
BOD incubator may be used).
-Wash bottles and prepare them using the
chlorine demand-free procedure. -Pull samples at designated times and mea-
sure using experimental procedure.
-Prepare reagents for experimental methods
and work area. -Record time and result of the experimental
procedure.
-Prepare laboratory notebook for recording
experimental conditions and results. 4. Processing Data
2. Sample Collection -Plot data.
-Collect water from the clearwell as it enters -Process data to determine rate coefficient.
the distribution system. (Summers, Hooper, Shukairy, Solarik, and Owen,
-Fill and cap bottles headspace-free. 1996; Rossman, Clark, and Grayman, 1994; and
Vasconcelos, Rossman, Grayman, Boulos, and
-Start the master clock.
Clark, 1996)
The frequency of the sampling should be proportional to the rate of the reaction. Typ-
ically, the sampling frequency should be more rapid at the start of the experiment
(every 30 minutes for a fast reaction and once every two hours for a slow one) and can
gradually decrease to a lower level (once or twice a day). After a schedule of samples
is determined, bottles, reagents, and other experimental equipment can be gathered.
Bottle tests can be used to determine bulk reaction rates for different types of reac-
tions (for example, disinfectant decay or DBP formation). The size of the bottles
depends on the volume of water required by the experimental procedure. Methods for
determining disinfectant concentrations can require anywhere from 20 to 100 ml.
Methods for determining DBP formation typically require a smaller sample volume.
In either case, the volume and number of bottles should also include any duplicates
taken.
It is important for the modeler to appreciate the precision (or lack thereof) when mea-
suring disinfectant residual concentrations. Each analytical method has its own mini-
mum and maximum detection limits, and each person performing a method may have
a bias or error associated with them as well. For example, attempting to measure con-
centrations of 0.08 mg/l when the analytical method is only accurate to 0.20 mg/l can
produce misleading data. Duplicates and replicates can be used to quantify these
types of errors.
Bottles should be washed prior to the experiment in accordance with the experimental
procedure. Frequently, bottles are prepared improperly and the experiment yields
worthless data. For example, if disinfectant decay is being measured, the bottle should
be prepared so that it does not contribute to the decay reaction. This can be accom-
plished by soaking the bottles for 24 hours in a strong solution of the disinfectant
(10 mg/l) and then rinsing with laboratory-clean water (Summers, Hooper, Shukairy,
Solarik, and Owen, 1996). Reagents should be gathered and prepared in accordance
with the experimental procedure specific to the constituent being measured. Once the
experiment has been planned and the laboratory prepared, the test can begin.
For the purposes of determining rates of reaction in the distribution system, water is
typically collected as it leaves the clearwell and enters the network, though this need
not always be the case. The water should be gathered and the bottles quickly filled
and capped with no airspace in the bottle. The experiment starts when the last bottle is
capped. At scheduled times, samples should be pulled and tested using
the constituent-specific experimental procedure. Between sampling times, samples
should be stored in complete darkness and at a constant temperature because reaction
rates (and thus reaction coefficients) are temperature dependent, and some reactions
are influenced by ambient light.
Example — Bottle Test Data Analysis. When all measurements have been taken and
the experiment is over, the data will describe the constituent concentration for each of the samples as
a function of time. The data can then be graphed. The constituent concentrations are charted along the
y-axis (the dependent variable), and the time is charted along the x-axis (the independent variable).
Figure 5.13 and Table 5.3 show an example of data collected from a bottle test for which the constitu-
ent was chlorine.
Figure 5.13
A best-fit straight line 10
drawn through the
charted results where
the slope of the line is
the bulk reaction
coefficient
Concentration, mg/L
0.1
0 12 24 36 48 60 72 84 96
Time, hr
Table 5.3 Bottle test results

Observed Observed
Time Time
Concentration Concentration
(hours) (hours)
(mg/l) (mg/l)
0 2.2 54 0.9
6 2.1 60 0.9
12 2.0 66 0.8
18 1.7 72 0.7
24 1.4 78 0.6
30 1.3 84 0.5
36 1.2 90 0.5
42 1.0 96 0.5
48 1.0
If the substance in the bulk fluid exhibits a first-order reaction, then the reaction rate
coefficient can be found using linear regression techniques. A best-fit straight line is
drawn through the data collected from the bottle test, with concentration plotted on a
log axis as illustrated in Figure 5.13. The slope of the line for the data in Table 5.3,
0.0165 hr-1, becomes the bulk reaction coefficient. Note that the reaction coefficient is
negative since the constituent concentration decays over time.
The straight line shown in Figure 5.13 produces a very nice fit with the observed data.
A more likely scenario is that the data will not fit as well as that shown. In fact, there
may be a few data points that are widely scattered. Outliers, as these points are called,
should be carefully examined and possibly discarded if they are found to negatively
influence the results of the data analysis. If there is a large amount of scatter, another
bottle test may be performed in an attempt to collect more reliable data.
Bottle tests can be performed on any water sample, regardless of where the sample is
collected. Samples from treatment plants and other entry points to the distribution
system are particularly important because these facilities act as water sources and
therefore, have a strong influence on water quality. Raw water quality influences fin-
ished water quality as well. Therefore, if a system has multiple treatment facilities
each with a different raw water source, the bulk reaction rates for each of the finished
waters are likely to differ. Although storage tanks are not sources of finished water,
under some circumstances, the bulk decay rate can be different in tanks than in the
distribution system. As a result, a separate bulk reaction coefficient should be consid-
ered for tanks and reservoirs. Booster disinfection (when disinfectant is reapplied to
previously disinfected water) is another circumstance in which bulk reaction
coefficients are likely to change.
Bulk reaction coefficients are associated with pipes for purposes of a simulation, and
are assumed to remain constant throughout the simulation for a particular pipe. Since
the bulk reaction coefficient is, in reality, associated with the fluid itself, the bulk
reaction rate can change throughout the actual system as water from different sources
becomes mixed at nodes. When assigning bulk reaction coefficients for pipes, the
mixing can be considered by designing and conducting bottle tests with representative
source mixtures. A source tracing analysis can assist in determining the degree of
mixing in a system. Source blending can change over the course of the day for a par-
ticular pipe, thus the predominant source or mix of sources should be used in assign-
ing the bulk reaction coefficient.
For example, suppose that a tracing analysis is performed on a system that has two
treatment plants. Through the bottle tests, a bulk reaction coefficient is established for
each treatment facility and each storage tank. Through a source tracing analysis of a
junction node, it is found that 90 percent of the water for that node comes from a spe-
cific treatment plant. Accordingly, the bulk reaction rate coefficient for those pipes
that make up the path between the plant and the node would be equal to, or nearly
equal to, the rate coefficient for the plant. For a single-source network, it is useful to
remember that the bulk reaction coefficient is really a function of the water passing
through a pipe or stored in a tank, and not a function of the pipe or tank itself. Thus,
specifying a global bulk reaction coefficient is frequently the simplest and the best
method for modeling bulk reactions occurring in such networks.
Field Studies
Several different types of field studies may be performed to collect data used in cali-
brating a water quality model. The sections that follow describe three types of studies.
The first type of study is aimed at determining actual pipe diameters — an important
factor in water quality modeling. The second type of field study can be performed to
determine pipe wall reaction coefficients.
Determining Actual Pipe Diameters. In the United States, friction head
loss is usually predicted in network models by using the Hazen-Williams head loss
equation:
1.852
Cf Q L
h L = ---------------------------------
- (5.19)
1.852 4.87
C D
where hL = head loss due to friction (ft, m)

Cf = unit coversion factor (4.73 English, 10.7 SI)
Q = flow (cfs, m3/s)
L = length (ft, m)
Most water distribution system models use pipe diameters based on the originally
stated nominal diameters (see page 255). Even for new pipes these values are only
approximations, and for older pipes these values can seriously overstate the effective
diameter due to tuberculation. For example, a new Class 50 ductile iron pipe with a
nominal diameter of 6 in. has an actual internal diameter of 6.4 in. Tuberculation is
most common in older metallic pipes and can result in significant reductions in the
effective pipe diameter.
For most hydraulic applications, the use of the nominal diameter is acceptable. This is
especially the case if C-factor tests were performed, and the nominal diameter was
used in the process of estimating C-factors. In effect, errors in both the true diameter
and the C-factor offset each other and, as a result, head loss and flows are calculated
to a generally accepted level of accuracy. This can be illustrated by examining the
denominator in Equation 5.19 and observing that any combination of C-factor and
diameter that result in the same value of the denominator will result in the same value
for head loss. For example, a C-factor of 83 and a pipe diameter of 12 in.(305 mm)
results in the same head loss as a C-factor of 134 and a pipe diameter of 10 in. (254 mm).
However, use of an incorrect pipe diameter can result in significant errors in predict-
ing velocity. Because velocity is a major factor in water quality modeling both in
terms of its effect on travel times and on calculation of chlorine wall demand, a pre-
mium exists on correctly estimating velocity. Therefore, one should ensure that the
diameters used in a hydraulic model when used as part of water quality modeling
more closely reflect actual values. Several methods are available that can be used to
estimate the true diameter of a pipe:
• In-situ direct measurement of diameters: A specially designed set of cali-
pers can be used to directly measure the diameter of a pipe at a particular
location. Figure 5.14 depicts a particular design for these calipers. The cali-
pers are inserted into a pipe at a corporation stop and adjusted so that they
directly measure the inside diameter of the pipe. It should be noted that this
method can provide relatively accurate point measurements. However, if the
diameter varies significantly along the pipe due to irregular tuberculation,
the point measurements may or may not be representative of the true diame-
ter for a pipe segment.
• Measurement of flow and velocity and calculation of diameters: Flow
can be calculated as the product of velocity and cross-sectional area. There-
fore, if the flow and velocity in a pipe are known, the actual pipe diameter
can be calculated. Various methods are available for measuring flow and
velocity. In order to use these measurements to calculate true diameter, how-
ever, the measurements must be independent. Thus, if a flow meter is mea-
suring velocity and using an assumed diameter to convert to flow, this
measurement can be used only as a single measurement, and a second inde-
pendent measurement is needed. A Pitot gage is frequently used as part of a
hydrant test to determine flow. Injection of a pulsed slug of tracer at an
upstream corporation stop and measurement at a downstream hydrant can
provide an accurate travel time and velocity calculation (Wright and Nevins,
2002).
• Development and use of an inventory of actual pipe diameters: It is good
practice to keep a database on pipes that have been taken out of service. The
database should include the pipe material, date of installation, pipe location,
nominal diameter, and notes on the condition of the pipe including the effec-
tive diameter of the pipe. This information can be used to define representa-
tive effective diameters for pipes of a particular age and material and entered
into the model as an alternative to the original, nominal diameter. Also, this
information can be used as a basis for analyzing pipe breaks in a system.
Figure 5.14
Calipers for
measuring pipe
diameter
Measuring Chlorine Wall Demand. Wall demand can be indirectly mea-

sured in the field in a manner that is analogous to C-factor tests (Grayman, Rossman,
Li, and Guastella, 2002). A homogeneous pipe segment (constant diameter, material,
and age) with a length of at least 1,000 ft (305 m) is selected for analysis. The pipe
segment is isolated by valving off major laterals and at the downstream end, as shown
in Figure 5.15.
Figure 5.15
Setup for performing
field chlorine wall
demand tests
The downstream hydrant is then flowed at a constant rate and chlorine measurements
are taken at an upstream hydrant (C1) and at the flowed hydrant (C2). The chlorine
measurement at the downstream hydrant should be taken TT minutes after the mea-
surement at the upstream hydrant where TT is the travel time between the two
hydrants. The process can be repeated at different flow rates. After accounting for the
bulk decay of chlorine (usually negligible in a short pipe segment), the wall demand
coefficient can be calculated based on the field data in a spreadsheet or by iteratively
running a water quality model of the pipe segment until the model results match the
field data. This method is most appropriate for pipes that are expected to have rela-
tively higher wall demands such as smaller diameter, unlined cast iron pipes. Larger
pipes and nonferrous pipe materials, such as PVC or cement, generally have relatively
low wall demands. It is likely that little or no chlorine loss would be measured in short
segments.
Intensive Water Quality Surveys. Intensive hydraulic and water quality

surveys that extend over a multiday period can be conducted to calibrate or validate an
extended-period hydraulic and/or water quality model. In such a survey, information
on how the system was operated is compiled along with time-series data collected in
the field or through remote telemetry. Surveys of this type can be quite expensive and
involve significant personnel, equipment, and laboratory analyses. Therefore, they
should be carefully planned and executed. A properly conducted survey can yield a
wealth of information that is invaluable in understanding the movement of water
through the study area and for calibrating and validating a model. Clark and Grayman
(1998) provide a detailed description of how to prepare for and conduct an intensive
water quality field survey. The information in this section draws heavily upon the
description provided in their book, Modeling Water Quality in Drinking Water Distri-
bution Systems.
Within the modeling context, the purpose of an intensive field survey is to collect suf-
ficient information in order to adjust model parameters or validate model parameters
by mimicking the results observed in the field. Operational information that is not col-
lected during the study becomes unknowns, introducing uncertainty in the modeling
process. Ideally, a detailed model of the study area is available prior to planning a
field study. If that is the case (and even if it is not a well-calibrated model), the model
can be used to predict how the system will behave and thus to determine what data to
collect, and when and where, in order to make the best use of the study results to cali-
brate or validate the model.
In the ideal case, it is best to predefine how the distribution system should be operated
during the field study in order to design an experiment that best serves the purpose of
the work. For example, one objective may be to operate the system in a manner that is
typical for that season. Another objective could be to operate the system in a stream-
lined manner (for example, avoid turning pumps on for a few minutes at a time) in
order to simplify the modeling process.
Intensive surveys conducted as part of a water quality modeling effort generally focus
on collection of data related to the hydraulic nature of the system, field data on disin-
fection residuals, and other water quality constituents such as temperature, pH, and
disinfection by-products (DBP). In some studies, a tracer is used to determine the travel
times and the patterns of movement of water through the distribution system. The
mixing patterns within storage facilities may also be studied during an intensive survey.
A field study is divided into three phases: designing the field study, executing the field
study, and analyzing and using the resulting set of data. Arguably, the most important
phase is the preplanning stage. Prior to performing a sampling study, a detailed writ-
ten sampling plan should be prepared. Clark and Grayman (1998) provide a list of
issues that should be addressed in a sampling plan and specific actions that should be
taken prior to a sampling study:
• Sampling location: Selection of sampling sites is typically a compromise

between selecting sites that provide the greatest amount of information and
sites that are most amenable to sampling. Sites should be spread throughout
the study area and should reflect a variety of situations of interest, such as
transmission mains and local lines, areas served directly from a source, and
areas under the influence of tanks. Because sampling is usually performed
around the clock, the sites must be accessible at all times. Also, sampling
taps should be placed close to mains so that the samples reflect the water
quality in the mains.
If automatic samplers are used, the sites may need to be secure (automatic
sampling equipment is expensive), and electric power and a method for dis-
posing of waste flow may be required. For manual sampling, the travel time
for crews between sampling sites is a consideration. Dedicated sampling
taps; water utility and public buildings, such as firehouses; and hydrants are
frequently used as sampling sites. Pump stations and valve vaults are also
good locations because it is possible to directly sample the main. Care must
be taken when sampling through a hydrant or service line to minimize or
account for the lag between the sample time and the time the water left the
main.
• Sampling frequency: For automated samplers, such as chlorine monitors,

pH meters, and pressure gages, the sampling frequency can generally be set
and is usually performed every few minutes. In the case of manual sampling,
a circuit is usually defined for a sampling crew to follow. The crew takes
samples at one site, analyzes them, and then moves to the next site, and con-
tinues this circuit for their entire shift. These practical constraints and the
project budget result in trade-offs between the number of samplers, the num-
ber of sites, and the sampling frequency.
Sampling frequency is usually on the order of once per hour to once every
several hours. The rate at which characteristics change at a site is an impor-
tant factor in choosing a sampling frequency. Thus, if the chlorine residual is
expected to change graduallyover the course of a day, less frequent sampling is
needed than if the residual is expected to vary rapidly over time. Dress
rehearsals in which a crew drives a circuit to determine driving times and
tests the sampling process to define the amount of time that must be spent at
a site provides important information when designing the sampling program.
In following a front through the systems (when the fluoride feed is turned
off, for example), it is desirable to sample more frequently at the time that
the front is expected to arrive.
• System operation: The sampling plan should specify the general system
operation that is expected to occur during the sampling study and discuss the
methodology for capturing the information on system operation. For exam-
ple, the plan should indicate if any unusual operations are planned during the
study period.
• Tracer study: If a tracer study is to be performed, details on the tracer study

should be discussed. This includes the type of tracer to be used, regulatory
requirements to use the tracer, the quantity of tracer needed, where the tracer
will be purchased, how and at what rate the tracer will be injected, how the
tracer will be measured in the field, and equipment. A meeting with the reg-
ulatory agency may be required.
• Preparation of sampling sites: Various activities should be planned to pre-

pare sampling sites prior to the sampling study. These may include testing
and flushing hydrants, installing sampling appurtenances, determining
required flushing times, and notifying personnel located at sampling sites in
buildings. Automated monitoring devices should be installed and thoroughly
tested several days before the actual sampling commences.
• Sample collection procedures: The sampling plan should include specific

procedures to be followed during the sampling program. Topics to be dis-
cussed include the flow rate and length of time that the tap should be flowed
before each sample is taken, the filling and sealing of the sampling contain-
ers, preservation of samples and required reagents, labeling of sampling con-
tainers, data recording, and delivery of samples to laboratories.
• Analytical procedures: Specific analytical procedures should be deter-

mined and described for any analyses that are conducted in the field. This
includes chlorine measurements and measurements of other water quality
constituents. Procedures to be used in the laboratory should be specified. It is
important for the modeler to know the accuracy and minimum detection lim-
its for each test so that he or she can correctly assess the field results.
• Personnel organization and schedule: Intensive sampling surveys can

involve a large number of personnel working around the clock in shifts for
several days. Work schedules should be determined for each member of the
survey team. This discussion should include logistical arrangements
(where to meet, who will bring what equipment, emergency phone num-
bers, and so on).
• Safety issues: The safety of the survey team is of paramount importance

when planning a study. Potential for accidents is very high due to the around-
the-clock effort, possible bad weather, unusual surroundings, sampling loca-
tions in close proximity to traffic, and countless other factors. Crews should
be clearly identified and carry proper identification cards, should wear
reflective vests or other paraphernalia to make them more visible, use
marked vehicles (if possible), carry flashlights, and be advised how to
respond to various emergency or unusual situations. The police department
and other government agencies should be notified prior to the sampling
study, and public notification through newspapers and television should be
considered.
• Data recording: The information being collected is very valuable, and a

procedure for recording and safeguarding the information should be devel-
oped in the study plan. Data forms should be prepared that include the sam-
pling location, sampling time (consider using military time to avoid
ambiguity), sampler’s name, field measurements, samples taken for labora-
tory analysis, and any comments or observations. Forms should be filled in
ink and safeguarded until they are delivered to a central location.
• Equipment and supply needs: A complete list of equipment and supplies

should be developed along with a schedule and plan for obtaining the materi-
als well before the start of the sampling survey.
• Calibration and review of analytical instruments: All instruments should

be properly calibrated and thoroughly checked out prior to the study. The
study plan should specify the methodology and schedule for performing this
task.
• Training requirements: Before the start of the survey, all survey personnel
should observe the sampling sites and receive hands-on training in the use of
the equipment and protocols.
• Contingency plans: Over the course of the several-day survey, some aspect
of the survey will most likely not go according to plan. The purpose of con-
tingency planning is to be prepared for such events. The contingency plan
should address equipment malfunction, severe weather, illness, changes in
the operation of the system, and other potential events.
• Communications: During the sampling survey, it is important that sampling

crews, personnel at the operations center, and the overall supervisor for the
study be in communication in case of questions or changes. Using cell
phones and two-way radios is recommended.
The study plan serves as a blueprint for conducting the water quality survey. During
the actual survey, data and information are assembled and assessed at a central loca-
tion on a near real-time basis. If some aspect of the study is not proceeding as origi-
nally intended (the tracer is spreading at a quicker rate through the system then
expected, for example), modifications can be made in real-time. All aspects of the
Section 5.7 Sampling Distribution System Tanks and Reservoirs 215
survey should be documented during the survey and following the completion of the
field work.
Although intensive water quality sampling studies can be expensive, they are valuable
in developing and validating water quality parameters and ensuring that the hydraulic
calibration is adequate for water quality modeling tasks.
5.7 SAMPLING DISTRIBUTION SYSTEM TANKS AND

RESERVOIRS
The primary water quality issues in distribution system tanks and reservoirs are con-
tamination entering the facility, long residence times, and poor mixing conditions.
Monitoring can be an effective mechanism for identifying contaminants and studying
the mixing and water quality behavior in existing tanks or reservoirs. There are three
categories of sampling and monitoring studies:
• Water quality studies provide data on the temporal and spatial variation of
water quality parameters within the storage facility and in the inflow and
outflow.
• Tracer studies provide information on the mixing behavior in the tank.
• Temperature studies gather information on how the temperature may vary at
different locations and depths within the tank over time.
These studies can be performed as part of an integrated study to develop a better
understanding of how reservoirs and tanks behave. The sampling results can also be
used to help calibrate or validate a mathematical or scale model of the storage facility.
The design and implementation of monitoring programs for distribution system tanks
and reservoirs was examined in a recent AWWA Research Foundation sponsored
study (Grayman et al., 2000).
Water Quality Studies

Water quality monitoring studies of tanks and reservoirs can provide data on the tem-
poral and spatial variation of water quality parameters within the facility and in the
inflow and outflow. Information on the state of the reservoir (is it filling or draining,
for example) should be collected during a water quality study in order to interpret the
water quality monitoring data. Internal sampling of a tank or reservoir provides infor-
mation on the actual spatial variation of water quality parameters inside the facility.
Water quality data in combination with information on the inflow and outflow history
furthers the understanding of the water quality behavior of the tank.
Water quality studies in tanks and reservoirs can be conducted to meet many different
goals. Routine and regulatory sampling is performed at storage facilities throughout a
distribution system to satisfy regulatory requirements, define the general water quality
in the facilities, and identify potential problems. Water quality sampling studies can
also be designed specifically to identify the variations in water quality over time and
location within the facility, the water quality transformations that occur during stor-
age, or the mixing processes that occur during inflow and outflow.
In addition to sampling at the inlet and outlet, samples taken at different locations
within the storage facility and at different depths provide information on spatial vari-
ability. Grab samples can be supplemented with automated monitors that perform and
log analyses at a pre-set sampling frequency. Chlorine residual and temperature are
typically measured. Bulk chlorine decay tests are usually performed in order to under-
stand the kinetics of chlorine decay within the facility. The effects of wall demand in a
tank are usually minimal because of the large ratio of volume to wall surface. Other
water quality constituents may be sampled to meet regulatory requirements or in
response to specific water quality concerns.
Internal samples are frequently taken through hatches located on the top of a tank.
When sampling elevated tanks or standpipes, tank climbing and sampling issues
become more substantial, and safety concerns become more of an issue.
Tracer Studies
Tracer studies in tanks serve a similar purpose as tracer studies in a distribution sys-
tem: to define the movement of water through the vessel. In a tank tracer study, a
chemical tracer is added to the inflow and its movement is monitored through analysis
of the tracer within the facility or in the outflow. Water quality sampling studies are
usually performed in conjunction with tracer studies.
Tracer studies of distribution system tanks and reservoirs can provide information
regarding the detention time of water in the storage facility, as well as the mixing con-
ditions as it fills and drains. The objective is to determine how influent water entering
the reservoir mixes and subsequently leaves the facility. In addition, tracer study data
can be used to develop, calibrate, or validate computational fluid dynamics (CFD)
models (see page 358) or physical scale models. When these models exist for a given
reservoir, modifications in design or operation can be tested before implementing a
costly change at full scale.
Grayman et al. (2000) describe the procedures involved in planning a tracer study.
They include selection and injection of the tracer chemical, logistical considerations
in performing the study, and the collection of various ancillary hydraulic and water
quality data. The following sections discuss these topics.
Tracer Chemicals. The most frequently used tracer chemicals include fluoride
and salt solutions (calcium chloride, sodium chloride, lithium chloride, and potassium
chloride). In the United States, acceptable chemicals for a potable water supply are
usually limited to National Sanitation Foundation (NSF) approved chemicals or food
grade chemicals that have been approved by the state regulatory agency. It is impor-
tant to make sure that the tracer does not affect the density of the inflow water because
this occurrence can give misleading results.
Tracer Injection. The step dose method of injection is generally used in distribu-
tion system storage facilities in which the tracer is fed into the influent over one or
more fill periods at a relatively constant concentration. The injection point should be
located far enough from the reservoir inlet so that the tracer has fully mixed with the
influent prior to its entry into the reservoir but close enough so that the tracer does not
Section 5.7 Sampling Distribution System Tanks and Reservoirs 217
enter the distribution system directly. A sample tap downstream of the injection point
before the water enters the reservoir is desirable so that the actual tracer concentra-
tions entering the storage facility can be monitored.
Tracer Dosage. The tracer dosage should be calculated so that variations in con-
centration can be clearly measured in the tank. It should not result in concentrations
exceeding regulatory requirements.
Monitoring Locations and Frequency. Monitors or grab samples should be

taken at the inlet and outlet of the tank and, ideally, at locations within the facility. If
stratification is suspected, sampling at varying depths is important. Internal sampling
is generally limited by access points and the location of permanent sampling ports. If
grab samples are taken, a sampling frequency of approximately once per hour is gen-
erally sufficient with more frequent sampling at the inlet sample tap during the fill
period and less frequently during the draw period. With automated monitors, more
frequent sampling is recommended. Samples should be collected over several fill and
draw cycles or until the water exiting the reservoir approaches the background con-
centration of the tracer chemical.
Regulatory Approval. Policies of state agencies concerning the addition of

tracers varies significantly around the country. For example, some may not allow the
addition of fluoride while others may not allow normal fluoridation to be turned off. It
is good practice to obtain written approval from the state regulatory agency before
performing the tracer study.
Flow Measurements. Inflow and outflow rates are required to assess the
behavior of the reservoir and to interpret the tracer results. If the reservoir operates in
a fill and draw mode (as opposed to simultaneous inflow-outflow), inflow and outflow
rates can be estimated from water level measurements during the study. For simulta-
neous fill and draw, and in situations where more accurate flow rates are needed, flow
meters on the inlet and/or outlet can be used.
Temperature Monitoring
Temperature variations in a tank or reservoir can affect the mixing characteristics in
the facility and, in extreme cases, lead to stratification. Spatial and temporal variations
in temperature can result from changes in inflow temperatures, differential heating in
the tank, and insufficient mixing. Flow patterns in a tank or reservoir can sometimes
be affected by temperature differences of less than 1.0oC. Temperature can vary in
both the vertical and horizontal directions and may change over the course of a fill
and draw period, over a few days, or between seasons.
Temperature can be measured manually with a thermometer or probe, or automati-
cally by a thermistor and data logger. Measuring temperature manually is inexpensive
and easy to implement but labor intensive for longer sampling periods. Measuring
temperature automatically requires equipment that costs a few thousand dollars. For
either method, temperature measurements should be quite accurate (to the nearest
0.1oC, if possible) because small variations in temperature are generally observed.
With manual sampling, samples can be collected from a sampling tap or can be drawn
from different locations by using a pump or sampling apparatus. Collection of sam-
ples from different depths by using a pump or depth sampler requires access from
above the reservoir or tank through access hatches.
Long-term temperature measurements can be taken using an apparatus composed of a

series of thermistors and a data logger. The thermistors are positioned to the required
depth and attached to a data logger which can be set to take a reading at a preset fre-
quency (generally every 15 minutes to 60 minutes is adequate). Figure 5.16 is a sche-
matic representation of a set of thermistors and a data logger. In this case, thermistors
are attached to a chain and set at specific depths. Additional thermistors are attached
to floats to measure temperatures at fixed preset depths below the water surface as the
water level varies. Internal sampling equipment should be removed before the winter
in areas where ice can form in tanks.
5.8 QUALITY OF CALIBRATION DATA

Users will sometimes try to calibrate models where the velocity and head loss are
very low, and thus the hydraulic grade line is essentially flat. Under such conditions,
the heads in the system are essentially the same as those at the boundary conditions
and virtually any value of the roughness coefficient or demand can be used to produce
similar results (Walski, 2000). McBean, Al-Nassari, and Clarke (1983) used first-
order analysis to determine the accuracy of pressure measurements needed for field
data to be useful for model calibration.
Referring to the head loss equations presented in Chapter 2 (see page 34), the head
loss depends heavily on the flow and the C-factor. Most model calibration eventually
comes down to adjustments in a parameter like the C-factor, according to the equa-
tion:
k Q r 'Q -
C = ---------------------------------- (5.20)
0.54
h L r 'h L
where C = Hazen-Williams C-factor

k = factor depending on units and distribution system
Q = estimated flow (gpm, m3/s)
'Q = error in measuring Q (gpm, m3/s)
hL = estimated head loss due to friction (ft, m)
' hL = error in measuring head loss due to friction (ft, m)
If the flows and heads are small, errors in measuring these quantities will be on the
same order of magnitude as the quantities themselves, making them of little use in the
calibration process. If such data are used, the value of parameters found by calibration
will be poor.
Section 5.8 Quality of Calibration Data 219
Figure 5.16
Typcial thermistor:
data logger
configuration for
temperature
measurements
Based on Grayman et al., 2000.
The key to successful calibration is to increase the flows and the head losses such that
these values are significantly greater than errors in measurement. The best way to do
this is by conducting hydrant flow tests as described in Section 5.2. If the model can
match conditions under normal demands and fire flow tests, then the user can feel
confident that the model can be applied to other conditions.
In larger pipes [for example, greater than 16 in. (400 mm)], hydrant flow tests will not
generate significant velocities. For these larger pipes, the engineer needs to create
head loss either by measuring it over very long lengths of pipe, or by artificially
increasing flow by allowing tank water levels to drop significantly and then filling the
tank quickly. Figure 5.17 shows a hydrant flow test.
Figure 5.17
Hydrant flow test
Impact on Optimized Calibration. With the availability of powerful optimi-

zation software (see page 268) that can accurately and automatically calibrate a water
distribution model, it is more important than ever to have high quality field data. This
is because the optimization software is guided by the field measurements. With per-
fect field measurements of head loss, optimization programs can give excellent cali-
bration results for pipe roughness, water demand, and element statuses. However,
with error in the head loss measurements, the optimization programs will deliver mis-
leading calibration results because they do not distinguish between good and bad field
measurements.
The denominator in Equation 5.20 serves as the key to the basic rule for acceptance of
data for use in calibration as shown in Equation 5.21. As the error in head loss
approaches the actual head loss, the usefulness of an observation for calculating head
loss is diminished.
h L >>'h L (5.21)
This rule is applicable for evaluating data for either manual or automated calibration
but is especially critical for automated calibration. For manual calibration, HGL val-
ues are used as a check of results, and odd data can be discounted. However, with
automated calibration optimization, every HGL observation is treated as if it is
exactly true and can drive the solution to erroneous values, which the program would
claim are optimal. Therefore, only HGL observations that meet the criterion stated in
Equation 5.21 should be used in the calibration.
Data collected when head loss is small can be used to check if the roughness and
demand are wrong but cannot be used to determine what values are correct. For exam-
ple, if the measured pressure is 65 psi (448 kPa) and the model predicts 75 psi (517
kPa) during low demand, the user can be certain there is something wrong with the
model (most likely not roughness or demand). But when the model predicts 65 psi
(448 kPa) and the measured pressure is 65 psi (448 kPa), the user cannot conclude the
model is correct because even incorrect roughness or demand may yield that result.
Compounding the problem is the fact that although model users have an appreciation
for pressures or HGL values throughout the system, they usually do not think in terms
of head loss between two points. For example, a modeler may know that the pressure
in one part of the system is 50 psi (340 kPa) and that the HGL in that area is 960 ft
(293 m), but will have very little intuitive feel for how much head loss there is
between that point and the nearby tank.
Example — C-Factor Sensitivity. For a given distribution system, error in flow is usually
not larger than flow measurements so Equation 5.20 can be simplified and the head loss can be related
to C-factor by:
k
C = ----------------------------------
-
0.54
h L r 'h L
Obviously, as the error term becomes large with respect to the actual head loss, the confidence bound
on any calculated C-factor becomes large. For example, Figure 5.18 shows that for a system with an
actual C-factor of 100 and a 10 ft (or m) head loss between the boundary node and the measurement
Section 5.8 Quality of Calibration Data 221
point, the confidence bounds are wide, and it does not take a very small error in measuring head loss
to make a huge difference in C-factor. For example, if the head loss is 10 ft and the error in measure-
ment is 5 ft, one can only conclude that the C-factor is somewhere between 124 and 69. On the other
hand if the head loss is 40 ft and the error in measurement is 2 ft, then the C-factor is between 103 and
97.
Figure 5.18
Impact of error in
160 head measurement
0 ft
140 head loss = 1
s when
bound
Upper
120
head loss = 40 ft
Upper bounds when
100
Lower bounds when
hea d loss = 40 ft
C-factor
80
Low
er b
oun
60 ds
whe
n he
ad
lo
40 ss
=
10
ft
20
0
0 2 4 6 8 10
Error in Head Measurement, ft
Sources of Error in Calibration Data. The inequality provided in Equation

5.21 can be viewed as the basic law for screening data for use in automated calibra-
tion and can be satisfied by increasing the left side or decreasing the right side. Errors
in the right side can be due to inaccuracies in pressure readings, elevations, or bound-
ary conditions, as discussed in the following.
• Pressure readings: Pressure gages must be accurate and readable to +/- 1

psi (6.8 kPa) and preferably should have an accuracy better than that. Even
good quality gages can drift out of calibration by several psi (kPa), rendering
the entire calibration effort in doubt, so they must be calibrated frequently.
• Elevation: Elevation data are usually the largest source of error; however,
unlike pressure data, once an elevation is accurately established, it does not
change over time. Elevations used for normal model nodes can have a fair
amount of error and still be useful; however, for calibration, elevations
should be known to within 1 ft (0.3 m). This means that elevations of the
pressure gage (not the ground) should be determined by surveying, using a
high quality global positioning system, or using contour maps from digital
orthophotogrammetry with an accuracy on the order of 1 ft (0.3 m) (Walski,

1998). Other possible sources for elevation data are elevations surveyed from
sewer manhole lids if their elevations are considered accurate and readings
from a high quality altimeter that is regularly calibrated and sufficiently
accurate.
• SCADA data during flow tests: Some engineers trust data from SCADA
systems without question. However, SCADA data may be taken from inac-
curate sensors at inaccurate elevations. Concerning SCADA data, Akel
(2001) notes that even though they were digitally generated, they are not
precluded as sources of error. SCADA data may also pick up errors in
transmission and polling intervals. Most SCADA systems are not
continuously wired into a sensor; they poll the sensor periodically (interval
of minutes to seconds). Therefore, the pressures being displayed on the
SCADA may not correspond to the pressure in the system at that time. (See
Chapter 6 for more information on sources of error in SCADA data.)
• Chart recorders during flow tests: Chart recorders have similar problems
in capturing flow test data. Because the chart speed is slow, hydrant flow
tests usually show up as a vertical line. Unless the test was run for a long
time, it is impossible to get an accurate flow test pressure reading from a
chart recorder. Human operators viewing a gage at the hydrant, pump sta-
tion, or control valve of interest are the safest means of obtaining pressure
data during flow tests. If personnel are available, it is best to station an
individual at a key pressure control valve or pump station to read the gages
during the test. Data loggers with a fairly high speed of data capture may
also be used to capture pressure and flow readings during a flow test. It is
essential during flow tests to run hydrants long enough so that all transient
effects have dissipated, otherwise they may mask the actual values. It may
also take a while for a pressure-regulating valve to fully adjust itself during a
flow test.
• Tank water levels: Operators are usually more interested in the fluctuations
of tank water levels than the accuracy of the level. As a result, it is not
uncommon to find tank level readouts off by several feet. Tank water level
sensors need to be checked before calibration data are collected. In addition,
the water level and pump status must be taken at exactly the moment when
pressure readings are taken. Using the average level of the tank during the
afternoon when data were collected can lead to errors in calibration.
If the benefits of optimal calibration are to be realized, the modeler needs to carefully
plan the data collection effort and recognize instances where optimal calibration may
not be the best alternative. For example, in some situations, such as larger transmis-
sion mains, it may be better to run a C-factor test on the pipe and use that value, rather
than perform optimal calibration. While optimal calibration programs can greatly
simplify the adjustments needed for calibration, the software does not have the capa-
bility to judge and ignore/discount questionable data. It is the responsibility of the
user to ensure that the model is fed accurate and useful data.
References 223
REFERENCES
Akel, T. (2001). “Best Practices for Calibrating Water Distribution Hydraulic Models.” Proceedings of the
AWWA Annual Conference, American Water Works Association, Washington, D.C.
American Water Works Association (1989). “Installation, Field Testing, and Maintenance of Fire
Hydrants.” AWWA Manual M-17, Denver, Colorado.
Clark R. M., and Grayman, W. M. (1998). Modeling Water Quality in Drinking Water Distribution Systems.
AWWA. Denver, Colorado.
Clark, R. M., Grayman, W. M., Goodrich, R. A., Deininger, P. A., and Hess, A. F. (1991). “Field Testing
Distribution Water Quality Models.” Journal of the American Water Works Association, 84(7), 67.
Grayman, W. M. (2001). “Use of Tracer Studies and Water Quality Models to Calibrate a Network Hydrau-
lic Model.” Current Methods, 1(1), Haestad Methods, Inc. Waterbury, Connecticut.
(2000). “Water Quality Modeling of Distribution System Storage Facilities,” AWWA and AWWA
Research Foundation, Denver, Colorado.
Grayman, W. M., Rossman, L. A., Li, Y., and Guastella, D. (2002). “Measuring and Modeling Disinfectant
Wall Demand in Metallic Pipes.” Proceedings of the ASCE Environmental Water Resources Institute
Conference, American Society of Civil Engineers, Roanoke, Virginia.
Hudson, W. D. (1966). “Studies of Distribution System Capacity in Seven Cities.” Journal of the American
Insurance Service Office (ISO) (1963). Fire Flow Tests. New York, New York.
Lamont, P. A. (1981). “Common Pipe Flow Formulas Compared with the Theory of Roughness.” Journal
McBean, E. A., Al-Nassari, S., and Clarke, D. (1983). “Some Probabilistic Elements of Field Testing in
Water Distribution Systems.” Proceedings of the Institute of Civil Engineers, Part 2, 75-143.
McEnroe, B. M., Chase, D. V., and Sharp, W. W. (1989). “Field Testing Water Mains to Determine Carry-
ing Capacity.” Miscellaneous Paper EL-89, U.S., Army Engineer Waterways Experiment Station,
Vicksburg, Mississippi.
Morin, M., and Rajaratnam, I. V. (2000). Testing and Calibration of Pitot Diffusers. University of Alberta
Hydraulics Laboratory, Alberta, Canada.
Ormsbee, L. E., and Lingireddy, S. (1997). “Calibrating Hydraulic Network Models.” Journal of the Amer-
ican Water Works Association, 89(2), 44.
Rossman, L. A., Clark, R. M., and Grayman, W. M. (1994). “Modeling Chlorine Residuals in Drinking-
Sharp, W. W., and Walski, T. M. (1988). “Predicting Internal Roughness in Water Mains.” Journal of the
Summers, R. S., Hooper, S. M., Shukairy, H. M., Solarik, G., and Owen, D. (1996). “Assessing DBP Yield:
Uniform Formation Conditions.” Journal of the American Water Works Association, 88(6), 80.
Vasconcelos, J. J., Rossman, L. A., Grayman, W. M., Boulos, P. F., and Clark, R. M. (1996). Characteriza-
tion and Modeling of Chlorine Decay in Distribution Systems. AWWA Research Foundation, Denver,
Colorado.
Walski, T. M. (1984a). Analysis of Water Distribution Systems. Van Nostrand Reinhold, New York, New
York.
Walski, T. M. (1984b). “Hydrant Flow Test Results.” Journal of Hydraulic Engineering, ASCE, 110(6),
847.
Walski, T. M. (1985). “Correction of Head Loss Measurements in Water Mains.” Journal of Transportation
Walski, T. M. (1988). “Conducting and Reporting Hydrant Flow Tests.” WES Video Report, U.S. Army
Engineer Waterways Experiment Station, Vicksburg, Mississippi.
Walski, T.M. (1998). “Importance and Accuracy of Node Elevation Data.” Essential Hydraulics and
Hydrology, Haestad Press, Inc., Waterbury, Connecticut.
Walski, T. M. (2000). “Model Calibration Data: The Good, The Bad and The Useless.” Journal of the
Walski, T. M., Edwards, J. D., and Hearne, V. M. (1989). “Loss of Carrying Capacity in Pipes Transporting
Softened Water with High pH.” Proceedings of the National Conference on Environmental Engineering,
American Society of Civil Engineers, Austin, Texas.
Walski, T. M., Gangemi, Kaufman, and Malos. (2001). "Establishing a System Submetering Project." Pro-
ceedings of the AWWA Annual Conference, Washington, DC.
Walski, T. M., and Lutes, T. L. (1990). “Accuracy of Hydrant Flow Tests Using a Pitot Diffuser.” Journal of
the American Water Works Association, 82(7), 58.
Walski, T. M., and O’Farrell, S. J. (1994). “Head Loss Testing in Transmission Mains.” Journal of the
Water and Environment Federation (WEF). (1997). “Energy Conservation in Wastewater Treatment Facili-
ties.” WEF Manual of Practice MFD-2, Alexandria, Virginia.
Wright, C., and Nevins, T. (2002). “In-situ Tracer Testing for Determining Effective Inside Pipe Diame-
ters.” Proceedings of the ASCE Environmental Water Resources Institute Conference, American Society
of Civil Engineers, Roanoke, Virginia.
5.1 English Units: Compute the HGL at each of the fire hydrants for the pressure readings presented
below and complete the table.
Elevation Pressure Reading HGL

Location
(ft) (psi) (ft)
FH-1 235 57
FH-5 321 42
FH-34 415 15
FH-10 295 68
FH-19 333 45
FH-39 412 27
SI Units: Compute the HGL at each of the fire hydrants for the pressure readings presented below
and complete the table.
Elevation Pressure Reading HGL
Location
(m) (kPa) (m)
FH-1 71.6 393
FH-5 97.8 290
FH-34 126.5 103
FH-10 89.9 469
FH-19 101.5 310
FH-39 125.6 186
5.2 English Units: A tank is used to capture the flow from a fire hydrant as illustrated in the figure. The
tank is 50 ft long, 30 ft wide, and 12 ft high. What is the average discharge from the fire hydrant if
the container is filled to a depth of 10 ft in 90 minutes?
12 ft (3.7 m)
30 ft
(9.1 m)
50 ft
(15.2 m)
SI Units: A tank is used to capture the flow from a fire hydrant as illustrated in the figure. The tank is
15.2 m long, 9.1 m wide, and 3.7 m high. What is the average discharge from the fire hydrant if the
container is filled to a depth of 3.0 m in 90 minutes?
5.3 English Units: A fire flow test was conducted using the four fire hydrants shown in the following fig-
ure. Before flowing the hydrants, the static pressure at the residual hydrant was recorded as 93 psi.
Given the data for the flow test in the following tables, find the discharges from each hydrant and
finish filling out the tables. Flow was directed out of the 2 ½-in. nozzle, and each hydrant has a
rounded entrance where the nozzle meets the hydrant barrel.
Residual
Hydrant FH-1 FH-2 FH-3
Q1 Q2 Q3
//=//= //=//= //=//=
a) Would you consider the data collected for this fire flow test to be acceptable for use with a
hydraulic simulation model? Why or why not?
b) Based on the results of the fire flow tests, do you think that the hydrants are located on a trans-
mission line or a distribution line?
c) Would these results typically be more consistent with a test conducted near a water source (such
as a storage tank) or at some distance away from a source?
d) If the needed fire flow is 3,500 gpm with a minimum residual pressure of 20 psi, is this system
capable of delivering sufficient fire flows at this location?
Residual Pressure Pitot Reading Hydrant Discharge

Flowed Hydrant
(psi) (psi) (gpm)
Residual Hydrant 88 N/A
FH-1 N/A 58
FH-2 N/A 52
FH-3 N/A Closed

Flowed Hydrant
(psi) (psi) (gpm)
FH-1 N/A 65
FH-2 N/A Closed
FH-3 N/A Closed

Flowed Hydrant
(psi) (psi) (gpm)
FH-1 N/A 53
FH-2 N/A 51
FH-3 N/A 48
SI Units: A fire flow test was conducted using the four fire hydrants shown in the figure. Before
flowing the hydrants, the static pressure at the residual hydrant was recorded as 641 kPa. Given the
data for the flow test in the following tables, find the discharges from each hydrant and finish filling
out the tables. Flow was directed out of the 64 mm nozzle, and each hydrant has a rounded entrance
where the nozzle meets the hydrant barrel.
a) Would you consider the data collected for this fire flow test to be acceptable for use with a
hydraulic simulation model? Why or why not?
b) Based on the results of the fire flow tests, do you think that the hydrants are located on a trans-
mission line or a distribution line?
c) Would these results typically be more consistent with a test conducted near a water source (such
as a storage tank), or at some distance away from a source?
d) If the needed fire flow is 220 l/s with a minimum residual pressure of 138 kPa, is this system
capable of delivering sufficient fire flows at this location?
Flowed Hydrant
(kPa) (kPa) (l/s)
FH-1 N/A 448
FH-2 N/A Closed
FH-3 N/A Closed

Flowed Hydrant
(kPa) (kPa) (l/s)
FH-1 N/A 400
FH-2 N/A 359
FH-3 N/A Closed

Flowed Hydrant
(kPa) (kPa) (l/s)
FH-1 N/A 365
FH-2 N/A 352
FH-3 N/A 331
5.4 English Units: A two-gage head loss test was conducted over 650 ft of 8-in. PVC pipe, as shown in
the figure. The pipe was installed in 1981. The discharge from the flowed hydrant was 1,050 gpm.
The data obtained from the test are presented in the following table.
Elevation Pressure
(ft) (psi)
Fire Hydrant 1 500 62
a) Can the results of the head loss test be used to determine the internal roughness of the pipe? Why
or why not?
b) If the test results cannot be used, what is most likely causing the problem?
Flowed
//=//= //=//= //=//=
Closed Valve
SI Units: A two-gage head loss test was conducted over 198 m of 203-mm PVC pipe, as shown in
the figure. The pipe was installed in 1981. The discharge from the flowed hydrant was 66.2 l/s. The
data obtained from the test are presented in the following table.
Elevation Pressure
(m) (kPa)
or why not?
b) If the test results cannot be used, what is most likely causing the problem?
5.5 English Units: A different two-gage head loss test was conducted over the same 650 ft of 8-in. PVC
pipe shown in Problem 5.4. In this test, the pressure at Fire Hydrant 1 was 65 psi, and the pressure at
Fire Hydrant 2 was 40 psi. The discharge through the flowed hydrant was 1,350 gpm.
or why not?
b) What is the Hazen-Williams C-factor for this line?
c) How can the results of this test be used to help calibrate the water distribution system?
d) Is this a realistic roughness value for PVC?
SI Units: A different two-gage head loss test was conducted over the same 198 m of 203-mm PVC
pipe shown in Problem 5.4. In this test, the pressure at Fire Hydrant 1 was 448 kPa, and the pressure
at Fire Hydrant 2 was 276 kPa. The discharge through the flowed hydrant was 85.2 l/s.
or why not?
b) What is the Hazen-Williams C-factor for this line?
c) How can the results of this test be used to help calibrate the water distribution system?
d) Is this a realistic roughness value for PVC?

5.6 The table below presents the results of a chlorine decay bottle test. Compute the bulk reaction rate
coefficient for this water sample.
Time Concentration
(hr) (mg/l)
0 1.5
3 1.4
6 1.2
9 1.0
12 1.0
15 0.9
18 0.7
21 0.7
24 0.6
27 0.5
30 0.5
33 0.5
36 0.4
39 0.4
42 0.3
45 0.3
48 0.3
51 0.3
54 0.2
57 0.2
60 0.2
5.7 English Units: Data from a pump test are presented in the following table. Fortunately, this pump
had a pressure tap available on both the suction and discharge sides. The diameter of the suction line
is 12 in. and the diameter of the discharge line is 8 in. Plot the pump head-discharge curve for this
unit.
Suction Pressure Discharge Pressure Pump Discharge

(psi) (psi) (gpm)
10.5 117 0
10.1 116 260
9.3 114 500
8.7 111 725
7.2 101 1,250
5.7 93 1,500
4.4 85 1,725
3.0 76 2,000
1.6 65 2,300
-0.2 53 2,500
-2.0 41 2,700
SI Units: Data from a pump test are presented in the following table. Fortunately, this pump had a
pressure tap available on both the suction and discharge sides. The diameter of the suction line is
300 mm, and the diameter of the discharge line is 200 mm. Plot the pump head-discharge curve for
this unit.
Suction Pressure Discharge Pressure Pump Discharge
(kPa) (kPa) (l/s)
72.4 803 0
69.6 798 16.4
64.1 784 31.5
60.0 764 45.7
49.6 694 78.9
39.3 644 94.6
30.3 586 108.8
20.7 522 126.2
11.0 451 145.1
-1.4 366 157.7
-13.8 283 170.3
5.8 A C-factor test is conducted in a 350 ft length of 12-in. pipe. The upstream pressure gage is at eleva-
tion 520 ft, and the downstream gage is at 524 ft.
a) The Hazen-Williams equation can be rearranged to solve for C as
C =KQ/hL0.54
where C = Hazen-Williams roughness coefficient

K = constant
Q = flow (gpm)
hL = head loss due to friction (ft)
What is the expression for K if length (L) is in feet and diameter (D) is in inches? All of the terms
in K are constant for this problem, so determine the numerical value for K.
b) What is the expression for head loss between the upstream and downstream pressure gages if the
head loss (h) and elevations (z1 and z2) are in feet, and the pressures (P1 and P2) are in psi?
c) The elevations are surveyed to the nearest 0.01 ft and the pressure gage is accurate to +/- 1 psi.
Opening a downstream hydrant resulted in a flow of 800 gpm (accurate to +/- 50 gpm) with a
measured upstream pressure of 60 psi and a measured downstream pressure of 57 psi. Determine
the possible range of actual Hazen-Williams C-factors and fill in the following table.
Hint: For the roughest possible C-factor, use 800 – 50 gpm for flow and h + 5 ft for head loss. For
the smoothest possible C-factor, use 800 + 50 gpm for flow and h – 5 ft for head loss.
Measured Values Roughest Possible C Smoothest Possible C
Q (gpm)
h (ft)
C
d) What can you conclude about the C-factor from this test?
e) Which measurement contributed more to the error in this problem, head loss or flow?
f) What could you do to improve the results if you ran the test over again?
5.9 English Units: A hydrant flow test was performed on a main line where a new industrial park is to tie
in. The following hydrant flow test values were obtained from a 2 ½-in. nozzle in the field. First, use
Equation 5.1 to determine the hydrant discharge for a discharge coefficient of 0.90.
Determine if the existing system is able to handle 1,200 gpm of fire flow demand for the new indus-
trial park by using the equation given in the sidebar on page 189 entitled Evaluating Distribution
Capacity with Hydrant Tests.
Fire Hydrant
Static Pressure Residual Pressure Pitot Pressure
Number
200 48 psi 33 psi 12 psi
SI Units: A hydrant flow test was performed on a main line where a new industrial park is to tie in.
The following hydrant flow test values were obtained from 64-mm nozzle in the field. First, use
Equation 5.1 to determine the hydrant discharge for a discharge coefficient of 0.90.
Determine if the existing system is able to handle 75.7 l/s of fire flow demand for the new industrial
park by using the equation given in the sidebar on page 189 entitled Evaluating Distribution Capac-
ity with Hydrant Tests.
Fire Hydrant
Static Pressure Residual Pressure Pitot Pressure
Number
200 331 kPa 227.5 kPa 82.7 kPa
5.10 A utility performed a C-factor test on a pipe with a nominal diameter of 8 in. and calculated the C-
factor as 40. Later, tests showed that the true diameter was 6 in. due to severe tuberculation. Using
the correct diameter, what would the corrected C-factor be? If the flow in the pipe is 200 gpm (0.446
cfs), what would the velocity be using the 8-in. nominal diameter and the 6-in. actual diameter?
5.11 A chlorine field test is conducted to estimate the wall demand for a 6-in. diameter (actual diameter)
1500-ft length of pipe. The flow rate during the test is 300 gpm. The chlorine bulk decay rate was
determined to be –0.2/day based on a bottle test. Chlorine residual at the upstream and downstream
ends of the segment during the test was measured as 0.80 mg/l and 0.55 mg/l respectively. Calculate
the following values: velocity, travel time, chlorine loss due solely to bulk demand, and chlorine loss
due to wall demand (that is, the difference between observed chlorine loss and loss due to bulk
decay). Then set up a model of this link as shown in the figure and iteratively run the model to find
the wall demand coefficient that results in the observed chlorine loss. Assume a water temperature of
15oC.
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
6
Using SCADA Data for
Hydraulic Modeling
Supervisory Control and Data Acquisition (SCADA) systems enable an operator to

remotely view real-time measurements, such as the level of water in a tank, and
remotely initiate the operation of network elements such as pumps and valves.
SCADA systems can be set up to sound alarms at the central host computer when a
fault within a water supply system is identified. They can also be used to keep a his-
torical record of the temporal behavior of various variables in the system such as tank
and reservoir levels. Appendix E provides an in-depth introduction to SCADA sys-
tems and their components.
When working with SCADA data, the modeler often has access to more data than can
be easily processed. For example, the modeler may have several weeks of data from
which to calibrate an extended-period simulation (EPS) model and must pick a repre-
sentative day or days to use as the basis for calibration. Selecting the best modeling
analysis period from these thousands of numbers, which may be in several sources, is
extremely difficult. Usually, there is no day when all of the instrumentation is func-
tioning properly, so selecting that day is often based on finding the day with the few-
est problems.
Another challenge of working with SCADA data is that incorrect readings, time-scale
errors, or missing values may not be readily apparent in the mass of raw data. Fortu-
nately, the modeler can use any of several procedures to compile and organize
SCADA information into a more usable format, usually in the form of a spreadsheet.
The tables and graphs developed using these procedures can then be used directly for
a range of applications, including EPS model calibration, forecasting of system oper-
ations, and estimating water loss during main breaks.
236 Using SCADA Data for Hydraulic Modeling Chapter 6
This chapter provides guidance for addressing these challenges and discusses the
types of SCADA data, different data collection techniques and formats, interpretation
and correction of errors in SCADA data, verification of the validity of SCADA data,
and other general procedures for the handling and managing of SCADA data for the
purpose of hydraulic modeling.
6.1 TYPES OF SCADA DATA

Data received from SCADA systems fall into one of the following categories:
• Analog data (real numbers): Analog data are usually represented by inte-
gers or IEEE floating-point numbers (these are numbers that have no fixed
number of digits before and after the decimal point and that follow the popu-
lar Institute of Electrical and Electronics Engineers standard). It may be
trended (placed in charts that show variation over time) or used to generate
alarms should the data indicate an abnormal condition.
• Digital data (on/off or open/closed): Digital data may be used to sound
alarms, depending on the state (on/off or open/closed) reflected by the data.
• Pulse data: Pulse data, such as the number of revolutions of a meter, are
accumulated at either the site collection point or at the SCADA central host
computer. They are typically converted to the same number format as analog
data; however, they are physically derived in a different manner from pure,
real-number analog data obtained from field instrumentation.
• Status bits (or flags): Status bits are usually ancillary to analog data. For
example, a data flag can accompany an analog input if the SCADA system
determines that a value is possibly invalid.
Although SCADA data are useful for many hydraulic modeling applications, the gen-
eral composition of SCADA data — time-based records of flows, pressures, levels,
and equipment status — is especially well-suited for EPS analyses. However, steady-
state modeling investigations also can benefit from SCADA data. For example, the
data can be useful for setting model boundary conditions. As with any information
used for hydraulic modeling, SCADA data require careful handling and processing to
maintain their usefulness and should not be accepted blindly.
6.2 POLLING INTERVALS AND UNSOLICITED DATA

SCADA systems are typically deployed over large geographic areas using communi-
cations links such as radio or telephone lines. In comparison with local, hard-wired
computer links, such communication channels can be relatively slow. Therefore,
many SCADA systems employ some form of data acquisition scheduling to conserve
bandwidth. Consequently, data are often not collected continuously from all devices
in the field, and thus it is not uncommon for a SCADA system to display analog data
as some form of average values rather than continuous instantaneous values. There
are two major ways that data are obtained from the field:
Section 6.2 Polling Intervals and Unsolicited Data 237
Using Hydraulic Models to

Assist in SCADA Setup
For the most part, this book considers SCADA as Before a SCADA system comes on line, it is usu-
a source of data to help support hydraulic model- ally tested by simulating events such as tank lev-
ing efforts. However, a hydraulic model can also els and valve statuses using EPS model runs.
be used to assist SCADA operators with setting Results from these tests can be used to deter-
up controls for an existing SCADA system or for mine control set points, levels, and variable-fre-
entirely new SCADA installations. Rather than quency-drive settings. With model output linked to
experimenting with the real system, the operator the man-machine interface of the SCADA system,
can test out different control strategies in the it becomes possible to simulate much more realis-
model and determine if the new controls are an tic sequences of events to better test the SCADA
improvement or if there are adverse impacts. system.
• The central host computer “polls” the field devices

• The field devices send “unsolicited” data to the central host
In a polled system, the SCADA central host sequentially polls the remote terminal
units (RTUs) (see Appendix E for more information on RTUs), each of which respond
in order, reporting the latest analog data values, alarms, and equipment status
(whether pumps are off or on, whether valves are open or closed, and so on). If no
change of state has occurred, polled RTUs can report by exception, in which case they
respond with “nothing to report.” If a change of state has occurred, however, the RTU
reports the appropriate information. This approach reduces the bandwidth require-
ments of the system; however, because each RTU must wait for its turn to be polled,
the duration of each polling cycle may vary depending on the number of RTUs that
have something to report. Many systems can be configured to poll selected RTUs at
fixed intervals, allowing enough time for the system to collect all information from
each RTU.
In an unsolicited response system, the RTUs generate all reporting messages as
required in a “random” fashion. Typically, such messages report a change of state or a
fault, or simply pass data to the SCADA central host. The information itself is termed
unsolicited data because the central host has not called for the information. In such
systems, the RTUs download a collection of analog data values and past equipment
statuses that were being held in local memory but did not warrant an unsolicited
response to the central host. RTUs may also be configured to generate an unsolicited
data transfer when the data memory has reached full capacity.
Some SCADA systems use a combination of the polled and unsolicited response
mechanisms. They use periodic polling to ascertain the health of the field devices and
rely on unsolicited messages to be the vehicle by which field alarms are sent to the
central host and displayed. These systems are known as hybrid systems. The practical
result of hybrid systems is that data are often compressed in the field collection
devices before it is transmitted to the central host. This minimizes data traffic, making
optimum use of the restricted bandwidth associated with the communication chan-
nels, and allows greater amounts of data to be held in memory in the field RTUs.
6.3 SCADA DATA FORMAT

SCADA systems generally allow some form of data transfer to external applications.
For example, data may be exported as ASCII text, as a spreadsheet file, or to a propri-
etary “data historian” software package. Table 6.1 illustrates flow meter and valve
position data that have been collected from a SCADA system and fed directly into a
Microsoft Excel spreadsheet using a standard ODBC (Open Database Connectivity)
link.
Once the data are in tabular format within a spreadsheet, it is possible to manipulate
the data to investigate the behavior of the instrument or the associated plant being
monitored. In Table 6.1, the flow measurements over time are being used to assess the
performance of a valve used to throttle the flow through the pipe.
Table 6.1 Flow meter data imported directly into a spreadsheet

Flow Valve Position
Time
(ML/day) (% Open)
8/22/01 21:56 8.52 10.00
8/22/01 21:57 8.70 10.00
8/22/01 21:59 8.70 10.00
8/22/01 22:00 8.76 10.00
8/22/01 22:01 8.76 10.00
8/22/01 22:02 8.52 10.00
8/22/01 22:14 8.52 18.40
8/22/01 22:15 10.26 19.24
8/22/01 22:16 13.20 20.08
8/22/01 22:17 13.26 20.92
8/22/01 22:19 13.26 22.61
8/22/01 22:20 16.74 23.45
8/22/01 22:21 17.58 24.29
8/22/01 22:22 19.32 25.13
8/22/01 22:23 19.92 25.97
8/22/01 22:24 19.68 26.81
8/22/01 22:25 19.68 27.65
8/22/01 22:26 22.62 28.49
8/22/01 22:27 22.50 29.33
8/22/01 22:29 22.50 31.01
Section 6.4 Managing SCADA Data 239
6.4 MANAGING SCADA DATA

To process SCADA data, the modeler must perform an overview of the records, orga-
nize the information into hydraulically related groups, review the data in detail to
identify and resolve potential problems (such as timing problems, missing data, and
so on), and develop a model time-step scale for EPS analyses.
The first step in using SCADA records is to perform an initial review of the informa-
tion downloaded from the SCADA system(s) and any other data sources. The begin-
ning and ending times and dates for each parameter must be identified to determine
the maximum common period of record for all of the parameters. Obvious errors in
the records, such as blanks or default values, should be factored into the decision
regarding the extent of the period of record.
At this point, it is also sensible to classify data into two groups: (1) measurements
averaged over the SCADA time increment, and (2) values reported on the SCADA
time stamp. An example of averaged SCADA data are pump station flows determined
from totalizing flow meters (devices that measure the total quantity of flow). Exam-
ples of time stamp SCADA data are tank levels or system pressures. The difference in
these two categories of data is evident in several modeling applications, in particular
when preparing diurnal demand curves. For example, a pressure of 61.2 psi from a
sensor at 10:15:23 a.m. may be the pressure at that instant or the average pressure
since the last reading. The modeler needs to check which type of value is being dis-
played.
In large complex systems, the SCADA information should be organized into groups
corresponding to distribution gradients that usually corresponds to pressure zones in
the hydraulic model. In some cases, it may be necessary to combine several model
pressure zones into one SCADA data group where SCADA information is not col-
lected at pumps or valves between the zones. After the SCADA groups are estab-
lished, the modeler should identify the records that should match, such as certain
pump station flows, pressures, and tank levels, in order to check and confirm the
SCADA records. Note that it is generally necessary to place some system facilities in
more than one SCADA group. For example, a flow meter record can represent an out-
flow from one zone and an inflow to a second zone.
6.5 SCADA DATA ERRORS

Errors in data obtained from SCADA systems can be caused by system failure or
scheduled downtime in the system and can include gaps in trended data or missing
digital event or alarm points. The following sections concentrate on errors in analog
data collected from the field during normal operation of a SCADA system. Such
errors may not impact the creation and tuning of a water distribution system model,
but the modeler should be aware of any data inaccuracies and take them into account
when comparing the model results with the data obtained from the SCADA system.
Data Compression Problems

Errors in SCADA data are commonly caused when data in the RTUs are compressed.
The data are compressed in order to reduce the amount of data that are transferred
between field-based devices and the SCADA central host, thereby making best use of
the available bandwidth associated with the communication links. The effect of this
compression is that trended analog data received from the field may not consistently
reproduce the behavior of an actual field variable. In particular, data received via
SCADA may not vary as quickly as the actual field variable. It is important that users
of the data understand the magnitude of error inherent in the data obtained from the
particular SCADA system in use. When using SCADA data for model validation, for
example, it is important that the quality of the collected data is understood before a
model is tuned to adequately reflect the field variable in question.
Common compression techniques use various transmission methods, but they trans-
mit information only when there is significant variation in the behavior of the vari-
able. For example, one such technique tracks the gradient of a trended variable over
time and transmits information to the central host only when the gradient has changed
by greater than a preset amount. Other techniques involve transmitting a new value
only when the variable has altered by a certain preset percentage of the total span of
the variable. Analog data obtained from a SCADA system may therefore exhibit
“step” behavior, which does not correspond precisely with the variation of the vari-
able in the field. This behavior can occur whether information is accessed from a poll
to the field or from unsolicited data transfer.
Other SCADA systems may average the field data between polls or between the
opportunities for unsolicited data transfers. In each of these cases, the resolution of
the data on display at the SCADA operator’s terminal is of a lesser quality than the
actual variation of the field variable. Users of the data should therefore understand
whether a value is instantaneous, and therefore at what time it was collected, or an
average value over the polling interval. It is important that users of the data under-
stand the particular mechanisms by which the data have been collected and take this
into account when analyzing the data.
Timing Problems
Other sources of error in SCADA trends may be temporary. When data are held in the
field memory and then transferred to the central host, such as is the case for an unso-
licited data transfer, the SCADA system may employ a “back-filling” mechanism.
When this happens, data are collected from the RTUs and then past values in the trend
display on the SCADA central host computer are updated to indicate the values
uploaded from the field. Therefore, there is a period when the values in the displayed
trend may not represent a complete record from the field, and the most recent data on
display is yet to be updated from the latest field information. Operators and modelers
viewing the data should be aware of how far in the past a back-filling mechanism may
affect trended data. Practically, data may appear as constant until updated. The time
period during which the displayed trend is inaccurate is set by the configuration of the
SCADA system, which is intended to optimize the use of communication bandwidth
and field-based data memory.
Section 6.5 SCADA Data Errors 241
Integrating SCADA Systems and Hydraulic

Models: Two Sample Applications
ESTIMATING PARAMETERS AT NON-SCADA This type of procedure is typically automated and
LOCATIONS accomplished with some form of dynamic data
SCADA systems monitor the water distribution linkage between the SCADA system and model-
system performance at discrete stations scattered ing software.
throughout the service area. However, there may
be locations in the distribution system, such as ESTIMATING WATER LOSS DURING MAIN
meter pits, that lack the power or communication BREAKS
connections required for a functional SCADA sta- Tracking the water discharged from the system
tion. For these situations, the flows and pressures during main breaks can help quantify losses.
can be estimated from SCADA information at Generally, a significant main break will show up in
nearby stations. When these calculations are not SCADA records as low pressures readings, an
complicated, they can be performed within the unexplained decline in tank levels, excessive
SCADA software (for example, by offsetting pres- pump flow, or other unexplained data inconsisten-
sure readings from other stations based on differ- cies.
ences in elevation). However, when the situation SCADA information for the time period surround-
is more complex, a hydraulic model interfaced ing the break can be downloaded to a hydraulic
with the SCADA software is required to obtain model and the model can then be executed to
parameter estimates. simulate system conditions at the time of the
The steps involved in calculating information for break.
non-SCADA sites include the following: By adjusting the demands (or emitter coefficients)
• Export data on boundary conditions from at the break location and trying different start and
SCADA. finish times for the break, the modeler should be
• Configure the hydraulic model to match those able to match modeling results to the SCADA
specific conditions. records during the break and thus determine the
quantity of water lost. These results can then be
• Execute the model.
used in estimates of unaccounted-for water.
• View the results or import results from the
model back into SCADA.
The data back-filling mechanism is a useful function that allows optimization of a

SCADA system, but it can cause difficulties for utility data collection. Therefore,
when an organization employs an automatic data historian database product to facili-
tate deployment of SCADA data into the utility-wide computing environment, it is
important that the software be compatible with the data back-filling mechanism of the
SCADA system. This ensures that trend data is not lost to other users of the data.
In addition, a temporal error can occur in the SCADA data itself. Some data may be
time-stamped with the time at which it was received by the central SCADA host com-
puter as opposed to the time at which it was collected in the field. As a result, events
and trend time stamps may not correspond with the times of the actual event occur-
rences in the field. If a SCADA system is configured to behave in this manner, the
user must be aware of the time lag inherent in the communication channels and vari-
ous data collection mechanisms. For many SCADA systems, this time lag may only
be several seconds or less, but for systems incorporating large remote data storage, it
is conceivable that under extreme circumstances, the time stamp may be substantially
incorrect. This latter situation is most inconvenient for event and alarm data.
For example, consider a system that is configured to time-stamp events and alarms
when they are received at the SCADA central host computer and that has remote out-
stations that are configured to buffer alarms and events if communications links to the
central host are lost or the central host is out of service. This is desirable, because
operational staff must know about alarms that have occurred during an outage, and the
exact time at which they occurred may not be considered relevant, as long as the
alarm is received. When the system is restored to normal operation and the alarm or
event is received, it is time-stamped by the central computer. A user viewing the his-
torical data obtained from the SCADA system would then see that the alarm or event
occurred at a time that did not correspond to evidence gathered from other observa-
tions, such as site-based data loggers or back-filled trend data. Users of the data must
understand the mechanism by which the data under review are time-stamped and
know whether system outages occurred that might have caused a substantial error in
the time stamp. This information should be considered when the data are analyzed.
Authenticating SCADA information to confirm that it is usable and sufficiently accu-

rate for modeling purposes is greatly assisted by preparing plots of each data record
over the previously determined common time period. The time scale, plotted on the
horizontal axis, should use the time stamp placed on the records by the SCADA sys-
tem. Reviewing and comparing these plots, individually and in groups, helps deter-
mine whether there are missing records, instrumentation problems, or time-stamp
inaccuracies. These conditions are much easier to identify from the plotted data than
from reading columns of numbers in a database.
Certain system operations should coincide, such as a pump start at a booster station
that supplies a zone and an increase in the water level in a tank in that same zone. If
interrelated operations do not agree, such as if the tank level begins to rise prior to the
time of the pump start, then the time scale of the SCADA data needs to be checked. A
thorough examination for time-related problems is especially important if SCADA
and system information were derived from more than one source. Figure 6.1 shows an
example of SCADA timing problems.
Missing Data
Missing data in SCADA information can be caused by many factors, such as power
failures, communication failures between the RTU and the central host computer, or a
variety of temporary SCADA software glitches. Regardless of the reason, periods dur-
ing which system data are incomplete must be identified. In many SCADA systems,
missing data are flagged. Typically, either an RTU or the central host computer will
set a questionable data flag if it determines that an input or sensor is not performing
properly. The most common cause is an over-range or under-range value. Also, if the
central host computer cannot communicate with an RTU, it will set flags for all the
database entries, which would otherwise be supplied by that RTU. If a back-filling
feature is not available, the modeler can rely on the flag to indicate that data are not
available.
Figure 6.1
TIMING PROBLEMS Timing problems
36
34
Discharge Tank Level, ft
32
30
28
26
24
22
20
0 3 6 9 12 15 18 21 24
Time, hours
350
300 DATA FROM

SCADA SYSTEM
Pump Station Flow, gpm
250
200
150
100 SHIFT SCADA PUMP OPERATIONS

RECORDS SHOULD CORRESPOND
2 HOURS TO TANK LEVEL PLOTS
50
-50
0 3 6 9 12 15 18 21 24
Time, hours
36
34
32
Suction Tank Level, ft
30
28
26
24
22
20
0 3 6 9 12 15 18 21 24
Time, hours
If no status flags are used, the modeler can look for other signs of missing data. A
rapid drop to zero or a negative value at the beginning of the problem period serves as
evidence that data is missing. Similarly, a rapid increase at the end of the problem
period can serve as evidence. Some RTUs use “rate of change” alarms to highlight
such conditions.
Note that not all zero values indicate incorrect SCADA data; however, because of zero
drift (a shift in the zero point of sensors usually over the long term), some sensors
may give a nonzero reading even when the true value is zero. Pumps that are off may
show a flow rate of 2 gpm, for example, which should be converted to zero.
Another sign that records are missing from the SCADA information is the occurrence
of horizontal sections on the SCADA data plots. SCADA software can be pro-
grammed to “latch” data, such that the last reported parameter value is held in mem-
ory until a new, updated value is received. In this situation, the data record plot will
not exhibit a drop to zero when data is missing. Instead, the SCADA data plot will
“flat-line” during the missing record period and resume its normal appearance at the
end of the problem period. Figure 6.2 shows examples of missing data periods in a
SCADA data plot.
Missing data is typically converted to all zeroes or flat lines, but some systems initiate
corrective actions to replace missing data, such as linear interpolation and use of aver-
ages of a number of known good data points. Systems using such techniques often
include flags to indicate that the original data was missing or questionable. In addi-
tion, some SCADA systems use editing packages, which allow the user to determine
how to address missing data.
Instrumentation
Instrumentation-related problems include inaccurate data, extreme fluctuations in
readings, and inadequate instrument range. Inaccurate readings from field instruments
may occur as a result of uncalibrated equipment, signal interference in the input
cabling to the RTU, misinterpretation by SCADA software, or other factors. Inaccu-
rate data may be difficult to ascertain from downloaded SCADA information, in par-
ticular when a station is located in an isolated site in the system and there are no
nearby stations to correlate data to check accuracy. If it is suspected that a certain field
instrument is inaccurate, the first step should be to calibrate it. If there are further con-
cerns, a chart recorder or data logger can be used to document instrument operations.
Extreme fluctuations in readings, called data spikes, can result from hydraulic tran-
sients produced at pump starts/stops or valve open/close. (See Chapter 13 for a
detailed discussion of hydraulic transients.) Data spikes also can occur as a result of
power surges or intermittent interference from improperly shielded signal cables.
Some smart field instruments also use extreme values to show that internal diagnos-
tics have determined that the input value is questionable. Generally, spikes show as a
single extreme value or smaller, constantly repeating variations in the SCADA
records. These fluctuations do not usually reflect actual system operations and will
not be reproduced by hydraulic model simulation. Therefore, the modeler should fil-
ter SCADA records, as appropriate, to moderate spikes in the data.
Figure 6.2
Missing SCADA data
MISSING DATA PERIODS IN A SCADA RECORD
DATA PLOT DROPS TO ZERO
200
Missing
175 Data
150
Pressure, psi
125
100
75
50
25
0
0 3 6 9 12 15 18 21 24
Time, hours
MISSING DATA PERIODS IN A SCADA RECORD

LATCHED VALUES
200
MISSING DATA
175
150
Pressure, psi
125
100
75
50
25
0
0 3 6 9 12 15 18 21 24
Time, hours
Field instruments typically are set to operate over a certain range or span. If the span
is insufficient, there may be times when the system is performing at a point higher or
lower than the capability of the instrument. During those periods, the SCADA data
plot usually will “flat-line” until the system variable returns to a value within the
instrument range. Such behavior could also indicate that the instrument has failed or
is out of calibration, or that the signal is bad. If it is determined that a field instrument
has inadequate range, the span should be increased or a replacement instrument with
the appropriate range should be installed. Estimating values during a “flat-line”
period may be difficult. Correlating with data from nearby stations or installing tem-
porary chart recorders or data loggers may help fill in the “flat-line” records. Figure
6.3 shows SCADA data plots that reflect instrument-related problems.
Figure 6.3
Instrumentation
problems
INSTRUMENTATION PROBLEMS
200
175 DATA SPIKE
150 MAXIMUM
INSTRUMENT
RANGE
Pressure, psi
125
100
75 MINIMUM
INSTRUMENT
RANGE
50
25
DATA SPIKE
0
0 3 6 9 12 15 18 21 24
Time, hours
Unknown Elevations
The sensors in a SCADA system may be very accurate, but the elevation of the sensor
(not the elevation of the RTU or the ground) may be unknown (or only roughly
known). For data from that sensor to be useful, the exact elevation must be deter-
mined. For example, the zero reading of a water level sensor may be referenced to the
elevation of the transducer in a valve vault and not the floor of the tank.
Other Error Sources

Other sources of error in SCADA systems include the following:
• Failure of the communication system, central host, or RTU. This error
may be identified by physical gaps in the data recorded by the SCADA sys-
Section 6.6 Responding to Data Problems 247
tem. Normally, status flags would also indicate that there was a communica-
tion problem.
• Noise in the communication system. This error is not obvious from the data
recorded by SCADA but can be identified through a long-term comparison
with data collected directly from field-based data loggers or by complex sta-
tistical filtering techniques. However, most substantial noise errors typically
result in no information being received at the central host computer.
• Failure with the field instrumentation that may go undetected for a long
period of time. Such errors are usually caused by a drift in the calibration of
the field instrument and can be identified through model comparison or
through long-term comparison with data collected directly from field-based
data loggers.
• Insufficient resolution of the field data collection device. Many RTUs
employ only 16- or 32-bit resolution. If the instrument being used to measure
the field variable allows a higher data word length, then the resolution of the
data from that instrument is lost in the communication of that data. This
error can also be identified through long-term comparison with data col-
lected directly from field-based data loggers. Most new RTUs avoid this
problem by using IEEE floating point format; however, the problem is valid
for older RTUs or for programmable logic controllers (PLCs) (see Appendix
E for more information on PLCs), which use integer representations of ana-
log data.
6.6 RESPONDING TO DATA PROBLEMS

When incorrect SCADA readings are found, the modeler typically examines another
time period and set of data where the problems do not occur. However, if a unique dis-
tribution system event is to be analyzed or if collecting the SCADA data requires a
special effort by SCADA operators, the modeler may not have the option of selecting
another problem-free period. In these cases, many of the previously described prob-
lems can be resolved in order to improve the SCADA information and arrange it in a
format sufficient for hydraulic modeling applications. Discussions with SCADA
operators can provide insight into the causes of inconsistencies in the data and permit
the modeler to make appropriate allowances. Completing the data series with infor-
mation from chart recorders, data loggers, or other monitoring devices, and shifting
time scales where justified also can address SCADA data issues sufficiently to sup-
port modeling applications.
EPS modeling analyses require SCADA information to be divided into model time
steps. The duration of the model time step depends on the type of analysis being per-
formed and is usually based on separating the total analysis time period into a reason-
able and manageable number of steps. However, it may be difficult to download
SCADA data at time increments that directly correspond to model time steps. A gen-
eral guideline is
Model time step length t SCADA time increment length (6.1)

SCADA information listed at a higher frequency than the model time step may exhibit
minor oscillations, or signal flutter. Some type of averaging (or smoothing) method
(for example, three-point moving average) can be used to smooth the flutter, as shown
in Figure 6.4.
Usually, it is suitable to interpolate between SCADA values to calculate parameters at
the model time step. For example, a SCADA database with a time increment of 15
minutes may list a tank level of 25 feet at 10:50 a.m. and 31 feet at 11:05 a.m. The
model time steps are each one hour in length and begin on the hour. A tank level of 29
feet at 11:00 a.m. can be interpolated from these SCADA readings for use in model
applications. However, averaging or interpolating information that involves a change
in status may not produce acceptable results. For example, a pump start/stop or valve
open/close may require an individual model time step to pinpoint its occurrence, as
Figure 6.4
Smoothing data
fluctuations
6.7 VERIFYING DATA VALIDITY

Along with the convenience of remote monitoring via SCADA comes the drawback
of data consumers becoming overly reliant on the data received from the SCADA sys-
tem. A user of data may mistakenly assume the correctness of data received from a
SCADA system when in fact the only way to be assured of its integrity is through crit-
ical analysis. Software for the central host is available that offers automatic detection
of sensor data errors through continuous automatic analysis of data using such tech-
niques as neural network analysis. However, more conventional data techniques are
typically used to verify the validity of critical sensors and systems associated with a
SCADA system.
Section 6.7 Verifying Data Validity 249
Figure 6.5
Data averaging
problems
Important flow meters and other such sensing devices used in a SCADA system can
be checked for data integrity by using locally housed data recorders, such as paper
chart recorders, or, more recently, electronic data loggers. This equipment is physi-
cally connected to the data signal from the sensor device under investigation and then
left to accumulate data from that device. Such data recording techniques frequently
provide a better source of data than that offered by a SCADA system mainly because
the sources of communication error mentioned in the preceding sections do not exist.
Of course, the sensor itself may require calibration. Data obtained from a SCADA
system can be used to indicate when a sensor is not operating correctly. This is done
by comparing data from that sensor against an expected value derived through calcu-
lations incorporating data from other sensors in the system. Such comparisons can aid
in scheduling sensor maintenance. Correct calibration of the sensor may then be
achieved through comparison with temporary sensing equipment, maintained to a
high standard of accuracy and placed on site to mirror the performance of the sensor
under review. Many water utilities employing SCADA systems have found that such
temporary sensors used in conjunction with local data recording systems are useful in
the calibration of sensors as well as SCADA system equipment. A regular mainte-
nance program involving such temporary site-based sensing equipment is one method
by which a water utility can be assured of the reliability of the field data received from
a SCADA system.
Local data recording systems are not hindered by the restrictions of a limited band-
width communication channel. They are therefore particularly appropriate when col-
lecting a high rate of change detail, such as rapidly changing signal level transients on
a radio link or fast pressure changes due to hydraulic transients. To capture the same
level of detail from a digitized input to a SCADA system, a high sampling rate would
be required, which would probably not be appropriate for the available bandwidth
used on the SCADA communications system.
REFERENCES
Barnes, M., and Mackay, S. (1992). Data Communications for Instrumentation and Control. Instrument
Data Communications (IDC), Australia.
Haime, A. L. (1998). “Practical Guide to SCADA Communications.” SCADA at the Crossroads Conference
Workshop, The Institution of Engineers Australia, Perth, Western Australia.
Williams, R. I. (1992). Handbook of SCADA Systems for the Oil and Gas Industries. Elsevier Advanced
Technology Limited, 1st Edition, Great Yarmouth, United Kingdom.
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
7
Calibrating Hydraulic Network
Models
Even though the required data have been collected and entered into a hydraulic simu-
lation software package, the modeler cannot assume that the model is an accurate
mathematical representation of the system. The hydraulic simulation software simply
solves the equations of continuity and energy using the supplied data; thus, the quality
of the data will dictate the quality of the results. The accuracy of a hydraulic model
depends on how well it has been calibrated, so a calibration analysis should always be
performed before a model is used for decision-making purposes.
Calibration is the process of comparing the model results to field observations and, if
necessary, adjusting the data describing the system until model-predicted perfor-
mance reasonably agrees with measured system performance over a wide range of
operating conditions. The process of calibration may include changing system
demands, fine-tuning the roughness of pipes, altering pump operating characteristics,
and adjusting other model attributes that affect simulation results.
The calibration process is necessary for the following reasons:
• Confidence: Results provided by a computer model are frequently used to
aid in making decisions regarding the operation or improvement of a hydrau-
lic system. Calibration demonstrates the model’s capability to reproduce
existing conditions, thereby increasing the confidence the engineer will have
in the model to predict system behavior.
• Understanding: The process of calibrating a hydraulic model provides
excellent insight into the behavior and performance of the hydraulic system.
In particular, it can show which input values the model is most sensitive to,
so the modeler knows to be more careful in determining those values. With a
better understanding of the system, the modeler will have an idea of the pos-
sible impact of various capital improvements or operational changes.
252 Calibrating Hydraulic Network Models Chapter 7
• Troubleshooting: One area of calibration that is often overlooked is the

capability to uncover missing or incorrect data describing the system, such
as incorrect pipe diameters, missing pipes, or closed valves. Thus, another
benefit of calibration is that it will help in identifying errors caused by mis-
takes made during the model-building process.
This chapter begins with a discussion of data requirements and the reasons for dis-
crepancies between computer-predicted behavior and actual field performance of a
water distribution system. Variations can stem from the cumulative effects of errors,
approximations, and simplifications in the way the system is modeled; site-specific
reasons such as outdated system maps; and causes that are more difficult to quantify
such as the inherent variability of water consumption.
Next, the chapter addresses some of the specific methods used to calibrate models,
including manual and automated approaches such as genetic algorithms. The chapter
concludes with a discussion of the limits of calibration and how to know when the
model is sufficiently calibrated.
7.1 MODEL-PREDICTED VERSUS FIELD-MEASURED

PERFORMANCE
In making comparisons between model results and field observations, the user must
ensure that the field data are correct and useful. The details of field-testing were
explained in Chapter 5; this section focuses on identifying data useful for calibration.
Comparisons Based on Head

When comparisons are made between field and model results, there is no mathemati-
cal reason to use pressures instead of hydraulic grades, or vice versa. Because pres-
sure is just a converted representation of the height of the HGL relative to the ground
elevation datum, the two are essentially equivalent for comparison purposes. For cali-
bration purposes, however, there are several compelling arguments for working with
hydraulic grades rather than pressures (Herrin, 1997):
• Hydraulic grades provide the modeler with a sense of the accuracy and reli-
ability of the data. If computed and measured hydraulic grade values are
drastically different from one another, it should immediately signal the mod-
eler that a particular value may be in error. For example, an elevation may
have been entered incorrectly.
• Hydraulic grades give an indication of the direction of flow—insight that

pressures do not provide.
• Working with hydraulic grades makes it easier to work with pressure mea-
surements not taken exactly at node locations within the model, because it is
the elevation of the pressure gage, not the node, that is used to convert mea-
sured pressure into HGL.
Section 7.2 Sources of Error in Modeling 253
Although both HGL and pressure comparisons will lead to the same results if all other
factors are equal, pressure comparisons make it much easier to overlook errors and
much harder to track down inconsistencies between real-world observations and the
model results. Accordingly, the first step the modeler should complete upon collec-
tion of field data is to convert pressure and tank water level data into the equivalent
HGLs. Subsequent comparisons should be made between observed and modeled
HGLs.
Location of Data Collection

Errors in roughness coefficients and demands affect the slope of the hydraulic grade
line. If data are collected near the boundary nodes, the differences between the model
and the field data may appear to be small because of the short distance even though
the slope of the HGLs (and hence the roughness coefficient and demand) are signifi-
cantly in error. Head data for model calibration should generally be collected at a sig-
nificant distance from known boundary heads. Data should also be collected for pipes
that have not been removed from the model during skeletonization.
There should be at least one flow test conducted in each pressure zone, and the num-
ber of flow tests should be roughly proportional to the size of the pressure zone. In
general, more tests will increase the confidence the user will have in the model. One
approach to selecting sampling locations uses a special procedure to select locations
that minimize the uncertainty of the model’s predictions (Bush and Uber, 1998).
Another uses genetic algorithms to determine the best locations to conduct fire
hydrant flow tests to maximize the coverage of the pipe network (Meier and Barkdoll,
2000).
Data collection can be classified as either point reading (grab samples) or continuous
monitoring. Point reading involves collecting data for a single location at a specific
point in time, and continuous monitoring involves collecting data at a single location
over time. For point readings, samples should be collected at locations where the
parameter being measured is steady so that the sample measurement is representative
of the location over a fairly long period of time. To get the most out of continuous
monitoring, the data should be collected from locations where the parameter being
measured is dynamic. In situations where a point reading must be made at a dynamic
location, it is critical to carefully note the time and boundary conditions correspond-
ing to the data point.
7.2 SOURCES OF ERROR IN MODELING

The primary objective of a simulation is to reproduce the behavior of a real system
and its spatial and dynamic characteristics in a useful way. To accomplish this goal,
data are supplied that depict the physical characteristics of the system, the loads
placed on the system, and the boundary conditions in effect. Even if all the data gath-
ered describing the model match the real system exactly, it is unlikely that the pres-
sures and flows computed by the simulation model will absolutely agree with
observed pressures and flows. Significant mathematical assumptions are employed by
the simulation software to make the simulation computationally tractable, yet allow
the simulated results to be meaningful and useful. Thus, modeling is essentially a bal-
ance between reality, a simulated reality, and the effort necessary to make the two
agree. This section explores some of the sources of error in input data, as well as the
causes for discrepancies between field conditions and modeling results.
Some may assume that calibration can be accomplished by adjusting only internal
pipe roughness values or estimates of nodal demands until an agreement between
observed and computed pressures and flows is obtained. Generally speaking, the basis
for this claim is that unlike pipe lengths, diameters, and tank levels, which are directly
measured, pipe roughness values and nodal demands are typically estimated, and thus
have room for adjustment. Numerous factors, however, can contribute to disagree-
ment between model and field observations (Walski, 1990). Any and all input data
that have uncertainty associated with them are candidates for adjustment during cali-
bration to obtain reasonable agreement between model-predicted behavior and actual
field behavior.
A discrepancy found during the calibration process can also mean that the system
itself has problems. A review of the system should be done before any changes are
made to rationally developed model data. Possible system problems are large leaks,
unchartered services, previously undetected errors in the metered consumption, errors
in recorded pipe sizes, unknown throttled or closed valves, worn pump impellers, or
old construction debris left in pipes.
Types of Errors
Errors in input data can be broken down into two main categories, typographical
errors and measurement errors. Typographical errors, although simple to correct, can
be very difficult to uncover (for example, a pipe length of 2,250 ft is accidentally
entered as 250 ft). Fortunately, some of today’s graphically-based hydraulic network
models have tools that can help reduce the potential for typographical errors. For
example, some models include automatic validation of input values and/or the ability
to determine pipe lengths and vertices automatically by measuring the distance
between two nodes based on the drawing scale.
Unfortunately, these tools do not completely eliminate the possibility for human error.
After data entry is completed, it is recommended that the model be reviewed for pos-
sible typographical errors. One tip is to use the sorting and color-coding capabilities
available in many models to quickly identify very large or very small values for pipe
length, diameter, or internal roughness. At a minimum, such values should be verified
as accurate.
Compared to typographical errors, measurement errors can be much more difficult to
identify and correct. One example of such an error may result from variations in scale
on system maps. For instance, if a length of a pipe is measured with an engineering
scale from a system map that has a scale of 1 in (2.54 cm) = 1000 ft (304.8 m), the
measured length may only be within ± 50 ft (15.24 m) of the actual length. Depending
on the application of the model, this level of accuracy may or may not be sufficient for
calibration purposes.
Nominal versus Actual Pipe Diameters

As discussed in Chapter 3 (see page 92), the nominal and actual diameters of a pipe
typically differ. Determining the actual diameter of a pipe is further complicated by
the chemical processes of corrosion and deposition that occur over time after the pipe
has been installed. Therefore, for lack of a better value, nominal pipe diameters are
generally used for model development, and the roughness coefficient is adjusted to
compensate for the change in diameter due to pipe wall build-up.
With severe tuberculation, a Hazen-Williams C-factor as low as 20 or 30 may be nec-
essary to obtain a suitable calibration. Conversely, high roughness coefficients may be
needed for calibration of new piping. Because the actual diameter of new pipe is usu-
ally greater than the nominal diameter, an increased roughness coefficient may be
used to account for the difference.
The pipe diameter has a much greater influence on the head loss through a pipe than
the pipe roughness value does. According to the Hazen-Williams equation, the head
loss is a function of the pipe diameter raised to nearly the fifth power, while it is a
function of the roughness value raised to only the second power. The result is that a 10
percent increase in the pipe diameter will decrease head loss by nearly 40 percent,
while a 10 percent increase in the roughness coefficient will decrease head loss by
about 20 percent.
There is little advantage to be gained in adjusting both the roughness coefficient and
diameter in a model. For example, a 6-in. (150-mm) pipe with a roughness coefficient
of 100 gives the same head loss as a 5-in. (130-mm) pipe with a coefficient of 161. In
calibration, the user wants to minimize the number of variables to adjust, and consid-
ering diameter as an unknown would double the number of variables that must be
determined for each pipe. Accordingly, making adjustments to roughness coefficients
is the preferable means of fine-tuning a model calibration (except for certain situa-
tions in water quality calibration, as explained later in this chapter on page 281).
Roughness coefficient values can help identify other problems in a model. In general,
if C-factors less than 40 or greater than 150 are needed to calibrate the model, then
chances are that some other condition, such as a partially closed valve, may be caus-
ing the difference between observed and modeled heads. The status of the valve may
then be changed within the model, or the valve in the real system may require adjust-
ing. Either alternative is a valuable result of the calibration process.
Internal Pipe Roughness Values

A great deal of research has been done in the area of estimating pipe roughness val-
ues. Colebrook and White (1937) developed the theory behind the loss of carrying
capacity with age. Full-scale testing of pipes was done in several cities to document
the effect in real systems (California Section AWWA, 1962; and Hudson, 1966).
Later, Lamont (1981) compiled an extensive table documenting pipe C-factors for a
wide variety of pipe materials, sizes, and ages. The increase in pipe roughness as a
function of water quality was also evaluated (Walski, Edwards, and Hearne, 1989).
The research determined that two pipes of the same size, material, and age can have
different effective diameters and roughnesses based on the quality of the water histor-
ically flowing through the pipe.
Compensating Errors. Despite all of these variables, pressure data collected in

the field can be used to select appropriate roughness values for the pipes. However, in
calibrating a model, it is important to consider the potential for compensating errors;
that is, fixing one inaccuracy by introducing another one into the network. When cali-
brating, the adjustments made to the variables should be appropriate for a range of
operating conditions, and not just the individual case being considered.
Example — Compensating Errors. Consider the simple parallel pipe system shown in
Figure 7.1 for which the internal roughness values of each pipe are unknown. Suppose that pressure
measurements have been taken at the nodes on each end of the pipe segments, the total flow through
the system is known, and the elevations of the nodes on each end of the pipe loop are the same. Know-
ing that the head losses through Pipe 1 and Pipe 2 are the same, the expressions for head loss in Pipes
1 and 2 can be equated. (Note that the pressure drop of 7 psi translates into a head loss of 16.2 ft.)
L 1 § Q 1· 1.852 L 2 § Q 2· 1.852
-----------
- ------ = -----------
- ------
4.87 © C ¹ 4.87 © C ¹
D1 1 D2 2
where L = pipe length (ft)

D = pipe diameter (ft)
Q = pipe discharge (cfs)
Table 7.1 shows the results of a simple analysis performed on this system. Column 1 provides a range
of assumed C-factor values for Pipe 1. The flow through Pipe 1 resulting from the assumed C-factor,
known head loss, pipe length, and diameter is shown in Column 2. Column 3 provides the flow in
Pipe 2 assuming a total system flow of 1,350 gpm. Column 4 shows the C-factor for Pipe 2 back cal-
culated from the head loss, pipe characteristics, and pipe flow.
Figure 7.1
Simple parallel pipe
Pipe 1 system
L = 2,300 ft
P = 54 psi D = 10 in. P = 47 psi
1,350 gpm
Pipe 2
L = 2,800 ft
D = 8 in.
Table 7.1 Flow rate versus pipe roughness values for this parallel pipe system
Pipe #1 Pipe #1 Flow Pipe #2 Flow Pipe #2
Roughness (gpm) (gpm) Roughness
80 660 689 166
90 743 606 146
100 825 524 126
110 908 441 106
120 991 358 86
130 1073 276 66
140 1156 193 46
Clearly, there are multiple C-factor choices for Pipes 1 and 2 that will produce the same head loss
across the system. Although one set of C-factors may be correct for a particular case, the selection
may introduce error into the model for another case, a clue useful in identifying the presence of com-
pensating errors. The question then becomes, which is the correct set of C-factors? The flow in one of
the pipes must be measured to answer this question and establish the correct C-factors.
This problem illustrates compensating errors for a simple two-pipe system. Assume,
for the same system, that the flow into the system and the pressure at the upstream
node are both unknown. The various combinations of flow, pipe roughness values,
and upstream pressures that would match the single pressure measurement taken at
the downstream node are now essentially infinite. As more parallel paths from point A
to point B are added, the problem grows in complexity. This simple example illus-
trates that compensating errors are often hidden behind seemingly valid data. As the
problems get larger and hydraulic measurements become more sparse, they become
increasingly more difficult to find.
When velocities are low, it is possible to make a model appear to be calibrated even
when C-factors contain significant errors (Walski, 1986). The best way to reduce the
likelihood of compensating errors in pipe roughness values is to take head measure-
ments under a range of demand conditions. Because the head loss equations are non-
linear, it will be difficult for compensating errors to make the model look calibrated
when it is not.
Flow measurements provide another way to reduce the likelihood that the wrong
parameter is adjusted. For instance, in the preceding example, knowledge of the total
flow through the system eliminated one degree of freedom. Obviously, it is not practi-
cal to measure flows for each pipe in the field that corresponds to a pipe in the model.
Nevertheless, to minimize the potential for compensating errors and to aid in the cali-
bration process, as many flow measurements should be made as possible, particularly
at critical locations such as pipelines connected to treatment plants, pump stations,
tanks, reservoirs, and other water sources. Tests should also be conducted along major
transmission mains that carry a large portion of the total system flow and along distri-
bution lines that are considered to be representative of the overall system (Ormsbee
and Lingireddy, 1997).
Determining roughness coefficients for a representative sampling of pipes of varying
ages and sizes provides a good check on the reasonableness of the coefficients used in
calibration. Chapter 5 discusses the measurement of flow and roughness coefficients
in greater detail.
Distribution of System Demands

The water distribution modeling equations are based on the simplifying assumption
that water is withdrawn at a junction node. In reality, however, water usage occurs
along the entire length of a pipe, as shown in Figure 7.2. Spatially redistributing water
usages that occur along a length of pipe to the junction nodes in the model is known
as demand allocation. The demand allocation process is a possible source of error that
should be considered when calibrating a model. Another source of error, often more
significant with regard to demands, is related to how the demands change over time
(an important issue when performing an EPS). Both of these sources of error and their
impact on calibration are discussed in this section. In addition, Chapter 4 discusses
them more generally.
It is conceivable that a model could incorporate all of the locations where water is
withdrawn from the system by placing junction nodes where the service lines are con-
nected to the water main. This approach, however, would significantly increase the
number of pipes required in the model, thereby increasing its complexity. Model sim-
plification is achieved through spatial demand allocation (see the example on page
140). For example, in Figure 7.2, the sum of the water use associated with the eight
homes closest to J-23 can be assigned to J-23, and the sum of the water use for the 10
homes closest to J-24 can be assigned to J-24. By placing the demands properly, they
will be accounted for in the model even if the pipe between nodes J-23 and J-24 is
removed during skeletonization. However, the modeler must realize that the simulated
pressures at the model nodes are only an approximation of the actual pressures at the
homes.
Figure 7.2
Spatial demand
Demands Assigned Demands Assigned
To J-23 To J-24 allocation
Q J-23 QJ-24
Q Q Q Q Q Q Q Q Q
J-23 J-24
Q Q Q Q Q Q Q Q Q
Several expressions have been developed that equate uniform water withdrawal along
a pipe to point loads at junction nodes. In the early stages of modeling, a method was
developed for correcting head loss equations with multiple service lines when per-
forming manual calculations (Muss, 1960). Another method uses a stepwise combina-
tion of elements and a nonlinear representation of the entire network (Shamir and
Hamberg, 1988).
Grouping water usage at the junction nodes instead of at the actual locations where
water is withdrawn from the system produces relatively minor differences between
computer-predicted and actual field performance if the actual location of the customer
demand is in close proximity to the assigned node. Incorrect spatial demand alloca-
tion usually becomes problematic when demands from large customers are missed or
assigned to nodes in the wrong pressure zones. In most cases, however, errors in allo-
cating demands to exactly the right node are insignificant, especially when fire flows
used in design are significantly greater than normal demands.
When making comparisons between the model and field measurements, it is impor-
tant that the demands in the model correspond to the time that the field measurements
were taken. A common mistake is to compare model HGL for an average day demand
with a field HGL taken at an hour when the demand is actually larger.
Just as a modeler should be skeptical of needing to assign unrealistic pipe roughness
values to obtain a calibrated model, he or she should also be skeptical of needing to
use unrealistically high or low nodal demands to achieve calibration. If demand val-
ues that are significantly different from historical records are needed to calibrate the
model, then a logical explanation for this deviation should be provided. For example,
maybe the community swimming pool was being filled on the day pressures were
measured or a large water-using industry was temporarily shut down by a strike dur-
ing pressure testing. In conclusion, demands should be adjusted within reason to
match actual field conditions.
System Maps
As discussed in Chapter 3, water system maps (shown in Figure 7.3) are the primary
source of data for the physical characteristics of the distribution network. Network
topology, pipe/node connectivity, pipeline lengths, nominal diameters, and informa-
tion on fittings and appurtenances can all be determined from water system maps.
As new pipelines are installed and new connections are made, the maps of a system
should be updated to reflect the changes. The quality and format of water system
maps, however, can range from highly detailed and regularly updated CAD or GIS
systems, to sets of rolled-up plans that have not been maintained in years. In some
cases, the full extent of system mapping actually resides in the head of the system
caretaker. Regardless of the medium on which they are available, it is important to
realize that system maps do not necessarily reflect real-world conditions.
Figure 7.3
F O R D TU
Portion of as-built
system map
RNPIKE
HARTFOR
D TURNPIK
E
If the differences in pressures and flows between actual conditions and predicted con-
ditions are so great that unrealistic and unexplainable pipe roughness values (less than
30 or more than 150) or major adjustments in demands must be used to achieve cali-
bration, then chances are good that the discrepancy is the result of a closed or partially
closed valve or errors in system mapping. For example, suppose that during calibra-
tion the observed pressure at a location is consistently about 20 psi (138 kPa) higher
than simulated pressure, regardless of the pipe roughness values used. This result is an
indication that there are problems with the model. A pipeline in the service area
where the pressure measurement was taken may not be included in the model, or a
junction elevation may be incorrect.
If errors in the connectivity of the model are suspected, then it may be necessary to
look at detailed intersection maps to determine how pipes are connected, or to talk
with system operators and maintenance personnel to determine the location and status
of valves in the system. It may even be necessary to locate the original as-built draw-
ings or field construction notes.
Temporal Boundary Condition Changes

The effect of time can have a significant impact on calibration efforts because many
of the parameters describing a water system, such as demands and boundary condi-
tions, are time-dependent. As was the case with demands, synchronizing times at
which field measurements are taken and the calculation time step used for simulations
will improve the accuracy of the calibration.
When a simulation is conducted for the purpose of calibration, it is critical that the
model’s loading and boundary conditions reflect the actual conditions at the time that
pressure measurements were taken, and that the boundary condition measurements
are known with the same accuracy as the pressure measurements. Results from cali-
bration will be misleading if the boundary heads are not known exactly.
For example, consider that on a particular day, pressure measurements were taken
throughout the system at 6:00 a.m., 10:00 a.m., 12:00 p.m., 3:00 p.m., and 7:00 p.m.
Because the demands, tank levels, control valve settings, and pump and pipe statuses
can change over time, a unique set of boundary conditions reflecting system condi-
tions for each point in time that measurements are taken needs to be created. For this
example, five separate steady-state simulations must be performed, with each set of
system conditions corresponding to a time when pressures were measured.
Information on how tank water levels change over time is often collected from chart
recorders (refer to Figure 4.1) or a SCADA system (see Figure 7.4). This type of
information is frequently used for steady-state calibration and is particularly useful
for extended period simulation (EPS) calibration (see page 279). Based on data from
these sources, the flow rate to or from the tank can be determined using Equation 7.1.
Hi + 1 – Hi
Q i = A ------------------------- (7.1)
't
where Qi = flow into tank in i-th time step (cfs, m3/s)

A = cross-sectional area of tank (ft2, m2)
Hi = water level in tank at beginning of i-th time step (ft, m)
't = length of i-th time step (s)
When conducting fire flow tests for collecting calibration data, it is highly desirable to
have someone actually recording suction and discharge pressures at pumps and inlet
and outlet pressures at pressure reducing valves (PRVs). In this way, it is possible to
determine if the pressure drop is due to pipe roughness, pumps moving along the
pump curves, or head drop through a PRV. Relying on the SCADA system for these
data may not be accurate because the system may not have polled the pump.
Figure 7.4
SCADA data showing
tank water levels
Model Skeletonization
When a computer model of an existing system is constructed, a skeletonized version
of the system will often be analyzed. As discussed in Chapter 3 (see page 114), a skel-
etonized model may remove certain types of fittings and appurtenances, and typically
does not include small-diameter pipes nor those lines that do not have a significant
influence on the system hydraulics. Ideally, a skeletonized model should provide a
simplified but accurate representation of the system. Accordingly, it is very important
that the integrity of the network connectivity (or topology) be maintained during skel-
etonization.
The level to which a model is skeletonized can have a significant impact on calibra-
tion. It is possible to over-skeletonize a model, leaving out critical links in the system.
In these cases, details that have been removed may need to be added back to the skel-
etonized model to improve accuracy, especially in the vicinity of fire flow tests. Con-
sider a system that includes a dense grid of small-diameter mains. Excluding those
mains from a model based solely on diameter may be inappropriate if, as a group,
they have a significant hydraulic impact on the system. This specific type of condition
can be identified by the unrealistically large C-factors required in the remaining pipes
to achieve calibration, especially during high flows.
Geometric Anomalies
Even if the modeler supplies high-quality information on the physical attributes of the
system and provides good estimates of nodal demands, the degree of calibration still
may not be satisfactory. In these cases, anomalies in the geometry of the system are
usually to blame.
Section 7.3 Calibration Approaches 263
Placing a node at the intersection of two pipes in the model when they are not hydrau-
lically connected would obviously have the potential to cause problems with calibra-
tion because the model would not match the actual system. The modeler should pay
particular attention to this type of situation when extracting data from CAD and GIS
systems (see page 84 for additional information).
Pump Characteristic Curves

Recall that hydraulic simulation models require data concerning the pump head ver-
sus discharge relationship. Generally, these models use some type of interpolation
routine that fits a curve through selected points from the manufacturer’s pump head
characteristic curve. Because a curve-fitting method is used, the true head and dis-
charge of the pump may differ somewhat from the curve, and error is introduced into
the model. Numerical curve-fitting errors can be identified by comparing the manu-
facturer’s curve with the curve produced by the hydraulic model.
A more likely cause of error when modeling pumps results from the use of old or out-
dated pump curves. For instance, suppose that you are modeling a system that uses
25-year-old centrifugal pumps, but the head versus discharge relationship shown on
the manufacturer’s pump curves reflects the performance of the pump when it was
new. Normal wear and tear on a pump as it ages can cause the field performance to
deviate from the performance illustrated on the pump characteristic curve. In fact, the
pump impellers may have been changed several times since the pumps were
originally installed. If so, the original pump curves will have little value because the
head/discharge relationship of a pump is dependent on the characteristics of the pump
impeller. In such cases, new curves should be determined based on field tests.
Hydraulic network model calibration involves more than just adjusting pipe rough-
ness values and nodal demands until suitable simulation results are obtained. Model
calibration can involve a significant amount of detective work (such as locating closed
and partially closed valves) as clues are tracked down and errors between field and
simulation results are investigated (Walski, 1990). Some leads may yield results, and
others may not.
7.3 CALIBRATION APPROACHES

Identifying and addressing large discrepancies between predicted and observed
behavior is critical in the calibration effort. This step, referred to as rough-tuning or
macrocalibration, is necessary to bring predicted and observed system parameters
into closer agreement with one another. After larger discrepancies are corrected,
efforts can be focused on fine-tuning or microcalibration. Fine-tuning involves adjust-
ing the pipe roughness values and nodal demand estimates, and is the final step in the
calibration process.
The most challenging part of calibrating a model is making judgments regarding the
adjustments that must be made to the model to bring it into agreement with field
results. This section introduces methods for making these calibration judgments.
The following is a seven-step approach that can be used as a guide to model calibra-
tion (Ormsbee and Lingireddy, 1997).
1. Identify the intended use of the model.

2. Determine estimates of model parameters.
3. Collect calibration data.
4. Evaluate model results based on initial estimates of model parameters.
5. Perform a rough-tuning or macrocalibration analysis.
6. Perform a sensitivity analysis.
7. Perform a fine-tuning or microcalibration analysis.
Identifying the intended use of the model is the first and most important step because
it helps the designer establish the level of detail needed in the model, the nature of the
data collection, and the acceptable level of tolerance for errors between field measure-
ments and simulation results.
After the intended use of the model is established, the modeler can begin estimating
model parameters and collecting calibration data as discussed in the previous sec-
tions. The model can then be evaluated, and large discrepancies can be addressed sim-
ply by looking at the nature and location of differences between the model results and
the field data.
Next, a sensitivity analysis can be conducted to judge how performance of the calibra-
tion changes with respect to parameter adjustments. For example, if pipe roughness
values are globally adjusted by 10 percent, the modeler may notice that pressures do
not change much in the system, thus indicating that the system is insensitive to rough-
ness for that demand pattern. Alternatively, nodal demands may be changed by 15
percent for the same system, causing pressures and flows to change significantly. In
this case, time may be more wisely spent focusing on establishing good estimates of
system demands. If neither roughness coefficients nor demands have a significant
impact on system heads, then the velocity in the system may be too low for the data to
be useful for this purpose.
The final step in the calibration process, fine-tuning, can be time-consuming, particu-
larly if there are a large number of pipes or nodes that are candidates for adjustment.
Compensating errors, as discussed on page 256, can further complicate the fine-
tuning stage.
Manual Calibration Approaches

The trial-and-error or manual process generally involves the modeler’s supplying esti-
mates of pipe roughness values and nodal demands, conducting the simulation, and
comparing predicted performance to observed performance. If the agreement is unac-
ceptable, then a hypothesis explaining the cause of the problem should be developed,
modifications made to the model, and the process repeated.
The process is conducted iteratively until a satisfactory match is obtained between

modeled and observed values. If no satisfactory match can be obtained, then the
model is not a true representation of the part of the real system where discrepancies
remain. In such cases, further site investigations are usually made to identify discrep-
ancies between the model and the real system, such as incorrectly modeled valve set-
tings and unrecorded connections. The overlay of computed values on a contour map
can provide insight into this process.
Models can be calibrated using one steady-state simulation, but the more steady-state
simulations for which calibration is achieved, the more closely the model will repre-
sent the behavior of the real system. At a minimum, a steady-state calibration should
be performed for a range of demand conditions. To improve results further, the model
should be calibrated for time-varying conditions using an extended period simulation.
In an EPS, calibration is performed until there is a reasonable agreement between
modeled and observed pressures, flows, and tank water levels. EPS calibration is dis-
cussed later in this chapter on page 279.
What Should Be Adjusted. Depending on the flow conditions being simu-
lated, the model will have different reactions to different types of data changes. The
following provides some general guidelines.
• Average and low flows: For most water distribution systems, the HGL
throughout the system (also referred to as the piezometric surface) is fairly
flat during average-day demand conditions. The reason for these small head
losses is that most systems are designed to operate at an acceptable level of
service during maximum day demands while accommodating fire flows. As
a result, the pipe sizes are usually large enough that average-day head losses
are small. For this reason, calibration during average conditions does not
provide much information on roughness coefficients and water use. Average
conditions do, however, provide insights into boundary conditions and node
elevations.
• High flows: During periods when flows through the system are high, such as
fire flow conditions or peak hour flows, pipe roughness and demand values
play a much larger role in determining system-wide pressures. Therefore,
pipe roughness values, and to a lesser extent demands, should be adjusted
during periods of high flow to achieve model calibration.
Relatively speaking, when pipe flow or roughness is greater, there will be more head
loss. Based on this relationship, the following are some recommendations for making
adjustments to models (Herrin, 1997).
• If the model HGLs are higher than field-recorded values (as shown in Figure
7.5), then the model is not predicting enough head loss. To produce larger
head losses, try reducing the Hazen-Williams C-factor, increasing the junc-
tion demands in the area of the measurements, or both.
• If the model HGLs are lower than field-recorded values (see Figure 7.6),
then the model is probably predicting too much head loss. To produce
smaller head losses, try increasing the Hazen-Williams C-factor and/or
decreasing the junction demands in the area of the measurements.
Figure 7.5
Model HGLs are
higher than field
values
Model
Actual
//=//= //=//=
//=
//=
//=
//=
Figure 7.6
Model HGLs are
lower than field values
Actual
Model
//=//= //=//=
//=
//=
//=
//=
Occasionally, an agreement in head is obtained for all but one node in the model. In
that case, the elevation for that location should be questioned and verified. During
high-flow conditions, the system’s increased head loss can mask pressure discrepan-
cies caused by poor elevation data. Thus, inaccurate elevations are more easily identi-
fied during low-flow conditions, when the system’s head losses are smaller.
Adjusting Roughness Coefficients. In trying to determine whether to adjust
roughness or water use, the following procedure can be helpful (Walski, 1983). Using
the pressure and flow results from a fire flow test, the modeler can simulate the
hydrant flow test with the model and develop estimates of head loss during the static
and test conditions. The user can then calculate correction factors A and B as follows:
F
A = ------------------------------------------------
- (7.2)
b e a Qe + F – Qe
F
B = ---------------------------------------- (7.3)
b Q e + F – aQ e
where A = correction factor

B = correction factor
F = fire flow (gpm, m3/s)
h 0.54
b = §© ----2-·¹
h4
h 0.54
a = §© ----1-·¹
h3
Qe = estimate of demand in area of test (gpm, m3/s)
h1 = measured head loss over test section, static conditions (ft, m)
h2 = measured head loss over test section, flowed conditions (ft, m)
h3 = modeled head loss over test section, static conditions (ft, m)
h4 = modeled head loss over test section, flowed conditions (ft, m)
The correction factors are then applied to the estimated Hazen-Williams C-factor and
water use to develop better estimates.
Q c = AQ e (7.4)
C c = BC e (7.5)
where Qc = corrected value for demands (gpm, m3/s)

Cc = corrected value for C-factors
Ce = initial estimated value for C-factors
Note that the above equations are true for any units as long as the flows and heads are
kept in consistent units.
Example — Corrected Demand and Roughness. Given that the head at the nearest
tank is 970 ft, the demands in the vicinity of the fire flow test are 200 gpm, the test flow is 750 gpm,
and the C-factors in the vicinity of the test are estimated as 85, find the corrected values for demand
and C-factor based on the fire flow test observation shown in the following table.
Field Test HGL Model-Predicted HGL

(ft) (ft)
Static condition 962 958
Fire flow test 927 910
Computing the correction factors introduced above results in:
970 – 962 0.54

a = §© ------------------------·¹ = 0.80
970 – 958
970 – 927 0.54

b = § ------------------------· = 0.83
© 970 – 910¹
750
A = -------------------------------------------------------------------------
- = 0.95
0.83 e 0.80 200 + 750 – 200
750
B = ---------------------------------------------------------------------- = 1.20
0.83 200 + 750 – 0.80 200
Q c = 0.95 200 = 190
C c = 1.20 85 = 101
Accordingly, the user would decrease the demands to 190 gpm and increase the C-factors to 101 for
the next model run. Evaluating the previous table, these computed adjustments are consistent with
what would be intuitively expected.
Automated Calibration Approaches

Traditionally, model calibration has been a manual task where the modeler makes
changes to pipe roughness values or demands on a trial-and-error basis to achieve
convergence between model and field values. Because many potential combinations
of calibration parameters exist, finding the best set of parameters presents a challenge
to the engineer. Therefore, the modeler can calibrate the system much more efficiently
and consistently by using a computer-based, numerical optimization technique that is
able to identify the optimal or near-optimal combination of calibration parameters to
achieve as close a match as possible to the field data.
Table 7.2 provides an overview of the many calibration approaches for water distribu-
tion network models that have been developed since the 1970s (Kapelan, Savic, and
Walters, 2000). Generally, calibration approaches can be grouped into three catego-
ries:
• Iterative procedure models

• Explicit models (or hydraulic simulation models)
• Implicit models (or optimization models)
The first group of models is based on some specifically developed, iterative, trial-and-
error procedures (shown as IP in Table 7.2). In these calibration procedures, unknown
parameters are updated at each trial, or iteration, using heads and/or flows obtained
by running a simulation model. Without discussing the details of these models, the
following can be observed about them:
• Simplifying the water distribution model (by skeletonization, for example) is

typically necessary (see page 114).
• Only small calibration problems (problems that have a small number of cali-
bration parameters) can be effectively handled.
• Convergence rate of the iterative models is rather slow (Bhave, 1988).
The main benefit of having developed these iterative procedures is that fundamental
principles and guidelines regarding water distribution model calibration were estab-
lished as a result (Ormsbee, 1989, and Walski, 1995). These principles were used to
develop more sophisticated explicit and implicit calibration approaches.
Table 7.2 Water distribution calibration models

Hydraulic
Model Model2 Decision Optimization Objective
No. Model Reference
Type1 (number Variables3 Method Function4 (OF)
of LC)
1 Rahal, Sterling, and IP SS(1) RC - -
Coulbeck (1980)
2 Walski (1983), IP SS(2) FC - -
Walski (1986) DEM
3 Bhave (1988) IP SS(2) FC - -
DEM
4 Ormsbee and Wood EX SS(M) FC - -
(1986)
5 Ormsbee and IM SS(M) FC Extended WSAE(NS)
Lingireddy (1997) or DEM complex
EPS method of
Box
6 Boulos and Wood EX SS(1) Any - -
(1990), Boulos and parameter
Wood (1991) excluding
state
variables
7 Boulos and Ormsbee EX SS(M) Any - -
(1991) parameter
excluding
state
variables
8 Lansey and Basnet IM SS(M) FC Gradient- WSSE(H,Q,T)
(1991) or DEM based
EPS VS GRG2
9 Datta and Sridharan IM SS(M) FC Sensitivity WSSE(H,RD)
(1994) Analysis
Technique
10 Ferreri, Napoli, and EX SS(1) FC - -
Tumbiolo (1994)
11 Savic and Walters IM SS(M) FC GAs WSSE(H,Q)
(1995)
12 Reddy, Sridharan, and IM SS(M) RC Gauss- WSSE
Rao (1996) DEM Newton (H,h,Q,D)
13 Walters, Savic, Mor- IM SS(M) RC GAs WSSE(H,Q)
ley, de Schaetzen, and
Atkinson (1998)
1) IP – iterative procedure (trial and error), IM – implicit procedure, EX – explicit procedure
2) SS – steady-state, EPS – extended period simulation, TS – transient simulation, 1 – Single LC, 2 – two LC, M – multiple LC,
LC – loading condition, BC – boundary condition, TBC – tank BC, VBC – valve BC, PBC – pump BC
3) FC – friction coefficient, DEM – nodal demand, RC – generalized pipe resistance coefficient, VS – valve setting, PS – pipe status, LLC – lumped leak coefficient,
AV – acoustic velocity
4) SSE – sum of squared errors, WSSE – weighted SSE, ASSE – average SSE, SAE – sum of absolute errors, WSAE – weighted SAE, ASAE – average SAE, MMD –
minimize maximum error, WMME – weighted MME (H – nodal head error, h – link head loss error, Q – link flow error, T – tank levels error, D – nodal demand error,
RD – reservoir demand error, FC – friction coefficient error)
Table 7.2 (cont.) Water distribution calibration models

Hydraulic
Model Model2 Decision Optimization Objective
No. Model Reference
Type1 (number Variables3 Method Function4 (OF)
of LC)
14 Greco and Del IM SS(M) FC LINDO SSE(FC)
Guidice (1999) GINO (subject to
limited H)
15 Todini (1999) IM SS(M) FC Kalman filter SSE(H,D)
16 Pudar and Liggett IM SS(1) LLC Levenberg- WSSE(H)
(1992) Marquardt
17 Liggett and Chen IM SS(1) FC Levenberg- SSE(H)
(1994) LLC Marquardt
18 Chen (1995) IM TS FC Levenberg- SSE(H)
LLC Marquardt
AV
19 Vitkovsky and IM TS FC GA SAE(H)
Simpson (1997), Sim- LLC
pson and
Vitkovsky (1997),
Vikovsky, Simpson,
and Lambert (2000)
20 Tang, Karney, Pendle- IM TS FC GA Not specified
bury, and Zhang DEM
(1999)
21 Wu, Boulos, Orr, and IM SS(1) FC GAs ASSE(H)
Ro (2000) ASAE(H)
MMD(H)
22 Wu et al. (2002a), Wu IM, IP SS(M) FC DEM fmGA WSSE(H,Q)
et al. (2002b) TBC(M) VS PS WSAE(H,Q)
VBC(M) WMME(H,Q)
PBC(M)
1) IP – iterative procedure (trial and error), IM – implicit procedure, EX – explicit procedure
2) SS – steady-state, EPS – extended period simulation, TS – transient simulation, 1 – Single LC, 2 – two LC, M – multiple LC,
LC – loading condition, BC – boundary condition, TBC – tank BC, VBC – valve BC, PBC – pump BC
3) FC – friction coefficient, DEM – nodal demand, RC – generalized pipe resistance coefficient, VS – valve setting, PS – pipe status, LLC – lumped leak coefficient,
AV – acoustic velocity
4) SSE – sum of squared errors, WSSE – weighted SSE, ASSE – average SSE, SAE – sum of absolute errors, WSAE – weighted SAE, ASAE – average SAE, MMD –
minimize maximum error, WMME – weighted MME (H – nodal head error, h – link head loss error, Q – link flow error, T – tank levels error, D – nodal demand error,
RD – reservoir demand error, FC – friction coefficient error)
The second group of calibration models, called explicit models, is based on solving an
extended set of steady-state, mass-balance, and energy equations (shown as EX in
Table 7.2). These models are essentially pipe network models that solve for roughness
or demand as well as pressure and flow. Shamir and Howard advanced the first
explicit solutions during the 1960s while at MIT (Shamir and Howard, 1968). The
extended set consists of the initial set of equations (those normally used in network
simulation models) augmented by a set of equations derived from available head and
flow measurements (one additional equation per measurement), and it is solved
numerically. The number of unknown calibration parameters is limited by the number
of available measurements, so when the number of unknown calibration parameters is
larger than the number of available measurements (underdetermined problem), the

number of calibration parameters must be reduced by grouping (see page 274).
Explicit calibration methods have several disadvantages and limitations (Kapelan,
Savic, and Walters, 2000):
• The calibration problem must be even-determined; that is, the number of cal-
ibration parameters must be equal to the number of measurements.
• Measurement errors are not taken into account; it is assumed that measured
heads/flows are completely accurate.
• It is difficult to quantify the uncertainty of the estimated calibration parameters.
The third group of calibration models, implicit models, consists of optimization-based

models (see Table 7.2 for a list of references). In this case, the calibration problem is
represented as an optimization problem by introducing an objective function. The
problem is solved implicitly, usually by minimizing the objective function. Three
commonly used types of objective functions are (1) sum of squared errors, (2) sum of
absolute errors, and (3) maximum absolute error. Errors (residuals) are calculated as
differences between measured (observed) and output variables computed by the
hydraulic model. Head and flow errors are typically used, although other types of
errors may be used as well, such as tank level, head loss, or chlorine residual errors.
Hydraulic models linked to optimization methods are steady-state models (single- or
multiple-loading condition), extended-period simulation models, or unsteady (tran-
sient) models.
Optimization Problem Formulation. The optimization methods search for

a solution describing the unknown calibration parameters that minimizes an objective
function, while simultaneously satisfying constraints that describe the feasible solu-
tion region. The objective function usually minimizes the sum of the squares of differ-
ences between observed and model-predicted heads and flows. If the vector of those
unknown parameters is given as x (roughness, demand, control status), the objective
function may be given as
N
* 2
min f x = ¦ wi > yi – yi x @ (7.6)
x
i=1
where f = objective function to be minimized

x = vector of unknowns
N = number of observations
wi = weighting factors
*
y i = observation (head, flow)
yi(x) = model predicted system variable (head, flow)
As an example, the y vector of observations and predictions would consist of a set of
values such as “517 m, 34 l/s, and 510 m” where those values would be the measured
head at node J-11, the flow in pipe P-131, and the head on the discharge side of PMP-
4, respectively. The x vector of unknowns would consist of values such as “121, 98, 5
l/s, and open” where those values would be the C-factors at pipes P-22 and P-23, the
demand at node J-14, and the status of pipe P-224, respectively. The values for x will
vary from one iteration to another, but the values for y are constant for a given run.
Weightings are applied to reduce the influence of observations that are less accurate,
to increase the influence of observations that are more accurate, and to enforce unit
consistency in the working equation.
In vector notation, the preceding objective function becomes
* T *
min f x = > y – y x @ W > y – y x @ (7.7)
x
where y* = vector of observations (head, flow)

y(x) = vector of model predicted values (head, flow)
T = transpose operator
W = weighting matrix
The set of constraints associated with this problem are implicit hydraulic constraints
(continuity and energy loss relationships), known initial conditions (device statuses
and tank levels), and boundary conditions (reservoir levels). Rather than explicitly
incorporating the equations of conservation of mass and energy into the optimization
routine, later approaches have simply called out to a standard hydraulic simulation
program to evaluate the hydraulics of the solution (Ormsbee, 1989; and Lansey and
Basnet, 1991). Then the solution is passed back to the optimization routine, where the
algorithm computes the objective function, evaluates the constraints, and, if neces-
sary, updates the decision variables. New values of the decision variables are then
passed to the simulation routine, and the process is repeated until an acceptable cali-
bration is obtained.
The stochastic search procedures, more commonly referred to as genetic algorithms
(GAs), also work by closely coupling an optimization routine with the hydraulic
solver (see page 673 in Appendix D for more information). GA optimization, based
on the theory of genetics, works by generating successive populations of trial solu-
tions, the “fittest” of which survive to breed and evolve into increasingly desirable
offspring solutions (Savic and Walters, 1995, 1997; and Walters, Savic, Morley, de
Schaetzen, and Atkinson, 1998).
Genetic algorithms work by evaluating the fitness of each potential solution consist-
ing of values for the set of unknown network calibration parameters. Fitness is deter-
mined by comparing how well the simulated flows and pressures resulting from the
candidate solution match the measured values collected in the field. Several steady-
state simulations are run to simulate a variety of demand conditions, including the
operating conditions for minimum, maximum, and average demands. At each mea-
surement point and for each steady-state run, the differences between simulated and
observed data (head and/or flow) are calculated, and the objective function (an overall
error value for the network) is computed. Objective functions can be formulated in
many different ways to achieve different goals. Usually, a squared error or root mean
square error criterion is adopted. Different weightings between head and flow mea-
surements can also be incorporated within the objective function. The GA continues
to spawn generations of potential solutions until comparison of solutions from succes-
sive generations no longer produces a significant improvement.
In addition to eliminating most of the routine and tedious aspects of the calibration
process, GA will generally achieve a better fit to the available data if the user can
select the correct set of variables to be included in the solution and can establish the
correct range of possible solutions.
Issues with Calibration. The modeler should assess uncertainty in field obser-
vations before using the observations in any calibration, even when using optimiza-
tion. It is not uncommon for errors in measurement of head loss to be on the same
order of magnitude or larger than the actual head loss (Walski, 2000). Such values
should not be used in calibration because the calibration algorithm will dutifully try to
match the field observations even if they are erroneous (see page 218).
To ensure that head loss adequately exceeds measurement error, it is helpful to collect
data when pipe velocities are appreciable. In some systems sized for fire protection,
demands (and velocities and head losses) are so low most of the time that head loss
measurements are meaningless other than to check pressure gage elevations. This
leads to a typical case in which calibrated parameters are uncertain and can result in
uncertain model predictions. The calibration parameters need additional observed
information to be determined more accurately. However, the necessary information is
only obtained if field-testing is done when the demand is significantly higher than that
normally observed in the system.
Another problem that occurs when calibrating a model is that some of the parameters
determined, such as roughness and valve status, are fixed and knowable at the time the
data are taken, and others, such as water use, are merely random observations from a
stochastic process. If a C-factor is determined to be 90, then that value will be true in
the not-too-distant future. However, if water use during a pressure observation is
determined to be 100 gpm (6.3 l/s), it is not necessarily the demand that should be
used in calibration.
A problem common to all calibration approaches (even trial-and-error ones) deals
with identifiability, which means that different vectors of calibration parameters x
may lead to (almost) the same vector of model predictions y(x) that are close to field
observations y* (Kapelan, Savic, and Walters, 2001). The problem of identifiability
occurs when an underdetermined calibration problem is being solved. An undeter-
mined calibration problem is when the number of calibration parameters x is larger
than the total number of independent observations. In such a case, there are many
combinations of input parameters (for example, C-factors) that result in good agree-
ment between observed and modeled behavior (for example, pressures and flows) of
the system. Because of this fact, little confidence can be placed in the calibrated
model.
Similar difficulties may occur even for even-determined or over-determined problems
(that is, when there are at least as many observations as calibration parameters). Diffi-
culties occur because the set of available observations simply fails to provide suffi-
cient information for determination of one or more calibration parameters (for
example, pressure monitoring points may not be properly located to enable identifica-
tion of all or some of the parameters). Calibrated parameters are either insensitive to
field test data or their values are quite uncertain, though they may appear to be pre-
cisely determined by the calibration procedure.
Problems associated with identifiability can be overcome by grouping unknown
parameters (for example, pipe roughness coefficients for all pipes that share the same
material, diameter, age, and location), or by increasing the quantity of observed infor-
mation through additional field measurements. Grouping is based on the assumption
that pipes laid in roughly the same time period with the same material will have the
same roughness properties. Grouping greatly reduces the identifiability problem, but
it may introduce errors if the pipes and nodes in a given group should not have the
same adjustments applied. Grouping and collecting additional observed information
are not always possible. Kapelan, Savic, and Walters (2001) introduced another
approach to improving the identifiability based on prior estimates on parameters.
The phrase prior estimates on parameters refers to information that can be directly or
indirectly obtained about a calibration parameter before beginning the calibration
adjustments. Possible sources of prior estimates on parameters are data from existing,
typically isolated, field tests/measurements/inspections. (Note that data collected in
these tests should be independent of the data collected during field tests conducted to
measure and record heads and flows for model calibration.) Additional sources of
prior estimates include data resulting from specific analyses such as prior estimates of
demands based on demand allocation analysis, data from engineering knowledge/lit-
erature (for example, hydraulic tables for pipe roughness coefficients as a function of
pipe material, age, diameter, condition, and so on), results of C-factor tests or pump
curve tests, and experts’ knowledge and experience. This approach is not conceptu-
ally unusual since engineers have traditionally used this additional knowledge when
manually calibrating models. However, the development of a framework to utilize this
information in the optimization process is something that can significantly improve
the chances of identifying the appropriate set of calibration parameters through opti-
mization.
In order to introduce prior estimates on parameters to the optimization problem, the
range of adjustments for parameter values can be constrained to a narrow band. In a
more rigorous approach, the calibration objective function needs to be augmented by
adding the sum of weighted-squared prior estimate residuals to the classic sum of
weighted-squared observed information residuals:
* T * * T *
min f x = > y – y x @ W > y – y x @ + > x o – x o @ V > x o – x o @ (7.8)
x
*
where x o = vector of prior parameter estimates (pseudo measurements)
xo = vector of actual parameter values
V = weighting matrix
The principal difference between observed heads and flows and prior estimates is that,
in the vast majority of cases, observed information is more reliable. Starting from this
fundamental fact, a procedure for effective incorporation of prior estimates on param-
eters to the calibration of water distribution models is suggested by Kapelan, Savic,
and Walters (2001). The basic steps are as follows:
1. Solve the optimization problem with observed data only (that is, without use of
prior estimates).
2. Evaluate optimized parameter values for sensitivity and the uncertainty with
which each parameter value was determined.
3. Identify insensitive or uncertain parameters. These parameters need extra
observed information in order to be determined more accurately and therefore are
possible candidates for the use of prior estimates. It is not advisable to use prior
estimates for parameters that are considered sensitive yet possess unreasonable
values. The existence of such parameters usually indicates that there is something
wrong with either the model (a valve is modeled as completely open although it
is partially closed, for example) or observed information (incorrect or insufficient
data).
4. In order to reduce uncertainty and improve values of previously identified insen-
sitive model parameters, one can either collect additional field data, use prior
estimates, or both. The optimization problem needs to be solved again — this
time with incorporated prior estimates on parameters. Prior estimates should be
used carefully; weights should reflect the level of confidence in each prior esti-
mate. The best strategy is to start with small weights and increase them gradually.
Also, instead of fixing a parameter value in the optimization model, it is often
useful to define that parameter as an additional decision variable (see page 644)
with associated prior estimates and their weights.
Uncertainties in computed calibration parameter values will typically improve as a
result of incorporated prior estimates. However, the use of prior estimates on parame-
ters may not always lead to computed parameter values that are closer to the true val-
ues. Improvement will depend primarily on quality and quantity of prior estimates.
An erroneous prior estimate will lead to erroneous results.
Sampling Design for Calibration. Data collection plays an important role in
managing water distribution systems. The main aim of the field data collection plan-
ning exercise is to determine what, when, under what conditions, and where to
observe the behavior of the system and collect data that, when used for calibration,
will yield the best results. This is what is known as a sampling design problem. The
answers to what, when, and under what conditions are usually known and are
described in Chapter 5, but the last question, where to locate measuring devices, has
been the subject of numerous research studies. Walski (1983) suggests that pressure-
measuring devices should be located near points of high demand, near the perimeter
of the skeletonized network, and generally distant from water sources. Multiple fire-
flow tests should be performed with fire flows that are as large as practical at test
hydrants, and both head and flow measurement data should be collected.
The relationship between calibration and sampling location selection is a typical
chicken-and-egg relationship. To calibrate a model, one needs field test data; that is,
sampling locations need to be defined. On the other hand, to judge the success of cal-
ibration, one would like to calculate sensitivities for all potential measurement loca-
tions with respect to all possible calibration parameters. To calculate those
sensitivities, the number, structure, and value of calibration parameters should be
known. However, parameters can be obtained only if the calibration problem is
solved. Therefore, there is a difficulty in that to solve the calibration problem, the
sampling design problem must be solved first, but, to solve the sampling design prob-
lem, the calibration problem must be solved beforehand. An obvious way around this
difficulty is to perform sampling design and calibration in iterations.
Many research studies have been done to devise optimization procedures that auto-
mate the process of selecting locations for data collection (Lee and Deininger, 1992;
Yu and Powell, 1994; Ferreri, Napoli, and Tumbiolo, 1994; Bush and Uber, 1998;
Piller, Bremond, and Morel, 1999; Ahmed, Lansey, and Araujo, 1999; de Schaetzen,
Randall-Smith, Savic, and Walters, 1999; de Schaetzen, 2000; Meier and Barkdoll,
2000; and Lansey, El-Shorbagy, Ahmed, Araujo, and Haan, 2001). One common
characteristic of these studies is that a single criterion (such as minimization of the
uncertainty of the model’s predictions or maximization of the coverage) is considered.
Kapelan, Savic, and Walters (2001a), however, formulated sampling design as a mul-
tiobjective optimization problem with relevant constraints. Two main objectives were
identified as (1) maximization of the accuracy of calibration parameter estimates or
minimization of the model prediction uncertainties and (2) minimization of total sam-
pling design costs.
It is important to note that optimization may not be the best tool for determining sam-
pling locations for several reasons:
• Many factors that are difficult to quantify with the precision needed in opti-
mization are involved.
• The objectives are very different if one is determining permanent sampling
locations versus locations to be used for one-time sampling.
• The criteria for hydraulic monitoring are different from the criteria for water
quality monitoring.
• Locations used to collect data for EPS calibration need to be locations with
dynamic behavior of the parameter or interest.
Because of these obstacles, Walski (2002) proposed an approach for sampling design
that relies on thematic mapping with a GIS.
Using Optimized Calibration. Regardless of the type of algorithm used, sim-
ilar steps are involved in all calibration optimization methods.
1. Given an uncalibrated model, the user makes runs to ensure that the results are
reasonable and that there are no gross errors.
2. Field data are collected in accordance with the accuracy criteria described in
Chapter 5 and all boundary conditions known.
3. The field observations are entered into the optimization model, and the parame-
ters to be adjusted are identified and grouped to the extent possible. Limits for the
parameter values should be set rather broadly at first. For example, if the C-factor
is expected to be on the order of 80, a range of 40 to 120 may be tried initially by
increments of 10 (that is, possible values are 40, 50, 60, 70, 80, 90, 100, 110, and
120).
4. The user identifies the objective function to be used (for example, minimize least
squares, minimize absolute value of differences) and assigns any weighting
between different objectives or field measurements. The user has some control
over the way that the optimization is run. For example, in GA calibration, the GA
tries a large number of solutions until it can’t make improvements or until it
reaches a preset maximum number of steps. (The user can specify the maximum
number of trials the GA can perform or the number of trials without improve-
ment.) Because the GA can run for a very long period of time, it is best to start
with a relatively small (for GA) number of trials (say 10,000) until good results
are achieved. As one nears a final solution, it may be possible to use a higher
number of possible trials to ensure that the GA gets as close to the best solution
as possible.
5. The user runs the calibration and reviews the results. These results should be
fairly reasonable. If the results are not reasonable, the user should check whether
the right parameters are being adjusted, the data were sufficiently accurate, or
enough data were supplied. If the parameters are at the upper or lower limit of
their ranges, it may be an indication that the range needs to be increased.
6. The user should repeat the optimization until the results are reasonable. When
reasonable results are obtained, the user may want to narrow the search range to
obtain more precision. For example, if the optimization determines the C-factor
to be 90, but the increment in the initial calibration is 10, the user may want to
make a repeat run with possible values of 80, 85, 90, 95, and 100 to more pre-
cisely define the C-factor. The modeler may also want to fix the values of some
parameters once they are known reasonably well and divide some groups into
two or more groups.
7. After an acceptable set of roughness and demand adjustments have been deter-
mined, the user can transfer those results to the model. The values determined
correspond to the roughness and demands at the time the data were collected. In
particular in the case of demands, which change significantly over time, the user
then needs to determine the extent to which those calibrated values need to be
adjusted to represent average day, max hour, or some other demands. The overall
calibration process is summarized in Figure 7.7.
Model Validation
After a model is calibrated to match a given set of test data, the modeler can gain con-
fidence in the model and/or identify its shortcomings by validating it with test data
obtained under different conditions. In performing validation, system demands, initial
conditions, and operational rules are adjusted to match the conditions at the time the
test data were collected. For example, a model that was calibrated for a peak day may
be validated by its capability to accurately predict average day conditions as well.
Although it is desirable to validate every model, most utilities do not have the time or
money required to perform a thorough verification of the entire system. Consequently,
a modeler may want to perform a quick validation before applying the model to a new
problem. For example, before a three-year-old model is applied to the study of a pro-
posed water main on the east side of town, the utility may want to conduct a handful
Section 7.4 EPS Model Calibration 279
of fire flow tests or place some pressure recorders in the study area for use in validat-
ing the model in that portion of the system.
Figure 7.7
The calibration
Identify Collect optimization process
Construct Parameters Field
Model to Adjust Data
Identify
Identify Screen
Objective
Grouping Data
Function
Set
Optimization Set Ranges
Controls
Run
Optimization
Repeat Apply to
Until Model
Satisfied
7.4 EPS MODEL CALIBRATION

Before beginning the calibration of an EPS model, the user needs to be confident that
the steady-state model is calibrated correctly in terms of elevation, spatial demand
distribution, and pipe roughness. Once calibration on that level is achieved, the EPS
calibration procedure can begin and will consist primarily of the temporal adjustment
of demands. Depending on the intended use of the model, the focus of the EPS cali-
bration may vary. For example, for hydraulic studies, the comparison between field
and model conditions will be centered around the prediction of tank water levels and
flows at system meters. On the other hand, for an energy analysis, the capability of the
model to predict pump station cycling and energy consumption will be the focus.
Parameters for Adjustment

Most EPS calibration deals with the examination of plots of observed versus modeled
tank water levels. As a general rule of thumb, if the observed and modeled water lev-
els are both heading in the same direction but at slightly different rates, then the water
use in that pressure zone needs to be corrected. However, if the water levels are going
in opposite directions, then the on/off status at pumps or valves is usually the culprit.
Data from chart recorders placed at key locations in the system can provide insights
into what to adjust.
The magnitude of the demand adjustment required can be approximated by the differ-
ence in tank storage volumes between modeled and observed conditions. For exam-
ple, if the modeled tank and the observed tank both contain 1.24 MG (46,939 m3), but
at the end of a one-hour time step the modeled tank contains 1.63 MG (61,702 m3)
while the real tank contains 1.57 MG (59,431 m3), then the demands in the model may
need to be increased by 0.06 MG (2,271 m3) during that hour (1,000 gpm, 63 l/s). As
always with calibration, such adjustments need to be logical and justifiable, and the
calibration should result in a fairly smooth curve in agreement with the observed data.
Calibration Problems
Discrepancies between the model and observed values are not always a sign of model
inaccuracies. Even though data may have come from a SCADA system (see Chapter 6
for more information) with several digits of precision, it should not be assumed that
the data are accurate to that level. A study done in Vancouver, Canada (Howie, 1999)
documents difficulties in using SCADA data ranging from inconsistent data to prob-
lems involving time-logging. An EPS calibration done for the Wilkes-Barre/Scranton,
Pennsylvania, system documents additional problems, including improperly located
tank level sensors, inaccurate logging of pump switches, and differences between
instantaneous observations and time-averaged data (Walski, Lowry, and Rhee, 2000).
In most modeling, it has been assumed that once an EPS model has been calibrated,
the diurnal curve can be used on other days with minor adjustments to the base
demand. Walski, Lowry, and Rhee (2000) showed that demands in a given hour vary
by up to 20 percent between days in which one would expect to have virtually identi-
cal demand patterns.
Calibration Using Tracers

Although tracers are generally thought of as a tool used in calibrating water quality
models, they are helpful in calibrating EPS hydraulic models as well. For example, if
a conservative tracer is used, the only parameters that a modeler has to adjust are
those affecting the hydraulics of the system.
Before conducting a tracer study, it is best to use a model to simulate tracer movement
in the system. The simulation helps to locate areas in which the tracer is sensitive to
input parameters such as demand. These locations can then be used for monitoring the
tracer.
The tracer selected should be inexpensive, safe, and easy to detect. In the case of a
system with multiple sources, a water quality constituent present in one source at a
concentration different from that of other sources may be a good tracer. For example,
if one well contains water with a higher conductivity than other sources, conductivity
(which is related to the concentration of total dissolved solids) could be used as the
tracer. In other cases, turning off or adjusting the fluoride feed at a plant can create a
disturbance that can be traced through the system.
Section 7.5 Calibration of Water Quality Models 281
The study is conducted by changing the tracer concentration and determining if the
model can reproduce the fluctuations in concentration measured in the system. This
type of analysis provides a great deal of information about the way that water moves
through the system. It is very helpful in spatially allocating demands, identifying
closed valves, and finding pipes with incorrect diameters. (It is not quite as helpful in
identifying pipe roughness errors or pump curve errors, because these parameters do
not significantly affect flow patterns.) More information on using tracer studies can be
found in Grayman (1998).
Energy Studies
In calibrating models to be used for energy consumption studies, it is important to
understand the nature of the data being used. Pump stations not only require energy
for pumping, but also for nonpumping functions such as lighting, SCADA, HVAC,
and so on. Because pump stations may have one power meter for all energy usage
associated with the pump station, it may be necessary to subtract the nonpumping
uses of energy from the total usage to get an accurate field estimate of power used by
the pump.
For situations in which electrical power rather than energy is measured, it is important
to understand whether actual power (in kW) or apparent power (in kVA) is being mea-
sured. The difference between the two is the reactive power used to induce the mag-
netic field within the motor (WEF, 1997). The ratio of the actual to apparent power is
called the power factor.
PF = (actual power) / (apparent power) (7.9)
where PF = power factor
If only apparent power is measured, then this value must be converted to the actual
power for a comparison with the pump energy predicted by models.
Because of the complicated tariffs involved with converting power usage into power
charges, comparisons between the model and the observed power usage should be
made in terms of kilowatt-hours, not dollars.
7.5 CALIBRATION OF WATER QUALITY MODELS

Calibration is the process of adjusting a model so that the simulation reasonably pre-
dicts system behavior. The underlying philosophy of water quality calibration is the
same as that of hydraulic calibration, though some methodological details of the
approach differ. The goal of a water quality calibration is to capture the transient,
dynamic behavior of the network, making water quality calibration a more ill-defined
problem than the notoriously ill-defined hydraulic calibration problem. As a result,
expectations for agreement between simulations and real-world systems are consider-
ably lower.
Source Concentrations
Constituent sources define boundary conditions for water quality simulations, just as
tank and reservoir levels define boundary conditions for hydraulic simulations. For
the purpose of water quality modeling, sources describe how a constituent enters the
distribution system. For example, chlorine and fluoride typically enter a network from
a water treatment plant or other source of finished water, such as an interconnection
with an adjacent system. In the case of system contamination, a substance may be
introduced at any point in the network, such as at a cross-connection or a contami-
nated storage tank. Water quality models typically allow reservoirs, tanks, or junction
nodes to act as constituent sources for flexible modeling of different source types and
scenarios.
Constituent sources can be modeled as a constant influx into the distribution system, or
they can exhibit variation over time. Patterns can be provided to model the dynamic
behavior of constituent sources. Concentrations at sources can also behave like differ-
ent simple feedback controllers (e.g., flow pacing or concentration set point control-
lers).
Initial Conditions
Unlike initial conditions in hydraulic simulations that dissipate quickly, initial condi-
tions in water quality simulations can persist for the entire duration of a reasonably
long simulation. Thus, when calibrating water quality models, the dynamics that
result are always a function of the initial conditions. This increases both the impor-
tance and the difficulty of predicting initial conditions.
Initial conditions reflect the state of the network at the beginning of a water quality
simulation. To model the dynamics of a real system that has been running continu-
ously, events that have occurred prior to the start of the simulation must be accounted
for. For example, consider a scenario in which disinfectant additions made at a treat-
ment plant 72 hours ago are just arriving at a node in the network periphery. The mod-
eler could account for the effect of these historical additions by measuring and
assigning an initial condition at the node. Initial conditions are a mathematical way of
incorporating the historical chain of events that determines the state of the network at
the beginning of a simulation.
Every pipe, junction, tank, and reservoir in the network can be assigned an initial con-
dition related to the analysis being conducted. For constituent analyses, network com-
ponents are assigned an initial concentration. For source trace and water age analyses,
network components are assigned an initial percentage of water arriving from the
source and an initial water age, respectively. Initial conditions are assigned at nodes,
tanks, and reservoirs, and an interpolation method is typically applied to assign them
to pipes.
Predicting Initial Conditions. Assigning initial conditions using values deter-

mined in the field is extremely problematic for a number of reasons. For constituent
analyses, measuring the disinfectant concentration at every node, tank, reservoir, and
pipe is logistically impractical. In addition, measuring all of these concentrations at a
single instant in time (the instant before the simulation is scheduled to start) is impos-
sible. The same problem exists when determining initial conditions for hydraulic
simulations. However, it is not as severe because hydraulic initial conditions are typi-
cally only measured to establish network boundary conditions. The smaller number of
measurements greatly simplifies the logistical considerations associated with collect-
ing them. Predicting initial conditions for water age analyses is also difficult because
age is not a parameter that is easily measured in the field.
The dynamic behavior of the water quality model, however, can be used to eliminate
the problems associated with assigning initial conditions. As water quality processes
are modeled, they form a dissipative system. For example, disinfectant residuals
assigned as an initial condition are present at the beginning of a simulation. As the
simulation progresses, the effects of the initial conditions dissipate as disinfectant
reacts and is removed from the network as hydraulic demands. The disinfectant ini-
tially present is gradually replaced by disinfectant entering from source locations.
Disinfectant concentrations in the network will eventually reach a dynamic equilib-
rium for which concentrations are independent of initial conditions. Thus, if the simu-
lation is allowed to run for a sufficiently long period of time, the values of the initial
conditions assigned become irrelevant.
Setting Initial Conditions. This dissipative behavior of water quality models

can be used to the modeler’s advantage, eliminating the need to predict initial condi-
tions by specifying long simulation times. The exact simulation time depends on net-
work topology and hydraulics but can run anywhere from 3 to 10 times the length of
the diurnal demand pattern. (Typically, a 24-hour diurnal cycle is used.) Essentially,
the hydraulic scenario being modeled is assumed to repeat for all times into the
future. To conduct a long simulation, it is necessary to balance the network hydraulics
so that inflows equal outflows over the length of the demand pattern. Otherwise, tanks
may drain empty or overfill as the simulation progresses, or disturbances may develop
in what should appear as periodic pipe flows.
Once a model has been modified for a long-duration simulation, the initial conditions
(at pipes and nodes) can then be set to any value, including 0.00 mg/l. The time it
takes for concentrations in the distribution system to achieve equilibrium is influ-
enced by the initial conditions set at tanks and reservoirs. Initial conditions within the
tanks and reservoirs adjust very slowly. Conversely, initial conditions at junction
nodes dissipate quickly, and thus initial conditions at junction nodes can safely be set
to zero. Familiarity with a specific source operation scenario may allow the initial
conditions within tanks and reservoirs to be set closer to their equilibrium values, sig-
nificantly decreasing simulation times. If source operation or hydraulics change as
alternative scenarios are evaluated, however, the equilibrium concentrations within
the storage tanks are also likely to change.
Wall Reaction Coefficients

Finding wall reaction coefficients is much more difficult than establishing bulk reac-
tion coefficients (discussed in Chapter 5, page 204). Wall reaction coefficients are
similar to pipe roughness coefficients in that they can and do vary from pipe to pipe.
Like the C-factor (head loss) test for pipe roughness values (see page 191), wall reac-
tion coefficients cannot be directly measured but must be deduced by measuring val-
ues in the field—in this case chlorine residuals and other factors—and then
calculating the wall reaction coefficient that would result in the observed behavior.
The ideal experiment is to isolate a pipe of homogeneous characteristics (diameter,
material, age, flow) and measure chlorine residuals and other factors for that pipe. As
discussed on page 210, for metallic pipes with diameters less than 12 in., a pipe seg-
ment of 1,000 to 2,000 ft is frequently adequate to produce a measurable drop in chlo-
rine residual due to wall demand. However, for larger diameter pipes and pipes made
of materials that are less reactive, it may be necessary to study much longer pipes (of
maybe a mile or more). Because finding homogeneous pipes of that length that can be
isolated to perform a field test is difficult, other methods of calibrating the chlorine
model are needed.
Calibration/Validation Using Time-Series Data. An important step in

calibrating and validating an extended-period simulation hydraulic and/or water qual-
ity model is to compare time-series field data (data collected at intervals over a period
of a day or more) to model results. If the field data and model results are acceptably
close, the model is calibrated. If significant variations exist, adjustments can be made
to various model parameters in order to improve the match. Ideally, one set of data
should be available for calibration, and another set of data should be available to vali-
date that the model is properly calibrated.
A combination of three types of field time-series data can be used in the EPS calibra-
tion/validation process: hydraulic measurements, water quality data, and tracer data.
Chapter 5 describes methods of collecting these types of data. In this chapter, the use
of this data in the calibration/validation process is illustrated.
Figure 7.8 illustrates an example of time-series data collected in a small distribution
system. In this example, a combined field study was performed over a period of 24
hours in which data were collected on the water level in the tank, a tracer study was
conducted, and chlorine measurements were taken at stations in the distribution sys-
tem. The results of the tracer study and the water level measurements were first used
in the hydraulic calibration of the model. After the hydraulic calibration was com-
pleted, the chlorine data was used to calibrate the water quality model (that is, adjust
chlorine decay rates) for chlorine residual. In both phases of the calibration, the field
data were plotted for selected stations, along with the initial model results and the
model results after calibration.
As illustrated in Figure 7.8, tracer measurements were taken at three stations at
approximately three-hour intervals, and water levels were taken at the tank (Station
D) at about 1.5-hour intervals. Referring first to the water level measurements, the ini-
tially predicted pattern is not adequately reflecting the true water levels in the tank.
Because the flow leaving the water treatment plant is well-defined by flow measure-
ments at the plant, the incorrectly predicted water levels suggest that the overall tem-
poral pattern of demands throughout the system is not correct. After the temporal
pattern is adjusted globally throughout the system, the model produces a much better
prediction of tank water levels. Tracer concentrations are initially predicted quite well
at Station A, but some notable discrepancies exist at Stations B and C. The close
agreement at Station A is expected because the travel time from the plant (where the
tracer concentrations are known) to Station A is quite short, leaving little opportunity
to introduce errors in prediction.
Figure 7.8
Collection of time-
series data for model
calibration
At Station B, the lower predicted concentrations indicate that the model is initially
predicting travel times to Station B that are longer than the observed travel times.
Because flow in dead-end pipes is controlled primarily by demand, a moderate
change in the demand pattern at Station B led to improved results in the model.
At Station C, the model was initially predicting that the tracer would reach the node
much more quickly than the observed results. Because the distance from the plant to
this station is relatively short, the significant difference between the observed and pre-
dicted results suggests that there may be an inadvertent fully or partially closed valve
on the direct path to Station C.
A change in the model resulted in much better agreement, and a crew dispatched to
the field confirmed that the valve was closed. The calibration process resulted in a
hydraulic model that was considered to be acceptably calibrated. Because some sig-
nificant changes were made in the model, the utility may want to consider collecting a
second set of data to validate the model parameters.
After the hydraulic model was acceptably calibrated, the chlorine measurements were
used to calibrate the water quality model. Chlorine measurements had been taken at
approximately three-hour intervals at the three distribution system sampling stations
and in the combined inlet/outlet line of the tank. The predicted chlorine concentration
was uniformly slightly low at Station A throughout the day — a surprise because the
travel time from the plant to Station A through a large-diameter pipe was quite short.
The systematic difference suggested that this discrepancy might be due to measure-
ment error.
A check of the instruments showed that the meter that was used to measure chlorine at
the plant was not properly calibrated with the meter used in the field. After adjusting
the chlorine concentrations leaving the plant, the model results accurately reflected
the field results. At Stations B and C, the predicted chlorine concentrations were uni-
formly higher than the observed results. This suggested that reaction rates used in the
model might not have been correct. Because the bulk decay rate was carefully deter-
mined using bottle tests, adjustments were made in the wall demand coefficient with
resulting improvements in the model predictions as compared to field results. At the
tank, the model was predicting very wide swings in chlorine residual in the inlet/
outlet line between the fill and draw cycles. Long residence times in the tank can
cause this behavior. The fact that the observed chlorine residuals did not display such
wide variations led the modeler to surmise either that the true residence time was
shorter than the model suggested or that the tank was stratified and displaying a last-
in-first-out (LIFO) behavior. Additional field-testing of chlorine at different levels in
the tank confirmed that the tank was stratified. By utilizing the LIFO tank representa-
tion in the model, the predicted chlorine concentrations matched the field results. The
utility also took steps to modify the tank operation in order to reduce or eliminate the
stratification problem.
The calibration example illustrated in the preceding paragraph shows many of the typ-
ical calibration problems that are found in the EPS water quality calibration process.
In most cases, parameter adjustment based on field measurements at only a few sta-
Section 7.6 Acceptable Levels of Calibration 287
tions is a difficult and iterative process. Automated calibration using advanced tools
such as genetic algorithms (see page 268) holds significant promise in this area.
Reported values for wall reaction coefficients are in the range of 0 to 5 ft/day (0 to 1.5
m/day). Because these values are difficult to measure, estimates can be based upon
field concentration measurements and water quality simulation results as part of a cal-
ibration analysis. Vasconcelos, Rossman, Grayman, Boulos, and Clark (1997) postu-
lated that the wall reaction coefficient is related to pipe roughness according to the
following equation:
kw = D e C (7.10)
where kw = wall reaction coefficient (ft/day)

D = fitting coefficient
The fitting coefficient is determined for a given system by trial and error during cali-
bration. Assuming that a sufficient number of observed constituent concentrations
have been collected at various locations throughout the system, initial values of wall
reaction coefficients for each pipe can be estimated and the simulation performed.
The observed constituent concentrations can then be compared to concentrations pro-
vided by the computer model. If the two values do not agree within reason, then the
wall reaction coefficients should be adjusted until a suitable match is obtained.
7.6 ACCEPTABLE LEVELS OF CALIBRATION

Regardless of which approach to calibration is adopted, a realistic model should
achieve some level of performance criteria. In the United Kingdom, certain perfor-
mance criteria have been established, and designers strive to meet these standards
(Hydraulic Research, 1983). Table 7.3 outlines the criteria for flow and pressure.
Additional criteria exist, including those for extended-period simulations (WRc,
1989).
For an EPS, in addition to pressures and flows, the volumetric difference between
measured and predicted tank storage between two consecutive time steps should be
r 5 percent of the total tank turnover for significantly large tanks (tank turnover is
taken to be total volume in plus total volume out between two time intervals).
No such guidelines exist in the United States; however, many modelers agree that the
level of effort required to calibrate a hydraulic network model and the desired level of
calibration accuracy will depend upon the intended use of the model (Ormsbee and
Lingireddy, 1997; Cesario, Kroon, Grayman, and Wright, 1996; and Walski, 1995).
The true test of model calibration is that the end user (for example, the pipe design
engineer or chief system operator) of the model results feels comfortable using the
model to assist in decision-making. To that end, calibration should be continued until
the cost of performing additional calibration exceeds the value of the extra calibration
work.
Table 7.3 Calibration criteria for flow and pressure

Flow Criteria
(1) Modeled trunk main flows (where the flow is more than 10% of the total demand) should be
within r 5 % of the measured flows.
(2) Modeled trunk main flows (where the flow is less than 10% of the total demand) should be
within r 10 % of the measured flows.
Pressure Criteria
(1) 85% of field test measurements should be within r 0.5 m or r 5 % of the maximum head loss
across the system, whichever is greater.
(2) 95% of field test measurements should be within r 0.75 m or r 7.5 % of the maximum head
loss across the system, whichever is greater.
(3) 100% of field test measurements should be within r 2 m or r 15 % of the maximum head
loss across the system, whichever is greater.
Each application of a model is unique, and thus it is impossible to derive a single set
of guidelines to evaluate calibration. The guidelines presented below give some
numerical guidelines for calibration accuracy; however, they are in no way meant to
be definitive. A range of values is given for most of the guidelines to reflect the differ-
ences among water systems and the needs of model users. The higher numbers gener-
ally correspond to larger, more complicated systems, and the lower end of the range is
more relevant to smaller, simpler systems. The words “to the accuracy of elevation
and pressure data” mean that the model should be as good as the field data. If the
HGL is known to within 8 ft (2.5 m), then the model should agree with field data to
within the same tolerance. It is important to remember that these guidelines need to be
tempered by site-specific considerations and an understanding of the intended use of
the model.
• Master planning for smaller systems [24-in. (600-mm) pipe and smaller]:
The model should accurately predict hydraulic grade line (HGL) to within 5–
10 ft (1.5–3 m) (depending on size of system) at calibration data points during
fire flow tests and to the accuracy of the elevation and pressure data during
normal demands. It should also reproduce tank water level fluctuations to
within 3–6 ft (1–2 m) for EPS runs and match treatment plant/pump station/
well flows to within 10–20 percent.
• Master planning for larger systems [24-in. (600-mm) and larger]: The
model should accurately predict HGL to within 5–10 ft (1.5–3 m) during
times of peak velocities and to the accuracy of the elevation and pressure
data during normal demands. It should also reproduce tank water level fluc-
tuations to within 3 to 6 ft (1–2 m) for EPS runs and match treatment plant/
well/pump station flows to within 10–20 percent.
• Pipeline sizing: The model should accurately predict HGL to within 5–10 ft
(1.5–3 m) at the terminal point of the proposed pipe for fire flow conditions,
and to the accuracy of the elevation data during normal demands. If the new
pipe impacts the operation of a water tank, the model should also reproduce
the fluctuation of the tank to within 3–6 ft (1–2 m).
Section 7.6 Acceptable Levels of Calibration 289
• Fire flow analysis: The model should accurately predict static and residual
HGL to within 5–10 ft (1.5–3 m) at representative points in each pressure
zone and neighborhood during fire flow conditions and to the accuracy of the
elevation data during normal demands. If fire flow is near maximum fire
flow such that storage tank sizing is important, the model should also predict
tank water level fluctuation to within 3–6 ft (1–2 m).
• Subdivision design: The model should reproduce HGL to within 5–10 ft
(1.5–3 m) at the tie-in point for the subdivision during fire flow tests and to
the accuracy of the elevation data during normal demands.
• Rural water system (no fire protection): The model should reproduce
HGL to within 10–20 ft (3–6 m) at remote points in the system during peak
demand conditions and to the accuracy of the elevation data during normal
demands.
• Distribution system rehabilitation study: The model should reproduce
static and residual HGL in the area being studied to within 5–10 ft (1.5–3 m)
during fire hydrant flow tests and to the accuracy of the elevation data during
normal demands.
• Flushing: The model should reproduce the actual discharge from fire
hydrants or distribution capability [such as the fire flow delivered at a 20 psi
(138 kPa) residual pressure] to within 10–20 percent of observed flow.
• Energy use: The model should reproduce total energy use over a 24-hour
period to within 5–10 percent, energy consumption on an hourly basis to
within 10–20 percent, and peak energy demand to within 5–10 percent.
• Operational problems: The model should reproduce problems occurring in
the system such that the model can be used for decision-making for that par-
ticular problem.
• Emergency planning: The model should reproduce HGL to within 10–20 ft
(3–6 m) during situations corresponding to emergencies (for example, fire
flow, power outage, or pipe out of service).
• Disinfectant models: The model should reproduce the pattern of observed
disinfectant concentrations over the time samples were taken to an average
error of roughly 0.1 to 0.2 mg/l, depending on the complexity of the system.
In addition to these standards, the AWWA Engineering Computer Applications Com-
mittee (1999) posted some calibration guidelines on its web page. As mentioned pre-
viously in this section, however, each modeling application is unique and requires its
own unique set of calibration requirements. The AWWA guidelines are merely exam-
ples of what could be written; they have not been accepted as standards.
In summary, a model can be considered calibrated when the results produced by the
model can be used with confidence to make decisions regarding the design, operation,
and maintenance of a water distribution system, and the cost to improve the model
further cannot be justified.
REFERENCES
Ahmed, I., Lansey, K., and Araujo, J. (1999). “Data Collection for Water Distribution Network Calibra-
tion.” Proceedings of Water Industry Systems: Modelling and Optimisation Applications, Savic, D. A.,
and Walters, G. A., eds., Vol. 1, Exeter, United Kingdom.
American Water Works Association Engineering Computer Applications Committee (1999). “Calibration
Guidelines for Water Distribution System Modeling.” http://www.awwa.org/unitdocs/592/calibrate.pdf.
Bhave, P. R. (1988). “Calibrating Water Distribution Network Models.” Journal of Environmental Engi-
neering, ASCE, 114(1), 120.
Boulos, P. F., and Ormsbee, L. E. (1991). “Explicit Network Calibration for Multiple Loading Conditions.”
Civil Engineering Systems, 8(3), 153.
Boulos, P. F., and Wood, D. J. (1990). “Explicit Calculation of Pipe-Network Parameters.” Journal of
Hydraulic Engineering, ASCE, 116(11), 1329.
Boulos, P. F., and Wood, D. J. (1991). “An Explicit Algorithm for Calculating Operating Parameters for
Water Networks.” Civil Engineering Systems, 8, 115.
Bush, C. A., and Uber, J. G. (1998). “Sampling Design and Methods for Water Distribution Model Calibra-
tion.” Journal of Water Resources Planning and Management, ASCE, 124(6), 334.
Califonia Section AWWA (1962). “Loss of Carrying Capacity of Water Mains.” Journal of the American
Water Works Association, 54(10).
Cesario, A. L., Kroon, J. R., Grayman, W., and Wright, G. (1996). “New Perspectives on Calibration of
Treated Water Distribution System Models.” Proceedings of the AWWA Annual Conference, American
Water Works Association, Toronto, Canada.
Chen, L. C. (1995). “Pipe Network Transient Analysis—The Forward and Inverse Problems.” Thesis, Fac-
ulty of the Graduate School, Cornell University.
Colebrook, C. F., and White, C. M. (1937). “The Reduction of Carrying Capacity of Pipes with Age.” Pro-
ceedings of the Institute of Civil Engineers, 5137(7), 99.
Datta, R. S. N., and Sridharan, K. (1994). “Parameter Estimation in Water Distribution Systems by Least
Squares.” Journal of Water Resources Planning and Management, ASCE, 120(4), 405.
de Schaetzen, W. (2000). “Optimal Calibration and Sampling Design for Hydraulic Network Models.”
Ph.D. Thesis, School of Engineering and Computer Science, University of Exeter, United Kingdom.
de Schaetzen, W., Randall-Smith, M., Savic, D. A., and Walters, G. A. (1999). “Optimal Logger Density in
Water Distribution Network Calibration.” Proceedings of Water Industry Systems: Modelling and Opti-
misation Applications, Savic, D. A., and Walters, G. A., eds., Vol. 1, Exeter, United Kingdom.
Ferreri, G. B., Napoli, E., and Tumbiolo, A. (1994). “Calibration of Roughness in Water Distribution Net-
works.” Proceedings of the 2nd International Conference on Water Pipeline Systems, BHR Group,
Edinburgh, United Kingdom.
Grayman, W.M. (1998). “Use of Tracer Studies and Water Quality Models to Calibrate a Network Hydrau-
lic Model,” Essential Hydraulics and Hydrology, Haestad Methods, Inc., Waterbury, Connecticut.
Greco, M., and Del Guidice, G. (1999). “New Approach to Water Distribution Network Calibration.” Jour-
nal of Hydraulic Engineering, ASCE, 125(8), 849.
Herrin, G. (1997). “Calibrating the Model.” Practical Guide to Hydraulics and Hydrology, Haestad Press,
Waterbury, Connecticut.
Howie, D. C. (1999). “Problems with SCADA Data for Calibration of Hydraulic Models.” Proceedings of
the ASCE Annual Conference of Water Resources Planning and Management, American Society of
Civil Engineers, Tempe, Arizona.
Hudson, W. D. (1966). “Studies of Distribution System Capacity in Seven Cities.” Journal of the American
References 291
Hydraulic Research (1983). Tables for the Hydraulic Design of Pipes and Sewers. Wallingford, England.
Kapelan, Z., Savic, D. A., and Walters, G. A. (2000). “Inverse Transient Analysis in Pipe Networks for
Leakage Detection and Roughness Calibration.” Water Network Modelling for Optimal Design and
Management, CWS 2000, Centre for Water Systems, Exeter, United Kingdom, 143.
Kapelan, Z., Savic, D. A., and Walters, G. A. (2001). “Use of Prior Information on Parameters in Inverse
Transient Analysis for Leak Detection and Roughness Calibration.” Proceedings of the World Water and
Environmental Resources Congress, Orlando, Florida.
Kapelan, Z., Savic, D. A., and Walters, G. A. (2001a). “Optimal Sampling Design Methods for Calibration
of Water Supply Network Models.” Water Software Systems: Theory and Applications, Vol. 1, Ulanicki,
B., Coulbeck, B. and Rance, J.P., eds., Research Studies Press, Baldock, Hertfordshire, United King-
dom.
Lamont, P. A. (1981). “Common Pipe Flow Formulas Compared With the Theory of Roughness.” Journal
Lansey, K. E., and Basnet, C. (1991). “Parameter Estimation for Water Distribution Networks.” Journal of
Water Resources Planning and Management, ASCE, 117(1), 126.
Lansey, K. E., El-Shorbagy, W., Ahmed, I., Araujo, J., and Haan, C. T. (2001). “Calibration Assessment
and Data Collection for Water Distribution Networks.” Journal of Hydraulic Engineering, ASCE,
127(4), 270.
Lee, B. H., and Deininger, R. A. (1992). “Optimal Locations of Monitoring Stations in Water Distribution
Systems.” Journal of Environmental Engineering, ASCE, 118(1), 4.
Liggett, J. A., and Chen, L. C. (1994). “Inverse Transient Analysis in Pipe Networks.” Journal of Hydraulic
Meier, R. W., and Barkdoll, B. D. (2000). “Sampling Design for Network Model Calibration Using Genetic
Algorithms.” Journal of Water Resources Planning and Management, ASCE, 126(4), 245.
Muss, D. L. (1960). “Friction Losses in Lines with Service Connections.” Journal Hydraulics Division,
ASCE, 86(4), 35.
Ormsbee, L. E. (1989). “Implicit Pipe Network Calibration.” Journal of Water Resources Planning and
Management, ASCE, 115(2), 243.
Ormsbee, L. E., and Lingireddy, S. (1997). “Calibrating Hydraulic Network Models.” Journal of the Amer-
Ormsbee, L. E., and Wood, D. J. (1986). “Explicit Pipe Network Calibration.” Journal of Water Resources
Planning and Management, ASCE, 112(2), 166.
Piller, O., Bremond, B., and Morel, P. (1999). “A Spatial Sampling Procedure for Physical Diagnosis in a
Drinking Water Supply Network.” Proceedings of Water Industry Systems: Modelling and Optimisation
Applications, Savic, D. A., and Walters, G. A., eds., Vol. 1, Exeter, United Kingdom.
Pudar, R. S., and Liggett, J. A. (1992). “Leaks in Pipe Networks.” Journal of Hydraulic Engineering,
ASCE, 118(7), 1031.
Rahal, C. M., Sterling, M. J. H., and Coulbeck, B. (1980). "Parameter Tuning for Simulation Models of
Water Distribution Networks.” Proceedings of the Institute of Civil Engineers, Part 269, 751.
Reddy, P. V. N., Sridharan, K., and Rao, P. V. (1996). “WLS Method for Parameter Estimation in Water
Distribution Networks.” Journal of Water Resources Planning and Management, ASCE, 122(3), 157.
Savic, D. A., and Walters, G. A. (1995). “Genetic Algorithm Techniques for Calibrating Network Models.”
Report No. 95/12, Centre For Systems And Control Engineering, School of Engineering, University of
Exeter, Exeter, United Kingdom, 41.
Savic, D. A., and Walters, G. A. (1997). “Evolving Sustainable Water Networks.” Hydrological Sciences,
42(4), 549.
Shamir, U., and Hamberg, D. (1988). “Schematic Models for Distribution System Design. I: Combination
Concept.” Journal of Water Resources Planning and Management, ASCE, 114(2), 129.
Shamir, U., and Howard, C. D. D. (1968). “Water Distribution Systems Analysis.” Journal of the Hydraulic
Division, ASCE, 94(1), 219.
Simpson, A. R., and Vitkovsky, J. P. (1997). "A Review of Pipe Calibration and Leak Detection Methodol-
ogies for Water Distribution Networks.” Proceeding of the 17th Federal Convention, Australian Water
and Wastewater Association, Australia, 1.
Tang, K., Karney, B., Pendlebury, M., and Zhang, F. (1999). "Inverse Transient Calibration of Water Distri-
bution Systems Using Genetic Algorithms.” Water Industry Systems: Modelling and Optimisation
Applications, Research Studies Press Ltd., Exeter, United Kingdom, 1.
Todini, E. (1999). "Using a Kalman Filter Approach for Looped Water Distribution Network Calibration.”
Water Industry Systems: Modelling and Optimisation Applications, Research Studies Press Ltd., Exeter,
United Kingdom, 1.
Vasconcelos, J. J., Rossman, L. A., Grayman, W. M., Boulos, P. F., and Clark, R. M. (1997). “Kinetics of
Chlorine Decay.” Journal of the American Water Works Association, 89(7), 54.
Vitkovsky, J. P., and Simpson, A. R. (1997). “Calibration and Leak Detection in Pipe Networks Using
Inverse Transient Analysis and Genetic Algorithms.” Report No. R 157, Department of Civil and Envi-
ronmental Engineering, University of Adelaide, Australia.
Vitkovsky, J. P., Simpson, A. R., and Lambert, M. F. (2000). “Leak Detection and Calibration Using Tran-
sients and Genetic Algorithms.” Journal of Water Resources Planning and Management, ASCE, 126(4),
262.
Walski, T. M. (1983). “Technique for Calibrating Network Models.” Journal of Water Resources Planning
and Management, ASCE, 109(4), 360.
Walski, T. M. (1986). “Case Study: Pipe Network Model Calibration Issues.” Journal of Water Resources
Walski, T. M. (1990). “Sherlock Holmes Meets Hardy Cross or Model Calibration in Austin, Texas.” Jour-
Walski, T. M. (1995). “Standards for Model Calibration.” Proceedings of the AWWA Computer Conference,
American Water Works Association, Norfolk, Virginia.
Walski, T. M. (2000). “Model Calibration Data: The Good, The Bad and The Useless.” Journal of the
Walski, T. M. (2002). “Identifying Monitoring Locations in a Water Distribution System Using Simulation
and GIS.” Proceedings of the AWWA Information Management and Technology Conference, American
Water Works Association, Kansas City, Missouri.
Walski, T. M., Edwards, J. D., and Hearne, V. M. (1989). “Loss of Carrying Capacity in Pipes Carrying
Softened Water with High pH.” Proceedings of the ASCE National Conference on Environmental Engi-
neering, American Society of Civil Engineers, Austin, Texas.
Walski, T. M., Lowry, S. G., and Rhee, H. (2000). “Pitfalls in Calibrating an EPS Model.” Proceedings of
the Environmental and Water Resource Institute Conference, American Society of Civil Engineers, Min-
neapolis, Minnesota.
Walters G.A., Savic, D. A., Morley, M. S., de Schaetzen, W., and Atkinson, R. M. (1998). “Calibration of
Water Distribution Network Models Using Genetic Algorithms.” Hydraulic Engineering Software VII,
Computational Mechanics Publications, 131.
Water and Environment Federation (WEF) (1997). “Energy Conservation in Wastewater Treatment Facili-
ties.” WEF Manual of Practice MFD-2, Alexandria, Virginia.
Water Research Center (WRc) (1989). Network Analysis – A Code of Practice. WRc, Swindon, England.
References 293
Wu, Z. Y., Boulos, P. F., Orr, C. H., and Ro, J. J. (2000). “An Efficient Genetic Algorithms Approach to an
Intelligent Decision Support System for Water Distribution Networks.” Proceedings of the Hydroinfor-
matics Conference, Iowa.
Wu, Z. Y, Walski, T. M., Mankowski, R., Herrin, G., Gurrieri, R. and Tryby, M. (2002a). “Calibrating Water
Distribution Model Via Genetic Algorithms.” Proceedings of the AWWA Information Management and
Technology Conference, American Water Works Association, Kansas City, Missouri.
Wu, Z. Y. Walski, T. M., Mankowski, R., Herrin, G., Gurrieri, R., Tryby, M., and Hartell, W. (2002b).
“Impact of Measurement Errors on Optimal Calibration of Water Distribution Models.” Proceedings of
the International Conference on Technology Automation and Control of Wastewater and Drinking Water
Systems, Technical University of Gdansk, Gdansk, Poland.
Yu, G., and Powell, R. S. (1994). “Optimal Design of Meter Placement in Water Distribution Systems.”
International Journal of Systems Science, 25(12), 2155.
7.1 English Units: Calibrate the system shown in Problem 3.3 (see page 131) and given in Prob7-01.wcd
so that the observed pressure of 63.0 psi at node J-5 is obtained. Adjust nodal demands by using the
same multiplier for all demands (global demand adjustment).
a) By what factor must demands be adjusted to obtain the observed pressure?
b) Would you say that pressures in this system are sensitive to nodal demands? Why or why not?
SI Units: Calibrate the system shown in Problem 3.3 (see page 131) and given in Prob7-01m.wcd so
that the observed pressure of 434.4 kPa at node J-5 is obtained. Adjust nodal demands by using the
same multiplier for all demands (global demand adjustment).
a) By what factor must demands be adjusted to obtain the observed pressure?
b) Would you say that pressures in this system are sensitive to nodal demands? Why or why not?
7.2 Calibrate the system shown in Problem 4.1 (see page 171) so that the observed pressure of 54.5 psi is
obtained at node J-4. Adjust the internal pipe roughness using the same multiplier for all pipes (glo-
bal adjustment factor).
a) What is the global roughness adjustment factor necessary to obtain the pressure match?
b) Are the pressures in this system sensitive to pipe roughness under average day demands? Why or
why not?
c) Would you say that most water distribution systems are insensitive to pipe roughness values
under low flows?
d) Is it reasonable to expect that a field pressure can be read with a precision of ±0.5 psi? If not,
what would you say is a typical precision for field-measured pressures?
e) What can contribute to the imprecision in pressure measurements?
7.3 Use the calibrated system found from Problem 7.2 and place a fire flow demand of 1,500 gpm at
node J-4.
a) What is the pressure at node J-4?
b) Is a pressure of this magnitude possible? Why or why not?
c) If a pressure this low is not possible, what will happen to the fire flow demand?
d) What is the most likely cause of this low pressure?

7.4 Starting with the original pipe roughness values, calibrate the system presented in Problem 4.3 (see
page 177) so that the observed pressure of 14 psi is obtained at node J-11. Close pipes P-6 and P-14
for this simulation. Assume that the area downstream of the PRV is a residential area.
Hint: Concentrate on pipe roughness values downstream of the PRV.
a) What pipe roughness values were needed to calibrate this system?
b) Would you consider these roughness values to be realistic?
c) A fire flow of 1,500 gpm is probably more than is needed for a residential area. A flow of 750
gpm is more reasonable. Using the uncalibrated model, determine whether this system can
deliver 750 gpm at node J-11 and maintain a minimum system-wide pressure of 30 psi.
7.5 Calibrate the system completed in Problem 4.4 (see page 180) and given in Prob7-05.wcd so that the
observed hydraulic grade line elevations in the Central Tank (see the following table) are repro-
duced.
Hint: Focus on changing the multipliers in the diurnal demand pattern.

Time Central Tank HGL
(hr) (ft)
0.00 1,525
1.00 1,527
2.00 1,529
3.00 1,531
4.00 1,532
5.00 1,534
6.00 1,536
7.00 1,537
8.00 1,539
9.00 1,540
10.00 1,541
11.00 1,542
12.00 1,540
13.00 1,537
14.00 1,534
15.00 1,532
16.00 1,533
17.00 1,535
18.00 1,536
19.00 1,537
20.00 1,538
21.00 1,539
22.00 1,541
23.00 1,542
24.00 1,544
Fill in the table below with your revised diurnal demand pattern multipliers. Insert more times if
necessary.
Time of Multiplication
Day Factor
Midnight
3:00 a.m.
6:00 a.m.
9:00 a.m.
Noon
3:00 p.m.
6:00 p.m.
9:00 p.m.
Midnight
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
8
Using Models for Water
Distribution System Design
Engineers have designed fully functioning water distribution systems without using
computerized hydraulic simulations for many years. Why then, in the last several
decades, has the use of computerized simulations become standard practice for
designing water distribution systems?
First, computerized calculations relieve engineers of tedious, iterative calculations,
enabling them to focus on design decisions. Second, because models can account for
much more of the complexity of real-world systems than manual calculations, they
give the engineer increased confidence that the design will work once it is installed.
Finally, the ease and speed with which models can be used gives the engineer the abil-
ity to explore many more alternatives under a wide range of conditions, resulting in
more cost-effective and robust designs.
There is a price to pay for the extra capability that engineers now possess as a result of
high-quality hydraulic simulation software. The easiest part of that price to quantify is
the cost of the software itself. Another obvious cost is the time required to assemble
data and construct a network model. In addition, there are costs associated with train-
ing personnel to use a new tool and the time it takes to gain experience using it effec-
tively. The total cost is small, however, compared to the value of the projects being
considered and the repercussions of poor decisions.
A model that has been assembled properly is an asset to the water utility, much like a
pipe or a fire hydrant. The model should therefore be maintained so that it is ready to
be put to valuable use. The difficult part of valuing modeling lies in the fact that the
costs of modeling are incurred mostly in model development, and the benefits are
realized later in the form of quicker calculations and better decisions.
Because such a large investment in time and effort is needed to make a model usable,
a common mistake is to not leave enough time in a study (whether it is creating a
major master plan or checking the location of a proposed tank) to adequately analyze
298 Using Models for Water Distribution System Design Chapter 8
the design. To get the most out of a model, it is important to allow sufficient time to
try different alternatives and test these alternatives against a wide range of conditions.
Although the time spent performing additional analyses may seem to cause a delay,
good designs will save both time and money, provide insight into the workings of the
system, and improve the performance of a project.
8.1 APPLYING MODELS TO DESIGN APPLICATIONS

Up to this point, the book has addressed how to build and calibrate models. The
remainder focuses on using those models and the results they provide to build water
systems and to assist in operating them. An overview of model application is shown in
Figure 8.1. The sections that follow discuss each item in the figure in more detail.
Figure 8.1
Overview of model
application Existing Additional Model
Problem Ready for
Calibrated Calibration
Statement Problem
Model Data
Formulate
Alternatives
Make Cost Test

Decision Analysis Alternatives
Extent of Calibration and Skeletonization

Anyone who uses models regularly realizes that no model is ever perfectly calibrated
(see Chapter 7). Therefore, before using a model to solve a particular problem, the
engineer needs to verify that the model is sufficiently calibrated and has a high
enough resolution to be useful for the problem under consideration. A single model
may not work well for analyzing every problem without additional work. Therefore, it
may be necessary to have slightly different versions of the model for performing dif-
ferent analyses. Care should be taken to keep all versions up-to-date.
For example, a model may predict peak flow and fire flow behavior in a given part of
town very well even though it is somewhat skeletonized. If the problem being consid-
ered involves determining what piping or storage must be added to ensure service in
the event that the largest pipe supplying that section of town fails, the model may not
contain sufficient detail. Numerous smaller pipes may need to be added to the model
so that it will more accurately represent the secondary paths that the water can take
within the distribution system.
Section 8.1 Applying Models to Design Applications 299
Depending on the reason for the analysis, it may be worthwhile to install a few porta-
ble pressure recorders or run several hydrant flow tests in the area being considered,
before getting too far into the analysis. Sometimes a detail important to a specific sit-
uation may be left out of a general model, even though the model as a whole appears
well-calibrated.
In other situations, the model may have far too much detail for the analysis being con-
ducted. For example, a highly detailed model may be more than sufficient for setting
control valves or evaluating pump cycling. Also, too much detail may give a false
sense that the model will provide more accurate results simply because it contains
more information.
A similar consideration exists with demands. The model may have been set up for a
master planning study, with reasonable projected demands assigned to nodes; how-
ever, the question being posed may refer to a particular subdivision or new industrial
customer. The previous master plan projections should be replaced by the more pre-
cise demand projections for this new customer or land development.
When using the model to assess the capability of the distribution system to serve a
particular new customer, it is important to remember that the improvements being
installed may also be required to serve future customers. Therefore, projected future
demands must be accounted for in any sizing calculations. Often, the question then
becomes one of how the cost of the new facilities will be divided between the new
customer and the utility.
Design Flow
Ideally, each pipe, pump, and valve in a system has been sized using some design
flow. The design flow used is typically the peak flow that the facility will encounter in
the foreseeable future. In any study, the engineer must determine this flow for the
facility being designed. The design flow is usually based on a prediction, which is
problematic because it is almost always incorrect to some extent. Therefore, facilities
should be sized to operate efficiently, accounting for uncertainty in design flow esti-
mates.
Oversized facilities have pipes and pumps that are not fully utilized, with the associ-
ated inefficiencies and misallocations of capital resources. Oversized facilities may
also be plagued with water quality problems due to long residence times. Conversely,
undersized facilities are inadequate to meet demands, a situation that must later be
corrected by paralleling, replacing, or retrofitting facilities to expand capacity.
To some extent, the decision on design flow acts as a self-fulfilling prophecy. If distri-
bution capacity is installed, customers will eventually use that capacity. Today’s
“excess capacity” has a way of becoming a valuable resource that is quickly absorbed
through development.
While design flow is a useful concept for specifying equipment, the model should
also be used to simulate a large range of possible conditions and ensure robust
designs. EPS runs performed for a range of flows (such as current average day and
year 2020 peak day) are particularly useful for evaluating how the system will
respond under a variety of conditions.
Reliability Considerations
When designing or improving a system, the possibility that the system needs to func-
tion even when components are out of service (such as in the case of a pipe break,
power outage, natural disaster, or off-line equipment) should be considered. The dis-
tribution system cannot be expected to perform without some degradation of service
during an outage. However, when economically feasible, the system should be
designed to at least meet appropriate minimum performance standards during reason-
able emergencies and other circumstances in which facilities may be out of service.
To model the failure of a pipe, removing or closing off a single pipe link in a distribu-
tion model would be simple. The number of links and nodes removed, however,
depends on the locations of the valves necessary to close off that area or segment (the
smallest portion of a system that can be isolated using valves). Seven water distribu-
tion segments are shown in the map in Figure 8.2(a). The effect of a break on the
topology of the distribution model is shown for segments 1 and 2. For a break in seg-
ment 1, only a single pipe link is taken out of service as shown in Figure 8.2(b). For a
break in segment 2, several pipe links and junction nodes are removed from the model
as shown in Figure 8.2(c). The effect of failures on facility operation is further
described in Chapter 10 (see page 433).
A reliability analysis of an entire water distribution system has not yet proven to be
workable, partly because there are so many different ways of defining reliability (as
summarized by Wagner, Shamir, and Marks, 1988a, 1988b). Mays (1989) summa-
rized the state-of-the-art in reliability analysis in an ASCE Committee Report. More
recently, Goulter et al. (2000) provided an overview of reliability assessment methods
that included 81 references. Walski (1993) pointed out that the problem is not simply
one of hydraulic analysis, but is also closely related to operation and maintenance
practices.
Figure 8.2
Distribution system
2 segments
1
7
6 3
5 4
a.
b.
c.
Key Roles in Design Using a Model

After the model has been constructed and calibrated, it is ready to be used in design.
There are two distinct roles that need to be filled when using a model. The first is that
of the modeler who actually runs the program, and the second is that of the design
engineer who must make the decisions regarding facility sizing, location, and timing
of construction. In most cases, the models are sufficiently easy to use that both roles
can be filled by a single individual. When two individuals are involved, the task of the
design engineer is to decide on the situations and design alternatives to be modeled.
The modeler then runs the desired simulations.
To get the most benefit from the model, the designer should examine a broad range of
alternatives. Background investigation prior to beginning the modeling process is
often very helpful. A brainstorming session with utility employees can generate a
consistent understanding of the nature of the problem, a review of the facts surround-
ing the problem, and, most important, a wide range of alternative solutions. By
involving others in this initial meeting, difficulties that can arise later (such as ques-
tions about why a particular alternative was not considered) can be prevented. A team
approach also facilitates acceptance of designs that are developed using the model.
Types of Modeling Applications

There is no single correct way to use models. Walski (1995) described how model
application for design purposes differs depending on whether the model is being used
for master planning, preliminary design, subdivision development, or system rehabili-
tation. Each type of model has a specific goal and related characteristics, as summa-
rized in the following list:
• Master planning. Master planning models are used to predict what
improvements and additions to the distribution system will be necessary to
accommodate future customers. Therefore, these models have long planning
horizons (on the order of 20 to 40 years). The designs are controlled by
future demands, and emphasis is usually placed on larger transmission
mains, pump stations, and storage tanks, as opposed to small neighborhood
mains. Systems in master planning models can be highly skeletonized, and
future pumps may be represented by constant-head nodes (that is, modeled
as reservoirs).
• Preliminary design. In preliminary design, the engineer models the facili-
ties that will be required to serve a particular area of, or addition to, the dis-
tribution system service area. For this type of modeling, the focus is limited
to a small portion of the system. Actual pump curves should be included, but
detailed calibration is only needed in the section of the model from the
source to the project, while the remainder of the system can be highly skele-
tonized.
• Subdivision layout. When designing a subdivision (particularly in the U.S.),
the capacity requirements of fire flows usually dominate those of customer
demands, and the planning horizon is typically short (perhaps five years to
subdivision build-out). Calibration is only needed near the points of connec-

tion to the existing system, and although detail is required when modeling
the pipes in the development, the remainder of the system can be highly skel-
etonized.
• Rehabilitation. In a rehabilitation study of an area of a system, adequate
capacity for fire flows is usually the most important consideration. Many
more alternative scenarios are needed compared to the number required
when designing new pipe because a variety of possible solutions exist (for
example, relining, paralleling, or looping). Detail is needed only for the part
of the model that represents the study area; the remainder of the system
model can be skeletonized.
Cesario (1995) reported that the most common application of water distribution mod-
eling is long-range planning (referred to by some utilities as master planning, capital
budgeting, or comprehensive planning studies). The next most common uses are fire
flow studies and new development design. Models tend to be used more by planning
and design personnel, rather than operations personnel. Of course, after a water util-
ity’s personnel become familiar with a model, the application is limited only by the
time and imagination of the users.
Pipe Sizing Decisions

One of the most common uses of water distribution models is selecting pipe sizes.
With the exception of a minimum pipe size of 6 in. (150 mm) for mains providing fire
protection, there are few standards for pipe sizing. Instead, the standards are usually
expressed in terms of a minimum pressure that must be maintained in the system. It is
the responsibility of the engineer to establish the demands that must be met in the sys-
tem and perform the hydraulic calculations that will determine whether the proposed
solution is adequate. For each adequate solution, the engineer then considers whether
the cost is acceptable or whether more promising solutions are available. This process
is summarized in Figure 8.3. The design process can be improved in some cases by
using optimization as described in Section 8.11 on page 360.
Once the model of the existing system has been created, the engineer adds the new
pipes that are being sized. The initial pipe sizes for these new pipes can then be set at
either the minimum allowable size or a size estimated from Equation 8.1:
Cf Q
D = ---------- (8.1)
V
where D = initial estimate of diameter (in., ft, mm)

Cf = unit conversion factor
= 0.41 for Q in gpm, D in in., V in ft/s
= 1274 for Q in l/s, D in mm, V in m/s
= 1.27 for Q in cfs, D in ft, V in ft/s
Q = peak flow (gpm, cfs, l/s)
V = maximum allowable velocity (ft/s, m/s)
Using this equation will result in reasonably sized pipes. Next the engineer runs the
model for a variety of conditions (average day, peak hour, max day plus fire at key
locations, tank refilling at low demand times, etc.) and reviews the model results for
things such as
• high velocities (see page 307 for guidelines on maximum allowable veloci-
ties)
• pressures below minimum
• pumps not operating at desirable points on pump curve
• tanks not draining or filling at desirable rates
• unusually high pressures
• low velocities during peak demand periods
• low disinfectant residual or high water age if water quality analyses are run
Figure 8.3
Overview of pipe
sizing Model of Initial
Selection of Demand
Existing Estimates
System Sizes
Layout of
Model System
Proposed
Runs Outages
Piping
Reformulate No Compare Yes Estimate

Design with Costs
Standards
and
Guidelines
No Best
Costs
Yes
Present to
Decision
Makers
If the engineer notices pipes not performing well, he or she should adjust the diame-
ters to obtain acceptable behavior in the system. Next, the engineer should try differ-
ent alternative layouts to find low-cost alternatives. This usually involves finding
pipes with low velocities even during peak demand conditions and decreasing pipe
size to determine the potential for cost savings without violating standards.
Section 8.2 Identifying and Solving Common Distribution System Problems 305
In some cases, pump energy analyses and water quality analyses may be required to
evaluate those aspects of the design. Solutions should be presented to decision-makers
for review and discussion. Each of these topics is discussed in greater detail in the fol-
lowing sections.
8.2 IDENTIFYING AND SOLVING COMMON

DISTRIBUTION SYSTEM PROBLEMS
Most water distribution systems share a number of common concerns. For example,
the typical governmental standard for water system design is, “The system shall be
designed to maintain a minimum pressure of 20 psi (138 kPa) at ground level at all
points in the distribution system under all conditions of flow. The normal working
pressure in the distribution system should be approximately 60 psi (414 kPa) and not
less than 35 psi (241 kPa)” (GLUMB, 1992). Regulations do not typically spell out
how to meet this requirement, leaving such decisions up to the design engineer, who
will examine possible alternatives by using modeling techniques.
In general, poor pressures tend to be caused by inadequate capacity in a pipe or pump,

high elevations, or some combination of the two. Models are helpful in pinpointing
the cause of the problem. Figure 8.4 shows how an EPS model can help determine
whether the low pressure is due to capacity or elevation problems. Customers at high
elevations may experience constant problems with low pressure, while a capacity
problem may show up only during periods of high demand. The “Typical” line in Fig-
ure 8.4 represents pressure fluctuations in a typical system; the “Capacity Problem”
line shows pressures for a system with pump or main capacity problems; and the
“Elevation Problem” line shows pressure fluctuations in a portion of a system where
the utility is attempting to serve a customer at too high of an elevation.
Figure 8.4
EPS runs showing low
60 pressure due to
elevation or system
Typical capacity
50
40
Pressure, psi
Capacity Problem
30
20
Elevation Problem
10
0
0 10 20 30 40
Time, hr
Undersized Piping
An undersized distribution main will not be easy to identify during average-day con-
ditions, or even peak-day conditions, because demand and velocity are typically not
high enough during those times to reveal the problem. If a pipe is too small, it may
become a problem only during high-flow conditions such as fire flow. Fire flows are
much greater than normal demands, especially in residential areas. Therefore, fire
flow simulations are the best way to identify an undersized distribution main. If look-
ing for sizing problems in larger pipes, such as those leaving treatment plants, the best
time for diagnosing problems would likely be the peak hour or, in some cases, during
periods when tanks are refilling.
If undersized pipes are suspected, they can usually be found by looking for pipes with
high velocities. These pipes can be located quickly by sorting model output tables by
velocity or hydraulic gradient (friction slope), or by color-coding pipes based on these
parameters. It is important to note that when evaluating models for undersized pipes,
it is better to evaluate based on hydraulic gradient rather than head loss. Although one
pipe may have a much larger head loss than another, the hydraulic gradient may actu-
ally be lower, depending on the length of the pipes being compared.
No fixed rule exists regarding the maximum velocity in a main (although some utili-
ties do have guidelines), but pressures usually start to drop off (and water hammer
problems become more pronounced) when velocities reach 10 ft/s (3 m/s). In larger
pressure zones (several miles across), a velocity as low as 3 ft/s (1 m/s) may cause
excessive head loss. Increasing the diameter of the pipe in the model should result in a
corresponding decrease in velocity and increase in pressure. If not, then another pipe
or pump may be the reason for poor pressures.
Inadequate Pumping
In a pressure zone that is served by a pump, pressures that drop off significantly may
indicate a pump capacity problem. This drop will be most dramatic in situations in
which most of the pumping energy is used for lift rather than for overcoming friction
or in which there is no storage in the pressure zone (or the storage is located far from
the problem area). When the flow rate increases above a certain level, the head pro-
duced by the pump drops off, and pressures decrease by a corresponding amount.
At first, undersized pipes might be suspected as the cause of the problem, but increas-
ing pipe sizes has little impact in this case. A comparison of the pump’s production
with its rated capacity [for example, a 600 gpm (0.037 m3/s) pump trying to pass 700
gpm (0.044 m3/s)] will indicate the problem. Installing a larger pump (in terms of
flow, not head) or another pump in parallel corrects the problem if the pump flow
capacity is the real cause.
When the pressure zone has enough storage, diagnosing problems caused by under-
sized pumps may be difficult. Undersized pumps show up more clearly in EPS runs,
however, because the tank water levels do not recover during a multiday simulation.
For most pump station designs, the pumps should meet design flow requirements,
even with the largest pump out of service. For example, a three-pump station should
What’s the Maximum Permissible

Velocity in a Pipe?
A frequently asked question in water distribution Another factor to consider is that when velocity is
design is, “What is the maximum acceptable high, changes in velocity are also high, and these
velocity in a pipe?” The answer to this question accelerations can lead to harmful hydraulic tran-
can make pipe design easier because, knowing sients (that is, water hammer). One approach to
the maximum velocity and the design flow, an reducing transients is to reduce velocity. Hydraulic
engineer can calculate pipe diameter using transients are covered in detail in Chapter 13.
With these multiple complicating factors, there
D = Cf Q e V
cannot be a single maximum velocity that is opti-
where D = pipe diameter mal in every situation. On the contrary, designing
Q = design flow pipe sizes for velocity alone is not the correct
V = maximum velocity approach with water distribution systems. The
Cf = unit conversion factor (see page 303) velocities are useful only for spot-checking net-
There is no simple answer to this question work model output when locating bottlenecks in
because velocity is only indirectly the limiting fac- the system (that is, pipes with very high velocities,
tor in pipe sizing. It is really the head loss caused and therefore high head losses). The real test of a
by the velocity, not velocity itself, that controls siz- design’s efficiency is not velocity, but residual
ing. The problem is complicated by the fact that pressures in the system during peak demand
most water distribution systems are looped, so a times.
sizing decision in one pipe affects the size, and When checking designs for permissible velocities,
therefore flow velocity, in all other pipes. some engineers use 5 ft/s (1.5 m/s) as a maxi-
Technical papers going back to Babbitt and mum, others use 8 ft/s (2.4 m/s), and yet still oth-
Doland (1931) and Camp (1939) discuss econom- ers use 10 ft/s (3.1 m/s). Because velocity is not
ically sound values for this maximum velocity. This the real design parameter, there is no simple
work was extended by Walski (1983), who answer. Rather, velocity is simply another
showed that the optimal velocity in pumped lines parameter an engineer can use to check a design.
can range from 3 to 10 ft/s (1 to 3 m/s), depending
on the relative size of the peak and average flow
rates through the pipe and the relative magnitude
of construction and energy costs.
be able to meet demands using any two of the pumps; otherwise, additional capacity
may be needed.
Consistent Low Pressure

If pressures are consistently low in an area, then the problem is usually due to trying
to serve customers at too high an elevation for that pressure zone. This problem is
apparent even during low-demand periods. Changing pipe sizes or pump flow capac-
ity will not improve this situation. If this pressure zone has no storage tanks, it may be
possible to increase the head for a fixed-speed pump or increase the control point for a
variable-speed pump (provided this increase does not overly pressurize other portions
of the system).
In many cases, the best solution is to move the pressure zone boundary so that those
customers experiencing low pressures will be served from the next higher pressure
zone. When a zone of higher pressure does not exist, one must be created (see page
333). If a new zone is established, the hydraulic grade line in that zone should serve a
significant area, not just a few customers around the current pressure zone boundary.
Each succeeding pressure zone should be approximately 100 ft (30.5 m) higher than
the next lower zone. If they are less than 50 ft (15.2 m) apart in elevation, too many
pressure zones may complicate operation. If they are more than 150 ft (45.7 m) apart
in elevation, it is difficult to serve the highest customers without overly pressurizing
the lowest customers in that zone.
High Pressures During Low Demand Conditions

High pressures are usually caused by serving customers at too low an elevation for the
pressure zone. Some utilities consider 80 psi (550 kPa) to be a high pressure, although
most systems can tolerate 100 psi (690 kPa) before experiencing problems (for exam-
ple, increased leakage, increased breaks, water loss through pressure relief valves, and
increased load on water heaters and other fixtures). Portions of some distribution sys-
tems can bear significantly greater pressures because pipe with a high-pressure rating
has been installed. When dealing with high pressures, PRVs can be used to reduce
pressures for individual customers, although they may result in additional mainte-
nance issues.
Usually, high pressures are easiest to evaluate with model runs at low demands (say
40 to 60 percent of average flow). This range corresponds to minimum nighttime
demands for a typical system. If the engineer feels that pressures are too high, the
usual solution is to establish a new pressure zone for the lower elevation using system
PRVs (as opposed to individual home PRVs).
When a constant-speed pump is moving a substantially lower flow than its design
flow, high pressures within the pumped zone can result. Possible solutions include a
variable-speed pump, a storage tank, or a pressure relief valve that blows off water
pressure to the suction side of the pump when the discharge pressure becomes too
high.
Oversized Piping
Oversized piping can be difficult to identify because the system often appears to work
well. The adverse effects are excessive infrastructure costs and potentially poor water
quality due to long travel times. If a pipe is suspected of being too large during a
design study, its diameter should be decreased and the model rerun for the critical
condition for that pipe (peak hour or fire flow). If the pressures do not drop to an
unacceptable range, the pipe is a candidate for downsizing.
Figure 8.5 shows a comparison of 6-in. (150-mm), 8-in. (200-mm), 12-in. (300-mm),
and 16-in. (400-mm) pipes providing water to an area on a peak day. The pressure
graph shows that a 6-in. pipe is too small to deliver good pressure during peak times,
but the 8-in. pipe experiences an acceptable drop. Increasing the pipe size to 12 in.
(300 mm) or 16 in. (400 mm) does not result in significantly improved pressure for
the increased cost. This problem can also be viewed in terms of head loss (as in Figure
8.6), which shows there is virtually no head loss in the 12-in. (300-mm) or 16-in.
(400-mm) pipe, and the loss in the 8-in. (200-mm) pipe is acceptable.
Figure 8.5
Pressure comparison
for 6-, 8-, 12- and 16-
55
in. pipes
16 in. 12 in.
50
45
8 in.
Pressure, psi
40
6 in.
35
30
25
20
0 10 20 30 40
Time, hr
Figure 8.6
Head loss comparison
40 for 6-, 8- and 16-in.
pipes
35
6 in.
30
Friction Headloss, ft
25
20
15
10
8 in.
5 16 in.
12 in.
0
0 10 20 30 40
Time, hr
8.3 PUMPED SYSTEMS

Most water distribution systems are fed through some type of centrifugal pump. From
a modeling standpoint, the type of pump (for example, vertical turbine or horizontal
split-case) is not as significant as pump head characteristics, the type of system in
which the pump operates, and how the pump is controlled. Figure 8.7 shows some
pumping configurations for various systems.
Figure 8.7
Pumping
configuration
alternatives
PRV
1) Pump Directly into Closed System 2) Pump through
(Constant or Variable Speed) Pressure Reducing Valve
Tank
PSV 4) Pump into

3) Pump with Relief Valve Hydropneumatic Tank
to Control Pressure
PSV
5) Pump with Storage Tank

Floating on System 6) Pump with Pumped Storage
When serving a pressure zone through a pump station or by pumping directly from a
well, a number of different methods of operation may be used:
• Pump feeding directly into a closed system
• Pump feeding through a PRV
• Pump with a pressure relief valve
• Pump feeding a system with a hydropneumatic tank
• Pump feeding a system with a tank floating on the system
• Pump feeding a system with a pumped storage tank (not floating on the system)
Section 8.3 Pumped Systems 311
The early parts of this section refer to design problems in which the pumps will take
suction from a source with an adequate and relatively constant HGL [less than 20 ft (6
m) variation], such as a tank or treatment plant clearwell. Situations in which the suc-
tion HGL can vary significantly, or the NPSH available (see Figure 2.18 on page 49)
is marginal, raise other issues that are addressed at the end of this section.
In the initial modeling of most pumped systems, the engineer may first want to repre-
sent the pump discharge as a known HGL elevation that the pump station will main-
tain (that is, model it as a reservoir). Steady-state runs for high-demand or fire flow
conditions should be used to set this known HGL and to size pipes. The pipes should
be sized so that the head loss during peak times is acceptable [for example, velocity
less than 5 ft/s (1.5 m/s)], and the HGL set so that the pressures are within a desirable
range of 40 psi (280 kPa) to 80 psi (550 kPa) [30 psi (200 kPa) to 100 psi (690 kPa) in
hilly areas]. If a large range of elevations will be served, the system may be divided
into more than one pressure zone (see page 334).
Figure 8.8
Pump station
After the pump(s) have been selected using system head curves (see page 342), the
HGL in the model can be replaced with the actual pump curve data and the suction
side of the pump connected to the upstream system piping or boundary node. Then a
set of steady-state runs is made for minimum-, average-, and maximum-day demands.
Special consideration should be paid to situations in which the variability of flows is
large or new construction in the pressure zone is going to occur gradually. In such
cases, the design may include specifying several pumps of different sizes, or choosing
a pump station design with an empty slot so that an additional pump may be installed
at a later time.
The design engineer is primarily concerned with selecting the correct pump(s), and
pump control is an operational issue. However, the designer must have a good under-
standing of how the pumps may be operated to size them properly. An EPS run is the
best way to understand the effect of pump controls and to study pump operation. If the
system includes one or more tanks, the storage should be evaluated using EPS runs to
check tank turnover and pump cycling. A few fire flow scenarios can be used to
ensure that the tanks are able to recover relatively quickly after a high-demand period
or fire. The designer should also check pump suction pressures in the model. These
pressures are especially critical in situations involving long suction lines.
In summary, to model a new water distribution system with a pump, follow these
basic steps:
1. Choose an HGL elevation that will initially serve as the pump discharge head,
and locate tank(s).
2. Using steady-state runs for high-demand or fire flow conditions, size the pipes to
achieve acceptable head losses during high-demand conditions.
3. Develop system head curves from the steady-state runs and select the pump(s)
using these system head curves.
4. Replace the constant head node in the model with actual pump data.
5. Test the system using steady-state runs of minimum-, average-, and maximum-
day demands, and fire flow analyses.
6. If the system includes storage, also perform EPS runs for minimum, average day,
and maximum-day demands to check tank and pump cycling.
7. Perform EPS runs with fire flows to check tank recovery, pump cycling, and
pump suction pressures.
Pumping into a Closed System with No Pressure

Control Valve
Most pressure zones contain some storage or are fed by variable-speed pumps. Occa-
sionally, there are too few customers or there is not enough power consumption to jus-
tify a tank or variable-speed pump. This situation may occur in small systems such as
trailer parks, recreation areas, or isolated high points within larger distribution sys-
tems.
The simplest way to provide water to a closed pressure zone is by using a constant-
speed pump and no storage. Although this option is the least costly, the pump does not
function efficiently much of the time and can easily over-pressurize the system during
low-demand periods. For example, a pump selected to run efficiently at peak demand
may run at an efficiency of 30 to 50 percent during periods of low demand. Therefore,
constant-speed, dead-end pumping tends to minimize capital costs but results in
higher energy costs.
When designing a closed system with no pressure control, the engineer must pay spe-
cial attention to ensure that the pump selected does not over-pressurize or under-
pressurize the system. The pump should be selected such that the shutoff head is only
slightly higher than the head at the pump’s best efficiency point. After the pump curve
data have been entered into the model, system pressures should be checked at various
usage levels. These pressures can be examined using either multiple steady-state runs
or a small set of EPS runs having a wide range of demand patterns.
After checking the pressures, the power consumption at various pump operating
points should be examined. Using the power consumption and the cost of energy, the
designer can estimate the cost of running the pump at each operating point. If the
pump spends a great deal of time on inefficient operating points, then it may be cost-
effective to install storage tanks or pressure controls to increase efficiency. Alterna-
tively, the engineer may choose to use three small pumps rather than two large ones.
For example, if the peak flow is 200 gpm (0.0126 m3/s) and the average flow is about
75 gpm (0.0047 m3/s), then three 100-gpm (0.0063 m3/s) pumps, or two 200-gpm
(0.0126 m3/s) pumps with a 75 gpm (0.0047 m3/s) jockey pump (that is, a small pump
used to maintain pressure in a closed system) could be used instead of two 200-gpm
(0.0126 m3/s) pumps. The capital costs will be slightly higher, but operating costs will
be lower, resulting in a net savings over the life of operation.
Pumping into a Closed System with Pressure Control

If the pumps tend to over-pressurize the system during all but peak-use periods, then
some type of pressure control may be needed. The first option is to install a pressure
reducing valve (PRV) on the discharge side of the pumps. Although doing so is waste-
ful in terms of energy, the initial costs are fairly low, and the downstream pressure will
be corrected. This option is easily modeled by inserting a PRV onto the pump’s dis-
charge pipe or, if there are multiple pumps in parallel, downstream of the node where
the discharge pipes tie together.
A more effective solution may be to install a pressure relief valve that bleeds off water
and pressure from the discharge side to the suction side of the pump during low-
demand periods. One advantage of this solution is that this valve can be much smaller
than the PRV described above. For example, if the station pumps approximately 500
gpm (0.0316 m3/s), a 4-in. (100-mm) to 6-in. (150-mm) PRV will be needed, but the
relief valve can be as small as 2 in. (50 mm) and therefore less costly.
The relief valve can be modeled as a pressure sustaining valve (PSV) set to the pres-
sure (or HGL) that is to be maintained on the discharge side of the pump. At high flow
rates, the valve stays shut, and at lower flow rates, the valve opens enough to relieve
pressure. Examining the range of pressures from an EPS run is a good way to check
the valve operation. Different combinations of pump sizes and valve settings can be
evaluated using the model to determine which works best. Figure 8.9 shows how pres-
sures can climb in a system that does not have any storage or variable-speed pumping,
compared to a system equipped with a pressure relief valve (modeled as a pressure
sustaining valve).
Variable-Speed Pumps
Variable-speed pumps are frequently used in systems that do not have adequate stor-
age. Their use increases the initial capital cost of pumping stations as well as mainte-
nance expenses; thus, the capital and operating costs should be compared to other
alternatives before implementation. In small pressure zones, the expense of installing,
maintaining, and operating a variable-speed pump is dwarfed by the cost of installing
additional storage facilities.
Figure 8.9
Pressures when
140
pumping into a dead-
end system (with and
without a relief valve 120
on the pump) No Relief Valve
100
Pressure, psi
80
Relief Valve
60
40
20
0
0 6 12 18 24 30 36 42 48
Time, hr
Variable-speed pumps can prevent over-pressurizing of the water distribution system

in a pressure zone that has no storage floating on the system (that is, no tanks where
the HGL in the tank is the same as the HGL in the system). A variable-speed pump
can be reasonably efficient, although not as efficient as a properly sized constant-
speed pump with a storage tank.
The model can help the designer to select the pump and determine the HGL (pressure)
that the variable-speed pump will try to maintain. There are several ways to model
variable-speed pumps. Speeds can be set based on a time condition or a logic-based
control. Some models free the user from the need to specify the speed during model
input by enabling the user to specify the head that must be maintained at some other
node in the system. The model will automatically determine the speed necessary to
achieve that head while meeting demands. This problem is mathematically difficult
because two distinct sets of equations must be used: (1) equations for the pump run-
ning at full speed, and (2) equations for when the speed is controlled by the variable-
speed drive (Haestad Methods, 2002).
If the engineer is analyzing only a single pressure zone in a steady-state run and if the
model does not have a specialized feature for modeling variable-speed pumps, the
simplest approximation for a variable-speed pump is to treat the pump as a constant-
head node, setting the head equal to the discharge HGL that the pump is trying to
maintain. This approach works as long as the pump is never expected to reach 100
percent speed. If the pump is modeled in this fashion and achieves full speed, the
model does not accurately account for the pump running out on its curve.
If the pump is going to reach full speed during some situations and is controlled by
the pressure immediately downstream, then a corrected pump curve can be used to
model the pump as long as the suction head does not change markedly. This effective
pump curve is flat at low flows and curves downward once full speed is reached. The
designer must select a pump that can deliver the maximum flow without a significant
drop in pressure, and select the HGL to be maintained that will result in the best range
of pressures.
The development of an effective pump curve for a variable-speed pump is illustrated
with this example. Note that the full-speed pump curve is shown in Figure 8.10 as a
dashed line. Assuming that the suction head is 708 ft (216 m), the pump station is
located at an elevation of 652 ft (199 m), and the target discharge pressure set at the
variable-speed control is 90 psi (610 kPa), the effective pump curve can be deter-
mined as the solid line using Equation 8.2. Beyond a flow rate of 200 gpm (12.5 l/s),
the pump can no longer maintain 90 psi (610 kPa) so it behaves like a constant-speed
pump. Below that flow rate, it maintains a constant discharge head independent of
flow using the variable-speed drive. The total dynamic head (TDH) in the flat portion
is determined as
TDH = Z pump + 2.31P set – h suc (8.2)
where TDH = total dynamic head in flat portion of effective curve (ft, m)
Zpump = elevation of pump (ft, m)
Pset = discharge pressure setting (psi, kPa)
hsuc = suction head (ft, m)
For the case described previously, TDH = 652 + 2.31(90) - 708 = 152 ft.
Figure 8.10
Effective pump curve
250
head, pump
200
head, variable
150
Head, ft
100
50
0
0 100 200 300 400 500
Flow, gpm
An old approach for simulating the behavior of variable-speed pumps is to specify the
full-speed pump curve. Then a PRV downstream of the pump can be used to regulate
the head to the setting of the variable-speed pump.
Pumping into a System with a Storage Tank

A storage tank is considered to be “floating on the system” if the HGL in the tank is
generally the same as the HGL in the system. Pumping into a system with a storage
tank that floats on the system, whether that tank is an elevated tank or a ground tank
on a hill, usually represents very efficient operation.
A pump discharging into a closed system (meaning there is no storage) must respond
instantaneously to changes in flow because there is no equalization storage. This
immediate response is not necessary when pumping into a zone with a storage tank
that floats on the system. In such cases, a more efficient and less costly constant-speed
pump can be used. The pump can be selected to operate at its most efficient flow and
pressure, thus eliminating the inefficiencies associated with variable-speed drives.
Furthermore, if there is sufficient storage floating on the system, the pressure zone
can respond to power outages without the need for a costly generator and transfer
switch.
The pump should be selected so that the operating point will be very close to the best
efficiency point of the pump. EPS runs can be used to determine how pump controls
should be set and to ensure that the pump is operating efficiently under virtually all
conditions. EPS runs should be at least 48 hours in duration to show that the pumps
can refill the storage tank even during a stretch of two or more maximum or near-
maximum demand days. Performing EPS runs that show tank water levels recovering
after a fire or power outage is also helpful. The tank water level should be able to
recover within a few days of the emergency.
If there are several tanks in a single pressure zone, it may be difficult to efficiently
operate the system in a way that takes full advantage of both tanks without encounter-
ing a difficulty with preventing one tank from overflowing while keeping the other
from draining. These operation problems are discussed further on page 339.
Pumping into Closed System with Pumped Storage

With pumped storage, the distribution storage (not the well or plant clearwell) has a
head lower than the hydraulic grade line required by the system, so water must be
pumped out of the tank to be used. An example would be a ground-level tank in flat
terrain. Such tanks may be attractive in certain instances because they have a lower
capital cost and less visual impact than elevated tanks. At times, this type of arrange-
ment may be the only way of incorporating an existing tank into a larger system after
annexation or regionalization.
In these cases, pumping is required to move water from the tank into the distribution
system. Therefore, operating costs are greater when compared to systems with tanks
that float on the system. In addition, the expense of this type of tank configuration
includes the capital and operating costs of a generator, transfer switch, valving, and
controls so that the system can operate during power outages. Because the HGL of the
system is higher than the water surface elevation in the tank, filling the pumped stor-
age tank wastes energy that must be added again when water is pumped out of the
tank. The amount of energy lost depends on how much lower the water level in the
tank is compared to the system HGL.
Running steady-state models of a pressure zone with pumped storage is complicated
because there are really five different modes of operation, as presented in Table 8.1. In
this table, the term source pump refers to the pump from the well, clearwell, or neigh-
boring zone into the pressure zone where the storage facility is located. The pumped
storage pump pressurizes water from the storage facility for delivery to the customers
within the pressure zone.
Table 8.1 Pump operation modes when pumping into a closed system
Mode Source Pump Pumped Storage Pump Notes
1 On On Peak demand period
2 Off On Storage providing water
3 On Off Pumped storage filling
4 On Off Pumped storage full or off-line
5 Off Off Alternative supply
Note that the fifth case above is only feasible if there is another storage tank floating
on the system or an alternate water source (for example, a PRV from a higher zone);
otherwise, turning both sets of pumps off leaves customers without water.
The list of cases in this section is an oversimplification in that there may be numerous
combinations of source pumps and pumped storage pumps in a real system. In some
situations, there might not be a “source pump” at all, and the system may actually be
fed by gravity, such as from a treatment plant on a hill or through a PRV from a neigh-
boring pressure zone. In any case, areas of the system with very high or very low pres-
sures must be identified to determine the required pump discharge heads. Unless there
is floating storage, the pumps used in pumped storage systems are usually variable-
speed pumps, and the designer needs to consider how to set the controls, as well as
how to select the pumps.
The actual tank fill time is a very important consideration when planning the filling
cycle of a pumped storage tank. If the tank fills up too quickly, it will depress the
hydraulic grade line (and pressures) in its vicinity. Conversely, if the tank fills too
slowly, water may not be available for pumping when it is needed. The tanks are usu-
ally fed through a pressure sustaining valve, as shown schematically in Figure 8.11.
The engineer should experiment with different settings for the PSV to determine a set-
ting that fills the tank at an adequate rate without adversely affecting pressure. Fur-
thermore, the speed at which tanks fill and drain can have a significant impact on
water quality within the tank volume.
After pumps, pump controls, and PSV settings have been specified, an EPS run can be
performed for a duration of at least 48 hours, for both maximum and average day con-
ditions, to guarantee that the system will work as designed. In particular, the operating
points of the various pumps should be checked for problems. For example, a storage
pump may run efficiently when operating alone, but if it runs with the source pump
on, the elevated pressure on its discharge side may back it off to an inefficient point on
the pump curve. Conversely, the storage pump may run correctly when running with
the source pump, but it may then run out to a very high flow when the source pump
shuts off. These inefficiencies can overload motors and waste energy.
Figure 8.11
Pumped storage tank HGL
fed through a PSV
Head Loss
Through Valve
Min. HGL Ground

at PSV Tank
HGL
//=//=//=//= //=//=//=//=//=//=//=//= //=//=//=//= //=//= //=//= //=//= //=//=//=//=
Fill Line
PSV
Pumped storage systems are easy to run, but difficult to run efficiently. This ineffi-
ciency is due to the fact that the pumps may be working against one system head
curve when they are running by themselves, and against a much different system head
curve when they are running with the other pumps. Selecting a pump that is sized
according to the largest head and relying on the variable-speed drive to control the
pump at other times is usually the best solution.
Pumping into Hydropneumatic Tanks

Hydropneumatic tanks are pressure tanks that can be used to store water at the correct
HGL using pressure head rather than elevation head. Because pressure tanks are
expensive, they are used only for small systems that are not required to meet fire
flows. Capital costs for hydropneumatic tanks are high compared with variable-speed
pumping or the installation of a pressure relief valve. Using this type of tank, how-
ever, allows pumps to operate more efficiently than when using no storage at all. A
hydropneumatic tank also provides surge protection and additional storage in the
event of a power outage.
Once a model for the hydropneumatic tank has been developed, it can help in select-
ing pumps, determining pump control settings, and evaluating the active storage vol-
ume in the tank. Other methods (available from tank manufacturers) are required to
determine maximum and minimum air volumes in the tank.
EPS model runs allow the cycle times of pumps for various flow rates to be evalu-
ated. One of the criteria for selecting pumps is the maximum number of starts per
hour, and because hydropneumatic tank volumes are small, such criteria can be criti-
cal. Usually, the shortest cycle time occurs when the system demand is half of the
pump production.
Well Pumping
Well pumping is similar to most of the other types of pumping previously described in
this chapter. An important difference is that the pump suction head will vary due to
the drawdown of the water table (piezometric surface) in the vicinity of the well as
water is pumped from it. The greater the flow rate through the pump, the larger the
drop in water table elevation.
In a model, a well is represented as a reservoir that is connected to a pump by a very
short piece of suction pipe. In the actual well, the pump is submerged, so there is no
suction pipe; however, pumps must be connected to a pipe for modeling purposes.
The riser pipe that extends from the pump to the ground surface is usually smaller
than the distribution piping and can contribute significantly to head loss. Figure 8.12
is a schematic showing how to model a well.
In very porous aquifers, the amount the water table drops during pumping may be
negligible, and the well can be represented by the reservoir alone. In most cases, how-
ever, the water level in the well experiences significant drawdown due to pumping,
and this drawdown is relatively linear with respect to flow rate (that is, pump flow rate
divided by well drawdown is equal to a constant). To model a drawdown situation, the

pump curve is adjusted by subtracting the amount of drawdown from the pump curve to
create a new “effective” pump curve for use in the model, as shown in Figure 8.13.
Figure 8.12
Representing wells in
Actual
the model
//=//= //=//=//=//= //=//= //=//=//=//=//=//=//=//=//=//

Riser Pipe Distribution
Discharge Line System
Well
Water Table
Drawdown
Well Pump
Model
Distribution
Node
Discharge Line System
Riser Pipe
Reservoir
Well Pump
Short Pipe
Figure 8.13
Adjustment of pump
curve for well
250
pumping
200
Pump Curve
150
Head, ft
Effective Pump
100 Curve
50
Drawdown
0
0 50 100 150 200 250
Flow, gpm
One problem with modeling wells is that groundwater tables can fluctuate for a wide
variety of reasons. Seasonal variations in water usage and recharge and varying con-
sumption rates of neighboring wells that use the same aquifer can contribute to fluctu-
ations in groundwater tables. Regardless of the cause, the static groundwater table (or
the model’s reservoir level) must be adjusted for the situation being considered. For
cases in which the groundwater table fluctuates significantly during the year [say,
more than 20 ft (6 m)], the designer must use the model to check pumps against the
full range of water table elevations. The pump that is selected needs to work with the
lowest water table and yet not overload the motor or over-pressurize the distribution
system when the water table is high. When the water table’s elevation range is very
large, the designer may want to install flow control valves and/or pressure regulating
valves on the discharge piping from the well.
Usually, one of the key decisions in installing the well is whether to pump directly
into the distribution system or into a ground or elevated tank (see Figure 8.14). The
ground tank alternative is more expensive because a tank and distribution pump are
required in addition to the well pump. However, contact time requirements for disin-
fecting the water, if necessary, can be met through the use of ground tanks, eliminat-
ing the need for large buried tanks or pipes. Also, the ground tank can store more
water for fire protection at a lower cost than can an elevated tank. Using a ground tank
with a well also provides some reliability in case the well should fail, because the dis-
tribution pumps at the tank can be placed in parallel. Because of the space constraints,
well pumps cannot be placed in parallel without the construction of multiple wells.
Figure 8.14
Well pumped to
Storage or Distribution
ground storage tank Treatment Pumps
Distribution
System
//=//= //=//= //=//= //=//=//=//
Riser Pipe
Well
Well Pump
Pumps in Parallel
In general, pump stations should, at a minimum, be capable of meeting downstream

demands when the largest pump is out of service. In small pump stations, there are
usually two pumps, either of which can independently meet demands. In large sta-
tions, it is common to provide additional reliability and flexibility by having more
than two pumps. If the pump station is to be operated such that different combinations
of pumps will be run under different demand conditions, it is important that the
pumps be selected to work efficiently when operating both alone and in parallel with
the other pumps.
A major factor affecting pump efficiency is the capacity of the piping system
upstream and downstream of the pump station. This capacity is reflected in the system
head curve. A flatter slope on this curve for a given discharge reflects lower system
head loss and ample pipe capacity. Conversely, when the slope of the curve is steep,
the ability of the pump to supply adequate flows is limited by the system piping. The
steepness of the system head curve determines the efficiency of running several
pumps in parallel.
The simplest way to evaluate pumps in parallel is to run the model of the system for
each different combination of pumps. For each combination, the operating point of
each pump should be near its best efficiency point. If a pump’s efficiency drops
significantly, the utility may want to select different pumps or avoid running that
combination.
Viewing the system head curves and pump head curves for parallel pump operation
provides a better understanding of what occurs in the system. For example, Figure
8.15 shows pump curves for two identical pumps in parallel. If there is ample piping
capacity (that is, the system head curve is fairly flat), each pump can discharge 270
gpm when operating alone, and the pair running together can discharge 500 gpm. If
piping capacity is limited, however, each pump will produce 180 gpm individually,
and the two together will produce only 220 gpm. The reason for such a small increase
in discharge when the second pump is added is a lack of capacity in the distribution
piping, not a lack of pump capacity.
Figure 8.15
Two identical pumps
in parallel
200
System With
180 Limited Pipe
Capacity
160
140 Two Pumps
120
Head, ft
100 System With

Ample Pipe
Capacity
80
One Pump
60
40
20
0
0 100 200 300 400 500
Flow, gpm
The problem becomes more complicated when the pumps are not identical, as shown
in Figure 8.16. In this case, Pump A is run when high flows are needed, and Pump B
is run during low-flow conditions. When the system head curve is flat, Pump A alone
delivers 270 gpm, Pump B alone delivers 160 gpm, and the two together deliver 380
gpm. For the steeper system head curve, however, Pump A alone produces 180 gpm,
Pump B alone produces 120 gpm, but the pair only produce 180 gpm. The reason for
this lack of increase in discharge is that Pump A produces pressures in excess of
Pump B’s “shutoff head,” so Pump B cannot contribute. Such a combination of sys-
tem head characteristics and pumps should be avoided. The model run with these
pumps will show a zero or very low discharge from Pump B.
Figure 8.16
Two different pumps
in parallel 200
180 System With

Limited Pipe
Capacity
160
System With
140 Ample Pipe
Pump A+B Capacity
120
Head, ft
100
80
Pump A
60
Pump B
40
20
0
0 100 200 300 400 500
Flow, gpm
Head Loss on Suction Side of Pump

The discussion thus far in this section has centered on the hydraulics of pumps as con-
trolled by the downstream system. Upstream piping is not usually as critical, because
designers typically try to locate pumps near the storage facility or source from which
they are drawing water to help reduce head losses. If the suction piping is long or
lacks capacity, however, problems may occur due to the elevation of the pumps and
the head losses in the suction piping. If the suction head is too low, the pump can
experience problems with cavitation. Additionally, there could be difficulties keeping
the pump primed.
The design of any pump requires that the suction head available be compared to the
net positive suction head (NPSH) required. The NPSH available depends on the
hydraulic grade elevation of the source, the elevation of the pump, and the head loss
on the suction side of the pump. The NPSH required is a function of flow rate and
pump properties as measured and documented by the manufacturer.
NPSH available is equal to the sum of atmospheric pressure at the pump and the static
head (gage pressure) measured on the suction side of the pump, minus the water vapor
pressure and the sum of the head and minor losses (velocity head is often negligible).
For the simple situation in which the pump takes suction directly from a tank, the
NPSH available is given by Tchobanoglous (1998).
NPSH a = H bar + H s – H vap – h loss (8.3)

where NPSHa = net positive suction head available (ft, m)

Hbar = atmospheric pressure (at altitude of pumps) (ft, m)
Hs = static head (ft, m) (water el. on suction side of pump – pump el.)
Hvap = water vapor pressure (corrected for temperature) (ft, m)
hloss = sum of head and minor losses (from suction tank to pump) (ft, m)
For the situation in which the distance from the suction tank is large and the suction
piping is complex, the model can be used to determine the (Hs - hloss) term by subtract-
ing the pump elevation from the HGL on the pump’s suction side.
NPSH a = h suc – Z pump + H bar – H vap (8.4)
where hsuc = HGL at suction side of pump as calculated in model (ft, m)

Zpump = elevation of pump (ft, m)
To determine the HGL on the suction side of the pump, it is helpful to model a node
immediately upstream of the pump. The value of barometric pressure is primarily a
function of altitude, although it also varies with weather. Vapor pressure is primarily a
function of temperature. Standard values are listed in Tables 8.2 and 8.3 (Hydraulic
Institute, 1979).
Table 8.2 Standard barometric pressures

Elevation Elevation Barometric Pressure Barometric Pressure
(ft) (m) (ft) (m)
0 0 33.9 10.3
1,000 305 32.7 9.97
2,000 610 31.6 9.63
3,000 914 30.5 9.30
4,000 1,220 29.3 8.93
5,000 1,524 28.2 8.59
6,000 1,829 27.1 8.26
7,000 2,134 26.1 7.95
8,000 2,440 25.1 7.65
Table 8.3 Standard vapor pressures for water

Temperature Temperature Vapor Pressure Vapor Pressure
(qF) (qC) (ft) (m)
32 0 0.20 0.061
40 4.4 0.28 0.085
50 10.0 0.41 0.12
60 15.6 0.59 0.18
70 21.1 0.84 0.26
80 26.7 1.17 0.36
90 32.2 1.61 0.49
100 37.8 2.19 0.67
If the NPSH available is found to be less than the NPSH required, the designer must
choose one of the following options to correct the problem and avoid cavitation:
• Lower the pump
• Raise the suction tank water level
• Increase the diameter of the suction piping to reduce head loss
• Select a pump with a lower NPSH requirement
The problem of meeting NPSH requirements can be particularly tricky when the suc-
tion line from the nearest tank is long. Using a model, the designer can try different
piping and pump station location combinations to prevent NPSH problems from
occurring in the real system.
8.4 EXTENDING A SYSTEM TO NEW CUSTOMERS

One of the most common water distribution system design problems is laying out and
sizing an extension to an existing system. This section focuses on modeling new pip-
ing that will become part of an existing system (without a meter or backflow preven-
ter) and that has a connection point that is not a tank or pump station. The new piping
might be for a residential subdivision, industrial park, shopping mall, mobile home
park, prison, school, or mixed-use land development.
Usually, the hydraulic demands placed on the new piping where the build-out is going
to occur are known with greater certainty than master planning demand projections
can provide. Frequently, when sizing new piping for a system extension, fire flow
demands are more significant than peak hour demands.
Extent of Analysis
The difficult part of sizing new piping is that it cannot be sized independently of the
existing distribution system. HGLs in a new extension are a function of existing pip-
ing and customer demands, as well as new and future customers in the same general
area. Ignoring the existing network performance during the design process would
yield poor results. Therefore, the best approach is to add the new piping to a cali-
brated model of the existing system.
Elevation of Customers
Before beginning the process of pipe sizing, the engineer needs to determine the ele-
vations of the properties that will receive service to ensure that the water pressures
there will be within a satisfactory range. Ideally, if a model of the existing system is
available, EPS runs can help the designer to define the range of HGLs and pressures
that may occur in the vicinity of the new piping. With or without a model, pressure
readings should be taken where the new system will connect to the existing system so
the general HGL can be determined. Knowing this range and the maximum and mini-
mum acceptable pressures during non-fire situations, it is possible to approximate the
Section 8.4 Extending a System to New Customers 327
elevations (Figure 8.17) of the highest and lowest customers that can be served using
Equations 8.5 and 8.6:
Elmin = HGL max – C f P max (8.5)
El max = HGL min – C f P min (8.6)
where Elmin = minimum allowable elevation of customers in zone (ft, m)

HGLmax = maximum expected HGL in pressure zone (ft, m)
Pmax = maximum acceptable pressure (psi, kPa)
Elmax = maximum allowable elevation of customers in zone (ft, m)
HGLmin = minimum expected HGL in pressure zone (ft, m)
Pmin = minimum acceptable pressure (psi, kPa)
Figure 8.17
Maximum Pressure Minimum Pressure Limits of pressure
zone
HGLmax
HGLmin
Highest Customer
//=//= //=//=
Lowest Customer
//=//= //=//=
Example — Range of Customer Elevations. If the HGL in a pressure zone normally

varies between 875 and 860 feet, the pressure varies between 35 and 100 psi, and the maximum pres-
sure is 100 psi, what is the range of ground elevations that can be served?
Elmin = 875 - 2.31(100) = 644 ft
Elmax = 860 - 2.31(35) = 780 ft
If some of the area served is above the determined elevation range, then pumping or
an alternative water source will most likely be required. Conversely, if some of the
area is below the elevation range, a PRV or alternative source may be required (see
page 427).
In Figure 8.18, suppose that the lowest customer is at an elevation of 700 ft and can-
not have pressures above 100 psi. According to Equation 8.5, in order to experience
these pressures, the HGL must be less than 931 ft. Suppose also that the highest cus-
tomer is at an elevation of 890 ft and must have pressures greater than 30 psi. Accord-
ing to Equation 8.6, in order to maintain this minimum pressure at this elevation, the
HGL must be at least 960 ft. Because no value of HGL exists that is greater than 960
ft but less than 931 ft, the two customers must be served from different pressure
zones.
Performing this calculation before modeling helps give the modeler an appreciation of
the types of problems that are likely to be encountered. Thus, some alternatives can be
immediately eliminated, such as trying to serve a customer at too high an elevation for
the pressure zone by using very large pipes. See page 334 for more information on
creating new pressure zones.
Figure 8.18
Customers must be Minimum Pressure
served from separate Pressure
pressure zones Zone
Boundary
HGLmin In Upper Zone
Highest Customer
Maximum Pressure
//=//= //=//=
HGLmax In Lower Zone
Lowest Customer
//=//= //=//=
Assessing an Existing System

Because of the interaction between the new piping and the existing system, an impor-
tant first step in analyzing a system extension is to conduct a hydrant flow test in the
vicinity of the connection to the existing system. This test provides data for model
calibration and a quick preliminary assessment of system strength. Performing a
hydrant flow test is described in detail in Chapter 5 (see page 184), and in AWWA
(1989).
A minimum of three values must be recorded during a hydrant flow test: static pres-
sure (Ps), test pressure (Pt), and test flow(s) (Qt). The pressures should be measured at
the residual hydrant, and the flows should be measured at the flowed hydrant. To con-
vert the pressure to hydraulic grade line, the elevation of the residual hydrant must be
accurately determined. In addition, it is helpful to record which pumps were running
during testing, the tank water levels, and information regarding any special conditions
in the system when the test was run (for instance, pipe breaks or fires).
Figure 8.19 shows the results of an example flow test. The static pressure of the exam-
ple flow test was 60 psi or 139 ft (614 kPa or 42 m), the hydrant flow was 500 gpm
(31.5 l/s), and the residual pressure was 35 psi or 81 ft (241 kPa or 25 m). The flow
value for the point where the graph intersects the horizontal axis (or any other pres-
sure) can be determined from the following equation:
P s – P 0 0.54
Q 0 = Q t §© -----------------·¹ (8.7)
Ps – Pt
where Q0 = flow at pressure P0 (gpm, m3/s)

Qt = hydrant test flow (gpm, m3/s)
Ps = static pressure during test (psi, kPa)
P0 = pressure at which Q0 is to be calculated (psi, kPa)
Pt = residual pressure during test (psi, kPa)
Note that this equation may be used with any units, as long as they are consistent (that
is, all flow units and all pressure units are the same). By inserting P0 = 0 into the pre-
ceding equation, the horizontal intercept can be back-calculated. The horizontal inter-
cept is determined to be 802 gpm (50.6 l/s), as shown in Figure 8.19.
Figure 8.19
Plot of hydrant flow
test data
160
Static HGL
140
120
100
Pressure Head, ft
Residual HGL
80
60
40
20
Flow at Po
0
0 200 400 600 800 1000
Flow, gpm
Using a P0 value of 20 psi (140 kPa), the value of Q0 will give an indication of the
strength of the system. The value of Q0 at 20 psi (140 kPa) should be significantly
greater than the maximum demand for the new pipes. Otherwise, considerable
improvements may be needed in the existing system.
When using hydrant tests to assess the existing system for expansion, the locations of
the hydrants being tested are important. The residual hydrant should be located
between the source and the flowed hydrant such that most of the water being dis-
charged from the flowed hydrant passes by it, as shown in Figure 8.20(a). Otherwise,
the results can be misleading, especially if the flowed hydrant is on a significantly
larger main than the residual hydrant.
Figure 8.20
Reservoir/tank and
pump approximation
Flowed
Hydrant
New Customers
Residual
Hydrant
Source
a. Actual Layout
Pipe to
Tank and Pump at New Customers
Residual Hydrant Location
Representing Remainder of System
b. Representing System by Tank and Pump

After a fire hydrant flow test has been conducted, the data can be used to model the
new piping by using one of the following three basic approaches:
• Add the proposed pipes to a current model of the existing system or an
appropriately skeletonized version of the model and verify the model with
the fire flow test data.
• Build a skeletal model of the existing system and add the proposed pipes
onto it.
• Use the hydrant flow test results to approximate the existing system by an
equivalent reservoir and pump.
The sections that follow discuss these approaches in detail. Note that setting an arbi-
trary HGL (based on a single pressure reading) where the system extension ties to the
existing system is almost never the correct way to model the existing system.
Building onto an Existing Model. The best way to model an extension of a
water system is to build the new pipes and customers into a calibrated model of the
existing system. In this way, it is possible to model both the effect of the existing sys-
tem operation on the new piping and the effect of the new piping on the existing sys-
tem. Having a calibrated model of the system also allows a wide variety of situations
to be simulated, such as future year peak day demands, fires at various locations, and
failures of important pipes.
The engineer designing the system extension, however, may have been hired by a land
developer and will probably not have much interest in studying the existing system. If
the utility already has a calibrated model of the system, the most straightforward solu-
tion would be to make it available to developers to add on to the existing model.
Unfortunately, it is often administratively difficult for the design engineer to utilize
the existing model, either because of incompatible water distribution modeling soft-
ware, or because the utility may not want to share its model. Because the utility can
use information that the engineer does not have to evaluate the proposed system, sev-
eral design and review iterations may be required to develop the best solution.
Skeletal Model of Existing System. If the model of the existing system
cannot be made available to the design engineer, the designer should construct a skel-
etal model of the portion of the existing system affecting the new design. The skeletal
model must begin at a real water source, such as the pump or tank, which will serve as
the primary water source for the new extension pipes. It should be calibrated using the
results of fire hydrant flow tests, especially the tests conducted near the location
where the new extension will tie in.
This approach is not as accurate as using a more detailed existing model because of
the high degree of skeletonization involved, and because assigning future demands to
the model is a somewhat arbitrary process. For instance, the skeletal model probably
will not be calibrated as well since a model that was used previously will have been
tested under a wider variety of conditions. It is possible that the inherent inaccuracies
of this method could lead to substantial modeling errors, even if the initial tests and
calibration process indicated that the model seemed to be a reasonable representation
of the system. If, for example, the engineer for the developer was not informed about
planned projects and utility growth projections, the skeletonized model would yield
little design value.
Approximating a System as a Pump Source. The simplest technique

for modeling the existing system is to use a pump and reservoir to simulate conditions
at the tie-in point, as shown in Figure 8.20(b). As shown in Figure 8.19, the results of
a fire hydrant flow test look like a pump head curve. For modeling purposes, a reser-
voir node placed at the location of the residual hydrant from the flow test, with the
hydraulic grade set to the hydrant’s elevation, is sufficient for modeling an existing
system. This reservoir is then connected to the new system through a short pipe and a
pump. The pump is modeled using a three-point pump curve that is established using
hydrant flow test data. Using notation similar to that in Equation 8.7, the three points
from the pump curve are shown in Table 8.4. Note that the value 2.31 converts the
pressures (in psi) from the hydrant tests to head (measured in feet) for the pump
curve.
Table 8.4 Points on simulated pump curve

Head Flow
(ft) (gpm)
2.31Ps 0
2.31Pt Qt
2.31P0 Q0
Flow Qo is calculated by using Equation 8.7 with a given Po, generally assumed to be
20 psi (140 kPa) (other reasonable values for pressure can also be used). The hydrant
test can also be repeated with three different flow rates to obtain data from which to
generate a curve.
For rural systems without fire hydrants, an approximate test can be conducted by
opening a blow-off valve or flushing a hydrant and measuring the flow with a cali-
brated bucket and stopwatch. If this approach is not possible, then the best test that
can be performed is to place a chart recorder at the connection point and monitor fluc-
tuations in HGL. The model can then be calibrated to reproduce those conditions.
Using the pump approximation method can present problems because this approxima-
tion of the existing system only accounts for the exact boundary conditions and
demands that existed at the time that the test was run (for example, the afternoon on
an average day with one pump on at the source). Therefore, determining the effect of
changing any of the demands or boundary conditions is difficult. An EPS that is per-
formed using the pump approximation method will be less accurate and may not pro-
vide reliable data regarding projected changes in consumption. The pump
approximation approach only works well if the existing system is fairly built-out near
the connection point and the demand and operation conditions are expected to remain
essentially the same in the long run. The hydrant flow test is useful for predicting
changes in pressure when downstream demands change but not for evaluating other
types of system changes such as the addition of new pipes, or operational alternatives
such as fire pumps starting up.
Section 8.5 Establishing Pressure Zones and Setting Tank Overflows 333
8.5 ESTABLISHING PRESSURE ZONES AND SETTING

TANK OVERFLOWS
Selecting a tank overflow elevation is one of the most fundamental decisions in water
distribution system design. This decision sets the limits of the pressure zone that can
be served and the overall layout of the distribution system. Once a tank has been con-
structed, the limits of the hydraulic grade line within a pressure zone are fixed. The
only way to change the hydraulic grade line limits would be to replace, raise, or lower
the existing tank (an expensive proposition).
Before developing a design, the engineer needs to look at the terrain being served
with short-term and long-term usage projections in mind. The designer should con-
sider what the distribution system may look like 20 to 50 years in the future when the
area is completely built-out. This is true for dead-end systems that are served by
pumps or pressure reducing valves, and especially true of new pressure zones with
tanks.
Unlike PRVs that can be reset or pumps that can be replaced easily, tanks are rela-
tively permanent. Even for systems without tanks, customers become accustomed to a
certain pressure, or, more important, industrial equipment and fire protection systems
may have been designed and constructed to work with a given HGL. Any change to a
network boundary condition like a tank overflow can change the dynamics of the sys-
tem and must therefore be carefully analyzed and designed.
Establishing a New Pressure Zone

The decision to create a new pressure zone may be triggered by
• Construction of a new isolated system
• Customers moving into an area with an elevation that is too high or too low
to be adequately served from the existing pressure zone
• The utility wanting better control over an area
Choosing the boundaries for the pressure zone is done manually before beginning to
model the system. When laying out pressure zones, the designer should examine the
elevations of the highest and lowest customers to be served. If customers are less than
approximately 120 ft (37 m) apart vertically, then most likely a single pressure zone
can serve them. If the elevation difference is significantly greater, more pressure
zones are needed. In general, the elevation of the lowest and highest customers in the
service area and the limits of the range of acceptable pressures are used to determine
the HGL in a pressure zone. Equations 8.8 and 8.9 provide some useful guidelines for
selecting a HGL:
HGLmin > (Elevation of highest customer) + Cf Pmin (8.8)
HGLmax < (Elevation of lowest customer) + Cf Pmax (8.9)

where HGLmin = minimum HGL (ft, m)

HGLmax = maximum HGL (ft, m)
Pmin = minimum acceptable pressure (psi, kPa)
Pmax = maximum acceptable pressure (psi, kPa)
The first criterion (Equation 8.8) ensures that the highest customer will have at least
minimum pressure, while the second (Equation 8.9) ensures that the lowest customer
will not experience excessive pressures. In flat terrain, there will usually be a band of
possible HGL values that meet both criteria. In hilly terrain, however, because the ele-
vations of the highest and lowest customers are very different, it may be impossible to
find an HGL that satisfies both inequalities (see page 327). Usually, this much differ-
ence means that the proposed pressure zone should actually be two (or more) pressure
zones, or the lowest customers will have pressures in excess of Pmax.
The above rules pertain to pressures during normal conditions, not during fire flows
when head loss becomes significant. Additional analysis is needed to size piping and
ensure adequate pressures for such conditions. If there are only a small number of
customers with excessive pressures, some utilities require the customer to install indi-
vidual PRVs.
Laying Out New Pressure Zones

The need for pressure zones can be visualized as shown in Figure 8.21, which depicts
how pressure zones can be set up along a hill 500 ft (152 m) high. In this example, the
step size between pressure zones is set at 100 ft (30 m). Normally, the difference
between pressure zones should be between 80 ft and 120 ft (24 m and 37 m). Large
step sizes will either over-pressurize the lower customers in a zone or under-
pressurize the higher ones. Smaller step sizes require too many zones and, conse-
quently, an excessive number of pumps, tanks, and PRVs.
The topography in most areas does not generally look like the smooth slope in Figure
8.21, but instead has ridges and valleys. A good way to get a feel for the layout of
pressure zones is to choose the nominal HGLs of the pressure zones and identify the
elevation contour that corresponds to the boundary between pressure zones. A sample
of such a map is shown in Figure 8.22. The figure shows a topographic map with the
high and low limits of a new pressure zone. The areas between the high and low pres-
sure limits will be served by the zone. Those above the solid line will need to be
served by a higher zone, while those below the dashed line will need to be served by a
lower zone. The boundary lines are not ironclad limits, but they give a suggestion of
how the system should be laid out.
As with all the elements in a water distribution model, a naming convention should be
developed for pressure zones. Some possibilities are
• The part of town in which they are located
• The nominal HGL (overflow level of the tank)
• The primary pump station/PRV serving the zone
• The name of the tank in the zone
• The relative HGL in the zone
Figure 8.21
Profile of pressure
2,590 ft zones
2,500 ft
2,490 ft
40 psi
Pressure Zone HGL 82 psi
2,400 ft
2,390 ft
40 psi
82 psi
2,300 ft
2,290 ft
40 psi
82 psi
2,200 ft
2,190 ft
40 psi
82 psi
2,100 ft
PRV or Pump*
40 psi
2,000 ft
82 psi
* Depending on direction of flow
Table 8.5 provides some alternative naming schemes for the zones in Figure 8.21. It is
important to be consistent with a naming convention in order to avoid confusion down
the road. For example, suppose that the Oakmont pump station takes suction from the
Oakmont tank. In this case, is the Oakmont pressure zone the one the pump dis-
charges to or the zone the tank floats on?
With a map like Figure 8.22, the designer can begin to lay out transmission lines.
Major transmission mains within a pressure zone (not including those connecting a
zone to adjacent zones) should be laid roughly parallel to the contours and remain
within the elevations that can be served by that pipe. Of course, the layout of roads
and buildings may prevent this from actually occurring, but with a map such as this,
the designer can roughly determine where the mains ought to be laid, thus avoiding
having pipes far outside the defined pressure zone.
Figure 8.22
Pressure zone
topographic map
Upper Limit
of Zone
Lower Limit
of Zone
Pressure Zone
Table 8.5 Alternative naming conventions for pressure zones

HGL Part of Town Tank Name Pump/PRV Name Relative HGL
2,190 South Wilson St. Mundy’s Glen Low Service
2,290 Central Downtown Hillside Medium
2,390 North Oakmont Flat Road High
2,490 Northwest Liberty Hill Oakmont Very High
2,590 Far Northwest Hanover Industrial Park Rice Street Top
There will be situations in which it may be more economical to serve a new customer
through a PRV or a pump from a different pressure zone, rather than from a tank in
the same pressure zone. Figure 8.23 shows an example of a situation in which, at least
in the short run, it is better to serve customers at the location labeled “New Develop-
ment” from the higher pressure zone B rather than from the lower pressure zone A.
(Serving the area from zone A would require the costly installation of the long main
labeled “proposed pipe.”) Even though serving this area through the PRV wastes
pumping energy, the PRV is necessary so that new customers in the development can
be served at an HGL similar to that of pressure zone A. In this way, customers can be
served at the “correct” pressure, without the expense of installing the proposed pipe.
Figure 8.23
920 ft New development
near “wrong” pressure
zone
Zone B
820 ft
800 ft
Zone A
Pump
Zone B
720 ft Zone A
640 ft
PRV
610 ft
600 ft
PROPOSED PIPE
New Development
Tank Overflow Elevation

After the tank overflow elevation has been selected, the tank dimensions must be
determined. First, tank height is set based on an appropriate water surface elevation
range because this range has the greatest effect on defining the maximum and mini-
mum pressures within the pressure zone. The upper and lower elevations are set such
that adequate system pressures can be maintained at all tank levels. Because large
fluctuations in tank water level correspond to similar fluctuations in pressure, most of
the water in the tank should be stored within 20 to 40 ft (6 to 12 m) of the tank over-
flow. The analysis outlined here is the basis for determining the best location for a
tank and what its overflow elevation should be. It is the role of the modeling analysis
described later to determine if the piping is adequate to move the water through the
pressure zone.
For the tank in Figure 8.24, if the water that is stored below an elevation of 869 ft (265
m) is used, the pressure will drop below 30 psi (207 kPa). If the water level drops
below 846 ft (258 m), the water pressure will fail to meet the 20 psi (140 kPa) stan-
dard. Therefore, the storage volume below 846 ft (258 m) is “dead” storage, useful
only in that it elevates a portion of the tank’s volume to the acceptable elevation
range. The water level cannot drop into that range without adversely affecting pres-
sure. Furthermore, this excess volume can lead to long detention times and contribute
to water quality problems.
To avoid water quality problems and wasted tank volume, storage must be placed at
the correct elevation. In general, designers try to install all storage as “effective” stor-
age that can provide pressures of at least 20 psi (140 kPa), and therefore favor ele-
vated storage tanks. Although standpipes (see page 88) will cost less than elevated
storage tanks with the same total storage volume, not all of the storage in a standpipe
is useful. The existence of dead storage at the bottom of the tank can lead to water
quality problems.
Figure 8.24
Tank profile 900 ft
Overflow
Effective Storage 869 ft
Emergency Storage 846 ft
Dead Storage 43
//=//= //=//=
30 psi
/=
20 psi
=/
//=
//
//= psi
/=
=/ //=
//=
//
800 ft
//=//= //=//=
Tank Water Level Fluctuations. A calibrated hydraulic model can be used to

check water level fluctuations in storage tanks. The pumps should have a design flow
capacity such that peak day demand can be met even when the largest pump is out of
service, and a head such that the band of system head curves for the pressure zone
passes close to or through the best efficiency point of the pump.
The combination of pump, piping, and distribution tank is best evaluated using an
EPS model. The EPS should be run based on projections for peak-day, average-day,
and minimum-day demands. If pumps fill the tanks, it is usually easy to cycle the
pumps so that the tank operates in a reasonable range. It is important to run the model
for at least 48 hours to determine if the tank can refill after a peak day. If the tank can-
not recover, then the weak link in the system (either pumping or piping) needs to be
upgraded.
If a distribution storage tank is filled from a plant clearwell, either by gravity or
through a PRV, then it becomes more difficult to get the tank to fluctuate as desired. If
the system is small, or the tank is close to the water source, the tank may not turn over
sufficiently. The model can be used to simulate corrections for these conditions by
simulating the closing or throttling of a valve for a few hours a day, or by switching to
an alternate pilot valve to control the PRV for a few hours.
As the system and the head loss across the system become larger, the HGL tends to
slope much more steeply during peak-use periods. In terms of tanks, this lower HGL
means that a tank with an overflow elevation selected to work well on a peak summer
day may have too low an elevation to work effectively during an average or minimal-
use winter day. The results of an EPS run for such a situation would look like Figure
8.25, which shows a tank that would operate in a different band during low-, average-
, and maximum-demand days. There are several possible solutions to this problem:
• Operate the tank as pumped storage with the tank overflow below the HGL
(this tends to be the most expensive solution as it wastes energy and needs a
generator for reliability).
• Construct a tall tank that operates in the upper range during the winter and the
lower range during the summer (with large seasonal pressure fluctuations).
• Significantly increase piping capacity across the system so that the HGL
does not drop as much during the summer (costly from a capital cost stand-
point).
• Construct the tank at an elevation that works well during a maximum use
day. Use a control valve on the major system’s main to control the filling rate
on other days (to use this type of control effectively, a SCADA system is
needed).
Figure 8.25
1,005 Tank water level
fluctuations
1,000 Low Demand
995
Avg. Demand
990
Water Elevation, ft
985
Max. Demand
980
975
970
965
960
955
0 12 24 36 48 60 72
Time, hr
An EPS model provides the designer with a tool to compare these approaches and
determine how each could be used (for example, by finding the correct pipe size or
effective control settings for a valve). Then, the benefits and costs of each solution can
be compared.
Tank Behavior During Emergencies. Another design question relates to
how well the tank can recover after a fire or power outage that draws down the tank
water level down. A tank cannot be expected to recover instantly, but it should not
take more than a few days to bring water levels back to their normal cycle. While this
is partly an operational problem, the designer needs to provide sufficient capacity for
emergencies.
Multiple Tanks in a Pressure Zone. When there is more than one tank in a
pressure zone, the problem of designing and operating the zone becomes much more
complicated. For modeling purposes, it is difficult to get all of the tanks to fluctuate in
the desired range. In general, all the tanks in a pressure zone should be constructed
with the same overflow elevation. In that way, the full range of each can be used. With
multiple tanks, it is also helpful to construct each tank with an altitude valve. This is
especially true for the tank that is hydraulically closer to the source, because the HGL
will be higher in this area.
The usual problem is that the tank near the source fills up quickly and drains slowly,
while the tank at the perimeter of the system fills more slowly and drains quickly. One
solution is to fill the tank that is closer to the source using a throttle control valve to
throttle the flow when the tank is nearly full, enabling more water to flow to the dis-
tant tank. The nearer tank should drain through a separate line with a check valve so
that the tank can drain easily even if the power should fail while the control valve is in
the throttled or closed position (Figure 8.26). This situation can be modeled with a
check valve and a throttling control valve.
Figure 8.26
Multiple tanks in
pressure zone HGL - Pump On
Near Distant
Tank Tank
HGL - Pump Off
Throttling
Control
Valve Check Valve
Source Pump
EPS models can be used to determine if there will be problems in pressure zones with
multiple tanks and to test alternative strategies for operating these zones. It is espe-
cially important to test the hydraulics under a wide range of demands. These demands
should include seasonal variations and future projections, since a shift in the size or
location of the population (for example, more demand in suburbs near a new tank)
will change how the system will operate.
Regionalization. When water systems are combined, whether due to regional-
ization, annexation, or acquisition, the adjacent systems usually do not have the same
HGL elevation; that is, they are in different pressure zones and cannot simply be con-
nected. Therefore, integrating the distribution systems becomes problematic. The eas-
iest way to integrate the systems is to install a pump or PRV at the boundary. Usually,
one of the systems will no longer use its original source, or will use it only as a
backup. Instead, it will use the other system as its source. Large pipes are usually
needed at the connection point linking a system and its new water source. At points
where two systems meet at their perimeters, the pipes are typically small. Substantial
improvements consisting of pipe paralleling and/or replacement are usually required
in one or both systems.
If new piping is to be installed, the designer has a unique opportunity to establish

pressure zones as they should be, rather than as they have evolved out of necessity.
Usually, the sizes of the pipes near the interconnection points are the limiting factors,
and paralleling or replacing those pipes becomes the focus of the modeling analysis.
Tank Volume Considerations. The discussion thus far has emphasized the
importance of water level in a tank. The tank cross-section (and therefore volume) is
also important. To a great extent, tank volume sizing can be done outside of the model
by considering the amount of water that is needed for equalization storage, fire stor-
age, and emergency storage. Too much storage, however, may contribute to water
quality problems. Tank sizing requirements are described in more detail in the Ten
State Standards (GLUMB, 1992) and by Walski (2000).
The volume of a tank is usually dictated by the volume needed for equalization and
the larger of fire and emergency storage. These volumes can be viewed as areas under
the curve as shown in Figure 8.27. The area between production and peak-day
demand is the volume needed for equalization, and the volume between the peak-day
demand and the peak day plus fire curves is the volume needed for fire protection. In
some systems, emergency storage volumes exceed volumes needed for fire fighting
and may predominate.
Figure 8.27
Determining tank
3,000 volumes
Peak Day + Fire
2,500
Fire Storage
2,000 Equalization
Storage
Flow, gpm
1,500
Production
1,000
500
Peak Day
Demand
0
0 6 12 18 24
Time, hrs
The water level in a tank routinely fluctuates through a fill and drain cycle. Ideally, the
tank water level should fluctuate by at least several feet during its cycle, whether that
cycle is a full day or the time until the pump starts again. The EPS capability of mod-
eling software is a valuable tool for predicting performance when comparing alterna-
tive tank and pipe sizes for various designs.
If the level does not drop much, the tank may be too large, or the pump may be set to
cycle too frequently. If the tank water level drops very quickly during peak demands,
then the tank may be too small. Increasing the volume of the tank to the next larger
standard size may correct the problem. For the situation in which the tank cannot
recover after maximum day or emergency conditions, the distribution system serving
the tank may have insufficient capacity to satisfy demands. Several runs may be nec-
essary to determine the problem (for example, an inadequate pump or a small pipe)
and correct it.
The tank should not fully drain during emergency demand conditions [for example, a
2-hour, 1,500-gpm (0.095 m3/s) fire]. If the tank drains completely during this time,
then either the tank is too small or another source of water (such as a pump station or
treatment plant) may not have performed as expected. Fire flow requirements can be
found in AWWA M-31 (1998) and other sources from the fire insurance industry, as
described on page 166.
If water quality problems due to chlorine decay are expected, then a water quality
analysis should be conducted to determine whether or not the chlorine decay is due to
the piping or the tank. When significant disinfectant decay is found to occur due to
residence times within storage tanks, the volume may need to be reduced, or operat-
ing procedures may need to be modified (Grayman and Kirmeyer, 2000).
8.6 DEVELOPING SYSTEM HEAD CURVES FOR PUMP

SELECTION/EVALUATION
The system head curve is a graph of head versus flow that shows the head required to
move a given flow rate through the pump and into the distribution system. Prior to
purchasing a pump, the system head curve that the pump will need to pump against
must be determined. In a simple situation with a single pipe connecting two tanks, a
system head curve can be generated without a model. When selecting a pump that will
be used in a complicated water distribution system, especially one with looping and
branching between the tanks on the suction and discharge sides of the pump, manual
calculations only give a rough approximation to the system head curve. In such cases,
a model is needed to derive a more exact solution.
The system head curve depends on tank water levels, the operation of other pump sta-
tions in the distribution system, the physical characteristics of the piping system, and
the demands. Therefore, the system head curve uniquely reflects the system condi-
tions at the time of the run. As a consequence, for any pump station, there is actually a
band of multiple system head curves similar to those shown in Figure 8.28. The high-
est curves correspond to low suction-side tank levels, high discharge-side tank levels,
low demands, other pumps/wells running, and possibly even throttled or closed
valves. The lowest system head curves correspond to high suction-side tank levels,
Section 8.6 Developing System Head Curves for Pump Selection/Evaluation 343
low discharge-side tank levels, no other sources operating, high demands (especially
fire demands near the pump discharge), and all valves being open. For more informa-
tion on system head curves, see Walski and Ormsbee (1989).
Figure 8.28
System head curves
300
250
200
Low Use
Head, ft
150
Avg. Use
100
High Use
50
0
0 2 4 6 8 10 12 14
Flow, gpm
A single run of a water distribution model with a pump identifies a single point on a
system head curve. Generating a system head curve for the full range of potential
flows requires multiple steady-state runs of the model, with each steady-state run rep-
resenting one point on the curve. The easiest way to arrive at the system head curve is
to remove the proposed pump from the model, leaving the suction and discharge
nodes in place, as shown in Figure 8.29. For the curve to be computed properly, a tank
or reservoir must be present on each side of the pump.
Figure 8.29
Tank Profile of system for
Filling system head curve
HGL Tank
Stable
HGL
Pump Discharge
Head Tank
HGL Tank Draining HGL/
Suction Pump Off
Tank
Discharge
Sytem
Suction
Suction Discharge
System
Node Node
Generating a System Head Curve

Some models can automatically calculate system 5. Run the model and determine the HGL eleva-
head curves. The approach for manually generat- tions at the suction and discharge nodes. For
ing system head curves is provided in the follow- instance, the suction node HGL is 715 ft (218
ing steps: m), and the discharge node HGL is 890 ft
1. Calibrate the model and identify suction and (271 m).
discharge nodes, but do not specify the pump 6. Subtract the suction HGL from the discharge
between them yet. HGL to obtain the coordinates for a point on
2. Set the demands, tank water levels, and other the system head curve [in this case, 100 gpm
operational conditions [for example, suction (0.006 m3/s), and 175 ft (53 m)].
tank at 720 ft (220 m), discharge tank at 880 ft 7. Repeat steps 4 through 6 until all the flow
(270 m), average demands, well number 2 points have been generated.
turned off]. 8. Plot these points and connect them to obtain a
3. Identify the range of flows that the pump may system head curve.
produce. For example, if selecting a 300-gpm 9. If additional system head curves are desired,
(0.019 m3/s) pump, use 0, 100, 200, 300, 400, return to step 2 to set up new boundary condi-
and 600 gpm (0, 0.006, 0.013, 0.019, 0.025, tions and demands and repeat steps 3 through
0.038 m3/s). 8 until all desired curves are obtained.
4. Select the first flow and insert it as a demand
on the suction node and an inflow on the dis-
charge node.
The water that leaves the suction-side pressure zone is identified as a demand on the
suction node, while the water that enters the discharge-side pressure zone is identified
as an inflow (or negative demand) on the discharge node. The difference in head
between the suction and discharge nodes as determined by the model is the head that
must be added to move that flow rate through the pump (that is, between the two pres-
sure zones). The flow rate at both the suction and discharge nodes is then changed and
the model re-run to generate additional points on the curve, continuing until a full
curve has been developed.
The curves should cover a reasonable range of conditions that the pump will experi-
ence. Once the system head curves are available, pump manufacturers can be con-
tacted to determine which pumps (comparing model, casing, impeller size, and speed)
will deliver the needed head at the desired flow rate with a high efficiency and suffi-
cient net positive suction head (NPSH).
Overlaying the pump head curve that was obtained from the manufacturer with the
system head curves will identify the pump operating points. The operating points can
also be determined by inserting the proposed pump into the model and performing a
series of runs for different conditions. The designer should check efficiency and
NPSH for the range of operating points the pump is likely to encounter.
Usually, several pumps from different manufacturers will function properly. The deci-
sion about which one to buy will be based on a variety of factors, including the pump
station floor plan, type of pump, operation and maintenance personnel preferences,
Section 8.7 Serving Lower Pressure Zones 345
cost, familiarity with a particular brand, and projected life-cycle energy cost. Chapter
10 explains how to calculate the energy cost. After a pump is selected, the designer
should use EPS runs to determine how it will operate in the system over a variety of
demands and emergency conditions.
8.7 SERVING LOWER PRESSURE ZONES

As a system expands into lower-lying areas, the customer elevations may not be
within the serviceable range of the pressure zone containing the water source. Creat-
ing a lower pressure zone will prevent the delivery of excessive pressures to custom-
ers at low elevations. There is no consensus on the exact limit at which it becomes
necessary to reduce pressure. However, for the majority of systems, the upper limit is
set around 100 psi (690 kPa). Some systems, especially in hilly areas, may distribute
water at up to 200 psi (1,380 kPa) and rely on PRVs in the service lines of individual
customers to reduce pressures.
PRV Feeding into a Dead-End Pressure Zone

Installing a PRV to feed a dead-end zone is usually the easiest solution for controlling
pressures in a low-elevation pressure zone. The key to this approach is to find a pres-
sure (HGL) setting that will keep pressures within a reasonable range. The PRV is
often installed initially to serve a small extension to the system. The PRV setting (or
downstream pressure to be maintained) should be established, however, based on cur-
rent projections of population growth and business development expected in the low
area. The drop in pressure across the PRV should be approximately 40 to 60 psi, or 90
to 110 feet (275 to 413 kPa, or 27 to 34 m). A smaller cut in pressure will cause limi-
tations on the size of the pressure zone to be served. Too large a cut may cause unac-
ceptably high or low pressures for a band of customers along the divide.
The PRV should be located as close as possible to the contour line defining the bound-
ary of the pressure zone. When the PRV is located far from the boundary, parallel
pipes are often required to serve the two pressure zones. In this case, one parallel pipe
would be located in the higher pressure zone, and the other parallel pipe would be in
the neighboring, lower pressure zone.
Figure 8.30 shows the boundary line between a 1,810-ft (552 m) HGL zone and a
1,690-ft (515 m) HGL zone. The boundary is located at a ground elevation of 1,600 ft
(488 m). If the PRV is placed exactly at 1,600 ft (488 m), then the upstream pressure
will be 91 psi (627 kPa) and the downstream pressure will be 39 psi (269 kPa), both of
which are reasonable pressures.
If the PRV is placed at 1,640 ft (500 m), the upstream pressure will be 74 psi (510
kPa), while the downstream pressure will be 22 psi (152 kPa). The lower pressure is
only marginally acceptable, and with normal head losses, may prove unacceptable.
If the PRV is placed at 1,560 ft (475 m), the pressure will be cut from 108 psi (745
kPa) on the upstream side to 56 psi (386 kPa) on the downstream side. In this case, the
pressure upstream of the valve is high enough that two pipes may be required in the
street so that customers in this area can continue to be served by the lower pressure
zone and not receive unacceptably high pressures.
Figure 8.30
Locating PRVs
When selecting and modeling the PRV itself, it is important to note that a large PRV
may not be able to precisely throttle small flows. Better control at low flows can be
obtained by using a smaller PRV [for example, a 4-in. (100-mm) PRV for an 8-in.
(200-mm) main]. When a PRV has to pass much higher flows (to meet fire capacity
requirements, for instance), the specifications should be checked to ensure that the
smaller PRV will not significantly restrict flow. This restriction can also be examined
by modeling the PRV’s minor losses at high flow; however, some models do not
account for a valve’s minor losses when it is in the control (throttling) mode. When
using such a model, the minor loss corresponding to the open PRV may be assigned to
a connecting pipe. If the minor losses through the small PRV are too large at high
flows, specifying a small PRV for normal flows and a larger PRV in parallel for
higher flows can solve the problem.
In some cases, a PRV is used along a pipeline that carries water from a high source to
a low area with few customers. The model may show that the PRV can successfully
produce a very high pressure cut [say, greater than 100 psi (690 kPa)], but it is impor-
tant to check the specifications for the PRV to ensure that it can pass the flow rate
with the required pressure cut without cavitating or eroding.
Lower Zone with a Tank

If there is a tank with its water level floating on the lower zone, setting the PRV
becomes more difficult. If the PRV is set too high, the tank may overflow or be shut
off by the altitude valve so that it no longer drains. If it is set too low, the tank may not
fill adequately.
Usually, if the tank is far from the PRV, the head loss across the zone and the diurnal
fluctuation in demand may be sufficient to make the tank cycle over the desired range.
It may be necessary, however, to alter the PRV settings seasonally to achieve this
range. The designer can find the right PRV settings and determine seasonal changes
for those settings by using information obtained from EPS runs.
If the tank is close to the PRV, it will be virtually impossible to get the tank water level
to fluctuate adequately using a single PRV setting. To get an adequate water level
Section 8.7 Serving Lower Pressure Zones 347
fluctuation, the designer could use a control valve, as described in the next section.
Another option is to use a PRV with dual pilot controls, so that when the tank is in a
“fill” mode, the higher setting prevails, and when the tank is in a “drain” mode, a
lower setting prevails. The switching can be done using a timer so that the tank is in
the fill mode when demands are low (typically at night). With this approach, there is
no need for sophisticated programmable logic controllers; a simple timer will suffice.
The model can check if the desired turnover of the tank can be achieved and will cal-
culate the amount of time it takes for the tank to refill. If the tank fills too quickly, the
PRV can cause problems in the upstream pressure zone. If the tank fills too slowly,
tank recovery may take longer than the designated time. The designer needs to check
pressure fluctuations in the higher pressure zone and velocities in the transmission
mains to ensure that they are acceptable.
Lower Zone Fed with Control Valves

If the utility desires controls that are more sophisticated than a PRV, a control valve
can be used instead to regulate the filling and draining of the tank. A control valve can
be programmed to operate based on fairly sophisticated logic by using information
from the tank and other points in the system. The disadvantages of using a control
valve are its reliance on expensive telemetry or SCADA equipment to provide infor-
mation about tank water levels and its need for a distributed logic controller (or
remote control by an operator) to operate the valve. Also, a backup power supply is
required so that it can function properly during a power outage, whereas a PRV
requires no power. The control valve also requires programming or alarming to handle
any loss of signal or sensor and is susceptible to lightning interference.
The control valve can have either a simple on/off control or an analog control. For
example, the on/off control would open the valve when the tank is filling and close it
when the tank is draining. In an analog control, the flow through the valve or the valve
position is determined by tank level or some other analog input. The valve can be pro-
grammed to open wider as the tank water level falls. Conversely, the valve can hold a
setting until the tank fills and then switch to a more closed setting to enable the tank to
drain. Using an on/off control is somewhat risky in that power could fail while the
valve is shut. A throttled valve can at least pass some water during a power outage.
The controls can be tested by running simulations; however, valves are operated
somewhat differently in models than they are in reality. With a real valve, the opening
size is controlled (for example, it can be set at 40 percent open). In most models,
though, the value that is controlled is the maximum flow, as with a flow control valve,
or the minor loss coefficient, as with a throttling control valve.
Use of a flow control valve in the model to set a flow rate is the simplest approach to
making sure that the piping and tank are sized correctly. This approach will not verify
the sizing of the control valve or be very helpful in selecting the valve opening that
corresponds to a given set of conditions. To find this information, the valve should be
modeled as a TCV with the minor loss coefficient being the variable controlled. The
relationship between the minor loss coefficient and percent valve opening must be
developed outside of the model using data supplied by the valve manufacturer.
Butterfly valves are usually used in this application because of their low cost and
good throttling characteristics, provided the head loss in not too high. Ball valves are
even better at control but are more expensive.
Conditions Upstream of the PRV or Control Valve

In some situations, the head loss in the upper zone above the boundary valve (whether
it is a PRV or control valve) can become excessive [for example, if the demand of the
lower zone is 500 gpm (0.032 m3/s) and the water is delivered through a 6-in. (150-
mm) pipe]. This head loss can cause pressures in the upper zone to drop to unaccept-
able levels. The long-term solution to this problem is to increase the carrying capacity
of the pipes in the upper zone. This type of project is usually expensive, however, and
an immediate solution may be necessary even if a large budget is not available.
The short-term solution may be the use of a combined PRV/PSV in the regulating
vault. EPS runs can indicate what the PSV setting should be to adequately maintain
pressures at the most critical (usually the highest) points in the upper zone. The com-
bined PRV/PSV can be simulated in the model by placing a PSV immediately above
the PRV. The designer should also look for any bottlenecks in the upper zone that can
be inexpensively corrected to help feed the PRV.
8.8 REHABILITATION OF EXISTING SYSTEMS

Models are often used to assess the rehabilitation of older water distribution systems.
The rehabilitation work may be necessary because of
• The cumulative effect of tuberculation and scaling
• Increased demands due to new customers
• Excessive leakage
• Infrastructure improvements, such as street reconstruction or sewer replace-
ment, in vicinity of distribution system piping
• Water quality problems
The problems associated with rehabilitation are somewhat more difficult than design-
ing a new system. Problems include
• Working with existing piping
• Numerous conflicts with other buried utilities
• The added importance of the condition of the paving
• The larger range of alternatives to be considered
The one way in which rehabilitation analysis is simpler than other design applications
is that pressure zones and their boundaries are already defined and are usually not
being adjusted.
Section 8.8 Rehabilitation of Existing Systems 349
Instead of simply deciding on the size of pipe, the utility is faced with additional
choices when performing rehabilitation. The utility can either replace existing pipes
or keep them in place and parallel them for added capacity. In addition to new piping,
the designer has a range of other options to choose from, including pigging (forcing a
foam “pig” through the pipes using water pressure), cleaning with cement mortar or
epoxy lining, installing inversion liners, sliplining, and pipe bursting. Each of these
options will need to be modeled in a slightly different way, as described in the sec-
tions that follow.
Data Collection
Before modeling improvements, the designer needs to thoroughly analyze the existing
system to determine its strengths and weaknesses. Because the system already exists,
data are readily available. As was the case with model calibration, fire hydrant flow
tests (see page 184) provide a great deal of information on the hydraulic capacity of
the system. Other valuable tests include pipe roughness coefficient tests (see page
191) and the use of pressure recorders to obtain readings at key locations. The pipe
roughness coefficient tests provide a view of the carrying capacity of individual pipes,
and the chart recorders show how the system handles present day demand fluctuations.
Any severe drops in pressure that occur during peak hours reveal capacity problems.
More information on testing and calibration is available in Chapters 5 and 7.
Both the hydraulic capacity of the existing pipe and the structural integrity of the pip-
ing are important. If adequate metering is available, a water audit comparing water
delivered to an area with metered water consumption can give an indication of the
unaccounted-for water. A leak detection survey of the study area using sonic leak
detection equipment can also be conducted. A review of past work orders for pipe
repairs and interviews with maintenance personnel can indicate if there are structural
problems with the pipe. If a section of pipe must be excavated in that area for any rea-
son, the designer should examine the inside of the pipe for tuberculation and scale,
and the outside of the pipe for signs of corrosion damage. Graphitic corrosion of cast
iron pipe may require scraping or even abrasive blast cleaning to reveal pits.
Records on service lines should also be assessed to determine if any service lines need
to be replaced in conjunction with the mains, or if the old service lines can just be
reconnected to the new mains. In some instances, it may be worthwhile to keep old
mains in service simply to avoid the cost of renewing a large number of service lines.
Other utilities have a policy of retiring old, parallel mains and connecting old service
lines to the new main when one has been installed.
Modeling Existing Conditions

With the results of fire flow and roughness coefficient tests, a detailed model of the
study area can be calibrated. The calibration effort often reveals problems that are eas-
ily or inexpensively corrected, such as closed or partly closed valves. Clearly, opening
a closed valve is very inexpensive compared with rehabilitation techniques or new
piping.
The model of the existing system will also reveal which pipe segments are bottle-
necks. These will usually be the segments with the highest velocities or highest
hydraulic gradients. Field data should then be collected to corroborate the simulation
results. Those segments that are bottlenecks will need to be replaced, paralleled, or
rehabilitated. In general, peak hour demands and fire flow demands are the control-
ling conditions, and steady-state runs may be used to solve this type of problem.
Overview of Alternatives
Usually the decision to replace piping is the most expensive alternative and should not
be selected unless the existing piping is in poor structural condition. The decision of
whether to parallel or rehabilitate the existing piping depends upon the design flow in
the area. Rehabilitating the existing pipes will restore more of their original carrying
capacity but will not greatly increase the nominal diameter of the pipe. Pipe bursting,
a technique in which the old pipe is broken in place, allows a slightly larger pipe to be
pulled through the opening where the old pipe once lay. If the future flows are going
to be significantly greater than the original flows in the pipes, then rehabilitation will
not provide sufficient capacity, and new pipes roughly paralleling the old system are
needed. A schematic of the evaluation process recommended for replacement deci-
sions is illustrated in Figure 8.31 and is discussed in the following sections.
Figure 8.31
Overview of pipe
rehabilitation options
Replacement. Most utilities will not have sufficient resources to replace large
portions of the distribution system. With this limitation in mind, the designer must be
extremely selective in identifying pipes for replacement. The model can answer ques-
tions as to which pipes are inadequate from a hydraulic standpoint. Information from
the simulation needs to be combined with other information, such as which pipes have
experienced breaks, leakage, and water quality problems in order to make informed
decisions. In this way, the worst pipes in the system are identified for replacement. By
examining fire flows at different points in the study area, the model will indicate those
pipes with the most severe hydraulic limitations. In most cases, these will be old,
unlined, 4-in. (100-mm) and 6-in. (150-mm) pipes. These pipes also tend to have the
highest break rates because they have the lowest beam strength. The designer should
selectively replace these pipes in the model and re-run it. Subsequent runs will then
indicate the next-worst bottlenecks. The designer should also be mindful that in some
cases the worst hydraulic limitations may be outside of the study area.
Paralleling. If the existing piping is found to be in sound structural condition,
pipes will not need to be replaced. The emphasis should then be placed on determin-
ing which lengths of pipe have the poorest hydraulic carrying capacities when com-
pared with the required capacities. Fire hydrant flow tests can indicate which portions
of the study area have problems, but the exact pipe(s) causing the problems is best
determined through model runs.
Rather than paralleling or replacing pipes sporadically throughout the study area, the
best solution will typically consist of installing a loop or a “backbone” of larger pipe
[such as 16-in. (400 mm) pipe] through the heart of the study area. This new main will
reduce the distance that water must travel from a large pipe to a fire hydrant. If the
new parallel pipe is significantly larger than the existing pipe, hydrants should be
transferred from the existing pipe to the new pipe to take advantage of the greater flow
capacity. The designer also needs to remember that the available fire flow at a node in
the model is not the same thing as the available fire flow from a hydrant near that
node, because of the distance between the node and the hydrant and the associated
losses (see page 191).
Pipe Cleaning and Lining (Nonstructural Rehabilitation). Pipe clean-
ing may be an economical and effective alternative to installing additional pipe to
restore lost carrying capacity if
• The pipe is structurally sound
• Future demands are not expected to be significantly greater than the
demands for which the system was designed
• The loss in carrying capacity due to pipe tuberculation, scaling, or other
deposits is significant
For smaller pipe sizes [4-, 6-, and 8-in. (100-, 150-, and 200-mm) pipe], installing
new pipe is usually only slightly more expensive than pipe rehabilitation. However, as
pipe diameters increase, the economics of pipe rehabilitation by cleaning become
very attractive. Pipe cleaning by scraping or pigging is most economical in situations
in which the installation of new pipe is unusually expensive due to interference with
other buried utilities or because of expensive pavement restoration costs.
The effects of pipe cleaning can be simulated by making the roughness factor of the
cleaned pipe more favorable. Usually, the Hazen-Williams C-factor can be increased
to values on the order of 100 to 120, with the higher values usually achieved in larger
pipes. A C-factor increase from 90 to 110 will not justify the costs of pipe cleaning,
but an increase from 40 to 110 is likely to correct a hydraulic deficiency at a reason-
able cost.
The decision whether to cement-line a main after the pipe is cleaned is usually based
on water quality considerations. If the pipe is not expected to experience corrosion,
scaling, or deposition problems in the future, the pipe may be left without a liner. In
most cases, however, it is better to line the pipe to maintain the benefits of the clean-
ing. Cement mortar or epoxy, which does not decrease the inner diameter signifi-
cantly, is typically the preferred lining material for distribution mains, but sliplining
can also be used.
In modeling any kind of rehabilitation involving a liner, it is important to use the
actual inner diameter of the liner pipe in any model runs.
Sliplining (Structural Rehabilitation). Several methods of rehabilitation
that also increase the structural strength of the pipe are available. These include fold-
and-form piping, swagelining, and sliplining. Fold-and-form-pipes are folded for easy
insertion within the existing pipe, and then expanded once in place. Swagelining
involves pulling the liner pipe through a die, temporarily reducing its size so that it
can be easily inserted in the existing pipe. Sliplining is performed by pulling a slightly
smaller pipe through the cleaned water main (Figure 8.32). Inversion lining (a type of
sliplining) utilizes sock-like liners that must be cured in place. This procedure is usu-
ally practical only in low-pressure applications, because the thickness of the required
inversion liner becomes excessive as pressures increase.
The liner used in sliplining is usually plastic and quite smooth (C-factor of 130). The
diameter of the liner pipe, however, is somewhat different from the original pipe; thus,
the actual inner diameter of the liner pipe must be used in the model. Structural reha-
bilitation is less attractive than cement-mortar lining for situations in which there are
numerous services that must be reconnected to the pipe. Structural rehabilitation can
be modeled by decreasing the effective roughness of the pipe and decreasing the
diameter of the pipe being lined to coincide with the correct values for the liner.
Figure 8.32
Sliplining procedure
Pipe Bursting. Pipe bursting is the only rehabilitation technique that actually can
increase the inner diameter of a pipe. In this technique, a mole is passed through a
pipe made from a brittle material (cast iron or asbestos cement), and the pipe is burst.
The fragments are forced into the surrounding soil and a liner pipe of the same size as
the original pipe (or slightly larger) is then pulled through the resulting hole. Excava-
tion work is necessary if there are service lines that must be reconnected to the new
pipe.
Evaluation
In distribution system rehabilitation studies, the number of alternative solutions tends
to be much larger than in most other distribution design problems. Exploring all the
possibilities is the only way to ensure that the most cost-effective solution, based on
whole-life-cycle cost, is chosen. More than one alternative can solve a design prob-
lem, and each alternative has its own costs and benefits.
8.9 TRADEOFFS BETWEEN ENERGY AND CAPITAL

COSTS
Near pumping stations, it may be worthwhile in some cases to use larger pipes in
order to reduce head losses and, consequently, energy costs. The distribution system
costs affected by pipe sizing include capital costs for piping and the present worth of
energy costs. These costs can be determined using Equation 8.10.
– 4.87
TC = ¦ f D x + PW ³ k 1 Qp h 1 + k 2 D ee (8.10)
allpipes
where TC = total life-cycle costs ($)

f(D, x) = capital cost function
x = set of pipe-laying conditions
PW = present worth factor for energy costs
k1 = unit conversion factor for energy
Q = actual flow over time (gpm, l/s)
p = price of energy ($/kW-hr)
h1 = lift energy (ft, m)
k2 = coefficient describing characteristics of system
e = wire-to-water efficiency (%)
The preceding equation cannot be solved analytically for a complex network. How-
ever, the problem can be simplified, because most of the pipes in a typical distribution
system have only a negligible effect on energy costs. It is usually only the pipes from
the pump stations to the nearest tanks that can have a significant impact on energy
costs. For a series of pipes between two tanks, Walski (1984) provides an analytical
solution for optimal pipe sizes. For the simplest case of a single pipe between two
tanks, the solution can be viewed as shown in Figure 8.33.
Even though the present worth of total energy cost is fairly large when compared to
construction costs, most of this energy is being used to overcome the difference in
head between pressure zones (that is, it is used for static lift). Because only the energy
cost used in overcoming friction losses is a function of pipe diameter, only this por-
tion of the energy cost is involved in the tradeoff with capital costs. Furthermore, the
cost of the pump station itself, including the building, land, piping, valves, SCADA,
and engineering are independent of head loss. The initial construction cost of pump
stations is not very sensitive to head loss and need not be considered in this cost
analysis.
Section 8.10 Use of Models in the Design and Operation of Tanks 355
Figure 8.33
Example of
450
relationship between
capital and energy
400 costs in a pumped
pipeline
350
300
Present Worth Cost, $/ft
250
Total
200
150
Capital
100
Energy
50
0
6 8 10 12 14 16 18 20
Diameter, in.
The optimal velocity to be used in pipe sizing depends on the relative costs of energy
and construction, the interest rate, the efficiency of the pumps, and the ratio of peak
flow in the pipes to average flow. For medium-size pumps [approximately 1,000 gpm
(60 l/s)], Walski (1983) showed that the optimal velocity would be on the order of
6 ft/s (2 m/s) at peak flow when the ratio of peak flow to average flow is 2. For pipes
with a ratio of peak to average flow of 1.25, the optimal velocity at peak flow would
be on the order of 4.5 ft/s (1.5 m/s).
Several investigators (Murphy, Dandy, and Simpson, 1994; and Walters, Halhal,
Savic, and Ouazar, 1999) have applied genetic algorithms to determine optimal pipe
sizes in pumped systems.
8.10 USE OF MODELS IN THE DESIGN AND OPERATION

OF TANKS
Storage facilities are an essential component of water distribution systems. Tradition-
ally, they have been designed and operated to meet the hydraulic requirements of the
distribution system, including providing emergency storage, equalizing pressure, and
balancing water use throughout the day. However, the implications of the design
and operation of the facility on water quality must also be considered to avoid water
quality deterioration in the facility and in the distribution system served by the tank
or reservoir.
Mixing and aging are two related phenomena that affect the water quality changes
that can occur within finished water storage tanks and reservoirs. In these facilities,
water quality deterioration is frequently associated with the age of the water. Long
residence times depress disinfectant residuals and can promote bacterial regrowth.
Uneven mixing can result in zones of older water. As a result, an implicit objective in
both the design and operation of distribution system storage facilities is the minimiza-
tion of detention time and the avoidance of parcels of water that remain in the storage
facility for long periods of time. The allowable detention time depends on the quality
of the water, its reactivity, the type of disinfectant that is used, and the travel time
before and after the water’s entry into the storage facility.
Mathematical models can serve a useful role in the design and operation of tanks and
reservoirs (Grayman et al., 2000). They can be used to answer “what if” questions
such as how water will mix in the tank, whether stratification will occur, and the
effects of a fill and drain pattern on water age and chlorine residual. A variety of types
of mathematical models can be applied to provide insights into the mixing and aging
behavior of tanks and reservoirs. Two primary categories, systems models and com-
putational fluid dynamics (CFD) models, are discussed in the following sections.
Physical scale models can also be used to study the mixing phenomena in tanks and
reservoirs.
Systems Models
Systems models (also called black-box models or input-output models) are a class of
models in which physical processes (that is, the mixing phenomena in the tank or res-
ervoir) are represented by highly conceptual, empirical relationships. Systems models
have been used to represent tanks and reservoirs that operate in a fill-and-draw mode
or with continuous (simultaneous) inflow and outflow. These models include com-
plete-mix models, plug-flow models, last-in/first-out (LIFO) models, and multicom-
partment models. Both conservative substances and nonconservative substances can
be simulated.
Systems models actually do not explicitly simulate the movement of water within the
tank, but rather determine the water quality (or water age) of the outflow of the tank
based on the inflow water quality and an assumed macrobehavior within the tank. It is
the user’s responsibility to choose a behavior pattern based on field studies, more
detailed modeling, or past experience. For example, complete and instantaneous mix-
ing in tanks is a standard modeling assumption, but the mixing actually occurring is
likely to be more complex.
Systems models of tanks and reservoirs are available as part of all water distribution
system models and as stand-alone models. The tank modules that are part of water
distribution system models are useful for examining the behavior of the tank and its
impacts on water quality and water age within the distribution system for extended peri-
ods of time (typically a few days). For example, the plots in Figure 8.34 generated by a
water distribution system model illustrate the variation in water age and chlorine resid-
ual leaving a tank over a period of a few days. The effects of the tank are then trans-
ferred to the distribution system so that the impacts of the tank can be determined.
Figure 8.34
Water age versus time,
and chlorine residual
versus time
However, for design and operational purposes, it is useful to study the effects of the
tank on water quality over a longer period of time (that is, several months representing
different seasonal conditions). Although this type of study could be done with a water
distribution system model, long, extended-period simulations of this type are difficult
to perform. An alternative is the use of a stand-alone tank model that only simulates
the tank behavior based on an assumed temporal inflow-outflow record. For existing
tanks, an inflow-outflow record can usually be constructed directly from water level
or flow records that are routinely collected by a SCADA system.
CompTank, a stand-alone tank model available from the AWWA Research Founda-
tion (Grayman et al., 2000) includes several different systems models of tanks and
reservoirs. An example of the use of CompTank to evaluate different tank designs is
shown in Figure 8.35. In this example, the effects of tank volume on water age were
studied using an assumed inflow-outflow pattern for a critical month. As illustrated, if
a 2-million-gallon tank was constructed, the water age varies between approximately
15 and 22 days, whereas a 1-million-gallon tank results in water ages between 7 and
12 days. Because even the lower range of water age is generally considered to be too
old, the model could be used to further explore the impacts of alternative operating
patterns on water age.
Figure 8.35
Use of tank model to
study the impacts of
tank size on water age
Computational Fluid Dynamics Models

Unlike systems models, which assume a particular ideal mixing regime, Computa-
tional Fluid Dynamics (CFD) models are based on modeling the physics of fluid
motion. Coupled nonlinear partial differential equations representing conservation of
mass, conservation of momentum, and conservation of energy are solved numerically
to simulate the movement of water within a storage facility. Although CFD modeling
has been used widely in chemical, nuclear, and mechanical engineering, its use in the
drinking water industry is a relatively recent development. CFD models can be used
to simulate the effects of temperature variations, unsteady hydraulic and water quality
conditions, and decay of constituents in storage facilities.
The primary factor that influences the mixing within a tank is the jet behavior as water
enters the tank during a fill period. As water flows through the inlet into the tank, a jet
is formed that expands as it moves through the tank, entraining the surrounding water.
When the jet reaches a surface within the tank (the water surface or an opposite wall),
the direction of the jet changes and flow patterns develop. This behavior is illustrated
in Figure 8.36 in a series of diagrams adapted from Okita and Oyama (1963). CFD
models employ numerical solution techniques to solve the mathematical equations
that represent this mixing process.
Figure 8.36
Views of jet mixing
Plan Views
Profile Views
Used by permission of the Society of Chemical Engineers, Japan (SCEJ)
Many commercial CFD software packages are available. They generally require a sig-
nificant investment in terms of both purchase or lease cost and learning how to effec-
tively build models and utilize the software. Training in fluid mechanics/
hydrodynamics is considered to be a prerequisite for the effective use of CFD model-
ing.
In studying tanks, CFD models can be used to determine the flow and velocity pat-
terns within a tank and to illustrate the behavior of a tracer as it moves through the
tank. Graphical depictions of the tracer over time can show areas where flow is stag-
nant and where older parcels of water would be expected to be found. By examining
different inlet configurations (location, diameter, angle), different flow rates, and
potential temperature effects with the CFD model, the engineer can select a proper
tank design and operation. An example plot generated by a CFD model of a tank is
Figure 8.37
Plot generated by a
CFD model
Courtesy of Red Valve Company, Inc.
8.11 OPTIMIZED DESIGN AND REHABILITATION

PLANNING
The capital costs of a water distribution system combined with the cost of maintaining
and repairing the system are often immense. Researchers and practitioners are con-
stantly searching for new ways to create more economical and efficient designs. How-
ever, the design and management of water distribution systems, if they are to be
completed in the most effective and economic manner, are complex tasks requiring a
systematic and scrupulous approach backed by skillful engineering judgment and sig-
nificant capital resources. Optimization, as it applies to water distribution system
design and rehabilitation planning, is the process of finding the best, or optimal,
solution to the problem under consideration. (See Appendix D on page 643 for more
information on optimization processes and techniques.)
The earliest attempts at optimization date back to Babbitt and Doland (1931) and
Camp (1939). The first computerized optimization was attempted by Schaake and Lai
(1969). By 1985, Walski (1985) had documented nearly one hundred papers on the
subject, and the number of papers has increased significantly since that time (Lansey,
2000). The methods used have included such techniques as linear programming,
dynamic programming, mixed integer programming, heuristic algorithms, gradient
search methods, enumeration methods, genetic algorithms, and simulated annealing
(see Appendix D). The models have been tested against standard modeling problems
Section 8.11 Optimized Design and Rehabilitation Planning 361
such as the New York tunnel problem (Schaake and Lai, 1969) and the Anytown prob-
lem (Walski et al., 1987), and even on real systems (Jacobsen, Dishari, Murphy, and
Frey, 1998; Savic, Walters, Randall-Smith, and Atkinson, 2000), and found to give
reasonable answers.
Despite such extensive research, optimization has not found its way into standard
engineering practice, partly because existing algorithms have not typically been pack-
aged as user-friendly tools. More significantly, with any model, differences always
exist between reality and the model. This is certainly true for optimization models —
the algorithms do not fully capture the design process (Walski, 2001). However, opti-
mization should not be viewed as an automated process by which only one solution is
identified. Rather, it is a process by which alternative solutions that provide, ideally, a
range of cost and benefits are generated. The full involvement of design engineers is
required. (See Appendix D on page 643 for more information on how optimization is
to be used.)
Most optimization algorithms set up the problem as one of minimizing costs subject
to (1) hydraulic feasibility, (2) satisfaction of demands, and (3) meeting of pressure
constraints.
In reality, design engineers need to consider these plus many more factors, including
• A reasonable level of redundancy and reliability
• Budgetary constraints
• Tradeoffs between different objectives (for example, fire protection versus
water quality)
• Uncertain future
As Walski, Youshock, and Rhee (2000) pointed out using illustrative example prob-
lems, optimization models have not yet been able to fully address many of these con-
siderations. For this reason, most design engineers prefer to use a combination of
steady-state and EPS model runs and engineering judgment as they develop their
designs.
Among the techniques that show promise, genetic algorithms (GA), discussed on
page 365 (and in more detail on page 673), are the most capable of meeting the needs
of the design engineer without contorting the problem to fit the algorithm. Although
optimization may increasingly serve as another tool for design engineers, it is not
likely to replace good engineering judgment.
Optimal Design Formulation

The design of water distribution systems is often viewed as a least-cost optimization
problem with pipe diameters acting as the primary decision variables. However,
although the cost of operating a water distribution system can be substantial (arising
from maintenance, repair, water treatment, energy costs, and so on), the costs of some
items often do not greatly depend on pipe size. In most situations, pipe layout, con-
nectivity, and imposed minimum head constraints at pipe junctions (nodes) are taken
as fixed design targets.
Clearly, other elements (such as service reservoirs and pumps) and other possible
objectives (reliability, redundancy, flexibility in the face of uncertain future demands,
and satisfactory water quality) can be included in the optimization process. But the
difficulties of including reservoirs and pumps and quantifying additional objectives
for use within the optimization process have focused researchers on determining pipe
diameters while maintaining the single objective of least cost. Typically, pumping and
storage alternatives are taken as entirely separate approaches that are considered out-
side of the optimization process (see Sections 8.9 and 10.8). Even this somewhat lim-
ited formulation of optimal network design offers a difficult problem to solve (Savic
and Walters, 1997). The objective function of the pipe-sizing problem is assumed to
be a cost function of pipe diameters and lengths:
min f x = ¦ ci xi li (8.11)

x
i=1

x = vector of unknown diameters xi
N = number of pipes
ci = cost function for pipe i
li = length of pipe i
The set of constraints associated with this problem consist of continuity and energy
loss equations, which can be satisfied by running a standard hydraulic simulation pro-
gram to evaluate the hydraulics of the solution. Other constraints may include
• The minimum and maximum head constraint at each or selected nodes
• The minimum and maximum velocity in pipes
• The minimum reliability and redundancy constraints
• Other operational constraints, such as balancing reservoirs within 24 hours
or any other period, or ensuring at least a minimum turnover of water in stor-
age
Rehabilitation Planning. The rehabilitation planning problem can be formu-

lated similarly to the pipe-sizing problem because rehabilitated pipes (cleaned,
replaced, duplicated, and so on) will acquire new discrete diameters and new friction
characteristics, as already included in the design formulation outlined in this section.
In addition to pipe sizing and rehabilitation, other system elements, such as tanks,
valves, and pumps, should be considered for a systematical design (Wu and Simpson,
2001). A comprehensive optimization of water distribution systems is formulated by
including pipe sizing, rehabilitation, and tank and pump design for all of the system
components under steady-state and extended-period simulation conditions.
Staged Development. For most optimization models, it is assumed that a total
distribution system is built all at once with a single design flow. However, distribution
systems evolve over many decades in response to demands that the original system
designers may or may not have anticipated. An individual project for an estimated
design flow may be optimized with relative ease, but staging the construction of a sys-
tem over a span of years is much more complicated.
Halhal, Walters, Savic, and Ouazar (1999) developed a network optimization method-
ology that accomplishes optimal scheduling of water distribution network improve-
ments. This methodology introduces the notion of time, allowing the method to
consider, in the evaluation of different designs, various time-dependent influencing
factors, such as inflation, interest rate, variation of network characteristics, and so on.
The method defines the design alternatives to be undertaken in the network and their
timing in the planning period, such that the accumulated benefit along the planning
period is maximized while the sum of the present value of the different investments is
minimized and maintained below the total fund allocated to the whole project.
Optimal Design Methods

Methods for finding the best design solution include both trial-and-error approaches
and formal optimization methods. The term optimization methods often refers to
mathematical techniques used to automatically adjust the details of the system in such
a way as to achieve the best possible system performance or, alternatively, the least-
cost design that achieves a specified performance level. The following are just a few
methods that have been applied to optimized design and rehabilitation planning of
water distribution models. A longer list and more details of the optimization methods
can be found in Appendix D.
Trial-and-Error Approach. In practice, the experienced design engineer will

adopt rules of thumb and leverage personal experience to avoid analyzing every possi-
ble configuration. This allows him or her to focus on schemes that are reasonably
cost-effective. With the aid of a hydraulic network solver, the designer traditionally
adopts a trial-and-error approach to produce a few feasible solutions (solutions that
satisfy design constraints), which can then be priced. On large systems, a number of
factors limit the effectiveness of the manual design method:
• A problem can have many possible solutions; therefore, the number of alter-
native solutions considered is limited by available time and financial
resources.
• The changes made at one location may influence the performance at another,
resulting in a highly nonlinear system where it is hard to manually relate
cause and effect.
Therefore, it is likely that solutions developed using the trial-and-error procedure are
successful in meeting design criteria with respect to constraints (pressure, velocity,
and so on) but are less successful at delivering these benefits at least cost.
Partial Enumeration Method. In 1985, Gessler suggested a simplified

approach based on enumeration of a limited number of alternatives. In his work, he
devised tests to eliminate certain inferior solutions from evaluation by a hydraulic
Perspectives on System Design

Part of the difficulty in applying optimization to Construction: “If you’re going to tear up a street
water distribution design lies in describing the and dig a hole, it doesn’t cost much more to put in
design objectives. Different parties in the deci- a big pipe.”
sion-making process have different perspectives, Operations: “Give us flexibility and redundancy
as summarized in the following list: so we aren’t hanging on a single pipe or pump.”
Customers: “We want great service at low price.” Fire Protection: “Have you ever had to carry a
Upper Management: “Provide adequate capacity body bag out of a building because you didn’t
but remain within the capital budget.” have enough water to fight the fire? Give us plenty
Planning: “Meet demands even though there is a of water.”
great deal of uncertainty in forecasts.” These different perspectives make it difficult to
Engineering: “When in doubt, build it stout.” mirror the decision-making process with a com-
puterized optimization.
simulation model (that is, testing for pressure constraints). In the process of testing,
the technique takes advantage of two considerations:
• After a combination of pipe sizes that gives a hydraulically feasible solution
has been found, there is no need to test any other pipe size combination that
is significantly more expensive.
• After an infeasible solution has been encountered, any other size combina-
tion, with all sizes equal to or less than these is an infeasible solution.
Pipe-link grouping by pipelines was another innovation in this work. Utilities are
unlikely to change diameter from block to block and insert bottlenecks in the system,
even though they might save a few dollars by doing so and still meet minimum pres-
sure requirements.
Because of the considerations just described, the partial enumeration method does not
require calculation of flow and pressure distribution for all pipe combinations. Indeed,
the larger the total number of combinations, the smaller the percentage of combina-
tions for which the pressure distribution needs to be calculated. However, Murphy and
Simpson (1992) subsequently dhowed that the approach failed to identify the optimal
design for a moderately small network expansion problem even though it did better
than any traditional optimization approach in the 1985 “Battle of the Network Mod-
els” (Walski et al., 1987).
Linear Programming Methods. Linear programming approaches (see page

655) were also used to reduce the complexity of the original nonlinear nature of the
problem by solving a sequence of linear sub-problems (Alperovits and Shamir, 1977;
Goulter and Morgan, 1985; and Fujiwara and Khang, 1990). The decision variables
are the lengths of pipe of a specific diameter:
xij = length of pipe i of size j (8.12)

An additional constraint is introduced to ensure that the sum of all segments of pipe
between any two nodes is equal to the length between those nodes:
¦ xij = li (8.13)
j=1
where J = the total number of pipe sizes

The optimum solution obtained by this method consists of one or two pipe segments
of different discrete sizes between each pair of nodes. It is known, however, that so-
called split-pipe solutions are not desirable when short pipe lengths of varying diame-
ters are used. For a more realistic solution, the split-pipe design should be altered so
that only one diameter is chosen for each pipe.
Nonlinear Programming Methods. Nonlinear programming methods (El-
Bahrawy and Smith, 1985; Duan, Mays, and Lansey, 1990) have also been tried on
pipe network design problems (see page 659). These methods rely on knowledge of
the functional relationship between the objective function value and the decision vari-
ables, thus requiring calculation of partial derivatives of the objective function with
respect to the decision variables. For pipe design problems, this is only possible if
pipe diameters are considered continuous.
Because they treat pipe diameters as continuous variables, tend to get stuck in local
optima, and have limitations to the size of problem they can handle, the use of nonlin-
ear programming methods for pipe design problems is limited.
Search Methods. Although the linear and nonlinear methods are good for find-
ing local optima, in real problems it quickly becomes inconvenient to invert matrices
(linear programming) or calculate the partial derivatives with respect to the decision
variables (nonlinear programming). In such a situation, knowledge of the functional
relationship between the objective function value and the decision variables either
does not exist or is too complex to be usable. Automated search methods are then
used instead of computationally intensive mathematical programming approaches.
The feature common to all of these methods is a generate-and-test strategy in which a
new point is generated and its function value tested. Depending on the particular
method, a new point (or set of points) is generated, and the search for the best solution
continues.
Genetic Algorithms. Most real network models are too large or too complex to
be handled by any of the previously discussed optimization methods without making
significant simplifications. Among the techniques that show promise, genetic algo-
rithms (GAs) are most capable of meeting the needs of the design engineers without
the necessity of contorting the problem to fit the algorithm (Dandy, Simpson, and
Murphy, 1996; Savic and Walters, 1997; Walters, Halhal, Savic, and Ouazar, 1999;
Wu et al., 2002). See page 673 for more details on genetic algorithms.
GAs have a relatively short but promising history, although the basic principles date
from the beginning of life on earth. In simple terms, the GA uses a computer model of
Darwinian evolution to “evolve” good designs or solutions to highly complex prob-
lems for which classical solution techniques such as linear programming or gradient-
based methods are often inadequate. The GA incorporates ideas such as a population
of solutions to a problem, survival of the fittest (most suitable) solutions within a pop-
ulation, birth, death, breeding, inheritance of genetic material (design parameters) by
children from their parents, and occasional mutations of that material (thereby creat-
ing new design possibilities).
A GA developed for distribution system optimization uses
• An objective function defined on a set of decision variables (pipe diameters,
for example)
• A calibrated model of the system to simulate its hydraulic behavior and to
ensure that continuity and head-loss equations are satisfied at all times (hard
constraints)
• A penalty term to penalize insufficient levels of service (soft constraints),
such as pressures at nodes, imbalance of reservoir flows, or low/high veloc-
ity in pipes.
Optimization Issues
Optimization methods, as presented so far, deal mostly with the pipe-sizing problem.
This is a simplified approach to solving design and rehabilitation planning problems.
The following sections discuss a number of key points related to the use of optimiza-
tion methods. These include cost data implications, reliability and redundancy of
designs, uncertainty, pipe sizing decisions influencing future development and
demands, and treatment of pumps and reservoirs.
Cost Data Implications. Cost data are often the most overlooked part of the
design analysis. Depending on the problem being solved, there may be thousands of
solutions that differ in cost by only one or two percent, yet the costs are only accurate
to +/– 20 percent. This might seem like a fatal flaw in optimization, but the effects of
this type of error are not dramatic as long as relative costs are consistent. For example,
if a 24-in. (600 mm) pipe costs 10 percent more than a 20-in. (500 mm) pipe, even if
the absolute costs are significantly in error, the larger pipe is still likely to be 10 per-
cent more expensive than the smaller one. Therefore, optimization is good at selecting
between different sizes along a given route.
However, when optimization is used to compare different routes, the differences in

cost are not caused simply by size but also factors such as excavation and paving con-
ditions and right-of-way costs such that the uncertainty in comparing costs can lead to
misleading solutions if all factors are not accurately considered. The effects of differ-
ent paving costs can be much greater than the effects of diameter, and uncertainty in
those costs can lead to a misleading “optimal” solution.
The differences in cost-estimating procedures are most critical when one is compar-
ing a traditional method, such as new pipe installation for which the engineer has rel-
atively accurate data, with a more novel approach, such as directional drilling for
which the engineer must rely on a very limited cost database and considerable uncer-
tainty in construction difficulty. In this case, the optimization may be comparing a
cost of $85,000 +/– $5,000 with $80,000 +/– $20,000. Some owners may prefer a
higher cost alternative, which minimizes risk. There are, however, ways to use optimi-
zation to find efficient, robust designs that are adaptable to a range of “wait and see”
strategies, with some economic efficiency or optimality traded in favor of adaptability
and robustness (Watkins and McKinney, 1997). Multiobjective optimization provides
methodologies for generating such tradeoffs and has been applied in water resources
(Haimes and Allee, 1982) and water distribution design (Walski, Gessler, and Sjos-
trom, 1990; Halhal, Walters, Savic, and Ouazar, 1999; Dandy and Engelhardt, 2001).
More about multiobjective optimization can be found on page 677 in Appendix D.
In summary, the cost of pipe installation is not simply a function of length and diame-
ter. The cost functions (cost versus diameter) used in any optimization must reflect the
actual costs of the installed pipe. Using a single cost function, which does not account
for differences in laying conditions, surface cover, and the need to acquire right-of-
way and traffic control, will result in a misleading “optimal” solution.
Reliability/Redundancy. Regardless of the optimization method, optimization

reduces costs by reducing the diameter of pipes or by completely eliminating them.
This tends to leave the system with barely sufficient capacity to meet the demands
placed on it and no ability to respond to pipe breaks or demands that exceed design
values without failing to achieve required performance levels. This consideration is
extremely important but difficult to incorporate into design studies.
Numerous researchers have sought methods to incorporate reliability and capacity

considerations as summarized by Mays (1989) and Goulter et al. (2000). Methods
usually involve forcing the closure of loops by fixing minimum diameters in pipes or
evaluating solutions with key pipes out of service.
As discussed previously, loops provide extra reliability only if there is adequate valv-
ing to isolate areas affected by pipe breaks and maintenance work. Each leg of a loop
should be able to carry significant flow. A loop with a 24-in. pipe on one side and a 2-
in. pipe on the other is really not capable of providing sufficient flow if the 24-in. pipe
should be out of service.
Forcing adequate reliability while minimizing costs is especially difficult because no
universally accepted quantifiable definition of water distribution system reliability
exists. Failure can be defined in terms of length or number of service interruptions,
number of customers interrupted, duration and magnitude of pressure shortfalls, or
some combination thereof.
Uncertainty in System Planning. The greatest source of uncertainty in sys-
tem planning is the uncertainty of demand projections. In smaller systems, an eco-
nomic downturn, the closing of a factory, the decision of an irrigator to switch to
reclaimed water, the installation of a fire sprinkler system in a school, or the opening
of a new dormitory at a school or cell block at a prison can make the most “optimal”
plans incorrect. For large systems, although a single event may not impact demand
projections, great uncertainty still exists with the projection of future demands.
Erring on the conservative side by oversizing pipes can lead to higher capital costs as
well as longer travel times through the system that can adversely impact the water
quality. However, oversized pipes also mean that the system can deliver more fire
flow and can provide extra capacity for future growth. Erring on the low side by
undersizing pipes can result in low pressures, inadequate fire flows, and moratoriums
on new construction. Both undersizing and oversizing pipes can have serious conse-
quences, and every effort should be made to balance the risk against cost savings and
benefits in the system.
Accounting for uncertainty in demand forecasting makes pipe design a tradeoff
between least-cost design and designs that maximize capacity. Optimization methods
that allow the user to examine the tradeoff between capacity and cost are referred to as
multiobjective optimization techniques (see page 370 and page 677) and have been
applied in water resources (Haimes and Allee, 1982) and water distribution design
(Walski, Gessler, and Sjostrom, 1990; Halhal, Walters, Savic, and Ouazar, 1999;
Dandy and Engelhardt, 2001).
Pipe Sizing Controlling Demands. Another factor that complicates optimal
design is that optimization methods often assume that pipe sizing does not have a
measurable effect on demands. That is, demands are considered as driving pipe siz-
ing, and it is assumed that there is no feedback. In reality, however, the location of
piping capacity significantly affects demand developments. Indeed, real estate devel-
opers are more likely to develop land parcels that are located near utilities with capac-
ity to serve the planned development.
For example, consider a town that is growing to its west along two main roads, Green
Road and Red Road. If the water and wastewater utilities install large pipes along
Green Road and nothing along Red Road, then development will occur much more
rapidly along Green Road. The provision of water system capacity is essentially a
“self-fulfilling prophecy,” and to some extent, the water utility’s pipe size and location
decisions will be correct regardless of where it installs the pipelines.
Treatment of Pumps and Reservoirs. The presence of pumps requires that

both the design and the operation of the network should be considered in the optimi-
zation. This means that the cost of a solution must include not only the capital costs of
pipes, pumps, and tanks but also the operating expenditure over a specified period,
with all the costs expressed in equivalent present value (see Sections 8.9 and 10.8).
The method developed by Savic, Walters, Randall-Smith, and Atkinson (2000) allows
for the optimal selection of pumps for installation in new or upgraded pumping sta-
tions.
Provision of new service reservoirs or expansion of existing ones can also be incorpo-
rated. Including service-reservoir storage requires simulation of the filling and empty-
ing of the reservoirs through the daily (or even longer) cycle of demands, in addition
to analysis of the instantaneous peak and emergency flows. Full simulation of the sys-
tem’s response to the variations in demand over a day is currently too time-intensive
for use in an optimal design program that requires evaluation of a very large number
of designs. Hence, only a small number of representative periods of the day may be
considered for the evaluation of the system performance during the optimization, with
a full 24-hour simulation used to check the feasibility of the final designs. The use of
the small number of steady-state spatial demand distributions (minimum average-
hour and maximum average-hour demand or several others) rather than a full 24-hour
simulation requires an approximate technique to be used for ensuring consistency
between storage volumes, reservoir levels, and reservoir inflows and outflows. This is
obviously a simplification of the problem and, as with any model, involves a tradeoff
between realism and efficiency.
In addition to introducing new variables (causing an increase in the dimensions of the
problem), the treatment of reservoir and pump flows within optimization requires the
algorithm to deal with not only discrete variables but also with a mixture of discrete
and continuous variables. This introduces yet another difficulty in trying to find a glo-
bal optimum. The modeling and solution of such optimization problems has not yet
achieved the maturity of linear programming techniques; however, such problems
have a rich area of application in design. Genetic algorithms and other adaptive search
techniques (see Appendix D on page 643 for more information on these techniques)
offer a potential solution to the problem.
Multiple Objectives and the Treatment of the Design

Optimization Problem
Like many real-world engineering design or decision-making problems, design of
water distribution networks needs to achieve several objectives: minimize risks, max-
imize reliability, minimize deviations from desired (target) levels, maximize water
quality, minimize cost (both capital and operational), and so on. The principle of mul-
tiobjective optimization is different from that of single-objective optimization. In the
latter, the goal is to find the best solution, which corresponds to the minimum or max-
imum value of a single objective function that lumps all different objectives into one.
In multi-objective optimization, the interaction among different objectives gives rise
to a set of compromised solutions, largely known as the Pareto-optimal solutions.
These solutions are also known as nondominated solutions — that is, there is no other
solution which is better with respect to all objectives. In other words, in going from
one solution to another in this Pareto-optimal set of alternatives, it is not possible to
improve on one objective without making the other objective worse. (See page 678 for
a definition of the Pareto-optimal set of solutions.)
The consideration of many objectives in the design process provides three major
improvements to optimization as a tool that directly supports the decision-making
process (Cohon, 1978):
• A wider range of alternatives is usually identified when a multiobjective
methodology is employed.
• Consideration of multiple objectives promotes more appropriate roles for the
participants in the planning and decision-making processes; the modeler
generates alternative solutions, and the decision-maker uses the solutions
generated by the model.
• Models of a problem are more realistic if many objectives are considered,
because real design problems are almost inevitably multiobjective.
Multiobjective Decision-Making
Least-cost optimization works by reducing pipe sizes and other infrastructure require-
ments. However, as additional infrastructure is added (such as larger pipes, larger
tanks, more pumps, and so on), the benefits of a project in terms of capacity and abil-
ity to deal with uncertainty increase. The tradeoffs can be seen in Figure 8.38 (Walski,
2001), which illustrates that when the benefits of additional infrastructure are
included in the optimization, the “best” decision moves from the least-cost decision
to one that favors some excess capacity. Walski, Youshock, and Rhee (2000)
showed that in a real study, decision-makers consistently favored more robust solu-
tions over least-cost solutions, and the final decision hinged on much more than
finding the least-cost combination of physical infrastructure improvements that meet
the hydraulic constraints.
Figure 8.38
Objective function for
maximizing net
benefits
Evolutionary algorithms such as genetic algorithms make it very easy to investigate

the tradeoffs involved in pipe sizing and other decisions because they evaluate the
benefits at every trial, not just gradients. Calculating benefits and cost is usually a triv-
ial step in an evaluation when compared with the computational time involved in solv-
ing the hydraulic equations. Thus, optimal cost-benefit tradeoff can be easily included
in a GA optimization process (Wu et al., 2002). By quantifying the hydraulic benefit
resulting from providing flow or pressure in excess of the absolute minimum
required, the GA can determine the optimal capacity design by maximizing the bene-
fit while meeting the hydraulic constraints and the budget available for a design. The
problem is posed as one of “Given a fixed budget, how much capacity can we add?”
The difficulty in multiobjective design is trying to quantify the benefits of an alterna-
tive. Usually, the flow that can be delivered to a number of nodes during fire events
can be a good indicator of capacity. Similarly, the amount of pressure in excess of
some minimum pressure at several indicator nodes can be used.
Enumeration methods generate large numbers of solutions, many of which are “infe-
rior.” That is, there is another solution that can give greater benefits at a lower cost.
Enumeration techniques can discard these inferior solutions and save only the “non-
inferior” (or Pareto-optimal) solutions. These are the solutions the user is most likely
to choose. Figure 8.39 shows typical results of a multiobjective analysis in terms of
two objectives. The points in Figure 8.39 represent trial solutions, and the solid line
represents a theoretical set of solutions that would hold true if discrete pipe sizes were
not needed.
Figure 8.39
Typical tradeoffs
between capacity and
cost
Using multiobjective analysis, the decision-makers can better assess the tradeoffs
between cost and capacity, and although they cannot identify a clear “best” solution,
they have reasonable grounds to make decisions and know which decisions are poor.
Halhal, Walters, Savic, and Ouazar (1997) developed two multiobjective optimization
methods based respectively on a standard genetic algorithm and an improved, so-
called structured messy, genetic algorithm. Both use the concept of Pareto-optimal
selection (see page 677 for more information on multiobjective optimization) and
were developed to find the best way to invest judiciously some or all of an available
budget by providing a tradeoff curve between different objectives. The main objec-
tives were to improve the carrying capacity of the water distribution system, the phys-
ical integrity of its pipes, the water quality, and the system flexibility. The structure of
the problem studied was such that only a small subset of the design variables (pipeline
upgrading options) would be selected in feasible solutions due to funding constraints.
The progressive building up of solutions from simple elements developed for struc-
tured messy genetic algorithm (GA), combined with the multiobjective approach,
which keeps a range of good solutions with varied costs throughout the process,
proved very effective. Wu et al. (2002) advanced this with the user-friendly Darwin
model, which enables the user to define a wide variety of fitness measures for trial
solutions.
Using Optimization
Using a water distribution optimization model is very similar to simulation models
used in more traditional design analyses. Figure 8.40 illustrates the steps involved.
Figure 8.40
A water distribution
optimization model
References 373
REFERENCES
Alperovits, E., and Shamir, U. (1977). “Design of Optimal Water Distribution Systems.” Water Resources
Research, 13(6), 885.
American Water Works Association (1989). “Installation, Field Testing, and Maintenance of Fire
Hydrants.” AWWA Manual M-17, Denver, Colorado.
Babbitt, H. E., and Doland, J. J. (1931). Water Supply Engineering. McGraw-Hill, New York, New York.
Camp, T. R. (1939). “Economic Pipe Sizes for Water Distribution Systems.” Transactions of the American
Society of Civil Engineers, 104, 190.
Cesario, A. L. (1995). Modeling, Analysis, and Design of Water Distribution Systems. AWWA, Denver,
Colorado.
Cohon, J. L. (1978). Multiobjective Programming and Planning. Academic Press, New York, New York.
Cunha M. D., and Sousa, J. (1999). “Water Distribution Network Design Optimization: Simulated Anneal-
ing Approach.” Journal Of Water Resources Planning And Management, ASCE, 125(4), 215.
Dandy, G. C., and Engelhardt. (2001). “Optimum Rehabilitation of a Water Distribution System Consider-
ing Cost and Reliability.” Proceedings of the World Water and Environmental Resources Congress,
Orlando, Florida.
Dandy, G. C., Simpson, A. R., and Murphy, L. J. (1996). “An Improved Genetic Algorithm for Pipe Net-
work Optimization.” Water Resources Research, 32(2), 449.
Duan, N., Mays, L. W., and Lansey, K. E. (1990). “Optimal Reliability-Based Design of Pumping and Dis-
tribution Systems.” Journal of Hydraulic Engineering, ASCE, 116(2), 249.
El-Bahrawy, A., and Smith, A. A. (1985). “Application of MINOS to Water Collection and Distribution
Networks.” Civil Engineering Systems, 2(1), 38.
Fujiwara, O., and Khang, D. B. (1990). “A Two-Phase Decomposition Method for Optimal Design of
Looped Water Distribution Networks.” Water Resources Research, 26(4), 539.
Gessler, J. (1985). “Pipe Network Optimization by Enumeration.” Proceedings of the Specialty Conference
on Computer Applications in Water Resources, American Society of Civil Engineers, New York, New
York.
Goldberg, D. E., Korb, B., and Deb, K. (1989). “Messy genetic algorithms: Motivation, analysis, and first
results.” Complex Systems, 3, 493.
Goulter, I. C., and Morgan, D. R. (1985). “An Integrated Approach to the Layout and Design of Water Dis-
tribution Networks.” Civil Engineering Systems, 2(2), 104.
Goulter, I. C., Walski, T. M., Mays, L. W., Sekarya, A. B. A., Bouchart, R., and Tung, Y. K. (2000). “Reli-
ability Analysis for Design.” Water Distribution Handbook, Mays, L. W., ed., McGraw-Hill, New York,
New York.
Grayman, W. M., and Kirmeyer, G. J. (2000). “Water Quality in Storage.” Water Distribution Handbook,
Mays, L. W., ed., McGraw-Hill, New York, New York.
(2000). Water Quality Modeling of Distribution System Storage Facilities. AWWA and AWWA
Research Foundation, Denver, Colorado.
Great Lakes and Upper Mississippi River Board of State Public Health & Environmental Managers
(GLUMB) (1992). Recommended Standards for Water Works. Albany, New York.
Haestad Methods, Inc. (2002). “Interview with Dr. Ezio Todini: GGA Inventor,” ClienCare Newsletter,
Mar/Apr 2002.
Haimes, Y. Y., and Allee, D. J. (1982). Multiobjective Analysis in Water Resources. American Society of
Civil Engineers, New York, New York.
Halhal, D., Walters, G. A., Savic, D. A., and Ouazar, D. (1999). “Scheduling of Water Distribution System
Rehabilitation using Structured Messy Genetic Algorithms.” Evolutionary Computation, 7(3), 311.
Hydraulic Institute (1979). Engineering Data Book. Hydraulic Institute, Cleveland, Ohio.
Jacobsen, Dishari, Murphy, and Frey (1998). "Las Vegas Valley Water District Plans For Expansion
Improvements Using Genetic Algorithm Optimization." Proceedings of the AWWA Information Man-
agement and Technology Conference, American Water Works Association, Reno, Nevada.
Lansey, K. E. (2000). “Optimal Design of Water Distribution Systems.” Water Distribution Systems Hand-
book, Mays, L. W., ed., McGraw-Hill, New York, New York.
Maier, H. R., Simpson, A. R., Foong, W. K., Phang, K. Y., Seah, H. Y., and Tan, C. L. (2001). “Ant Colony
Optimization for the Design of Water Distribution Systems.” Proceedings of the World Water and Envi-
ronmental Resources Congress, Phelps, D., and Sehlke, G., eds., Orlando, Florida.
Mays, L. W., ed. (1989). Reliability Analysis of Water Distribution Systems. ASCE Task Committee on
Risk and Reliability Analysis, New York, New York.
Murphy, L. J., Dandy, G. C., and Simpson, A. R. (1994). “Optimal Design and Operation of Pumped Water
Distribution Systems.” Proceedings of the Conference on Hydraulics in Civil Engineering, Australian
Institute of Engineers, Brisbane, Australia.
Murphy, L. J., and Simpson, A. R. (1992). “Genetic Algorithms in Pipe Network Optimization.” Research
Report No. R93, Department of Civil and Environmental Engineering, University of Adelaide, Austra-
lia.
Okita, N., and Oyama, Y. (1963). “Mixing Characteristics in Jet Mixing.” Japanese Chemical Engineering,
1(1):94-101.
Savic, D. A., and Walters G. A. (1997). “Genetic Algorithms for Least-Cost Design of Water Distribution
Networks.” Journal of Water Resources Planning and Management, ASCE, 123(2), 67.
Savic, D. A., Walters, G. A., Randall-Smith, M., and Atkinson, R. M. (2000). “Large Water Distribution
Systems Design Through Genetic Algorithm Optimisation.” Proceedings of the ASCE Joint Conference
on Water Resources Engineering and Water Resources Planning and Management, American Society of
Civil Engineers, Hotchkiss, R. H., and Glade, M., eds., proceedings published on CD, Minneapolis,
Minnesota.
Schaake, J. C., and Lai, D. (1969). “Linear Programming and Dynamic Programming Applied to Water
Distribution Network Design.” MIT Hydrodynamics Lab Report 116, Cambridge, Massachusetts.
Tchobanoglous, G. (1998). “Theory of Centrifugal Pumps.” Pumping Station Design, Sanks, R.L., ed., But-
terworth, Boston, Massachusetts.
Wagner, J., Shamir, U., and Marks, D. (1988a). “Water Distribution System Reliability: Analytical Meth-
ods.” Journal of Water Resources Planning and Management, ASCE, 114(2), 253.
Wagner, J., Shamir, U., and Marks, D. (1988b). “Water Distribution System Reliability: Simulation Meth-
ods.” Journal of Water Resources Planning and Management, ASCE, 114(2), 276.
Walski, T. M. (1983). “Energy Efficiency through Pipe Design.” Journal of the American Water Works
Association, 75(10), 492.
Walski, T. M. (1984). Analysis of Water Distribution Systems. Van Nostrand Reinhold, New York, New
York.
Walski, T. M. (1985). “State-of-the-Art: Pipe Network Optimization.” Computer Applications in Water

Resources, Torno, H., ed., ASCE, New York, New York.
Walski, T. M. (1993). “Practical Aspects of Providing Reliability in Water Distribution Systems.” Reliabil-
ity Engineering and Systems Safety, Elsevier, 42(1), 13.
References 375
Walski, T. M. (1995). “Optimization and Pipe Sizing Decisions.” Journal of Water Resources Planning and
Walski, T. M. (2000). “Water Distribution Storage Tank Hydraulic Design.” Water Distribution Handbook,
Mays, L. W., ed., McGraw-Hill, New York, New York.
Walski, T. M. (2001). “The Wrong Paradigm—Why Water Distribution Optimization Doesn’t Work.”
Accepted for Journal of Water Resources Planning and Management, ASCE, 127(2), 203.
Walski, T. M., Brill, E. D., Gessler, J., Goulter, I. C., Jeppson, R. M., Lansey, K., Lee, H. L., Liebman, J. C.,
Mays, L. W., Morgan, D. R., and Ormsbee, L. E. (1987). “Battle of the Network Models: Epilogue.”
Journal of Water Resources Planning and Management, ASCE, 113(2), 191.
Walski, T. M., Gessler, J., and Sjostrom, J. S. (1990). Water Distribution Systems: Simulation and Sizing.
Lewis Publishers, Ann Arbor, Michigan.
Walski, T. M., and Ormsbee, L. (1989). “Developing System Head Curves for Water Distribution Pump-
ing.” Journal of the American Water Works Association, 81(7), 63.
Walski, T. M., Youshock, M., and Rhee, H. (2000). “Use of Modeling in Decision Making for Water Distri-
bution Master Planning.” Proceedings of the ASCE EWRI Conference, Minneapolis, Minnesota.
Walters, G. (1998). “Optimal Design of Pipe Networks: A Review.” Proceedings of the International Con-
ference on Computer Methods and Water Resources in Africa, Computational Mechanics Publications,
Springer Verlag.
Walters, G. A., Halhal, D., Savic, D., and Ouazar, D. (1999). “Improved Design of ‘Anytown’ Distribution
Network Using Structured Messy Genetic Algorithms.” Urban Water, 1(1), 23.
Watkins, D.W. Jr., and McKinney, D. C. (1997). “Finding Robust Solutions to Water Resources Problems.”
Wu, Z. Y., Boulos, P. F., Orr, C. H., and Ro, J. J. (2001). “Rehabilitation of Water Distribution System
Using Genetic Algorithms.” Journal of the American Water Works Association, 93(11), 74.
Wu, Z. Y, Walski, T. M., Mankowski, R., Tryby, M., Herrin, G., and Hartell, W. (2002). “Optimal Capacity
of Water Distribution Systems.” Proceedings of the 1st Annual Enviromental and Water Resources Sys-
tems Analysis (EWRSA) Symposium, Roanoke, Virginia.
Yates, D. F., Templeman, A. B., and Boffey, T. B. (1984). “The Computational Complexity of the Problem
of Determining Least Capital Cost Designs for Water Supply Networks.” Engineering Optimization,
7(2), 142.
8.1 English Units: For the system in the figure, find the available fire flow at node J-7 if the minimum
allowable residual pressure at this node is 20 psi. Assume that pumps P1 and P2 are operating and
that pump P3 is off. (This network is also given in Prob8-01.wcd.)
Hint: Connect a constant-head (reservoir) node to junction node J-7 with a short, large-diameter
pipe. Set the HGL of the constant-head node to the elevation of node J-7 plus the required residual
pressure head, and examine the rate at which water flows into it.
Clearwell N
P3-Suc
P2-Suc
P3
P1-Suc P2
P2-Dis P3 -Dis
P1
J-1 P-11
Main Pump P1 -Dis
Station
P-1 J-8 P-9

J-7
J-3 P-2 J-2

P-8
P-10
Industrial Park
P-3 P-6
J-4 P-4
J-5 J-6
P-7
P-5
West Side Tank
Elevation Demand
Node Label
(ft) (gpm)
Clearwell 630 N/A
West Side Tank 915 N/A
J-1 730 0
J-2 755 125
J-3 765 50
J-4 775 25
J-5 770 30
J-6 790 220
J-7 810 80
J-8 795 320
P1 627 N/A
P2 627 N/A
P3 627 N/A

Pipe Label
(ft) (in.) C-factor
P1-Suc 50 18 115
P1-Dis 120 16 115
P2-Suc 50 18 115
P2-Dis 120 16 115
P3-Suc 50 18 115
P3-Dis 120 16 115
P-1 2,350 12 110
P-2 1,500 6 105
P-3 1,240 6 105
P-4 1,625 12 110
P-5 225 10 110
P-6 1,500 12 110
P-7 4,230 6 105
P-8 3,350 6 105
P-9 2,500 6 105
P-10 2,550 6 105
P-11 3,300 4 85
Pump Curve Data
P1 P2 P3
Head Flow Head Flow Head Flow
(ft) (gpm) (ft) (gpm) (ft) (gpm)
Shutoff 305 0 305 0 305 0
Design 295 450 295 450 295 450
Max Operating 260 650 260 650 260 650
a) Which node has the lowest pressure under the fire flow condition?
b) Is the available fire flow at node J-7 sufficient for the industrial park?
c) If the available fire flow is insufficient, what are the reasons for the low available fire flow?
d) Analyze alternatives for improving the available fire flow to node J-7.
SI Units: For the system in the figure, find the available fire flow at node J-7 if the minimum allow-
able residual pressure at this node is 138 kPa. Assume that pumps P1 and P2 are operating and that
pump P3 is off. (This network is also given in Prob8-01m.wcd.)
Hint: Connect a constant-head (reservoir) node to junction node J-7 with a short, large-diameter
pipe. Set the HGL of the constant-head node to the elevation of node J-7 plus the required residual
pressure head, and examine the rate at which water flows into it.
Pipe Label
(m) (mm) C-factor
P1-Suc 15.2 457 115
P1-Dis 36.6 406 115
P2-Suc 15.2 457 115
P2-Dis 36.3 406 115
P3-Suc 15.2 457 115
P3-Dis 36.6 406 115
P-1 716.3 305 110
P-2 457.2 152 105
P-3 378.0 152 105
P-4 495.3 305 110
P-5 68.6 254 110
P-6 457.2 305 110
P-7 1,289.3 152 105
P-8 1,021.1 152 105
P-9 762.0 152 105
P-10 777.2 152 105
P-11 1,005.8 102 85
Elevation Demand
Node Label
(m) (l/s)
Clearwell 192.0 N/A
West Side Tank 278.9 N/A
J-1 222.5 0
J-2 230.1 7.9
J-3 233.2 3.2
J-4 236.2 1.6
J-5 234.7 1.9
J-6 240.8 13.9
J-7 246.9 5.0
J-8 242.3 20.2
P1 191 N/A
P2 191 N/A
P3 191 N/A
Pump Curve Data
P1 P2 P3
Head Flow Head Flow Head Flow
(m) (l/s) (m) (l/s) (m) (l/s)
Shutoff 93.0 0 93.0 0 93.0 0
Design 89.9 28.4 89.9 28.4 89.9 28.4
Max Operating 79.2 41.0 79.2 41.0 79.2 41.0
a) Which node has the lowest pressure under the fire flow condition?
b) Is the available fire flow at node J-7 sufficient for the industrial park?
c) If the available fire flow is insufficient, what are the reasons for the low available fire flow?
d) Analyze alternatives for improving the available fire flow to node J-7.
8.2 English Units: A disadvantage associated with branched water systems, such as the one given in
Problem 3.3, is that more customers can be out of service during a main break. Improve the reliabil-
ity of this system by adding the pipelines in the following table. (This network can also be found in
Prob8-02.wcd.)
Start End Length Diameter Hazen-Williams

Pipe Label
Node Node (ft) (in.) C-factor
P-20 J-1 J-8 11,230 12 130
P-21 J-2 J-4 3,850 8 130
P-22 J-5 J-7 1,500 8 130
P-23 J-11 J-10 680 6 130
a) Complete the tables below for the new looped system.

Pipe Label
(gpm) (ft/1000 ft)
P-1
P-2
P-3
P-4
P-5
P-6
P-7
P-8
P-9
P-10
P-11
P-12
HGL Pressure
Node Label
(ft) (psi)
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
J-11
J-12
b) You can simulate a main break by closing a pipeline. Complete the tables below for the looped
system if pipe P-3 is closed.
Pipe Label
(gpm) (ft/1000 ft)
P-1
P-2
P-3
P-4
P-5
P-6
P-7
P-8
P-9
P-10
P-11
P-12
HGL Pressure
Node Label
(ft) (psi)
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
J-11
J-12
SI Units: A disadvantage associated with branched water systems, such as the one given in Problem
3.3, is that more customers can be out of service during a main break. Improve the reliability of this
system by adding the pipelines in the table below. (This network can also be found in Prob8-
02m.wcd.)
Start End Length Diameter Hazen-Williams
Pipe Label
Node Node (m) (mm) C-Factor
P-20 J-1 J-8 3422.9 305 130
P-21 J-2 J-4 1173.5 203 130
P-22 J-5 J-7 457.2 203 130
P-23 J-11 J-10 207.3 152 130
a) Complete the tables below for the new looped system.

Pipe Label
(l/s) (m/km)
P-1
P-2
P-3
P-4
P-5
P-6
P-7
P-8
P-9
P-10
P-11
P-12
HGL Pressure
Node Label
(m) (kPa)
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
J-11
J-12
b) You can simulate a main break by closing a pipeline. Complete the tables below for the looped
system if pipe P-3 is closed.
Pipe Label
(l/s) (m/km)
P-1
P-2
P-3
P-4
P-5
P-6
P-7
P-8
P-9
P-10
P-11
P-12
HGL Pressure
Node Label
(m) (kPa)
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
J-11
J-12
8.3 Analyze the following changes to the hydraulic network for the system shown in Problem 4.3.
a) Increase the diameters of pipes P-16, P-17, and P-19 from 6 in. to 8 in. Are head losses in these
lines significantly reduced? Why or why not?
b) Increase the head of the High Field pump to 120 percent of current head. Is this head increase
sufficient to overcome the head produced by the Newtown pump? What is the discharge of the
High Field pump station?
c) Decrease the water surface elevation of the Central Tank by 30 ft. Recall that the tank is modeled
as a reservoir for the steady-state condition. How does the overall system respond to this change?
Is the High Field pump station operating? Is the pump operating efficiently? Why or why not?
8.4 English Units: Analyze each of the following conditions for the hydraulic network given in Problem
4.2 (see page 174). Use the data provided in Problem 4.2 as the base condition for each of the scenar-
ios in the following list. Complete the table for these scenarios.
a) Increase the demand at nodes J-7, J-8, J-9, and J-10 to 175 percent of base demands.
b) Increase the demand at node J-6 to 300 gpm.
c) Change the diameter of all 6-in. pipes to 8 in.
d) Decrease the HGL in the West Carrolton Tank by 15 ft.
e) Increase the demands at nodes J-7, J-8, J-9, and J-10 to 175% of base demands, and change the
diameter of all 6-in. pipes to 8 in.
f) Decrease the HGL in the West Carrolton Tank by 15 ft and increase the demand at node J-6 to
300 gpm.
g) Increase the demands at nodes J-7, J-8, J-9, and J-10 to 175 percent of base demands, change the
diameter of all 6-in. pipes to 8 in., and drop the HGL in the West Carrolton Tank by 15 ft.
Miamisburg
Pump Pressure at Pressure at
Time Tank
Scenario Discharge J-1 J-3
(hr) Discharge
(gpm) (psi) (psi)
(gpm)
Part (a) Midnight
Part (b) 2:00 am
Part (c) 7:00 pm
Part (d) Noon
Part (e) 6:00 am
Part (f) 9:00 pm
Part (g) Midnight
SI Units: Analyze each of the following conditions for the hydraulic network given in Problem 4.2
(see page 174). Use the data provided in Problem 4.2 as the base condition for each of the scenarios
listed below. Complete the table for each of the scenarios presented.
a) Increase the demand at nodes J-7, J-8, J-9, and J-10 to 175 percent of base demands.
b) Increase the demand at node J-6 to 18.9 l/s.
c) Change the diameter of all 152-mm pipes to 203 mm.
d) Decrease the HGL in the West Carrolton Tank by 4.6 m.
e) Increase the demands at nodes J-7, J-8, J-9, and J-10 to 175 % of base demands, and change the
diameter of all 152-mm pipes to 203 mm.
f) Decrease the HGL in the West Carrolton Tank by 4.6 m and increase the demand at node J-6 to
18.9 l/s.
g) Increase the demands at nodes J-7, J-8, J-9, and J-10 by 175%, change the diameter of all 152-
mm pipes to 203 mm, and drop the HGL in the West Carrolton Tank by 4.6 m.
Miamisburg
Pump Pressure at Pressure at
Time Tank
Scenario Discharge J-1 J-3
(hr) Discharge
(l/s) (kPa) (kPa)
(l/s)
Part (a) Midnight
Part (b) 2:00 am
Part (c) 7:00 pm
Part (d) Noon
Part (e) 6:00 am
Part (f) 9:00 pm
Part (g) Midnight
8.5 English Units: Determine the available fire flows at node J-8 for each of the conditions presented
below. Assume that the minimum system pressure under fire flow conditions is 20 psi. Use the sys-
tem illustrated in Problem 8.1.
a) Only pump P1 running.
b) Pumps P1 and P2 running.
c) Pumps P1, P2, and P3 running.
d) The HGL in the West Side Tank increased to 930 ft and pumps P1 and P2 running.
e) Pipe P-11 replaced with a new 12-in. ductile iron line (C=120) and pumps P1 and P2 running.
f) Pipe P-11 replaced with a new 12-in. ductile iron line (C=120) and all pumps running.
Available Fire Flow at Node J-8
Scenario
(gpm)
Part (a)
Part (b)
Part (c)
Part (d)
Part (e)
Part (f)
SI Units: Determine the available fire flows at node J-8 for each of the conditions presented below.
Assume that the minimum system pressure under fire flow conditions is 138 kPa. Use the system
illustrated in Problem 8.1.
a) Only pump P1 running.
b) Pumps P1 and P2 running.
c) Pumps P1, P2, and P3 running.
d) The HGL in the West Side Tank increased to 283.5 m and pumps P1 and P2 running.
e) Pipe P-11 replaced with a new 305-mm ductile iron line (C=120) and pumps P1 and P2 running.
f) Pipe P-11 replaced with a new 305-mm ductile iron line (C=120) and all pumps running
.
Available Fire Flow at Node J-8

Scenario
(l/s)
Part (a)
Part (b)
Part (c)
Part (d)
Part (e)
Part (f)
8.6 English Units: A new subdivision is to tie in near node J-10 of the existing system shown in the fig-
ure. Use the information from the data tables below to construct a model of the existing system, or
open the file Prob8-06.wcd. Answer the questions that follow.
J-9 J-10
Crystal Lake P -15
P -16 P -14
Suction
J-1 J-8 J-7
Discharge
P -12 P -13
West Carrolton
Tank
P -1 P -11 P -10
J-2 J-3 J-6
P -2 P -9 P -8
P -3
J-4
Miamisburg
Tank P -4
P -5
J-5
P -6 P -7
(Not To Scale)
Diameter Length Hazen-Williams

Pipe Label
(in.) (ft) C-factor
Suction 24 25 120
P-1 6 1,250 110
P-2 6 835 110
P-3 8 550 130
P-4 6 1,010 110
P-5 8 425 130
P-6 8 990 125
P-7 8 2,100 105
P-8 6 560 110
P-9 8 745 100
P-10 10 1,100 115
P-11 8 1,330 110
P-12 10 890 115
P-13 10 825 115
P-14 6 450 120
P-15 6 690 120
P-16 6 500 120
Elevation Demand
Node Label
(ft) (gpm)
J-1 390 120
J-2 420 75
J-3 425 35
J-4 430 50
J-5 450 0
J-6 445 155
J-7 420 65
J-8 415 0
J-9 420 55
J-10 420 20
Minimum Initial Maximum Tank

Tank Label Elevation Elevation Elevation Diameter
(ft) (ft) (ft) (ft)
Miamisburg Tank 535 550 570 50
West Carrolton Tank 525 545 565 36
Elevation
Reservoir Label
(ft)
Crystal Lake 320
Pump Curve Data

Maximum Maximum
Shutoff Design
Design Head Operating Operating
Pump Label Head Discharge
(ft) Head Discharge
(ft) (gpm)
(ft) (gpm)
PMP-1 245 230 1,100 210 1,600
a) Determine the fire flow that can be delivered to node J-10 with a 20 psi residual.
b) Given the range of possible water level elevations in West Carrolton Tank, what is the approxi-
mate acceptable elevation range for nearby customers to ensure adequate pressures under normal
(nonfire) demand conditions?
c) What can be done for customers that may be above this range?
d) What can be done for customers that may be below this range?
SI Units: A new subdivision is to tie in near node J-10 of the existing system shown in the figure.
Use the information from the data tables below to construct a model of the existing system, or open
the file Prob8-06m.wcd.
Pipe Label
(mm) (m) C-factor
Discharge 533 67.1 120
Suction 610 7.6 120
P-1 152 381.0 110
P-2 152 254.5 110
P-3 203 167.6 130
P-4 152 307.9 110
P-5 203 129.5 130
P-6 203 301.8 125
P-7 203 640.1 105
P-8 152 170.7 110
P-9 203 227.1 100
P-10 254 335.3 115
P-11 203 405.4 110
P-12 254 271.3 115
P-13 254 251.5 115
P-14 152 137.2 120
P-15 152 210.3 120
P-16 152 152.4 120
Elevation Demand
Node Label
(m) (l/s)
J-1 118.9 7.6
J-2 128.0 4.7
J-3 129.5 2.2
J-4 131.1 3.2
J-5 137.2 0.0
J-6 135.6 9.8
J-7 128.0 4.1
J-8 126.5 0.0
J-9 128.0 3.5
J-10 128.0 1.3
Maximum Initial Minimum Tank

Tank Label Elevation Elevation Elevation Diameter
(m) (m) (m) (m)
Miamisburg Tank 173.7 167.6 163.1 15.2
West Carrolton Tank 172.2 166.1 160.0 11.0
Reservoir Label Elevation (m)

Crystal Lake 97.5
Pump Curve Data

Maximum Maximum
Shutoff Design
(m) Head Discharge
(m) (l/s)
(m) (l/s)
PMP-1 74.7 70.1 69.4 64.0 100.9
a) Determine the fire flow that can be delivered to node J-10 with a 138 kPa residual.
b) Given the range of possible water level elevations in West Carrolton Tank, what is the approxi-
mate acceptable elevation range for nearby customers to ensure adequate pressures under normal
(non-fire) demand conditions?
c) What can be done for customers that may be above this range?
d) What can be done for customers that may be below this range?
8.7 A distribution system for a proposed subdivision is shown in the figure. Construct a model of the
system using the data tables provided, or the file Prob8-07.wcd. This system will tie into an existing
water main at node J-10. The water main hydrant flow test values measured at node J-10 are given
below. Flow was directed out of a 2 ½-in. nozzle having a discharge coefficient of 0.9.
Fire Hydrant Static Pressure Residual Pressure Pitot Pressure

Number (psi) (psi) (psi)
2139 74 60 20
J -100 P -105 J -200
P -100 P -200
J -10 P -15 J -110 P -115 J -210
P -10 P -110 P -210
J -20 P -25 J -120 P -125 J -220
P -120
P -220
P -35 J -130
P -130
J -140 P -135 J -230
Determine if the new subdivision will have adequate pressures for a 750 gpm fire flow at each node.
All pipes are new PVC with a Hazen-Williams C-factor of 150.
Model the existing system as a reservoir followed by a pump, with the elevation of the reservoir and
the pump set to the elevation of the connecting node J-10. Use the results of the hydrant flow test as
described on page 329 to generate a pump head curve for this equivalent pseudopump.
Elevation Demand
Node Label
(ft) (gpm)
J-10 390 20
J-20 420 20
J-100 420 20
J-110 415 20
J-120 425 20
J-130 430 20
J-140 450 20
J-200 420 20
J-210 425 20
J-220 445 20
J-230 460 20
Diameter Length
Pipe Label
(in.) (ft)
P-10 6 625.0
P-15 6 445.0
P-25 6 417.5
P-35 6 505.0
P-100 6 250.0
P-105 6 345.0
P-110 6 665.0
P-115 6 412.5
P-120 6 275.0
P-125 6 372.5
P-130 6 212.5
P-135 6 596.5
P-200 6 225.0
P-210 6 550.0
P-220 6 453.5
8.8 Given the two existing systems 2,000 ft apart shown in the figure, develop a system head curve to
pump from a ground tank in the lower, larger system to the smaller, higher system. The pump will be
placed between the “Suction Node” and “Discharge Node” as shown in the network diagram.

Develop additional system head curves for water levels in the discharge tank of 1,170 ft and 1,130 ft.
J -5 P -25 J -12
J -6 J -11 P -65
Discharge Node P-80 Discharge Tank
Suction Node P-35
P-20 P-60 Main 30
Main 10
Main 25 P-10
Supply Main 15 Main 20
J -7 J -13
J -10
J -1 P-90
P-55
P-40 Most of upper system
Suction Tank P-85
P -45
J -9
J -14 J -8

Pipe Label
(in.) (ft) C-factor
Main10 12 2,000 130
Main15 12 5,878 130
Main20 12 3,613 130
Main25 12 2,670 130
Main30 12 3,926 130
P-10 12 29 130
P-20 6 3,514 130
P-25 6 4,988 130
P-35 6 2,224 130
P-40 6 3,276 130
P-45 6 3,198 130
P-55 6 3,363 130
P-60 6 2,345 130
P-65 6 23 130
P-80 6 1,885 130
P-85 6 3,475 130
P-90 6 6,283 130
Supply 12 60 130
Elevation Demand
Node Label
(ft) (gpm)
Suction Node 995 N/A
Discharge Node 995 N/A
J-1 1,082 10
J-5 1,095 10
J-6 1,100 10
J-7 1,098 10
J-8 1,098 10
J-9 1,112 10
J-10 1,115 10
J-11 1,077 10
J-12 1,124 10
J-13 1,122 10
J-14 1,075 10
Most of upper system 1,150 700
Elevation
Tank Label
(ft)
Suction Tank 1,000
Discharge Tank 1,130
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
9
Modeling Customer Systems
Most water distribution system modeling is done by or for water utilities. In some
instances, there are entire water systems that are served by other water utilities
through wholesale agreements, such that the water source is actually the neighboring
system. This type of situation is shown in Figure 9.1 where the source water utility
delivers to an adjacent customer water utility through a meter and a backflow preven-
ter. Some examples are military bases, prisons, university campuses, and major indus-
tries. These systems can include domestic water use, industrial process water, cooling
water, irrigation water use, and fire protection systems. Most water system design
work is the same within a customer’s system as it is within the water utility’s system.
Figure 9.1
Customer water
Backflow
system using utility's
Source Meter Preventer Customer system as a water
Water M Water source
Utility System
There are several principal differences between working for a customer water system
and a utility system. When working for a customer water system, the designer does
not control the source of water, and therefore must model back into the utility system.
More information regarding the extent to which the water utility’s system must be
modeled can be found in Chapter 8 (see page 326). In addition, the designer must
account for head losses in meters and backflow preventers in the customer water sys-
tem, which are usually not an issue for the utility engineer.
394 Modeling Customer Systems Chapter 9
9.1 MODELING WATER METERS

A customer’s water meter is usually a positive displacement technology meter used on
lines sized from 5/8 in. to 2 in., or a turbine technology meter (shown in Figure 9.2)
for lines sized 1-1/2 in. to 20 in. For some applications in which the flow rate varies
greatly, a compound meter is used. This meter houses a positive displacement element
for the low flows and a turbine meter element for the high flows.
Figure 9.2
Turbine meter
6-in. (DN 150-mm) Cold Water Recordall ® Turbo Series Meter courtesy of Badger Meter Inc.
A single register meter can be represented in the model as a minor loss or an equiva-
lent pipe; however, most meter manufacturers do not provide a minor loss coefficient
(KL) for use in modeling. Instead, they provide a curve relating pressure drop to flow
rate, as shown in Figure 9.3. The designer must calculate the KL by finding the flow
and pressure drop for a point on the curve, and then substituting those values into the
Equation 9.1. A point at the high end of the flow range is usually chosen.
4 2
K L = C f 'PD e Q (9.1)
where KL = minor loss coefficient

' P = pressure drop (psi, kPa)
D = diameter of equivalent pipe (in., m)
Q = discharge (gpm, m3/s)
Once KL is determined for a given type of meter, it can be applied to different size
meters of similar geometry. Table 9.1 lists some typical KL values for several types of
meters in representative sizes. AWWA M-22 (1975) discusses meter sizing.
Section 9.1 Modeling Water Meters 395
Figure 9.3
Typical manufacturer
water meter head loss
curve
Courtesy of Hersey Products, Inc.
Table 9.1 Minor loss KL values for various meter types
Size
Type of Meter Minor Loss KL
(in.)
Displacement Meter 5/8 4.4
2 8.3
6 17.2
Turbine 1.5 6.7
4 9.4
12 14.9
Compound 2 3.9
4 18.1
10 33.5
Fire Service Turbine 3 4.1
6 4.1
10 4.3
Multijet 5/8 5.1
1 5.3
2 12.6
In the case of a compound meter, a single KL value does not adequately describe the
pressure drop versus flow relationship. When modeling high-flow conditions, the
larger meter is in operation, and the diameter and KL value for the larger meter are
used. When an accuracy of 2 to 3 psi (13.7 to 20.6 kPa) is required for lower flow
runs, the data for the smaller meter should be used instead. For example, when run-
ning simulations to look at tank cycling, pump operation, or energy consumption, the
flow would typically be passing through the smaller meter. During a fire flow condi-
tion, the larger meter is active and should be used in the model so that head loss is not
overestimated.
If accuracy over the full range of flows is necessary, the compound meter can be mod-
eled as two parallel equivalent pipes using the appropriate sizes and KL values. In the
model, the pipe representing the smaller meter will always be open. For a steady-state
run, the designer must specify whether the larger meter is also open. For an EPS run,
the pipe representing the meter can be opened or closed based on the flow rate
through the pipe immediately upstream, or based upon the head loss across the small
meter. For example, the controls could specify, “If the flow rate is greater than 30 gpm
(0.002 m3/s), or if the head loss is greater than 10 ft (3 m), then open the larger meter.”
Figure 9.4 shows an approximation of an actual compound meter head loss curve, and
Figure 9.5 shows how that meter can be represented in the model. An alternative
approach to modeling compound meters is to use the generalized head loss versus
flow curve definition capability available with some simulation software.
Figure 9.4
Approximation of
compound meter head 12
loss curve
10
8
Pressure Drop, psi
1-in. meter alone
4 X 1 Compound
4 Meter
2 4-in. meter alone
0
0 100 200 300 400 500 600
Flow, gpm
Figure 9.5
Model representation
of compound meter 4 in. Equivalent Pipe
1 in. Equivalent Pipe

Section 9.2 Backflow Preventers 397
9.2 BACKFLOW PREVENTERS

A utility-approved backflow prevention assembly (shown in Figure 9.6) is typically
required for large customers to prevent cross-connections (AWWA M-14, 1990). The
distinguishing feature of backflow preventers is that they require a fairly large pres-
sure drop across the valve before they even begin to open. Consequently, the head loss
through the device can be more significant, especially at low flow, than pipe friction
losses in the service line or minor losses through meters.
Figure 9.6
Backflow preventer
Courtesy of CMB Industries, Inc.
A typical pressure drop curve for a reduced pressure backflow preventer or a double
check backflow preventer is shown in Figure 9.7. Because of the significant drop in
HGL required to open the valve, modeling backflow preventers is more complex than
inserting a check valve on an equivalent pipe with an additional minor loss. There are
several ways to model a backflow preventer.
Figure 9.7
Pressure drop curve
for reduced pressure
backflow preventer
Courtesy of Hersey Products, Inc.

The backflow preventer can be modeled as a pressure breaker valve with a head loss
of Pmin in series with an equivalent pipe that has a check valve and a minor loss coeffi-
cient (see Figure 9.8). The minor loss coefficient is determined from the initial pres-
sure drop plus a single representative point (Q, P) on the valve curve using the
following equation:
4 2
k = C f P – Pmin D e Q (9.2)
where k = minor loss coefficient

P = pressure at representative point on curve (psi, kPa)
Pmin = minimum pressure drop through backflow preventer (psi, kPa)
D = diameter of valve (in., m)
Q = discharge at representative point on curve (gpm, m3/s)
The value of Pmin is the point where the pressure drop curve intersects the vertical axis.
The model approximation of the backflow preventer is shown in Figure 9.9. Backflow
preventers with head loss curves that are difficult to approximate can be modeled
using generalized user-defined head loss versus flow relationships. Some models
enable the user to insert generalized valves for which the user can enter points
describing the relationship between head loss and flow. This feature works best when
the head loss versus flow relationship is strictly increasing (that is, no dips).
Figure 9.8
Model network Pipe with Check Valve
representation of and Minor Loss 'K'
backflow preventer
Inlet Node Pressure Outlet Node

Breaker
Valve
9.3 REPRESENTING THE UTILITY’S PORTION OF THE

DISTRIBUTION SYSTEM
As was the case with an ordinary extension to the distribution system, the model of a
customer’s system cannot simply start at an arbitrary point in the distribution system
serving it. Unless the impact of the customer’s load on the utility’s system is negligi-
ble, the head loss from the source, tank, or pump station that controls pressure must
be accounted for.
Section 9.4 Customer Demands 399
Figure 9.9
25.00 Model approximation
of head loss for
backflow preventer
Model
20.00
Actual
Pressure Drop, psi
15.00
10.00
5.00
0.00
0 200 400 600 800 1,000 1,200 1,400 1,600 1,800 2,000
Flow, gpm
The best way to model the connection depends on the relative size of the customer’s
system compared to the size of the utility’s system in the pressure zone providing ser-
vice. If the customer uses half of the water in the source utility’s system, and causes
half of the head loss, then it is important to model the utility’s system back to a rea-
sonably known source. On the other hand, if the customer’s system represents a negli-
gible percent of the demand, then it may be possible to model the utility’s system as a
reservoir and pump, using the results of a hydrant flow test (see page 332). Of course,
if fire flows are to be provided in the customer system, then the loads cannot be con-
sidered negligible.
9.4 CUSTOMER DEMANDS

The material on demand estimation in Chapter 4 is applicable to customer systems.
When working with a customer’s system, demands may be assigned more precisely
than when modeling an entire system. For small industrial complexes, recent water
usage rates can be determined directly by using meter readings.
Commercial Demands for Proposed Systems

Engineers for commercial developments such as hotels and office buildings may also
want to use modeling for their projects but will not have data on existing customers as
a water utility would. This problem was addressed by the National Bureau of Stan-
dards during the 1920s and ’30s and resulted in the Fixture Unit Method for estimat-
ing demands (Hunter, 1940).
This method consists of determining the number of toilets, sinks, dishwashers, and so
on, in a building and assigning a fixture unit value to each. Fixture unit values are
shown in Table 9.2. Once the total fixture units are known, the value is converted into
a peak design flow using what is called a Hunter curve (see Figure 9.10).
The basic premise of the Hunter curve is that the more fixtures in a building, the less
likely it is that they will all be used simultaneously. This assumption may not be
appropriate in stadiums, arenas, theaters, and so on where extremely heavy use occurs
in a very short time frame, such as at halftime or intermission.
The values in Table 9.2 are somewhat out-of-date, as they were prepared before the
days of low-flush toilets and low-flow shower heads, but a better method has not yet
been developed. This technique is used in the Uniform Plumbing Code (International
Association of Plumbing and Mechanical Officials, 1997) and a modified version is
included in AWWA Manual M-22 (1975). Although the fixture unit assigned may
require some adjustment to reflect modern plumbing practice, the logic behind the
Hunter curve still holds true.
Table 9.2 Fixture units

Fixture Type Fixture Units Fixture Type Fixture Units
Bathtub 2 Wash sink (per faucet) 2
Bedpan washer 10 Urinal flush valve 10
Combination sink & tray 3 Urinal stall 5
Dental unit 1 Urinal trough (per ft) 5
Dental lavatory 1 Dishwasher (1/2'') 2
Lavatory (3/8'') 1 Dishwasher (3/4'') 4
Lavatory (1/2'') 2 Water closet (flush valve) 10
Drinking fountain 1 Water closet (tank) 5
Laundry tray 2 Washing machine (1/2'') 6
Shower head (3/4'') 2 Washing machine (3/4'') 10
Shower head (1/2'') 4 Kitchen sink (1/2'') 2
Hose connection (1/2'') 5 Kitchen sink (3/4'') 4
Hose connection (3/4'') 10
Hunter (1940)
The peak demand as determined by the Fixture Unit Method must be increased to
account for any sprinkler, cooling, and industrial process demands that are not other-
wise included.
Additional work on residential and small commercial demands was conducted under
the Johns Hopkins Residential Water Use Program in the 1950s and '60s (Linaweaver,
Geyer, and Wolff, 1966; Wolff, 1961).
Section 9.5 Sprinkler Design 401
Figure 9.10
Determining peak
500 demand from fixture
units using a Hunter
curve
400
300
Demand, gpm
200
Flush
Valves
100
Toilet
Tanks
0
0 500 1000 1500 2000 2500 3000
Fixture Units
9.5 SPRINKLER DESIGN

Water distribution models can also be used to help design irrigation and fire sprinkler
systems. The principal difference between modeling sprinklers and modeling a typi-
cal water distribution system is that pressure dictates what the sprinkler discharge will
be (the demands are “pressure-based”), while in distribution systems, demands are
typically modeled as if they are independent of pressure.
Starting Point for Model

One of the most important questions in sprinkler studies is where to start the model.
For cases in which sprinklers are fed by pumps from wells, tanks, or ponds, the model
should start at the source. Modeling a sprinkler system that is fed from a larger water
distribution system is more complex. In such a situation, it may be difficult to deter-
mine if the model should begin at the main in the street, or be taken back to the actual
source or tank that will be providing water. The key to this decision is determining the
extent to which sprinkler flows, when combined with other demands, will draw down
the hydraulic grade line in the distribution system. If the effect on pressures in the dis-
tribution system is significant, then it will be necessary to extend the model into the
system. For further explanation on using fire hydrant flow tests to make this determi-
nation and modeling the customer’s connection to the main, see page 332.
If a small pipe, such as a 2-in. (50 mm) rural water system main, feeds the sprinkler
system, then it will almost certainly be necessary to include this pipe in the model,
because the head loss at higher flows will be significant. If the sprinkler system is
being fed from a typical water distribution system, then the meter and backflow pre-
vention assembly must be included in the model by using the techniques described
previously in Section 9.1.
It is important to be conservative when estimating the pressure that will be available
to operate a sprinkler system. The water distribution system can change over time as a
result of tuberculation, additional customers, or changes in pressure zone boundaries.
Water utilities cannot guarantee that they will maintain a specified pressure in the
main permanently (AWWA M-31, 1998).
Sprinkler Hydraulics
Flow out of a sprinkler is governed by the equation for orifice flow:
Q = C d A 2gh (9.3)
where Q = discharge (gpm, m3/s)

Cd = discharge coefficient
A = orifice area (in.2, m2)
g = gravitational acceleration constant (32.2 ft/s2, 9.81 m/s2)
h = head loss across orifice (ft, m)
Rather than explicitly stating the area and discharge coefficient, sprinkler manufactur-
ers usually employ a nominal size and a coefficient, K (not to be confused with the
minor loss KL ). K is a function of the size and type of sprinkler and relates discharge
and pressure according to
Q = K P (9.4)
where K = sprinkler coefficient

P = pressure (psi, kPa)
Table 9.3 shows K-factors for fire sprinklers. It is best to obtain sprinkler K-factors
from the sprinkler suppliers. It is also possible to calculate K from a chart of pressure
drop versus Q.
Table 9.3 Sprinkler discharge characteristics

Nominal Orifice Size Nominal K-factor K-factor Range K-factor Range
(in.) (mm) (gpm/(psi)1/2) (gpm/(psi)1/2) (dm3/min/(kPa)1/2)
1/4 6.4 1.4 1.3–1.5 1.9–2.2
5/16 8.0 1.9 1.8–2.0 2.6–2.9
3/8 9.5 2.8 2.6–2.9 3.8–4.2
7/16 11.0 4.2 4.0–4.4 5.9–6.4
1/2 12.7 5.6 5.3–5.8 7.6–8.4
17/32 13.5 8.0 7.4–8.2 10.7–11.8
5/8 15.9 11.2 11.0–11.5 15.9–16.6
3/4 19.0 14.0 13.5–14.5 19.5–20.9
- - 16.8 16.0–17.6 23.1–25.4
- - 19.6 18.6–20.6 27.2–30.1
- - 22.4 21.3–23.5 31.1–34.3
- - 25.2 23.9–26.5 34.9–38.7
- - 28.0 26.6–29.4 38.9–43.0
Reprinted with permission from NFPA Automatic Sprinkler Systems Handbook, Copyright 1999, National Fire Protection Association, Quincy, MA 02269.
This reprinted material is not the complete and official position of the NFPA on the referenced subject which is represented only by the standard in its entirety.
Approximating Sprinkler Hydraulics

Many hydraulic models can simulate sprinkler hydraulics using flow emitters (see
page 451). With a flow emitter, a modeler need only enter the sprinkler K-factor at a
junction, and the model will determine the discharge as a function of pressure at the
node. The emitter coefficient should be the same as the sprinkler K-factor and the
node corresponding to the sprinkler should be at the exact elevation of the sprinkler,
not the pipe. If the sprinkler is connected to a larger pipe through small branch pipes,
those small pipes must be included in the model as they can account for considerable
head loss.
However, some water distribution system models do not explicitly account for sprin-
kler K-factors. Instead, the sprinkler, and the associated losses, must be modeled as an
equivalent length of pipe discharging to the atmosphere. The atmospheric pressure
downstream of the sprinkler can be represented as a discharge from the equivalent
pipe into a reservoir, tank, or pressure source where the HGL setting of the down-
stream node is equal to the elevation of the sprinkler (see Figure 9.11). With this
approach, the designer must assign a length, diameter, and roughness to the equivalent
pipe representing the sprinkler. It is important to note that there are infinite combina-
tions of D, L, and C that will give the same head loss for the sprinkler. One solution is
to use a 1-in. (25-mm) diameter pipe with a length of 0.271 ft (0.083 m). With these
dimensions, the C-factor for the pipe equals the sprinkler K-factor (Walski, 1995).
Figure 9.11
Model representation
of sprinkler as
equivalent pipe
Pipe
Diameter=D
Length=L
C-factor=C
Sprinkler
K
a. Fire Sprinkler System
Upstream Pipe
D, L, C
Node Upstream
Sprinkler
Node
Equivalent Pipe
D=1 in.
L=0.275 ft
C=K
Atmospheric
Pressure
b. Model Representation
Piping Design
To reduce costs, sprinkler systems are usually laid out in a branched, tree-like pattern.
Unlike looped water distribution systems, which use isolation valves to isolate indi-
vidual segments, sprinkler systems have very few. If repairs are needed on sprinkler
piping, the entire system is generally shut down while repairs are made.
Choosing the Right Sprinkler System

Most sprinkler systems are wet-pipe systems in areas subject to cold temperatures. Two options
which the system is always full of pressurized available to address this situation are antifreeze
water. The individual sprinkler heads have fusible systems and dry-pipe systems.
or frangible links that cause the sprinkler to open In antifreeze systems, the wet-pipe system is filled
when exposed to heat. Although this design works with a mixture of antifreeze and water. The anti-
in the majority of situations, there are a number of freeze solutions recommended in systems that
variations to accommodate special conditions. are connected back into a potable water system
One common variation is a deluge system. In this are chemically pure glycerine or propylene glycol.
type of system, the sprinkler heads are always These systems are generally used to protect
open, and water is kept out of the piping by a main small, unheated areas.
control valve. When a fire occurs, the main valve Dry-pipe systems are filled with pressurized air.
is opened, and all of the sprinklers discharge When heat causes a sprinkler to open, the air
simultaneously. pressure in the system is reduced. This drop in
Deluge systems are typically used in places pressure causes a dry-pipe valve to open, allow-
where heat from a fire is unlikely to cause a sprin- ing pressurized water to travel through the system
kler to open, such as in a building with very high to the sprinkler heads. Because of the time delay
ceilings. When modeling this type of system, all in filling the sprinkler system piping, dry-pipe sys-
sprinkler heads must be modeled as open. A vari- tems are not quite as efficient as wet-pipe sys-
ation of the deluge systems is a preaction system tems in controlling fires. In a water distribution
which is equipped with a valve that opens based simulation, dry-pipe systems are modeled the
on some supplemental detection system. same way as wet-pipe systems (that is, it is
A challenge often encountered in designing sprin- assumed that the system is already filled with
kler systems is how to keep pipes from freezing in water).
Velocities are usually higher in sprinkler system piping than in other distribution pip-
ing. Therefore, the minor losses from each valve and fitting in the system must be
considered. Otherwise, discharge can be overestimated during design.
Sprinkler systems generally have small pipe diameters. With small pipe diameters, the
difference between nominal diameter and actual internal diameter can be significant,
depending on pipe material. For example, nominal 1-in. (250-mm) C901 HDPE pipe
can have an inner diameter ranging from 0.860 in. to 1.062 in. (21.8 mm to 27.0 mm)
depending on the DR (diameter ratio), and copper tubing with the same nominal
diameter can have an inner diameter ranging from 0.995 in. to 1.055 in. (25.2 to 26.8
mm) depending on the type. A 20 percent difference in inner diameter can result in a
40 percent difference in capacity. For this reason, it is important to use the actual
internal diameter when performing sprinkler design.
Sprinkler heads do not require a great deal of pressure to operate and pressures on the
order of 10 psi (70 kPa) are usually sufficient [7 psi (48 kPa) minimum]. While sprin-
klers will still deliver water at lower pressure, their ability to blow off the orifice cap
and produce desirable discharge patterns decreases at lower pressures.
The designer should monitor the pressure at the upstream end of the equivalent pipe to
determine if a particular set of pipe sizes results in adequate pressure. Because the
pressure at the sprinkler is so critical in design, it is important to determine the exact
elevation of the sprinkler heads when assigning the elevation of the nodes in the
model.
The information covered up to this point on sprinkler hydraulics and design is appli-
cable to both fire and irrigation sprinkler systems. There are many similarities
between these two types of systems, but each also has unique features, as detailed in
the following sections.
Fire Sprinklers
National Fire Protection Association (NFPA) Standards 13 (1999c) and 13D (1999b)
govern fire sprinkler design. Additional information is provided in NFPA (1999a) and
AWWA M-31 (1998).
Sprinklers are intended to control fires, not necessarily to completely extinguish them.
Therefore, some allowance is made for hose stream flows from hydrants or fire trucks
when determining sprinkler system performance requirements.
Sprinklers are usually laid out so that only a few need to open to control a fire.
Designs should not be based on the assumption that all sprinklers will operate simul-
taneously, unless there is reason to believe this is the case.
Sprinkler demand is based on the area of sprinkler operation and the associated occu-
pancy group. Using the area/density curves shown in Figure 9.12, the density of water
can be determined. Extensive tables are provided in NFPA (1999c) and Pucholvsky
(1999) describing the types of activities that fall into each occupancy group. See Table
9.4 for some examples of the occupancy types in each group.
Figure 9.12
Area density curve
Reprinted with permission from NFPA Automatic Sprinkler Systems Handbook, Copyright 1999, National Fire Protection Association, Quincy, MA 02269. This
reprinted material is not the complete and official position of the NFPA, on the referenced subject which is represented only by the standard in its entirety.
Table 9.4 Example occupancies

Occupancy Group Occupancy
Light hazard Churches, hospitals, museums, offices
Ordinary hazard 1 Bakeries, dairies, laundries
Ordinary hazard 2 Dry cleaners, post offices, repair garages, wood product assembly
Extra hazard 1 Aircraft hangars, printing, saw mills, rubber reclaiming/vulcanizing
Extra hazard 2 Flammable liquid spraying, plastics processing, solvent cleaning
By multiplying the sprinkler operation area by the density (both determined from Fig-
ure 9.12), the sprinkler demand can be computed. If the area is less than the minimum
area for the curve being used, then the minimum area should be used. For example, if
a light occupancy is less than 1,500 ft2 (139 m2), then the density for 1,500 is used.
In addition to the sprinkler demand, a hose stream allowance is also needed to extin-
guish the fire. Typical values for hose stream flow and duration for sprinklered facili-
ties are given in Table 9.5.
Table 9.5 Hose stream demand and water supply duration requirements
Total Hose Duration
Occupancy or Commodity Classification
Stream (gpm) (minutes)
Light hazard 100 30
Ordinary hazard 250 60–90
Extra hazard 500 90–120
Rack storage, Class I, II, and III commodities up to 12 ft (3.7 m) in height 250 90
Rack storage, Class IV commodities up to 10 ft (3.1 m) in height 250 90
Rack storage, Class IV commodities up to 12 ft (3.7 M) in height 500 90
Rack storage, Class I, II, and III commodities over 12 ft (3.7 m) in height 500 90
Rack storage, Class IV commodities over 12 ft (3.7 m) in height and plas- 500 120
tic commodities
General storage, Class I, II, and III commodities over 12 ft (3.7 m) up to 500 90
20 ft (6.1 m)
General storage, Class IV commodities over 12 ft (3.7 m) up to 20 ft 500 120
(6.1 m)
General storage, Class I, II, and III commodities over 20 ft (6.1 m) up to 500 120
30 ft (9.1 m)
General storage, Class IV commodities over 20 ft (6.1 m) up to 30 ft (9.1 500 150
m)
General storage, Group A plastics d 5 ft (1.5 m) 250 90
General storage, Group A plastics over 5 ft (1.5 m) up to 20 ft (6.1 m) 500 120
General storage, Group A plastics over 20 ft (6.1 m) up to 25 ft (7.6 m) 500 150
Reprinted with permission from NFPA Automatic Sprinkler Systems Handbook, Copyright 1999, National Fire Protection Association, Quincy, MA 02269.
This reprinted material is not the complete and official position of the NFPA on the referenced subject which is represented only by the standard in its entirety.
Sprinkler Pipe Sizing

Traditionally, sprinkler pipe sizing has been based on a “pipe schedule” method
where the pipe size is based on the number of sprinklers being served by the pipe.
While that approach is still allowed in some situations, “hydraulically calculated”
designs are the preferred method. Manual “hydraulically calculated” designs rely on
equivalent pipes to simulate the minor losses and can be cumbersome and approxi-
mate when many sprinklers are flowing. Automatic “hydraulically calculated”
approaches, on the other hand, rely on hydraulic models and can provide a much more
accurate evaluation of the system.
Usually, sprinkler systems are modeled through a series of steady-state runs, each run
corresponding to the operation of a different sprinkler or group of sprinklers. One or
two sprinklers on the top floor at the far end of the building from the service line, usu-
ally referred to as the hydraulically most distant, will typically control design calcula-
tions.
If the sprinkler system is not delivering sufficient flow, the engineer should first try to
increase pipe sizes, thereby reducing head losses. If increasing pipe sizes is ineffec-
tive, the head at the supply main may not be sufficient to operate the system. In this
case, the pressure to the sprinklers must be increased. In most instances, installing a
fire pump in the building is simpler and less expensive than raising the pressure in the
supply main.
Irrigation Sprinklers
Irrigation systems are operated frequently, and are designed so that more of the sprin-
klers can be used simultaneously. While irrigation sprinklers are different from fire
sprinklers, they can still be modeled using orifice flow equations.
For larger systems, opening all the sprinklers simultaneously will tend to use exces-
sive water and require larger piping and meters. To reduce pipe and meter sizes, these
systems are usually “zoned” so that only one set of sprinklers operates at a given time.
If the water source is plentiful and storage volume is not an issue, then the operation
of each zone can be modeled as a separate steady-state analysis. If the amount of stor-
age is an issue (for instance, water is taken from a small pond), then an EPS run
should be used in modeling to ensure that the water supply is adequate.
REFERENCES
American Water Works Association (1975). “Sizing Service Lines and Meters.” AWWA Manual M-22,
Denver, Colorado.
American Water Works Association (1990). “Recommended Practice for Backflow Prevention and Cross
Connection Control.” AWWA Manual M-14, Denver, Colorado.
Hunter, R. B. (1940). “Methods of Estimating Loads in Plumbing Systems.” Report BMS 65, National
Bureau of Standards, Washington, DC.
References 409
International Association of Plumbing and Mechanical Officials (1997). Uniform Plumbing Code. Los
Angeles, California.
Linaweaver, F. P., Geyer, J. C., and Wolff J. B. (1966). A Study of Residential Water Use: A Report Pre-
pared for the Technical Studies Program of the Federal Housing Administration. Department of Hous-
ing and Urban Development, Washington, DC.
National Fire Protection Association (NFPA) (1999). Fire Protection Handbook. Quincy, Massachusetts.
National Fire Protection Association (NFPA) (1999). “Sprinkler Systems in One- and Two-Family Dwell-
ings and Manufactured Homes.” NFPA 13D, Quincy, Massachusetts.
National Fire Protection Association (NFPA) (1999). “Standard for Installation of Sprinkler Systems.”
NFPA 13, Quincy, Massachusetts.
Pucholvsky, M. T. (1999). Automatic Sprinkler Systems Handbook. National Fire Protection Association,
Quincy, Massachusetts.
Walski, T. M. (1995). “An Approach for Handling Sprinklers, Hydrants, and Orifices in Water Distribution
Systems.” Proceedings of the AWWA Annual Convention, American Water Works Association, Ana-
heim, California.
Wolff, J. B. (1961). “Peak Demands in Residential Areas.” Journal of the American Water Works Associa-
tion, 53(10).
9.1 The water system for an industrial facility takes water from the utility’s system through a meter and
reduced pressure backflow preventer. The following figures show the industrial system piping con-
nected to the skeletonized utility system and the head loss curves for the meter and backflow preven-
ter. The total head at the water source in the utility’s system is 320 ft, and the elevation of the
backflow preventer is 90 ft. The meter and backflow preventer are located on pipe P-C-1, a nominal
6-in. pipe. Construct a model of the system under normal demand conditions using the data provided
below.
Utility Res. U-2
P- P-
2
U- U-
U-
1 3
P-
U-3
U-1
Utility System P-
4
U-
U-
5
P-
C-3 P-C-4 C-4

U-4 P-
U-
6 P
P-C-5
P-C-3
-C-
1
PBV-1 P-
C-
Meter and Backflow Valve 2
C-1 P-C-9 C-5
C-2
P-C-8
P-C-6
C-7 P-C-7
C-6
Industrial System
(Not to scale)
25
20
Pressure drop, psi
15
RPBP
10
5 Meter
0
0 300 600 900 1200 1500
Flow, gpm
Node Elevation Demand

Label (ft) (gpm)
C-1 90.0 0
C-2 120.0 5
C-3 100.0 5
C-4 135.0 5
C-5 140.0 5
C-6 135.0 5
C-7 130.0 5
U-1 100.0 200
U-2 95.0 500
U-3 80.0 700
U-4 100.0 200
Pipe Length Diameter Hazen-Williams

Label (ft) (in.) C-factor
P-C-1 1 6 130
P-C-2 50 6 130
P-C-3 500 6 130
P-C-4 500 6 130
P-C-5 500 6 130
P-C-6 500 6 130
P-C-7 500 6 130
P-C-8 500 6 130
P-C-9 500 6 130
P-U-1 3,000 16 130
P-U-2 2,000 12 130
P-U-3 2,000 12 130
P-U-4 2,000 12 130
P-U-5 2,000 12 130
P-U-6 100 12 130
a) Determine the minor loss K-values for the meter and backflow preventer and Pmin for the back-
flow preventer. Apply the minor losses to the pipe immediately downstream of the valve (P-C-1).
b) Determine the head immediately downstream of the backflow preventer and meter during normal
demand conditions.
c) Add a 1,500 gpm fire demand to the normal demand at node C-4 and determine the residual pres-
sure at this node. Under this demand condition, what is the HGL immediately downstream of the
meter?
d) For the 1,500 gpm fire flow condition, determine the head loss (in feet) for the following portions
of the system:
• Between the source and the meter/backflow preventer
• In the backflow preventer and meter
• Between the meter/backflow preventer and C-4
9.2 This problem uses the system from Problem 9.1. Suppose you do not want to model the utility’s sys-
tem at all, even as the skeletal system shown. Rather, you would like to model it as a constant head
node located downstream of the meter and backflow preventer at node C-2. Using the HGL deter-
mined in part (b) of the previous problem, insert a reservoir attached to node C-1 through a 1-ft pipe
with a 6-in. diameter and a Hazen-Williams C-factor of 130. Delete the valve and the utility part of
the system from the model or disconnect the systems.
a) Using this HGL, what is the residual pressure at C-4 for a 1,500 gpm flow?
b) Does deleting the utility system, backflow valve, and the meter and instead modeling it as a con-
stant head give an accurate representation of the system under fire demands?
9.3 An existing small irrigation system consists of five sprinklers in Area A, as shown in the figure
below. A new landscaped area (Area B) of roughly the same size is planned, requiring that an addi-
tional five sprinklers be installed.
Water for the existing irrigation system is pumped from a nearby pond. The owner would like to use
the existing 1.5 hp pump to supply the additional sprinklers as well. Manufacturer pump curve data
for this pump is provided in the following tables. The elevation of the pump is 97 ft.
Construct a steady-state hydraulic model of the sprinkler system if the pond water surface is at an
elevation of 101 ft. Pipes M-3 and M-4 are both equipped with a 1-in. gate valve (K = 0.39) for
isolating the system if necessary (for repairs and so forth) and a 1-in. anti-siphon valve (K = 14) to
prevent contamination of the pond with substances such as chemical fertilizers.
To model the sprinklers, attach a reservoir with an HGL elevation equivalent to the sprinkler head
elevation to the sprinkler node with an equivalent pipe.
According to the sprinkler manufacturer’s information, a pressure of 30 psi is required at the sprin-
kler head to produce a discharge of 1.86 gpm. At this flow, the radius of the sprinkler coverage is 15
ft. The sprinkler spacing was determined based on this radius. Given this information, use Equation
9.4 to solve for the sprinkler coefficient, K, and determine the characteristics of the equivalent pipe
as discussed in Section 9.5.
S-8 S-10
Landscape
Area A
L-1
L-8
.)
p. typ
0
(ty e (
S-7 S-9
lv valv
)
te hon
e
L-9
Ga sip
va
Landscape
L-7
ti-
An
J-3 Area B
PVC Lateral (typ.) L-6
M-4 J-1
S-6 S-4 L-5 S-5
M- J-2
3 L-4
L-1
PVC Main (typ.)

S-1 L-2 S-2 L-3 S-3
2
M-
Pump Sprinkler (typ.)

1
M-
Pond
Water Intake
Sprinkler Elevation
Label (ft)
S-1 115.45
S-2 115.40
S-3 115.25
S-4 115.15
S-5 115.10
S-6 115.75
S-7 116.00
S-8 116.10
S-9 115.55
S-10 115.80
Node Elevation
Label (ft)
J-1 115
J-2 115
J-3 115
Main Length Hazen-Williams

Label (ft) C-factor
M-1 19 150
M-2 80 150
M-3 12 150
M-4 12 150
Lateral Length Hazen-Williams

Label (ft) C-factor
L-1 17 150
L-2 26 150
L-3 26 150
L-4 16 150
L-5 26 150
L-6 16 150
L-7 26 150
L-8 26 150
L-9 17 150
L-10 26 150
Pump Curve Data
Head Flow
(ft) (gpm)
Shutoff 230 0
Design 187 10
Max Operating 83 20
a) The existing system uses ¾-in. laterals and 1-in. mains. Run the model with only the existing sys-
tem in operation (that is, close pipe M-3). Is the pump able to adequately supply all of the sprin-
klers? What is the minimum sprinkler discharge?
b) Re-run the model with all sprinklers (existing and proposed) open. Use ¾-in. laterals and 1-in.
mains for both existing and proposed piping. Is the pump able to adequately supply all of the
sprinklers? What is the minimum sprinkler discharge?
c) If you were designing the entire system from scratch (no pipes have been installed yet), what
minimum size must the mains and laterals be to meet the minimum flow/pressure requirement?
Assume all sprinklers are discharging simultaneously and the pump is the same as described in
the preceding tables.
d) The owner obviously prefers to continue using the existing piping and would like to save on
expenses by using smaller pipes in the new system as well. What could be done operationally to
make such a design work?
9.4 This problem is a continuation of Problem 9.3. The irrigation system will be used to water the land-
scaped areas for 2.5 hours each day. A schedule is established such that Area A will be watered from
4:00 a.m. to 6:30 a.m., and Area B from 6:30 a.m. to 9:00 a.m. The pipe sizes to be used are ¾-in.
laterals and 1-in. mains.
a) Using the existing pump and given the minimum system requirements from Problem 9.3, can
adequate flow/pressure be supplied at all of the sprinklers?
b) You are concerned about whether the pond has enough water for irrigation during a dry spell.
Volume data for the pond is provided in the following tables. Model the pond as a tank using this
volume data, and run an EPS to determine the total volume of water used by the irrigation system
in a daily cycle. Neglecting evaporation and infiltration, extrapolate this rate of consumption to
determine how long could a dry spell last before the pond runs dry.
Pond Data
Total Pond Volume 10,000 ft3

Maximum Pond Elevation 104 ft
Initial Pond Elevation 103 ft
Minimum Pond Elevation 98 ft
Pond Depth to Volume Ratios
Depth Volume
Ratio Ratio
0.0 (elev. = 98 ft) 0.0 (vol. = 0)
0.5 (elev. = 101 ft) 0.3 (vol. = 3,000 ft3)
0.8 (elev = 102.8 ft) 0.7 (vol. = 7,000 ft3)
1.0 (elev. = 104 ft) 1.0 (vol. = 10,000 ft3)
9.5 For a building with an Ordinary Hazard Group 1 occupancy classification, the required minimum
fire sprinkler capacity is 0.16 gpm/ft2 for a 1500 ft2 area. The coverage area for an individual sprin-
kler is 130 ft2.
a) Compute the number of sprinklers required to provide coverage for a 1500 ft2 area.
b) What is the minimum discharge required from each sprinkler to meet the capacity requirement
for the 1500 ft2 area?
c) If the type of sprinkler being used has a K-value of 4.0, what pressure must be supplied at the
sprinkler head to deliver the required flow?
9.6 Use the fixture unit method to estimate the peak design flow for a commercial office complex with
the following fixture totals:
32 urinals (flush valve)

60 water closets (flush valve)
50 sinks
shower rooms with eight shower
2
heads total
16 drinking fountains
2 dishwashers (3/4-in.)
4 kitchen sinks (3/4-in.)
4 hose connections (3/4-in.)
The complex has lawn irrigation, but it does not operate during peak demand times. The fire service
is through a separate line. Therefore, the fire and irrigation demands need not be included in the cal-
culation.
Determine the total number of fixture units and the design flow. If you would like a velocity of
5 ft/s in the service line during peak flow, roughly what size pipe would you use?
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
10
Operations
In the early days of water distribution computer modeling, simulations were primarily
used to solve design problems. Because models were fairly cumbersome to use, oper-
ators preferred measuring pressures and flows in the field rather than working with a
complicated computer program. Recent advances in software technology have made
models more powerful and easier to use. As a result, operations personnel have
accepted computer simulations as a tool to aid them in keeping the distribution system
running smoothly.
Using a model, the operator can simulate what is occurring at any location in the dis-
tribution system under a full range of possible conditions. Gathering such a large
amount of data in the field would be cost-prohibitive. With a model, the operator can
analyze situations that would be difficult, or even impossible, to set up in the physical
system (for example, taking a water treatment plant out of service for a day). A cali-
brated model enables the operator to leverage relatively few field observations into a
complete picture of what is occurring in the distribution system.
10.1 THE ROLE OF MODELS IN OPERATIONS

Models can be used to solve ongoing problems, analyze proposed operational
changes, and prepare for unusual events. By comparing model results with field oper-
ations, the operator can determine the causes of problems in the system and formulate
solutions that will work correctly the first time, instead of resorting to trial-and-error
changes in the actual system.
Physically measuring parameters such as flow in a pipe or HGL at a hydrant is some-
times difficult. If the operator needs to know, for instance, the flow at a location in the
system where no flow meter is present, the location must be excavated, the pipe
tapped, and a Pitot rod or other flow-measuring device installed (a substantial and
costly undertaking). Also, because operators must deal with the health and safety of
418 Operations Chapter 10
customers, they cannot simply experiment with the actual system to see what effects a
change in, say, pressure zone boundaries will have on them. With a model, however, it
is easy to analyze many types of operational changes and plan for unusual events.
Unlike design engineers working with proposed systems, operators can obtain field
data from the actual system, including pressures, tank water levels, and flows. Solving
operational problems involves the integrated use of these field data with simulation
results. Some of the most useful types of data that will be referred to frequently in this
chapter are hydrant flow tests, pressure chart recorders, and data collected through
SCADA and telemetry systems.
Assuming that the model is well-calibrated, contradictions between field data and val-
ues computed by the model can indicate system problems and provide clues about
how they may be solved. Problems that may be discovered by comparing field obser-
vations with model predictions include closed valves, water hammer, and pumps not
operating as expected. The models discussed in this book do not explicitly examine
short-term hydraulic transients (such as water hammer). However, when the other
causes of unusual pressure fluctuations are ruled out, water hammer can be diagnosed
through a process of elimination.
Models are also useful as training tools for operators. Just as pilots train on flight sim-
ulators, operators can train on a water distribution system simulation. It is much less
costly to have an operator make mistakes with the model than the real system. In addi-
tion, operators can determine how to handle situations such as catastrophic pipe fail-
ures or fires before they occur. Cesario (1995) described how models can allow
operators to attempt changes they might otherwise be reluctant to try.
Unusual situations that occur in a real system often give the modeler an opportunity to
further calibrate the model according to conditions that the operators purposely would
not want to duplicate because of their disruptive and unwanted influence on the sys-
tem. When these events occur, it is important to gather as much information as possi-
ble regarding pressures, system flows, consumer complaints, tank elevations, operator
statements, and so on and record it before it is lost or forgotten. In many instances,
experienced field crews and plant operators will log all the information while strug-
gling with the problem, and then file or discard it after the crisis is over.
The following sections describe how a water distribution model can be used to
address some operational problems. This chapter assumes that there is already an
existing calibrated model of the system, and covers the following topics:
• Solutions to common operating problems

• Preparation for special events
• Calculation of energy efficiency
• Flushing
• Metering
• Water quality investigations
• Impact of operations on water quality
Section 10.2 Low Pressure Problems 419
10.2 LOW PRESSURE PROBLEMS

The most frequently occurring operational problem associated with water distribution
systems is low or fluctuating pressures. Although confirming that the problem exists
is usually easy, discovering the cause and finding a good solution can be much more
difficult.
Identifying the Problem

Customer complaints, modeling studies, and field measurements obtained through
routine checks can indicate that a portion of the system is experiencing low pressure.
The pressure problem can be verified by connecting a pressure gage equipped with a
data logging device or chart recorder to a hydrant or hose bib to continuously record
pressure. Occasionally, a customer may report a low-pressure problem when the pres-
sure at the main is fine. In such cases, the low pressure may be due to a restriction in
the customer’s plumbing, or a point-of-use/point-of-entry device that is causing con-
siderable head loss.
If measurements indicate that pressure in the main is low and a problem in the distri-
bution system is suspected, the next step is to examine the temporal nature of the
problem. If the test readings show that the pressure is consistently low, then the cause
is probably that the elevation of the area is too high for the pressure zone serving it.
Pressure drops that occur only during periods of high demand are usually due to
insufficient pipe or pump capacity, or a closed valve. If the problem occurs at off-peak
times, nearby pumps may be shutting off once remote tanks have been filled, lowering
the pressure on the discharge side of the pumps.
Modeling Low Pressures

When data are available, the problem can be re-created and simulated in the model.
Steady-state runs with instantaneous pressure readings can be used, but more infor-
mation can be gained by attempting to reproduce a pressure-recording chart using an
extended-period simulation (EPS).
Though some idea of the cause of the low-pressure problem (elevation, inadequate
capacity, and so on) can be quickly gained from the chart recorder readings alone, the
model is needed to accurately identify the weak system component causing the prob-
lem. For instance, to locate pipes potentially acting as bottlenecks in the system, the
model can be checked for pipes with high velocities. The capacity of the high-velocity
pipe(s) can then be increased in the model by adding parallel pipes, changing diame-
ters, or adjusting roughness to see if the problem is then solved. If the problem is due
to insufficient pump capacity, then pump curve data for a new pump or impeller size
can be entered. If elevation is the culprit, then the pressure zone boundaries may need
to be adjusted, or booster pumping added.
If a pressure problem exists only in a remote part of the system during peak demands,
then adding a storage tank may be the solution. In urban situations, the need for stor-
age is typically driven by fire–fighting concerns, and in rural systems, it may be
driven by peak hour demands. The need for storage is discussed further in Section 8.3.
Finding Closed Valves

Very often, the model does not agree with the chart recorder, especially during high
water-use periods or fire flow tests. The model may indicate a much smaller pressure
drop than is observed. This discrepancy usually occurs when there is a closed or par-
tially closed valve in the actual system that causes the system to have a significantly
reduced capacity when compared to the model. The problem component can typically
be located by measuring the HGL throughout the actual system and looking for abrupt
changes that cannot be associated with pressure zone divides. If the HGL is signifi-
cantly lower than model predictions downstream of a specific point, it is likely that a
closed valve is located there.
As discussed previously, the HGL is equal to the sum of the elevation and the pressure
head. Because HGL accounts for both elevation and pressure, it is easier to pinpoint
closed valve locations by comparing HGLs than by comparing pressures only.
Though the accuracy of the pressure gages affects the computed HGL, incorrect ele-
vation is likely to contribute more significantly to error, because it is the more difficult
of the two parameters to measure. Topographic maps usually have too great of a con-
tour interval to provide the required precision and are not the most accurate source of
data. Elevations can be more accurately obtained using sewer system manhole eleva-
tions, altimeters, or global positioning systems (GPS) (Walski, 1998).
A good way to review the data is to compare plots of the measured and predicted
HGLs from the water source (or nearby tank) to the area with the pressure problems.
As previously noted, an abrupt drop in the measured HGL indicates that potentially
there is a closed valve.
Because head losses are smaller at lower flows, an abrupt drop in HGL may not occur
under normal demand conditions (under normal conditions, the HGL may slope only
one to two feet per thousand feet), making diagnosis of the problem difficult, as
shown in Figure 10.1. Compounding this difficulty is the inaccuracy of the HGL val-
ues themselves, which may be off by more than five feet unless the elevations of the
test locations were precisely determined through surveying or GPS. Therefore, the
errors in measurement may be greater than the precision of data necessary to draw
good conclusions.
Figure 10.1
Slope of HGL at
875 lower flows
870
High Measure
865
860 Measured Line
855
Model
850
HGL, ft
845 Low Measure
840
835
830
825
820
0 2,000 4,000 6,000 8,000 10,000
Distance from Tank, ft
To overcome these inaccuracies, flow velocities in the pipes must be increased to pro-
duce a sufficiently large head loss. By opening a hydrant, or, if necessary, a blow-off
valve, the head loss is increased so that the slope of the HGL is significantly greater
than the measurement error, as shown in Figure 10.2. Even with the “noisy” data
shown in the figure, the model and field data clearly diverge approximately 3,200 ft
(975 m) from the tank. For the rest of the range, the slope of the HGL is the same for
both model and measured data. Therefore, either some type of restriction exists at
around 3,200 ft (975 m), or the model has an error in that area.
Figure 10.2
Slope of HGL at
higher flows 870
High Measure
860
850
Model
840
Low Measure
830
820
HGL, ft
Measured Line
810
800
790
780
770
760
0 2,000 4,000 6,000 8,000 10,000
Distance from Tank, ft
Because it is usually difficult to keep a hydrant or blow-off valve open for a long
period of time while pressure gages are moved from location to location, it may be
best to run several fire hydrant flow tests with the pressures at multiple residual
hydrants being tested simultaneously. In this way, three tests can yield a significant
number of data points. The same hydrant should be flowed at the same rate while
pressures are taken at several residual hydrants.
Solving Low Pressure Problems

After the cause of the pressure problem has been identified and confirmed, the possi-
ble solutions are usually fairly straightforward, and include the following:
• Making operational changes such as opening valves
• Changing PRV or pump control settings
• Locating and repairing any leaks
• Adjusting pressure zone boundaries
• Implementing capital improvement projects such as constructing new mains
• Cleaning and lining pipes

• Installing pumps to set up a new pressure zone
• Installing a new tank
Some pressure problems are difficult for the utility to resolve. For example, an indus-
trial customer may demand a very high pressure, or a resident may experience low
pressure due to the customer’s own plumbing. When the utility cannot justifiably
spend large sums of money to make changes to the system to meet customer expecta-
tions or improve customer plumbing, the problem becomes one of customer relations.
Using the model, the utility can determine the cause of the problem, study ways of
increasing pressure, better decide if the costs of system changes exceed the benefits,
or discover that the problem is not in the utility’s system. For example, a commercial
customer who has installed a fire sprinkler system requiring 60 psi (414 kPa) to oper-
ate will be upset if the utility makes operational changes that cause the pressure to
drop to 45 psi (310 kPa), even though this pressure meets normal standards. In
another case, the model and field data may show 60 psi (414 kPa) in the utility sys-
tem, but the customer has only 15 psi (103 kPa). The problem may be caused by an
undersized backflow preventer valve, meter, or a point-of-entry treatment unit with
excessive head loss.
In some cases, customers near a pump station can become accustomed to high pres-
sures, as shown in Figure 10.3. When a pump cycles off, those customers are fed
water from a tank that may be some distance away. If the pump should cycle off dur-
ing a high demand time, the pressure can drop significantly. The model can confirm
this drop, and whether or not pump cycling is the cause. If so, it may be necessary to
always run some pumps during periods of high demand, even if the tank is full.
Figure 10.3
Tank Water Level Effect of pump
operation on
HGL Pump On customers near the
{
pump
Pressure
Drop at
Pump
Switch
Tank
HGL Pump Off
Pump
Problems with the model may also surface when trying to reproduce low pressure
problems that exist in the actual system. For instance, the model and field data may
agree when pumps are off, but not when they are on, indicating that the pump perfor-
mance curves may be incorrect in the model (for example, a pump impeller may have
been changed and not updated in the model). Another possible discrepancy between
the model and field data could be the control setting on the PRV. Settings can drift by
themselves over time, or may be changed in the field but not in the model. To be use-
ful, a model must reflect operating conditions current at the time the field data were
collected.
Leak Detection. Unless accurate demands are known, this approach is only
effective if the leakage is very large compared to demands. Unless demands are
known very accurately, however, this approach is really only effective if the leakage is
very large compared to demands [say, a 100 gpm (0.006 m3/s) leak in a pressure zone
with 200 gpm (0.013 m3/s) usage]. In such cases, these large leaks show up as surface
water, making locating them with a model unnecessary.
As an exception to this scenario, modeling can be used to help locate leaks when very
high nighttime demands are required to accurately reflect diurnal water usage in part
of the system. Unless there are large nighttime water users, such as industries that
operate at night, water use will typically drop to about 40 percent of average day con-
sumption. Because leakage is not reduced at night, if very high nighttime usage is
required for the model to match historical records, leakage can be suspected in that
part of the system.
In some instances, the cause of low pressure is a large, nonsurfacing water loss result-
ing from a pipe failure. In this case, the water may be lost through a large-diameter
sewer, a stream, or a low area that is not easily observed. An indicator of this type of
leak could be an increase in demand on a pump or an unusual drop in an elevated tank
level if it is a very large loss or occurs near one of these facilities. If the leak is on one
of the smaller grid mains, its effect would be felt over a smaller area and may not be
noticed as an increase in demand. These leaks are generally found by checking sewer
manholes for exceptional flows and/or listening on hydrants and valves.
10.3 LOW FIRE FLOW PROBLEMS

Low fire flows are another common operational problem. Solving this problem in an
existing system is different from designing pipes for new construction, in that the util-
ity cannot pass the cost of improvements onto a new customer. Rather, the operator
must find the weak link in the system and correct it.
Identifying the Problem

The possible reasons for poor fire flows in an existing system are
• Small mains
• Long-term loss of carrying capacity due to tuberculation or scaling
• Customers located far from the source
• Inadequate pumps
Section 10.3 Low Fire Flow Problems 425
• Closed or partly closed valves (as discussed in Section 10.2)

• Some combination of the preceding
Fire flow tests can reveal the magnitude of the problem, but the model will help to
determine and quantify the cause and possible solutions. Fire flow tests should first be
used to more precisely calibrate the model in the area of interest. If the predicted pres-
sures are higher than observed pressures during fire flow tests, the problem is usually
that there are closed valves (or occasionally pressure reducing valves failing to oper-
ate properly). Plotting the actual and modeled hydraulic grade line during high flow
events can help locate and determine the reason for the low fire flows.
If the model can be calibrated and no closed valves are found, then the cause of the
poor flow is either pipe capacity or distance between the problem area and the water
source. The model provides the operator with a tool for examining velocity in each
pipe. If velocities are greater than 8 ft/s (2.4 m/s) for long pipe runs, then the issue is
small piping.
If the model requires a friction factor corresponding to extremely rough pipe (for
instance, a Hazen-Williams C-factor of less than 60), and this value is verified by
visual inspection of internal pipe roughness or through testing, then loss of carrying
capacity due to tuberculation is to blame.
The main size and roughness affect the slope of the HGL, but the length affects the
magnitude of the pressure drop. For instance, it is possible to get a much larger flow
through 100 ft (30.5 m) of an old 6-in. (150 mm) main than through 10,000 ft (3,050
m) of the same pipe without significantly affecting pressure.
In models that have been skeletonized, it may be necessary to add pipes that have
been removed back into the model to get an accurate picture of fire flow.
Solutions to Low Fire Flow

The best solutions to low fire flows depend on the problems as identified by the model
and field data collection. In general, the solutions consist of some combination of
• New piping
• Rehabilitation (cleaning and lining, sliplining, or pipe bursting)
• Booster pumping
• Additional storage near the problem area
Each of these options affects the system in different ways and has different benefits,
so the comparison of alternatives should be performed based on a benefit/cost analy-
sis as opposed to simply minimizing costs. The modeling for this evaluation can usu-
ally be performed with steady-state runs. Only if the volume of storage or the ability
of the system to refill storage is in question should an EPS model be used.
New Piping and Rehabilitation. Sizing new pipes (adding capacity) and
rehabilitating existing pipes flattens out the slope of the hydraulic gradient for a given
flow rate, as shown in Figure 10.4. By allowing the modeler to examine the HGL, the
model can help locate which individual pipes are bottlenecks in need of repair or
rehabilitation.
Figure 10.4
Effect of pipe Source HGL
improvements on
HGL New or Rehabilitated Pipe
HGL
Small or Rough Pipe Minimum

Required
HGL
Booster Pumping. Booster pumping is usually the least costly method of cor-
recting low pressure problems from an initial capital cost standpoint. Booster pumps
can significantly increase operation and maintenance costs, however, and do not allow
as much flexibility in terms of future expansion as the other available options. Booster
pumping increases the HGL at the pump location but does not reduce the hydraulic
gradient, as shown in Figure 10.5. Therefore, booster pumps should, in general, be
used only to transport water up hills, not to make up for pipes that are too small. Fur-
thermore, booster pumps can over-pressurize portions of the system and even cause
water hammer, especially when there is no downstream storage or pressure relief.
Figure 10.5
Increasing HGL using Booster
booster pumping Pump
Source HGL HGL With Pump
HGL
Minimum
Required
HGL With HGL
Inadequate Piping
Adding Storage. Adding storage at the fringe of the system tends to be a costly
alternative, but it provides the highest level of benefit. Storage greatly increases fire
flows and pressures, because water can reach the fire from two different directions
(both original and new sources). This splitting of flow significantly reduces velocities
in the mains. Since head loss is roughly proportional to the square of the velocity, cut-
ting the velocity in half (for example) reduces the head loss to one-quarter of its prior
value, as shown in Figure 10.6. Storage also increases the reliability of the system in
the event of a pipe break or power outage, and helps to dampen transients.
Section 10.4 Adjusting Pressure Zone Boundaries 427
Figure 10.6
//=//= Effect of additional
storage location on
Source HGL
HGL
Storage
HGL with Storage
Inadequate
Piping
Minimum
Required
HGL
//=//=//=//=
10.4 ADJUSTING PRESSURE ZONE BOUNDARIES

In spite of efforts to properly lay out water systems as discussed on page 333, utilities
occasionally find themselves with pockets of very high or low pressures. Low pres-
sure problems are usually identified and corrected quickly, as described in Section
10.2 on page 419, because of customer complaints. High pressure problems, on the
other hand, can persist because most customers do not realize that they are receiving
excessive pressures.
When it is determined that pressures are too high in an area, it is best to move the
pressure zone boundary so that the customers receive more reasonable pressures from
a lower zone. However, in some cases, this may not be a practical solution. For exam-
ple, in Figure 10.7, the pocket of customers on the right may be too far from the low
pressure zone on the left to economically be connected, so they must be served
through a PRV from the higher zone. Reducing pressure should reduce leakage and
improve the service life of plumbing fixtures. It is important that the performance of
the PRV be modeled before installation to determine the impacts on existing custom-
ers and fire flows.
The utility may want to adjust the boundaries of the pressure zones to reduce the
energy costs associated with pumping and avoid over-pressurizing the system. The
best way to start this work is to decide on the elevation contour that should be the
boundary between pressure zones, and close valves in the model along that boundary.
For example, in Figure 10.8, the utility has selected the 1,200 ft (366 m) elevation
contour as the boundary between pressure zones having HGLs of 1,300 ft and 1,410 ft
(396 m to 430 m). The model can aid in pointing out problems that will result from
closing valves in the system, and it can help the operator identify solutions. In this
case, some of the customers between valves B and D should receive water from the
higher pressure zone. However, the 12-in. (300-mm) pipe through the middle of the
figure may be an important transmission main for the lower pressure zone. If valves B
and D are closed, the pipe can no longer serve this purpose, so alternative solutions
must be explored.
Figure 10.7
Serving high pressure
pocket areas
Modified
Pressure Zone
Boundary
High
Pressure Zone
Reduced
Pressure Zone
Pump to PRV
Higher Zone
Low Pressure
Zone
Feed to
Lower Zone
Previous
Pressure Zone
Boundary
Figure 10.8
A single pressure
zone
ft
10 ft
1,2 20
0
1,
6"
A F E
ft
190
1,
H
12"
B D
G
C
6"
6" 6" 6"

Section 10.4 Adjusting Pressure Zone Boundaries 429
If the elevation of the area near valve G is only slightly higher than 1,200 ft (366 m),
then it may be possible to simply move the pressure zone divide back to valve H, leav-
ing valves B, C, and D open, as in Figure 10.9. If, however, this solution causes cus-
tomers near G to receive pressures that are too low, then a crossover must be
constructed between G and H (Figure 10.10). In addition, a small service line parallel-
ing the 12-in. (300 mm) main and extending to the limits of the 1,410 ft (430 m) pres-
sure zone will be necessary to serve the higher-elevation customers. Note that even
though the pressure is lower, any hydrants near G should remain connected to the 12
in. (300 mm) line because of its greater capacity.
Figure 10.9
Relocating a pressure
ft zone boundary
10 ft
1,2 200
1,
6"
A F E
ft
190
1,
H
12"
B D
G
C
6"
6" 6" 6"
Low Pressure Zone Pipes

High Pressure Zone Pipes
This example illustrates just how complex adjustments to pressure zone boundaries
can become. The model is an excellent way to test alternative valving and determine
the effects of the adjustments. If the model was originally skeletonized, it will be nec-
essary to fill in the entire grid in the area being studied to obtain sufficient detail for
the analysis. All pipes and closed valve locations along the pressure zone boundary
are important. In Figure 10.10, for example, the node at the intersection near valves E
and F is in the higher zone, and the valves themselves represent the ending nodes for
two pipes in the lower zone.
After the steady-state runs have demonstrated that pressures are in a desirable range
for a normal day, the results of hydrant flow simulations near the boundary are com-
pared with actual hydrant test data. This comparison will identify any fire flow capac-
ity problems resulting from a potential boundary change that may, for instance,
reduce the number of feeds to an area. In some cases, it will be necessary to add pipes
or to close loops. Also, PRVs and/or check valves can be installed between the zones
to provide additional feeds to the lower zone, improving reliability.
Figure 10.10
Isolating a pressure ft
zone and installing a 10 ft
1,2 200
crossover pipe 1,
6"
A F E
ft
190
1,
H
Install
Crossover
12"
B D
G
Parallel
Pipes
C
6"
6" 6" 6"
Low Pressure Zone Pipes

High Pressure Zone Pipes
Adjusting pressure zone boundaries may result in some long dead-end lines with little
flow. These dead ends should be avoided because of the potential for water quality
problems. Blow-offs (bleeds) may need to be installed at the ends of such lines.
10.5 TAKING A TANK OFF-LINE

Occasionally, water distribution storage tanks must be taken off-line for inspection,
cleaning, repair, and repainting. Even a simple inspection can cause a tank to be out of
service for several days while the time-consuming tasks of draining, removing sedi-
ment, inspecting, disinfecting, and filling are completed.
Because tanks are so important to the operation of the system, taking one out of ser-
vice markedly affects distribution system performance (see page 435 for information
on the impact on pressures). The reduction in system capacity can be dramatic; con-
versely, the system can over pressurize during off-peak periods when demands fall
well below pumping capacity.
Section 10.5 Taking a Tank Off-Line 431
Fire Flows
While tanks are important for flow equalization, their main purpose is to contribute
capacity for peak hour demand and fire flows. Taking a tank out of service removes a
major source of water for emergencies. Fire flows at several locations in the system
should be analyzed using the model to determine what effect taking the tank out of
service will have, and how much flow can be delivered from other sources (for exam-
ple, the plant clearwell, pumps, or through PRVs).
If the loss of water for emergencies is significant, then the utility may want to do one
or more of the following:
• Install temporary emergency pumps
• Prepare to activate an interconnection with a neighboring utility, if necessary
• Install a pair of hydrants at a pressure zone divide so that a temporary fire
pump can be connected in an emergency
These emergency connection alternatives can be simulated to determine the amount
of additional flow provided. Sometimes, adding an emergency connection or pump
may only provide a marginal increase in flow because of bottlenecks elsewhere in the
system.
The utility can use the results of the simulations to better present the impacts of
removing a tank from service on fire departments and major customers. In this way,
fire departments can prepare for the tank to be out of service, and make appropriate
arrangements to supply the water that may be needed if there is a fire in the affected
area. For instance, the fire department may be prepared to use trucks to carry water in
an emergency rather than rely on the distribution system. Alternatively, they can
research the feasibility of laying hose to hydrants in a neighboring pressure zone that
has adequate storage.
Low Demand Problems

While problems meeting fire demands are most obvious when taking a tank off-line,
problems can crop up even during times of normal or low water usage. When a pres-
sure zone is fed by a pumping station with constant-speed pumps and no other stor-
age, the pumps will move along their pump curves to match demand. In off-peak
times, the pumps must deliver very low flow compared to design flow (for example,
40 percent of average), and this flow will therefore be delivered at a higher head.
Depending on the shape of the pump curve, very high system pressures can result.
The worst problems occur with deep well pumps designed to pump against very high
heads (that is, their pump curves are very steep), since a slight change in demand can
result in dramatic pressure changes.
An example of this situation is shown in Figure 10.11 and the corresponding data in
Table 10.1. The figure shows pressures at a representative node in the system (for
example, the pump discharge) and how pressures at other points vary with elevation.
The comparatively low nighttime demands indicate that the system is probably expe-
riencing very little head loss during this time, compounding the effect that the
increased pump discharge head has on the system pressures. If the piping of an area
normally operates at a pressure of 85 psi (586 kPa), a 15 psi (103 kPa) increase can
cause marginal piping to break [pipes that could not withstand 85 psi (586 kPa) would
have broken previously]. If a tank is off-line, the utility can ill afford a major pipe
break.
Figure 10.11
Effect of a change in
demand on a constant-
speed pump in a
closed system
Table 10.1 Sample discharges and corresponding pressures for pump shown in Figure 10.11
Pressure Flow
Point Demand
(psi) (gpm)
A Peak demand 80 520
B Normal operating point with tank 100 420
C Average demand 110 350
D Nighttime demand 125 180
An EPS model can be used to simulate the pressures and flows that occur over the
course of a day to determine if the pressures may be too high during off-peak demand
hours, or too low during peak demand hours. If the pressures become too high, then a
PRV can be installed on the pump’s discharge piping, or a pressure relief valve can be
installed to release water back to the suction side of the pump. The PRV can be mod-
eled as a pressure sustaining valve set to open when the pressure exceeds a given set-
Section 10.6 Shutting Down a Section of the System 433
ting. If the pressures are too low during peak times, then additional pumping or a
modification of pump controls may be necessary.
10.6 SHUTTING DOWN A SECTION OF THE SYSTEM

Pipes are occasionally taken out of service to install a connection with a new pipe,
repair a pipe break, or perform rehabilitation on a pipe section. Without a model, the
engineer must either make an intelligent guess about the effect on system perfor-
mance or perform a trial shutdown to see what happens. Simulations are an excellent
alternative or supplement to these options.
Representing a Shutdown
The correct location of valves in the model is critical to determining the impact of a
shutdown on distribution system performance. Figure 10.12 shows pipe P-140 con-
necting nodes J-37 and J-38. In shutdown scenario A, pipe P-140 is removed from
service, along with nodes J-37 and J-38 and all pipes connected to these junctions. In
shutdown scenario B, with valves G and H in service, only P-140 is taken out of ser-
vice. Models are not usually set up to include small sections of pipe such as the one
between valve G and node J-37, so the operator needs to carefully analyze how to
modify the nodes and pipes to simulate the shutdown correctly. Most hydraulic simu-
lation packages allow you to simulate pipe shutdowns as a function of the pipe instead
of including all of the valves in the model.
Figure 10.12
Simulating a pipe
Shutdown A
shutdown
B Closed Valve
D
J-37 J-38
A P-140 E
C F
Shutdown B
Closed Valve
J-37 J-38
G P-140 H
When shutting down a large transmission main with many taps, the problem can be
much more difficult. In this case, all smaller pipes that run parallel to the main must
be included in the model, even if they have not been included under the current level
of skeletonization (they may not have been considered important when the large main
was in service). In Figure 10.13, the 6-in. (150-mm) pipe represented by the dashed
line may have originally been excluded from the model. During a simulation of the
shutdown of the 16-in. (400-mm) pipe, however, the smaller pipe becomes very
important and must be included in the model.
Figure 10.13
Shutdown of a
transmission main
16”
Closed Valve
12"
6"
Closed
Main Shutdown Valves Add to Model
of Shutdown
8"
16” 8"
Valve
Closed Valve
12"
16”
Simulating the Shutdown
After the model is configured correctly, it can be used to simulate the shutdown. Prior
to performing any EPS runs, steady-state simulations are used to determine if any cus-
tomers will be immediately without water. This problem may show up in the model
output as “disconnected node” warnings, or as nodes with negative pressures. Note
that negative pressures do not actually exist in a water distribution system, rather they
typically indicate that the specified demand cannot be met.
Section 10.7 Power Outages 435
After the steady-state runs are successfully completed and it is clear which customers,
if any, will be out of water, the operator will move on to EPS runs. An EPS run of the
shutdown will show whether or not the affected portion of the system is being served
with water from storage. A system that is relying on its storage can have tank water
levels that drop quickly. In such cases, EPS runs are then used to study the range of
tank water levels and their effect on system pressures.
Usually, in short-term shutdowns, the system can be supplied by storage. The model
can determine how long it will take before storage is exhausted. If a long-term shut-
down is required, the utility can identify the feasibility of alternative sources of water
for the area using the EPS results.
This additional water supply may be provided by cracking open valves along pressure
zone boundaries to obtain water from adjacent zones, opening interconnections with
neighboring utilities, using portable pumps, or laying temporary pipes or hoses to
transport water around the shutdown area. When a project is planned in which pipes
will be out of service for cleaning and lining, temporary bypass piping laid on the
ground (called highlining) can be used to transport water and supply customers in the
area being taken out of service.
When modeling a tank shutdown, the effect that pressure has on demand needs to be
considered. In actuality, water demands are a function of system pressure. When pres-
sure drops below normal, less water is used and leakage decreases. When evaluating
demands as a function of pressure drop, the model must be adjusted for this change in
demand or the simulation results will be conservative. To quantitatively determine the
amount to compensate for a change in demand, a model with pressure-dependent
demands can be used (some hydraulic models allow you to specify demand as a func-
tion of pressure). In practice, however, demand is decreased by a factor related to the
change in pressure, usually involving considerable judgment.
10.7 POWER OUTAGES

A power outage may affect only a single pump station, or it can impact the entire sys-
tem. Typically, utilities attempt to tie important facilities to the power grid with redun-
dant feeds in several directions. No electrical system is perfectly reliable, however,
and power may be interrupted due to inclement weather or extreme power demands.
Most utilities rely on some combination of elevated storage, generators, or engine-
driven pumps to protect against power outages. Generators may be permanently
housed in pump stations, or temporarily stored elsewhere and transported to the area
when they are needed. The generators located in pump stations may be configured to
start automatically whenever there is a power outage, or they may require manual
starting.
Modeling Power Outages

A power outage can be modeled by turning off all the pumps that do not have a gener-
ator or engine. This action will probably result in nodes that are hydraulically discon-
nected from any tank or reservoir, or in negative pressures. A model that is structured
this way will generate error messages and may not compute successfully. One possi-
ble way to work around this problem is to identify pressure zones that are discon-
nected and feed them from an imaginary reservoir (or fixed grade node) through a
check valve at such a low head that all the pressures in the zone are negative. The neg-
ative pressures can then be used to indicate customers that will be without water.
Steady-state model runs of the power outage are performed before EPS runs. The
steady-state runs identify the nodes that will immediately be without service. After
those problems have been addressed, EPS runs are used to study the effects of storage
on service during a power outage. Zones with storage will first experience a drop in
pressure as water levels fall, and then a loss of service once the tanks are empty. As in
the case for systems with no tanks, each zone should have a reservoir at low head con-
nected to the system through a very small pipe, so that nodes will not become discon-
nected in the model. Customers at lower elevations within a pressure zone may
experience very little deterioration in service until the tank actually runs dry.
Demands will most likely decrease during a power outage. A factory that does not
have power may have to shut down, and so will not use much water. Dishwashers and
washing machines also do not operate during power outages. The modeler needs to
estimate and account for this effect to the extent possible.
In areas where the utility uses portable generators to respond to power outages, the
system will operate solely on storage for the amount of time that it takes to get the
generator moved, set up, and running. Time-based controls can be set up in the model
to turn on the pump after the time it would take to put the generator in place (probably
a few hours after the start of the simulation).
Duration of an Outage
Estimating the duration of a power outage is one of the most difficult decisions in this
type of analysis. Generally, the simulation should be based on the longest reasonable
estimate of the outage duration (that is, model the worst-case situation). If the outage
is shorter, the system performance will be that much better.
The EPS should be run for a duration that would give the tanks enough time to
recover their normal water levels after the outage. Therefore, the EPS does not simply
end at the time that the power is restored. Full tank level recovery may take hours or
days. Although the system may have performed well up to this point, problems some-
times arise in the recovery period. For instance, because tank levels are lower, there is
less head to pump against, so the flow rate of the pumps may be too high, causing the
motor to overload and trip. Also, after lengthy system-wide outages, recovery may be
limited by source capacity.
10.8 POWER CONSUMPTION

One of the largest operating costs for water utilities is the cost of energy to run pumps.
Unfortunately, many utilities do not realize that an investment in a few small pump
Section 10.8 Power Consumption 437
modifications or operational changes is quickly recovered through significant energy

savings. Many pump stations provide an excellent opportunity for significant savings
with minimal effort. Because so much energy is required for pumping, a savings of
only one or two percent can add up to several thousand dollars over the course of a
year. Some stations operate as much as 20 to 30 percent under optimal efficiency.
Common operational problems that contribute to high energy usage are
• Pumps that are no longer pumping against the head for which they were
designed
• Pumps which were selected based upon a certain cycle time and are being
run continuously
• Variable-speed pumps being run at speeds that correspond to inefficient
operating points
In modeling pump operation, a highly skeletonized model can be used because only
the large mains between the pump station and tanks are important in energy calcula-
tions. Adding detail usually has little impact on the results for this type of application.
In addition to the information in the following sections on using models for determin-
ing energy costs, publications with guidelines for minimizing energy costs are avail-
able (Arora and LeChevallier, 1998; Hovstadius, 2001; Reardon, 1994; Walski, 1993).
Many researchers have attempted to apply optimization techniques to energy manage-

ment, with some success. Energy management continues to be a very active research
area (Brion and Mays, 1991; Chase and Ormsbee, 1989; Coulbeck and Sterling, 1978;
Coulbeck, Bryds, Orr, and Rance, 1988; Goldman, Sakarya, Ormsbee, Uber, and
Mays, 2000; Lansey and Zhong, 1990; Ormsbee and Lingireddy, 1995; Ormsbee,
Walski, Chase, and Sharp, 1989; Tarquin and Dowdy, 1989; and Zessler and Shamir,
1989).
Determining Pump Operating Points

Many pumps are selected based on what the operating points ought to be but are run
at operating points that can vary greatly over the course of a day. The simplest kind of
analysis for a pump is to calculate pump production versus time of day and compare it
with the flow rate at the pump’s best efficiency point. Figure 10.14 shows pump dis-
charge versus time for a two-day period for a pump discharging into a pressure zone
with no storage. If the pump’s best efficiency point is approximately 400 gpm (0.025
m3/s), then the pump is running efficiently. If, however, the pump is a 600 gpm (0.038
m3/s) pump, the pump is not being run efficiently and is wasting energy.
Figure 10.14
Pump discharge 600
versus time in closed 600 gpm pump
system
500
Flow, gpm
400 400 gpm pump
300
200
100
0
0 6 12 18 24 30 36 42 48
Time, hr
Variable-speed pumps do not run efficiently over a wide range of flows, as shown in
Figure 10.15. For instance, a variable-speed pump discharging against 150 ft (46 m)
of head may run efficiently at 500 gpm (0.032 m3/s), but not at 250 gpm (0.016 m3/s).
This type of analysis will help the operator determine whether the pump is operating
efficiently. Operators will also want to know exactly how much money they are
spending on a given pump versus a more efficient pump. The following sections
explain the options to correct inefficiencies in pump operation and the calculations
necessary to compute the associated costs.
Figure 10.15
250 Method for
determining pump
ncy
operating points
Efficie
ncy
y
ienc
40%
icie
200 100% Speed
ffic
Eff
E
%
95%
55%
70
90%
150 85%
Head, ft
80%
100 System
Head
Curve
50
0
0 100 200 300 400 500 600
Flow, gpm
Calculating Energy Costs

The cost of pumping depends on the flow, pump head, efficiency, price, and the dura-
tion of time that the pump is running. The cost for pumping energy over a given time
period can be determined using the following equation:
C=Cf QhP pt/(epemed ) (10.1)
where C = cost over time duration t ($)

Q = flow rate (gpm, l/s)
hP = total dynamic head (TDH) of pump (ft, m)
p = price of energy (cents/kW-hr)
t = duration that pump is operating at this operating point (hrs)
ep = pump efficiency (%)
em = motor efficiency (%)
ed = variable-speed drive efficiency (%)
To determine the daily and annual energy costs for continuously running pumps, use 24
hours and 8,760 hours, respectively for t. The product of pump, motor, and variable-
speed drive (if applicable) efficiency is generally referred to as the wire-to-water effi-
ciency or overall efficiency, because it accounts for the efficiency of the motor and the
pump combined, and is commonly measured directly in the field (see page 198).
Because most stations have a single pump running at a time, and if there is storage
downstream it operates at roughly the same operating point over the course of the day,
this equation can be used to estimate the energy consumption over a given time
period. For example, to calculate the cost of running a pump for a day, t would repre-
sent the number of hours the pump is running per day, and C would be the daily
energy cost. Determining the value of C can be complicated by the fact that some util-
ities pay different rates for power at different times of the day. If t is the number of
hours that the pump is running per year, then C will be the annual energy cost.
Example – Energy Costs. Consider a pump that discharges 600 gpm at a TDH of 230 ft (as
determined by the model) and runs 12 hours per day. From the pump efficiency curve, we know that
the pump efficiency is 62 percent. The motor is 90 percent efficient, and the power costs 8 cents per
kW-hr. The energy cost equation (Equation 10.1) is applied as follows.
C (in $/day) = (0.0189 x 600 gpm x 230 ft x 8 cents/kW-hr x 12 hours/day) / (62% x 90%)
C = $44.87/day
If the utility were to change to a 400 gpm pump that ran for 18 hours per day at a pump efficiency of
70 percent, and replace the motor with a premium efficiency motor having an efficiency of 95 percent,
then the cost for energy would be $37.65/day. The annual savings would be $2,635, which may pay
for the change in pumping equipment.
In the case of variable-speed pumps, the user needs to account for the efficiency of the
variable-speed drive. This efficiency varies with the pump speed, drive manufacturer,
and load, among other factors. Modelers are encouraged to obtain the drive efficiency
data from the manufacturer for the specific drive they are using.
A wide variety of variable-speed drives have been used over the years to adjust the
speed of a pump. They include variable frequency drives (VFDs) (also called adjust-
able frequency drives), eddy current couplings, hydraulic couplings, and wound rotor
slip. Most of these involve allowing some slip between the motor output and pump
input, but the VFDs actually change the electrical input to the motor to make it turn at
a different speed. Today most variable-speed pumps use VFDs.
Some typical data for variable-speed drive efficiency is given in Table 10.2; however,
it is important to note that the efficiency of newer variable speed drives is significantly
better than older drives. The values for the VFD shown in the table were obtained
from several different manufacturers and the eddy current coupling and hydraulic
coupling data are from TREEO (1985).
Multiple Distinct Operating Points

Some pumps do not stay on a single operating point when running, but will operate at
several different operating points depending on tank water levels, demands, control
valve settings, and the status of other pumps. Equation 10.1 would then be solved for
each operating point by using the model to determine the Q and h for that operating
point and the duration of time the pump will be running at that operating point.
Table 10.2 Sample variable-speed drive efficiency data

Variable-Speed Drive Efficiency
(%)
% of Full Speed VFD Eddy Current Coupling Hydraulic Coupling
100 97-95 85 83
90 95-92 78 75
70 93-88 59 56
50 92-81 43 33
As the flow changes over the course of a day, the head and efficiency of the pump will
change, affecting the amount of energy used. Therefore, each operating point and the
duration the pump operates at that point must be considered separately when comput-
ing energy costs. The costs associated with each duration composing the period being
studied are then totaled to arrive at the total cost for the time period.
For the example shown in Table 10.3, the total of all the individual durations during a
day, including the time that the pump is not in operation, is 24 hours. For each operat-
ing point, the model will determine Q and h, and the pump efficiency curve will give a
value for pump efficiency. The energy cost for the duration of each operating point is
then determined. Costs are then summed to give daily or annual costs, as shown in
Table 10.3, for a pump that can be approximated with three operating points.
Table 10.3 Example calculation of energy costs with different operating points
Given Information Calculated Data
Energy Price: $0.08 kW-hr Motor efficiency: 92% (using Equation 10.1)
Q h ep t C C
(gpm) (ft) (%) (hrs/day) ($/day) ($/yr)
200 150 65 4 3.03 n/a
220 135 68 14 10.05 n/a
250 125 63 6 4.89 n/a
TOTAL 24 17.97 6,561
Continuously Varying Pump Flow

For constant-speed pumps discharging into a dead-end system, the flow will vary con-
tinuously. These pumps do not have a single pump operating point or even a handful
of them. In determining energy costs associated with these pumps, each hour (or other
time step) corresponds to a separate operating point. For example, if the table above
were made for a pump with a continuously varying pump flow, it would have 24 oper-
ating points instead of three. A spreadsheet or a model that automatically computes
energy costs then becomes the method of choice for the calculation. The time scale
over which the flow varies would determine the time step size. For example, if there
were large fluctuations within an hour, a smaller time step would be required.
The difficulty in using a spreadsheet is that the efficiency must be read from the pump
curve for each flow in the spreadsheet. Ideally, a function could be developed relating
flow to efficiency, as described in the next section.
Developing a Curve Relating Flow to Efficiency

A curve representing the relationship between pump discharge (Q) and pump effi-
ciency (ep) can usually be described by the equation for an inverted parabola, as
shown in Equation 10.2.
2
ep = ao + a1 Q + a2 Q (10.2)
The key, therefore, is to determine the values for the coefficients ao, a1, and a2. The
easiest way is to enter at least three pairs of Q and e values into a polynomial regres-
sion program (or spreadsheet), and have the program determine ao, a1, and a2. This
functionality is also available in some hydraulic models.
A quick approximation to the regression coefficients can be obtained knowing only
the best efficiency point, the fact that the derivative is zero at that point, and the fact
that the efficiency curve should pass through the point (0,0). Based on this knowl-
edge, the coefficients can be approximated as
ao = 0 (10.3)
2e
a 1 = -------o-
Qo
–e
a 2 = -------o-
2
Qo
where eo = efficiency at best efficiency point (%)

Qo = flow at best efficiency point (gpm, l/s)
While the coefficients above are less accurate than a parabola determined from regres-
sion using multiple points, it may be adequate for some calculations.
For example, given the efficiency versus flow data in the following table, a polyno-
mial regression program can be used to produce the equation relating flow and effi-
ciency as shown in Equation 10.4 and illustrated in Figure 10.16.
Q Efficiency
(gpm) (%)
200 55
400 85
600 60
Figure 10.16
100 Flow versus efficiency
relationship
80
Efficiency, %
60
40
20
0
0 100 200 300 400 500 600 700
Flow, gpm
ep = -0.00069 Q2 + 0.5625 Q - 30 (10.4)

Using only the best efficiency point data, the coefficients would have been -0.00053,
0.42, and 0. For better accuracy, higher degree polynomials can be used. In general,
they can be written as:
N
i
ep = ¦ ai Q (10.5)
i=0
where N = degree of the polynomial used
Variable-Speed Pumps
The procedure for computing the costs of variable-speed pumping is similar to the
procedure for continuously-varying pumps, except that the relationship is not simply
between flow and efficiency but also a function of total dynamic head. The equation
can be approximated by
2
ep = a2 Q e n + a1 Q e n + ao (10.6)
where n = ratio of pump speed/pump test speed

For higher degree polynomials, a more general form of the equation is
N
i
ep = ¦ ai Q e n (10.7)
i=0
Extending Efficiency Curves

Although centrifugal pumps can operate at rates Pump efficiencies can be calculated from this data
that vary from no flow to flows well beyond the by using the following equation:
best efficiency point, most pump manufacturers
100C f Qh P
show the pump efficiency curve only over a nar- e = -------------------------
-
row range of flow rates. However, in performing BHP
energy calculations, the modeler may need to
determine the efficiency of a pump at a very low where e = efficiency (%)
or very high flow rate. Q = flow rate (gpm, m3/hr)
Plotting a curve of efficiency versus flow outside hP = head added at pumps (ft, m)
the range of the given efficiencies leaves a great BHP = brake horsepower (hp, kW)
deal of room for judgment and therefore error. Fit- Cf = unit conversion factor (2.53 x 10-4 English,
ting the points on the efficiency curve to a parab- 2.72 x 10-3 SI)
ola and extrapolating that parabola provides a A plot of this equation should exactly match the
logical approximation of the data, but it is not nec- manufacturer’s efficiency data in the range pro-
essarily correct. vided, and it provides a means of computing effi-
ciencies for points beyond the limits of this curve
Although manufacturers provide only a narrow
range of efficiency data, they do provide head and when necessary.
power data over a much wider range of flows, usu-
ally all the way down to no flow.
A model can be used to determine a table of Q and h versus time of day. It is then pos-
sible to look up the efficiency for each time increment and use a spreadsheet program
to sum up the costs for the entire period.
The results of these calculations usually surprise operators, who often assume that
variable-speed pumps operate efficiently over the full range of speeds. Actually,
variable-speed pumps are only slightly more efficient than a constant-speed pump
when pumping into a dead-end zone. When considering both the cost of the variable-
speed drive and the drive’s energy costs, variable-speed pumping is not always a cost-
effective option.
Using Pump Energy Data

The methods described previously will enable the operator to identify inefficient
pumps and estimate the cost savings if they are replaced with pumps and motors that
are more efficient. In some cases, the problem may be in the system rather than the
pumps, so modifying the system can be the more cost-effective option. For example,
it may be helpful to install a hydropneumatic tank in a small dead-end system so that
the pump operates against a fairly steady head. Also, in a system with an existing
hydropneumatic tank, changing the pump controls may save energy.
In general, pumps with a steady head on both the suction and discharge sides will be
more efficient than pumps feeding a dead-end zone. Tanks (even hydropneumatic
tanks in small systems) can be very helpful in reducing energy costs.
The model can be used to simulate various strategies. The energy costs associated
with different pumps, operating points, and operating strategies are calculated by the
simulation program, so that the most efficient and cost-effective solution can be cho-
sen. Once the changes are instituted, energy savings pay back the utility year after
year. The present worth of energy savings can be compared with the capital costs to
achieve them.
Whenever there is a major change in the water distribution system, pumps must be re-
examined. Installing a new parallel or larger pipe on the discharge side of a pumping
station lowers the system head curve and moves the operating point to a higher flow
rate. In some cases, this change can move the pump to a less efficient operating point.
In some cases, it may be necessary to counteract the change by trimming the impel-
lers. Another alternative is to throttle valves to move the operating point to a more
efficient location. The model can be used to determine the pump operating points
before and after any changes.
Understanding Rate Structures

To get the most out of pump efficiency studies, it is essential for the utility to under-
stand the power rate structure. Some power utilities have time of day energy pricing,
especially during peak energy consumption periods. This pricing can be accounted for
by using the applicable energy price for each time period. Others use a block rate for
which the unit price drops as the total power consumed increases. Still others use
block rates that change based on the peak energy consumption rate.
Not all energy costs are structured as described in the preceding paragraph. Some
costs are demand charges or capacity charges related to the maximum rate at which
the water utility uses energy. The time period on which the maximum rate is based
may be the peak 15-minute period over the last three months, or the peak hour over
the last year. It may be calculated meter-by-meter, by customer, or by some other cri-
terion. The modeler can estimate the demand charge by simulating peak pumping
periods and determining which one yields the highest power consumption rate. In
some cases, total cost may be more sensitive to the peak demand than the energy use.
The water utility needs to understand this peaking charge and operate the system in a
way that will minimize it. Often, the demand charge is set on a peak day or a day with
a large fire. As a goal, the utility should try not to exceed peak day pumping rates.
The demand charge can be calculated using the peak flow rate in the following equa-
tion with the corresponding head and efficiencies.
Demand (kw) = C f Qh p e e m e p (10.8)
where Cf = unit conversion factor (1.88 English, 0.0098 SI)
Note that time is not considered in the above equation, because the demand is a rate of
energy usage, not the total energy used.
Optimal Pump Scheduling

Pump scheduling is the process of choosing which of the available pumps within a
water supply system are to be used and for which periods of the day they are to be run.
The aim of pump scheduling is to minimize the marginal cost of supplying water
while keeping within physical and operational constraints, such as maintaining suffi-
cient water within the system’s reservoirs, to meet the required time-varying con-
sumer demands.
At the heart of any attempt to improve pump scheduling is the calculation of energy
costs as defined in Equation 10.1. According to the equation, the best way for opera-
tors to minimize cost is to
• Maximize efficiency, e, by having the pumps operating at the best efficiency
point
• Minimize head, h, by operating the pumps at a low head
• Minimize energy price, p, by operating the pumps when energy costs are
lowest
If the pumps were installed correctly for the system under consideration, there is not a
great deal of room for energy cost reduction. However, by trying different pump
scheduling scenarios, the operator can find operating strategies that squeeze cost sav-
ings out of pumping operation. Likewise, once the pumps have been selected, there is
very little room for optimization in systems with no storage because the pumps must
be operated to meet demands. In systems with storage tanks, however, there is much
more room for optimization, such as storing energy in tanks at higher pressure zones
to enable pumps to operate in the most efficient way.
One way to evaluate the vast number of operational approaches is through the appli-
cation of optimization. Operational optimization looks at the day-to-day operation of
a water distribution system with the goal of striking a balance between minimizing
the running costs of pumps and maintaining an acceptable level of service to the cus-
tomer (Brdys and Ulanicki, 1994; and Jowitt and Germanopoulos, 1992).
The important features of pumping costs to be considered in an optimization problem
are the electricity tariff structure (with both consumption and demand charges), the
relative efficiencies of the available pump sets, the head through which they pump,
and the marginal treatment costs. The objective function for optimization is therefore:
min f x = ¦ > cc x + cd x @ (10.9)

x
i=1

x = vector representing pump schedules
N = total number of pump stations
cc = consumption (unit) charge in a pump station
cd = demand (capacity) charge in a pump station
Pump Scheduling Optimization for the

London Main Ring
A successful example of using optimization for What made this more complicated was that in
pump scheduling is the London Ring Main (LRM) order to minimize operational costs, the pump
scheduling system in the United Kingdom. The scheduling for the LRM needed to be considered
LRM, built by Thames Water, is a 50-mile (80-km) in conjunction with the pump scheduling of the
long tunnel, 7.5 ft (2.5 m) in diameter, running 130 London distribution system.
ft (40 m) below London. Built as a loop, the Ring The solution was found by combining hydraulic
Main feeds into the local distribution system via simulation and optimization routines. Two main
eleven vertical shafts and serves as an alternative optimization techniques were used: linear pro-
way of delivering water to the London system. gramming and the traveling salesman algorithm.
Thames Water inherited a water distribution net- Linear programming was used to provide different
work for London that was developed in the Victo- schedules of water delivery, and the traveling
rian times. Rather than simply replacing the old salesman algorithm was used for making changes
pipes (which would have caused enormous dis- to the schedules in order to minimize the number
ruption in the city), Thames Water built the LRM to of pump switches. The scheduling system has
transform water distribution in London. Treated yielded savings of a few percent on pumping
water is fed into the LRM from London’s five main costs. It has also been used extensively for train-
water treatment works and then pumped into Lon- ing staff (Simons, 1996).
don’s distribution system via 11 output shafts.
The problem was therefore to determine how
much water to bring into the LRM and when, and
how much water to pump out of the LRM and
when.
If only hydraulic requirements are considered in the optimization problem, the impor-
tant constraints include consumer demand, reservoir capacities, abstraction limits,
pumping capacity, and treatment works throughput. However, if water quality require-
ments are also considered, the problem becomes much more complicated (see page
467). For example, the best chlorination strategy needs to determine not only the best
pump operation policy but also the best concentration pattern at the sources and at the
booster stations of a distribution system.
Methods for Finding Optimal Pump Schedules. Several types of opti-
mization methods are available that have been used to find optimal pump schedules.
The main methods used at present are linear programming, nonlinear programming,
dynamic programming, and heuristic methods. (See Appendix D for more informa-
tion on different optimization techniques.)
Linear programming: Linear programming (LP) is a widely used operational opti-
mization technique in water distribution systems and related fields, such as water
resources management (Jowitt and Germanopoulos, 1992). The advantages of LP
include the ability to accommodate relatively high dimensionality with comparative
ease, the obtaining of the universal optima, and the availability of standard LP soft-
ware. However, various disadvantages are also present, including the loss of informa-
tion through the linearization of nonlinear relationships, the problem-specific nature

of the method, and the high computational costs for large problems. See page 655 for
more information on linear programming.
Nonlinear programming: Nonlinear programming (NP), such as quadratic program-
ming and separable programming, do not have the same popularity as LP in water
systems analysis. This is particularly due to the fact that the optimization process is
usually slow and computationally expensive when compared with other methods.
Nevertheless, Yu, Powell, and Sterling (1994) and Percia, Oron, and Mehrez (1997)
have applied NP to the operational optimization of distribution systems. See page 659
for more information on nonlinear programming.
Dynamic programming: Dynamic programming (DP) is used extensively in the
optimization of water resources systems (Ormsbee, Walski, Chase, and Sharp, 1989;
Zessler and Shamir, 1989; and Lansey and Awumah, 1994). To apply DP to the oper-
ational optimization of water distribution systems, the problem is decomposed into a
number of stages and is analyzed from one state to the next for all valid operational
states (see page 663 for more information). Once all outcomes have been calculated,
the most economical one is selected as a solution. However, the problem suffers from
the so-called curse of dimensionality, which limits the size of the system that can be
optimized using DP. For example, DP formulations may become inadequate when
there are more than two reservoirs in the system or even for one-reservoir systems that
have several different pump combinations. The amount of necessary calculation
increases so rapidly with the number of reservoirs and possible pump combinations
that the computing requirements become unacceptable.
Heuristic methods: Heuristic methods used for operational optimization of water
distribution systems include genetic algorithms (Goldberg and Kuo, 1987; Esat and
Hall, 1994; Savic, Walters, and Schwab, 1997; Wu, Boulos, Orr, and Ro, 2000; and
Wu, Boulos, Orr, and Moore, 2001) and nonlinear heuristics (Ormsbee and Lini-
gireddy, 1995; and Pezeshk and Helweg, 1996). Genetic algorithms are especially
useful for developing a series of pump on/off decisions to minimize operating costs.
Steps for Optimizing Pump Scheduling. The general steps involved in

optimizing pump scheduling are as follows:
1. Construct a calibrated EPS model.
2. Enter a price tariff for energy and the pump efficiency curves.
3. Develop constraints on tank water levels and system pressures.
4. Estimate the actual demand pattern for the day under consideration.
5. Run the operational optimizer.
6. Check for reasonableness of results.
7. Communicate results to the operator.
Most of these steps are fairly straightforward. However, there can be considerable
uncertainty in the demand estimates on a given day, and small errors in estimating
demand patterns on that day can yield significantly different operating strategies
which may or may not be optimal.
Section 10.9 Water Distribution System Flushing 449
10.9 WATER DISTRIBUTION SYSTEM FLUSHING

Water distribution system flushing is an important tool for helping operators to control
distribution system water quality. Flushing stirs up and removes sediments from
mains and removes poor quality water from the system, replacing it with fresher water
from the source. Procedures for flushing are discussed in detail in Antoun, Dyksen,
and Hiltebrand (1999); California-Nevada AWWA (1981); Chadderton, Christensen,
and Henry-Unrath (1992); Oberoi (1994); Patison (1980); and Walski (2000). The
term directional flushing or unidirectional flushing is used to describe the operation
of the valves during flushing to maximize velocity and control flow direction.
Flushing is usually accomplished by opening one or more hydrants in a planned pat-
tern. The usual rule of thumb for flushing is to always flush with clean water behind
you, meaning that hydrants should be operated to pull the freshest water into the area
being flushed. Flushing programs usually start at the source and move out through the
system.
Unfortunately, operators conducting the flushing program cannot see what is occur-
ring in the mains, or measure parameters like velocity or flow rate in pipes. Water dis-
tribution models provide a way to look into the pipes and obtain an indication of how
a flushing program will work.
Modeling Flushing
Because every pipe can carry a good deal of water during flushing, all pipes in the
area being flushed should be included in the model. If the pipe being flushed has a
single directional feed, as in a branched system, a model is not needed to determine
velocity; however, model runs can still help to identify the system’s ability to main-
tain adequate service. In areas of a system with multiple sources or looping flow
paths, determining the direction of flow and velocity is more challenging. Flow in
pipes will often differ from what the operator anticipates in these cases, so analysis
using a model is very helpful.
Representing a Flowed Hydrant

The easiest way to simulate flushing is to change the demand at the node being
flowed to that expected during flushing, which can be called the free discharge. In
concept, this approach is fairly simple. However, most utilities do not know the free
discharge flow they can expect to get from each hydrant, so this value must be esti-
mated. Closed valves can complicate the process of flow estimation, since valves are
sometimes closed to facilitate directional flushing. Some utilities measure and keep a
record of the discharge from each hydrant during flushing. This information can then
be very helpful in future simulations, providing good estimates for nodal demands
unless there have been significant changes in the system since the last flushing.
Estimating demand is also difficult because of the interrelationship of discharge and
pressure. Pressure in the mains affects hydrant discharge, and hydrant discharge
affects pressure. In larger mains, the flow is almost completely controlled by head
losses in the hydrant lateral and the hydrant itself. In smaller mains, the distribution
system may account for more head loss than the hydrant.
The hydrants must be modeled in a way that accounts for the pressure drop due to
flushing. One way is to represent the hydrant discharging to the atmosphere as an
equivalent pipe discharging to a reservoir located at the elevation of the hydrant. This
representation is shown in Figure 10.17. The first step in this process is to estimate the
head loss in the hydrant, and the conversion of static head to velocity head in the out-
let. As a quick approximation, a 2½-in. (60-mm) hydrant outlet orifice can be repre-
sented by about 1,000 ft (305 m) of equivalent 6-in. (150-mm) pipe added to the
hydrant lateral, while a 4½-in. (115-mm) outlet can be represented by 250 ft (76 m) of
6-in. (150-mm) pipe (Walski, 1995). Because the velocity is very high in the hydrant
lateral, all minor losses must be accounted for, either by using minor loss k-factors or
an equivalent pipe.
Using a reservoir with an equivalent pipe increases the size of the model since each
hydrant represents an additional pipe. The equivalent pipe representing the hydrant
should be closed unless the hydrant is flowing. Provided the modeler uses appropriate
equivalent lengths to represent the hydrant, this method can be more accurate than
simply placing a demand on a node.
Figure 10.17
Representing a
Reservoir Set at hydrant as a reservoir
Elevation of Outlet
(i.e., Atmospheric Pressure)
Node
Equivalent Pipe
for Hydrant and Lateral
Estimating Hydrant Discharge Using Flow Emitters

Another way to model the discharge from a hydrant is to use flow emitters. A flow
emitter is a property of a model node that relates the discharge from the node to the
pressure immediately upstream of the emitter node using:
Q = K P (10.10)
where Q = flow through emitter (gpm, l/s)

K = overall emitter coefficient (gpm/psi0.5, l/s/m0.5)
P = pressure drop across emitter (psi, m)
It should be possible to simply model a hydrant as a flow emitter and enter the appro-
priate value for K. However, not all of the energy available immediately upstream of
the hydrant is lost. Instead, some of the energy is converted into increased velocity
head, especially for the smaller (2.5 in, 63 mm) hydrant outlet.
To accurately model a hydrant, the modeler must provide an overall K value, which
includes both the head loss through the hydrant and the conversion of pressure head to
velocity head. AWWA Standards C502 (AWWA, 1994a) and C503 (AWWA, 1994b)
specify the allowable pressure drop through a hydrant. For example, the standards
state that a 2.5 in. outlet must have a pressure drop of less than 2.0 psi (1.46 m) when
passing 500 gpm (31.5 l/s). These values can be used to relate pressure drop to flow
through the hydrant similar to the way this is done in Equation 10.10.
Q = k P (10.11)
where k = pressure drop coefficient for hydrant (gpm/psi0.5, l/s/m0.5)

P = pressure drop across hydrant (psi, m)
The difference between K and k is that K needs to be solved for and includes the head
loss through the hydrant and accounts for the conversion of velocity head to pressure
head; k, on the other hand, accounts only for the head loss through the hydrant and is
a known value.
To solve for K, the energy equation can be written between a pressure gage immedi-
ately upstream of the hydrant and the hydrant outlet (see Figure 10.18) by using the
following formula:
2 2
v v Q 2
-----p- + z p + P p = -----o- + z o + P o + C f § ----· (10.12)
2g 2g © k¹
where vp = velocity immediately upstream of hydrant (ft/s, m/s)

g = gravitational acceleration constant (ft/s2, m/s2)
zp = elevation where Pp is measured (ft, m)
Pp = pressure immediately upstream of hydrant (psi, m)
vo = velocity at hydrant outlet (ft/s, m/s)
zo = elevation where Po is measured (ft, m)
Po = pressure at hydrant outlet (psi, m)
Cf = unit conversion factor (2.31 for pressure in psi, 1 for pressure in m)
The pressure at the outlet is essentially atmospheric, so Po can be set to 0. By placing

the elevation of the pressure gage on the pipe at the same elevation as the outlet, then
the z terms will cancel out.
Equations 10.10 and 10.12 can both be solved for the pressure immediately upstream
of the hydrant and set equal to one another to give Equation 10.13.
2 2
Q 2 vo – vp Q 2
C f §© ----·¹ - + C f §© ----·¹
= --------------- (10.13)
K 2g k
Figure 10.18
Representing a
hydrant as a flow Pressure
emitter Gage Hydrant
Outlet
//=//= //=//=
Main
Lateral Hydrant
Boot
Because the diameter is known instead of the velocity, the velocity can be replaced by
Q-
v = ---------- (10.14)
2
cf D
where D = hydrant outlet diameter

cf = unit conversion factor
(2.44 for Q in gpm, D in in., 0.0785 for Q in l/s, D in mm)
Substituting Equation 10.14 into 10.13 and solving for K results in the following:
1
K = --------------------------------------------------------------------
- (10.15)
§ 1 § 1 1 · 1 ·1 e 2
¨ ----------------- ¨ ------ – ------¸ + -----¸
© 2gC f c 2f © D 4 D 4¹ k 2¹
o p
A typical hydrant lateral in North America is 6 in. (150 mm) in diameter and typical
outlet sizes are 2.5 in. (63 mm) and 4.5 in. (115 mm) in diameter. AWWA standards
(AWWA, 1994a; AWWA, 1994b) give the maximum allowable head loss through a
hydrant at specified flow rates. Most hydrants have considerably less head loss. Using
these Q and P values from the standards or manufacturer tests, k can be calculated
from Equation 10.11. Table 10.4 provides values for K for a range of k values for 6-in.
(150-mm) pipes.
Table 10.4 Emitter K values for hydrants

Outlet Nominal k k K K
(in.) (gpm/psi0.5) (l/s/m0.5) (gpm/psi0.5) (l/s/m0.5)
2.5 250–600 18–45 150–180 11–14
2–2.5 350–700 26–52 167–185 13–15
4.5 447–720 33–54 380–510 30–40
The coefficients given are based on a 5-ft (1.5-m) burial depth and a 5.5-in. (140-mm)
hydrant barrel. A range of values is given because each manufacturer has a different
configuration for hydrant barrels and valving. The lowest value is the minimum
AWWA standard.
The elevation of the junction node representing the hydrant should be the elevation of
the hydrant outlet, not the main or the lateral. Because of the high velocity typically
occurring in hydrant laterals, minor losses in the lateral are important and should not
be ignored.
Hydrant Location Relative to Nodes

Although hydrants are located close to intersections, they are usually not located pre-
cisely at model nodes. Therefore, modeling a hydrant flow by placing the flow at the
model node is not entirely correct. Usually, this problem is only significant if the
hydrant is on a small main close to a junction with a large main [for example, a
hydrant located on a 6-in. (150-mm) pipe 30 ft (9 m) from a connection to a 24-in.
(600-mm) main]. In the model, almost all of the flow will arrive at the hydrant from
the 24-in. (600-mm) pipe. In reality, the flow must also travel through 30 ft (9 m) of 6-
in. (150-mm) pipe before reaching the hydrant. Because the 6-in. (150-mm) pipe
restricts the flow much more than the 24-in. (600-mm) pipe, it can have a significant
effect on hydrant discharge.
In addition, when the utility is practicing directional flushing (that is, closing valves to
control the direction of flow), differences between the actual position of the hydrant
and the position of the hydrant model node may be significant. To complicate matters,
there may be a closed valve between the hydrant and the node in the model during
directional flushing. A hydrant that is close to a node may not be directly connected to
the node, as shown in Figure 10.19. Placing the demand at the existing node may not
be accurate, and another node might need to be added to depict the situation accu-
rately. If the utility will be using directional flushing, then numerous additional nodes
may be needed to accurately represent closed valves.
Figure 10.19
Accurately modeling
Hydrant
a hydrant near a
closed valve
Closed Valve
System Map
Hydrant
Node Discharge Closed Pipe
Modeling Hydrant at Node
Hydrant
Discharge
Node
Modeling Hydrant Beyond Closed Valve
Steady-State versus EPS Runs

If the operator is interested only in velocities and is not trying to track water quality,
then a steady-state run can be used to simulate each hydrant being flowed for flush-
ing. The maximum velocity in each pipe over all of the steady-state simulations can
then be used as an indicator in evaluating the effectiveness of flushing in increasing
velocities and thereby cleaning the pipes. Individual water quality constituents can
also be tracked using EPS runs to simulate flushing.
Some utilities do not fully open the fire hydrant for the entire flushing period. Instead,
they fully open it to stir up sediment, and then back off the flow to remove the sedi-
ment from the pipe. This method is effective as long as the flow continues for a suffi-
cient period of time, and continues in the same direction. When a hydrant is located
between a tank and a pump, most of the flow may first come from the tank because
the flow from the pump is limited by the pump curve. As the tank level drops and the
hydrant flow backs off, however, more water may be delivered to the flowed hydrant
from the pump rather than the tank. Therefore, the water stirred up during the higher
flow period may not be removed effectively from the system and may inadvertently
enter the tank. Figure 10.20 illustrates this situation. Model runs would show that it
would be necessary to flush at a high rate or shut valves to control the direction of
flow.
Figure 10.20
Tank Effect of flushing a
hydrant located
between pump and
tank
900 gpm
Pump
300 gpm 600 gpm
High Flow
Tank
100 gpm
Pump
200 gpm 100 gpm

Lower Flow
Indicators of Successful Flushing

When using simulations to determine the effectiveness of the flushing technique, the
operator selects the parameters to use as indicators of success. The velocity is gener-
ally the best indicator of flushing effectiveness. There is no fast rule, however, for
determining the velocity needed for successful flushing. In a neighborhood distribu-
tion grid of small pipes, a velocity of 2 ft/s (0.6 m/s) is considered a good value. How-
ever, raising velocity to values on the order of 5 ft/s (1.5 m/s) will result in even better
flushing. It is difficult to achieve high velocities in large transmission mains, so flush-
ing has little effect there.
Another important consideration is the change in velocity in a pipe relative to the
pipe’s normal velocity. For example, a pipe that normally experiences velocity of 2 ft/s
(0.6 m/s) would require a substantially higher velocity for flushing. Any sediment in
that pipe that could be suspended by a 2 ft/s (0.6 m/s) velocity has already been moved,
and the material that is left is of a size and density that requires velocities greater than
2 ft/s (0.6 m/s) to be moved.
Velocity is the best indicator of flushing success, but the flushing must be carried out
for a long enough period of time to allow the water disturbed by flushing to be trans-
ported out of the system. Flushing can be tracked using the water quality tracing fea-
ture in a model, and the user can track chlorine as an indicator of how effective the
flushing is.
Alternatively, the modeler can tag the water initially in the system and the tank with a
concentration of 0.0 mg/l and tag the water at the source with a concentration of 100.0
mg/l and monitor how long it takes for the fresh water to move through the system
during flushing (concentration in all pipes equal to 100 mg/l). Using a hydraulic sim-
ulation program’s color-coding feature, the fresh water from the source can be repre-
sented by the color blue, while the water present in the pipes at the start of the
simulation can be represented by red, and a spectrum of colors can be used for the
concentrations in between. In this way, the model user can view the way in which the
old water is displaced by the fresher water. This type of simulation can serve as an
indication of how long the flushing needs to continue.
Ideally, the models would be able to track turbidity during flushing events. However,
the effect of flushing on turbidity cannot be accurately predicted since turbidity does
not conform to the law of conservation of mass (Walski, 1991). Some insights can be
gained, however, by correlating turbidity with flushing velocity (Walski and Draus,
1996).
One of the problems in reviewing the results of flushing runs is determining the extent
to which they were successful. When a hydrant is flowed, there are actually three lev-
els of effectiveness that can be used to judge the success or failure:
• Areas cleaned by flushing

• Areas undisturbed by flushing
• Areas that are stirred up but not successfully cleaned
The goal is to maximize the cleaned areas and minimize areas that are disturbed and
not successfully cleaned. The latter areas contribute to turbidity complaints after
flushing and are characterized by velocities that increase slightly over normal condi-
tions, but not significantly enough to successfully be flushed, and by insufficient
durations that prevent suspended sediment from making its way to the flushed
hydrant. The model can help locate these areas and minimize them by adjusting the
valving and sequence of flushing.
Section 10.10 Sizing Distribution System Meters 457
10.10 SIZING DISTRIBUTION SYSTEM METERS
Subsystem Metering
Many water systems have been constructed with little or no metering between the mas-
ter meters at the treatment plant (or well) and the customer meters. System operators
may like to have more information about distribution system flows at locations
throughout the system to better understand water use, quantify available capacity in the
system, and compute the amounts of unaccounted-for water in different parts of the
system. Understanding unaccounted-for water can be helpful in managing leak detec-
tion and repair, and in instituting water main replacement programs. Once the meter is
in place, it can also provide additional information used in fine-tuning model calibra-
tion. Subsystem metering is commonly practiced in the United Kingdom, where indi-
vidual customers are not universally metered.
Because of cost and difficulty of installation, system flow meters are only placed in a
few select points throughout the system. Pump stations and PRV vaults at pressure
zone boundaries are the most common locations for metering. Other key metering
locations include pipelines that carry virtually all of the flow into an area.
Sizing the meters is primarily a problem of understanding the range of flows that the
meter will experience. The meter needs to be selected to pick up both the high and the
low range of flows. Unfortunately, unlike the situation with customer metering, there
is no fixture unit method (see page 399) that can be used for large areas within the dis-
tribution system. The next section provides guidelines for using models to size
meters.
Using Models for Meter Sizing

Before computerized modeling, the operator would estimate the range of flows based
either on the number of customers or on readings from a temporary meter, such as a
Pitot rod. Use of the Pitot rod required excavation to the top of the main and tapping
the pipe. Even a less accurate clamp-on meter requires access to the pipe and long
straight runs of pipe upstream of the meter, and when pipe wall properties are
unknown, clamp-on meters need to be calibrated in place to have any validity. With
temporary metering, the operator cannot be certain that the full range of flows has
been captured because the temporary meter may not have been installed during the
extreme event.
Running an EPS model can generate a demand pattern for any demand condition,
including projected conditions. By making multiple runs of the model and plotting
them, the operator can see the kinds of flows the meter is likely to encounter.
The meter can be sized based on the pipe size and the type of meter. For example,
after the range of flows is determined, the engineer can select the Beta ratio (the ratio
of throat diameter to pipe diameter) for a venturi meter, or determine whether the
velocity is high enough for an electromagnetic meter. The flow patterns are usually
quite different for a meter on a pipe that serves an area continuously (for example, a
pipe from a PRV or a variable-speed pump with no storage) and a meter that is located
at a pump that cycles on and off as it is discharging into a zone with storage. These
different patterns are shown in Figure 10.21. Meters that are located at a pump in a
system having storage can have a more limited range, and those on lines where flows
must vary continuously to meet demands require a much greater range.
Implications for Meter Selection

After the range of flows has been estimated, the type of meter can be selected. Small
flows, such as those on 4- or 6-in. (100- to 150-mm) pump discharge lines, can be
metered by turbine meters equipped with some type of pulse counter that produces
rate of flow information in analog form.
Figure 10.21
Variations in flow 250
through meter to On-Off Pumping
system with and
without storage
200
Discharge, gpm
150
100 Continuous
Pumping
50
0
0 12 24 36 48
Time, hr
As flows become greater, the turbine meters are used less often than electromagnetic
(mag) meters, differential head meters (Venturi, orifice, flow tube, and nozzle types),
or ultrasonic meters. Differential head meters are usually the most reliable and least
expensive, and can be run without power. Unfortunately, they are limited (with some
exceptions) to unidirectional flow and can produce significant head loss as the veloc-
ity increases. Also, recent advances in PRV technology have produced several types
of PRVs that can also serve as flow meters.
Ultimately, the selection of a meter depends on the nature of the flow, the site, and the
preference of the operator. The pipe network model can be used to provide informa-
tion on the range of flows, and after the meter is in place, the model can be improved
with information from the flow meter.
Section 10.11 Models for Investigation of System Contamination 459
10.11 MODELS FOR INVESTIGATION OF SYSTEM

CONTAMINATION
Models can also be used to analyze historical water system contamination events.
Usually, when water distribution system contamination is suspected, there is not
enough data to completely reconstruct the events. Using available data as a starting
point, it is possible to reconstruct the events that caused contamination and identify
locations in the system that received contaminated water using EPS models. Because
it is not possible or desirable to re-create the actual contamination event(s), the mod-
eler usually does not have all the data that would be available for modeling the
current-day system, especially when the contamination events happened years earlier.
The results involve greater uncertainty than modeling current day conditions, but rea-
sonable conclusions can nevertheless be drawn from the model.
The first application of water quality modeling to distribution systems was the now
famous case of Woburn, Massachusetts (Murphy, 1991; and Harr, 1995). The model
was used to substantiate contamination that had happened years earlier. Therefore,
several different versions of the model were needed to reflect how the system evolved
over the time of the alleged contamination. Murphy also applied this type of modeling
to San Jose, California.
The USEPA applied water quality modeling to investigate contamination first in
Cabool, Missouri (Escherichia coli serotype 0157:H7) and Gideon, Missouri (Salmo-
nella typhimurium). The Cabool outbreak involved 243 cases and six deaths and was
attributed to two water line breaks (Geldreich, 1996). In Gideon, water quality model-
ing was used to identify contamination of a tank by bird droppings as the likely source
(Clark, et al., 1996; Clark and Grayman, 1998). Of recent interest is the water system
contamination in Walkerton, Ontario, Canada.
Kramer, Herwald, Craun, Calderon, and Juranek (1996) reported that 26.7 percent of
waterborne disease outbreaks in the United States in 1993-94 were due to distribution
system deficiencies. For those cases due to source deficiencies, water quality models
can be used to track the area of influence of each water source if the system has sev-
eral sources.
Water quality models have also been used in forensic studies (also referred to as hind-
casting) to identify responsible parties in litigation. While many such studies have
been conducted, few have been published because of legal issues surrounding such
applications. Two cases of forensic applications of water quality models are docu-
mented in Harding and Walski (2000) for Phoenix/Scottsdale, Arizona, and Maslia, et
al. (2000) for Dover Township (Tom’s River), New Jersey. The Phoenix study showed
that contaminant concentrations can fluctuate widely at a particular location, both
hourly and seasonally, due to changes in pump operation caused by varying demands.
Forensic studies are different from most planning and design applications in that
instead of running the EPS model for an average day or a peak day, the model must be
run for a large number of actual days to reconstruct historical events. This type of
modeling involves a great deal of research into how the system was being operated
when the contamination is alleged to have occurred. Because the system evolved over
many years, several versions of the model are required to reflect the different topol-
ogy, facilities, and loading.
While typical water quality modeling studies reported in the literature have focused
on disinfectant decay (Clark, Grayman, Goodrich, Deininger, and Hess, 1991; and
Rossman, Clark, and Grayman, 1994), forensic studies can involve a variety of con-
taminants, including VOCs, microbes, and inorganic chemicals. Although behavior of
these chemicals is often treated in the models as conservative (that is, no mechanism
exists for volatile chemicals to leave a pipe), the chemicals can undergo a variety of
transformations. For example, volatile chemicals can leave water stored in tanks. Wal-
ski (1999) determined that only a small portion of TCE (Trichloroethylene) is
expected to be lost to the air in storage tanks because of the lack of turbulence in a
tank.
Even when it is not possible to accurately model the behavior of a chemical or
microbe in a distribution system, water quality models can provide a picture of how
water moves in the system, and therefore which portions of the system are exposed to
water from particular sources, tanks, or pipe breaks.
10.12 LEAKAGE CONTROL

Distribution system water losses can be split into two basic categories: losses due to
pipe bursts and losses resulting from background leakage. Bursts are characterized by
a sudden loss of water limited in duration to the time reported and unreported bursts
are allowed to discharge. Background losses are characterized by a continual seeping
of water from pipe fittings and from mains that are cracked or perforated through cor-
rosion.
The quantity of leakage from a water distribution system is related to the system pres-
sure, thus reducing pressures during off-peak hours can reduce leakage. Although
controlling leakage by reducing pressures is not common practice in North America,
this approach is used in some European systems (Goodwin, 1980; Germanopoulos,
1995). Pressure reduction is accomplished through valve operation.
In addition to water pressure management, active leakage control involves the disag-
gregation of large networks into smaller areas (called District Metering Areas in the
United Kingdom) that can be better monitored (Engelhardt, Skipworth, Savic, Saul,
and Walters, 2000). Water audits, detailed accounting of water flow into and out of
portions of the distribution system, are then used to identify areas having excessive
leakage. Unfortunately, they do not provide specific information about the location of
leaks. To pinpoint leaks, detection surveys are required.
Modeling can be used to help manage leak reduction by determining the effects that
either the disaggregation of networks into smaller areas or the adjustment of the con-
trolling valves has on pressure and flow through the system. The modeler should note
that there is a practical limit to which pressure can be reduced before it adversely
affects customers at higher elevations.
Although predicting pressure as a result of valve operation is fairly straightforward,
using a model to estimate the amount of leak reduction is far less precise. This estima-
tion involves modeling leakage through the use of a flow emitter element or a reser-
voir connected to the system through a small pipe (see page 403 for information on a
Section 10.12 Leakage Control 461
similar technique used for modeling sprinklers). A reduction in pressure results in a

reduction in flow through this element. Whenever a leak is repaired, the coefficient of
this element is then changed to reflect the reduced number of leaks.
Several studies have been conducted on leakage modeling using steady-state simula-
tions (Martinez, Conejos, and Vercher, 1999; Pudar and Ligget, 1992; and Stathis and
Loganathan, 1999), but the hidden nature of leaks limits the precision. The use of
inverse transient models to locate leaks has also been reported in literature (Tang,
Brunone, Karney, and Rosetti, 2000; Kapelan, Savic, and Walters, 2000).
Leaks in water systems are somewhat pressure dependent. If the pressure decreases,
then leakage should decrease. If the modeler wants to study leakage as a function of
pressure, then placing the leaks in the system as known demands will not be useful.
Instead pressure dependent leakage should be modeled using flow emitters (see page
451). The difficulty is, of course, determining where to place this (or these) emitter(s)
and how to set the coefficients. While it would be most accurate to assign small emit-
ter coefficients to all nodes, it is easier to place the leakage at one or two nodes which
will accomplish roughly the same effect.
To estimate the emitter coefficient, the modeler must estimate the leakage flow and
then, using an average pressure in the zone, calculate the overall emitter coefficient K
using
Q
K = ------- (10.16)
P
where K = emitter coefficient

Q = leakage flow (gpm, l/s)
P = average zone pressure (psi, kPa)
Using an average pressure is not completely accurate because leakage is not a linear
function of pressure. However, there is so much uncertainty in understanding leakage
that this error is not significant.
If the leakage is placed at a single node, the K determined in Equation 10.16 should
be used at that node. If the leakage is spread out over N nodes, then the K for each
node can be given as
Kn = K
---- (10.17)
N
where Kn = emitter coefficient at nth node

K = overall emitter coefficient
N = number of nodes with leakage
Example – Computing Leakage Using Flow Emitters. Consider a pressure zone

with an estimated leakage of 200 gpm and an average pressure of 65 psi. The emitter coefficient
would be
200- = 24.8
K = ---------
65
If the leakage is to be spread around to 40 nodes, each node would have a coefficient of 0.62.
10.13 MAINTAINING AN ADEQUATE DISINFECTANT

RESIDUAL
A primary disinfectant is applied at the treatment plant to oxidize inorganics, control
taste and odor, and achieve disinfection requirements. Secondary disinfectants are
applied to maintain a protective disinfectant residual during distribution. Maintenance
of a residual within the distribution system is widely employed to inhibit the growth
of bacteria, pathogens, and biofilms. The most popular disinfectant in the United
States is chlorine, due to its oxidizing power, favorable residual characteristics, and
economy. Other commonly used secondary disinfectants include chloramines, and
chlorine dioxide.
When disinfected water moves through a distribution system, it reacts with natural
organic matter (NOM) in the bulk fluid, and biofilms and pipe materials at the pipe
wall, causing disinfectant demand and residual loss in the distribution system. The
complex oxidation and substitution reactions between NOM and residual disinfec-
Section 10.13 Maintaining an Adequate Disinfectant Residual 463
tants result in the formation of a class of compounds referred to as disinfection by-

products (DBPs). DBPs are recognized as having potential negative health impacts. In
fact, various potential health risks have been attributed to DBPs, including cancer,
birth defects, and spontaneous abortion (Boorman et al., 1999; Waller, Swan,
DeLorenze, and Hopkins, 1998). Therefore, allowable concentrations are regulated
within the United States and many other countries.
In the United States, the Stage 1 Disinfectants and Disinfection Byproducts Rule
establishes maximum contaminant levels (MCLs) in the form of maximum annual
average values for disinfectants and disinfectant by-products. These include
• Chlorine: 4 mg/l (as Cl2)
• Total trihalomethanes: (TTHM) 80 Pg/l
• Haloacetic acids (5): 60 Pg/l

Total trihalomethanes is the sum of the concentrations of chloroform, bromodichlo-
romethane, dibromochloromethane, and bromoform. Haloacetic acids (5) is the sum
of the concentrations of mono-, di-, and trichloroacetic acids and mono- and dibro-
moacetic acids (U.S. EPA, 1998 and 2000).
The most common DBPs formed by chlorine are trihalomethanes (THM) such as
chloroform and bromoform, and haloacetic acids (HAA) such as dichloroacetic acid.
Other disinfectants form DBPs as well. Therefore, water utilities must balance the
risks associated with too little or too much disinfectant being added. The generally
accepted practice (or goal) is to add sufficient disinfectant at the treatment plant in
order to maintain a disinfectant residual throughout the distribution system while at
the same time not over-disinfecting, which causes the formation of DBPs. Some water
utilities rechlorinate within the distribution system in order to maintain residual levels
in zones of the distribution system that contain older water.
Disinfectant Residual Assessment

Traditionally, disinfectant residual levels in a distribution system are assessed by
using a field sampling program. Monitoring requirements for disinfectants in the dis-
tribution system are influenced both by federal and state regulations. EPA’s Surface
Water Treatment Rule (SWTR) requires water systems using surface water to have a
disinfectant residual of at least 0.2 mg/l at entry points into the system and to maintain
a detectable residual throughout the distribution system. However, it does not specify
how many or where the residual is to be measured, nor does it apply to groundwater
systems. In most cases, water utilities utilize monitoring sites that were selected for
mandatory monitoring of coliform. EPA’s Total Coliform Rule (TCR) specifies the
number of required monitoring sites based on the number of customers.
Note that because of sampling limitations, monitoring cannot provide a complete pic-
ture of the spatial and temporal variation of disinfectant and DBP concentrations
within the distribution system. For example, disinfectant residual can vary signifi-
cantly between nearby sites and over the course of the day based on water demands
and the fill and draw operations of tanks.
Figure 10.22 illustrates the diurnal variation in chlorine residual at two locations
within a distribution system. The solid line corresponds to a junction in close proxim-
ity to a treatment plant. As would be expected, water reaches the junction very
quickly and as a result, little loss of chlorine residual occurs (the chlorine dosing rate
at the treatment plant is 1 mg/l), and the residual remains relatively constant through-
out the entire day. The dashed line is a junction that is quite a distance from the treat-
ment plant and is affected by a nearby tank. As a result, (1) travel time to the junction
is longer, resulting in increased loss of chlorine residual, and (2) during the times that
the junction is receiving water that had been in the tank, the chlorine residual is very
low. If the junction near the tank were routinely sampled during the daytime, monitor-
ing results would not reflect the very low chlorine residuals that occur during the
night.
Figure 10.22
Example diurnal
variation in chlorine
residual at two
junctions 1.0
Chlorine Residual, mg/L
0.8
Near WTP
0.6
0.4
Near Tank
0.2
0
0 6 12 18 24
Time, hours
Another shortcoming of depending exclusively on monitoring is that monitoring pro-

vides little information to help assess the impacts of alternative operating or disinfec-
tant options on the distribution of disinfectant and DBP concentrations throughout the
system prior to actually implementing these options. On the other hand, properly per-
formed distribution system modeling is an excellent tool for doing a wide range of
“what-if” scenarios. The prerequisite for modeling disinfectant residual and DBP con-
centrations in a distribution system is a calibrated, extended-period simulation (EPS)
water quality model of the distribution system.
Models can be used to evaluate a wide range of operating and design factors that can
affect the disinfectant and DBP concentrations within the distribution system. The
following steps are generally followed in such analyses:
1. Select one or more seasonal demand conditions. During high demand summer
conditions, residence time in the distribution system is generally lower than dur-
ing lower demand, winter conditions. However, due to the higher temperatures,
disinfectant decay rates are greater in the summer. As a result, it is frequently
necessary to evaluate both seasons to determine the more critical conditions.
Section 10.13 Maintaining an Adequate Disinfectant Residual 465
2. Simulate the disinfectant and DBP behavior over a period of several days.
Because residence time in a system may extend over periods of several days or
even weeks, the system should be simulated over a period long enough to capture
the dynamics of the system that cause low residuals or high DBP concentrations.
A calibrated EPS hydraulic model can be used to estimate residence times, and
thus for determining a sufficient duration for a simulation.
3. Evaluate the effects of different operating procedures on disinfectant residual and

DBP concentration. Operational plans that may affect disinfectant residuals
include dosage rates at the treatment plant, tank and reservoir operations that
affect the storage residence times (such as lowering water levels in the winter),
and dead-end flushing programs. Operational changes should be evaluated in
terms of their cost, hydraulic reliability, and impact on disinfectant residuals and
DBP concentrations throughout the system.
4. Evaluate structural changes on disinfectant residual and DBP concentrations. If

operational changes are not effective, then structural changes can be investigated
that will better maintain the disinfectant residual or reduce DBP formation.
Examples of potential structural changes include installation of booster chlorina-
tion facilities in the distribution system, modifications to pumps or tanks to
reduce residence times in storage facilities, or modifications in pressure zones.
Examples of such studies are described in Clark and Grayman (1998); Vasconcelos et
al. (1996); Tryby et al. (1999); Vandermeyden and Hartman (2001); Elton, Brammer,
and Tansley (1995); Sekhar and Uber (2001); Prentice (2001); and Kiene and Hemery
(1999).
Booster Chlorination
The lowest disinfectant residuals are frequently found in parts of the distribution sys-
tem that experience long travel times from the disinfectant dosing location. Examples
of locations that are most susceptible to low residuals include areas served by tanks
and especially those served by cascading tanks (several pressure zones and tanks in
series), long dead ends, and mixing zones that receive water from multiple sources.
Areas that are served by pipes with a high wall demand for disinfectant (for example,
smaller diameter, older, unlined cast iron pipes) frequently see a reduced disinfectant
residual as well.
Booster chlorination is a means of adding chlorine in the distribution system at the

areas that experience low residuals. The advantages of booster chlorination (versus
increasing the dosage rate at the treatment plants) are that
• It avoids the natural decay of the residual in traveling from the plant to the
points of low residual;
• It results in an effective redistribution of disinfectant dosages from the treat-

ment plant to the periphery of the distribution system where it is needed;
• It has the potential to reduce the formation of disinfectant by-products.

Although booster chlorination is practiced by many water utilities, many other utili-
ties have avoided this practice because of the operations, maintenance, and security
issues associated with remote chlorination stations. There have been several recent
research studies that have investigated the optimal location and control of booster
chlorination stations in the distribution system (Tryby et al., 1999; Nace, Harmant,
and Villon, 2001; Propato, Uber, Shang, Polycarpou, 2001; Wang, Polycarpou,
Shang, and Uber, 2001).
Representation of booster chlorination in earlier water quality models was difficult
and approximate because a constituent could only be introduced by adding flow with
a specified concentration. Therefore, in order to simulate the addition of chlorine
(without adding flow), one had to create a demand and a nearby inflow of the same
flow magnitude and specify the concentration in the inflow. This representation lim-
ited the ability to simulate the actual behavior of the booster facility. Recent models
have added several options that provide greater ease and flexibility in simulating
booster chlorination. The user can now select the method and appropriate chlorine
concentration that most closely represents the operation of the particular booster chlo-
rination facility being simulated. The following sections discuss three options that are
provided in some models.
Mass Booster Source. This case is used to represent a feeder, which is manu-
ally set to feed a constant mass feed rate. The concentration out of the feeder node is
dependent on flow past the feeder node and is given as
6Q i C i + M
C o = --------------------------
- (10.18)
6Q i
where Co = concentration at and out of node (M/L3)

Qi = flow from i-th inflow into node (L3/T)
Ci = concentration of i-th inflow into node (M/L3)
M = mass feed rate (M/T)
Flow Paced Booster. This case corresponds to raising the concentration a set
amount even as the flow changes and can be used to model a flow paced chemical
feed:
6Q i C i
C o = --------------
- + Cf (10.19)
6Q i
where C f = increase in concentration at node (M/L3)
Setpoint Booster. The final case represents a feed rate that is controlled to main-
tain a fixed output concentration from a node and is typical of a feedback control sys-
tem:
6Q i C i
C m = --------------
- (10.20)
6Q i
Maintaining an Adequate Disinfectant Residual 467
where Cm = concentration that would occur with no feed

due to mixing alone (M/L3)
If Cm < C (setpoint), then

Co = C (setpoint)
where
C (setpoint) = outflow concentration setpoint
If Cm = C (setpoint), no chemical is fed, and thus

Co = Cm
DBP Formation
The formation of disinfectant by-product compounds is related to water age such that
areas containing old water are expected to experience higher DBP concentrations.
This relationship with water age is strongest for the formation of trihalomethanes and
considerably weaker, or nonexistent, for the formation of haloacetic acids (Chen and
Weisel, 1998).
The topic of disinfectant dynamics and DBP formation is a very active area of
research and regulatory action. One particular area of general interest to the drinking
water industry is the U.S. EPA Stage 2 Disinfectants and Disinfection Byproducts
Rule and the Initial Distribution System Evaluation (IDSE) process that will be
required over the coming years to define sampling locations. As part of that process,
water distribution system modeling will be an option in the selection of future sam-
pling locations in the distribution system.
Optimization Techniques
Good water quality corresponds to maintaining disinfectant concentrations between
lower and upper bounds to ensure good disinfecting properties and avoid bad tasting
water. This problem is compounded if different water sources with different water
quality are used and water blending requirements at certain control points in a water
distribution system are imposed.
The concentrations of chlorine are controlled by scheduling the chlorine patterns at
treatment plants, by scheduling pumping operations, and by injecting additional chlo-
rine at intermediate nodes of a network, the booster stations. Unfortunately, the phe-
nomena involved are very complex. Therefore, the determination of the best
chlorination strategy is best approached by optimization techniques to determine the
best pump operation policy, the best concentration pattern at the sources, and the best
concentration patterns at the booster stations in the distribution system.
As opposed to considering only water, the product to be delivered in sufficient quan-
tity to all demand nodes is residual disinfectant. In addition to the cost of operating
pumps, the new objective that can be introduced is related to the cost of chlorination
or minimization of deviations between the optimized and target concentration values.

Tryby and Uber (1999) combined simultaneous minimization of the number of
booster stations and the total mass of disinfectant being applied. They used mixed-
integer linear programming to solve the problem in which the integer variables are the
location of the booster stations and the continuous variables are the periodic mass
injections. Constans, Brémond, and Morel (2000) considered a linear programming
formulation to determine the locations of booster stations and optimize the injection
patterns.
An attempt to use discrete-time optimal control to identify efficient operations of
water supply systems is reported by Sakarya and Mays (1999). They examined the
optimal operation of pumping in a system to minimize the deviations of concentra-
tions of a substance from desired levels, as well as to minimize pump operating times
and total energy cost. The model applies penalty factors to the pressures, substance
concentrations, and water storage heights in the system if they deviate from desired
values. Goldman and Mays (1999) carried out a similar study using simulated anneal-
ing (see page 670 in Appendix D) to optimize system operation. This model was
applied to the North Marin Water District, Novato, California (USA), to determine
optimal operations to minimize power costs while satisfying hydraulic and water
quality constraints.
The most complete approach to considering both design and operation of a water dis-
tribution system (including water quantity and quality) was proposed by Dandy and
Hewitson (2000) and was applied to the Yorke Peninsula, a rural area situated some
50 km west of Adelaide, Australia. They analyzed the use of the genetic algorithm
model to optimize the total social cost of the system, including
• The capital cost of new pipes, pumps, and tanks
• The present value of the electricity cost for pumping
• The present value of the likely cost to the community due to waterborne dis-
eases caused by low chlorine levels
• The present value of the likely community cost due to disinfection by-
products (cancer risk)
• The present value of the community cost of chlorine levels that exceed
acceptable limits (aesthetic value)
• The present value of the cost of disinfection
In contrast to the traditional approach where a distribution system is designed consid-
ering only hydraulic factors, and disinfection is treated as an operational consider-
ation, this study demonstrated the advantages of including optimized design,
operations, and water quality in a single analysis. Many of the aforementioned quanti-
ties are difficult to quantify (aesthetics, cancer risk). To the extent that the costs can be
quantified, they can be included in the problem formulation as benefits to be maxi-
mized or minimized, or constraints to be met.
Goldman, Sakarya, Ormsbee, Uber, and Mays (2000) present several additional models
to assist in optimal operation, but these models are still not widely used by utilities.
References 469
REFERENCES
American Water Works Association (1994a). “Standard for Dry Barrel Fire Hydrants.” AWWA C502-94,
Denver, Colorado.
American Water Works Association (1994b). “Standard for Wet Barrel Fire Hydrants.” AWWA C503-94,
Denver, Colorado.
Antoun, E. N., Dyksen, J. E., and Hiltebrand (1999). “Unidirectional Flushing.” Journal of the American
Arora, H., and LeChevallier, M. W. (1998). “Energy Management Opportunities.” Journal of the American
Boorman, G. A., Dellarco, V., Dunnick, J. K., Chapin, R. E., Hunter, S., Hauchman, F., Gardner, H., Cox,
M., Sills, R. C. (1999). “Drinking Water Disinfection By-products: Review and Approach to Toxicity
Evaluation.” Environmental Health Perspectives, 107, Supplemental 1:207.
Brdys, M. A., and Ulanicki, B. (1994). Operational Control of Water Systems : Structures, Algorithms, and
Applications. Prentice Hall.
Brion, L., and Mays, L. W. (1991). “Methodology for Optimal Operation of Pumping Stations in Water
Distribution Systems.” Journal of Hydraulic Engineering, ASCE, 117(11), 1551.
California-Nevada AWWA (1981). Distribution Main Flushing and Cleaning. California-Nevada AWWA,
Rancho Cucamonga, California.
Cesario, A. L. (1995). Modeling, Analysis, and Design of Water Distribution Systems. AWWA, Denver,
Colorado.
Chadderton, R. A., Christensen, G. L., and Henry-Unrath, P. (1992). Implementation and Optimization of
Distribution Flushing Programs. AWWARF, Denver, Colorado.
Chase, D. V., and Ormsbee, L. E. (1989). “Optimal Pump Operation of Water Distribution Systems with
Multiple Tanks.” Proceedings of the AWWA Computer Conference, American Water Works Association,
205.
Chen, W. J., and Weisel, C. P. (1998). “Halogenated DBP Concentrations in a Distribution System.” Jour-
Clark, R. M., Geldreich, E. E., Fox, K. R., Rice, E. W., Johnson, C. H., Goodrich, J. A., Barnick, J. A., and
Abdesaken, F. (1996). “Tracking a Salmonella Serovar Typhimurium Outbreak in Gideon, Missouri:
Role of Contaminant Propagation Modeling.” Journal of Water Supply Research and Technology-Aqua,
45(4), 171.
Clark, R. M., and Grayman, W. M. (1998). Modeling Water Quality in Drinking Water Distribution Sys-
tems. AWWA, Denver, Colorado.
Clark, R. M., Grayman, W. M., Goodrich, R. A., Deininger, P. A., and Hess, A. F. (1991). “Field Testing
Distribution Water Quality Models.” Journal of the American Water Works Association, 84(7), 67.
Constans, S., Brémond, B., and Morel, P. (2000). “Using Linear Programs to Optimize the Chlorine Con-
centrations in Water Distribution Networks.” Proceedings of the ASCE Joint Conference on Water
Resources Engineering and Water Resources Planning and Management, Minneapolis, Minnesota.
Coulbeck, B., Bryds, M., Orr, C., and Rance, J. (1988). “A Hierarchal Approach to Optimized Control of
Water Distribution Systems.” Journal of Optimal Control Applications and Methods, 9(1), 51.
Coulbeck, B., and Sterling, M. (1978). “Optimal Control of Water Distribution Systems.” Proceedings of
the Institute of Electrical Engineers, 125, 1039.
Dandy, G., and Hewitson, C. (2000). “Optimising Hydraulics and Water Quality in Water Distribution Net-
works using Genetic Algorithms.” Proceedings of the ASCE Joint Conference on Water Resources
Engineering and Water Resources Planning and Management, Minneapolis, Minnesota.
Elton, A., Brammer, L. F., and Tansley, N. S. (1995). “Water Quality Modeling in Distribution Networks.”
Engelhardt, M. O., Skipworth, P. J., Savic, D. A., Saul, A. J., and Walters, G. A. (2000). “Rehabilitation
Strategies for Water Distribution Networks: A Literature Review with a UK Perspective.” Urban Water,
Vol. 2, No. 2, 153-170.
Esat, V., and Hall, M. (1994). “Water Resources System Optimization Using Genetic Algorithms.”
Hydroinformatics '94, Vol. 1, Rotterdam, Netherlands.
Geldreich, E. E. (1996). Microbial Quality of Water Supply in Distribution Systems. Lewis Publishers, Boca
Raton, Florida.
Germanopoulos, G. (1995). “Valve Control Regulation for Reducing Leakage.” Improving Efficiency and
Reliability in Water Distribution Systems, Kluwer Academic Press, London, United Kingdom, 165.
Goldberg, D. E., and Kuo, C. H. (1987). “Genetic Algorithms in Pipeline Optimization.” Journal of Com-
puting in Civil Engineering, 1(2), 128.
Goldman, F. E., and Mays, L. W. (1999). “The Application of Simulated Annealing to the Optimal Opera-
tion of Water Systems.” Proceedings of the 26th Water Resources Planning and Management Confer-
ence, American Society of Civil Engineers, Tempe, Arizona.
Goldman, F. E., Sakarya, B., Ormsbee, L. E., Uber, J., and Mays, L. W. (2000). “Optimization Models for
Operations.” Water Distribution Systems Handbook, Mays, L. W., ed., McGraw-Hill, New York, New
York.
Goodwin, S. J. (1980). “The Results of the Experimental Program on Leakage and Leakage Control.” Tech-
nical Report TR 154, Water Research Centre.
Harding, B. L., and Walski, T. M. (2000). “Long Time-series Simulation of Water Quality in Distribution
Systems.” Journal of Water Resources Planning and Management, ASCE, 126(4), 199.
Harr, J. (1995). A Civil Action. Vintage Books, New York, New York.
Hovstadius, G. (2001). “Pump System Effectiveness.” Pumps and Systems, 9(1), 48.
Jowitt, P. W., and Germanopoulos, G. (1992). “Optimal Pump Scheduling in Water Supply Networks.”
Kiene, L., and Hemery, A. (1999). “Monitoring of Water Quality in Distribution Systems Using Optimized
Models, On-line Sensors and Booster Chlorination Stations.” Proceedings of the Water Resources Plan-
ning and Management Conference, American Society of Civil Engineers, Reston, Virginia.
Kramer, M. H., Herwald, B. L., Craun, G. F., Calderon, R. L., and Juranek, D. D. (1996). “Waterborne dis-
ease—1993 and 1994.” Journal of the American Water Works Association, 88(3), 66.
Kapelan, Z., Savic, D.A., and Walters, G.A. (2000). “Inverse Transient Analysis in Pipe Networks for
Leakage Detection and Roughness Calibration.” Water Network Modeling for Optimal Design and
Management, CWS 2000, Centre for Water Systems, Exeter, United Kingdom, 143.
Lansey, K. E., and Awumah, K. (1994). “Optimal Pump Operations Considering Pump Switches.” Journal
of Water Resources Planning and Management, ASCE, 120(1), 17.
Lansey, K. E., and Zhong, Q. (1990). “A Methodology for Optimal Control of Pump Stations.” Proceedings
of the Water Resources Planning and Management Specialty Conference, American Society of Civil
Engineers, 58.
Martinez, F., Conejos, P., and Vercher, J. (1999). “Developing an Integrated Model of Water Distribution
Systems Considering both Distributed Leakage and Pressure Dependent Demands.” Proceedings of the
ASCE Water Resources Planning and Management Division Conference, Tempe, Arizona.
Maslia, M. L., Sautner, J. B., Aral, M. M., Reyes, J. J., Abraham, J. E., and Williams, R. C. (2000). “Using
Water-Distribution Modeling to Assist Epidemiologic Investigations.” Journal of Water Resources
References 471
Murphy, P. J. (1991). Prediction and Validation in Water Distribution Modeling. AWWARF, Denver, Colo-
rado.
Nace, A., Harmant, P., and Villon, P. (2001). “Optimization of Location and Chlorine Dosage of the
Booster Chlorination in Water Distribution Network.” Proceedings of the World Water and Environmen-
tal Resources Congress, The Environmental & Water Resources Institute, Reston, Virginia.
Oberoi, K. (1994). “Distribution Flushing Programs: The Benefits and Results.” Proceedings of the AWWA
Annual Conference, American Water Works Association, New York, New York.
Ormsbee, L. E., and Lingireddy, S. (1995). “Nonlinear Heuristic for Pump Operations.” Journal of Water
Resources Planning and Management, ASCE, 121(4), 302.
Ormsbee, L. E., Walski, T. M., Chase D. V., and Sharp W. W. (1989). “Methodology For Improving Pump
Operation Efficiency.” Journal of Water Resources Planning and Management, ASCE, 115(2), 148.
Patison, P. L. (1980). “Conducting a Regular Main Flushing Program.” Journal of the American Water
Percia, C., Oron, G., and Mehrez, A. (1997). “Optimal Operation of Regional System With Diverse Water
Quality Sources.” Journal of Water Resources Planning and Management, ASCE, 123(2), 105.
Pezeshk, S., and Helweg, O. J. (1996). “Adaptive Search Optimization in Reducing Pump Operating
Costs.” Journal of Water Resources Planning and Management, ASCE, 122(1), 57.
Prentice, M. (2001). “Disinfection Modeling Study for the Owen Sound Water Distribution System.” Pro-
ceedings of the AWWA Annual Conference, American Water Works Association, Denver, Colorado.
Propato, M., Uber, J. G., Shang, F., Polycarpou, M. M. (2001). “Integrated Control and Booster System
Design for Residual Maintenance in Water Distribution Systems.” Proceedings of the World Water and
Environmental Resources Congress, The Environmental & Water Resources Institute, Reston, Virginia.
Pudar, R. S., and Liggett, J. A. (1992). “Leaks in Pipe Networks.” Journal of Hydraulic Engineering,
ASCE, 118(7), 1031.
Reardon, D. (1994). “Audit Manual for Water-Wastewater Facilities.” CR-104300, EPRI, St. Louis, Mis-
souri.
Rossman, L. A., Clark, R. M., and Grayman, W. M. (1994). “Modeling Chlorine Residuals in Drinking-
Sakarya, A. B., and Mays, L. W. (1999). “Optimal Operation of Water Distribution Systems for Water
Quality Purposes.” Proceedings of the 26th Water Resources Planning and Management Conference,
American Society of Civil Engineers, Tempe, Arizona.
Savic, D. A., Walters, G. A., Randall-Smith, M., and Atkinson, R. M. (2000). “Large Water Distribution
Systems Design Through Genetic Algorithm Optimisation.” Proceedings of the ASCE Joint Conference
on Water Resources Engineering and Water Resources Planning and Management, American Society of
Civil Engineers, Hotchkiss, R. H., and Glade, M., eds., proceedings published on CD, Minneapolis,
Minnesota.
Savic, D. A., Walters, G. A., and Schwab, M. (1997). “Multiobjective Genetic Algorithms for Pump Sched-
uling in Water Supply.” AISB, Lecture Notes in Computer Science 1305, Springer Verlag.
Sekhar, W. W., and Uber, J. G. (2001). “Field Study of Chlorine Decay and Disinfection Byproduct Forma-
tion in Water Distribution Systems.” Proceedings of the World Water and Environmental Resources
Congress, The Environmental & Water Resources Institute, Reston, Virginia.
Simons, R. (1996). “Simulation and Optimisation Join Forces to Schedule London’s Water.” The Newslet-
ter of Mathematical Programming in Industry and Commerce, http://www.eudoxus.com/mpac9601.pdf.
Stathis, J. A., and Loganathan, G. V. (1999). “Analysis of Pressure-Dependent Leakage in Water Distribu-
tion Systems.” Proceedings of the ASCE Water Resources Planning and Management Division Confer-
ence, Tempe, AZ.
Tang, K. W., Brunone, B., Karney, B., and Rossetti, A. (2000). “Role and Characterization of Leaks Under
Transient Conditions.” Proceedings of the ASCE Joint Conference on Water Resources Engineering and
Water Resources Management, Minneapolis, Minnesota.
Tarquin, A., and Dowdy, J. (1989). “Optimal Pump Operation in Water Distribution.” Journal of Hydraulic
TREEO (1985). “Operations and Training Manual on Energy Efficiency in Water and Wastewater Treat-
ment Plants.” TREEO Center, University of Florida, 1986.
Tryby, M. E., Boccelli, D. L., Koechling, M. T., Uber, J. G., Summers, R. S., and Rossman, L. A. (1999).
“Booster Chlorination for Managing Disinfectant Residuals.” Journal of the American Water Works
Association, 91(1), 86.
Tryby, M. E., and Uber, J. G. (1999). “Development of a Booster Chlorination Design Using Distribution
System Models.” Proceedings of the 26th Water Resources Planning and Management Conference,
American Society of Civil Engineers, Tempe, Arizona.
U. S. Environmental Protection Agency (1998). Stage 1 Disinfectants and Disinfection Byproducts Rule.
Office of Water, http://www.epa.gov/safewater/mdbp/dbp1.html.
U. S. Environmental Protection Agency (2000). Stage 2 Microbial/Dininfection Byproducts Federal Advi-

sory Committee. Office of Water, http:// http://www.epa.gov/safewater/mdbp/st2faca.html.
Vandermeyden, C., and Hartman, D. (2001). “Water Quality Modeling of a Complete Distribution System.
Development, Results, and Practical Applications.” Proceedings of the AWWA Annual Conference,
American Water Works Association, Denver, Colorado.
Vasconcelos, J. J., Boulos, P. F., Grayman, W. M., Kiene, L., Wable, O., Biswas, P., Bhari, A., Rossman, L.
A., Clark, R. M., and Goodrich, J. (1996). Characterization and Modeling of Chlorine Decay in Distri-
bution Systems, AWWA Research Foundation, Denver, Colorado.
Waller, K., Swan, S. H., DeLorenze, G., Hopkins, B. (1998). “Trihalomethanes in Drinking Water and
Spontaneous Abortion.” Epidemiology, 9(1), 134.
Walski, T. M. (1991). “Understanding Solids Transport in Water Distribution Systems.” Water Quality
Modeling in Distribution Systems, AWWARF, Denver, Colorado.
Walski, T. M. (1993). “Tips for Energy Savings in Pumping Operations.” Journal of the American Water
Walski, T. M. (1995). “An Approach for Handling Sprinklers, Hydrants, and Orifices in Water Distribution
Systems.” Proceedings of the AWWA Annual Conference, American Water Works Association, Ana-
heim, California.
Walski, T. M. (1998). “Importance and Accuracy of Node Elevation Data.” Essential Hydraulics and
Hydrology, Haestad Press, Waterbury, Connecticut.
Walski, T. M. (1999). “Modeling TCE Dynamics in Water Distribution Tanks.” Proceedings of the 26th
Water Resources Planning and Management Conference, ASCE, Tempe, Arizona.
Walski, T. M. (2000). “Water Quality Aspects of Construction and Operation.” Water Distribution System
Handbook, Mays, L. W., ed., McGraw-Hill, New York, New York.
Walski, T. M., and Draus, S. J. (1996). “Predicting Water Quality Changes During Flushing.” Proceedings
of the AWWA Annual Conference, American Water Works Association, Toronto, Canada.
Wang, Z., Polycarpou, M. M., Shang, F., Uber, J. G. (2001). “Design of Feedback Control Algorithm for
Chlorine Residual Maintenance in Water Distribution Systems.” Proceedings of the World Water and
Environmental Resources Congress, The Environmental & Water Resources Institute, Reston, Virginia.
Wu, Z. Y., Boulos, P. F., Orr, C. H., and Moore, M. (2001). “Using Genetic Algorithm for Water Distribu-
tion System Optimization.” Proceedings of the World Water & Environmental Resources Congress,
Orlando, Florida.
References 473
Wu, Z. Y., Boulos, P. F., Orr, C. H., and Ro, J. J. (2000). “An Efficient Genetic Algorithms Approach to an
Intelligent Decision Support System for Water Distribution Networks.” Proceedings of the Hydroinfor-
matics Conference, Iowa City, Iowa.
Yu, G., Powell, R. S., and Sterling, M. J. H. (1994). “Optimized Pump Scheduling in Water Distribution
Systems.” Journal of Optimization Theory Applications, 83(3), 463.
Zessler, M. L., and Shamir, U. (1989). “Optimal Operation of Water Distribution Systems.” Journal of
Water Resources Planning and Management, ASCE, 115(8), 735.
10.1 Complete Problem 4.4 and add

controls to the Newtown pump sta-
tion so that the pump runs from
midnight to 4:30 a.m., is off from
4:30 a.m. to 11:00 a.m., and then
runs from 11:00 a.m. to midnight.
a) Produce a plot of HGL versus

time for the Central Tank.
b) Produce a plot of discharge ver-

sus time for the Newtown pump
station.
c) Does the system work properly

under these controls? If not,
what is the problem with the
system operation?
10.2 Complete Problem 4.4 and add the following pump station controls:
• The Newtown pump station enters service when the water elevation in the Central Tank
falls below 1,515 ft, and turns off when the water level in the tank reaches 1,535 ft.
• The High Field pump station enters service when the HGL in the tank falls below
1,510 ft, and turns off when the water elevation reaches 1,530 ft.
Initially, the Newtown pump station is running and the High Field pump station is off.
a) Produce a plot of HGL versus time for the Central Tank.
b) Produce a plot of discharge versus time for the Newtown pump station.
c) Is it necessary to pump from the High Field reservoir to fill the tank under this typical day
demand condition?
10.3 Complete Problem 10.2 and add a fire flow of 2,250 gpm to node J-5. The fire demand begins at
11:00 a.m. and lasts for three hours.
a) Plot HGL versus time for the Central Tank.
b) Plot discharge versus time for the Newtown pump station.
c) Plot discharge versus time for the High Field pump station.
d) Plot pressure versus time for node J-5.
e) Are pressures maintained within acceptable limits at J-5? Throughout the entire system?
f) Once the High Field pump station activates, how many hours does it take for the Central Tank to
recover to its initial elevation at the start of the simulation (1,525 ft)?
10.4 The system shown in the figure is the same system given in Problem 4.3. However, a PRV has been
added to pipe P-6, and the status of pipe P-14 is closed. Note that adding a PRV to pipe P-6 will
necessitate splitting the line into two pipes. The setting for the PRV (elevation = 1,180 ft) in pipe P-6
is 74 psi. For this simulation, assume that both pumps are running.
Perform a fire flow analysis to find the maximum flow, in addition to the original demands, that can
be supplied at each node in the system [that is, the fire flow will only be applied to one node at a time
(in addition to the original demands), while the other nodes maintain their original demands]. The
minimum allowable pressure anywhere in the system (other than on the suction side of a pump) is 30
psi.
Hint: If you are using WaterCAD, you can use the automated fire flow feature to automatically per-
form this analysis for all nodes in a single model run.
Lower Pressure Zone Newtown Res.

J-10 P-16 J-9 P-15 J-8
P-1
PRV-1 P-14
P-17 P-19
PMP-1
P-2
J-11 P-18 J-3 P-5 P-4 J-2 P-3
PRV-2 J-1
P-6
Central Tank P-13
J-4 J-7 P-9

Pressure Zone Boundary P-20
P-7
P-12
J-5 P-8 J-6

PMP-2
P-11
Higher Pressure Zone
P-10
High Field Res.
a) Complete the table below.

Available Fire Flow Residual Pressure at Flowed Node
Node Label
(gpm) (psi)
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
J-11
b) Do you think this network is capable of meeting the fire flow requirements of a system that sup-
plies water to a residential community and an industrial park? Assume the needed fire flow is
3,500 gpm.
c) How many fire hydrants have to be opened to deliver the available fire flow at node J-6? Assume
that each fire hydrant can be represented by the system shown in Figure 10.17 and the hydrant
opening size is 2 1/2 in. Also, recall that the HGL of the pseudo-reservoir must be set equal to the
elevation of the node of interest, in this case, node J-6.
10.5 Given the layout of the system in the figure and the data below (and in file Prob10-05.wcd), examine
the tradeoffs between the two possible routes for a pipeline connecting the pump station with tank R-
2. One route runs three miles along a highway with high backfilling and restoration costs. The alter-
native route is 3,000 ft longer, but the pipe can be laid in the shoulder of a back road at a lower cost.
R -2 P -4
PMP -1 Highway J -2
P -1 P -2
J -1
R -1 P -6 P -7 Backroad
PMP -2
Node Elevation Demand

Label (ft) (gpm)
J-1 980 500
J-2 950 1,200
R-1 1,000 N/A
R-2 1,100 N/A
Pump Curve Data
Maximum Maximum
Shutoff Design Design
Pump Operating Operating
Head Head Discharge
Label Head Discharge
(ft) (ft) (gpm)
(ft) (gpm)
PMP-1 160 140 2,000 100 3,000
PMP-2 160 140 2,000 100 3,000

Pipe Label Material
(ft) (in.) C-factor
P-1 40 12 Ductile Iron 130
With only one pump at the source running, determine whether to use a 12-, 14- or 16-in. pipe, and
which route (highway or back road) is preferable. Information on energy and construction costs is
provided in the following paragraph.
The fraction of time during which the pump is running can be calculated as
f = Qdemand / Qpump
where Qdemand = total demand in the system (gpm)
Qpump = calculated pump flow rate (gpm)
For this system, the energy cost can then be calculated as
Cenergy = 0.22 u Qpump u Hpump u f

where Cenergy = energy cost ($/year)
Hpump = calculated pump head (ft)
The present worth (PW, in $) of the energy cost over the next 20 years, assuming 7 percent interest
per year, is given by
PW = 10.6 x Cenergy
Use the values in the table below to compute construction costs for the pipe.
Pipe Size Cost (in highway) Cost (in back road)
(in.) ($/ft) ($/ft)
12 90 60
14 95 65
16 100 70
Fill in the table below to give the construction cost and present worth of energy costs for each sce-
nario.
Total
Pipe Present Worth Construction
Flow Head Present
Size Route f of Energy Cost Cost
(gpm) (ft) Worth
(in.) ($) ($)
($)
12 Back road
14 Back road
16 Back road
12 Highway
14 Highway
16 Highway
Which scenario is the most economical for the 20-year period being considered?
10.6 English Units: Find fractions of the water coming from the Miamisburg tank and the West Carrolton
tank for the system presented in Problem 4.2. Add the following pump controls before conducting
the simulation. The pump is initially on, but turns off at hour 8 of the simulation, and then back on at
hour 16. Assume that the initial fraction at each junction node is zero. Complete the following table
for conditions at hour 12 of the simulation.
Fraction of Water from Fraction of Water from West

Node Label
Miamisburg Tank Carrolton Tank
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
SI Units: Find the fractions of water from the Miamisburg tank and the West Carrolton tank for the
system presented in Problem 4.2 (SI). Add pump controls before conducting the simulation. The
pump is initially on, but turns off at hour 8 of the simulation, and back on at hour 16. Assume that
the initial fraction at each junction node is zero. Complete the following table for conditions at hour
12 of the simulation.
Fraction of Water from Fraction of Water from West
Node Label
Miamisburg Tank Carrolton Tank
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
10.7 English Units: Determine the age of water in the system presented in Problem 4.2. Add pump con-
trols before conducting the simulation. The pump is initially on but turns off at hour 5 of the simula-
tion and back on at hour 12. Complete the following table for conditions at hour 12 of the
simulation. The initial age of the water at each node is given in the table.
Initial Age Average Age

Node Label
(hr) (hr)
J-1 0.0
J-2 0.0
J-3 0.0
J-4 0.0
J-5 0.0
J-6 0.0
J-7 0.0
J-8 0.0
J-9 0.0
J-10 0.0
PMP-1 0.0
Miamisburg Tank 15.0
West Carrolton Tank 7.0
SI Units: Determine the age of water in the system presented in Problem 4.2 (SI). Add pump
controls before conducting the simulation. The pump is initially on but turns off at hour 5 of the
simulation and back on at hour 12. Complete the following table for conditions at hour 12 of the
simulation. The initial age of the water at each node is given in the table.
Initial Age Average Age
Node Label
(hr) (hr)
J-1 0.0
J-2 0.0
J-3 0.0
J-4 0.0
J-5 0.0
J-6 0.0
J-7 0.0
J-8 0.0
J-9 0.0
J-10 0.0
PMP-1 0.0
Miamisburg Tank 15.0
West Carrolton Tank 7.0
10.8 English Units: Determine the concentrations of chlorine for the system presented in Problem 4.2.
For this constituent, the bulk reaction rate is –2.6 day-1, the wall reaction coefficient is –1.25 ft/day,
and the diffusivity is 1.3 x 10-8 ft2/s. The initial chlorine concentrations are 1.0 mg/l at each junction
node, 2.3 mg/l at Crystal Lake reservoir, 0.7 mg/l at the Miamisburg Tank, and 0.9 mg/l at the West
Carrolton Tank. Perform a 72-hour extended-period simulation with the diurnal demands repeating
each day. Complete the following table.
Chlorine Concentration at 60 hr

Node Label
(mg/l)
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
Miamisburg Tank
West Carrolton Tank
SI Units: Determine the concentrations of chlorine for the system presented in Problem 4.2 (SI). For
this constituent, the bulk reaction rate is –2.6 day-1, the wall reaction coefficient is –0.38 m/day, and
the diffusivity is 1.2 x 10-9m2/s. The initial chlorine concentrations are 1.0 mg/l at each junction node,
2.3 mg/l at Crystal Lake reservoir, 0.7 mg/l at the Miamisburg Tank, and 0.9 mg/l at the West Carrol-
ton Tank. Perform a 72-hour extended-period simulation with the diurnal demands repeating each
day. Complete the following table.
Chlorine Concentration at 60 hr
Node Label
(mg/l)
J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
Miamisburg Tank
West Carrolton Tank
10.9 English Units: Given the network shown in the figure and the model data given in the following
tables or saved in file Prob10-09.wcd, determine whether complaints of low pressure at node J-7 are
due to limited capacity in the system or the elevation of the customer. Maintain a 35-psi residual
pressure even during peak (nonfire) demands. To check the cause of the problem either
• Set up model runs with demand multipliers of 0.6, 1.0, and 1.33 to determine if the
pressure drops; or
• Set up an EPS run for a 24-hour period.
a) How much did the pressure and HGL change at J-7 from peak to low demand times? If an EPS
run was made, what was the range of water level in the West Side Tank?
b) Did the HGL (pressure) change markedly when demands changed, or was the pressure relatively
insensitive to demands?
c) Was the pressure highly dependent on tank water level?
d) How many pumps need to be run to meet demands over the course of the day?
e) What recommendations would you make to improve pressures?
Clearwell P3
P3 -SUC
P2 -SUC
P3-DIS
P2
P1-SUC
P2 -DIS
J -1 P -11
P1
P1 -DIS
P-1
J -8 P -9 J -7
J -3 P -2 J -2
P -8
P -3 P-6 P-10
J -4 P -4 P -7 J -6
J -5
P -5
West Side Tank


Pipe Label
(ft) (in.) C-factor
P1-Suc 50 18 115
P1-Dis 120 16 115
P2-Suc 50 18 115
P2-Dis 120 16 115
P3-Suc 50 18 115
P3-Dis 120 16 115
P-1 2,350 12 110
P-2 1,500 6 105
P-3 1,240 6 105
P-4 1,625 12 110
P-5 225 10 110
P-6 1,500 12 110
P-7 4,230 6 105
P-8 3,350 6 105
P-9 2,500 6 105
P-10 2,550 6 105
P-11 3,300 4 85
Elevation Demand
Node Label
(ft) (gpm)
J-1 730 0
J-2 755 125
J-3 765 50
J-4 775 25
J-5 770 30
J-6 790 220
J-7 810 80
J-8 795 320
Elevation
Reservoir Label
(ft)
Clearwell 630
Minimum Maximum
Initial Elevation Tank Diameter
Tank Label Elevation Elevation
(ft) (ft)
(ft) (ft)
West Side Tank 900.0 917.0 920.0 50.0
Pump Curve Data
P1 P2 P3
Head Flow Head Flow Head Flow Head
(ft) (gpm) (ft) (gpm) (ft) (gpm) (ft)
Shutoff 305 0 305 0 305 0
Design 295 450 295 450 295 450
Max 260 650 260 650 260 650
Operating
Hydraulic Pattern (Continuous)

Time
Multiplier
(hr)
0 0.60
3 0.75
6 1.20
9 0.90
12 1.15
15 1.00
18 1.33
21 0.90
24 0.60
Controls for Pump
P1 P2 P3
Initial Setting ON ON OFF
ON if West Side Below 916.0 ft Below 913.0 ft Below 905.0 ft
Tank is
OFF if West Side Above 919.5 ft Above 917.0 ft Above 914.0 ft
Tank is
SI Units: Given the network shown in the figure and the model data given in the following tables or
saved in file Prob10-09m.wcd, determine whether complaints of low pressure at node J-7 are due to
limited capacity in the system or the elevation of the customer. Maintain a 240 kPa residual pressure
even during peak (non-fire) demands. To check the cause of the problem either:
• set up model runs with demand multipliers of 0.6, 1.0, and 1.33 to determine if the
pressure drops; or
• set up an EPS run for a 24-hour period.
a) How much did the pressure and HGL change at J-7 from peak to low demand times? If an EPS
run was made, what was the range of water level in the West Side Tank?
b) Did the HGL (pressure) change markedly when demands changed, or was the pressure relatively
insensitive to demands?
c) Was the pressure highly dependent on tank water level?
d) How many pumps need to be run to meet demands over the course of the day?
e) What recommendations would you make to improve pressures?


Pipe Label
(m) (mm) C-factor
P1-SUC 15.2 457 115
P1-DIS 36.6 406 115
P2-SUC 15.2 457 115
P2-DIS 36.6 406 115
P3-SUC 15.2 457 115
P3-DIS 36.6 406 115
P-1 716.3 305 110
P-2 457.2 152 105
P-3 378.0 152 105
P-4 495.3 305 110
P-5 68.6 254 110
P-6 457.2 305 110
P-7 1289.3 152 105
P-8 1021.1 152 105
P-9 762.0 152 105
P-10 777.0 152 105
P-11 1,006 102 85
Elevation Demand
Node Label
(m) (l/s)
J-1 222.50 0.0
J-2 230.12 7.9
J-3 233.17 3.2
J-4 236.22 1.6
J-5 234.70 1.9
J-6 240.79 13.9
J-7 246.89 5.0
J-8 242.32 20.2
Elevation
Reservoir Label
(m)
Clearwell 192.0
Maximum Minimum
(m) (m)
(m) (m)
West Side Tank 280.42 279.50 274.32 15.24
Pump Curve Data

Maximum Maximum
Shutoff Design
(m) Head Discharge
(m) (l/s)
(m) (l/s)
P1 93.0 89.9 28.4 79.3 41.0
P2 93.0 89.9 28.4 79.3 41.0
P3 89.9 77.7 15.8 48.8 28.4
Hydraulic Pattern (Continuous)
Time
Multiplier
(hr)
0 0.60
3 0.75
6 1.20
9 0.90
12 1.15
15 1.00
18 1.33
21 0.90
24 0.60
Controls for Pump

P1 P2 P3
Initial Setting ON ON OFF
ON if West Side Below 279.2 m Below 278.3 m Below 275.8 m
Tank is
OFF if West Side Above 280.3 m Above 279.5 m Above 278.6 m
Tank is
10.10 English Units: This problem uses the calibrated model from Problem 10.9. A customer near J-8
having an elevation of 760 ft complains of low water pressure. A pressure of 46 psi was measured
at a hydrant near J-8. Is the problem in the distribution system or in the customer's plumbing?
a) Make an EPS run of the model and plot pressure at J-8 versus time. Does a pressure of 46 psi
look reasonable in that part of the system?
b) To get an idea of the cause of the problem, look at the HGL when the pressure is at its worst. Is
it very different (>20 ft) from the HGL at the tank? What does that tell you about head loss?
c) To determine the source of the head loss, look at the velocities in the system at times of low
pressure. Which pipes have the highest velocities?
d) Is there much that can be done operationally to correct the problem?
e) Is 46 psi a low pressure?

SI Units: This problem uses the calibrated model from Problem 10.9 (SI). A customer near J-8
having an elevation 231.65 m complains of low water pressure. A pressure of 317 kPa was mea-
sured at a hydrant near J-8. Is the problem in the distribution system or in the customer's plumb-
ing?
a) Make an EPS run of the model and plot pressure at J-8 versus time. Does a pressure of 317 kPa
look reasonable in that part of the system?
b) To get an idea of the cause of the problem, look at the HGL when the pressure is at its worst. Is
it very different (>6.1 m) from the HGL at the tank? What does that tell you about head loss?
c) To determine the source of the head loss, look at the velocities in the system at times of low
pressure. Which pipes have the highest velocities?
d) Is there much that can be done operationally to correct the problem?
e) Is 317 kPa a low pressure?
10.11 A customer near J-6 in the figure complains of low pressure, and pressure chart recorder data was
collected as shown in the chart. Construct a model using the data tables that follow, or open
Prob10-11.wcd.
a) Determine if the low pressure is due to elevation, inadequate pump capacity, undersized piping,
or some large demand.
b) How would you use a model to confirm this problem? Perform the simulation to confirm it.
J-9 J-10
Crystal Lake P -15
P -16 P -14
PMP-1
Suction
J-1 J-8 J-7
Discharge
P -12 P -13
P -1 P -11 P -10
J-2 J-3 J-6
P -2 P -9
P -3
J-4
Miamisburg
Tank P -4
P -5
J-5
P -6 P -7
33
32
31
Pressure, psi
30
29
28
27
26
0 2 4 6 8 10 12 14 16 18 20 22 24
Time, hr
Elevation Demand
Node Label
(ft) (gpm)
J-1 390.00 120
J-2 420.00 75
J-3 425.00 35
J-4 430.00 50
J-5 450.00 0
J-6 485.00 155
J-7 420.00 65
J-8 415.00 0
J-9 420.00 55
J-10 420.00 20
Length Diameter
Pipe Label
(ft) (in.)
Discharge 220 21
Suction 25 24
P-1 1,250 6
P-2 835 6
P-3 550 8
P-4 1,010 6
P-5 425 8
P-6 990 8
P-7 2,100 8
P-9 745 8
P-10 1,100 10
P-11 1,330 8
P-12 890 10
P-13 825 10
P-14 450 6
P-15 690 6
P-16 500 6
Pump Curve Data
Maximum Maximum
Shutoff Design
(ft) Head Discharge
(ft) (gpm)
(ft) (gpm)
PMP-1 245 230.0 1,100 210.0 1,600
Elevation
Reservoir Label
(ft)
Crystal Lake 320.0
Minimum Maximum
Initial HGL Tank Diameter
(ft) (ft)
(ft) (ft)
Miamisburg Tank 535.0 550.0 570.0 50.0
Hydraulic Pattern (Stepwise)
Time
Multiplier
(hr)
0 1.00
6 0.75
12 2.00
18 1.20
24 1.00
10.12 If customers near J-5 in the figure complain about having low pressures between 2:00 p.m. and
5:00 p.m. each day (hours 14 to 17 of the simulation), determine if their complaints are due to the
utility’s system. If so, use an EPS model run to formulate alternatives to improve the system. The
customer has an alarm on a fire protection system that sounds when the pressure drops below 37
psi. (This network can be found in Prob10-12.wcd.)
Lake Anderson
Suction P-15 J-10

Lag - suction J-9
PMP -1
Discharge P-16
PMP -2
Lag - discharge J-1 P-14
P-12
J-8
P-13
P-1 J-7 East Tank
P-11
J-2 P-10
P-2
J-3
P-9
P-3 J-6 P-8
P-4 J-4
P-5
J-5
P-7
Elevation Demand
Node Label
(ft) (gpm)
J-1 390 120
J-2 420 75
J-3 425 35
J-4 430 50
J-5 460 20
J-6 445 155
J-7 420 65
J-8 415 0
J-9 420 55
J-10 420 20

Pipe Label
(in.) (ft) C-factor
P-1 6 1,250 110
P-2 6 835 110
P-3 8 550 130
P-4 6 1,010 110
P-5 8 425 130
P-7 8 2,100 105
P-8 12 560 110
P-9 8 745 100
P-10 10 1,100 115
P-11 8 1,330 110
P-12 10 890 115
P-13 10 825 115
P-14 6 450 120
P-15 6 690 120
P-16 6 500 120
Suction 24 25 120
Lag-discharge 21 220 120
Lag-suction 24 25 120
Pump Curve Data
Maximum Maximum
Shutoff Design
(ft) Head Discharge
(ft) (gpm)
(ft) (gpm)
PMP-1 245 230 600 210 1,000
PMP-2 245 230 600 210 1,000
Controls for Pump
PMP-1 PMP-2
Initial Setting ON OFF
ON if East Tank Below 560 ft Below 545 ft
OFF if East Tank Above 564 ft Above 555 ft
Maximum Minimum
(ft) (ft)
(ft) (ft)
East Tank 565.0 553.0 525.0 54
Elevation
Reservoir Label
(ft)
Lake Anderson 320.0
Hydraulic Pattern (Stepwise)
Time
Multiplier
(hr)
0 1
6 0.75
12 2.0
18 1.2
24 1
10.13 Start with the same network used in Problem 10.12. Determine what kind of fire flows can be
delivered with a 20 psi residual at every node at hour 8, first with the tank on-line, then with the
tank off-line. Find the same information during peak demands at hour 12.
Hint: To use WaterCAD’s automated fire flow feature for this problem, check the boundary and
operational conditions in Problem 10.12 at the specified times. Create a steady-state run for these
conditions, and set up and run the fire flow analysis.
Fire Flows at 20 psi at hour 8
Node Label Tank On-line Tank Off-line

J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
Fire Flows at 20 psi at hour 12
Node Label Tank On-line Tank Off-line

J-1
J-2
J-3
J-4
J-5
J-6
J-7
J-8
J-9
J-10
10.14 Using the same model from Problem 10.12, determine how long it will be before the system
reaches full recovery when there is a power outage from 2:00 pm to 8:00 pm, if hour 0 in the
model represents midnight. Full recovery means that the tank level reaches the same level as
before the power outage. With the data from Problem 10.12, simulate a power outage by control-
ling the open/closed status of the discharge pipes of the pumps at hour 14 and hour 20.
10.15 English Units: Given the existing small system shown in the figure, perform three runs to deter-
mine the impact on the pressure at node J-8 (highest elevation customer) if the pump is allowed to
come on when the tank drops to levels of 25, 20, or 15 ft (relative to tank base elevation). Use the
stepwise demand patterns shown in the chart. (This network can be found in Prob10-15.wcd.)
T-1
J-8 P-15
P-17
J-3 J-20
P-40
J-9
P-16
R-1
P-53 P-18
P-50 P-31 P-41
Suction Discharge J-1 P-67
J-16
Pump J-26 P-51 P-52 J-10
P-26
P-66
P-22
J-14
J-27 P-24
P-54 J-13
P-55
J-12
P-28
P-59
P-56
P-58 J-29
J-15 P-57
J-28
Elevation Demand
Node Label Pattern
(ft) (gpm)
J-1 25.00 100 Residential
J-8 90.00 80 Commercial
Length Diameter
Pipe Label
(ft) (in.)
Discharge 38 12
Suction 33 12
P-15 970 12
P-16 850 12
P-17 955 12
P-18 905 6
P-22 1,145 12
P-24 1,195 12
P-26 1,185 4
P-28 1,215 4
P-31 1,023 8
P-40 570 6
P-41 645 6
P-50 1,080 6
P-51 870 6
P-52 630 6
P-53 585 6
P-54 360 6
P-55 370 6
P-56 540 6
P-57 400 6
P-58 320 6
P-59 560 6
P-66 956 12
P-67 570 6
Pump Curve Data
Maximum Maximum
Shutoff Design
(ft) Head Discharge
(ft) (gpm)
(ft) (gpm)
PUMP 210 160 600 100 900
Controls for Pumps
PUMP
Initial Setting ON
ON if T-1 Below 175.0 ft
OFF if T-1 Above 179.5 ft
Base Maximum Initial Minimum Tank

Tank Label Elevation Elevation Elevation Elevation Diameter
(ft) (ft) (ft) (ft) (ft)
West Tank 150.0 180.0 176.0 150.0 40.0
Elevation
Reservoir Label
(ft)
R-1 20.00
Stepwise Pattern
Residential
1.4
1.2
1.0
M ultiplie r
0.8
0.6
0.4
0.2
0.0
0.0 2.0 4.0 6.0 8.0 10.0 12.0
Time, hr
Stepwise Pattern
Commercial
1.4
1.2
1.0
M ultiplie r
0.8
0.6
0.4
0.2
0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
Time, hr
a) To observe the impact on J-8, prepare a plot of pressure versus time at node J-8 for each pump
control scenario.
b) Make a 480-hr (20-day) water quality run and graph the average water age in the tank versus
time for each operating alternative.
c) Based on your calculations, revise the initial water age in the tank to 60 hours and rerun the
scenarios. Plot the minimum pressure at J-8 vs. average water age for each scenario, and con-
sider the water pressure/quality trade-offs.
d) How do you recommend this system be operated?

SI Units: Given the existing small system shown in the figure, perform three runs to determine the
impact on the pressure at node J-8 (highest elevation customer) if the pump is allowed to come on
when the tank drops to levels of 7.6, 6.1, or 4.6 m (relative to tank base elevation). Use the step-
wise demand patterns shown in the chart. (This network can also be found in Prob10-15m.wcd.)
Elevation Demand
Node Label Pattern
(m) (l/s)
J-1 7.62 6.31 Residential
J-8 27.43 5.05 Commercial
Length Diameter
Pipe Label
(m) (mm)
Suction 9.8 305
Discharge 11.9 305
P-15 295.7 305
P-16 259.1 305
P-17 291.1 305
P-18 275.8 152
P-22 349.0 305
P-24 364.2 305
P-26 361.2 102
P-28 370.3 102
P-31 311.8 203
P-40 173.7 152
P-41 196.6 152
P-50 329.2 152
P-51 265.2 152
P-52 192.0 152
P-53 178.3 152
P-54 109.7 152
P-55 112.8 152
P-56 164.6 152
P-57 121.9 152
P-58 97.5 152
P-59 170.7 152
P-66 291.4 305
P-67 173.7 152
Pump Curve Data
Maximum Maximum
Shutoff Design
(m) Head Discharge
(m) (l/s)
(m) (l/s)
PUMP 64.0 48.8 37.85 30.5 56.8
Controls for Pumps
PUMP
(levels)
Initial Setting ON
ON if T-1 Below 7.62 m
OFF if T-1 Above 8.99 m
Base Maximum Initial Minimum Tank

Tank
Elevation Elevation HGL Elevation Diameter
Label
(m) (m) (m) (m) (m)
T-1 45.72 54.86 53.34 45.72 12.2
Elevation
Reservoir Label
(m)
R-1 6.10
a) To observe the impact on J-8, prepare a plot of pressure versus time at node J-8 for each pump
control scenario.
b) Make a 480-hr (20-day) water quality run and graph the average water age in the tank versus
time for each operating alternative.
c) Based on your calculations, revise the initial water age in the tank to 60 hours and rerun the
scenarios. Plot the minimum pressure at J-8 vs. average water age for each scenario, and con-
sider the water pressure/quality tradeoffs.
d) How do you recommend this system be operated?

ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
11
Water System Security
The security of water systems has long been a concern in the water industry. The
potential for natural, accidental, and purposeful contamination or other events that
would hinder the ability of the system to provide a safe water supply has been the sub-
ject of many studies. In May 1998, President Clinton issued Presidential Decision
Directive (PDD) 63, which outlined a policy on critical infrastructure protection,
including our nation’s water supplies. However, it wasn’t until the terrorist events of
September 11, 2001, that the water industry truly focused on the vulnerability of our
nation’s water supplies to terrorist activities.
11.1 WATER SYSTEM VULNERABILITY

Because water systems are spatially diverse (see Figure 11.1), they are inherently vul-
nerable to a variety of activities that can compromise the system’s ability to reliably
deliver sufficient water at an acceptable level of quality. There are several areas of
vulnerability as water travels to the customer. These areas include (1) the raw water
source (surface or groundwater); (2) raw water canals and pipelines; (3) raw water
reservoirs; (4) the treatment facilities; (5) connections to the distribution system
pipes; (6) pump stations and valves; and (7) finished water tanks and reservoirs. Each
of these system elements presents unique challenges to the water utility in safeguard-
ing the water supply. These challenges include
• Physical disruption that prevents sufficient water flow at an acceptable pres-
sure to all customers
• Contamination of the water delivered to the customer by a chemical or bio-
logical agent such that the product is not safe to use or is not of an acceptable
quality to the customer
• Loss of confidence by customers in the ability of the water utility to deliver a
safe and secure water supply
500 Water System Security Chapter 11
Figure 11.1
Major points of
vulnerability in a
Reservoir
water supply system
Tanks
Canals &
Pumps
Pipes
Wells
Distribution System
River
Water systems are vulnerable to natural, purposeful, or accidental events that can
challenge the security of the system. Examples of each of these types of events
include the following:
• Natural events: Floods, earthquakes, fire, severe weather (droughts, hurri-
canes, tornadoes, and so on), sinkholes, and natural contamination of surface
or groundwater sources
• Purposeful events: Terrorist or criminal contamination, vandalism, or
destruction
• Accidental events: Accidental discharges of pollutants into source waters,
cross-connections between the water distribution system and waste collec-
tion system, vehicle and pipeline accidents, and explosions
11.2 POTENTIAL WATER SECURITY EVENTS
Physical Disruption
The ability of a water supply system to provide water to its customers can be compro-
mised by damage or destruction of key physical elements of the water system. Key
elements include raw water facilities (for example, dams, reservoirs, pipelines, and
canals), treatment facilities, and finished water elements (such as transmission lines
and pump stations).
In general, physical disruption can result in disruption of service, significant eco-
nomic cost, inconvenience, and loss of confidence by customers, but the direct threat
to human health is generally limited. Exceptions to this generalization include (1) the
destruction of a dam that causes loss of life and property in the accompanying flood
wave and (2) an explosive release of chlorine gas at a treatment plant that produces a
cloud injurious or lethal to nearby populations.
Section 11.2 Potential Water Security Events 501
Some Prominent Historic Events

in Water System Security
Bibilical times – The Nile River is turned to 1980s – Lawsuit and trial in case of industrial
“blood” in the first plague in the Book of Exodus contamination in Woburn, Massachusetts
requiring Egyptians to turn to wells as an alter- (USA) was subsequently profiled in the book
nate water supply. and movie A Civil Action.
19th Century – Cholera outbreaks in London 1993 – Flooding in Iowa (USA) disrupted water
led to the identification of water supply as the service to 250,000 customers of the Des
major culprit by Dr. John Snow. Moines Water Works.
1860s – During the Civil War, soldiers shot and 1993 – 400,000 people in Milwaukee (USA)
left farm animals in ponds to poison the water became ill from Cryptosporidiosis carried in the
supply so that advancing troops couldn’t use it. public water supply.
1941 – J. Edgar Hoover, director of the FBI, 1998 – President Clinton issued Presidential
acknowledged that “water supply facilities offer Decision Directive 63, which designated water
a particularly vulnerable point of attack to the systems as one of the country’s critical infra-
foreign agent.” structures and outlined a policy for protecting it.
1940s – During World War II, water supplies September 11, 2001 – Attack on the World
were purposely contaminated in China (bacte- Trade Center and Pentagon demonstrated con-
ria) and Bohemia (sewage). clusively that the United States is vulnerable to
1978 – Carbon Tetrachloride spill on the terrorist attacks.
Kanawha River resulted in contamination of
water supplies on the Ohio River and led to
establishment of an early warning system.
Water utilities should examine their physical assets, determine their areas of vulnera-
bility, and increase security accordingly. For example, switching from chlorine gas to
liquid hypochlorite, especially in less secure locations, decreases the risk of exposure
to poisonous chlorine gas. Also, redundant system components can limit disruption of
service by providing backup capability in case of accidental or purposeful damage to
facilities.
Contamination
The ability to deliver water of an acceptable quality can be compromised by the pres-
ence of contaminants in the raw or finished water supply. Contaminants can enter the
water supply through natural causes, accidental spills or events, or purposeful terrorist
or criminal acts. As illustrated in Figure 11.1, there are many potential locations
where a contaminant can enter the water system. Locations include raw water sources
(surface or groundwater), raw water delivery systems, the treatment facility, or the
actual distribution system. Water treatment is the primary barrier to contaminants
reaching the customer but may not be effective for some constituents found in the raw
water and is not effective for contaminants that enter the finished water in the distribu-
tion system. Maintenance of a secondary disinfectant residual in a distribution system
provides protection from some bacterial agents but is not effective for all constituents.
Contamination has long been viewed as a serious potential terrorist threat to water
systems (Hoover, 1941). Chemical or biological agents could spread throughout a dis-
tribution system and result in sickness or death among people drinking the water. For
some agents, the presence of the contaminant would not be known until emergency
rooms reported an increase in patients with a particular set of symptoms. Deliberate
chemical and biological contamination of water supply systems is nothing new (Hick-
man, 1999; Deininger, 2000; Clark and Deininger, 2000). Such terrorist activities
were reported in ancient Rome (cyanide), in the United States during the Civil War
(animal carcasses in farm ponds), in Europe and Asia during World War II (anthrax,
cholera, and sewage), and, more recently, in Kosovo (paints, oil, and gasoline in
wells). Deininger and Meier (2000) discuss the topic of deliberate sabotage of water
supply systems.
Accidental contamination of water systems has resulted in many fatalities and ill-
nesses, as well. Examples of such outbreaks include (but are not limited to) E. coli
contamination in Walkerton, Ontario (Canada) (Haestad Methods, 2002), cholera
contamination in Peru (Craun et al., 1991), cryptosporidium contamination in Mil-
waukee, Wisconsin (USA) (Fox and Lytle, 1996), and salmonella contamination in
Gideon, Missouri (USA) (Clark et al., 1996).
The U.S. Army has compiled information on potential biological agents (Burrows and
Renner, 1998). Table 11.1 summarizes this information on agents that may have an
impact on water systems.
There are many factors that should be considered in evaluating the potential threat
from different chemical and biological agents. Following is a summary of some of
these factors:
• Availability: Is the agent readily available or difficult to obtain?

• Monitoring response: Can the agent be detected by monitoring equipment?
• Physical appearance: Is there a telltale odor, color, or taste associated with
the agent?
• Dosage/health effects: What dosage is required to have effects on human
health?
• Chemical and physical stability in water: How long will the agent be sta-
ble in water?
• Tolerance to chlorine: Are chlorine or other disinfectants effective in neu-
tralizing the agent?
Deininger and Meier (2000) ranked various agents and compounds in terms of their
relative factor of effectiveness, R, based on lethality and solubility using the following
equation:
R = solubility in water (in mg/L) / (1000 x lethal dose (in mg/human)) (11.1)
Table 11.2 lists values of R for various biological agents and chemicals by decreasing
level of effectiveness (that is, decreasing degree of toxicity in water).
Table 11.1 Potential threat of biological weapons agents

Agent Type Weaponized Water Threat Stable in Water Chlorine1 Tolerance
Anthrax Bacteria Yes Yes 2 years (spores) Spores resistant
Brucellosis Bacteria Yes Probable 20-72 days Unknown
C. perfringens Bacteria Probable Probable Common in sewage Resistant
Tularemia Bacteria Yes Yes Up to 90 days Inactivated, 1 pm, 5 min
Glanders Bacteria Probable Unlikely Up to 30 days Unknown
Meliodosis Bacteria Possible Unlikely Unknown Unknown
Shigellosis Bacteria Unknown Yes 2-3 days Inactivated, 0.05 ppm, 10 min
Cholera Bacteria Unknown Yes ‘Survives well’ ‘Easily killed’
Salmonella Bacteria Unknown Yes 8 days, fresh water Inactivated
Plague Bacteria Probable Yes 16 days Unknown
Q Fever Rickettsia 2
Yes Possible Unknown Unknown
Typhus Rickettsia Probable Unlikely Unknown Unknown
Psittacosis Rickettsia-like Possible Possible 18-24 hours, seawater Unknown
Encephalomyelitis Virus Probable Unlikely Unknown Unknown
Hemorrhagic fever Virus Probable Unlikely Unknown Unknown
Variola Virus Possible Possible Unknow Unknown
Hepatitis A Virus Unknown Yes Unknown Inactivated, 0.4 ppm, 30 min
Cryptosporidiosis Protozoan 3
Unknown Yes Stable days or more Oocysts resistant
Botulinum toxins Biotoxin4 Yes Yes Stable Inactivated, 6 ppm, 20 min
T-2 mycotoxin Biotoxin Probable Yes Stable Resistant
Aflatoxin Biotoxin Yes Yes Probably stable Probably tolerant
Ricin Biotoxin Yes Yes Unknown Resistant at 10 ppm
Staph enterotoxins Biotoxin Probable Yes Probably stable Unknown
Microcystins Biotoxin Possible Yes Probably stable Resistant at 100 ppm
Anatoxin A Biotoxin Unknown Probable Inactivated in days Unknown
Tetrodotoxin Biotoxin Possible Yes Unknown Inactivated, 50 ppm
Saxitoxin Biotoxin Possible Yes Stable Resistant at 10 ppm
Based on Burrows and Renner (1998)
1 – Ambient temperature, <1 ppm free available chlorine, 30 minutes, or as indicated
2 – Parasites that are pathogens from humans and animals
3 – Consisting of one cell or of a colony of like or similar cells
4 – Toxic to humans
Source Water Contamination. Contamination of source water is of signifi-

cant concern because of the many potential locations where contaminants can enter
surface water or groundwater sources and the difficulty in providing security over an
entire source water area. Fortunately, a contaminant entering source water is subject
to dilution, chemical reactions, exposure to sunlight, and treatment, and thus the con-
centration may be significantly reduced by the time it enters a water distribution sys-
tem. Contamination of source waters may be caused by natural hydrologic processes,
accidental releases of contaminants, or purposeful dumping of pollutants into the
water.
Table 11.2 Relative toxicity of poisons in water

Compound R
Botulinus toxin A 10,000
VX 300
Sarin 100
Nicotine 20
Colchicine 12
Cyanide 9
Amiton 5
Fluoroethanol, sodium, fluoroacetate 1
Selentie 1
Arsenite, arsenate 1
Based on Deininger and Meier (2000)
The primary mechanism for identifying and responding to source water contamina-
tion is an early warning system. An early warning system is a combination of equip-
ment and institutional arrangements and policies that are used in an integrated manner
to achieve the goal of identifying and responding to contaminants in the source water.
An effective early warning system includes the following components (Grayman,
Deininger, and Males, 2002):
• A mechanism for detecting the likely presence of a contaminant in the

source water
• A means of confirming the presence of the contamination, determining the
nature of the contamination event, and predicting when (and for how long)
the contamination will affect the source water at the intake sites and the
intensity (concentration) of the contamination at the intake
• An institutional framework generally composed of a centralized administra-
tive unit that coordinates the efforts associated with managing the contami-
nation event
• Communication linkages for transferring information related to the contami-
nation event
• Various mechanisms for responding to the presence of contamination in the
source water in order to mitigate its impacts on water users
The central component of an early warning system is the detection of a contaminant

event so that mitigative actions can be taken if needed. The three basic mechanisms
for detecting spills and other contaminant events are monitoring, self-reporting by the
facility causing the spill, and siting and reporting an event by the public or outside
groups. An effective early warning system generally combines all three mechanisms.
A wide range of online monitoring equipment is available that can be used as part of
an early warning system (AwwaRF, 2002). Conventional sensors, which include dis-
solved oxygen, pH, conductivity, temperature, and turbidity, are relatively inexpen-
sive, widely available, and easily used, but they provide little useful information in
identifying the presence of most transient contaminants. More advanced online moni-
tors, such as gas chromatographs and spectrophotometers are more expensive and
require greater expertise and maintenance, but they are more effective as part of an
early warning system.
Examination of online monitors around the world has shown that given sufficient
resources, almost all analyses can be automated if there is a perceived need. Biomon-
itors utilize living aquatic organisms (such as fish, mussels, daphnia, and bacteria) and
measure the stresses placed on the organisms by contaminants in the water. They do
not provide information on the specific contaminant but rather “raise a red flag” that
there is something unusual in the water that is affecting the organisms used in the
biomonitor. Although they have been in use now in many places for almost 20 years,
the field of biomonitoring still appears to be an emerging technology. Other emerging
technologies that may have future utility in early warning systems are generally being
developed in other fields and include electronic noses, DNA chips, flow cytometry,
immuno-magnetic separation techniques, and online bacteria monitors. The current
state of the art in identification of microbial contamination in drinking water has been
summarized by Sobsey (1999).
If the presence of a contaminant is identified in source water, several responses can be
taken to mitigate the impacts of a spill event:
• Closure of water intakes and use of alternate sources
• Cleanup of the spill prior to impacting water intakes
• Enhanced temporary chemical treatment at the treatment plant
• Public notification
Closure of water intakes provides the most absolute barrier to a contaminant impact-
ing a drinking water supply but is effective only if the presence of a contaminant is
identified prior to entering the water intake and is limited by the length of time that a
water utility can close their intake and still provide a sufficient supply of water.
Cleanup within the source water body itself is practical for only a limited range of
pollutants (such as some petroleum products, which can be physically separated from
the water body). Enhanced chemical treatment can include addition of coagulants,
carbon, disinfectant, or other chemicals. In order to be effective, the chemical nature
of the contamination must be known and the proper chemical dosage determined.
If contamination has not been identified in the source water, the only line of defense is
the normal treatment processes that are online at the time of the event. Table 11.3 con-
tains typical contaminant removal ranges for various treatment types and contaminant
categories. Actual removal rates depend on specific designs, the chemical characteris-
tics of the raw water, temperature, and so on. Other treatment options that provide
additional barriers to contaminants reaching the water user include regular use of
granular activated carbon, and groundwater injection and subsequent harvesting of the
water. Another mechanism, raw water storage, provides a time lag between the water
intake and water treatment, thus potentially allowing for additional testing prior to the
use of the water.
Table 11.3 Effectiveness of processes for contaminant removal

Taste &
Bacteria Viruses Protozoa VOC SOC TOC
Odor
Aeration, air P P P G-E P-F F F-E
stripping
Coagulation, G-E G-E G-E G-E P P-G P-G
sediment/filtration
Lime softening G-E G-E G-E P-F P-F G P-F
Ion exchange P P P P P G-E -
Reverse osmosis E E E F-E F-E G -
Ultra filtration E E E F-E F-E G -
Disinfection E E E P-G P-G G-E P-E
Granular F F F F-E F-E F G-E
activated carbon
Powdered P P P P-G P-E F-G G-E
activated carbon
UV irradiation E E E G G G G
Based on AWWA (1990)
Abbreviations: P – Poor (0–20% removal); F – Fair (20–60% removal); G – Good (60–90% removal); E – Excellent (90–100% removal);
- Insufficient Data; VOC – volatile organic chemicals; SOC - synthetic organic chemical; TOC - total organic carbon
Contamination of Distribution System. Water distribution systems are

usually closed systems that are generally more secure than raw water sources. How-
ever, the implications of contamination directly into the distribution system are poten-
tially much more severe for the following reasons:
• No treatment exists to provide a barrier (other than residual disinfectant)
• Monitoring within a distribution system is generally quite limited
• The travel time between a point of contamination in the distribution system

and customers can be quite short
• If a contaminant is not detected before it reaches a distribution system stor-

age facility, contaminated water may reach many customers over an
extended period of time
Potential points of contamination within the distribution system include the follow-
ing:
• Water treatment plant: Treatment plants are the interface between the raw
water source and the distribution system. They are especially vulnerable
because the entire water supply goes through the plant. Clearwells at the end
of the treatment should receive special attention. In the clearwells, a large
amount of supply is concentrated in a single location. A contaminant added
at that point can impact the entire water supply leaving the plant for a period
of many hours. Treatment plants generally have some form of security, and
large plants have around the clock personnel.
• Pump stations and valves: These components are potential points of con-
tamination because a large amount of flow may be passing through these
points at any given time. Thus, any contaminant injected at these locations
could impact many downstream customers.
• Finished tanks and reservoirs: Tanks and reservoirs are vulnerable because
they contain a relatively large quantity of water at a single location, are fre-
quently located in isolated locations, are generally maintained at atmo-
spheric pressure, and usually have access to the interior for maintenance
purposes. Open reservoirs are especially vulnerable because of the open
access to the contents, but they are also relatively rare. Because of the quan-
tity of water in storage in a tank or reservoir, a significant amount of contam-
inant is needed to impact the water quality. Once a sufficient amount of
contaminant is added and mixed with the water in the tank, the water quality
in the distribution system fed by the tank can be impacted for many days
until the fill and draw process has sufficiently diluted the contaminant. This
process is shown graphically in Figure 11.2, where it would take approxi-
mately 12 days for the concentration to be reduced by 90 percent from its
initial peak concentration.
• Hydrants: Hydrants are located on relatively major lines throughout water

systems providing fire protection. There are no recorded instances of terror-
ists pumping contaminated water into the system through a hydrant, but
there have been cases of accidental backflow from tankers that were being
filled directly from hydrants.
Figure 11.2
Decline in
concentration of a 0.30
well-mixed
contaminated tank
0.25
Calculated Concentration, mg/l
0.20
0.15
0.10
0.05
0.00
0 48 96 144 192 240 288
Time, hr
• Distribution system connections: A pump that is capable of overcoming

the system pressure could inject contaminants into the system affecting
nearby customers or larger areas of the distribution system depending on the
location and amount of water pumped into the system.
11.3 ASSESSMENT OF VULNERABILITY

The purposes of a vulnerability study or assessment are
• To examine facilities to determine their vulnerability to various events that
could threaten their ability to achieve their specific operational purposes
• To assess and prioritize the risks associated with the facilities and operations
of the water system
• To develop a program to reduce risks and respond in case of events that
could threaten the water system
Formal vulnerability assessments have been applied to facilities that have been recog-
nized as vulnerable to terrorist or natural events that could result in significant conse-
quences. Examples of such facilities include nuclear power plants, military bases, and
federal dams. Transfer of such methods to water systems is a recent and ongoing
development (Sandia National Laboratories, 2001). Such assessments can employ
formalized information collection and analysis tools, or more simply, a set of check-
lists, protocols, and procedures.
Section 11.3 Assessment of Vulnerability 509
Water System Vulnerability Checklist

SOURCE WATER STORAGE
Wells • Is the structure covered, fenced, locked, well
• Are the wells covered, preferably in perma- lit, with an area clear of vegetation and obsta-
nent, secured structures? cles?
• Are the wellheads covered and locked? • Can the structure be isolated from the sys-
tem?
• Are the wells with vents not easily accessi-
ble? • Are manholes/access ports and vents locked
and secured?
• Is the area fenced, well lit, and clear of vege-
tation and obstalces?
DISTRIBUTION
Surface Sources • Is the system above ground or below ground?
• Are intakes accessible? If so, can their vul- Can it be tampered with?
nerability be reduced? • Are there valve pits on the system? Are the
facilities fenced, locked, and secured? Who
Local Supplies has access to them?
• Who operates and maintains them? • Are pits located upstream from critical facili-
• Are there controlling valves between the sys- ties?
tems? Who controls them?
• Do they have an adequate vulnerability PERSONNEL
reduction program?
• Who has access to any components of the
water system?
Alternate Emergency Sources Identified
• Are such personnel reliable? Have they been
• Is bottled water available? Does it meet pre-
ventive medicine standards? checked out?
• How are keys and combination locks secured
• Is there an adequate supply? Is it secure?
and managed?
TREATMENT
SECURITY PATROLS
• How is water disinfected? Is chlorine used?
• Are they regularly conducted? If so, how
• Are chlorine levels checked throughout the often and on what sites? Is this sufficient?
system, by whom, and how often?
• Is there remote monitoring of critical points
• What other chemicals are added? and/or processes? Is it necessary?
• How and by whom are the process and the
concentration monitored?
SAMPLING/DETECTION
• Could treatment process(es) be modified to
• How often are chlorine residual measure-
prevent injection of unwanted material?
ments done and where?
• Is the facility secured, is it monitored, how
• What type of monitoring capability exists?
often?
• Are there adequate contingency plans?
• Are the grounds fenced, locked, with an area
clear of vegetation and obstacles, and is it • Do medical facilities identify and track a
well lit? waterborne epidemic?
(Hickman, 1999)
Whichever approach is adopted, the vulnerability analysis can be framed in the con-
text of managing the risks within the water system. The basic framework for consider-
ing risk analysis consists of the interwoven areas of risk assessment and risk
management. Risk analysis recognizes that it is impossible to eliminate all risks but
that risks can be identified, assessed, managed, and balanced with other resources so
that the resulting risks are within an acceptable level. Quite simply, risk analysis can
be embodied in a series of questions, such as the following.
Assessment
• What can go wrong?
• How can it happen?

• How likely is it to happen?
• What are the consequences?
Management
• What questions should the risk assessment answer?
• What can be done to reduce the impact of the risk described?
• What can be done to reduce the likelihood of the risk described?
• What are the trade-offs of the available options?

• What is the best way to address the described risk?
Inspections and Checklists

In conducting a vulnerability analysis, water utilities can develop a set of inspection
protocols and a checklist of items. The inspection protocol includes items such as
examination of design plans, field inspection of the facilities, and review of the opera-
tional and maintenance procedures within the water utility.
Formal Assessment Tools and Methods

Vulnerability assessments of water systems can involve a large number of facilities
and a wide range of potential risks and solutions. There are various methods and tools
available to help a water utility to organize, manage, and assess such information.
Examples of such tools include fault trees, event trees, decision trees, Monte Carlo
simulations, and computer simulations.
Fault Trees. Sandia National Laboratories in conjunction with the EPA and
AwwaRF developed a water system vulnerability assessment training package. This
stemmed from a performance-based vulnerability assessment methodology initially
developed by Sandia for the nuclear security area and later applied to security of fed-
eral dams. The central component of this analysis is a risk assessment process shown
schematically in Figure 11.3.
Section 11.3 Assessment of Vulnerability 511
Figure 11.3
Risk assessment
process applied to
Characterize
Context water systems (Sandia
Assets
National Laboratories,
2001)
Determine
Prioritize Targets
Consequence
Define
Protection Goals
Threats
Define
Protection Systems
Safeguards
Analyze
System Risks
No Risks Yes
Proposed Upgrades Acceptable END
?
One of the tools applied in the Sandia assessment is a fault tree. Figure 11.4 illustrates
an example of a generic fault tree applied to a water system. A fault tree is a top-down
method of analyzing system design and performance. The top event is an undesired
state of the system—in this case, defined as an occurrence that will result in the water
utility being unable to meets its mission of supplying safe water to its customers. In
the second level, specific credible ways that could result in not meeting this mission
are enumerated, such as interruption of the raw water supply, treatment capability, or
distribution.
The fault tree continues down with events being combined either through AND or OR
gates. An AND gate indicates that multiple events must occur simultaneously for the
higher level event to occur while OR gates indicate that either of two or more events
could lead to the higher level event occurring. The lowest level of detail depicted in
the graphical fault tree is referred to as a basic event. In the example shown, the basic
events are referred to as “undeveloped” events because they may be broken down into
more detail. (That level of detail is not shown in the diagram.)
The fault tree can be viewed as a graphical depiction of events that would lead to the
overall system failure. Alternatively, it may take on more of a quantitative tool by
assigning probabilities of occurrence to the basic events and then, by following the
tree upward, calculating the overall probability of the top event occurring.
Figure 11.4
Sample fault tree for
water system (Sandia
National Laboratories,
2001)
Monte Carlo Simulation. Monte Carlo simulation is a well-known technique

for analyzing complex physical systems where probabilistic behavior is important.
Grayman and Males (2001) developed a Monte Carlo simulation of a source water
early warning system to test the effectiveness of alternative designs. Spill events are
represented as probabilistic occurrences (that is, probability distribution of stream-
flows, probabilities of spills of different substances, magnitudes and duration, and so
on). The relationships of these events (for example, how a river responds to a spill) are
embodied in the model, which is then run many times, with varying inputs based on
the probabilities of the events. This model was applied to the Ohio River to study the
effectiveness of alternative early warning system designs (for example, monitor loca-
tions, monitoring frequency, response policies, and so on) on the resulting water qual-
ity of the finished water supply.
Computer Simulation Models. Simulation models are another tool that can
be used in vulnerability studies to help a water utility understand how their system
will respond to an accidental or purposeful physical or chemical event. This under-
standing can be used to identify the consequences of such events, to test solutions to
minimize the impacts of the events, or to learn how to respond if such events occur.
Simulation models are discussed in more detail in the following section.
11.4 APPLICATION OF SIMULATION MODELS

Models are representations of systems that are especially effective in examining the
consequences of “what if” scenarios. Within the context of water system security,
some examples of “what if” scenarios include the following.
Section 11.4 Application of Simulation Models 513
• If an oil tank adjacent to a river ruptures and discharges to a river used down-
stream as a source of raw water, when should the utility close its water intake
and for how long will they need to keep the intake closed?
• If a major main in the water system breaks, what happens to pressure
throughout the distribution system and will there be sufficient flow and pres-
sure to provide fire protection?
• If runoff contaminates a particular well, what customers would receive con-
taminated water and how quickly will the contaminant reach them?
• If a terrorist manages to dump a barrel of a particular chemical into a fin-
ished water tank, how will the chemical mix within the tank and, if the con-
tamination is not discovered, which customers will receive the contaminated
water, when will they receive it, and what will the concentration be?
In the area of water system security, computer models have been used to examine
three different time frames:
• As a planning tool to look at what may happen in the future in order to assess
the vulnerability of a system to different types of events and to plan how to
respond if such an event occurs
• As a real-time tool for use during an actual event to assist in formulating a

response to the situation
• As a tool for investigating a past event so as to understand what happened
The characteristics of models used and the type of information that is available in
these three time frames can vary significantly. The following sections discuss this in
the context of three types of computers models: water distribution system models,
tank and reservoir mixing models, and surface water hydraulic and water quality
models.
Water Distribution System Models

A water distribution system model can be used to simulate flows and pressures within
a distribution system, and the movement and transformation of a constituent after it is
introduced into the distribution system. Previous chapters of this book discuss the use
of water distribution system models. In this section, we will examine in greater detail
the specific cases of modeling contaminants that have entered a distribution system by
accidental or purposeful causes.
In order to simulate the movement of a contaminant in a distribution system, a
hydraulic extended-period simulation (EPS) model of the system is needed. The
model should reflect the hydraulic conditions of concern. In using the model in a vul-
nerability study, it would be appropriate to select several normal operating conditions
such as a typical summer day, a typical winter day, and a typical spring/autumn day. It
is impossible to simulate all possible conditions. In addition, selecting typical operat-
ing conditions provides information reflective of most situations rather than an
extreme demand day that represents conditions for only a few days per year.
A contaminant is represented in the model by describing the transformation character-
istics of the constituent in the distribution system, where it is introduced into the sys-
tem, and the time history of the amount of the constituent that is introduced. Most
modern distribution system models provide options for designating this information.
Typically, a constituent is represented as being a conservative substance, which means
that it does not change concentration unless dilution occurs, or a non-conservative
substance, which means that it follows some form of decay (for example, first order
exponential decay). For the purposes of a vulnerability study, it is generally assumed
that the constituent will be conservative.
Most modern distribution system models provide the user with several options for
introducing a contaminant at a source (see page 465 for more information). In each
case, a time-varying pattern can be applied to the source. These options include the
following:
• Concentration: A concentration constituent source fixes the concentration
of any external inflow entering the network at a node, such as flow from a
reservoir or from a source placed at a junction.
• Flow-Paced Booster: A flow-paced booster constituent source adds a fixed
concentration to the flow resulting after the mixing of all inflow to the node
from other points in the network.
• Setpoint Booster: A setpoint booster constituent source fixes the concentra-

tion of any flow leaving the source node, as long as the concentration result-
ing from the inflow to the node is below the set point.
• Mass Booster: A mass booster constituent source adds a fixed mass flow to
the flow entering the node from other points in the network.
• Initial Concentration: The initial concentration in a tank may be set with
the concentration changing over time due to either decay or dilution during
the fill cycle.
These options for specifying source concentrations allow the user to select the method
that most accurately represents the physical contamination event that he or she is sim-
ulating.
Use as a Planning Model. A water distribution system model can be applied
to a wide range of “what if” scenarios to determine the general vulnerability of the
distribution system. For example, the model can be used to determine the effects of a
major pipe break or the impacts of a purposeful or accidental contamination of the
system. With this information in hand, a water utility is better equipped to develop an
effective plan of action.
An example of the results of the application of a distribution system model to repre-
sent contamination tracking in the distribution system is shown in Figure 11.5. This
example illustrates the case where a contaminant has been pumped directly into the
distribution system at a constant flow and concentration rate over a 24-hour period.
The figure shows how the contaminant spreads over a significant portion of the sys-
tem during the 24-hour period.
An important part of any emergency planning is the simulation of a possible emer-
gency before it occurs. Using a model to simulate emergency drills with operators is
another facet of such preparedness. An operator can be given a contamination sce-
nario and asked to respond. The operator’s response can be simulated in the model,
showing where the contaminated plume would move and how the actions affected the
movement of the plume and exposure for customers.
Historical Modeling. What caused a sudden outbreak of debilitating and fatal

cases of diarrhea in Gideon, Missouri, in December 1993? Was an elevated number of
serious childhood illnesses in Dover Township, New Jersey, caused by industrial con-
tamination of the water supply dating back more than 40 years? These are the types of
questions that have been investigated by historical modeling (also referred to as retro-
spective or forensic modeling) using water distribution system models (see page 459).
The disease outbreak in Gideon was identified as salmonellosis and a private tank was
suspected as a source of the bacteria (Clark et al., 1996). The U.S. EPA applied a
water distribution system model of the Gideon system to investigate possible scenar-
ios by which the bacteria could propagate through the system and infect the custom-
ers. Ultimately, it was determined that it was likely that the tank, which was in bad
repair, was infected by birds. A sudden drop in temperature resulted in a temperature
inversion in the tank which mixed infected bird droppings and feathers throughout the
tank, and resulted in taste and odor problems. An aggressive flushing program then
drew the contaminated water deep into the distribution system. Similar historical
modeling has been done in the investigation of E. coli outbreaks in Cabool, Missouri,
and Walkerton, Ontario (Haestad Methods, 2002).
Figure 11.5
Example movement
of a contaminant
through a distribution
system
Improper handling and disposal of industrial chemicals led to the contamination of

some groundwater in the United States during the 20th century. In many cases, the
contaminated groundwater was used as a source of drinking water and, as a result,
customers were exposed to elevated levels of the contaminants. Many of these cases
have resulted in legal actions and governmental studies of the impacts of the contami-
nants on the population exposed to the substances. Distribution system hydraulic and
water quality models have played a key role in many of these cases in determining the
likely movement of the contaminants through the distribution system. Because some
of these incidents date back many years (for example, the 1950s through the 1980s),
the development of models and reconstruction of the operation of the water system
was required. Information on the model development and reconstruction process is
limited because of non-disclosure requirements associated with many legal cases.
A contamination case in Woburn, Massachusetts, has been documented in the book
and movie, A Civil Action (Harr, 1995), and was the subject of an early modeling
study in the 1980s (Murphy, 1986). Industrial sources were found to have improperly
discarded chemicals that seeped into the groundwater and contaminated two wells.
A Checklist of Security Measures

SHORT TERM • Limit access to water supply reservoirs.
• At your office, well houses, treatment plants, • Develop a clear policy so all employees know
and vaults, make it a rule that doors are how to deal with trespassers.
locked and alarms are set. • Fit hydrants and valve boxes with tamper
• Make security a priority, and emphasize it at proof caps and lids.
employee and safety meetings. • Install pass-code locks instead of keyed locks
• Tell your employees to ask questions and so access numbers can be changed as nec-
make note of strangers who are in your facili- essary—for example, when an employee is
ties or call in threats. terminated. (Make sure, however, that emer-
• Limit access to facilities. Indicate restricted gency response personnel and law enforce-
areas by posting Employees Only signs. ment personnel receive the updated access
codes.)
• Increase lighting in parking lots, treatment
bays, and other areas that seldom have peo- • Fence and lock vulnerable areas such as
ple present. wellheads, reservoir vents, and meter pits.
• DO NOT leave keys in equipment of vehicles • Mark equipment with logos and distinctive
at any time. paint jobs.
• Invite local law enforcement to become famil- • Integrate early warning monitoring systems
iar with your facilities, and establish a proto- into water transport, treatment, and distribu-
col for reporting and responding to threats. tion systems so that an operator will be noti-
fied immediately of changes in chemical
• Discuss detection, response, and notification characteristics, flows, pressures, and temper-
issues with public health officials, and estab-
ature.
lish protocols for them.
• Design and install valves that open and close
• Establish a chain of command and an emer- slowly.
gency call list to be used in emergency situa-
tions. • Know your employees and who it is you are
hiring. Make it a standard procedure to con-
• Provide copies of your operational proce- duct background checks on all new employ-
dures to local law enforcement and emer-
ees.
gency management personnel.
• Install firewalls in computer systems, and
LONG TERM change access codes frequently.
• Install motion sensors and video cameras to • Conduct and attend training activities to pre-
monitor, detect, and record events. They can pare your staff to detect, delay, and respond
be tied into a supervisory control and data appropriately.
acquisition system for remote monitoring. (Denileon, 2001)
• Install intrusion alarms that cover remote
buildings and grounds.
A series of steady-state distribution system models were used to identify areas of

Woburn that received the contaminated water under different well operating patterns
and demand conditions.
A groundwater contamination case in Phoenix and Scottsdale, Arizona, involving the
chemical trichloroethylene (TCE) was studied using extended-period simulations of
the operation of the water systems and the movement of water through those distribu-
tion systems. Most notable in this study was the use of long-term continuous simula-
tions (multiple years) of the hydraulics and water quality in the distribution system
(Harding and Walski, 2000).
A recently completed detailed study of the water system in the Dover Township, New
Jersey, area by the U.S. Agency for Toxic Substances and Disease Registry identified
the paths between wells and customers over a multi-decade period (ATSDR, 2001). In
this study, a water distribution system model of the present day system was first
developed and calibrated. Subsequently, models of the distribution system were
developed that covered the period from 1962 to 1996. The models were used to trace
the percentage of water reaching nodes from the wells serving the system.
In all historical reconstruction modeling, the challenge is to utilize historical data to

determine the characteristics and operation of the water system during the period of
interest. Generally, the challenge increases as one models further back in time. One of
the key challenges is reconstructing the actual operation of pumps when records are
incomplete. For example, a typical well might be operated to discharge 100,000 gal-
lons over the course of a day (69 gpm, 4.4 l/s), but the well pump discharges 120 gpm
(7.6 l/s). This would mean that the well pump is only operating 57 percent of the time,
but to which hours does that correspond? The resolution of this problem makes a
great difference in determining which customers were exposed to which sources.
Real-Time Modeling. Suppose that the manager of a water system receives a

call from the police that an individual has been apprehended for dumping a poisonous
chemical into the water system. The location, chemical, and approximate time and
duration of the contamination are known. The first action the manager takes, of
course, is to notify the public. Next the manager needs to determine how to operate
the system to flush out the contaminant; that is, he needs to determine which hydrants
to open and how long to keep them open. The manager has a good idea of how water
moves through the system during a normal day, but he also knows the flushing pro-
gram could drastically change the normal flow patterns. Attempting to clean the sys-
tem by trial-and-error would be a long, risky, and uncertain proposition. The best and
easiest way to analyze the problem is to use a properly developed and maintained
water distribution system model.
Water distribution system models have been proposed as part of a real-time or near
real-time system to assist in many aspects of the operation of a water system includ-
ing energy management, water quality management, and emergency operation. The
major obstacle in such use of water distribution system models is the requirement that
the model must be calibrated for a wide range of conditions and is ready to apply
quickly and easily in an extended-period simulation mode. Information on the current
state of the system must be readily available to the model through direct ties to a
SCADA system. In addition, the model must be set up in an automated mode so that
operation is represented by a series of logical controls that reflect the existing operat-
ing procedures. Both information requirements are feasible based on existing technol-
ogy but there have been only limited demonstrations of this type of operation to date.
The key to using a model as part of a real-time response lies in having the model
ready to run. During an emergency, there is no time to construct a model. There is
only time to make some minor adjustments to an existing model.
Section 11.5 Security Measures 519
Tank and Reservoir Mixing Models

Distribution system models represent mixing in tanks and reservoirs using simplified,
hypothetical representations such as complete and instantaneous mixing, plug flow, or
by a last in-first out “short-circuiting” model (see page 356). Though this has proved
adequate for most planning and operational situations, a more accurate representation
of how the facility mixes may be needed when planning for emergency contamination
events.
Computational fluid dynamics (CFD) models (see page 358) use mathematical equa-
tions to simulate flow patterns, heat transfer, and chemical reactions and thus provide
a much truer picture of the actual mixing processes in a tank. The use of CFD models
has grown significantly in the drinking water industry in the past several years (Gray-
man et al., 2000), and the technology has been applied in many planning and design
studies to assess the mixing characteristics of a tank and its inlet-outlet configuration.
Several commercial CFD software packages are available. Significant experience is
required to apply CFD models, and model run times of many hours, days, or even
weeks are required for complex situations. In conducting a vulnerability study, knowl-
edge of the mixing characteristics in a tank is useful in assessing the likely impacts of
a contaminant being added to a storage facility.
Surface Water Hydraulic and Water Quality Models

Hydraulic and water quality models of surface water systems can be used to study the
movement of contaminants in a surface water body. This information is useful in
assessing the vulnerability of a water intake to contaminants that are accidentally or
purposely added to the water body. Such models are also useful as part of an early
warning system to predict the real-time movement of contaminants that have been
detected upstream of a water intake. There are many general purpose hydraulic/water
quality models and specially designed “spill models” that can be used in both vulner-
ability studies and in real-time prediction (Grayman, Deininger, and Males, 2000).
11.5 SECURITY MEASURES

Many security measures are available that can be applied to decrease the vulnerability
of a distribution system to purposeful (and accidental) events that threaten the ability
of a water utility to provide a safe water supply to its customers. Some of these mea-
sures can be applied quickly with minimal costs, and others may take significant time
and resources to apply. The applicability of specific measures varies from water sys-
tem to water system. In evaluating potential security measures, a water utility should
balance the reduction of risks with the costs of implementing the measures.
The following list describes a variety of potential security measures.
• Maintain a significant disinfectant residual. As illustrated in Table 11.1,
many of the potential biological agents are inactivated by exposure to chlo-
rine. The inactivation effects of other disinfectants (such as chloramines) are
not as well known. Actions that can be taken relative to maintaining an
acceptable disinfectant residual include (1) placing continuous chlorine

monitors throughout the system to report chlorine residual back to a central
control center and warn of low residuals; (2) increasing the chlorine dose at
the treatment plant during periods of higher alert (although this may result in
undesirable higher levels of disinfectant by-products); (3) adding booster
chlorination stations at locations in the distribution system that routinely
experience low disinfectant residuals; and (4) modifying operating policies
to reduce water age in the system, including changing the fill and draw pat-
terns for tanks.
References 521
• Increase the security surrounding key facilities in a water system. Dein-

inger and Meier (2000) recommend the following to increase the security of
a water system:
The intakes, pumping stations, treatment plants, and reservoirs should be fenced to secure
them against casual vandalism. Beyond that, there should be intrusion alarms that notify the
operator that an individual has entered a restricted area. An immediate response may be to
shut down part of the pumping system until the appropriate authorities determine that there
is no threat to the system. In underground reservoirs, the ventilation devices must be con-
structed in such a way as to not allow a person to pour a liquid into the reservoir. An above
ground reservoir with roof hatches should not have ladders on it that allow climbing. In addi-
tion, the hatches should be secured.
• Install secure backflow preventers or check valves at potential injection
sites. A contaminant can be injected at any connection to a water system if a
pump that is capable of overcoming the system pressure is available. Back-
flow preventers provide an obstacle and deterrent to such action, but in order
to be effective, the backflow preventer must be installed so that it cannot be
disengaged easily. Maintenance and cost issues associated with widespread
installation of such devices should be considered when evaluating this
option.
• Develop an early warning system for the raw water supply. If not
detected, contaminants in source waters (surface and groundwater) can pass
through a treatment plant and enter the distribution system. An early warning
system is a combination of equipment and institutional arrangements and
policies that are used to detect and respond to contaminants in the source
water.
• Install continuous monitors at key locations in the distribution system.
Monitoring can provide a means of identifying the presence of unwanted
contaminants in the distribution system. In order to be effective as a security
mechanism, monitors that sample continuously or very frequently at key
locations in the distribution system and are tied into a central operations cen-
ter are needed.
• Develop a detailed emergency response plan. An emergency response plan
is like an insurance policy—one hopes that he or she will never have to use it
but is very thankful that a plan is available if there is an emergency. Such a
plan should provide detailed information on how to respond under a wide
range of emergency situations. The plan should be kept up-to-date and per-
sonnel should be familiar with it so that it can be quickly implemented when
needed. The procedures for developing emergency response plans have been
developed by the American Water Works Association (AWWA, 2001).
REFERENCES
Agency for Toxic Substances and Disease Registry (ATSDR). (2001). Summary of Findings. Historical
Reconstruction of the Water-Distribution System Serving the Dover Township Area, New Jersey: Janu-
ary 1962-December 1996. Atlanta, Georgia.
American Water Works Association. (1990). Water Quality & Treatment A Handbook of Community Water
Supplies. 4th Edition, McGraw-Hill, New York, New York.
American Water Works Association (2001). “Emergency Planning for Water Utilities.” Manual of Practice
M-19, Denver, Colorado.
AwwaRF. (2002). Online Monitoring for Drinking Water Utilities. Edited by E. Hargesheimer, Denver, Col-
orado.
Burrows, W. D., and Renner, S. E. (1998). “Biological Warfare Agents as Threats to Potable Water.” Envi-
ronmental Health Perspectives, 107(12), 975.
Clark, R. M., and Deininger, R. A. (2000). “Protecting the Nation’s Critical Infrastructure: The Vulnerabil-
ity of U.S. Water Supply Systems.” Journal of Contingencies and Crisis Management, 8(2), 73.
Clark, R. M., Geldreich, E. E., Fox, K. R., Rice, E. W., Johnson, C. W., Goodrich, J. A., Barnick, J. A., and
Abdesaken, F. (1996). “Tracking a Salmonella Serovar Typhimurium Outbreak in Gideon, Missouri:
Role of Contaminant Propagation Modeling.” Journal Of Water Supply and Technology, 45(4), 171.
Craun, G., Swerdlow, D., Tauxe, R., Clark, R., Fox, K., Geldreich, E., Reasoner, D., and Rice, E. (1991).
“Prevention of Waterborne Cholera in the United States.” Journal of the American Water Works Associ-
ation, 83(11), 43.
Deininger, R. A. (2000). “Constituents of Concern. The Threat of Chemical and Biological Agents to Pub-
lic Water Supply Systems.” Appendix F in Pipeline Net User’s Guide, SAIC, McLean, Virginia.
Deininger, R. A., and Meier, P. G. (2000). “Sabotage of Public Water Supply Systems.” Security of Public
Water Supplies, Deininger, R. A., Literathy, P., and Bartram, J., eds, NATO Science Series 2, Environ-
ment - Volume 66, Kluwer Academic Publishers.
Denileon, G. P. (2001). “The Who, What, Why, and How of Counterterrorism Issues.” Journal of the Amer-
Fox, K. R., and Lytle, D. A. (1996). “Milwaukee’s Crypto Outbreak Investigation and Recommendations.”
Grayman, W. M., Deininger, R. A., and Males, R. M. (2002). Design Of Early Warning and Predictive
Source-Water Monitoring Systems. AWWA Research Foundataion, Denver, Colorado.
Grayman, W. M., and Males, R. M. (2001). “Risk-Based Modeling of Early Warning Systems for Pollution
Accidents.” Proceedings of the IWA 2nd World Water Congress, IWA, London, United Kingdom.
Grayman, W. M., Rossman, L. A., Arnold, C., Deininger, R. A., Smith, C., Smith, J. F., Schnipke, R.
(2000). Water Quality Modeling of Distribution System Storage Facilities. AWWA Research Founda-
tion, Denver, Colorado.
Haestad Methods, Inc. (2002). Proceedings of the Water Security Summit. Haestad Press, Waterbury, Con-
necticut.
Harding, B. L., and Walski, T. M. (2000). “Long Time-Series Simulation of Water Quality in Distribution
Systems.” Journal of Water Resources Planning and Management, ASCE, 126(4), 1199.
Harr, J. (1995). A Civil Action. Vintage Books, New York, New York.
Hickman, D. C. (1999). “A Chemical and Biological Warfare Threat: USAF Water Systems at Risk.” Coun-
terproliferation Paper No. 3, Maxwell Air Force Base, USAF Counterproliferation Center, http://
www.au.af.mil/au/awc/awcgate/cpc-pubs/hickman.pdf.
Hoover, J. E. (1941). “Water Supply Facilities and National Defense.” Journal of the American Water
Works Association, 33(11), 1861-1865.
Murphy, P. J. (1986). Water Distribution in Woburn, Massachusetts. University of Massachusetts, Amherst,
Massachusetts.
Sandia National Laboratories (2001). “Water Infrastructure Security Critical Infrastructure Protection.”
Webcast presented by AwwaRF, http://www.awwarf.com/whatsnew.html.
Sobsey, M. D. (1999). “Methods to Identify and Detect Contaminants in Drinking Water.” Identifying
Future Drinking Water Contaminates, National Academy Press, Washington, D.C.
11.1 The source (node A) in a distribution system has become contaminated and you have been asked to
determine when the contamination will reach a hospital located at node B. You have a steady-state
hydraulic model that has been calibrated for average day conditions that you feel is representative of
the current day.
A
1 2 6 14
3 5 7
4 8
9
11
10 12
B
13
The following table contains information on the network as represented under steady-state average
day conditions. Fill in the values for velocity and travel time in the table.
Diameter Length Discharge Velocity Travel Time
Pipe Label
(in.) (ft) (gpm) (ft/s) (hours)
P-1 8 8,000 500
P-2 8 8,000 300
P-3 8 8,000 200
P-4 8 8,000 180
P-5 6 8,000 50
P-6 8 8,000 240
P-7 8 8,000 40
P-8 8 8,000 220
P-9 8 4,000 240
P-10 6 4,000 120
P-11 6 4,000 120
P-12 6 4,000 100
P-13 6 4,000 100
P-14 8 4,000 200
a) What is the shortest travel time and longest travel time between nodes A and B?
b) Is it reasonable to expect that flows will remain relatively constant over the time it takes for the
contaminant to travel from node A to B?
c) What is likely to happen when the tank switches from the fill cycle represented in the diagram to
a draw cycle?
d) Based on your analysis do you feel that the use of a steady-state model is reasonable or should
you use an EPS hydraulic and water quality model to calculate travel times?
11.2 You are conducting a vulnerability analysis of a water system and want to determine the potential
impacts of a particular toxic substance if it was added to the city’s water supply. The city water sys-
tem serves 100,000 people with a daily average water use of 400 liters per person. Assume that only
1 liter is actually consumed by the average person per day. The average lethal dose for this toxic
contaminant for a person is estimated to be 1 milligram over a course of a day. Assuming that the
substance is evenly mixed throughout the water supply, how much of the substance must be added to
reach the lethal level?
Is it reasonable to assume that the substance would be evenly distributed throughout the system?
What other factors should you consider in making a more accurate assessment of the vulnerability of
the water system to this substance?
11.3 Assume you are given an EPS model of the water distribution system in the following figure and you
are asked to investigate the consequence of contaminants being added at several locations (Prob11-
03.wcd). The model is currently set up to represent average day conditions with a repeating daily
pattern for 288 hours. The system is served by a well (represented as a constant inflow at J-13) and
by a pump from a surface water source (R-1). The pump is controlled by the water level in tank T-1.
Run the model for the following three situations to simulate the movement of the contaminant and
answer the corresponding questions.
a) A conservative substance is dumped into the tank where it mixes completely resulting in an ini-
tial concentration of 100 mg/L. If the substance is not detected and the system is operated nor-
mally, how long will it take for the concentration in the tank to drop by 90% to 10 mg/L? Do all
demand nodes receive contaminated water at some time?
b) A contaminant is added at the well (node J-13) at a high concentration (1,000 mg/L) from hour 6
to hour 7. Does the contaminant reach the tank? How much of the system is affected by the con-
taminant? How long does it take until the entire system is cleared of the contaminant?
c) A contaminant is pumped into the system at node J-5 for a 24-hour period starting at hour 6 at a
rate of 1 gpm with a concentration of 1,000 mg/L. Does the contaminant reach the tank? What is
the maximum concentration in the tank? How many nodes receive a concentration exceeding 1
mg/L at any time during the event? What demand node receives the highest concentration (aside
from node J-5)? When does the concentration drop below 1 mg/L throughout the entire system?
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


12
Integrating GIS and Hydraulic
Modeling
A geographic information system (GIS) is a powerful configuration of computer hard-

ware and software used for compiling, storing, managing, manipulating, analyzing,
and mapping (displaying) spatially-referenced information. It integrates database
operations such as data storage, query, and statistical analysis with visual and geo-
graphic analysis functions enabled by spatial data. A GIS can serve as an integral part
of any project that requires management of large volumes of digital data and the
application of special analytical tools.
GIS is becoming an increasingly valuable tool for the water distribution modeler both
as a source for modeling data and as a decision-support tool. Anderson, Lowry, and
Thomte (2001) reported that although approximately 15 percent of water utilities cur-
rently use GIS in their modeling, almost 80 percent plan to use GIS in the future.
In the past, most models were developed with a batch-run philosophy in which a text
file (or card deck) containing the model data was required to drive the model. Gradu-
ally, models gained the ability to interact with databases that were either internal to
the model or more general and commercial. Finally, because a GIS is essentially a
spatially aware database, ongoing development has led to models that are highly inte-
grated with GIS. Shamsi (2001) describes the evolution of model/GIS integration as a
three-step process:
1. Interchange: Data are exchanged through an intermediary file, which may be an
ASCII text file or a spreadsheet. Data is written to this intermediary file, where it
is reformatted for the model, if necessary, and then read into the model. The
model and GIS are run independently.
2. Interface: Links are built between the model and GIS. These links are used to
synchronize the model and GIS. The data are duplicated on each side of the link
528 Integrating GIS and Hydraulic Modeling Chapter 12
and the model and GIS are run independently. One common approach is the use
of shapefiles, which can pass data between the model and the GIS and optionally
update either based on data contained in the other.
3. Integration: A single repository for the data is used. The model can be run from
the GIS and vice versa.
This integration of the hydraulic model and the GIS leads to the following benefits:
• Time-savings in constructing models
• Ability to integrate disparate land use, demographic, and monitoring data
using GIS analysis tools to more accurately predict future system demands
• Visual, map-based quality control of model inputs

• Map-based display and analysis of model outputs in combination with other
GIS layers
The most powerful feature of a GIS, from a planner's perspective, is probably the abil-
ity of the GIS to integrate, through their spatial relationships, databases that would be
difficult or impossible to integrate outside of a GIS environment. For example, a GIS
can overlay soil data, repair data, and hydraulic modeling output to automatically
assign a condition rating to pipes.
This chapter contains background information on GIS development and uses for both
water model creation and other application areas. The experienced GIS practitioner
with a specific interest in applications of GIS technology to model development may
want to begin with Section 12.3, “Model Construction.” The experienced modeler
with access to GIS, but with no GIS experience, may want to begin with Section 12.1,
“GIS Fundamentals,” and then skip ahead to Section 12.4, “GIS Analysis and Visual-
ization.”
12.1 GIS FUNDAMENTALS

An easy way to think of GIS is to imagine it as a set of transparencies that are layered
in such a manner that any point in one layer would appear at the same location in any
other layer, as shown in Figure 12.1. In an actual GIS graphical user interface (GUI),
these layers appear together, and the user can manipulate the order in which they
appear.
Within a GIS, features (objects on a map) are not simply points and lines; they have
attributes (information about the feature) associated with them. In a water distribution
system, facilities such as pipes, tanks, and pumps are features possessing attributes.
For instance, a pipe is represented in a GIS as a feature, and the diameter of the pipe is
an attribute of that feature.
As shown in Figure 12.1, maps may contain more than one type of feature, each of
which is displayed as a layer on the GIS map. By selecting which layers are dis-
played, the order in which layers are displayed, and the symbology (size, shape, and
color of symbols), the user can control the appearance of the resulting map. Figure
12.2 shows a GIS map.
Section 12.1 GIS Fundamentals 529
Figure 12.1
A conceptual layout
of a GIS
Graphic image courtesy of ESRI
In addition to being used for map-making, a GIS can be used to perform system anal-
ysis, answering questions about:
• Location (using proximity, buffer, or overlay analysis)
• Condition
• Temporal and spatial patterns (trends)
• What-if scenarios (in modeling)
Figure 12.2
A GIS map
Data Management
Two primary and opposing data management paradigms are in use today: centralized
data management and decentralized data management. The mainframe computing
environment is an example of centralized data management, and the PC environment
is an example of decentralized data management. In the mainframe environment, all
applications and data reside on a central server. This data management approach is
very practical, but the hardware and software are typically very expensive to develop
and maintain. Within the PC environment, special-purpose applications and databases
are less expensive than their mainframe counterparts, but data and applications can
reside on different networked PCs. The decentralized management philosophy is
therefore very practical due to the economics involved; however, it does lead to the
creation of “data islands.”
The data developed within these “islands” are often generated and maintained redun-
dantly. For example, the diameter (6 in./150 mm), length (250 ft/76 m), and material
(ductile iron) for a pipe can be entered into the hydraulic modeling application, asset
management system, and maintenance management system, and the pipe can be
drawn and annotated on a map. Very few links to a master database or other data
islands are ever developed. As a result, the utility is unable to tap the knowledge and
efficiency that can be gained by analyzing and acting upon this information centrally.
Unfortunately, the data management situation of many water utilities around the
world is characterized by data islands.
The computer industry has created technology such as SQL and ODBC to assist in
centralizing these data islands. However, because they were designed and developed
Section 12.1 GIS Fundamentals 531
independently, they frequently do not have the key identifiers needed to form mean-
ingful data relationships. For example, the billing system may use account number as
its primary identifier, and the hydraulic modeling software may use pipe and node
numbers. It is therefore difficult to relate the billing information to the hydraulic
model for use in customer service and system planning.
What do the systems and databases for billing, customer service, asset management,
work management, inspections/permits, water quality testing, facility mapping,
hydraulic modeling, and document management have in common? The answer is
geography. All of these systems have information about items (for example, permits,
work orders, test reports) that can be tied to a geographic location such as parcel num-
ber, address, or facility number. Geography is the essence of GIS-centric data man-
agement, a compromise between centralized and decentralized data management;
features that cannot be linked explicitly through database table-to-table relationships
can be associated geographically by determining the proximity (connectivity, distance,
closeness) to each other. This aspect of GIS uniquely qualifies it as the preferred inte-
grating technology and unifying information resource within an organization.
Figure 12.3
Organizing the
desktop with GIS-
centric data
management
The compromise that GIS attains between centralized and decentralized data manage-
ment is this: not all data needs to be centralized; only the GIS layers need to be cen-
tralized. Therefore, the only limitation to the integration of specialized systems into
the GIS-centric model is that they must have some reliable means of geographic refer-
encing, such as facility ID, parcel number, or street address. Virtually any type of sys-
tem or database can be linked to a GIS layer, provided that a geographical association
and consistency and quality of data are present in both systems. Usually, expanding the
utility of a GIS system is primarily a data development and quality control exercise.
Geographic Data Models

Three representations have emerged to handle most geographic data: raster, vector,
and TIN. The following list describes these representations, which are shown in Fig-
ure 12.4.
• Raster: This representation stores data as discrete grids in which a single
value for each attribute is associated with each grid cell. Each grid cell has
an attribute value and location coordinate. Because the data is stored in a
matrix, the coordinates for each cell do not need to be stored explicitly
(ESRI, 2001). Rather, they can be determined “on the fly” because the origin
of the raster model, the grid cell size, and the rotation are known.
• Vector: This representation stores discrete features as points, lines, or poly-

gons. Each feature has attribute values and a set of x-y coordinates (and pos-
sibly a z coordinate) associated with it. In this way, vector data differ from
raster data, which have coordinates intrinsically associated with the cells
(ESRI, 2001).
• Triangulated irregular networks (TIN): TINs divide space into a set of

contiguous (non-overlapping) triangular faces. The triangular faces are
derived from irregularly spaced sample points, breaklines, and polygon fea-
tures, and each sample point has coordinate (x, y) and attribute (z coordinate
or other attribute to be modeled) information associated with it. Attribute
values at other locations (along breaklines and polygon features, or even on
the triangular faces themselves) are calculated using interpolation within the
TIN. TINs are sometimes considered a special case of a vector model, but
Zeiler (1999) presents them as a separate data model.
Figure 12.4
Geographic data
representations
Section 12.2 Developing and Maintaining an Enterprise GIS 533
Most data used in hydraulic modeling are vector data. For example, junction nodes
are points, pipes are lines, and node service areas are polygons. Although most mod-
eling data is made up of vector features, modelers also use raster and TIN data for
tasks such as extracting elevation data or using an aerial photo as a background for the
model.
12.2 DEVELOPING AND MAINTAINING AN

ENTERPRISE GIS
Many publications describe the process of developing a GIS, including Orne, Ham-
mond, and Cattran (2001) and Przybyla (2002). The pace of technologic advancement
and the diversity of commercial implementations outstrip the ability of any general
text to provide detailed guidelines on the configuration and management of a specific
GIS-based modeling system. Consequently, this section provides a general discussion
of the GIS development process and outlines the major steps involved in developing a
GIS.
GIS is generally implemented at one of four levels:
1. Project: Supporting a single project objective

2. Departmental: Supporting the needs of one department
3. Enterprise: Inter-departmental sharing of data that meets the needs of many
departments
4. Interagency: Sharing of application and data with external agencies
Building a GIS on a project basis for the sole purpose of using it with a hydraulic
model is quite rare because an organization derives a variety of benefits from a GIS,
including
• Elimination of redundant data maintenance activities

• Streamlining of workflow processes with GIS functions
• Improvements in access to quality information
• Ability to use spatial analysis to solve problems
For these reasons, a GIS is often implemented at the departmental or enterprise level.
The subsections that follow discuss some of the key steps that should be taken to
ensure a successful enterprise-level GIS implementation. For more information on
planning, developing, and maintaining a GIS for modeling and other applications, see
the references listed at the end of this chapter or the GIS section of this book's bibli-
ography (see page 727).
Keys to Successful Implementation

A community or water utility that commits to developing a GIS must consider several
key factors:
• Development of a GIS that is capable of supporting a hydraulic model

requires a high level of data quality, accuracy, and detail.
• Creation of a GIS necessitates a review of existing hardware and software
and does not imply that legacy systems will go away. Systems such as a cus-
tomer information system (CIS), CMMS, and SCADA are usually fixtures
on the landscape.
• Enterprise GIS development calls for a high level of interdepartmental coop-
eration. All departments involved in enterprise GIS development should
share the same vision for the system. The “people” aspect of GIS develop-
ment can be more challenging than the technological aspects.
• Enterprise GIS development demands a GIS leader — someone to champion
the effort. The GIS leader must build communication bridges between
departments so that people talk, cooperate, and share.
• The largest expense of the GIS project is usually the data development effort,
which often requires expensive conversion from paper map sources. If an
organization fails to consider all of the cartographic, asset database, and
hydraulic modeling needs during the database development, then its GIS
system will either (a) fail to meet future application demands or (b) necessi-
tate an expensive data upgrade effort in the future. For example, the GIS
must be able reproduce the map sources used in its creation, or all that will
have been accomplished is the creation of a new, redundant dataset to main-
tain.
Needs Assessment
Critical to a successful GIS implementation is a detailed understanding of the busi-
ness processes and operational and management needs of the organization. An under-
standing of the specific GIS functions required by the individual users is also crucial.
This information is gathered through a needs assessment.
A needs assessment has three components:
1. User needs assessment: The purpose of this assessment is to answer the follow-
ing questions: who are the people that are going to use the system; what roles do
they fulfill; what task or functions do they need to accomplish with the GIS; what
skill levels do they possess; where (physically) are they; and how often will they
use the GIS applications.
2. Data source assessment: The data source assessment determines what data
sources are available to support the GIS data development, including their for-
mats (e.g., electronic, paper), geographic extents, spatial (x, y, z) and attribute
accuracies, update frequencies, and dates last updated.
3. System design assessment: The purpose of this assessment is to determine the
types, locations, and characteristics of existing servers, individual workstations,
and computer networking components. This includes operating system platforms,
current applications being served, and current levels of utilization.
The needs assessment is an essential part of creating any IT system and is usually
accomplished by conducting detailed surveys and interviews with all potential GIS
users (including hydraulic engineers) and the appropriate organization decision-mak-
ers. It can also be accomplished as part of a business process workflow analysis (rec-
ommended).
A thorough needs assessment is crucial for developing a GIS that will provide ade-
quate service now and in the future. It should reveal specific problems or constraints
associated with present systems and identify project implementation requirements. A
properly conducted needs assessment will culminate in an integrated GIS with the fol-
lowing benefits and advantages:
• Increased operational and management efficiency and staff productivity
• Better sharing of data
• Quicker access to quality and timely information
• Full leveraging of the capabilities of the system
• Support for organizational operations that reflects the organization's mission
and priorities
• Staff endorsement and regular use
• Immediate value to the organization
• Functionality that supports all current and future needs
Design
The second phase of the GIS development process is the design, which potentially
includes the following tasks:
1. Application design: Describes the commercial software to be deployed and the
custom programs that will be combined to create the applications required to sup-
port user needs.
2. Database design: Describes the format for the layers, individual features, and
their attributes that will comprise the new GIS database.
3. Data development plan: Describes the techniques, methods, and procedures that
will be used to convert the data sources into the desired GIS database.
4. System design: Describes the hardware and software to be installed on new serv-
ers, workstations, and network components, as well as the reorganization and
redeployment of existing hardware and software components. (Beyond the scope
of this book.)
5. Implementation plan and schedule: Describes the tasks and provides a sched-
ule for developing the GIS. (Beyond the scope of this book.)
Application Design. The value of digital data in general, and in a GIS in partic-
ular, is the ability to create data one time and use it over and over for many purposes
without the need to manually handle the data. This recycling of data enables the data
user to work with much larger datasets than would be possible with manual data entry
and manipulation. Once developed, a GIS can serve many applications — beyond
hydraulic modeling — in a community or water utility.
In evaluating the responsibilities and workflow within a department, certain tasks that
can be performed more efficiently or effectively in a GIS will be identified. These
tasks will form the basis of GIS applications, and application descriptions prepared as
part of the needs assessment will document these tasks.
Popular GIS interfaces include:
• Interface to hydraulic and hydrologic modeling
• Interface to customer service and maintenance management systems
• Interface to customer information (billing) system
• Interface to laboratory information system
• Interface to SCADA (Supervisory Control and Data Acquisition) system
• Interface to Document Management/Workflow
Popular GIS-based applications (often utilizing/integrating the above system inter-
faces) include:
• Facility mapping (GIS data maintenance)
• Service request tracking/work management
• Asset management/GASB 34 reporting
• Crew dispatch/vehicle routing
• Field data collection/inspections

• Leak detection (compare master meter to individual account data)
• Link to as-built/intersection drawings (CAD or images)
• Isolation tracing/customer notification
• Demand projections/demographic data
• CIP planning/construction monitoring
• One call/underground service alert response
• New connection processing
• Cross-connection (backflow) test tracking
• Well monitoring/water resources analysis
Database Design. The database design for a GIS developed for a water utility
should strive to accomplish three fundamental goals that will enable the GIS to
become a strategic asset for the organization:
1. Cartographically represent the water distribution facilities (assets). This rep-
resentation can be used to create map products.
2. Inventory the network. The GIS is often the primary record for geographically-
distributed assets (that is, assets outside the plant).
3. Model the network. The GIS should be able to model the flow of water in the
system and support the integration of hydraulic modeling software.
Assuming that hydraulic modeling will be one of the activities supported by the GIS,
the GIS analyst should identify the entire range of related hydraulic applications that
potentially will be used. The analyst can then determine the types of data required for
these applications (for example, SCADA data and as-built drawings) and how the var-
ious types of data relate to one another. This information is necessary if the database
design is to meet all functional and interrelational requirements.
An important step in the database design process is the compilation of information
about the dataset, which is called metadata. Metadata provides the user with:
• Source of the data
• Data reliability, quality, and quality confidence levels
• Methods used in collecting and associating the data
• QA/QC (quality assurance/quality control) and validation procedures
• Other applications and software systems that the data might interact with
Metadata is becoming more important as the ability to share data between organiza-
tions over the Internet using eXtensible Markup Language (XML), a programming
language for structured information, becomes commonplace. Portal sites use meta-
data so that users can search for data layers and determine whether the datasets listed
meet their accuracy and spatial extent needs.
As GIS applications mature, the major GIS vendors have advanced their software
from relational-relational architectures (application/database) to object-relational
architectures. In the older relational-relational environment, connectivity, attribute
domain validation, and relationships between features were implemented as relation-
ships between tables (executed by the DBMS). In the new architecture, relationships
between features, connectivity, and attribute domain validation are implemented as
object-oriented components (executed by client-side software) with the raw objects
being stored in the RDBMS. The principal advantage of the new object-relational
structure is the compatibility of the GIS applications and data to other object-oriented
software applications.
This evolution provides new opportunities and poses some new challenges for inte-
gration with modeling. Objects can be more complex, with more sophisticated con-
nectivity, than with older GIS data types (points, lines, polygons). Because the data
modeling effort in an object world is so much more involved (Zeiler, 1999), the
hydraulic modeler has to be even more involved at the GIS database design stage
when using an object-relational GIS.
As an example, consider a GIS that has been developed for a water distribution sys-
tem using the older relational-relational system. In this model, each valve, hydrant,
water service, and service shutoff throughout the entire water distribution system is
required to be a node in the network in order to maintain connectivity. As a result, the
GIS might easily contain hundreds of thousands of short pipe segments (for example,
fitting to valve, valve to valve, and valve to fitting). To the hydraulic modeler, this
level of detail is unnecessary and even problematic.
Object-relational systems provide powerful new opportunities to the hydraulic mod-
eler. For instance, the connectivity between features is now controlled by the software
according to rules supplied by the user. For instance, valves and services can be part
of the network without creating nodes in the pipe segments (that is, pipes can be fit-
ting-to-fitting). In addition, connectivity rules can be established that enhance the
integrity of the database, such as “a pipe can only connect to another pipe through a
fitting, PRV, or pump station” or “only a customer meter or backflow device can be at
the end of a service line.” The result can be a GIS database that more closely meets
the needs of the hydraulic modeler.
In older data models, the modeler would likely need to develop code to perform the
tasks of discarding unnecessary point features and merging small pipe segments into
the larger segments that would form the link-node system in the model. This code
would rely not only on a programmer's expertise to properly process the myriad con-
ditions which exist, but on a high level of data quality to avoid unintentional errors
during the translation. With object-oriented data storage, the processing of data to
form new, merged features can be manifested through behaviors introduced at the
database design stage. Further, rules can be applied during dataset development to
ensure that, for example, a 2-inch pipe is not connected to a 16-inch pipe without
some type of required transition element. By designing these rules, properties, and
behaviors up front, extracting the features needed to model a system can become a
basic function of the GIS. The challenge in this system is that the technology is new,
and a significant investment in design time and funding will be required in the coming
years to turn the promise into reality.
Although it is relatively easy to define those characteristics of the buried piping sys-
tem that are needed to support hydraulic modeling, modeling the behavior of a com-
plex pump station within the GIS is more difficult and may not make sense. In many
cases, some of the information needed by the modeler will not be stored in the GIS
and will have to be acquired from other sources, such as as-builts or design docu-
ments. Ideally, all necessary input data for the model would exist in the GIS, and the
data format and content would be capable of supporting all modeling goals and appli-
cations identified in the needs assessment. In practice, however, all the data needed by
modelers is sometimes not in a readily available format or is not economical to
develop (for example, field data) during the initial GIS implementation. Therefore,
many hydraulic model interface implementations are staged to address the most
urgent or time-saving modeling priorities first. The GIS should be capable of generat-
ing, in a repeatable manner, a high percentage of the data required for model develop-
ment.
It is important to realize that GIS is often part of a wider information management
program that may include maintenance management systems, SCADA, facility auto-
mation, CAD, flow monitoring databases, a water asset database, as-built maps and
drawings, and other elements. Developing a database design that supports the multi-
ple needs of an organization is sometimes difficult to accomplish, but the end result
should not to be compromised.
Data Development Plan. As previously mentioned, development of the data

asset is often the most important and expensive step in the GIS development process.
The following subsections describe the main components and issues of the data devel-
opment plan.
Land Base. A water utility GIS must use some type of land base layer as a spatial ref-
erence. Some design issues that must be considered in the base map are the accuracy
(+/- x ft or m), the scale, the projection (latitude/longitude, state plane coordinates,
UTM [Universal Transverse Mercator], etc.), the vertical datum (NAD [North Ameri-
can Datum] 27, NAD83, etc.), the methods used to create the land base, the frequency
of updates, and the timeliness of the data.
The details of these decisions are beyond the scope of this book, but users must real-
ize that they cannot simply make a quick decision about using a United States Geo-
logical Survey (USGS) quadrangle map, aerial photo, or commercially produced map
without giving serious consideration to the long-term potential implications of the
land base choice.
Developing the land base is expensive and must be done correctly and accurately in
the initial stages of the project; otherwise, costs escalate when problems arise that
must be addressed after the fact. The lack of an accurate land base is the most com-
mon problem in developing an accurate GIS because multiple datasets developed
without a common land base will almost always be inconsistent with each other and
therefore not suitable to be used together.
For example, if spatially accurate water facilities (such as those located with a differ-
ential global positioning system [GPS]) are used in combination with inaccurate base
map layers, a good map will not be produced (for example, pipes may not fall on the
Case Study: Columbia, South Carolina

The water system in the City of Columbia—the • Beginning junction node number
largest water system in the central part of South • Ending junction node number
Carolina—provides potable water to a service
• Pipe segment number
area of approximately 320 square miles and
350,000 customers. The city worked to create a • Pipe segment diameter (in.)
GIS and construct and calibrate a system-wide • Pipe segment length (ft)
water model to evaluate and develop system
• Pipe segment friction coefficient
improvements.
• Elevations of connecting junction nodes
The city conducted a system-wide inventory by
using GPS and pen-based PCs to locate • Demand at connecting junction nodes
hydrants, water valves, water tanks, standpipes, • Minor loss coefficients
and pump stations, as well as the following
Included in the database are attributes for high-
attributes: feature identifier, feature type, condi- service pumping stations, distribution system
tion, and a flag indicating whether the feature was
tanks, and pressure reducing valves (PRVs).
located in the field. The city converted existing
water system mapping into a GIS format that was The GIS was used to assign elevations to model-
spatially accurate to the city’s planimetric and ing nodes. The city obtained elevation coverages
inventoried structure data. Existing water line data from the USGS, created a 3-D plane of USGS ele-
was corrected based on GPS data and record vation data, and assigned elevations based on the
drawing information. location of modeling nodes on the plane.
A skeletonized water network was then devel- The GIS also was used to assign demands to
oped. Water pipes and nodes were extracted from modeling nodes. (Nodes associated with pumps
CAD files and created in a GIS system. Included and tanks were temporarily removed so they
were all water lines with a 10-in. or larger diame- would not have demands assigned.) Geocoding
ter, plus selected 6-in. and 8-in. pipes required for was used to geolocate customer meters to the
system looping. All breaking points in a line were TIGER (Topologically Integrated Geographic
removed unless a diameter changed or an inter- Encoding and Referencing) street centerline file.
section between pipes occurred. Once conversion During geocoding, the GIS matched the address
was complete, a map of the skeletonized water of the customer meter with address ranges in the
network was created and inspected for connectiv- TIGER street centerline file. Demands were then
ity concerns and gaps in the distribution system. assigned based on the proximity of the customer
meter to a model node. Aggregate demands were
A database of water system model attributes was computed and then applied to the node.
created using GIS data, as-builts, and the city’s 1
in. = 1,200 ft scale schematic map. The database
contains information for each pipe segment
included in the model:
correct side of the street centerline or property line). Not being able to produce reli-
able maps is more than a nuisance; it could be a legal liability.
If an inaccurate street-centerline file is used as the spatial reference for placing water
facilities (say, for putting pipes four feet to the right of the street centerline), and a
new, more-accurate data source for the street centerline is provided in the future, all of
the facilities will have to be moved in the GIS to match the new spatial reference —
an expensive proposition.
Data Conversion. GIS database design must be matched to the specific needs of the
applications that the GIS is to serve, such as those of hydraulic models. Through data
conversion, data is made to conform to a uniform format that supports all functional
requirements. If necessary, GIS database design is modified to effectively integrate it

with the hydraulic modeling software, information management systems, as-built
drawing records, GPS, and CAD.
In supporting the data development process, the GIS analyst must develop and main-
tain data accuracy standards and implement QA/QC procedures to ensure data integ-
rity. The higher the accuracy standard, the higher the cost of the data conversion.
In some cases, the information necessary to support current and future hydraulic mod-
eling applications is not available in digital format. It will therefore be necessary to
convert paper maps to digital format or use GPS to collect data in the field. If multiple
data sources exist, the analyst should identify the sources that are best-suited for the
intended use, as well as the method for entering this data into the GIS.
Raster conversion, or rasterization, is the conversion of vector data (that is, points,
lines, and polygons) to cell or pixel data. Vector conversion, or vectorization, is the
conversion of cell or pixel data into points, lines, and polygons. Scanning of paper
maps produces raster files that must be converted to vector files to be useful for model
creation.
Rasterization and vectorization are only used when the quality of the data sources
supports it, which means that the information to be captured is easily distinguishable
by the vectorization software. Dealing with the errors of the vectorization process can
be onerous for the majority of GIS development projects. Heads-up digitizing is more
efficient, produces excellent results, and is rapidly becoming the industry standard.
Converting map data from CAD files into the desired GIS format can also pose a
number of challenges. For example, a single line in CAD file may actually represent
several pipe segments in the GIS. When this element is converted to GIS, the line
must be divided into multiple lines representing individual pipe segments. As an addi-
tional hurdle, map annotations such as pipe diameter and valve size in a CAD drawing
often do not have any data records and are not linked to the CAD features being anno-
tated. To be turned into attribute information in the GIS, these floating graphics must
be associated with the nearest pipe, and the text strings must be filtered to produce the
correct attribute. For instance, “1987 - 3 ½” may need to be converted into “1987” for
the year installed attribute and “3.50” for the pipe diameter attribute.
Pilot Study
After design, the next phase of the GIS development is usually to perform a pilot
study, which can include the following activities:
1. Create a pilot database following the data development plan.
2. Develop prototypes of high-priority applications following the application
design.
3. Provide core software training to key staff.
4. Test the applications and data during several pilot review sessions with end-users
and management.
5. Finalize the database design, data development plan, and system design docu-
ments as appropriate, incorporating what was learned from the pilot review ses-
sions.
Production
The next phase is the Production phase, which can include the following tasks:
1. Finalize the QA/QC software and techniques that will be used during the entire-
service-area data conversion.
2. Perform the service-area-wide data conversion following the data development
plan.
3. Procure new hardware and software.
4. Finalize the applications.
5. Develop end-user and system maintenance documentation.
6. Begin user training and rollout of high-priority applications (such as facility map-
ping).
Rollout
The final phase is a rollout phase, which can include the following tasks:
1. Installation of full complement of operational hardware and software.
2. User training and system maintenance training, which will likely be a combina-
tion of core GIS software courses and application training.
3. Acceptance testing (formal testing to determine whether the system satisfies the
acceptance criteria and thus whether the customer should accept the system).
4. Roll-out, which may include transition from any legacy systems being retired.
12.3 MODEL CONSTRUCTION

Constructing a water model and maintaining it over time can be one of the most time-
consuming, costly, and error-prone steps of a hydraulic modeling project. Prior to
widespread integration of GIS and modeling, building a water model was a special-
ized activity, separate from an organization's routine business procedures and work-
flows. Engineers created model input files by gathering, combining, and digitizing
data from a variety of hard-copy source documents, such as water system maps, topo-
graphic maps, and census maps, among others. If CAD data was available, features
required for modeling had to be extracted for use with the hydraulic modeling soft-
ware. The process was manual and required great attention to detail and many engi-
neering judgments along the way. Once a model was developed, calibrated, and run,
the modeler generated the required outputs and what-if scenarios and typically pro-
duced a master plan so that a community or utility could begin making the required
capital improvements.
Section 12.3 Model Construction 543
Despite rapid advances in hydraulic modeling software throughout the last decade,
which have included tools for the automatic translation of CAD data into modeling
data and for linking the model to external data sources, many communities and water
utilities have found it difficult to build, update, and maintain anything but highly skel-
etonized models. Because of the impracticality of using manual methods to gather and
manage large volumes of data, many communities and water utilities have not per-
formed routine modeling and have had no mechanism for water model maintenance.
Although organizations may have intended to keep models updated, time constraints
or business-process issues have often interfered. Even if CAD layers have been
updated or as-builts red-lined, model maintenance has usually been ignored because
the model input data has had to be maintained separately from system drawings.
Thus, in communities experiencing rapid development, water models could quickly
become outdated and have often had to be re-created from scratch when a current sys-
tem model was required.
A GIS professional can use a GIS to create a model more efficiently, more accurately,
and more cost-effectively than an engineer creating a model input file from scratch
inside a traditional modeling environment. Consider the following:
• Because GIS tools can automate the process, model building can be faster
and more efficient, especially for large models.
• Because GIS can manage large volumes of data, the model can incorporate
more detail.
• Both hydraulic modeling software and GIS have advanced editing tools. The
user needs to look at each task and decide if it is better done in the model or
the GIS.
• In the ideal case, where GIS data-entry has a consistent spatial reference and
a high level of quality control applied, the integrated model should contain
better data and should therefore be easier to calibrate and potentially lead to
better decision-making.
• Data collected and stored in the GIS primarily for other applications can be
extracted and incorporated into the model input file, if needed.
• As long as the GIS is maintained routinely, GIS data needed for reconstruct-
ing the model input file will be maintained routinely.
• Contour interpolations and digital elevation models (DEMs) in the GIS can
be used to assign elevations to model nodes automatically.
• Digital orthophotos available in the GIS can be overlaid with the model to
provide a base map reference.
• Georeferenced customer billing records can be used to generate and allocate
water demands for the model.
• If modeling results are returned to the GIS, further analyses can be run, and
other users such as planners and developers can manipulate modeling data in
conjunction with other GIS data.
Model Sustainability and Maintenance

By working with GIS professionals to build a water model from a properly con-
structed GIS, the engineer can spend time evaluating the water system and making
engineering decisions rather than constructing — and, over time, reconstructing —
the water model. If the GIS is used as the foundation for building the water model,
and assuming the GIS is maintained in the long term, the water model can be rebuilt
easily in the future using up-to-date GIS data (see Figure 12.5).
Figure 12.5
Maintaining the
hydraulic model
Graphic image courtesy of Jim McKibben, CH2MHill and ESRI
For instance, a model may be constructed by using the GIS to select and extract all
pipes that are 8 in. (200 mm) and larger, and by manually selecting additional pipes
that have smaller diameters but are necessary to close important loops or to service
large water users. All pipes and other network elements included in the model must be
marked appropriately in the GIS. Thus, as pipes, nodes, and associated features and
attributes in the GIS are updated over time, these elements can easily be selected
again and used to reconstruct the model with current data. Alternatively, the full sys-
tem can be imported into the model and then reduced by skeletonization.
The GIS professional and the modeler should determine where certain features
required for the model will reside — in the GIS or in the model. Ideally, the required
datasets should be detailed from the outset so that they can be stored and maintained
in the GIS, which helps to ensure that the model-building process can be repeated and
minimizes adjustments that the modeler must make to the data delivered by the GIS
professional.
Making good decisions about whether new pipes added to the distribution system
should be marked for inclusion in the model requires engineering judgment (based on
pipe size or other criteria). The GIS professional can use GIS tools to do much of the
sorting and selecting required for skeletonization, but the modeler must review the
entire network carefully to ensure that the model reflects reality.
Communication Between GIS and Modeling Staff

In some small systems, the modeler and the GIS professional are the same person, but
in most instances, the modeler and GIS professional are two individuals who are in
different departments or companies. To be successful, model development should be a

coordinated effort between the modeler and the GIS professional.
During the model-building process, the GIS professional must confer regularly with
the modeler to ensure that the model is developed efficiently. Both professionals
should acknowledge their differing perspectives to help ensure effective communica-
tion. The GIS professional understands GIS technology, what it can do, and how to
manipulate data from various systems and databases, and the modeler understands
how the model works, what data are required by the model, and how to ensure that the
model generates meaningful results.
Because the modeler is the user or consumer of the data to be developed or managed
by the GIS professional, he or she should take time to explain the data needs of the
model. Similarly, the GIS professional should explain GIS capabilities and limita-
tions, and what is technologically feasible.
Sufficient time spent in model planning and design ensures a smooth model-develop-
ment process. Specifically, the GIS professional and the modeler should discuss the
following issues at the outset of the project:
• Modeling basics: The modeler should share modeling basics with the GIS
professional (such as flow in or out of the system occurs only at nodes;
closed pipes do not have flow; the model predicts pressures at nodes; and so
on). Before using the GIS to develop the model input file, the GIS profes-
sional must have a fairly complete understanding of input data requirements,
types of model calculations, and how the model operates. In most instances,
the GIS professional can benefit from receiving some modeling training. By
understanding how the model works and having a vision of the intended out-
come, the GIS professional can determine which GIS tools should best serve
the modeler's needs. Merely listing required features and attributes will not
suffice; the modeler must discuss why these elements are crucial.
• Modeling terminology: For successful model development, each profes-

sional must have a basic understanding of the relevant terminology familiar
to the other. The GIS professional is comfortable discussing fields, tables,
and technology, and the modeler is comfortable discussing pipes and rough-
ness coefficients. Both professionals must discuss common modeling terms,
their meanings, and how they are applied. For example, they need to discuss
terms such as demand (the GIS professional may use demand to mean an
individual customer billing record, but the modeler may think of aggregate
demand at a modeling node) and junction (the GIS professional may use
junction synonymously with any node, and the modeler may think of junc-
tion as a location where two or more pipes meet).
• Standard units of measure: The modeler must define the standard units of
measure required and the calculations needed to make unit conversions.
Although the modeler understands intuitively how certain data must be rep-
resented in the model, the GIS professional may not. For example, if the
modeler needs average demand in gallons per minute, and the demand in the
billing system is in gallons per day, then the units must be converted either
by the model (preferred) or the GIS professional before the GIS data is use-
ful in the model. The modeler must therefore communicate very clearly how
the GIS professional should prepare the data for model input.
Using an Existing GIS for Modeling

A common misconception in the industry is that if a utility has a water system GIS
dataset, using the GIS to generate data needed for the model will be easy, if not auto-
matic. If hydraulic modeling was identified as a potential application of the GIS dur-
ing the needs assessment, this should be true. However, if the GIS was designed only
for purposes such as hard-copy mapping, maintenance management, or capital
improvement planning, it might not be easy to use the GIS to generate a hydraulic
model. This limitation is most pronounced when the GIS is used primarily for hard-
copy mapping and its “users” never interact directly with the underlying GIS data-
base. The GIS data are not necessarily bad, but they may not be suitable for hydraulic
modeling purposes, even though they satisfy the existing GIS needs.
Some typical challenges often faced by modelers using GIS data include the follow-
ing.
Database design incompatibilities or omissions:
• The GIS incorporates identification numbers for system elements that differ
from what can be used by the modeler.
• The GIS lacks critical valve information.
• The GIS lacks pump performance curve information.
• The GIS may have many short pipe segments that were introduced to provide
full topology, but that would unnecessarily add to the complexity of the
model.
Data errors:
• The GIS may not include connections that occur in the real system (because
components are not “snapped”), and may include connections that do not
occur in the real system (for example, at locations where pipes cross but do
not connect).
• The GIS may have artificial gaps between pipes that were introduced
because the data was created in a tiled system and pipes were not connected
across sheet boundaries.
Network Components
Even if hydraulic modeling was considered during the development of the GIS, unless
the GIS was created solely to support modeling, it is likely to posses a much greater
level of detail than what is needed by the model. This excess is especially true with
regard to the number of piping elements. It is not uncommon for the GIS to include
every service line and hydrant lateral. Such information is not needed for most model-
ing applications and should be removed to improve model runtime, reduce file size,
and save costs.
In addition to the extraneous service lines and hydrant laterals, a GIS may begin a
new pipe element at every isolation valve or fitting — an unnecessary level of detail
for most hydraulic modeling applications. Conversion of the GIS to the model there-
fore involves combining GIS elements to form a smaller number of model elements,
as shown in Figure 12.6.
Figure 12.6
GIS view versus
model view
Two steps exist at which the GIS data can be cleaned up for model use — importing
and skeletonization. In importing data from the model to the GIS, the extraneous GIS
features such as air release valves, in-line meters, open gate valves, and locations
where new pipe segments were installed in response to a break must be associated
with model elements. This association usually requires that mapping be established
between the GIS and model. The primary criterion for handling these elements is
whether or not they have appreciable head loss associated with them. Open gate
valves and pipe bends often have negligible head loss and can be represented as pres-
sure junctions, which can be eliminated during skeletonization. Those features that
have significant head loss either need to be assigned to an adjacent pipe or treated as a
control valve element or a general-purpose valve. Typical relationships are shown in
Table 12.1.
Trade-offs exist with each situation. For example, modeling an open gate valve or
water meter as a throttling control valve or general-purpose valve makes it easy to
assign minor losses to the element. However, the three elements (upstream pipe,
valve, and downstream pipe) could also be modeled as a single pipe element with a
minor loss, thereby reducing model size.
Table 12.1 GIS features and their corresponding model elements
GIS Feature Model Elements

Bend or other fitting with negligible Pressure junction to be skeletonized out
loss
Bend or other fitting with significant Valve or minor loss on adjacent pipe
loss
Isolating valve that is always open with Pressure junction to be skeletonized out
negligible loss
Isolating valve that is always open with Valve or minor loss on adjacent pipe
significant loss
Isolating valve that is normally closed Valve or a pipe segment with two nodes
Isolating valve that is always closed Valve or nodes connected by closed

pipe
Air release, blowoff, or surge relief Pressure junction to be skeletonized out

valve
Customer or hydrant lateral Pressure junction and pipe to be skele-
tonized out unless individual customer
is to be modeled
Check valve that is in-line Property of adjacent pipe
Check valve that is at a pump Usually automatically included in

pump, pipe on either side should be
combined
System water meter Valve or minor loss on adjacent pipe
Control valve PRV, PSV, TCV, or GPV, depending on

function of valve
Pump control valve Pressure junction to be skeletonized out
Reducer Pressure junction with different diame-

ter pipe on either side
Change in pipe material Pressure junction which may be skele-

tonized out depending on difference in
hydraulic properties
After the data are imported into a model, the number of elements can be further
reduced by skeletonization. Section 3.11 provides an overview of the skeletonization
process.
Some would argue that with the increasing power of models and the ease with which
models can share data with GIS, skeletonization has become less important and the
state-of-the-art is evolving toward more “all-pipe” models. Different levels of skele-
tonization are still appropriate, however, depending on how the model will be used.
The GIS professional and the engineer (or modeler) should discuss specific skeleton-
ization criteria in detail, including the importance of network connectivity. The GIS
professional must understand that a fully connected network is required for modeling.
Retrieval of Water Use Data

Utilities collect and compute water usage data by means of several possible methods,
ranging from the highly accurate to the more generalized. Three common techniques
are:
• Storing individual customer meter records for each billing period in a cus-
tomer information system
• Aggregating usage data for larger areas such as meter routes or pressure
zones
• Computing water usage estimates based on land-use or population
A GIS is not essential for loading the hydraulic model with water usage data, but it
can be used to effectively address each of these use cases and streamline the demand
allocation process. Much of the early work in using GIS for modeling focused on
accurately placing demands (Basford and Sevier, 1995; Buyens, Bizier and Conbee,
1996; Davis and Braun, 2000).
Node Service Polygons. Junction nodes are point features, but some demand
allocation methods require that the nodes have a polygon service area associated with
them. These polygons can be constructed manually, but automatic techniques, such as
construction of Thiessen polygons, exist. Thiessen polygons define the individual trib-
utary areas for each node. The space is divided such that any point within a particular
Thiessen polygon is nearer to that polygon's node than to any other node. A Thiessen
polygon for a node is created by connecting the perpendicular bisectors of lines drawn
between it and all adjacent nodes. The polygons for nodes along the outer edge of the
model will have no outer boundary, so it is essential to specify some method for clos-
ing the boundary to calculate areas. The boundary can be based on a buffering dis-
tance, but it is usually best to draw the outer boundary manually. Also, areas having
no customers may exist within the model area (for example lakes, parks, landfills).
Figure 12.7 shows a typical system with Thiessen polygons around each node.
Customer Meter Data. When water usage data is available for individual cus-
tomer meters, the GIS can be used to automatically geocode the customer location.
Geocoding is the process of matching an address data field or equivalent spatial refer-
ence, in this case a customer service address, against (usually) a street centerline file
or a parcel file that also contains address information. The resulting file is a set of
points that coincide with parcel centroids or the interpolated length along a line seg-
ment, depending on the source file used to geocode. It is important for the modeler to
understand how the meter location was geocoded. Using the underlying point coordi-
nates and the coordinates of the nodes in the model, the demand can be assigned to a
node, usually based on which node is nearest to the customer meter (see Figure 12.8).
When the actual service line connection point is stored in the GIS, then the demand
can be placed at one of the nodes for that main or proportioned based on distance
from the end nodes. Usually, the difference in model results due to different methods
of assigning demands along the pipe is negligible.
Figure 12.7
Thiessen polygons for
distribution system
nodes
Geocoding can be a very effective method of demand placement, but it relies on

source datasets (GIS and billing) containing address data that are seldom standardized
and validated for geocoding purposes. Therefore, it is important to understand that the
results of a geocoding effort may inaccurately attribute the load to one polygon versus
another due to geocoding inaccuracies. The geocoding results should be checked for
accuracy, and changes to the load assignments may need to be performed manually.
Another problem in using customer meter data is that CISs are set up to capture total
volume used for billing purposes, not flow rates that are required for models. Con-
sider Table 12.2. The CIS is likely to contain the first four columns, but the modeler
needs data from the final two columns (that is, flow rate). The conversion from vol-
ume billed to flow rate must be done either in the CIS software, the GIS, or by some
code written specifically for this calculation.
Figure 12.8
Meter aggregation
Table 12.2 Typical billing information from a Customer Information System CIS
Reading Billing Cycle Usage Flow Flow
Read Date
(100 ft3) (days) (1003 ft) (gpd) (gpm)
6,754.83 3/12/2002
6,770.25 4/9/2002 28 15.42 412 0.29
6,786.72 5/11/2002 32 16.47 385 0.27
6,805.99 6/11/2002 31 19.27 465 0.32
6,826.93 7/11/2002 30 20.94 522 0.36
6,850.74 8/9/2002 29 23.80 614 0.43
6,879.08 9/9/2002 31 28.35 684 0.48
6,900.10 10/9/2002 30 21.02 524 0.36
Cumulative 211 145.27 515 0.36
The modeler must also decide on whether the value to be loaded into the model is the
value from the most recent billing period or an average over a longer period. This
decision depends on whether the modeler is interested in loading the model with aver-
age annual flows or flows from a particular period for, say, a calibration exercise.
Other issues include the fact that not all customer meters are read on the same day.
For instance, “July water use” for one customer may be calculated from July 1 to July
31, but for another customer, it may be July 11 through August 12. In addition, some
utilities may use different units for commercial customers and residential customers.
Meters may have become stopped during the year or been replaced during the year
such that simply subtracting the last reading from the first reading will not give the
correct volume used. Sometimes corrections or adjustments are made to billed
amounts. The modeler must decide whether to use the raw value or corrected value.
Most of these cases must be dealt with manually. For the modeler, the key to working
with customer billing data is talking with the individual who understands the data to
determine its true meaning.
When nodal demands are determined from customer metering, the modeler must
remember that unaccounted-for water, by definition, is not included in the flow rates.
The modeler must determine how to assign unaccounted-for water to nodes. This
assignment may be done by simply dividing the unaccounted-for demand evenly
among the nodes. For greater accuracy in EPS runs, the modeler may want to set up a
different demand pattern for unaccounted-for water demands, although sufficient data
are not usually available for this.
Area Flow Data. When demand data are available for large areas such as pres-
sure zones and meter routes, these areas should be incorporated into the GIS as a sys-
tem meter polygon layer. Using overlay analysis, the water use within each polygon
can be equally distributed among the model nodes that fall within the polygon by
using point-in-polygon analysis. For customer demand data that have been placed
using a geocoding process, use of the meter route identifier to place points that could
not be geocoded is a common backup procedure.
Alternatively, if the individual model nodes have service area polygons associated
with them, the total demand from the pressure zone or meter route polygon can be
proportionally assigned to the service area polygons (and then to the model nodes)
based on the percentage of the larger polygon area taken up by the service area poly-
gon (see Figure 12.9). See page 549 (“Node Service Polygons”) for more information
about this method.
Figure 12.9
Proportional flow
distribution
Land-Use/Population Data. In some cases, the attribute data in a database

may be sufficient to get the bills out but may be very poor for geocoding purposes. In
other cases, demand data is only available as land-use or population information
based on census tracts, traffic analysis zones, or other similar areas. When data are
stored in the GIS in terms of the area of a given land use, population, number of struc-
tures, or population density, the modeler needs to develop the corresponding rates for
water usage. For example, if considering land use, rates need to be established for
water demand per unit of land use (for example, gal/day/ac or l/day/ha). If the number
of structures is being considered, then rates need to be established for water demand
for each type of dwelling unit. With this data, the aggregate demands can be automat-
ically computed within the GIS.
Figure 12.10 and Table 12.3 show how water usage can be determined based on land
use.
Figure 12.10
Determining water
use based on land use
Table 12.3 Computing water use based on land-use area

Total Node Land Use Unit
Land Use Demand Node Total
Node Area Area Demand
Type (l/day) (l/day)
(ha) (ha) (l/day/ha)
J-1 6.88 Industrial 6.88 11,200 77,100 77,100
J-2 7.69 Industrial 1.38 11,200 15,500 60,200
Commercial 0.92 4,700 4,300
Residential 5.38 7,500 40,400
J-3 7.69 Commercial 1.31 4,700 6,100 44,800
Residential 5.15 7,500 38,600
Undeveloped 1.23 0 0
J-4 8.50 Industrial 0.17 11,200 1,900 20,800
Commercial 0.10 4,700 470
Residential 2.45 7,500 18,400
Undeveloped 5.78 0 0
J-5 8.09 Industrial 6.48 11,200 72,500 80,100
Commercial 1.62 4,700 7,600
J-6 4.86 Industrial 0.20 11,200 2,200 33,400
Commercial 1.36 4,700 6,400
Residential 3.30 7,500 24,800
GIS can also be used to calculate demands for future conditions based on population
or land-use projections supplied to the modeler. Custom GIS operations enable mod-
elers to compute future water usage rates by overlaying data such as population pro-
jection and future land-use polygon layers with a modeling node layer.
Retrieval of Elevation Data

Many GISs contain elevation data, and numerous sources of elevation data in digital
form exist, such as published digital elevation models (DEMs). These models are
often used to develop the contours found in base mapping, but they can also be used to
derive the elevation for any other GIS feature, or for hydraulic model nodes. It is typi-
cally a very simple and fast process to drape, or overlay, the model nodes onto a sur-
face model and compute the elevation at each point. As with most GIS layers,
depending on the means used to develop it, a surface model can be very detailed or
quite coarse.
Elevation data can be stored in a GIS in several ways, two of which are TINs and ras-
ter DEMs. Figure 12.11 shows a shaded TIN of ground elevation data with the
hydraulic network draped over it, and Figure 12.12 shows a raster DEM with a
hydraulic model superimposed on it.
Figure 12.11
Network draped over
a TIN
Figure 12.12
Network super-
imposed on a DEM
For hydraulic modeling purposes, elevations are needed at specific points, including
junction nodes, pumps, valves, and tanks. To determine the elevation at these points,
the x-y coordinates of the node are passed to the GIS, and the GIS software uses the
elevations of DEM grid points or TIN vertices surrounding the node to determine the
node elevation, which can be passed back to the model. Use of DEMs is described in
more detail in Miller (1999), Price (1999), and Walski et al. (2001).
DEMs have been prepared for much of the United States by the U.S. Geological Sur-
vey and are available with grid spacing of 30 m (although 10-m grid spacing is avail-
able for an increasing number of locations). These maps are based on the contour
intervals found in the 7.5-minute quad sheets of that area (for example, 20 ft and 10
ft). Interpolating between these lines may only be accurate to several feet. In some
cases, it may be necessary to find sources of elevation data with higher accuracy in
order to achieve the model accuracy desired. It may be worthwhile to use free or low
cost data initially, and then determine if it is accurate enough. If it is not, more accu-
rate DEMs may need to be obtained.
Software that converts the data from a DEM to elevation attributes in a model usually
requires that the DEM data be in a specific format. Some type of function is usually
necessary to convert the raw DEM data into the required raster grid format and project
the data to the coordinate system being used in the GIS (most USGS data sources are
provided in Universal Tranverse Mercator [UTM] coordinates, which are not com-
monly used in municipal GIS applications).
Most models cover more than a single USGS quad sheet. Although elevation data
import software can work with one sheet at a time, it is usually easier to mosaic the
raster grid files together into a single file using a GIS function rather than deal with
numerous files. In recent years, improvements in LIDAR (Light Detection and Rang-
ing) technology have proven to be a source for higher resolution and accuracy in ele-
vation data. LIDAR data can be represented in a number of file formats, including
DEM and TIN.
Modeling GIS Versus Enterprise GIS

Few would doubt the value of using data from a GIS to create and update a model.
One issue that arises in modeling with GIS, however, is whether model results should
be carried back to the source GIS. Typically, the modeling data is maintained in a sep-
arate GIS layer created specifically for modeling.
Hydraulic modeling results in the enterprise GIS can be used for a number of pur-
poses, including:
• Pressure mapping (used by both engineering and customer service)
• Establishment of water main replacement priorities (when combined with
other GIS layers such as soils and repair data)
• Connection permit processing (available capacity can be reviewed and future
demands reserved)
• Contaminant isolation/remediation (contaminant is introduced accidentally
or intentionally)
Extract, Transform, Load. Modelers will leverage many data sources within
an enterprise. An enterprise GIS houses the centrally managed data that is shared by
individuals within the organization. Modelers typically envision the GIS as the pri-
mary data source or hub from which they will derive their models. Modelers also need
access to other important data sources managed outside of the GIS. Often, this infor-
mation will not reside in the same database or physical server that contains the GIS. It
may be maintained using a blend of database technologies and/or proprietary file for-
mats. The data sources may be distributed across the enterprise and hosted on various
servers and client workstations. Figure 12.13 presents a generalization of such an
enterprise-based GIS modeling system.
The figure shows the data pathways between the distributed data sources and the
modeling GIS. The key to successful modeling in the enterprise is to use automation
tools to accomplish the flow of data between these pathways whenever possible,
avoiding manual intervention or transcription of data between the GIS and the origi-
nal data sources. The modeler can accomplish this automation in several ways:
• By using general utilities to extract the data from one source, transform it as
required for the target source, and then load it into the target source. Such
utilities, called extract/transform/load utilities (ETLs), are readily available
and are extremely valuable to modelers.
• By using programming and script Application Programming Interfaces
(APIs) provided by the GIS and the databases to develop custom extensions
of the standard GIS commands for accomplishing the ETL steps.
• By using commercially available technologies that are specialized for mod-
eling within GIS. These technologies usually focus on some of the intensive
and key data transformation services (for example, automation of the proto-
cols for skeletonization, demand loading, and terrain extraction).
Modeling Features. Several aspects of hydraulic modeling must be taken into
account when using a GIS for this purpose, and these considerations often lead to the
separation of the enterprise GIS layers from the modeling GIS layers. These aspects
of modeling are:
• Network granularity
• Scenarios
• Time-series data
• Ownership
Network granularity. For many hydraulic modeling applications, the model network
does not need to contain every pipe in the actual system to obtain accurate results. For
example, a skeletonized version of the system is often sufficient to make informed
planning decisions and is often desirable to improve the efficiency of the hydraulic
modeling software. Providing fields in the enterprise GIS layers to manage the
hydraulic results would be wasteful because many of the GIS features are eliminated
during the skeletonization process. However, providing a single field for the modeling
identifier is possible, and this identifier allows the extraction or skeletonization soft-
ware to make the link between the GIS feature and model feature. This link then pro-
vides a trail back to the GIS feature for the analysis of hydraulic results.
Figure 12.13
Enterprise-based GIS
modeling system
Scenarios. In water system planning, the modeler is most often dealing with what-if
conditions, not as-built or in-service conditions. These conditions may include future
demands, proposed pipes and system facilities, or facility outages for emergency
response planning.
The GIS can be designed to incorporate the various what-if conditions and phases of a
water system facility. For example, a given pipe goes through many stages:
1. Alternative pipe in the model
2. Proposed pipe in a planning study or budget
3. A pipe under design, bid, and then construction
4. An installed pipe that has not been tested or placed in service
5. A pipe placed in service
Ideally, the enterprise GIS will be designed to manage this whole life cycle, because
many users need to see proposed pipes with other GIS data. A tremendous amount of
activity within a water utility revolves around planned or proposed pipes. The sharing
of data on planned or proposed pipes is one area where inter-agency GIS needs are
high. However, papers on modeling (for example, Deagle and Ancel, 2002) typically
describe how models use GIS, but few describe incorporating model information in
the GIS. An exception to the situations described in most papers is the Indianapolis
Water Company (Schatzlein and Dieterlein, 2002), which has a separate area in its
GIS for proposed projects.
Time-series data. Hydraulic models and water quality models are dynamic, meaning
they can be used to predict the response of the water system over an extended period
of time. The modeler needs to visually analyze this time-series data in an efficient
manner. Although an enterprise GIS can be designed to handle time-series data, they
typically lack the tools for working with time-series data efficiently because most
attributes for GIS features contain a single value (such as node elevation). With time-
series data, it is not uncommon to have 48 hourly values for each hydraulic parameter
for dozens of scenarios. The modeling software is set up to handle these large num-
bers of values, but the enterprise-wide GIS is usually not the best repository.
Ownership. In an enterprise-level GIS, the majority of the data is typically not

“owned” by the hydraulic modelers, and changes to this data are often outside the
modeler's direct control. Several issues arise from this lack of control. First, the enter-
prise GIS will often have data inaccuracies or omissions that are insignificant from a
maintenance management or asset management perspective, but that are important
from a hydraulic modeling viewpoint. The “owners” of the GIS may not be able to
make these corrections as quickly as the hydraulic modeler requires, so the modeler is
forced to make these corrections directly in the modeling GIS layers. An opposite but
equally problematic issue is that the hydraulic modeler often does not want changes to
the GIS to be immediately reflected in the model. For example, when calibrating
against fire hydrant flow test data, it is important that the model reflect the “as-built”
conditions at the time of the test. The GIS owners may be updating for recent water
main installations, and this could have a significant impact on the model's results. The
bottom line is that the modeler needs to be in control of the data used for modeling.
Section 12.4 GIS Analysis and Visualization 561
Case Study: Germantown, Tennessee

The City of Germantown created a GIS to get a Additional criteria were used to perform batch
complete and accurate digital mapping inventory queries and skeletonize the network. For exam-
of its water network. The city also wanted to use ple, because Germantown did not need to per-
the GIS to create a skeletonized water network form unidirectional flushing, all hydrant laterals
suitable for hydraulic modeling. Germantown’s were queried and removed from the model
approach to GIS development was to develop a dataset. Once the skeletonization was complete,
basic infrastructure GIS for a single application— any critical pipe spans omitted were added back
modeling—and then consistently enhance and into the skeletonized network.
build upon the GIS for additional applications over The GIS was used to assign customer demand
time. information, taken from the city’s billing database
In the past, Germantown had maintained its water and stored in parcel centroids, to model nodes.
data on a 1 in. = 500 ft scale, hard-copy map. The Each billing data address was matched with a par-
city used digital orthophotos to compile planimet- cel centroid address, and demands were then
ric data so a spatially accurate GIS could be cre- aggregated and assigned to modeling nodes. The
ated and used as the basis for the water model. procedure resulted in very precise demand
Valve and hydrant locations from the digital ortho- assignments, which would not have been possible
photos and planimetric mapping were used with without GIS. Elevations were also extracted from
the 1in. = 500 ft scale map as the basis for con- Germantown’s DEM and automatically assigned
necting water pipes. The result was a highly accu- to model nodes. The model was then calibrated
rate GIS dataset that could be used immediately and run to determine whether the system could
for modeling applications, inventory, and map meet the present-day demand, and where future
maintenance, and in the future, for building addi- capital improvements were needed.
tional applications. Based on the success of the GIS, Germantown is
GIS tools were used to extract a skeletonized now considering a GIS expansion that would pro-
water network for hydraulic modeling. The model duce an integrated information system featuring
was skeletonized to include water lines larger than an online parcel application and a document man-
6 in., plus all other lines critical for looping. agement system for managing the city’s as-built
drawings.
12.4 GIS ANALYSIS AND VISUALIZATION

This section illustrates ways that GIS might support a water utility in the hydraulic
model application cycle, from model development to capital planning, decision sup-
port, and operations support. It is important to note that a significant amount of plan-
ning and data conversion may be required to develop the datasets to empower a GIS to
perform all of these operations. Section 12.3 covers GIS development to support
water utilities, concentrating on the hydraulic modeling aspects.
Using Attributes to Create Thematic Maps

A classic and very simple use of GIS is to change the appearance of features in a
dataset based on an underlying attribute (attributes are the data that have been col-
lected and associated with each feature in a dataset). In Figure 12.14, this capability
illustrates an apparent error in GIS data that will be imported into a model, where a 6-
in. (150-mm) line still exists in the middle of all new 8-in (200-mm) piping. In this
figure, water pipes 4-in. (100-mm) and smaller are represented by dashed lines, 6-in.
(150-mm) pipes by thin lines, and 8-in. (200-mm) and larger pipes by thick lines.
Used in this manner, GIS can serve as an excellent quality-control tool for all manner
of data related to the water distribution system piping prior to model construction
(such as pipe length, C-value, diameter, and so on). Similarly, thematic maps can be
used to show which customers are late on payments or distinguish between different
types of water quality complaints (or any underlying attribute of any dataset).
Figure 12.14
Basic thematic
mapping — the thin
line between thick
lines indicates a
diameter error in the
GIS
Note: Most water distribution modeling software can perform basic thematic map-
ping, but the number of available settings to distinguish between unique attribute val-
ues is usually higher in a GIS environment.
A variation of basic thematic mapping for quality control or map production involves
color-coding data by underlying attribute ranges. For example, water distribution
nodes at elevations below 500 ft could be color-coded dark blue, 501-600 ft in light
blue, 601-700 ft in green, and so on. These types of thematic maps can also be effec-
tive in quality control and debugging operations.
Communicating the results of modeling to managers, regulators, the media, and the
general public is often difficult. A good map can convey the information much more
clearly than long oral explanations or text. Publications such as ESRI’s Map Book
series contain hundreds of examples of maps drawn using GIS (ESRI, series).
Some examples of thematic mapping are:
• Color-coding pressure zones and overlaying the model to show which unde-
veloped land parcels belong in which pressure zones
• Illustrating the areal extent of deficient fire flow capacity before and after the
proposed improvements are constructed
• Displaying concentrations of a contaminant in the system in conjunction
with epidemiological data to aid in correlating illness with water quality
• Showing a three-dimensional view of the hydraulic grade line overlaying a
system
Most GIS packages provide many alternatives within the framework of basic thematic
mapping that can be used to produce innumerable effects.
Using the Spatial Coincidence of Features to Assign

New Data
GIS software can analyze features on different layers to determine which coincide. In
this way, data from one set of features can be transferred into a second set of features.
A typical example in water model construction involves the overlay of model nodes
against a layer that contains elevation data. The process is more fully explored in Sec-
tion 12.3, but essentially, the elevation associated with the point at each node can be
incorporated as an attribute of the node based on this coincidence. Nodes that lie out-
side the extent of the elevation layer will not be assigned an elevation in this case.
Other examples of this type of analysis include
• Overlaying water distribution model nodes on meter routes to assign the
meter route number to each node
• Overlaying water distribution model pipes and nodes on pressure zone
boundaries to add the zone code to each model feature, enabling zone-spe-
cific models to be created (see Figure 12.15)
• Overlay a land-use map on a node service zone map to determine the land-
use in each node service zone
Using Spatial Relationships Between Features to Select

Certain Elements and Assign New Data
In addition to using the coincidence of features to assign new data, many GIS pack-
ages can select or isolate features based on their proximity to other features. Classic
cases include finding the closest feature in another dataset and finding all features
within a specified distance of a selected feature. Examples of this type of analysis
include
• Finding the water distribution node closest to a parcel centroid, so that the
billed water demand may be assigned to that node.
• Finding all parcels within 300 feet (91 m) of a set of water distribution pipes,
so that the owners can be notified that a construction project will take place
from October to November. In Figure 12.16, the dark pipes in the center of
the figure were selected and buffered by 300 ft (91 m) to select the parcels to
be notified.
Figure 12.15
Zone-specific model
Figure 12.16
Buffer area along
pipeline project
Using Relationships to Trace Networks

When structured properly, datasets such as those that make up a hydraulic model
(pipes and nodes), can be subjected to a process known as network tracing. In net-
work tracing, GIS software uses information on which links are connected to which
nodes to allow a system to be traversed. These functions are commonly applied to

street networks and utility networks. Examples of this type of analysis include:
• Generating shortest-path driving directions from point A to point B.
• Indicating the location of a hydrant flushing into a catch basin and identify-
ing the location at which the chlorinated water will enter a stream.
• Indicating the location of a water pipe break and identifying the valves that
must be closed to isolate the break (see Figure 12.17).
• Tracing the network to identify segments that are disconnected, either due to
inaccurate data or inadvertently closed valves. This analysis can be a great
aid in GIS and model input data quality control.
Figure 12.17
Using trace analysis
to simulate the
isolation of a main
break
Using Combinations of GIS Capabilities to Perform

Complex Analyses
Although many advanced types of GIS analyses can be performed, the examples
listed in the preceding sections illustrate some of the most common uses available
through most commercial GIS packages and can be easily mastered. These simple
capabilities can also be used in combination and series to perform more complex
analyses. Examples pertinent to the water distribution industry include:
• Developing or optimizing pressure zone boundaries: A polygon map of

the static hydraulic grade line in each pressure zone can be draped on an ele-
vation model to calculate theoretical static pressures across an entire service
area. The resulting pressures can then be color-coded and shown together
with piping and valve locations to indicate where the boundaries might be
adjusted for optimum service and to determine how undeveloped areas
should be incorporated into the zones (see Figure 12.18).
Figure 12.18
Pressure zone
topographic map
(darker shading
corresponds to higher
pressure zone)
• Locating potential sites for facilities: The GIS can be used to identify good
locations for water system facility sites. In this case, the GIS is not used as a
source of data for modeling but as a way to present alternatives to decision-
makers. As an example, Figure 12.19 illustrates the results of an analysis to
identify the top five ranked sites for the location of a new water storage tank.
About 13,000 parcels were ranked using various criteria, including parcel
size, land ownership, distance to a large water main, and parcel elevation.
The resulting scores were filtered to select the top five parcels. Better parcels
are shown in darker colors. Areas shown in white were eliminated from con-
sideration because of size or elevation constraints.
• Locating potential sites for monitoring equipment: Another example

described by Walski (2002) shows how model results can be imported into a
GIS and used with other data, such as property ownership and locations of
utility-owned buildings, to determine good locations for water quality or
pressure monitoring equipment. For example, siting a pressure monitor or
chlorine sensor in an area where the pressure or chlorine concentration is
almost always constant would not provide a great deal of information to the
system operators. Model runs can provide data on locations in the system
with widely fluctuating pressures or chlorine concentrations. These locations
can be displayed in the GIS by using color-coding and can be overlaid with
other layers to show desirability of monitoring locations in terms of utility
land ownership, location of power, and existence of pump stations and
vaults.
Section 12.5 The Future of GIS and Hydraulic Modeling 567
Figure 12.19
Locating the best sites
for a new water
storage tank
12.5 THE FUTURE OF GIS AND HYDRAULIC MODELING

Both GIS software and hydraulic models are embracing open computing standards
promulgated by the International Standards Organization (ISO), the Open GIS Con-
sortium (OGC), and others. These standards enable the once disparate systems of
models and GIS to share databases and objects, allowing users to quickly make more
informed decisions with less risk of using outdated or inaccurate information.
Users can look forward to even tighter integration between systems in the next few
years as vendors begin to incorporate object technologies and object-oriented pro-
gramming methods into their products. Ultimately, this technology will bring models
and GIS so close to one another as to be indistinguishable in certain applications.
REFERENCES
Anderson, J. L., Lowry, M. V., and Thomte, J. C. (2001). “Hydraulic and Water Quality Modeling of Distri-
bution Systems: What are the Trends in the US and Canada.” AWWA Annual Conference, Washington,
DC.
Basford, C., and Sevier, C. (1995). “Automating Maintenance of Hydraulic Network Model Demand Data-
base Using GIS and Customer Billing Records.” AWWA Computer Conference, Norfolk, Virginia.
Buyens, D. J., Bizier, P. A., and Combee, C. W. (1996). “Using a Geographical Information System to
Determine Water Distribution Model Demands.” AWWA Annual Conference, Toronto, Ontario, Canada.
Davis, A. L., and Braun, R. C. (2000). “General Purpose Demand Allocator (DALLOC).” EWRI Confer-
ence, ASCE, Minneapolis, Minnesota.
Deagle, G., and Ancel, S. (2002). “Development and Maintenance of Hydraulic Models.” AWWA IMTech,
Kansas City, Missouri.
ESRI. (2001). Dictionary of GIS Terminology. ESRI Press, Redlands, California.
ESRI (series). ESRI Map Book: Applications of Geographic Information Systems. ESRI Press, Redlands,
California.
Miller, D. (1999). “DEM-Based Terrain Modeling with ArcView 3-D Analyst.” ARC User, ESRI, July-Sep-
tember, 16.
Orne, W., Hammond, R., and Cattran, S. (2001). “Building Better Water Models.” Public Works, October.
Price, M. (1999). “Converting and Using SDTS Digital Elevation Models.” Arc User. ESRI, July-Septem-
ber, 10.
Przybyla, J. (2002). “What Stops Folks Cold from Pursuing GIS.” Public Works, April.
Schatzlein, M., and Dieterlein, J. (2002). “Finding Needles in a Haystack: IWC's Experience Optimizing
Integration with Hydraulic Models.” AWWA IMTech, Kansas City, Missouri.
Shamsi, U. M. (2001). “GIS and Modeling Integration." CE News, 13(6).
Walski, T. M. (2002). “Identifying Monitoring Locations in a Water Distribution System.” AWWA IMTech,
Kansas City, Missouri.
Walski, T. M., Toothill, B., Skronski, D., Thomas, D., and Lowry, S. G. (2001). “Using Digital Elevation
Models.” Current Methods, Haestad Press, 91.
Zeiler, M. (1999). Modeling Our World. ESRI Press, Redlands, California.
12.1 Match the definition with the GIS term on the left. Place the letter in the blank.
1) ___
Polygon a) Data structure in which features are represented by set of coordinates
2) ___
TIN b) A closed two-dimensional figure
3) ___
Raster c) Assign x-y coordinates to a location such as an address
4) ___
DEM d) Data structure made up of contiguous non-overlapping triangles
5) ___
Vector e) Type of projection of earth surface to Cartesian coordinates
6) ___
Analytical f) Assigns unique attribute value to even sized cells
paradigm
7) ___ Geocode g) Polygons generated around points
8) ___ UTM h) File used for storing elevation data
9) ___ Thiessen i) Using GIS as a place to store mapping data
12.2 Consider a small pressure zone with known flow rates. The demands are to be assigned to the nodes
based on the area of each node. A set of Thiessen polygons was created and their areas are listed in
the table below. Given that the long term average demand is 85 gpm and the peak is 215 gpm, find
the average and peak demand at each node as would be done by a GIS-based tool.
Average and peak demands
Area Average Demand Peak Demand

Node
(acres) (gpm) (gpm)
J-151 15
J-152 25
J-153 23
J-154 41
J-155 12
J-156 15
J-157 11
When is it logical to expect water use to be proportional to the service area of the node?
12.3 The interpolation for elevation used in a GIS is based on two-dimensional TINs using three points.
In the simple problem below, interpolate the elevation at point x, given the elevation at points x1 and
x2.
Extracting elevation data from points is based on the assumption that the points are close enough
such that interpolation gives accurate results. What would the elevation at x be if, between the points,
there is a steep embankment as shown by the dashed gray line?
12.4 Using the nearest node method based on customer meters, the demands assigned to a given node are
16 gpm. What demand should be placed on the node if unaccounted-for water is 20 percent of pro-
duction?
12.5 Given the demand (already corrected for unaccounted-for water) for the land-uses in the following
table:
Unit Demand
Demand Type
(L/day/hectare)
Single Family
1,400
Residential
Multi-family
1,800
Residential
Commercial 1,200
Light Industrial 2,500
Open Space 100
And the following land uses for each node:

Node Area % in Land
Land Use
(hectare) Use
J-201 5 Single Family Residential 100
J-202 8 Commercial 100
Single Family Residential 65
J-203 12 Commercial 20
Light Industrial 15
Single Family Residential 40
J-204 15
Multi-family Residential 60
Light Industrial 75
J-205 9
Open Space 25
Determine the demand at each node in Liters/day.
To perform the above calculation in a GIS, it is necessary to have two different polygon layers.
Describe them?
12.6 In building a GIS for water distribution system modeling, indicate whether the following items
should be vectors (also indicate if they should be point, line, or polygon), rasters, or TINs. There
may be more than one correct answer for each depending on the system.
Item GIS representation

Junction Node Vector (point)
DEM
Tank
Pipe
Aerial Photo Back Ground
DXF File Background
Raw Water Reservoir
Pump
Node Service Area
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


C H A P T E R
13
Transients in Hydraulic Systems
A hydraulic transient is the flow and pressure condition that occurs in a hydraulic sys-
tem between an initial steady-state condition and a final steady-state condition. When
velocity changes rapidly because a flow control component changes status (for exam-
ple, a valve closing or pump turning off), the change moves through the system as a
pressure wave. If the magnitude of this pressure wave is great enough and adequate
transient control measures are not in place, a transient can cause system hydraulic
components to fail.
This chapter presents the basic concepts associated with transient flow, discusses var-
ious methods to control hydraulic transients, and introduces aspects of system design
that should be considered during transient analysis. Special attention is given to the
specification of system equipment and devices that are directly related to causing and
controlling hydraulic transients.
The primary objectives of transient analysis are to determine the values of transient
pressures that can result from flow control operations and to establish the design crite-
ria for system equipment and devices (such as control devices and pipe wall
thickness) so as to provide an acceptable level of protection against system failure due
to pipe collapse or bursting. Because of the complexity of the equations needed to
describe transients, numerical computer models are used to analyze transient flow
hydraulics. An effective numerical model allows the hydraulic engineer to analyze
potential transient events and to identify and evaluate alternative solutions for control-
ling hydraulic transients, thereby protecting the integrity of the hydraulic system.
13.1 INTRODUCTION TO TRANSIENT FLOW

System flow control operations are performed as part of the routine operation of a
water distribution system. Examples of system flow control operations include open-
ing and closing valves, starting and stopping pumps, and discharging water in
response to fire emergencies. These operations cause hydraulic transient phenomena,
especially if they are performed too quickly. Proper design and operation of all
574 Transients in Hydraulic Systems Chapter 13
aspects of a hydraulic system are necessary to minimize the risk of system damage or
failure due to hydraulic transients.
When a flow control operation is performed, the established steady-state flow condi-
tion is altered. The values of the initial flow conditions of the system, characterized by
the measured velocity (V) and pressure (p) at positions along the pipeline (x), change
with time (t) until the final flow conditions are established in a new steady-state con-
dition.
The physical phenomenon that occurs during the time interval TT between the initial
and final steady-state conditions is known as the hydraulic transient. In general, tran-
sients resulting from relatively slow changes in flow rate are referred to as surges, and
those resulting from more rapid changes in flow rate are referred to as water hammer
events.
Evaluating a hydraulic transient involves determining the values during the time inter-
val TT of the functions V(x, t) and p(x, t) that result from a flow control operation per-
formed in a time interval TM. Changes in other physical properties of the liquid being
transported, such as temperature and density, are assumed to be negligible.
The evolution of a transient is represented at incremental positions in the system
through a graph like the one shown in Figure 13.1. In this graph, pressure (p) is repre-
sented as a function of time (t) resulting from the operation of a flow control valve.
Note that the figure represents a view of the transient at a fixed point (x) just upstream
of the valve that is being shut. In the figure, p1 is the initial pressure at the start of the
transient event, p2 is the final pressure at the end of the event, pmin is the minimum tran-
sient pressure, and pmax is the maximum transient pressure.
Impacts of Transients
A wave is a disturbance that transmits energy and momentum from one point to
another through a medium without significant displacement of matter between the
two points. For example, a wave caused by a boat moving across a lake will disturb a
distant boat, but water is not directly transported from the moving boat to the other
boat. As can be seen in Figure 13.1, a transient pressure wave subjects system piping
and other facilities to oscillating high and low pressure extremes. These pressure
extremes and the phenomena that accompany them can have a number of adverse
effects on the hydraulic system.
If transient pressures are excessively high, the pressure rating of the pipeline may be
exceeded, causing failure through pipe or joint rupture, or bend or elbow movement.
Excessive negative pressures can cause a pipeline to collapse or groundwater to be
drawn into the system. Low-pressure transients experienced on the downstream side
of a slow-closing check valve may result in a very fast, hard valve closure known as
valve slam. This low-pressure differential across the valve can cause high-impact
forces to be absorbed by the pipeline. For instance, a 10-psi (69-kPa) pressure differ-
ential across the face of a 16-in. (400-mm) valve results in a force in excess of 2,000
lb (8,900 N). This situation is common where hydropneumatic tanks are used on
pump station header systems, but it can also result from elevated tanks that are in
close proximity to pump stations.
Section 13.1 Introduction to Transient Flow 575
Figure 13.1
Hydraulic transient at
position x in the
system
Some flow control operations that initially cause a pressure increase can lead to sig-
nificant pressure reductions when the wave is reflected. The magnitude of these pres-
sure reductions is difficult to predict unless appropriate transient analysis is
performed. If subatmospheric pressure conditions result, the risk of pipeline collapse
increases for some pipeline materials, diameters, and wall thicknesses. Although the
entire pipeline may not collapse, subatmospheric pressure can still damage the inter-
nal surface of some pipes by stripping the interior lining of the pipe wall.
Even if a pipeline does not collapse, column separation (sudden vaporous cavitation)
caused by differential flow into and out of a section could occur if the pressure in the
pipeline is reduced to the vapor pressure of the liquid. Two distinct types of cavitation
can result. Gaseous cavitation involves dissolved gases such as carbon dioxide and
oxygen coming out of the water, and vaporous cavitation is the vaporization of the
water itself. When the first type of cavitation occurs, small gas pockets form in the
pipe. Because these gas pockets tend to dissolve back into the liquid slowly, they can
have the effect of dampening transients if they are sufficiently large.
With vaporous cavitation, a vapor pocket forms and then collapses when the pipeline
pressure increases due to more flow entering the region than leaving it. Collapse of
the vapor pocket can cause a dramatic high-pressure transient if the water column
rejoins very rapidly, which can in turn cause the pipeline to rupture. Vaporous cavita-
tion can also result in pipe flexure that damages pipe linings. Cavitation can and
should be avoided by installing appropriate protection equipment or devices in the
system, as described later in this chapter on page 607.
When pressure fluctuations are very rapid, as is the case with water hammer, the sud-
den changes can cause pipelines and pipeline fittings (bends and elbows) to dislodge,
resulting in a leak or rupture. In fact, the cavitation that commonly occurs with water
hammer can—as the phenomenon’s name implies—release energy that sounds like
someone pounding on the pipe with a hammer.
Overview of Transient Evaluation

For typical water distribution main installation, transient analysis may be necessary
even if velocities are low. System looping and service connections may amplify tran-
sient effects and need to be studied carefully. Transient analysis should be performed
for large, high-value pipelines, especially those with pump stations. A complete tran-
sient analysis, in conjunction with other system design activities, should be performed
during the initial design phases of a project. Normal flow control operations and pred-
icable emergency operations should of course be evaluated during the design. How-
ever, uncommon flow control activities can occur once the system is in operation,
making it important that all factors that could affect the integrity of the system be con-
sidered.
Evaluating a system for potential transient impacts involves determining the values of
head (Hmax and Hmin) at incremental positions in the system. These head values corre-
spond to the minimum and maximum pressures of the transient pressure wave,
depicted as pmax and pmin in Figure 13.1. Computation of these head values through the
system allows the engineer to draw the grade lines for the minimum and maximum
hydraulic grades expected to occur due to the transient. If the elevation (z) along the
pipe is known, then the pipe profile can be plotted together with the hydraulic grades
and used to examine the range of possible pressures throughout the system.
Figure 13.2 shows a pumping system in which an accidental or emergency pump
shutdown has occurred. The extreme values indicated by the hydraulic grade lines in
Figure 13.2 were developed by reviewing the head versus time data at incremental
points along the pipeline.
The grade lines for Hmin and Hmax, which define the pressure envelope or head enve-
lope, provide the basis for system design. If the Hmin grade line drops significantly
below the elevation of the pipe, as shown in a portion of the system in Figure 13.2,
then the engineer is alerted to a vacuum pressure condition that could result in column
separation and possible pipeline collapse. Pipe failure can also result if the transient
pressure in the pipe exceeds the pipe’s pressure rating. Maximum (or minimum) tran-
sient pressure can be determined for any point in the pipeline by subtracting the pipe
elevation (z) from Hmax (or Hmin) and converting the resulting pressure head value to the
appropriate pressure units.
Specialized programs are necessary to perform transient analysis in water distribution
systems. The extended-period simulation (EPS) discussed elsewhere in this book does
not consider momentum in the system and is therefore incapable of detecting or ana-
lyzing hydraulic transients. Such simulations are sufficient to analyze hydraulic sys-
tems that undergo velocity and pressure changes slowly enough that significant
inertial forces are not mobilized. If a system undergoes large changes in velocity and
pressure in relatively short time periods, then transient analysis is required.
Section 13.2 Physics of Transient Flow 577
Figure 13.2
Grade lines for a
pumping system
during an emergency
shutdown
13.2 PHYSICS OF TRANSIENT FLOW

When a flow control device is operated rapidly in a hydraulic system, the flow
momentum changes as a result of the acceleration of the liquid being transported and
a transient is generated. This hydraulic transient is analyzed mathematically by solv-
ing the velocity [V(x, t)] and pressure [p(x, t)] equations for a well-defined elevation
profile of the system, given certain initial and boundary conditions determined by the
system flow control operations. In other words, the main goal is to solve a problem
with two unknowns, velocity (V) and pressure (p), for the independent variables posi-
tion (x) and time (t). Alternatively, the equations may be solved for flow (Q) and head
(H).
The continuity equation and the momentum equation are needed to determine V and p
in a one-dimensional flow system. Solving these two equations produces a theoretical
result that usually reflects actual system measurements if the data and assumptions
used to build the numerical model are valid. Transient analysis results that are not
comparable with actual system measurements are generally caused by inappropriate
system data (especially boundary conditions) and inappropriate assumptions.
Hydraulic transients can be analyzed using one of two model types: a rigid model or
an elastic model. These models and their limitations are discussed in the next subsec-
tions.
The Rigid Model

The rigid model assumes that the pipeline is not deformable and the liquid is incom-
pressible; therefore, system flow control operations affect only the inertial and fric-
tional aspects of transient flow. Given these considerations, it can be demonstrated
using the continuity equation that any system flow control operations will result in
instantaneous flow changes throughout the system, and that the liquid travels as a sin-
gle mass inside the pipeline, causing a mass oscillation. In fact, if the liquid density
and the pipeline cross-section are constant, the instantaneous velocity is the same in
all sections of the system.
These rigidity assumptions result in an easy-to-solve ordinary differential equation;
however, its application is limited to the analysis of surge (see the next subsection on
Limitations). The rigid model is established for each time instant (t) of the transient
period using the fundamental rigid model equation:
fL L dQ
H 1 – H 2 = ----------------2- Q Q + ------- ------- (13.1)
2gDA gA dt
where H1 = total head at position 1 in a pipeline (ft, m)

H2 = total head at position 2 in a pipeline (ft, m)
L = length of pipe between positions 1 and 2 (ft, m)
A = area (ft2, m2)
dQ/dt = derivative of Q with respect to time
If a steady-state flow condition is established — that is, if dQ/dt = 0 — then Equation

13.1 simplifies to the Darcy-Weisbach formula for computation of head loss over the
length of the pipeline. However, if a steady-state flow condition is not established
because of flow control operations, then three unknowns need to be determined: H1(t)
(the upstream head), H2(t) (the downstream head), and Q(t) (the instantaneous flow in
the conduit). To determine these unknowns, the engineer must know the boundary
conditions at both ends of the pipeline.
Using the fundamental rigid model equation, the hydraulic grade line can be estab-
lished for each instant in time. The instanteneous slope of this line indicates the
hydraulic gradient between the two ends of the pipeline, which is also the head neces-
sary to overcome frictional losses and inertial forces in the pipeline. For the case of
flow reduction caused by a valve closure (dQ/dt < 0), the slope is reduced. If a valve is
opened, the slope increases, potentially allowing vacuum conditions to occur in the
pipeline.
Limitations. The rigid model has limited applications in hydraulic transient analy-
sis because the resulting equation does not accurately interpret the physical phenome-
non of pressure wave propagation caused by flow control operations, and because it is
not applicable to rapid changes in flow. With the rigid model, the slope change is
directly proportional to the flow change. According to the model, if an instantaneous
flow change (even a minor one) occurs as a result of a rapid flow control operation,
the resulting head is immediately and excessively changed. Therefore, rigid model
results are not realistic for analyzing rapid changes in the system.
The slow-flow transient phenomenon to which the rigid model may be applied is
called surge. With surge, head changes occur slowly and are relatively minor in mag-
nitude, allowing changes of the liquid density and/or elastic deformation of the pipe-
line to be neglected.
The Elastic Model

The elastic model assumes that changing the momentum of the liquid causes defor-
mations in the pipeline and compression in the liquid. Because liquid is not com-
pletely incompressible, it can experience density changes. Based on these model
assumptions, a wave propagation phenonemon will occur. The wave will have a finite
velocity that depends on the elasticity of the pipeline and of the liquid.
Elasticity of a Liquid. The elasticity of any medium is characterized by the

deformation of the medium due to the application of a force. If the medium is a liquid,
this force is a pressure force. The elasticity coefficient (also called the elasticity index,
constant, or modulus) describes the relationship between force and deformation and is
a physical property of the medium.
Thus, if a given liquid mass in a given volume (V) is submitted to a static pressure rise
(dp), a corresponding reduction (dV < 0) in the fluid volume occurs. The relationship
between cause (pressure increase) and effect (volume reduction) is expressed as the
bulk modulus of elasticity (Ev) of the fluid, as shown in Equation 13.2:
dp dp
E v = – -------------- = ------------------ (13.2)
dV eV dU e U
where E v = volumetric modulus of elasticity (M/LT2)

dp = static pressure rise (M/LT2)
dV eV = incremental change in liquid volume with respect to initial volume
dU eU = incremental change in liquid density with respect to initial density
A relationship between a liquid’s modulus of elasticity and density yields its charac-
teristic wave celerity, as shown in Equation 13.3.
E dp-
a = -----v = ----- (13.3)
U dU
where a = characteristic wave celerity of the liquid (L/T)

The characteristic wave celerity (a) is the speed with which a disturbance moves
through a fluid. Its value is approximately equal to 4,716 ft/s (1,438 m/s) for water
and approximately 1,115 ft/s (340 m/s) for air. For water with a one-percent volume
of free air, this value is approximately 410 ft/s (125 m/s) due to the decreased elastic-
ity of the air–liquid mixture. The next subsection explains the physical meaning of the
characteristic wave celerity and the reason that the speed varies so widely depending
on the medium.
Example — Computing Modulus of Elasticity for a Fluid. Assume that a 0.26-gal

(1-liter) volume of water at ambient temperature with a density of 1.94 slugs/ft3 (1,000 kg/m3) is sub-
jected to a pressure of approximately 290 psi (20 bar). In this case, the volume would decrease by
approximately 0.055 in3 (0.9 cm3), or by 0.09%. Compute the modulus of elasticity for water.
Using Equation 13.2, the modulus of elasticity can be computed as
Ev = -290 psi/-0.0009 = 3.2 u 105 psi
or
Ev = -20 bars/-0.0009 = 2.2 u 104 bars = 2.2 u 109 Pa = 2.2 GPa
Wave Propagation in a Liquid. Temporal changes in liquid density and

deformations of system pipelines are not considered in steady-state flow analysis,
even if considerable spatial changes in pressure exist due to frictional head losses or
elevation differences in the system. Steady-state flow analysis assumes that to move a
molecule of liquid in the pipeline system, a simultaneous displacement of all other
liquid molecules in the system must occur. It also assumes that the liquid density is
constant throughout the system.
In reality, however, some distance exists between molecules, and a small disturbance
to a fluid molecule is transmitted to an adjacent molecule only after traveling the dis-
tance that separates them. This movement produces a small local change in the den-
sity of the fluid, which in turn produces a wave that propagates through the system.
The approach used to analyze transient waves depends on the perspective from which
the equations are written. They can be written from the perspective of a stationary
observer, an observer traveling with the velocity of the water, or an observer traveling
with the velocity of the wave.
For example, consider a liquid flowing with a velocity (V) in a nondeformable pipe
that is subjected to a pressure force (dp) in the direction of flow caused by a system
operation at the left end of the pipe (see Figure 13.3). The force applied to the liquid
molecules on the left transmits as a molecular action to the adjacent molecules on the
right, which characterizes a mechanical wave propagating in the direction of the flow.
In Figure 13.3, the flow is to the right at velocity V, and the observer and the distur-
bance are moving to the right at velocity c. [The term c represents the speed of the
wave relative to a fixed point, and is equal to the characteristic wave celerity (a) plus
the velocity of the moving fluid (V).] The flow velocity in front of the moving
observer relative to the observer is therefore (c - V) = a.
Figure 13.3
Wave propagation in a
liquid, assuming the
observer is moving at
velocity c
After a period of time, the wave will have traveled a distance (x), and a disturbed zone
will exist behind the wave. In front of the wave, the initial flow condition is not yet
affected and maintains its initial properties. The flow properties in the pipeline will
appear variable to a stationary observer because the flow conditions change along the
length of the pipeline. The observer moving with a control volume at velocity c will
see the liquid flowing into the control volume at a velocity (c - V) and out of the con-
trol volume at velocity [c - (V + dV)], where dV is the disturbance of the absolute
velocity of the flow caused by the pressure force.
Water Hammer Theory. Water hammer refers to the transient conditions that
prevail following rapid system flow control operations. It can be used beneficially, as
in the case of a hydraulic ram, which is a pump that uses a large amount of flowing
water to temporarily store elastic energy for pumping a small amount of water to a
higher elevation. More commonly, the destructive potential of water hammer is what
attracts the attention of water engineers.
The concept of propagation of a wave in a liquid within a pipeline is needed to under-

stand the water hammer phenomenon. The preceding subsection on “Elasticity of a
Liquid” described pressure waves propagating in fluid only. This explanation can be
used to describe wave speed in a completely rigid pipeline; however, most pipelines
are made of deformable materials for which elasticity must be taken into account.
To generate equations describing the water hammer phenomenon, the unsteady

momentum and mass conservation equations are applied to flow in a frictionless, hor-
izontal, elastic pipeline. First, the momentum equation is applied to a control volume
at the wave front following a disturbance caused by downstream valve action. The fol-
lowing equation may be developed, which is applicable for a wave propagating in the
upstream direction:
a
'p = – Ua 'V or 'H = – --- 'V (13.4)
g
where 'p = change in pressure (psi, Pa)

U = fluid density (slugs/ft3, kg/m3)
a = characteristic wave celerity of the fluid (ft/s, m/s)
'V = change in fluid velocity (ft/s, m/s)
'H = change in head (ft, m)
The equation makes intuitive sense in that a valve action causing a positive velocity
change will result in reduced pressure. Conversely, if the valve closes (producing a
negative 'V ), the pressure change will be positive.
By repeating this step for a disturbance at the upstream end of the pipeline, a similar
set of equations may be developed for a pulse propagating in the downstream direc-
tion:
a
'p = Ua 'V or 'H = --- 'V (13.5)
g
These equations are valid at a section in a pipeline in the absence of wave reflection.
They relate a velocity pulse to a pressure pulse, both of which are propagating at the
wave speed a. To be useful, a numerical value for the wave propagation velocity in the
fluid in the pipeline is needed.
Assume that an instantaneous valve closure occurs at time t = 0. During the period
L/a (the time it takes for the wave to travel from the valve to the pipe entrance), steady
flow continues to enter the pipeline at the upstream end. The mass of fluid that enters
during this period is accommodated through the expansion of the pipeline due to its
elasticity and through slight changes in fluid density due to its compressibility.
The following equation for the numerical value of a is generated by applying the
equation for conservation of mass to the entire pipeline for L/a seconds and combin-
ing it with Equation 13.4
Ev
-----
U
a = ----------------------- (13.6)
E v 'A
1 + -------------
A 'p
where E v = volumetric modulus of elasticity of the fluid (lbf/ft2, Pa)

'A = change in cross-sectional area of pipe (ft2, m2)
For the completely rigid pipe, the pipe area change, 'A , is zero and Equation 13.6
reduces to Equation 13.3. For real, deformable pipelines, the wave speed is reduced,
since a pipeline of area A will be deformed 'A by a pressure change 'p . The solid
mechanics problem of finding this area change for a given pressure change is all that
is needed to determine the wave propagation speed a of any pipeline. On page 586 of
this chapter, Korteweg’s equation (Equation 13.11) presents the form of this equation
for a thin-walled elastic pipeline.
By using Equation 13.6 to calculate a numerical value for the wave speed in the pipe-
line, Equations 13.4 and 13.5 may be used with confidence at any section in the pipe-
line in the absence of reflections. Given that a is roughly 100 times as large as g, a 1-
ft/s (0.3-m/s) change in velocity can result in a 100-ft (30-m) change in head. Because
changes in velocity of several feet or meters per second can occur when a pump shuts
off or a hydrant or valve is closed, it is easy to see how large transients can occur
readily in water systems.
Full Elastic Water Hammer Equations. Derivation of the complete equa-

tions for transient analysis is beyond the scope of this book but can be found in other
references, such as Almeida and Koelle (1992) and Wylie and Streeter (1993). The
water hammer equations are one-dimensional unsteady pressure flow equations given
by
2
wH a wQ
------- + ------- ------- = 0 (13.7)
wt gA wx
------- + gA ------- + fQ
wQ wH Q = 0
------------- (13.8)
wt wx 2DA
Transient modeling essentially consists of solving these equations for a wide variety
of boundary conditions and system topologies. The equations cannot be analytically
solved, so various approximate methods have been developed over the years. Today,
solutions for all but the simplest problems are performed using computers. The fol-
lowing subsection describes some of the approaches that have been used.
History of Transient Analysis Methods

Various methods of analysis were developed for the problem of transient flow in
pipes. They range from approximate analytical approaches whereby the nonlinear
friction term in the momentum equation is either neglected or linearized, to numerical
solutions of the nonlinear system. These methods can be classified as follows:
Arithmetic method: This method neglects friction (Joukowski, 1904; Allievi, 1903
and 1925).
Graphical method: This method neglects friction in its theoretical development but
includes a means of accounting for it through a correction (Parmakian, 1963).
Method of characteristics: This method is the most popular approach for handling
hydraulic transients. Its thrust lies in its ability to convert the two partial differential
equations (PDEs) of continuity and momentum into four ordinary differential equa-
tions that are solved numerically using finite difference techniques (Gray, 1953;
Streeter and Lai, 1962; Chaudhry, 1987; Elansary, Silva, and Chaudhry, 1994).
Algebraic method: The algebraic equations in this method are basically the two
characteristic equations for waves in the positive and negative directions in a pipe
reach, written such that time is an integer subscript (Wylie and Streeter, 1993).
Wave-plan analysis method: This method uses a wave-plan analysis procedure that
keeps track of reflections at the boundaries (Wood, Dorsch, and Lightner, 1966).
Implicit method: This implicit method uses a finite difference scheme for the tran-
sient flow problem. The method is formulated such that the requirement to maintain a
relationship between the length interval ' x and the time increment ' t is relaxed
(Amein and Chu, 1975).
Linear methods: By linearizing the friction term, an analytical solution to the two
PDEs of continuity and momentum may be found for sine wave oscillations. The lin-
ear methods of analysis may be placed in two categories: the impedance method,
which is basically steady-oscillatory fluctuations set up by some forcing function, and
the method of free vibrations of a piping system, which is a method that determines
the natural frequencies of the system and provides the rate of dampening of oscilla-
tions when forcing is discontinued (Wylie and Streeter, 1993).
Perturbation method: With this method, the nonlinear friction term is expanded in a
perturbation series to allow the explicit, analytical determination of transient velocity
in the pipeline. The solutions are obtained in functional forms suitable for engineering
uses such as the determination of the critical values of velocity and pressure, their
locations along the pipeline, and their times of occurrence (Basha and Kassab, 1996).
Section 13.3 Magnitude and Speed of Transients 585
13.3 MAGNITUDE AND SPEED OF TRANSIENTS

Using Equations 13.4 and 13.5, an engineer can calculate the magnitude of the change
in pressure for a given change in velocity. These pressure changes can be very large.
For example, for water in a pipeline with a = 3,200 ft/s (980 m/s), a change in velocity
of 3.3 ft/s (1 m/s) (not uncommon at every pump switch) can result in a pressure surge
of 330 ft (100 m) or 143 psi (980 kPa). With distribution system pressures on the
order of 60 psi (410 kPa), a positive pressure wave of this magnitude can raise the
pressure beyond the bursting strength of the pipe, while a negative pressure wave can
drop the pressure below the vapor pressure of the liquid.
A criterion used in determining which equation to use to evaluate a transient is the

pipeline characteristic time. The significance of this attribute is explained in the first
subsection. Next, this section introduces the Joukowsky equation, which is a formula
used to predict the magnitude of a transient. Transient magnitude depends on the
wave speed, which was introduced previously. This presentation is followed by a
comparison of rigid and elastic water hammer magnitude calculations and a discus-
sion of boundary and reflection methods.
Characteristic Time
The pressure wave generated by a flow control operation propagates with speed a and
reaches the other end of the pipeline in a time interval equal to L/a seconds. The same
time interval is necessary for the reflected wave to travel back to its origin, for a total
of 2L/a seconds. The quantity 2L/a is termed the characteristic time for the pipeline. It
is used to classify the relative speed of a maneuver that causes a hydraulic transient.
If a flow control operation produces a velocity change dV in a time interval (TM) less
than or equal to a pipeline’s characteristic time, the operation is considered “rapid.”
Flow control operations that occur over an interval longer than the characteristic time
are designated “gradual” or “slow.” The classifications and associated nomenclature
are summarized in Table 13.1.
Table 13.1 Classification of flow control operations based on system characteristic time
Operation Time Operation Classification
TM = 0 Instantaneous
T M d 2L e a Rapid
T M ! 2L ea Gradual
T M » 2L ea Slow
The characteristic time is significant in transient flow analysis because it dictates

which method is applicable for evaluating a particular flow control operation in a
given system. The rigid model provides accurate results only for surge transients gen-
erated by slow flow control operations that do not cause significant liquid compres-
sion or pipe deformation. Instantaneous, rapid, and gradual changes must be analyzed
with the elastic model.
Joukowsky’s Equation
In 1897, Joukowsky demonstrated the applicability of Equations 13.4 and 13.5 by
correctly predicting maximum line pressures and disturbance propagation times in his
experiments to determine the maximum velocities that should be allowed in the Mos-
cow water system with valves and other protection devices (surge tanks and relief
valves). Noting that Ugh = p , Equations 13.4 and 13.5 can be rewritten to relate
explicit head and flow changes as
a a
dH = r --- dV = r ------- dQ = r BdQ (13.9)
g gA
where H = head (ft, m)

a = characteristic wave speed of the liquid (ft/s, m/s)
V = fluid velocity (ft/s, m/s)
A = area (ft2, m2)
B = characteristic impedance, a/gA (s/ft2, s/m2)
The characteristic impedance factor, B, relates head changes to changes in flow. The
value of B depends on liquid and pipe characteristics and is defined as equal to (a/gA).
If the flow control change is executed rapidly (that is, the duration of the control
change is less than 2L/a), the time interval can be subdivided into shorter intervals,
and the individual head changes are then added to determine the total head change:
'H = ¦ dH = r B ¦ dQ = r B 'Q (13.10)
Celerity and Pipe Elasticity

In 1848, Helmholtz demonstrated that wave celerity in a pipeline varies with the elas-
ticity of the pipeline walls. Thirty years later, Korteweg developed an equation similar
to Equation 13.11 that allowed for determination of wave celerity as a function of
pipeline elasticity and liquid compressibility. When performing transient analyses
today, an elastic model formulation with a correction to account for pipeline elasticity
should be used.
E
-----v
U -
a = ----------------------- (13.11)
DE v
1 + ---------- \
eE
where a = characteristic wave speed of the liquid (ft/s, m/s)

E v = bulk modulus of elasticity for the liquid (lbf/ft2, Pa)
U = liquid density (slugs/ft3, kg/m3)

D = diameter (in., mm)
e = wall thickness (in., mm)
E = Young’s modulus for pipe material (lbf/ft2, Pa)
\ = pipeline support factor
Equation 13.11 is valid for thin walled pipelines (D/e > 40). The factor \ depends on
pipeline support characteristics and Poisson’s ratio. If a pipe is anchored throughout
against axial movement, \ = 1 - P 2, where P is Poisson’s ratio. If the pipe has func-
tioning expansion joints throughout, \ = 1. If the pipe is supported at only one end
and allowed to undergo stress and strain both laterally and longitudinally, \ = 5/4 - P
(ASCE, 1975). For thick-walled pipelines, there are theoretical equations proposed to
compute celerity; however, field investigations are needed to verify these equations.
The values shown in Table 13.2, and Table 13.3 for various pipeline materials and liq-
uids are useful to calculate celerity during transient analysis. Figure 13.4 provides a
graphical solution for celerity, given pipe wall elasticity and various diameter/thick-
ness ratios.
Table 13.2 Physical properties of some common pipe materials

Young’s Modulus
Material Poisson’s Ratio, P
(10 lbf/ft2)
9
(GPa)
Steel 4.32 207 0.30
Cast Iron 1.88 90 0.25
Ductile Iron 3.59 172 0.28
Concrete 0.42 to 0.63 20 to 30 0.15
Reinforced Concrete 0.63 to 1.25 30 to 60 0.25
Asbestos Cement 0.50 24 0.30
PVC (20 ) o
0.069 3.3 0.45
Polyethylene 0.017 0.8 0.46
Polystyrene 0.10 5.0 0.40
Fiberglass 1.04 50.0 0.35
Granite (rock) 1.0 50 0.28
Table 13.3 Physical properties of some common liquids
Temperature Bulk Modulus of Elasticity Density

Liquid
(oC) 6 2
(10 lbf/ft ) (GPa) 3
(slugs/ft ) (kg/m3)
Fresh Water 20 45.7 2.19 1.94 998
Salt Water 15 47.4 2.27 1.99 1,025
Mineral Oils 25 31.0 to 40.0 1.5 to 1.9 1.67 to 1.73 860 to 890
Kerosene 20 27.0 1.3 1.55 800
Methanol 20 21.0 1.0 1.53 790
Figure 13.4
Celerity versus pipe
wall elasticity for
various D/e ratios
For pipes that exhibit significant viscoelastic effects (for example, plastics such as
PVC and polyethylene), Covas et al. (2002) showed that these effects, including
creep, can affect wave speed in pipes and must be accounted for if highly accurate
results are desired. They proposed methods that account for such effects in both the
continuity and momentum equations.
Comparing the Elastic and Rigid Models

To compare the elastic and rigid models, it is only necessary to consider the friction-
less flow condition and verify that the pressure changes using each model follow the
relation:
dV
'p rigid L -------
dt -
--------------------- v -------------- (13.12)
'p elastic a dV
where 'p rigid = change in pressure computed with rigid model

'p elastic = change in pressure computed with elastic model
dV edt = fluid acceleration
Although similar, the differences between the elastic and rigid model equations are
significant and can be compared by examining the effect of a flow control operation
performed at the end of a pipe over a time interval dt that causes a velocity change dV.
In the elastic model, pressure changes depend on the flow control operation’s execu-
tion time compared to the pipeline’s characteristic time. In the rigid model, when a
rapid flow control operation occurs (dt o 0), the computed pressure change will be
excessive and will increase with pipe length (dx o L), even for small velocity
changes.
Both models produce similar results when dV o 0, which corresponds to a near

steady-state flow condition, or to a transient flow condition characterized by slow
pressure changes of the same magnitude as the head loss in the pipeline. Thus, as
stated earlier, the rigid model is acceptable for slow operational changes for which TM
is much longer than the characteristic time.
If a flow control operation produces a velocity change dV in a time interval shorter
than the characteristic time 2L/a (that is, the operation is “rapid”), the corresponding
pressure change is practically the same as an “instant” flow control change in low fric-
tion systems. This pressure change can be determined by using Equation 13.4 or 13.5.
Wave Reflection and Transmission

In addition to the equations describing transient flow, it is necessary to know about
the boundaries—such as tanks, dead ends, and pipe branches—that control the behav-
ior of the transient phenomena.
Hydraulic systems commonly have interconnected pipelines with differing character-
istics such as material and diameter. These pipeline segments and connection points
(nodes) define a system’s topology.
When a wave, defined by a head pulse ' Ho and traveling in a pipe, comes to a node, it
transmits itself with a head value ' Hs to all other connected pipes and reflects in the
initial pipe with a head value ' HR. The wave reflection occurring at a node changes
the head and flow conditions in each of the pipes connected to the node.
Figure 13.5 shows a node with four pipes connected to it. Part (a) shows the node as
the transient wave approaches, and part (b) shows the node after wave reflection and
transmission. If the distances between the pipe connections are small, the head at all
connections can be assumed to be the same (that is, the head loss through the node is
negligible), and the transmission (s) and reflection (r) factors can be defined as
A0
2 ------
'H s a0
s = ---------- = --------------
- (13.13)
'H 0 n A
¦ ---- -i
ai
i=0
where s = transmission factor (dimensionless)

'H s = head of transmitted wave (ft, m)
'H 0 = head pulse (ft, m)
A 0 = incoming pipe area (ft2, m2)
a 0 = incoming wave speed (ft/s, m/s)
A i = area of i-th pipe (ft2, m2)
a i = wave speed of i-th pipe (ft/s, m/s)
n = number of outgoing pipes
i = pipe number index
Figure 13.5
Pipe connections
'H
r = ----------R- = s – 1 (13.14)
'H 0
where r = reflection factor (dimensionless)
'H R = head of reflected wave (ft, m)
These factors are used to determine how waves are reflected and transmitted at each
branch and boundary. The expressions for r and s are obtained using Joukowsky’s
equation (Equation 13.9) for several pipes connected to the same node and consider-
ing flow continuity before and after the arrival of the wave.
Calculation approaches for evaluating transmitted and reflected waves in typical

hydraulic system scenarios follow.
• Pipe connected to a reservoir: In this case, n = 1 and A 1 o f. So, s = 0 and

r = -1. In other words, a wave reaching a reservoir reflects with the opposite
sign. Because 'HR = - 'H 0 and H f = H 0 + 'H 0 + 'H R , H f = H o in this case.
( H f represents final head—the head after wave transmission/reflection.) This
scenario is depicted in Figure 13.6(a).
• Pipe connected to a dead-end or closed valve: In this case, n = 1, and

through the derivation of an equation for r similar to Equation 13.13, it can
be shown that r = 1. In other words, a wave reflects at a closed extremity of a
pipe with the same sign, and therefore, head amplification occurs at that
extremity. If a flow control operation causes a negative pressure wave that
reaches a closed valve, the wave’s reflection causes a further reduction in
pressure. This transient flow condition can cause liquid column separation
and in low head systems, potential pipeline collapse. Figure 13.6(b) shows
that at a dead end, the wave is reflected with twice the pressure head of the
incident wave.
• Pipe diameter reduces (celerity increase): In this case, A1 < A0, and s > 1,
so the head that is transmitted is amplified. For example, if A1 = A0/4 (or D1 =
D0/2), then s = 8/5=1.6 and r = s – 1 = 0.6, and the head transmitted to the
smaller pipeline is 60 percent greater than the incoming. The larger pipeline
will also be subjected to this head change after the wave partially reflects at
the node. The effect of a contraction is illustrated in Figure 13.6(c).
• Pipe diameter increases (celerity decrease): In this case, an attenuation of

the incident head occurs at a pipeline diameter increase. The smaller pres-
sure wave is transmitted to the larger pipeline, and after the reflection, the
smaller pipeline is subjected to the lower final head. Figure 13.6(d) shows
that at an expansion, only some of the wave is reflected.
Figure 13.6
Transmission and
reflection factors
• Pipe with a lateral withdrawal: This case refers to a pipeline arrangement

in which a withdrawal pipe or surge tank designated as pipe 1 is connected to
a secondary network. In this case, n = 3, and with a0 = a1 = a2 and A0 = A2, it
can be shown that
1
s = ------------------- (13.15)
2
D1
1 + --------- -
2
2D 0
1
r = s – 1 = – ------------------- (13.16)
2
2D 0
---------- + 1
2
D1
where D1 = diameter of the lateral pipe (ft, m)

D0 = incoming pipe diameter (ft, m)
The existence of a lateral withdrawal or feed connection will always reduce the head
transmitted and decrease the system head (0 < s < 1). In this case, the transmitting fac-
tor can be referred to as a smoothing factor and can be used to determine the prelimi-
nary diameter of a surge tank needed to absorb incident waves without significant
pressure wave transmission downstream.
For example, if 0.05 < s < 0.10, then 6.2D0 > D1 > 4.2D0. With this preliminary esti-
mate of the surge tank size, the engineer can apply the rigid model to analyze the tran-
sient flow condition downstream, because the elastic propagation effect is minimal.
The example below illustrates the relationship between head and flow for the case of a
reservoir and valve on either end of a frictionless pipeline (that is, for an ideal case).
Example — Relationship Between Head and Flow. Figure 13.7 shows the evolu-
tion of a hydraulic transient that is initiated by the complete and instant closure of a valve and causes
expansion and contraction of the pipeline and the liquid, which has a specific weight J 0 . A single
wave is followed through a period 4L/a as it travels through a single frictionless pipe with the closed
valve at one end and a reservoir at the other. The wave reflections at the reservoir and at the closed
valve show the head and flow direction changes that occur with time.
Descriptions of the individual steps in the progression of the transient wave follow. These steps corre-
spond to those shown in Figure 13.7.
a) At time 0 < t < L/a, the wave front is moving toward the reservoir. To the right of the front, the
water has stopped and the pressure has increased. To the left of the front, the water does not yet
“know” that the valve was shut, so it continues to move to the right at the initial head.
b) At time t = L/a, the wave front has reached the reservoir and all the water in the pipe has stopped
and is compressed. However, the head in the pipe is above the water level in the reservoir. This dif-
ference in head must be relieved, so the water begins to move to the reservoir.
c) At time L/a < t <2L/a, the wave front moves toward the valve, and water to the left of the front
moves toward the reservoir. Water to the right of the front is motionless and is compressed.
d) At time t = 2L/a, the wave front has reached the valve and water is moving away from the valve
toward the reservoir. Of course, the water cannot continue to move away from a dead end, so
another wave cycle begins.
Figure 13.7
Valve closure in a
frictionless system
e) At time 2L/a < t < 3L/a, the wave front is moving away from the valve. To the right of the front,
pressures are below static pressure and velocity is zero. To the left, velocity continues in the direc-
tion of the reservoir, but the pressure is static.
f) At time t = 3L/a, the wave has again reached the reservoir. However, the head in the pipe is below
the water level in the reservoir and the water is at a low density. Another wave cycle must start.
g) At time 3L/a < t < 4L/a, the wave is once again moving back toward the valve. This time, the pres-
sure to the left is at the static value and water is moving into the pipe. To the right, velocity is zero
and the pressure is below static.
h) At time t = 4L/a, the wave has reached the closed valve again and conditions are the same as they
were at t = 0. The wave will start again and would continue indefinitely if not for friction and other
energy dissipation mechanisms that will eventually dampen the wave.
Figure 13.8 shows the change in head over the course of the transient at two key locations. Figure
13.8(a) illustrates the situation at the valve including the heads that would be measured at the valve
during the event as the wave moves back and forth through several cycles. Figure 13.8(b) shows the
head variation at a point D, which is located roughly midway along the pipeline, s units from the
valve.
Figure 13.8
Variation of head at
key points during
valve closure
Figure 13.9 summarizes the results and associates the transient heads to the x-t plane. Each point in
this plane corresponds to a location and time within the transient event. The line for “First distur-
bance” shows the position of the wave front given that the valve closure occurs at position L at time
zero. Points in the gray area will not “know” that the valve has been closed. The lines with slope a or
–a show the position of the wave front and are referred to as characteristic lines.
Figure 13.10 shows a three dimensional plot of head as a function of time and position for a friction-
less pipeline with no packing or attenuation. Similar plots could be made for pressure, flow, and
velocity.
At any point in the system, the head and flow change with time in an uneven cyclic manner, known as
the characteristic period (4L/a) of the transient phenomenon.
In an actual system, friction causes the pressure wave to decay; however, the period of the wave is the
same. A plot of pressure versus time allows the engineer to determine the celerity in a system of
known length using the equation T = 4L/a, as shown in Figure 13.11.
Figure 13.9
Characteristic lines in
the x-t plane
Figure 13.10
Three-dimensional
representation of the
characteristic lines
Figure 13.11
Celerity determination
Attenuation and Packing

In a system without friction or tanks to dampen transients, transients could conceiv-
ably persist indefinitely. However, viscous and friction effects and loss of momentum
in tanks typically cause transients to attenuate within seconds to minutes.
Joukowsky’s equation (Equation 13.9) enables the engineer to compute the increase
in head that occurs due to the rapid closure of a downstream valve in a frictionless
system having an initial flow of Q0. The increase in head 'Ho = aQ0/gA = aV0/g is
referred to as the potential head change or potential surge.
Because friction does exist in an actual system, the potential head change calculated
using the Joukowsky equation underestimates the actual head rise. This underestima-
tion is due to packing—an additional increase in head occurring at the valve as the
pressure wave travels upstream.
Consider a pipe with a liquid flowing at V0 that connects a reservoir to a valve, similar
to the system shown at the top of Figure 13.12. As the wave travels upstream from a
closed valve, packing occurs because the hydraulic gradient in the pipe still exists
after the wave front passes. For an instantaneous closure or pump trip, an abrupt wave
moves upstream. Once the wave passes a point, the velocity downstream of the front
ostensibly goes to zero. However, a pressure gradient still exists in that section of the
pipe, and it causes a small (but nonzero) flow to continue toward the closed valve.
This additional water packs against the closed valve, resulting in an additional pres-
sure increase above the potential surge, a(V0)/g.
The small velocity behind the wave front means that the velocity difference across the
wave front is less than V0, with the effect that the pressure change is progressively less
than the potential surge as the wave travels upstream. This effect, which is concurrent
with line packing, is called attenuation or reduction.
Both line packing and attenuation are present continuously in real hydraulic systems.
The effect of attenuation and packing can be observed by solving the elastic wave
equations with and without the friction term. The difference between the two solu-
tions indicates the effect of packing and attenuation.
Figure 13.12 illustrates the change in system head over successive time increments.
The system consists of a 20-km-long pipeline with a diameter of 500 mm that carries
water from a reservoir with a water level of 100 m to a distribution reservoir at sea
level (reference datum). The internal roughness of the pipeline is 0.25 mm ( f # 0.0175
for turbulent flow), the celerity is 1,000 m/s (2L/a = 40 seconds), and the flow rate is
approximately 330 l/s.
A transient is caused by a rapid linear valve closure (20 seconds) of a 500-mm globe
valve located at the downstream end of the pipeline. Because the friction loss coeffi-
cient (fL/D = 700) is considerably larger than the loss coefficient (K = 10) of an open
globe valve, most of the available energy is being lost along the pipeline in this sys-
tem. Thus, the system may be categorized as a high-friction system. Additionally, we
know that the effective stroke during closure is primarily at the end of the closure
period, due to the relatively low value of K over much of the valve’s travel. Although
the valve closure time for the system being considered is 20 seconds, the effective clo-
sure time is much less. A rapid and significant rise in head at the valve is therefore
anticipated at a time of approximately 20 seconds. The series of graphs in Figure
13.12 shows the evolution of the wave front traveling along the pipeline with packing
occurring near the valve and attenuation occurring at the wave front.
Figure 13.12
Wave propagation,
packing, and
attenuation due to
valve closure
In Figure 13.13, the head change at the valve is plotted versus time. In this figure, it
can be verified that the effect of closing the valve occurs within the 20-second closing
period and the effective valve closure time is less than 2 seconds.
Figure 13.13
Head change versus
time at the valve
13.4 NUMERICAL MODEL CALIBRATION

Transient flow problems are usually solved using hydraulic transient analysis com-
puter models. Given descriptions of the system and the event triggering the transient
and information on the boundary conditions, the model can determine fluid velocity V
(and flow Q) and pressure p (and head H).
Comparisons of computed results from a transient analysis program with results mea-
sured during laboratory experiments or field tests from actual systems have been used
to validate transient analysis programs. These comparisons require that well-defined
flow control operations be used to allow proper simulation and calibration of the com-
puter model.
For systems having free gas in the liquid and the potential for liquid column separa-
tion, the theoretical transient analysis is more complex, and the computed results have
more uncertainty. It is impossible to develop a theoretical model that accurately simu-
lates every physical phenomenon that can occur in an actual hydraulic system. There-
fore, all modeling of transients involves some approximation and simplification of the
real problem.
The differences between computer model results and actual system measurements are
caused by several factors, some of which are highlighted in the following list:
• Precise determination of the celerity of the piping system is impossible. This
is especially true for buried pipelines, which are influenced by bedding con-
ditions and the compaction of the surrounding soil.
• Precise modeling of dynamic system elements (such as valves, pumps, and
protection devices) is difficult because they are subject to deterioration with
age and adjustments made during maintenance activities.
• Differences exist in the friction coefficients for a steady-state condition ver-
sus a transient condition. In a transient condition, flow direction changes,
Section 13.4 Numerical Model Calibration 601
and changes in velocity gradients modify the shear stress in the pipeline. The
friction coefficient used in the model should ideally account for the localized
and transport accelerations.
• Prediction of the presence of free gases in the system liquid is sometimes
impossible. These gases can significantly affect the celerity and propagation
of waves. In addition, the occurrence of column separation and vapor cavity
formation are difficult to accurately simulate.
The first two items can be eliminated as error factors by calibrating the model data
and by carefully representing the operational characteristics of dynamic system ele-
ments. Unsteady flow friction coefficients and the effects of free gases are more diffi-
cult to account for in theoretical transient analysis. Available computer models are
limited in their ability to accurately model these factors, but model improvements are
ongoing.
Fortunately, in water supply and distribution systems, friction effects are usually
minor and vaporization conditions can be avoided by installing proper protection
devices. For this reason, the numerical transient model, although limited, is an ade-
quate and essential tool in the analysis, design, and operation of the hydraulic system.
The operational risks can be evaluated, flow control operations specified and opti-
mized, and protection devices sized such that the extreme transient heads are con-
trolled to acceptable limits for each particular system.
Analysis of field measurements will clearly indicate the evolution of the transient. If
the period of the transient (4L/a) shown in Figure 13.11 is recorded, and the length (L)
between measurement locations is known, then the celerity can be determined. If air
is in the system, the celerity measured may be much lower than the theoretical celer-
ity.
If friction is significant in a system, the transient attenuation measured in the actual
system will usually be greater than the attenuation computed by the theoretical
numerical simulation, particularly during longer time periods (t > 2L/a). Poor friction
representation will not explain lack of agreement of the initial potential surge in a
rapid transient. With respect to timing, there should be close agreement between the
computed and measured periods of the system, regardless of what flow control opera-
tion initiated the transient. With a well-calibrated model of the actual system, it is pos-
sible to use the model in the operational control of the system and anticipate the
impacts of specific flow control operations.
A calibrated model can also be used as part of an “outflow” detection system. Model
simulation results from a specific flow control operation can be compared to actual
system results at particular nodes in the system. Significant differences between com-
puted and measured values may indicate an outflow (such as an open valve, leak, or
pipeline break) that can be rapidly located, evaluated, and corrected.
In general, if model peaks arrive at the wrong time, the wave speed needs to be
adjusted. If model peaks have the wrong shape, the description of the control event
(pump shutdown or valve closure) should be adjusted. If the transient dies off too
quickly or slowly in the model, the friction losses need to be adjusted. If there are sec-
ondary peaks, important loops and diversions may need to be included in the model.
13.5 GATHERING FIELD MEASUREMENTS

Calibration of transient numerical models depends on adequate field instrumentation
for dynamic measurement of pressures and velocities in the actual system. Figure
13.14 shows a schematic of field equipment and instrumentation for obtaining infor-
mation on a particular location in a piping system.
Rather than conventional pressure gages and SCADA systems, high-speed data log-
ging equipment is needed to accurately track transient events. The pressure transducer
should be very sensitive, have a high resolution, and be connected to a control and
data acquisition unit. It should be connected to the system pipeline with a device to
release air, because air can distort the pressure signal transmitted during the transient.
The control unit allows a pressure measurement interval to be established and the
electronic signal transmitted from the pressure transducer to be converted to a pres-
sure reading.
The high-resolution data acquisition system records pressure information during the
transient event for later review and interpretation. With additional programming of the
data acquisition system, system celerity and extreme pressure heads can be deter-
mined, and, if necessary, the data acquisition system software can perform a harmonic
analysis to determine the predominant frequencies of the water distribution system.
Recording of pressure should not begin until all air is released from the pipeline con-
nection and the pressure measurement interval is defined. Typically, at least two pres-
sure measuring locations should be established in the system, and the flow control
operation should be closely monitored. The timing of all recording equipment must
be synchronized. For valves, the movement of the position indicator is recorded as a
function of time. For pumps, rotation or speed is measured with time. For protection
devices such as one-way and two-way surge tanks and hydropneumatic tanks, the
level is measured with time.
13.6 TRANSIENT CONTROL

Ideally, a system will be designed and operated to minimize the likelihood of damag-
ing transient events. However, in reality, transients will still occur; thus, methods for
controlling transients are necessary. This section has two goals: (1) to make the
hydraulic engineer aware of the system conditions that lead to the development of
undesirable transients, such as pump and valve operation and air pockets that can
form at high points, and (2) to present the protection methods and devices that should
be used during design and construction of particular systems and discuss the practical
limitations of these devices.
Two possible strategies for controlling transient pressures exist. The first is to focus
on minimizing the possibility of transient conditions during project design by specify-
ing appropriate system flow control operations and avoiding the occurrence of emer-
gency and unusual system operations. The second is to install transient protection
devices to control potential transients that may occur due to uncontrollable events
such as power failures and other equipment failure.
Section 13.6 Transient Control 603
Figure 13.14
Instrumentation to
record transient
pressures
Systems that are protected by adequately designed surge tanks (see page 608 for more
information) are generally not adversely impacted by emergency or other unusual
flow control operations because operational failure of surge tank devices is unlikely.
In systems protected by hydropneumatic tanks, however, an air outflow or air com-
pressor failure can occur and lead to damage from transients. Consequently, potential
emergency situations and failures should be evaluated and avoided to the extent possi-
ble through the use of alarms that detect device failures and control systems that act to
prevent them.
With most small, well-gridded water distribution network piping, sufficient safety
factors are built into the system, such as adequate pipe wall thickness and sufficient
reflections (tanks and dead ends) and withdrawals (water use). The effects of tran-
sients are most likely to result in pipe failures in long pipelines with long characteris-
tic times (large values of 2L/a), high velocities, and few branches. Filion and Karney
(2002) found that water usage and leaks in a distribution system can result in a dra-
matic decay in the magnitude of transient pressure effects.
Piping System Design and Layout

When designing water distribution systems, the engineer needs to consider economic
and technical factors such as acquisition of property, construction costs, site topogra-
phy, and geological conditions of the land where the piping system will be con-
structed. In addition, emergency flow control scenarios should be analyzed and tested
during the design phase because they affect the piping system design and the specifi-
cation of system equipment.
Pipeline system layouts with undulating topographic profiles are common. For these
systems, it may be desirable to change the route of the pipeline to avoid high points
that are prone to air accumulation or exposure to low pressures (or both). If the mini-
mum transient head grade line is above the topographical profile of the piping system,
then transient protection devices are most likely unnecessary, thus minimizing con-
struction costs and operational risks.
Low-head systems are more prone to experience transient vacuum conditions and liq-
uid column separation than are high-head systems. If the system designer does not
account for the occurrence of low transient pressures in low-head systems, then a
pipeline with inadequate wall thickness may be specified, potentially leading to pipe-
line collapse even if the pipeline is buried in a well-compacted trench. For example,
low-head systems with buried steel pipelines and diameter/thickness ratios (D/e) more
than 200 should be avoided because of the risk of structural collapse during a transient
vacuum condition, particularly if the trench fill is poorly compacted.
Steel, PVC, HDPE, and thin-wall ductile iron pipes are susceptible to collapse due to
vapor separation, but any pipe that has been weakened by repeated exposure to these
events may experience fatigue failure. A pipe weakened by corrosion may also fail.
Where very low pressures are possible during transient events, the engineer may
choose to use a more expensive material to preclude the chance of collapse. For exam-
ple, for large-diameter pipes under high pressures, steel is usually more economical
than ductile iron. However, the engineer may select ductile iron because it is less sus-
ceptible to collapse. It is always best to avoid vapor pressure conditions through surge
protection measures regardless of the type of pipe used.
Piping systems constructed above ground are more susceptible to collapse than are
buried pipelines. With buried pipelines, the surrounding bedding material and soil
provide additional resistance to pipeline deformations and help the pipeline resist
structural collapse.
Another important consideration when designing a system to protect against hydrau-
lic transients is the use of air valves. Using air valves to avoid vacuum conditions
requires careful analysis of possible transient conditions to ensure that the air valve is
adequately sized and designed. Several cases cited in literature describe the collapse
of piping systems due to the failure of an air inlet valve that was poorly sized,
designed, or maintained. The potential for operational failures in air valves should not
be ignored.
Other factors that influence extreme transient heads are wave celerity and liquid
velocity. Selecting larger diameters to obtain lower velocities with the purpose of
minimizing transient heads is acceptable for short pipeline systems delivering rela-
tively low flows. However, for long pipeline systems, the diameter should be selected
to optimize construction and operating costs. Long piping systems almost always
require transient protection devices.
After considering these factors during the conceptual and preliminary designs of the
system, the project should move into the final design phase. Any changes to the sys-
tem during final design should be analyzed with the transient model to verify that the
previous analysis results and specifications are still appropriate.
Example — Analysis of a Piping System. A pumping station located at an elevation

of 690 m (2,263 ft) delivers 1 m3/s (35.3 ft3/s) of water from a suction well with a water surface eleva-
tion of 700 m (2,296 ft), as shown in Figure 13.15. The water is delivered through a check valve and
2,500 m (8,200 ft) of 800-mm (31-in.) pipe to a reservoir with a water surface elevation of 765 m
(2,510 ft). The wave speed a is approximately 980 m/s (3,220 ft/s). The pump station includes a dou-
ble suction pump that operates at 880 rpm and is driven by a 1,000 kW (1,341 HP) motor. The com-
bined inertia of the pump and motor is approximately 150 kg-m2 (3,562 lbm-ft2).
Figure 13.15
Elevation view of
pumping system
The pump is started at time t = 10 seconds and takes approximately 4 seconds to ramp up to full speed
[see Figure 13.16(a)]. A blow-off valve located at the pump discharge opens to relieve flow during
pump start-up, and then gradually closes to direct water down the transmission main [Figure
13.16(b)]. At time t = 80 seconds, a pump shutdown caused by a loss of electric power occurs. This
incident is indicated by the abrupt drop-off in speed and flow in Figures 13.16(a) and (b). The shut-
down is considered an emergency condition. Figure 13.16(c) shows the simulation results from a tran-
sient analysis computer program for the period including the pumping operations at time t = 10
seconds (pump start) and t = 80 seconds (pump failure).
The simulation results show that, following the pump start at t = 10 seconds, a steady-state flow con-
dition is established at approximately 60 seconds and lasts for about 20 seconds. Following the emer-
gency pump shutdown, the flow and head reduction produced at the discharge of the pump station
travels through the downstream transmission main. Depending on the elevation profile of the down-
stream transmission main, vacuum pressures and column separation could occur. This potential is
illustrated in Figure 13.16(d), which gives the envelope of heads that the pipeline will experience. The
line Hmax shows the maximum heads along the pipeline, Hmin shows the minimum heads, Ho shows the
initial head, and HE shows the ending head (that is, the head after the pump shuts off). The value of
820 m (2,690 ft) is the maximum head that will be experienced during the event provided that column
separation does not occur. To fully utilize this envelope, it is necessary to know the relative elevations
of the pipeline.
For this project, two options for routing the downstream transmission main exist, as shown in Figures
13.15 and 13.16(d). Profile A follows a lower elevation [approximately 660 m (2,165 ft)] route. For
this route, the piping system will not undergo transient vacuum pressures and will experience a maxi-
mum transient head of approximately 16 bars (232 psi), which is 4 bars (58 psi) higher than the 12-bar
(174-psi) head experienced during steady-state conditions. The maximum transient head is therefore
approximately 33 percent higher than the static pressure. The fact that the Hmin line remains above the
pipe profile in Figure 13.16(d) indicates that pressures will remain positive throughout the event.
Profile B follows a higher elevation [approximately 720 m (2,360 ft)] route. After the pump is shut
down, transient vacuum pressure conditions will occur with column separation in the piping system.
This separation is indicated by the Hmin line dropping below profile B in Figure 13.16(d). If this profile
is selected, a protection device(s) will certainly be required to control vacuum pressure conditions and
avoid the high-pressure transients that could result from the collapse of vapor cavities. The example
on page 609 will consider a protection device for this system.
Figure 13.16
Transients in a
pumping system
Protection Devices
To the extent possible, the engineer would like to design flow control equipment such
that serious transients are prevented. Using a transient model, the engineer can try dif-
ferent valve operating speeds, pipe sizes, and pump controls to see if the transient
effects can be controlled to acceptable levels. If transients cannot be prevented, spe-
cific devices to control transients may be needed.
Some methods of transient prevention include
• Slow opening and closing of valves: Generally, slower valve operating
times are required for longer pipeline systems. Operations personnel should
be trained in proper valve operation to avoid causing transients.
• Proper hydrant operation: Closing fire hydrants too quickly is the leading
cause of transients in smaller distribution piping. Fire and water personnel
need to be trained on proper hydrant operation.
• Proper pump controls: Except for power outages, pump flow can be slowly
controlled using various techniques. Ramping pump speeds up and down
with soft starts or variable-speed drives can minimize transients, although
slow opening and closing of pump control valves downstream of the pumps
can accomplish a similar effect, usually at lower cost. The control valve
should be opened slowly after the pump is started and closed slowly prior to
shutting down the pump.
• Lower pipeline velocity: Pipeline size and thus cost can be reduced by
allowing higher velocities. However, the potential for serious transients
increases with decreasing pipe size. It is usually not cost-effective to signifi-
cantly increase pipe size to minimize transients, but the effect of transients
on pipe sizing should not be ignored in the design process.
To control minimum pressures, the following can be adjusted or implemented:
• Pump inertia
• Surge tanks
• Air chambers
• One-way tanks
• Air inlet valves
• Pump bypass valves
To control maximum pressures, the following can be implemented:
• Relief valves
• Anticipator relief valves
• Surge tanks
• Air chambers
• Pump bypass valves
The items in the preceding lists are discussed in the subsections that follow. These
items can be used singly or in combination with other devices.
Pump Inertia. Pump inertia is the resistance the pump has to acceleration or
deceleration. Pump inertia is constant for a particular pump and motor combination.
The higher a pump’s inertia, the longer it will take the pump to stop spinning follow-
ing pump shutoff. Larger pumps have more inertia because they have more rotating
mass.
Pumps with higher inertias can help to control transients because they continue to
move water through the pump for a longer time as they slowly decelerate. This behav-
ior slows transient generation and can reduce the overall transient experience in a sys-
tem with a short pipeline if the generation time is longer than the characteristic time
(period) of the system. Pump inertia can be increased through the use of a flywheel.
For long systems, the magnitude of pump inertia needed to effectively control tran-
sient pressures makes this control impractical due to the mechanical problems associ-
ated with starting high inertia pumps. Therefore, increasing pump inertia is not
recommended as an effective option for controlling transient pressures for long piping
systems.
Air Chambers and Surge Tanks. Air chambers and surge tanks work by
bleeding water out of the system during high-pressure transients and adding water
during low-pressure transients. An air chamber is a pressure vessel that contains water
and a volume of air that is maintained by an air compressor [see Figure 13.17(a)].
When pumps are shut down and the flow and pressure decrease at the pump dis-
charge, the air in the chamber expands as a result of the pressure drop, and water
enters the system from the chamber.
A surge tank is a relatively small tank located such that the normal water level eleva-
tion is equal to the hydraulic grade line elevation [see Figure 13.17(b)]. The tank
feeds the system by gravity, and the outflow of water from the tank controls the mag-
nitude of the low-pressure transient generated at the pump discharge following a shut-
down.
Figure 13.17
Air chamber and
surge tank
The piping connection between the air chamber or surge tank and the system is sized
to provide adequate hydraulic capacity when the chamber is discharging, as well as to
cause a head loss sufficient to dissipate transient energy and prevent the chamber or
tank from filling too quickly. Both of these requirements are met through the use of a
piping bypass as depicted in Figure 13.17(a).
Example — Use of an Air Chamber for Transient Control. This example contin-
ues to examine the system on page 606. To make profile B viable, an air chamber with the preliminary
dimensions indicated in Figure 13.18(a) is suggested to control minimum and maximum transient
heads. Following the pump shutdown at t = 80 seconds, flow quickly discharges from the chamber to
minimize the drop in pressure.
The air chamber impacts the discharge from the pump as shown in the graph in Figure 13.18(b). Fast-
acting check valves must be used at the discharge of the pumps to avoid flow reversal through the
pumps. (For more information on check valves, see Combined Devices on page 611.)
The results for head at the discharge side of the pump are shown in Figure 13.18(c), and the pressure
envelope for the pipeline is shown in Figure 13.18(d). The pressure wave frequency and amplitude are
reduced in the system protected by the air chamber; therefore, a comparison of the heads shown in
Figures 13.16(d) and 13.18(d) indicates a much narrower range of heads when the air chamber is
used. Similarly effective results may be obtainable with other protection devices.
An economic analysis of the two profiles will help determine the best option. It should be noted that
even though option A does not require the installation of a protection device, it does not provide the
“smooth” changes in transient head that option B provides, which may have an impact on system
operation and maintenance.
As discussed in the previous example, the pressure responses resulting from system
options A and B should be considered with project costs when evaluating which
option is “best.” The engineer should compare the life-cycle costs of the alternative
routes with the costs for additional transient protection.
It is important to note that using air chambers and surge tanks in treated drinking
water systems can result in water quality deterioration and loss of disinfectant resid-
ual. These devices should be equipped with a mechanism for circulating the water to
keep it fresh. A further complication occurs when the tanks are located in cold cli-
mates where the water can freeze. If freezing is an issue, smaller air chambers that can
be housed in heated buildings are preferable.
One-Way Tank. A one-way tank is a storage vessel under atmospheric pressure

that is connected to the system with a check valve that is normally closed and only
allows flow from the tank into the system (see Figure 13.19). When a low-pressure
transient in the system reaches a one-way tank that has a greater head than the low-
pressure transient, the tank check valve opens to feed water from the tank into the sys-
tem. This action controls the magnitude of the low-pressure transient. After the tank
discharges into the system, a float switch triggers the opening of a valve to refill the
tank from the system through a separate connection. The significant advantage of
using a one-way tank rather than a surge tank is that the check valve allows the one-
way tank to have a much lower height.
Figure 13.18
Transient protection
due to an air chamber
Figure 13.19
A one-way tank
Combined Devices. If an air chamber has an air inlet valve installed on top of it,
it can double as a one-way tank. This combined device admits air into the chamber
during an extreme low-pressure transient and acts as an air cushion during a high-
pressure transient to control maximum pressure (see Figure 13.20). In some cases,
this combined protection device allows the size of the air chamber to be optimized.
The combined protection device can allow the use of a smaller air chamber sized for
“normal” low-pressure transients, but it can still protect the system against “extreme”
low-pressure transients when the volume of air in the chamber is insufficient.
Figure 13.20
Air chamber with an
air inlet valve
These protection devices require special attention during the design and specification
of the equipment and the control system. One area of concern is temperature. In an air
chamber, temperature changes cause the expansion and compression of air in the
chamber. Excessive temperature variations can cause damage to the chamber’s inter-
nal coating and should be considered during design.
Another item that requires special attention is the pump check valve. The pump check
valve should have a fast closing time to prevent flow reversal through the pump and
the valve slam that can occur with delayed valve closure and where surge tanks are
incorporated into the pump station design. Valve slam can damage the valve, pump, or
system piping upstream of the air chamber. If it is not possible to have a check valve
that closes before the surge tank responds and slams the valve, some type of dampen-
ing device, such as a dash pot, is necessary to control valve closure during the last 5 to
10 percent of the valve travel.
Figure 13.21 shows how air chambers, one-way tanks, and combined devices affect
transient conditions. Without these devices, the system pressure would drop so low
that column separation would occur. With the devices, system pressures remain posi-
tive. The initial HGL in the systems corresponds to the gently sloping line labeled
“steady-state regime.” When the pump is turned off, the HGL begins to drop (repre-
sented by the decreasing heads of the solid curved lines in the drawings), starting on
the left side. Without protection devices, the HGL can drop well below the pipe, indi-
cating the potential for column separation and severe damage. With the devices, the
HGLs remain above the ground.
In Figure 13.21(a), which shows the impacts of protection with an air chamber, the
HGL never drops below the actual pipeline elevation. However, without the air cham-
ber, extremely low pressures with column separation can occur (see dashed HGL).
Figure 13.21(b) shows the impact of protecting a system with one-way surge tanks.
Although they allow the HGL to drop lower than the air chamber, they do maintain a
positive pressure in the pipe. In addition, they are usually less expensive than pressur-
ized air chambers, but requires continuous maintenance on the installed system.
Figure 13.21(c) shows the impact of protecting a system with a combination of air
chambers and one-way surge tanks. With both measures in place, a positive pressure
is maintained in the pipe. However, with only the one-way surge tank in place, col-
umn separation could occur in the area where the HGL drops below the pipe profile).
Pressure Relief and Other Regulating Valves. Typically, if the decrease
in pressure caused by a transient is insufficient to cause vacuum conditions in the sys-
tem, the resulting positive transient may not be excessive and additional high pressure
protection devices might not be required. In some cases, however, pressure relief
valves must open quickly if the system pressure reaches a pre-established maximum
pressure setting. The relief valve opens to discharge water, thus controlling the maxi-
mum system pressure. After the high pressure is relieved, the valve closes slowly to
avoid creating a transient condition. If a storage facility exists on the suction side of
the pumps, water is usually discharged to the tank, though it could also be discharged
into the pump suction line or even to the atmosphere for systems without a tank.
An anticipator relief valve can be used instead of a pressure relief valve to control
high-pressure transients. This type of relief valve starts to open immediately follow-
ing an emergency pump shutdown in anticipation of a high-pressure transient. The
anticipator valve is already open when the high-pressure transient reaches the valve;
sensing of high pressure is not required to initiate the opening of the valve. This type
of valve is more effective when high-pressure transients occur quickly and the limited
opening time of a relief valve is not adequate. Care must be taken in setting the low-
pressure activation point to avoid premature opening before the pump has spun down,
which can cause a very steep negative transient wave.
Air inlet valves are installed at high points along the pipeline system to control vac-
uum conditions and potential column separation. Following the low-pressure tran-
sient, the air that enters the pipeline should be expelled slowly to avoid creating
another transient condition. An adequate period of time should be allowed for the air
to be expelled before the pumps are restarted. A wide variety of valves are available
that enable air to enter and leave the system. (Their names vary with the manufac-
turer.) These valves include air inlet valves, air release valves, vacuum relief valves,
air vacuum valves, and vacuum breaker valves. Engineers should carefully review the
air valve manufacturer’s technical information when selecting an appropriate air valve
for transient control.
Figure 13.21
Effect of protection
devices on system
HGL
Booster Pump Bypass. Another type of protection device is the pump bypass.
Figure 13.22 shows a booster pumping system with bypass. When the booster pumps
shut down, the resulting reduction in flow generates two pressure waves. The wave
traveling upstream is a positive transient, and the wave that travels downstream is a
negative transient.
Figure 13.22
Booster pumping
system with bypass
Depending on the relative lengths of the upstream pipeline (LS) and the downstream
pipeline (LR) and the magnitude of the velocity changes, a pump bypass connection
can act as a transient protection element. Water continues past the booster station if
the downstream pressure falls below the upstream pressure, thus limiting the pressure
rise upstream of the booster station and the pressure drop downstream of the booster
station.
Figure 13.23 shows the transient analysis results for the system shown in Figure
13.22. These results show that the bypass opened to transfer water from the upstream
pipeline to the downstream pipeline, which helped to attenuate or control the maxi-
mum and minimum pressure transients on the upstream and downstream sides of the
station.
The effectiveness of using a booster station bypass depends on the specific booster
pumping system and the relative lengths of the upstream and downstream pipelines. If
the low-pressure surge generated on the discharge side of the pump is still greater than
the high-pressure surge generated on the suction side of the pump (tends to occur if LR
< LS), the bypass will not open. For systems in which the bypass may not open, other
transient protection devices are necessary. Each system should be individually ana-
lyzed to assess the occurrence of excessively high- and/or low-pressure transients and
determine strategies to control potentially excessive pressures.
Section 13.7 Operational Considerations 615
Figure 13.23
Booster pump
shutdown
13.7 OPERATIONAL CONSIDERATIONS

In order to use a transient computer model, the hydraulic engineer must have a
detailed understanding of the system equipment, operations, and operational con-
straints. He or she needs to ensure that the equipment and protection devices specified
for the system will in fact operate as predicted by the transient analysis.
The use of computer models to perform transient analysis of a hydraulic system
requires the following:
• Obtaining accurate information about the system equipment and operations
• Determining the operational characteristics of system flow control equip-
ment and transient protection devices
• Verifying the operational limitations of flow control equipment and transient
protection devices
This section discusses the valves most commonly used to control flow in hydraulic
systems. The key factors that need to be considered when evaluating valves are head
loss, cavitation, and valve operator effects. Each of these factors influences both the
steady-state and transient flow conditions for flow control valves.
There are many documented cases of poorly designed control valves. Some of these
valves do not operate adequately because of excessive head loss or cavitation during
steady-state flow conditions; others are inadequate to control hydraulic transients
because of poor valve selection or poor operation. When designing control valves for
flow control and/or pumping stations, the type, number, and size of valves to be used
should be carefully evaluated to ensure that adequate steady and transient flow regula-
tion characteristics are provided.
Even with a comprehensive understanding of the system equipment and operations,
the engineer should realize that it may not be possible to precisely model the actual
system and system components. Therefore, it is the engineer’s responsibility to recog-
nize these modeling limitations, use appropriate safety factors, and apply good engi-
neering judgment when performing transient analysis.
Flow Control Stations

Flow control stations typically include a flow meter, flow control valve, and valves to
isolate the station during maintenance activities. Flow control stations are sometimes
equipped with a remote terminal unit (RTU), which communicates with a Supervisory
Control and Data Acquisition (SCADA) system, to monitor and control the station
remotely. (For more information about SCADA systems, see Chapter 6 and Appendix
E.)
The transient pressures that result from the operation of flow control valves depend on
the design of the flow control station, particularly:
• The time period of the valve position change
• The valve type and its hydraulic characteristics
• The system hydraulic characteristics (for example, head loss in the piping
relative to head loss through the valve)
When considering valve position change, it is important to consider that the reduction
in flow due to valve closure is not proportional to the valve travel distance (stroke). In
fact, with most valves (including hydrants), most of the change in velocity occurs
when the valve is barely open. It is at this time that a quick turn of the valve can lead
to a significant water hammer event. For example, if it takes 20 turns to close a valve
and the initial velocity through the valve is 16 ft/s (5 m/s), the velocity may change to
6.6 ft/s (2 m/s) over the first 19 turns. The velocity is then reduced from 6.6 ft/s to
zero over the last turn (known as the “effective stroke” of the valve). The change of
velocity over the last interval having a duration equal to the characteristic time (2L/a)
determines the magnitude of the transient.
One of the most important considerations when selecting the flow control valve type
is cavitation. Cavitation occurs when the minimum pressure at critical points within
the valve reaches the vapor pressure (pv) of the liquid, and vapor bubbles form. If the
differential pressure across the valve is excessive, or if the pressure downstream of the
valve is minimal, cavitation can occur during the steady-state flow condition. Cavita-
tion can damage the valve and cause excessive noise, especially if an inappropriate
valve is selected. Control valves specifically designed to minimize the potential for
cavitation should be selected for these cases.
Depending on its severity, cavitation can also affect the hydraulic capacity of the
valve. When the flow stream expands immediately downstream of the valve, the pres-
sure increases, causing the vapor bubbles to collapse. This dynamic vaporization and
collapse phenomenon causes noise and vibration and can erode the interior of the
valve.
To completely eliminate valve cavitation, the head loss across the valve must be
reduced, or the downstream pressure must be increased. However, these requirements
may not be feasible for a particular valve station. Limited cavitation during critical
flow conditions is acceptable. To avoid excessive maintenance and repairs, valve
materials that are resistant to cavitation, such as stainless steel, should be specified in
these cases. If the required head loss across the valve cannot be reduced, and the
downstream pressure cannot be increased to reduce valve cavitation, a valve specially
designed for these extreme hydraulic conditions, such as a multijet valve, should be
used.
The last key design consideration concerns system hydraulic characteristics. The fric-
tional losses in the piping system can cause an attenuation of the wave front and pack-
ing near the valve, which in turn influences the transient heads that are produced and
propagated through the system. The friction of the piping system also impacts the
valve position necessary to produce the desired flow rate.
Figure 13.24 compares the hydraulic characteristics of various flow control valves,
including a Howell Bunger valve, which is designed for free discharge energy dissipa-
tion. The graph shows how the discharge coefficient for different types of valves
changes with the size of the opening as a percentage of the fully open position. The
discharge coefficients can be used in calculations to determine flow rate as a function
of valve position.
Automatic Control Valves. Valves that do not require an external source of

energy to operate are referred to as automatic control valves. Automatic control
valves open and close based on system pressure. The valve body usually has a globe
or angle pattern, and the internal operator is connected to a diaphragm or a piston sit-
uated in the valve body. Hydraulic pilot controls and piping use the system pressure to
control valve position by directing water to either side of the diaphragm or piston
operator. Depending on the selection and arrangement of the hydraulic pilot controls,
the valve can be designed to perform specific functions, such as maintaining a con-
stant downstream or upstream pressure or maintaining a constant flow rate.
For an automatic control valve to operate properly, it must be installed at a location

within the hydraulic system that has adequate pressure to overcome the weight and
friction associated with the internal valve operator. Valve location is typically not an
issue because it does not take much pressure spread over the area of the diaphragm or
piston operator to produce a force sufficient to overcome this internal operator resis-
tance. Periodic valve inspections are necessary to ensure that nothing that obstructs
operation has become lodged in the valve seat and that strainers or filters in the con-
trol piping are kept clean.
The inherent design of automatic control valves limits their ability to quickly respond
to rapid transients. Upon sensing the system pressure change due to a transient, the
hydraulic pilot controls direct control water to the internal valve operator. The time
required for the direction of the water limits the response time of automatic control
valves, usually in proportion to valve size, as larger valves require greater volumes of
control water. However, some automatic control valves can be equipped with fast-act-
ing features to stop reverse flow.
Figure 13.24
Flow control valve
characteristics
For control valve stations that are required to break a considerable amount of head
across the valve, significant turbulence can occur in the valve control piping if the pip-
ing is not adequately designed. This turbulence can cause an oscillatory phenomenon
that inhibits the valve’s ability to maintain a steady-flow condition in the system.
Also, cavitation can occur in the control piping, causing excessive wear and prema-
ture failure of the control pilot valves.
An automatic control valve can be designed to operate as a pressure relief valve that
discharges water from the system if a maximum system pressure is exceeded. How-
ever, because of its limited response time, the effectiveness of this type of pressure
relief valve in controlling system transients is also limited.
Check Valves. There are several types of check valves available for the preven-
tion of reverse flow in a hydraulic system. Required check valves should be carefully
selected to ensure that their operational characteristics (such as closing time) are suf-
ficient for the transient flow reversals that can occur in the system. Some transient
flow reversal conditions can occur very rapidly; thus, if a check valve cannot respond
quickly enough, it may slam closed and cause the valve or piping restraints to fail.
Check valves that have moving discs and parts of significant mass have a higher iner-
tia and therefore tend to close more slowly upon flow reversal. Check valves with
lighter checking mechanisms have less inertia and therefore close more quickly.
External counterweights present on some check valves (such as swing check valves)
are intended to assist the valve in closing following stoppage of flow. However, for
systems that experience very rapid transient flow reversal, the additional inertia of the
counterweight can slow the closing time of the valve. Spring-loaded check valves can
be used to reduce closing time, but these valves have higher head loss characteristics
and can induce an oscillatory phenomenon during some flow conditions.
It is important that the modeler understands the closing characteristics of the check
valves being used. For example, ball check valves tend to close slowly, swing check
valves close somewhat faster (unless they are adjusted otherwise), and nozzle check
valves have the shortest closing times. Modeling the transient event with different
types of check valves can indicate whether a more expensive nozzle-type valve is
worthwhile.
In summary, transient analysis is needed to evaluate water column deceleration at
check valve installations in order to understand how quickly the water column will
reverse and thus ensure that a check valve with an adequate closing time is selected to
prevent flow reversal.
Air Release Valves

Air that accumulates at high points in a water supply system and air that exists in a
vertical pump column before the pump is started should be appropriately released
from the system to prevent flow restrictions and transients that can occur when air
pockets in the system are dislodged.
Air can enter a water system in several ways. Water that contains dissolved air will
slowly release this air, which gradually accumulates at system high points. Also, vor-
texing at the open water surface on the suction side of some pumping stations will
admit air into the system. If portions of the system are drained and taken out of ser-
vice for repair or maintenance, air will often be trapped in the system before it is
returned to service. Air can also be admitted into the system through air inlet or relief
valves, which are commonly used to control low-pressure transients and prevent
excessive vacuum conditions in the system.
An air release valve (see Figure 13.25) used to control low-pressure transients should
be designed to exhaust the air that was admitted to the system at a slow and controlled
rate. If this air is allowed to discharge from the system at an uncontrolled rate, signif-
icant high-pressure transients can occur in the system at the air valve when the air is
exhausted and the water column rejoins. Several cases of system failure due to inade-
quate air valve design have been documented.
Limited closure control can be obtained by including a check valve with holes drilled
in the check disc at the base of the combination air valve. This check valve device
normally stays open due to the spring and will only close during an outflow condition
after all air has been discharged from the system. When the check valve closes, water
slowly passes through the check disc holes and slowly closes the main air valve. This
type of device is typically used where rapid air expulsion could cause damage to the
air valve float.
Figure 13.25
Air release valve
Courtesy of Bermad Control Valves
Additional air outflow control for low-pressure transient conditions can be obtained
by utilizing a combination air valve that allows air to enter the system through a large
orifice that closes during air outflow, forcing the air out of a small orifice air valve.
A hydraulic air valve is a speciality air valve that is available to control system tran-
sients. This type of air valve quickly opens in a vacuum condition to allow air into the
system through a large orifice and thus control the low-pressure transient. When the
system starts to repressurize, the air that entered the system is quickly exhausted
through the same large orifice, followed by the system liquid. After the liquid begins
to discharge, the valve slowly closes at a set rate to avoid causing a high-pressure tran-
sient in the system. A disadvantage of this type of device is that an energy dissipation
and collection structure usually has to be constructed to drain the discharged water
away from the air valve station.
References 621
REFERENCES
Allievi, L. (1903). Teoria Generale del Moto Perturbato Dell’acqua Nei Tubi in Pressione. Annali Della
Societa Degli Ingegneri ed Architetti Italiani, Milan.
Allievi, L. (1925). Theory of Waterhammer (translated by E. E. Halmos). Riccardo Garoni, Rome.
Almeida, A. B., and Koelle, E. (1992). Fluid Transients in Pipe Networks. Elsevier Applied Science,
Southampton, UK.
Amein, M., and Chu, H. L. (1975). “Implicit Numerical Modeling of Unsteady Flows.” Journal of Hydrau-
lics Division, ASCE, 101, 717.
ASCE (1975). Pressure Pipeline Design for Water and Wastewater. ASCE, New York, New York.
Basha, H. A., and Kassab, B. G. (1996). “A Perturbation Solution to the Transient Pipe Flow Problem.”
Journal of Hydraulic Research, 34, 633.
Chaudhry, M. H. (1987). Applied Hydraulic Transients. Van Nostrand Reinhold, New York.
Covas, D., Stoianov, I., Graham, N., Maksimovic, C., Ramos, H., and Butler, D. (2002). “Hydraulic Tran-
sients in Polyethylene Pipe.” Proceedings of the ASCE Environmental Water Resources Institute Confer-
ence, American Society of Civil Engineers, Roanoke, Virginia.
Elansary, A. S., Silva, W., and Chaudhry, M. H. (1994). “Numerical and Experimental Investigation of
Transient Pipe Flow.” Journal of Hydraulic Research, 32, 689.
Filion, Y., and Karney, B. W. (2002). “A Numerical Exploration of Transient Decay Mechanisms in Water
Distribution Systems.” Proceedings of the ASCE Environmental Water Resources Institute Conference,
American Society of Civil Engineers, Roanoke, Virginia.
Gray, C. A. M. (1953). “The Analysis of Dissipation of Energy in Waterhammer.” Proceedings of the Amer-
ican Society of Civil Engineers, 119, 1176.
Joukowski, N. (1904). Waterhammer (translated by Miss. O. Simmin). Proceedings of the American Water
Works Association, 24, 341-424.
Parmakian, J. (1963). Waterhammer Analysis. Dover Publications, Inc., New York, New York.
Streeter, V. L., Lai, C. (1962). “Waterhammer Analysis Including Fluid Friction.” Journal of Hydraulics
Division, ASCE, 88, 79.
Wood, D. J., Dorsch, R. G., and Lightner, C. (1966). “Wave-Plan Analysis of Unsteady Flow in Closed
Conduits.” Journal of Hydraulics Division, ASCE, 92, 83.
Wylie, E. B., and Streeter, V. L. (1993). Fluid Transients in Systems. Prentice-Hall, Englewood Cliffs, New
Jersey.
13.1 A 12-inch ductile iron transmission main delivers water from an elevated tank with a water surface
elevation of 500 ft to a ground storage tank with a water surface elevation of 450 ft. A control valve
located at the inlet to the ground storage tank is throttled to maintain a flow rate of 1,000 gpm. The
transmission main has a wave celerity of 4,000 ft/s.
a) How long will it take for a pressure wave to travel from one end of the system to the other?
b) What is the characteristic time for the system?
c) If the control valve is completely closed in 4 seconds, what will the potential head change be
immediately upstream of the control valve?
13.2 A horizontal pump delivers 10 million gallons/day from a water treatment plant clearwell with a
water surface elevation of 50 feet to an elevated storage tank with a water surface elevation of 375
feet. The water is conveyed through 20,000 feet of 30-in. steel transmission main. The transmission
main has a wave celerity of 3,500 fps.
a) What is the characteristic time for the system?
b) If an instantaneous emergency pump shutdown occurs due to a power loss, is water column sepa-
ration likely to occur at location A?
c) Would a pump bypass line be effective in preventing a vacuum condition at location A following
an emergency pump shutdown? If not, what could be done to prevent it?
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


A P P E N D I X
D
Model Optimization Techniques
Optimization, as it applies to water distribution system modeling, is the process of

finding the best, or optimal, solution to a water distribution system problem. Exam-
ples of possible problems are the design of new piping or determination of the most
efficient pumping schedule.
Use of the model for design applications follows the process shown in Figure D.1
(formulate alternatives, test alternatives, cost analysis, and make decision). Before
this process can begin, however, the engineer must develop a descriptive network
model, or simulation, that predicts the behavior of the system under various condi-
tions. The calibration required as part of model development can be aided by optimi-
zation techniques that enable automated adjustments in input parameters that result in
closer agreement between computed and observed values.
Figure D.1
Overview of model
application
Existing Additional Model
Problem Ready for
Calibrated Calibration
Statement Problem
Model Data
Formulate
Alternatives
Make Cost Test

Decision Analysis Alternatives
644 Model Optimization Techniques Appendix D
When the simulation is ready for the problem being analyzed, the engineer formulates
alternative solutions by altering inputs to the decision-making model, tests and costs
those alternatives, and presents the results to the decision-maker. The alternatives pre-
sented to the decision-maker will ideally provide a similar level of technical and eco-
nomic compliance (for example, pressure compliance, reliability, and cost) with the
design brief, but will differ in the design parameters (such as pipe diameters, pump
characteristics, valve settings, and reservoir sizes).
The idea of rerunning a simulation model under all possible conditions to determine
the most favorable alternatives is simple. However, the number of alternatives quickly
becomes enormous, and the associated time and cost can become prohibitive. Optimi-
zation techniques provide a way to efficiently examine a broad range of alternative
solutions by automatically altering the details of the system to generate new,
improved solutions. A decision-making model that uses these automated techniques is
called an optimization model. The first overview of how optimization can be applied
to water distribution design was prepared by deNeufville, Schaake, and Stafford
(1971), and the principles they presented still apply.
D.1 OVERVIEW OF OPTIMIZATION

The typical optimization problem consists of finding the maximum or minimum value
of an objective function, subject to constraints. This section provides a general over-
view of the optimization process, including key terminology and principles.
Optimization Terminology
Objective Function. When optimizing a system or process, it is important to
quantify how good a particular solution is. A mathematical function called an objec-
tive function is used to measure system performance and indicate the degree to which
the objectives are achieved. If multiple objectives exist, then there will be multiple
objective functions.
The term optimization refers to mathematical techniques used to automatically adjust
the details of the system in such a way as to achieve, for instance, the best possible
system performance (that is, the best value of the objective functions), or the least-
cost design that achieves a specified performance level. The best or most advanta-
geous solution (or solutions in multiobjective analysis) is called the optimal solution.
Decision Variables. In order to improve the performance of a system, the
parameters that can be changed must be known. These quantifiable parameters are
called decision variables, and their respective values are to be determined. For exam-
ple, in pipe-size optimization, the decision variables are the diameter for each of the
pipes being considered. Any restrictions on the values that can be assigned to decision
variables should be clearly stated in the optimization model. In the case of pipe sizes,
each discrete pipe size available should be defined.
Constraints. When judging systems and solutions, it is necessary to consider the
limits or restrictions within which the system must operate. These limits are called
Section D.1 Overview of Optimization 645
constraints. If one’s objective is to attain a minimum-cost solution, for example, one

must also consider the constraints on system performance and reliability. Constraints
serve to define the decision space from which the objective function can take its val-
ues. The decision space is the set of all possible decision variables, and the solution
space is the set of all possible solutions to the problem.
Constraints may be further classified as hard constraints, which may not be exceeded
without failure or severe damage to the system, and soft constraints, which may be
exceeded to a certain extent, although it is generally not desirable to do so. An exam-
ple of a hard constraint is the maximum pressure that a pipe can withstand without
jeopardizing the structural integrity of the system. A minimum pressure requirement
for all water system nodes and a maximum permissible velocity for system pipes are
possible soft constraints. Constraints may be applied explicitly to decision variables
(for example, pipe sizes are discrete) or implicitly to other system parameters (for
example, net head loss around a loop must be zero).
In evaluating systems and possible solutions, it is also important to understand how
the various constraints interact. The limits for a given constraint often prevent one
from obtaining a better value in another. For instance, a larger pipe size may aid in
meeting required fire flows, but at the same time may be detrimental to water quality.
The Optimization Process

Figure D.2 shows the basic steps in creating (formulating) an optimization model.
Figure D.2
Optimization model
formulation process
Based on Figure D.2, we can say that optimization involves

1. The selection of a set of decision variables to describe the decision alternatives.
2. The selection of an objective or several objectives, expressed in terms of the deci-
sion variables, that one seeks to optimize (that is, minimize or maximize).
3. The determination of a set of constraints (both hard and soft), expressed in terms
of the decision variables, which must be satisfied by any acceptable (feasible)
solution.
4. The determination of a set of values for the decision variables so as to minimize

(or maximize) the objective function, while satisfying all constraints.
In a more formal mathematical fashion, an optimization problem is said to be given in

the standard form if the above elements of the problem are presented as:
Objective function: Max f(x) (D.1)
Subject to constraints: g(x) d 0
h(x) = 0
xX
where f = objective function

x = vector of decision variables
f, g, h = functions of x
X = set of all possible solutions
Problem Visualization
If we use an example in two dimensions (that is, with two decision variables), it is
possible to visualize an optimization problem in terms of a landscape where f is the
elevation of the ground surface at a point. It is then possible to plot elevation contours
at equal increments of f. Such a contour plot is shown in Figure D.3.
Figure D.3
A contour plot in two
dimensions
Section D.1 Overview of Optimization 647
The goal of optimization in this case is to find the values of the two decision vari-
ables—the horizontal coordinates x1 and x2—for which the objective function f is
maximized (that is, the coordinates of the top of the hill).
This problem can also be visualized by considering a hiker’s search for the top of a
thickly wooded hill emerging from a flat plain. The problems begin when the density
of the forest prevents the hiker from seeing the summit, or even the general shape of
the hill. The hiker could eventually reach the top simply by continuing to gain eleva-
tion as he or she walks. However, there are easier (faster) and/or more perilous routes
to the top. If there are multiple peaks, the problem becomes even more difficult for the
hiker. This landscape representation of the decision space is discussed in more detail
later in this appendix.
Why Use Optimization?

All modelers like to believe that they produce well-calibrated models or good, safe
designs that represent value for money. However, it is rare for a modeler to have the
time or resources to consider more than a handful of solutions to a problem. In the
planning phase of a project, there are often many alternatives for each component of a
scheme. The number of possible designs for a complete scheme can be very large, if
not infinite. Even in the final stages of design, an enormous number of possibilities
usually exists—far too many to be considered and evaluated individually.
To illustrate the size of the optimization problem, consider a simple network design
example in which only 10 pipes must be sized. If we assume that there are 10 discrete
diameters to choose from for each pipe, then theoretically, the total number of possi-
ble design alternatives is 1010 (or 10,000,000,000).
When dealing with complex problems in practice, the experienced design engineer
will of course adopt rules of thumb and use personal experience to focus on possible
alternatives that are reasonably cost-effective, thereby dramatically reducing the deci-
sion space size. With the aid of a hydraulic network model, the modeler adopts a trial-
and-error approach to produce a few feasible solutions, which can then be priced.
Figure D.4 shows how these solutions often satisfy one criterion (in this case, accept-
able water quality risk) but fail to reach the optimum point at which the standard is
met and capital expenditure is minimized. With larger or more complex problems,
finding even one feasible solution may take a great deal of effort, and it is clearly
impossible to explore a wide range of alternatives using this manual, trial-and-error
technique.
It is not only the size of the problem that limits the effectiveness of the manual design
method. The behavior of the network simulation model is nonlinear, whereby changes
made at one component of the system may influence the performance at another,
resulting in a system where it is hard to intuitively relate cause and effect. Similarly,
most analysts are taught to make one system change at a time, which makes it difficult
for them to identify novel solutions that combine interrelated actions. Although engi-
neering solutions developed using the manual technique can be successful in meeting
technical design criteria, they are generally less successful at delivering these benefits
for the least cost.
Figure D.4
Manual solutions
versus optimal trade-
off point
Optimization is successful to the extent that it can account for all important factors in
a problem. Optimization approaches often have difficulty accounting for subjective
considerations such as uncertainty in demand projections, the need for reliability, dif-
ferent uncertainties in costs for different alternatives, the value of excess capacity,
uncertain budgetary constraints, and subjective preferences for pipe routes. In theory,
optimization can account for these considerations, but it is cumbersome to do so in
practical problems. Therefore, caution must be exercised in the use of optimization
techniques, but equally it must be recognized that computer-aided algorithms enable
problems to be optimized to a degree that cannot be achieved through a trial-and-error
exercise.
D.2 HOW TO USE OPTIMIZATION

A very simplified view of the decision-making process is that it involves two types of
actors: analysts (modelers) and decision-makers. Analysts are technically capable
people who provide information about a problem to the decision-makers responsible
for choosing which course of action to take. Modeling and optimization techniques
are tools that analysts may use to develop useful information for the decision-makers.
The main reason to rely on any model in a decision-making process is to provide a
quantitative assessment of the effects of management decisions on the system being
considered. A model also provides an objective assessment as opposed to subjective
opinions of system behavior. Thus, models should be used in support of decision-
making.
Optimization is simply another type of modeling, and the logic that applies to the
application of other computer models applies to optimization models as well. Optimi-
zation tools therefore should be used for supporting decisions rather than for making
them — they should not substitute for the decision-making process.
Section D.2 How to Use Optimization 649
Single-Objective Optimization
Many real-world engineering design or decision-making problems need to achieve
several objectives: minimize risks, maximize reliability, minimize deviations from
desired (target) levels, minimize cost (both capital and operational), and so on. How-
ever, the mathematical representation of the optimization problem given by Equation
D.1 considers just one objective — that is, f(x). The goal of such single-objective opti-
mization problems is to find the “best” solution, which corresponds to the minimum
or maximum value of an objective function. In practice, problems with multiple
objectives are often reformulated as single-objective problems by either lumping dif-
ferent objectives into one (that is, by forming a weighted combination of the different
objectives), or by replacing all but one of the objectives with constraints.
When multiple objectives are to be evaluated using a single-objective model, all
design objectives must be measurable in terms of a single objective function. Some a
priori ordering of different objectives (that is, a scheme to weight the relative impor-
tance of the objectives) is therefore necessary to integrate them into a single fitness
function. Because this weighting of priorities (for instance, determination of the cost
equivalent of failure risk) typically falls on the analyst, it is really the analyst who car-
ries the burden of decision-making. Even if the decision-makers are technically capa-
ble and willing to provide some a priori preference information, the decision-making
role is taken away from them.
Consider the following example with two objectives. In a pipe-sizing problem, the
two objectives are identified as: (1) minimize costs (min f1) and (2) maximize benefits
(max f2). First, both objectives need to be either minimized or maximized. This is eas-
ily done by multiplying one of them by -1. For example, max f2 is equivalent to min
(-f2) = min f2'. Next, the two objectives must be lumped together to create a single
objective function. This lumping is possible if, for example, both objectives can be
expressed in monetary terms. If a dollar figure can be attached to both of them, then
the multiobjective problem can be reduced to a single-objective one. Weighting (con-
version) factors w1 and w2 are used in order to obtain a single, combined objective
function [max f = max (w1 f1 + w2 f2')]. This combining is possible because the objec-
tive function now has a single dimension (dollars, in this case). Note that another set
of weights can be used for the two objectives if some predetermined preference
between the objectives is expressed. Finally, given this single objective function and a
set of constraints, one can find a single optimal solution.
This type of optimization is useful as a tool for providing decision-makers with
insights into the nature of the problem, but it cannot easily provide a set of alternative
solutions that trade different objectives against each other. The next subsection dis-
cusses approaches for evaluating multiple objectives.
Multiobjective Optimization
The principle of multiobjective optimization is different from that of single-objective
optimization. The main difference is that the interaction among different objectives
gives rise to a set of compromised solutions, largely known as tradeoff, nondomi-
nated, noninferior, or Pareto-optimal solutions.
For example, if the two objectives are given in the objective space as depicted in Fig-
ure D.5, the alternative C is dominated by D because this alternative (D) gives more of
(that is, maximizes) both objectives f1 and f2. Alternatives A and B belong to the
Pareto set because each solution on the Pareto-optimal curve is not dominated by any
other solution. In going from one solution to another, it is not possible to improve on
one objective without making the other objective worse.
The consideration of many objectives in the design or planning stages provides three
major improvements to the decision-making process (Cohon, 1978):
• A wider range of alternatives is usually identified when a multiobjective

methodology is employed.
• Consideration of multiple objectives promotes more appropriate roles for the

participants in the planning and decision-making processes. The analyst or
modeler generates alternative solutions, and the decision-maker uses these
alternative solutions to make informed decisions.
• Models of a problem will be more realistic if many objectives are consid-

ered.
Figure D.5
Graphical
interpretation of
dominance
Applications of Optimization
Various problems from water distribution modeling practice can be formulated as
optimization problems. This section provides a brief overview of the main areas
where optimization has been applied to water distribution management problems.
Section D.2 How to Use Optimization 651
Automated Calibration. Before hydraulic network models can be used for

predictive purposes with any degree of confidence, they need to be calibrated against
field data. Optimization can automate the process of adjusting the parameters of the
network model by using a measure of the match between modeled and observed val-
ues as the objective. In other words, optimization in this case is aimed at determining
the pipe roughness, nodal demand, and link status data that will minimize the differ-
ence between modeled and observed values (see page 268 for more information).
Sampling Design for Calibration. The selection of field test locations in a
water distribution system for collection of data to be used in calibration is called sam-
pling design. Sampling design is often done by subjective judgment, which can lead
to system calibration that is based on insufficient monitoring data, or that has redun-
dant information for calibration.
Optimization provides an alternative way of selecting field test locations for use in
calibrating a water distribution system. One of the key difficulties in sampling design
via optimization, however, is choosing objective functions that faithfully represent the
monitoring objective (for example, that maximize model accuracy or minimize the
cost of data collection). See page 276 for more information on using optimization for
sampling design.
Operational Optimization. Pump operating costs make up a large proportion
of the expenses of water utilities. It is therefore important to plan the operation of
pumps to minimize energy consumption while maintaining the required standard of
service and reliability. For water distribution systems, the objective function is nor-
mally defined to minimize the operational cost of the system over a period of time
(typically 24 hours), and the decision variables are the times for which each pump is
run (see page 446).
Design/Expansion. Design of new water distribution networks or expansion of
existing ones is often viewed as a least-cost optimization problem with pipe diameters
being the decision variables. Pipe layout, connectivity, and imposed head and velocity
constraints are considered known. Obviously, other elements (such as service reser-
voirs and pumps) and other possible objectives (reliability, redundancy, and water
quality) exist that could be included in the optimization process. However, difficulties
with including reservoirs and pumps and with quantifying additional objectives for
use within the optimization process have historically kept most optimization research-
ers focused on pipe diameters and the single objective of least cost. Even so, some
attempts have been made to include these additional factors in formal optimization.
See page 360 for more information on using optimization for system design.
Rehabilitation. Improvements in a water distribution system’s performance can
be achieved through replacing, rehabilitating, duplicating, or repairing some of the
pipes or other components (pumps, tanks, and so on) in the network, and also by add-
ing completely new components. It is likely that funding will be available to modify
or add only a small number of components to a network at any one time. The multiob-
jective optimization problem is therefore formulated to choose which components to
add or improve (and how to improve them) in order to maximize the benefits result-
ing from the system changes, while minimizing the costs, possibly subject to budget
constraints.
D.3 OPTIMIZATION METHODS

Optimization search methods range from analytical optimization of one variable; to
linear, nonlinear, and dynamic programming approaches; to numerous search meth-
ods. The most sophisticated search methods mimic various natural processes in their
approaches; these techniques are called adaptive search methods. Adaptive methods
known as genetic algorithms mimic the natural selection process and have been suc-
cessfully applied to distribution network optimization.
Figure D.6 shows how the search techniques described in this section can be used
within an optimization framework that is coupled with a hydraulic simulation model.
For example, in the case of water distribution pipe sizing, an optimal solution to the
problem is obtained by interfacing a hydraulic simulation model with a search
method. The hydraulic simulation model is used to implicitly solve for the hydraulic
constraints that define the flow phenomena (for example, continuity and energy bal-
ance) each time the search method needs to evaluate these constraints. The search
process starts by generating one or more initial solutions (for example, pipe diameters
assigned to each of the pipes). Each solution is then tested by solving the hydraulic
simulation model to compute the flows and pressures in the system. Based on the cost
and performance (minimum pressure at nodes, minimum/maximum velocities, travel
times, and so on), the solution is assessed and the search method generates a new test
solution. The procedure is repeated until some convergence criterion is reached.
Figure D.6
Search-simulation
framework
Analytical Optimization
Analytical optimization techniques are often introduced in calculus courses. These
techniques usually deal with unconstrained problems for which one is trying to obtain
the optimal solution to a problem that consists of an objective function alone (that is,
without constraints imposed on the solution). Although it is unlikely that any substan-
tial water distribution system optimization could be formulated as an unconstrained
optimization problem, this type of problem is important to the development of more
advanced optimization techniques, as constrained optimization algorithms are often
Section D.3 Optimization Methods 653
extensions of unconstrained ones. Many algorithms used to optimize a function of

multiple variables attempt to do so by solving successive one-dimensional problems.
In the unconstrained optimization problem, a maximum (largest) value of a real-
valued function, f(x), where x is a real-valued variable, is sought. In other words, the
optimization seeks a value x* such that f(x*) t f(x) for all x. To solve this problem, a
few terms must be defined.
The largest and smallest values that a function assumes can be defined locally or glo-
bally. Figure D.7 shows that a local optimum can be defined at a point x* if, for a local
maximum, f(x*) is greater than or equal to f(x) for all x adjacent to x*. A global opti-
mum exists at point x* if, for a maximum, f(x*) is greater than or equal to f(x) for all x
(that is, there are no greater values anywhere in the search area). If the objective is
instead to find the minimum value, a global optimum exists at point x* if f(x*) is
smaller than f(x) for all x. Global maxima and minima are also called absolute
extrema. Finding a global optimum is usually much more difficult than finding a local
optimum.
Figure D.7
local and global
maxima and minima
Assuming f(x) to be a continuous function, it can be readily seen from Figure D.7 that
the extrema must occur at one of the following:
• The endpoint a or b [f(x) = A or f(x) = B]
df
• A point where the tangent is horizontal (that is, a root of f '(x) = ------ = 0)
dx
[f(x) = C or f(x) = D]
df
• A discontinuity of f '(x) = ------ [f(x) = F]
dx
Solving for the x values that satisfy the equality in the second bullet above (that is,
points where the tangent is horizontal) yields candidate points for local minima and
maxima inside the boundaries; however, these values alone are not sufficient to indi-
cate if the point is a minimum or maximum. It can be seen immediately that point E is
neither a minimum nor a maximum, though the first derivative at that point is equal to
zero because it is an inflection point.
If the first derivative for a point is equal to zero, and if the values of f '(x) adjacent to
that point are negative for smaller values of x and positive for greater values of x (that
is, the slope of the tangent changes from negative to positive), then this point is a local
minimum (see Figure D.8). Likewise, if f '(x) changes from positive to negative at a
point, then this point is a local maximum. For the point of inflection, the sign of the
first derivative does not change from one side of the point to the other.
Figure D.8
First derivative test for
local extrema
Local maxima and minima can also be determined using the second derivative of f(x).
The function has a local maximum at a point where the second derivative is less than
zero, and a local minimum where the second derivative is greater than zero. These
conditions can be extended to the constrained case (Finney, Weir, and Giordano,
2001).
The problem of unconstrained optimization of a function of more than one variable
(the multivariate case) is concerned with finding the correct combination of the values
of the variables to obtain the best value of the objective function. The criteria used for
selecting the combination of variables are similar to those used for single-variable
functions but require more complex mathematical techniques. The nonlinear pro-
gramming section of this appendix (see page 659) covers these techniques. For more
information on unconstrained optimization of functions of a single variable and multi-
ple variables, see introductory texts on operations research (Hillier and Lieberman,
1995; Wagner, 1975).
The problem of finding the optimum of a constrained problem (that is, a problem for
which the decision variables are subject to various constraints) can be approached in
two ways. The first approach consists essentially of adapting those methods that have
been developed for the unconstrained problem to make them suitable to use in the
constrained case. The second approach treats the constraints as the essential part of
the problem. The latter approach was first used in connection with linear functions
and linear constraints and was given the name linear programming. Later, these meth-
ods were generalized to deal with nonlinear problems. The term mathematical pro-
gramming is frequently used as a label to include all of the variations.
Analytical optimization methods do not usually work well for network problems
because of the large number of pipes involved. However, some problems can be posed
in such a way that the pipes can be considered one at a time. Camp (1939) first
applied analytical optimization to pipe sizing. Bhave (1983), Cowan (1971), Dancs
(1977), Deb (1973 and 1976), Swamee and Khanna (1974), and Watanatada (1973)
developed similar approaches. Walski (1984) showed how analytical methods can be
applied for a number of simple pipe sizing problems. As systems become more com-
plicated, these methods become cumbersome.
Shamir and Howard (1979) and Walski and Pelliccia (1982) applied such techniques
to decisions to replace pipes due to breakage, and Walski (1982 and 1985) developed
a method for determining whether to rehabilitate a pipe based on energy costs or the
cost of parallel piping using an analytical solution. Section 8.9 describes an analytical
approach to selecting pipe size and pumping equipment based on life-cycle costs.
Linear Programming
Linear programming (LP) refers to the class of optimization problems in which both
the objective function and constraints in Equation D.1 are linear functions (that is, the
variable’s exponent is 1). Linear programming can be envisioned as a mountain with
straight-sloping sides. In maximizing an objective, LP locates the highest peak of the
mountain. Although modeling and decision-making problems connected with water
distribution systems are almost always nonlinear, an understanding of the basic prin-
ciples of linear programming is essential to understanding the methods of solving
nonlinear problems that developed from it.
A simple two-variable example will be used to study the geometry of a linear problem
in two dimensions and illustrate the linear programming technique. If the equation of
the optimization problem is given in the standard form of Equation D.1, then:
Max f(x) = x1 + 5x2
Subject to: x1 + x2 d 8 (i)

x 1 – 3x 2 t 0 (ii)
x1 t 0 (iii)
x2 t 0 (iv)
Clearly, the objective function and constraints are linear because no product terms
involving the decision variables exist [that is, there are no terms like (x1 u x2) or x12 in
the objective function or constraints].
A graphical solution technique requires that lines be drawn for the objective function
and the constraints. First, consider constraint (i). If this constraint is an equality, then
it can be plotted as a straight line in (x1, x2) space. The line is constructed by simply
determining the intersection points of the two axes and connecting them. In this case,
x2 = 8 when x1 = 0 [point (0, 8)], and x1 = 8 when x2 = 0 [point (8, 0)].
Because the constraint is an inequality, the question of which side of the line contains
points that satisfy the constraint arises. This determination is made by testing any
point not on the line. For example, the coordinates (0, 0) do satisfy the inequality
because 0 + 0 d 8 . All points on the same side of the boundary line as point (0, 0), as
well as all points on the line, will satisfy the inequality. The plane on the opposite
(infeasible) side of the line is shaded to remove these values from consideration (see
Figure D.9).
Figure D.9
LP constraint (i) in the
two-dimensional
example
In constructing the constraint boundary line for constraint (ii), x1 – 3x2 t 0 the inter-
section point for both axes is found to be (0, 0).The other point required for drawing
the line can be obtained by taking a value other than 0 for x1. For example, if x1 = 3,
then x2 = 1 and the other point for drawing the constraint boundary line is (3, 1).
Again, a point not on the line should be used to test for feasible region. For example,
for point (0, 2), x1 – 3x2 = –6, which is less than 0. Therefore, point (0, 2) does not sat-
isfy the inequality, and the plane on the same (infeasible) side of the line should be
shaded (see Figure D.10).
If the feasible regions for the two constraints are drawn together, then an intersection
of two spaces is obtained that consists of the set of all points whose coordinates sat-
isfy both inequalities. Graphing all of the inequalities simultaneously yields the geo-
metric representation of the space for the set of all decision-variable values that
satisfy all of the constraints, including the non-negativity constraints [(iii) and (iv)].
This space is shown in Figure D.10 and is called the feasible region. The corner points
of the feasible region (0, 0), (8, 0), and (6, 2) are called the extreme points because
they belong to the feasible region and lie on the intersection of two distinct boundary
lines.
Figure D.10
The feasible region
for the example
problem
The feasible region in Figure D.10 is also said to be convex because it contains a set of
points for which a line segment joining any pair of points in the region will lie com-
pletely within the region. Figure D.11 shows examples of convex and nonconvex fea-
sible regions in two-dimensional space.
Figure D.11
Examples of convex
and nonconvex
regions
The final step in solving the linear programming problem is to find the point in the
feasible region that maximizes the value of the objective function. It can be shown
that optimal values of a linear function on a convex set will occur at extreme points if
they occur at all. Therefore, a finite number of optimal values exist for a convex set.
To find the extreme point that produces the optimum solution, the objective function
line is graphed just like any constraint boundary line. Taking a point that is in the fea-
sible region, say (5, 1), yields f(x) = 5 + 5(1) = 10. The x1 and x2 axis intercept points
for this function are found to be (10, 0) and (0, 2), respectively. The function f(x) =
x1 + 5x2 = 10 can now be plotted by drawing a line through these points, as shown in
Figure D.12. If f(x) is assigned a series of arbitrary values, the equation in each case
represents a straight line with the slope –1/5 (that is, the objective function lines are
parallel to each other). As the value assigned to f(x) increases, the line representing
the function shifts upward. It can be seen that the maximum value of f(x) lying within
the feasible region will correspond to the equation for the line passing through the
extreme point (6, 2). (Recall that this point is the intersection of lines x1 + x2 = 8 and
x1 – 3x2 = 0.) The maximum value of the objective function is therefore f(x) = x1 +
5x2 = 6 + 5 u 2 = 16.
Figure D.12
The objective function
lines
Note that if the parallel lines representing the different objective function values are
plotted at equal intervals, the plot would resemble the contour plot of Figure D.3. The
differences in this case are that the contours are straight lines and that areas of the
decision space exist that are excluded from the acceptable (feasible) region due to lin-
ear constraints. For example, the line with the objective function value of 33 does not
pass through any point in the feasible region; therefore, the objective function cannot
reach this value while the solution is still feasible.
The knowledge that the optimal solution of an LP problem is an extreme point moti-
vates a special iterative procedure for reaching the optimum called the simplex
method. Starting from a feasible extreme point, the simplex method changes the vari-
ables to move to an adjacent extreme point where the objective function has a larger
value. Movement therefore occurs on the edge of the feasible region along a selected
constraint line that is connected to the current extreme point. The procedure selects
the next extreme point on the basis of the largest gain in the objective function value.
It continues in this way until an extreme point is reached for which no further
improvement in the objective function is possible. For a more complete discussion of
the simplex method, see Hillier and Lieberman (1995) and Wagner (1975).
The basic principles behind the LP method are illustrated by a simple example of siz-
ing a branched pipe network (that is, a network without closed loops). This technique
applies only to branched networks for which demands are known and therefore pipe
flows can be determined from continuity alone. Because head loss is a nonlinear func-
tion, the initial problem is nonlinear; however, it can be reformulated as a linear func-
tion for which LP can be used. This reformulation is possible because frictional head
loss is a linear function of pipeline length. The optimization problem is thus formu-
lated as minimization of cost subject to the constraint that each branch in the network
is comprised of various pipe lengths, each with its own fixed diameter. The decision
variable of the problem is the length of a particular segment of a particular diameter.
For example, the program solves for the length of a branch of 8-in. pipe necessary to
solve the problem, then the length of 10-in. pipe, 12-in. pipe, and so on. An additional
constraint ensures that the total sum of the segments of all the possible diameter com-
binations is equal to the length of the particular branch.
One of the greatest advantages of LP over nonlinear programming (NLP) algorithms

is that if the optimum exists, LP is guaranteed to find it. Once formulated, an LP is
easy to handle, especially because well-established, general-purpose, “off-the-shelf”
LP software is available. Thanks to the advances in computing of the past decade,
problems having tens or hundreds of thousands of continuous variables are regularly
solved. LP makes it possible to analyze the final solution and find its sensitivity to
changes in parameters. However, the assumption of linearity is often not appropriate
for real engineering systems, and it is certainly not appropriate for water distribution
system optimization.
Linear programming techniques work well for pipe sizing problems involving
branched systems with one-directional flow as demonstrated by Karmeli, Gadish, and
Meyer (1968); Salcedo and Weiss (1972); and Austin and Robinson (1976). For
looped systems, another method must be coupled with the LP solution to determine
the optimal flow distribution as demonstrated by Alperovits and Shamir (1977);
Bhave (1980); Quindry, Brill, and Liebman (1981); Kettler and Goulter (1983); and
Morgan and Goulter (1985). Jowitt, Garrett, Cook, and Germanopoulos (1988) also
applied LP to solving problems with optimal system operation. Boccelli et al. (1998)
and Constans, Brémond, and Morel (2000) used LP to optimize disinfection in water
systems.
Nonlinear Programming
Nonlinear programming (NLP) problems have the same structure as the general opti-
mization problem given in Equation D.1. However, nonlinear programming refers to
the class of optimization problems for which some or all of the problem functions,
f(x), g(x), and h(x), are nonlinear with respect to the variables. NLP models more real-
istically capture certain characteristics of system relations but introduce significant
computational difficulties. This type of problem is particularly difficult to solve
because
• The feasible region is not guaranteed to be convex due to a combination of

nonlinear constraints;
• An optimum solution does not necessarily occur at an extreme (stationary)

point of the feasible region;
• Multiple local optima may exist, making it difficult to identify the global
optimum.
Nonlinear programming problems come in many different forms and can be very hard
to solve. It is therefore unreasonable to expect a generic optimization methodology to
be able to solve all nonlinear programming problems. For example, if a problem
includes nonlinear constraints, and if an initial feasible point is not provided, there is
no guaranteed way to find a feasible point, or even to determine whether one exists. If
all the constraints are linear, however, determining a feasible point is no more difficult
than in linear programming. A large body of literature concerned with the various
NLP techniques exists, and it is beyond the scope of this text to provide a comprehen-
sive survey of these methods. However, this section will examine the basic extensions
to the calculus-based optimization for more than one variable in order to illustrate the
increased complexity of NLP problems.
Consider the optimization problem in two dimensions for which x1 and x2 are decision
variables and the objective function contour lines are as shown in Figure D.13. As in
the one-dimensional problem previously presented, the optimum value of f occurs
inside the permissible range only at a point where the tangent plane (plane in this case
because there are two variables) is horizontal, or at a point of discontinuity in the
gradient.
Figure D.13
Two-dimensional
maximization
problem
For a point where the tangent plane is horizontal, the first-order partial derivatives will
equal zero:
wf x = ------------
wf x = 0
------------
wx 1 wx 2
However, similar to one-dimensional problems, this information is necessary but not

sufficient to indicate whether the point is a maximum, minimum, point of inflection,
or saddle point [a point that is a maximum for the function in one direction (say, x1)
and a minimum in the other (x2) direction]. In order to determine which of the station-
ary points are maxima (or minima), it is necessary to examine the second-order partial
derivatives of f. These are contained in what is called the Hessian matrix, which in
this case is:
2 2
w f-
------- w f -
----------------
2 wx 1 wx 2
wx 1
H =
2 2
w f w f-
----------------
- -------
wx 2 wx 1 2
wx 2
In order for a stationary point to be a maximum, the following conditions must be

satisfied:
2
w f-
------- 0
2
wx 1
2
w f-
------- 0
2
wx 2
2 2
w f-
------- w f -
----------------
2 wx 1 wx 2
wx 1
det H = !0
2 2
w f - w f-
---------------- -------
wx 2 wx 1 2
wx 2
In other words, all of the diagonal elements of the Hessian matrix at the point in ques-
tion must be negative, and the determinant must be positive, which also means that the
Hessian matrix must be negative definite.
Conversely, for a point to be a minimum, the following conditions must be satisfied:

2
w f-
------- !0
2
wx 1
2
w f-
------- !0
2
wx 2
2 2
w f- w f -
------- ----------------
2 wx 1 wx 2
wx 1
det H = !0
2 2
w f -
---------------- w f-
-------
wx 2 wx 1 2
wx 2
These conditions translate into the requirement that the Hessian matrix at the point in
question be positive definite.
The development of tests for positive or negative definiteness becomes progressively
more difficult as the number of variables increases. Thus, such tests are rarely applied
in algorithms that do not already make use of this matrix.
The above approach for identifying potential extremum points and testing them for
optimality is built into several search techniques that assume that
• It is possible to calculate either or both of the first and second derivatives of
the function to be optimized;
• The starting decision variable values are near the desired extremum of the
objective function;
• The function is reasonably smooth.
An increasingly large number of methods known as steepest-ascent or gradient-
ascent methods (in the case of maximization) take search steps in the direction of the
most rapid increase in the objective function. However, these searches slow as they
near the optimum solution due to the decrease in gradient. Conversely, the well-
known Newton (or Newton-Raphson) method is quite good when the trial solution is
close to the optimum but can be quite unreliable for solutions far from the optimum.
One solution to this problem is offered by the Levenberg-Marquardt method, which
switches between the gradient method and the Newton method depending on how far
the current solution is from the optimum. However, this method still does not solve
the problem of decision spaces with multiple local optima, because the Levenberg-
Marquardt method (and any other hill-climbing method as they are mostly known)
will find only the local minimum closest to the starting condition.
Nonlinear problems become even more difficult to solve if one or more constraints are
involved. It is beyond the scope of this text to give details on different methods of
dealing with multiple optima and nonlinear constraints in NLP. See Hillier and Lie-
berman (1995) and Wagner (1975), where different nonlinear programming tech-
niques (such as the methods of substitution and the Lagrange multipliers for nonlinear
equality constraints, or specific cases on NLP such as quadratic and separable pro-
gramming) are presented in sufficient detail. Some NLP problems are further compli-
cated by the existence of decision variables that can only take on integer values.
Discrete pipe sizes are an example of this. Optimization problems that combine con-
tinuous and integer values are referred to as mixed-integer problems and require a spe-
cial set of techniques such as the branch-and-bound method (Hillier and Lieberman,
1995).
Because of the nonlinear nature of head loss equations and cost functions, various
nonlinear programming methods have been used for pipe sizing optimization. Jacoby
(1968); Lam (1973); Loganathan, Greene, and Ahn (1995); and Ormsbee and Con-
tractor (1981) proposed NLP solutions, but NLP has not been widely used because of
the dimensionality of the problems being solved. NLP has shown more promise with
optimal operation problems (Coulbeck and Sterling, 1978; Ormsbee and Chase, 1988;
and Ormsbee and Lingireddy, 1995) and pressure-sensitive leaks (Stathis and Loga-
nathan, 1999).
A very popular approach for nonlinear programming is the use of generalized reduced
gradients, which was made popular by Lasdon and Waren (1982). Generalized
reduced gradients have been used for design (Shamir, 1974; Lansey and Mays, 1989;
Duan, Mays, and Lansey, 1990; Cullinane, Lansey, and Mays, 1992), operation
(Sakarya and Mays, 2000), and calibration (Lansey, 1988; Brion and Mays, 1991).
Dynamic Programming
Dynamic Programming (DP) is a procedure for optimizing a multistage decision pro-
cess in which a decision is required at each stage. The technique is based on the sim-
ple principle of optimality of Bellman (1957), which states that an optimal policy
must have the property that regardless of the decisions leading to a particular state, the
remaining decisions must constitute an optimal sequence for leaving that state. For
instance, if point B is on the shortest path from A to C, then the shortest path from B
to C lies along that route. If there were a shorter path from B to C, then it can be used
in combination with the shortest path from A to B to give the shortest path from A to
C (see Figure D.14).
Figure D.14
Illustration of the
principle of optimality
This technique is appropriate for problems that have the following characteristics:
• The problem can be divided into stages with a decision required at each
stage;
• Each stage has a number of system states associated with it;

• The decision at one stage transforms that state into a state in the next stage
through a state transformation function;
• Given the current state, the optimal decision for each of the remaining states
does not depend on the previous states or decisions.
Consider the following example, which illustrates the basic principles behind the DP
method. Assume that the operation of a water distribution system needs to be opti-
mized over a 24-hour period under known demand conditions. If a day is divided into
24 periods, then the decision of how much water to pump into the central reservoir to
meet demands and minimize the cost of pumping is made at 24 stages. The reservoir
volume will go through different states during the 24 hours (for example, the reservoir
could be full initially, then empty, and then refill). Therefore, the system state or the
state variable is defined as the states of the reservoir (that is, the reservoir volume).
The main problem with this technique arises from the requirement that the state vari-
able be discretized. This requirement indicates that DP is highly dependent on the
level of discretization of the state variable (in this case, the reservoir storage). The
decision in the example is the volume pumped into the reservoir, and the state trans-
formation function is the continuity equation relating the storage in one time period to
the storage in the previous time period. Figure D.15 is a simple diagrammatic repre-
sentation of one stage in a DP procedure.
Figure D.15
One-stage DP
representation
The 24 stages of the example can be linked together as a succession of similar to the
one in Figure D.15. At each stage, the calculations involve three components:
• The cost at the beginning of the state (that is, the cost incurred up to the stage
in question)
• The cost of the decision in this stage (in this case, the cost of pumping)
• The minimum cost of the remaining stages (in this case, the minimum cost
of pumping water in the remaining time periods)
The transformation from the beginning state to the ending state is based on the state
transformation function (the continuity equation in this example). Each decision in
DP — in this case, the volume pumped into the reservoir — has an immediate conse-
quence cost associated with pumping in the current time period, as well as a long-
term consequence cost arising from the volume of water in the reservoir as a result of
that decision. The long-term cost should be minimized for the remaining period.
To solve a DP problem, it is necessary to evaluate both immediate and long-term con-
sequence costs for each possible state at each stage. This evaluation is done through
the development of the following recursive equation (for minimization):
* *
f i S i = min > c i S i d i + f i – 1 S i – 1 @ i = 1 ...,N (D.2)
di
where ci(Si,di) = cost of making decision di, which changes state from
state Si – 1 to state Si, with i – 1 states to go
In terms of the reservoir example, ci(Si, di) is the cost of pumping from one discrete
reservoir volume level to another in the next time step; fi*(Si) is the cost of the opti-
mum policy to the end of the final stage when starting in state Si with i stages to go;
and N is the total number of stages in the problem (24 stages in the example).
The preceding equation requires computation of the costs for each feasible state at
each stage. Because the state variable needs to be discretized, say, into n discrete lev-
els, evaluation of n u n = n2 computations is necessary at each stage to test all combi-
nations for moving from one storage state to another. If the reservoir’s volume is
discretized into 10 discrete levels, the total number of computations is 100. However,
if more than one reservoir exists in the system, the number of computations increases
dramatically. For a system with three reservoirs, the total number of states at each
stage increases to 1,000, and the total number of computations is 1,0003 =
1,000,000,000.
Thus, in a real system the problem becomes extremely large, providing another exam-
ple of the “curse of dimensionality.” Most convenient for problems with a small num-
ber of state variables, dynamic programming becomes increasingly unmanageable
with greater dimensionality of state variables. Despite this major disadvantage, DP is
probably the most widely used optimization technique in water resources problems
(Yeh, 1985). The popularity and success of this technique can be attributed to the fol-
lowing:
• It guarantees determination of the global optimum to within the accuracy
defined by discretization of the state space;
• The nonlinear and stochastic features that characterize a large number of

water resources problems can be easily translated into a DP formulation.
Because dynamic programming works for a series of stages, the only types of design
problems it is applicable to are single pipelines with multiple withdrawals as
described by Liang (1971). Dynamic programming is useful however for optimizing
temporal processes suh as those typical in system operation problems. Sterling and
Coulbeck (1975); Coulbeck (1984); Sabel and Helwig (1985); Ormsbee, Walski,
Chase, and Sharp (1989); and Lansey and Awumah (1994) applied dynamic program-
ming to determine optimal pumping operation for minimization of costs in a water
system. Dynamic programming works well as long as the number of decision vari-
ables is very small.
Nonadaptive Search Methods

Although the steepest-ascent methods are good for finding local optima, in real prob-
lems it quickly becomes inconvenient — and often impractical or impossible — to
calculate analytically the partial derivatives with respect to the decision variables. An
obvious solution would be to evaluate the function at a number of points around the
current location and use a numerical difference technique to find the partial deriva-
tives, but the number of function evaluations required at each step rapidly becomes
prohibitive in a high-dimensional case. In such a situation, knowledge of the func-
tional relationship between the objective function value and the decision variables
either does not exist or is not usable.
The problem of a hiker’s search for the top of a thickly wooded hill (or hills) pre-
sented earlier actually introduces optimization problems for such a case. The func-
tional relationship between the objective function value (that is, the height of a point
on the hill) and the adjustable (decision) variables (which describe the direction of the
walk) is not known because of the woods obstructing the view. The hiker example
considers a two-dimensional problem; a problem with a multi-dimensional search
space would be much more complicated.
Several search techniques for finding the optimal solution (in this case, the top of the
hill) are presented in this section. The feature common to all of these methods is a
generate-and-test strategy in which a new point is generated and its function value
tested. Depending on the particular method, a new point (or set of points) is gener-
ated, and the search for the top of the hill continues.
The overall aim of nonadaptive search methods in multidimensional search problems
is to find, after only a few trials, a set of decision variable values that yield a value of
the objective function that is close to the best value attainable. Such a search has two
purposes: to attain a good value of the objective function, and to give information use-
ful for locating future test points where desirable values of the objective are likely to
be found.
Random Search. Simple or random search is the least sophisticated method of
optimization that can be used to locate the top of the hill. Because the random search
method makes no use of a priori data about the problem, the results of the method
tend to be influenced by the laws of chance and probability. The method is also called
random walk because it assumes that the hiker is drunk and that each of his or her
steps has an equal probability of going in any direction. In a variation of this method
called the random multistart method, a number of drunk hikers are parachuted into
the area at random places with the hope that at least one of them will find the peak.
For even modestly large problems, the chance of finding the optimum by luck quickly
diminishes with the increase in the problem size and dimensions.
Hill-Climbing Strategies. A hill-climbing strategy is a manner of searching for
a maximum that corresponds to the way the hiker might try to climb up the hill, even
though the only information he or she has about the surface comes from past moves.
The hill-climbing methods try to solve the problem of how to reach the local maxi-
mum in as few trials as possible. Thus, the purpose of each trial solution is not only to
attain a good value of the function, but also to give information useful for finding
future test locations likely to correspond to desirable values of the function. Methods
that use heuristic schemes to systematically search the solution space from a given
starting point are called direct search methods. These methods make changes to vari-
ables, and the effect is tested by evaluating the objective function. These attempts are
called steps and are used in various ways to search the solution space. Steps in
“wrong” directions are inevitable. The attraction of direct search methods lies not in
theoretical proofs of convergence, but in their simplicity and the fact that they have
proved themselves in practice.
Fibonacci Coordinate Search. The Fibonacci method is named after a thir-
teenth-century mathematician who introduced a series of numbers on which the coor-
dinate search method is based. This search technique is considered here primarily
because it has been shown to be the best of all sequential interval division procedures.
With this type of procedure, sequential divisions of the original search space eliminate
intervals in which the optimum cannot occur (Schwefel, 1981). In a coordinate strat-
egy, a line search is performed sequentially on each parameter. The line search entails
three steps:
1. Determine the search direction.
2. Delimit the search interval.
3. Search for the optimal parameter value within the interval.
Because the gradient of the objective function is unknown, the search direction is
determined by taking a small step in the positive direction and, if required, also in the
negative direction. If the function value is improved, further steps are taken in the suc-
cessful direction until the optimal parameter value is overstepped (that is, a reduction
in the function value is found). This procedure is known as blocking the maximum,
and it returns an interval that must contain the optimal parameter value. Finally, this
interval is searched by repeatedly subdividing the interval into four, and then elimi-
nating one of the two outer intervals that cannot contain the optimal parameter value.
Van Zyl, Savic, and Walters (2001) and Van Zyl (2001) used the Fibonacci search
method for performing operational optimization of water distribution systems.
Hooke and Jeeves Pattern Search. The Hooke and Jeeves method is based
on two types of moves. At each iteration, there is an exploratory (pattern) move that
resembles a simplified coordinate search with one discrete step per coordinate direc-
tion. On the assumption that the line joining the first and last points of the exploratory
move represents an especially favorable direction, a move is made by extrapolating in
the direction of this line. Another exploratory move (local exploration) is made from
the extrapolated point, and then the new function value is compared to the function
value before the pattern step. The length of the pattern step is hereby increased at each
successive pattern move while the pattern search direction changes only gradually.
This approach is most advantageous where there are narrow valleys in the solution
space (Schwefel, 1981). When no further improvements are made through exploration
around the base point, the initial step size can be reduced and the process repeated if a
higher accuracy is required.
Downhill Simplex Search. The downhill simplex method relies on a simplex,

which is a geometric element having the minimum number of boundaries for the
number of dimensions in which it exists. The number of vertices that the figure has
will be one greater than its number of dimensions (that is, in N dimensions, N + 1 ver-
tices exist). Therefore, in one dimension, a simplex is a line; in two dimensions, it is a
triangle; in three dimensions, it is a tetrahedron; and so on.
Nelder and Mead (1965) developed a method that requires only function evaluations,
not derivatives. It starts by choosing an initial simplex (by picking a random location,
for example), taking unit vectors for the edges, and evaluating the function being
searched at the N + 1 vertices (points) of the simplex. The downhill simplex method
(used for minimization) now takes a series of steps, most of which just move the point
of the simplex where the function is largest (worst point) through the opposite face of
the simplex to a lower point (for minimization). These steps are called reflections. If
the current point can be extended even further, then larger steps are taken (Figure
D.16). If growing in this way yields a better point than just reflecting, then the move
is kept; otherwise, the original point obtained through reflection is selected. If, after
reflecting, the new point is still the worst, then the search has probably overshot the
minimum and, instead of reflecting and growing, reflecting and shrinking is intro-
duced. If better results are still not obtained, shrinking is tried on its own. When the
method reaches a minimum, it will shrink down around it and stop. An advantage of
this algorithm is that, other than stopping criteria, no adjustable algorithm parameters
exist.
Figure D.16
Downhill simplex
moves
Exploration and Exploitation

Throughout a search, one must continually decide whether to climb (by exploiting
information gathered from previously visited points in the search space to determine
which places might be desirable to visit next) or to explore (by visiting entirely new
regions of a search space to see if anything promising may be found there). This is
referred to as the battle between exploration and exploitation.
An example of exploitation is hill climbing, which investigates adjacent points in the
search space and moves in the direction giving the greatest increase in the objective
function value. Exploitation techniques are good at finding local optima. Unlike
exploitation, exploration involves leaps into the unknown. Problems that have many
local optima can sometimes only be solved by this sort of random search. In order to
find the global optimum among numerous local optima, it is necessary to achieve a
balanced search strategy. If all of the tests are expended on exploration, then one may
learn approximately where the top of the hill is but not locate it precisely. However, if
only previously generated points are used to select new points for testing, only the
local optimum can be found and the global optimum may not be reached.
A master search plan that properly combines exploration with exploitation will
change the character of the strategy as the search progresses. At the beginning of the
search, nothing at all is known about the function, and the search must explore in
some small, randomly chosen region so that the next move is in the direction for
which the function is higher. In the middle of a search, having left the very low
regions behind, the search should climb as fast as possible, exploring only when
strictly necessary to guide successive steps. Finally, toward the end of the search, it is
necessary to do more exploitation to attain any increase in elevation because the slope
of the function surface is often slight near the maximum.
Adaptive Search Methods

Over the course of the last two decades, computer algorithms that mimic certain prin-
ciples of nature have proven their usefulness in various domains of application.
Researchers have found especially worth copying those principles by which nature
has discovered “stable plateaus” in a “rugged landscape” of solution possibilities.
Such phenomena can be found in annealing processes, central nervous systems, and
biological evolution. These phenomena have led to the analogous new optimization
methods of simulated annealing, artificial neural networks, and evolutionary pro-
grams.
An algorithm is a fail-safe procedure guaranteed to find the optimum (for example,
the simplex algorithm for solving LP problems), but this guarantee of success often
proves to be prohibitively expensive for real-world, complex problems with time and
budget constraints. Heuristic methods, on the other hand, are procedures that produce
good solutions and “almost always” get to the right answer, which is often good
enough. Heuristic methods are important because of the lack of robust, rigorous meth-
ods that can efficiently optimize broad classes of difficult problems. For example, the
structure of many global optimization problems is unknown or too complex for ana-
lytic optimization approaches to be applied. These heuristic methods provide rules
that tend to give the global optimal answer, but are not guaranteed to do so. One of the
main characteristics of adaptive heuristic methods is their ability to move out of local
optima and sample the search space globally.
Genetic Algorithms. Over many generations, natural populations evolve
according to the principles first stated clearly by Charles Darwin. The main principles
are those of preferential survival and reproduction of the fittest members of the popu-
lation. Additional principles that characterize natural systems are the maintenance of
a population with diverse members, the inheritance of genetic information from par-
ents, and the occasional mutation of genes. Evolutionary Programs (EPs) are general
artificial-evolution search methods based on natural selection and the aforementioned
mechanisms of population genetics. This form of search evolves throughout genera-
tions, improving the features of potential solutions by means of biologically-inspired
operations. Although they represent a crude simplification of natural evolution, EPs
provide efficient and extremely robust search strategies.
Genetic algorithms are probably the best-known type of EP. Although called an algo-
rithm, a genetic algorithm is actually an adaptive heuristic method. It has achieved
fame among analysts and engineers for its ability to identify good solutions to previ-
ously intractable problems. During recent years, GAs have been applied to hydraulic
network optimization (calibration, design, and pump scheduling) with increasing suc-
cess. Commercial modeling software packages are now making this technology
widely available to engineering professionals. For this reason, this appendix covers
the subject of genetic algorithms in more detail in Section D.4.
Simulated Annealing. Annealing is a formal term for the ancient art of heating
and/or cooling materials to forge pottery, tools, weapons, and works of art. It is the
process of subjecting a substance to changes in pressure or temperature to achieve
desired material properties. As mentioned previously, simulated annealing (SA) is
another random-search technique that exploits the analogy with such a process
(Metropolis, Rosenbluth, Rosenbluth, Teller, and Teller, 1953; Kirkpatrick, Gelatt,
and Vecchi, 1983). This method is based on similarities between the way in which a
metal cools and freezes into a minimum energy crystalline structure (the annealing
process) and the search for a minimum in a more general system.
Simulated annealing approaches the optimization problem similarly to giving the
hiker (from the section on Problem Visualization, page 646) superhuman abilities to
jump from valley to valley and over hills in search for the deepest valley, if minimiz-
ing, or the highest hill, if maximizing. The hiker still cannot see these hills and val-
leys, which means that he cannot use information on slope values and therefore
cannot follow the steepest ascent/descent route.
Assume that the hiker is looking for the deepest valley among a large number of pos-
sible valleys in the landscape. The search begins at a high “temperature,” meaning
that the search allows greater flexibility of movement initially, which enables the
hiker to make very high (and long) jumps. Early in the process, this temperature
allows jumps to be made over any hill (even a mountain) and allows access to any val-
ley. As the temperature declines (that is, the search becomes less flexible), the hiker
cannot jump as high, and may become trapped in a relatively shallow valley. Random
moves produce other possible valleys or states to be explored, and if a new search
point is at a lower altitude (that is, a lower function value) than the previous one, the
heuristic always adopts it. However, not all moves to points higher than the current
“best” are immediately rejected. An acceptance criterion that depends on the differ-
ence between the height of the presently generated point and the last saved lowest val-
ley is used here. The acceptance distribution allows probabilistic decisions to be made
about whether to stay in a new lower valley or to jump out of it. The acceptance crite-
rion depends on the current “temperature,” and it uses a procedure based on the Bolt-
zmann probability distribution.
There are two aspects of the SA process that need more explanation. The first is the
annealing (cooling) schedule — the rules for lowering temperature as the search
progresses. The second important aspect is the determination of how many random
steps are sufficient at each temperature. If the temperature is decreased too slowly, the
search will be inefficient. However, if the cooling is too rapid, the search will be
trapped in a sub-optimal region of the search space. Two different cooling schedules
are generally used in practice. The first reduces the temperature by a constant amount
in each phase, while the second reduces the temperature by a constant factor (for
example, 10 percent). With the first method, the simulation proceeds for the same
number of search steps in the high-, intermediate- and low-temperature regimes,
while the second method causes the process to spend more time in the low-tempera-
ture regime.
Simulated annealing can deal with highly nonlinear models, chaotic and noisy data,
multiple local optima, and many constraints. It is a robust and general technique
whose main advantage over local search techniques is its ability to approach global
optimality; however, it does not guarantee global optimality.
Goldman (1998) and Goldman and Mays (1999) demonstrated how simulated anneal-
ing could be applied to system operation while Cunha and Sousa (1999) showed how
it could be used in design.
Ant-Colony Search. Ant-colony optimization (ACO) was inspired by the obser-
vation of real ant colonies. Ants are a classic example of social insects that work
together for the good of the colony. A colony of ants finds new food sources by send-
ing out foragers who explore the surroundings more or less at random. If a forager
finds food, it will return to the colony, leaving a pheromone trail as it goes. Ants can
smell the pheromone, and, when choosing their way, they tend to choose paths
marked by strong pheromone concentrations. This mechanism allows other ants to
follow the most promising trail back to the food and provides positive feedback to
other ants as more pheromone is left on the same trail. Introducing pheromone decay
provides an opposite, negative-feedback mechanism. Although the behavior described
is fairly simple, a group of ants can solve the difficult problem of finding the shortest
route among countless possible paths to a food source. This collective behavior that
emerges from a group of social insects is called swarm intelligence.
In the ACO search, a colony of artificial ants cooperates in finding good solutions to
difficult discrete optimization problems. Similar to the real ant colony problem of
finding the shortest path to the food source, the first optimization applications were
trying to solve the famous “traveling salesman problem” (Dorigo and Di Caro, 1999).
This problem is one of several classical combinatorial optimization problems in
which a person must find the shortest path by which to visit a given number of cities,
each exactly once. Each link between two cities i and j has a variable W ij, which is the
intensity of the artificial pheromone trail. At decision points, ants probabilistically
chose the path with the most pheromone, which is not always the best of the paths
found so far.
Ant colony optimization has already been used in many other areas of application,
including sequential ordering problems, quadratic assignment problems, vehicle
routing problems, and other scheduling and partitioning problems. This type of opti-
mization is particularly well-suited to routing and networking problems, which
involve finding the shortest (optimal) and most efficient path for directing resources,
data, and/or information. An important feature of such problems is the necessity for
backup plans. If a particular route is down due to a failure, then there is a need to
know another route that is almost as efficient as the initial, desirable route. By main-
taining pheromone trails and continuously exploring new paths, the ants can easily
respond to changes in their environment. This property, which may explain the eco-
logical success of real ants, is crucial for many applications.
Maier et al. (2001) and Simpson et al. (2001) demonstrated how ant colony optimiza-
tion could be applied to water distribution system design.
Tabu Search. The roots of tabu search (TS) go back to the 1970s, but the tech-
nique in its present form was first presented by Glover (1986). The systematic use of
memory is an essential feature of tabu search. Most exploration methods keep in
memory only the value of the best solution found up to that particular iteration; how-
ever, TS will also keep information on the itinerary through the last solutions visited.
Such information will be used to guide the move from the current to the next solution.
The role of memory is to restrict the choice of moves by forbidding those which were
not fruitful in the previous iterations. In that sense, tabu search is an extension of a
steepest descent method with the addition of the systematic use of memory.
To prevent the search from endlessly cycling between the same solutions, the
attribute-based memory of TS is structured at its first level to provide a short-term
memory function. If a running list of the attributes of all explored moves is created to
represent the trajectory of solutions encountered, then so-called tabu lists may be
introduced. Based on certain restrictions, these tabu lists keep track of moves by
recording complimentary attributes from the running list. These attributes will be for-
bidden from being embodied in moves selected in the next iteration because their
inclusion might lead back to a previously visited solution. Thus, the tabu list restricts
the search to a subset of admissible moves consisting of admissible attributes or com-
binations of attributes. The goal is to permit “good” moves in each iteration without
revisiting solutions already encountered. This mechanism forces the exploration of
the solution space outside of local optima.
Ribeiro and da Conceicao Cunha (2000) and Fanni, Liberatore, Sechi, Soro, and Zud-
das (2000) demonstrated how tabu search alogorithms could be used for water distri-
bution system optimization.
Section D.4 Genetic Algorithms 673
D.4 GENETIC ALGORITHMS

The theory behind GAs was proposed by Holland (1975) and further developed in the
1980s by Goldberg (1989) and others. These methods rely on the collective learning
process within a population of individuals, each of which represents a search point in
the space of potential solutions. Various applications have been presented since the
first works, and GAs have clearly demonstrated their capability to yield good solu-
tions even in the complicated cases of multi-peak, discontinuous, and non-differentia-
ble functions.
The following are the steps in a standard GA run:
1. Randomly generate an initial population of solutions.
2. Compute the fitness of each solution in the initial population.
3. Generate a new population using biologically inspired operators: reproduction

(crossover) and mutation.
4. Compute fitness of the new solutions.
5. Stop if the termination condition is reached, or repeat steps 3 through 5 to pro-

duce successive generations.
The analogy with nature is established by the creation within a computer of an initial
population of individuals (step 1) represented by chromosomes, which are, in essence,
a set of character strings that are analogous to the chromosomes found in human
DNA. Each of the chromosomes represents a possible location in a multidimensional
search space. In the function optimization example, a chromosome may represent a
set of parameter values x1 (being optimized) generated randomly within prespecified
bounds. In a pipe-sizing problem, a chromosome is a trial solution consisting of pipe
sizes, C-factors, pump start times, and so forth.
Standard GAs use a binary alphabet (characters may be zeros or ones) to form chro-
mosomes. Parameters being optimized are coded using binary strings. Suppose that
the length of a string is 8, and that the function f(x) to optimize is equal to the number
of ones in the chromosome. This is an extremely simple function from an example by
Mitchell (1999), which is presented here to illustrate the methodology. Figure D.17
shows a typical 8-character string for an 8-bit chromosome. Each bit is analogous to a
gene.
Figure D.17
A binary chromosome
It should be noted that not all EPs restrict representation to the binary alphabet. This
additional flexibility makes them more applicable to a variety of decision-making
problems.
Typical values for the population size in a GA are in the range of 50 to 1,000. After
the initial population has been generated, the individuals in the population go through
a process of evolution. In nature, different individuals compete for resources (food,
water, and shelter) in the environment. Some prove to be better at obtaining these
resources than others. Those that are better are more likely to survive, attract mates,
have and successfully rear offspring, and thus propagate their genetic material. The
measure of how good the individual is at competing in its environment is called the
fitness of the individual, which is analogous to the value of the objective function.
Consequently, the selection of who gets to mate is a function of the fitness of the indi-
vidual. The value of the function being optimized for a particular set of parameter val-
ues (as defined by the chromosome) is used for evaluating fitness. Because there are
four 1s in the chromosome in Figure D.17, the value of the fitness is 4.
In genetic algorithms, constraints are usually handled using a penalty function

approach in which solutions with constraint violations are penalized. Rather than
ignoring infeasible solutions and concentrating only on feasible ones, infeasible solu-
tions are allowed to join the population and help guide the search, but for a certain
price. A penalty term incorporated in the fitness function is activated for an infeasible
solution, thus reducing its fitness relative to the other solutions in the population. The
penalty function should be graded as a function of the distance from feasibility. The
penalty multiplier is usually chosen to normalize nominal values of the penalties to
the same scale as the basic fitness of the solution. The multiplier can also be made a
function of the generation number, which allows a gradual increase in the penalty
term to ensure that the final solution is feasible.
In nature, sexual reproduction allows the creation of genetically different offspring

that still belong to the same species as their parents. A simplified look at what hap-
pens at the molecular level reveals that paired chromosomes exchange pieces of
genetic information. This is the recombination operation, which is generally referred
to as crossover because of the way that genetic material crosses over from one chro-
mosome to another. During the reproductive phase of the GA (step 3), individuals are
selected from the population and recombined, producing offspring which will com-
prise the next generation. Crossover takes two individuals (parents in Figure D.18)
and cuts their chromosome strings at some randomly chosen point. The newly created
head segments stay in their respective places, and the tail segments are crossed over to
produce two new chromosomes. Typical values for the probability that a selected indi-
vidual will undergo crossover are in the range 0.5 to 1.0 (50 to 100 percent of the
selected individuals).
Mutation also plays a role in the reproduction phase, though it is not the dominant
role, as is popularly believed, in the process of evolution. In GAs, mutation randomly
alters individual genes with a small probability, thus providing a small amount of ran-
dom search (in Figure D.19 only one gene is changed). If the probability of mutation
is too high, the search degenerates into a random search. This degeneration should not
be allowed; a properly tuned GA is not a random search for a solution to a problem.
The typical values of the probability of mutation are in the range 0.001 to 0.01 (0.1 to
1 percent). As a simulation of a genetic process, a GA uses stochastic mechanisms,
but the result is distinctly nonrandom.
Section D.4 Genetic Algorithms 675
Figure D.18
Single-point crossover
Figure D.19
A single mutation
The simple procedure described above is the basis for most applications of GAs. To
balance exploration and exploitation, a number of parameters must be decided upon,
such as the size of the population and the probabilities of crossover and mutation.
To continue with the example of counting 1s in a chromosome, assume that the popu-
lation size is 4 (an unrealistically low value used only for illustration purposes), the
probability of crossover is 0.7, and the probability of mutation is 0.001. The initial,
randomly generated population might look like this:
Chromosome Number Chromosome String Fitness (# of 1s)

A 00000110 2
B 11101110 6
C 00100000 1
D 00110100 3
A commonly used selection method in GAs is fitness-proportionate selection, in

which the number of times an individual is expected to reproduce is proportional to
the ratio of its fitness to the total fitness of the population. A simple method of imple-
menting this method is known as roulette-wheel selection (Goldberg, 1989). Roulette-
wheel selection assumes that each individual will be assigned a slice on a roulette
wheel. However, unlike the roulette wheels found in casinos, these slices are of differ-
ent areas; fitter solutions will have a larger area and, consequently, a greater chance of
being selected for reproduction when the roulette wheel is spun. In the above exam-
ple, the wheel would be spun four times to select two pairs of parents. If these parents
were selected as B, D and B, C, respectively (where A is not selected because of the
probabilistic nature of the wheel), the next step would be to perform crossover on
these two pairs with a crossover probability of 70 percent (that is, there is a 70 percent
chance that crossover will occur). If we assume that due to chance, parents B and D
crossed over after the first bit position to form offspring E=10110100 and
F=01101110, and parents B and C do not cross over, then the following intermediate
population is created:

E 10110100 4
F 01101110 5
B 11101110 6
C 00100000 1
Finally, to finish one cycle (generation) of a GA run, it is necessary to test chromo-

somes for mutation, where each gene in a chromosome could be mutated with a low
probability of 0.1 percent. For example, if the sixth gene in offspring E is mutated to
form E' = 10110000, offspring F and C are not mutated at all, and the first gene of off-
spring B is mutated to form B' = 01101110 (note that simply by chance these two
mutations have reduced the fitness of the chromosomes), the new population will be
the following:

E' 10110000 3
F 01101110 5
B' 01101110 5
C 00100000 1
In this example, although the best string from the initial population (B, fitness = 6)
was lost, the average fitness of the population has increased from 12/4 to 14/4. Iterat-
ing this procedure will eventually result in a string with all ones.
Genetic algorithms are particularly suited for use with highly nonlinear combinatorial
problems, chaotic and noisy data, and search spaces with multiple local optima and
many constraints. They are a robust and general technique whose main advantage
over local search techniques is the ability to approach global optimality, as has been
Section D.5 Multiobjective Optimization 677
shown in many practical situations dealing with difficult problems. Although they are
not guaranteed to find the global optimum, GAs are generally good at finding
“acceptably good” solutions to problems “acceptably quickly.” Where specialized
techniques exist for solving a particular problem, they are likely to outperform GA in
both speed and accuracy of the final result. GAs can be computationally intensive
when objective function evaluation requires significant computational resources such
as those required for the hydraulic analysis of a large water distribution model. GAs
can also be more computationally consuming because the same individual may be
repeatedly evaluated in numerous generations, and because solutions improve slowly
as one gets near the global optimum.
Genetic algorithms have been widely used in recent years for a variety of water distri-
bution system problems. Walters and Lohbeck (1993); Simpson, Dandy, and Murphy
(1994); Dandy, Simpson, and Murphy (1996); Halhal, Walters, Savic, and Ouzar
(1997); Savic and Walters (1997); Lingireddy and Ormsbee (1999); Wu and Simpson
(2001); Wu, Boulos, Orr, and Ro (2001); and Wu et al. (2002a) have demonstrated the
use of GA for system design and rehabilitation (see page 360).
Savic and Walters (1995) and Wu et al. (2002b) showed that GA could be used for
model calibration (see page 268). Meier and Barkdoll (2000) used GA to identify
sampling locations for flow tests.
D.5 MULTIOBJECTIVE OPTIMIZATION

Many real-world engineering design or decision-making problems involve the simul-
taneous optimization of multiple objectives. The principle of multiobjective optimiza-
tion is different from that of single-objective optimization. In single-objective
optimization, the goal is to find the best solution, which corresponds to the minimum
or maximum value of the objective function. In multiobjective optimization with con-
flicting objectives, there is no single optimal solution. The interaction among different
objectives gives rise to a set of compromised solutions, largely known as the Pareto-
optimal (or trade-off, nondominated, or noninferior) solutions (see Figure D.20).
Each solution of the Pareto optimal set is not dominated by any other solution. In
going from one solution to another, it is not possible to improve on one objective (for
example, reduction of the amount of iron in water) without making at least one of the
other objectives worse (for example, failure to minimize cost).
All solutions above the Pareto optimal curve are feasible (in this particular example
where both objectives are minimized), while those below the curve are infeasible (see
Figure D.20). Feasible solutions that are dominated by one or more solutions are
called inferior. For example, solution C is clearly dominated by solution B (in the
Pareto set) because B achieves more of both objectives (for example, larger reduction
in water quality risk and lower cost). In fact, any solution in the shaded area to the
“southwest” of C dominates solution C. It is clear, however, that there is a need to
identify as many solutions as possible within the Pareto-optimal range to ensure that
an acceptable solution will be produced and selected by the decision-maker.
Figure D.20
Pareto solutions
Weighting Method
Many, if not all, of the single-objective optimization techniques can be used to gener-
ate a subset of the Pareto-optimal range if more objectives are identified and these
techniques are used appropriately. This subset can be generated by weighting the
objectives to obtain Pareto solutions. Point A (Figure D.20) represents a cheap solu-
tion (low f1), but has the highest level of water quality risk (high f2). On the opposite
side of the curve, point B represents the most expensive solution found (high f1) that
achieves the lowest water quality risk (low f2). Each of the points could be generated
by using one of the single-optimization methods presented previously.
The two objective functions must be lumped together to create a single objective func-
tion. This lumping is possible if, for example, the risk objective can be expressed in
monetary terms. If a dollar figure can be attached to the risk level, then the multiob-
jective problem could be reduced to a single-objective problem. The weight (conver-
sion) factor w is multiplied by the values of the objective f2 in order to obtain a single,
combined objective function f = f1 + wf2. Note that this objective function has a single
dimension (dollars, in this case). Next, given this single objective function and a set of
constraints, one can find a single optimal solution by using some of the previously
introduced optimization methods. In the run to identify point A, w takes its largest
feasible value (more weight is put on f2), and in the run to identify point B, w takes its
lowest feasible value (more weight is put on f1). All intermediate solutions on the
Pareto curve are generated by using the different weight ratio between the two objec-
tives. The procedure for weighting the objectives to obtain the Pareto solution is
called the weighting method (Cohon, 1978).
The method has several weaknesses. One problem is that there are cases in which one
objective value can vary by orders of magnitude more than another, and therefore
Section D.5 Multiobjective Optimization 679
large changes in the weights can lead to no corresponding change in the objective val-
ues. Similarly, there are cases when small changes in the weights can cause a large
change in the objective values. However, the weighting method’s most serious draw-
back is that it cannot generate proper members of the Pareto-optimal front when this
front is not convex (see Figure D.11 for an illustration of convex and non-convex
regions).
Constraint Method
The constraint method represents an alternative framework for generating Pareto-
optimal solutions. It operates by optimizing one objective while constraining the oth-
ers to some value (Cohon, 1978). In the preceding example (Figure D.20), instead of
expressing one objective in the same units as the other (that is, expressing the risk
objective in monetary terms), one objective can be expressed as a constraint. The sin-
gle optimization problem is then solved for the remaining objective. If single-objec-
tive optimization is run with “minimize risk (f2)” as an objective function and a
A
constraint that requires that f 1 d f 1 (that is, the cost must be less than or equal to a
certain budget, f1A), then the solution obtained should identify point A. In this process,
the original feasible region is first reduced to the area to the left of f1A. The minimiza-
tion of f2 over the new feasible region then leads to point A. Similarly, any other point
N
on the curve can be obtained by appropriately constraining the budget to f 1 d f 1 . This
type of approach fails to take advantage of some methods’ (for example, GA’s) ability
to thoroughly search the decision space.
The main disadvantage of both the weighting and constraint methods is that not all of
the points can be identified because the appropriate weights (or constraint levels) are
not known in advance. These methods also suffer from high computational costs
because the number of optimization runs increases exponentially with the number of
objectives.
The need to identify as many solutions as possible within the Pareto-optimal range
often represents a problem for standard solution-generating techniques. By maintain-
ing and continually improving a population of solutions, a GA can search for many
nondominated solutions at the same time (in a single run), which makes it a very
attractive tool for solving multiobjective optimization problems (Fonseca and Flem-
ing, 1997).
Examples of multiobjective optimization problems within water distribution model-
ing practice are abundant. In fact, it could be said that water distribution problems are
inherently multiobjective. However, the lack of appropriate user-friendly tools
(among other problems) has hampered the wider adoption of these techniques by
practitioners. Commonly encountered multiobjective problems are those of the design
of water distribution systems in which cost and capacity are the most obvious con-
flicting objectives. Other benefits, such as improved water quality, security of supply,
reliability, and so on, should also be considered when designing or rehabilitating a
water distribution system. Operational optimization can also benefit from the use of
multiobjective optimization. Examples of multiobjective analysis that consider the
energy cost of alternative pump schedules and the number of pump switches within
each schedule represent a better way of assessing tradeoffs between the energy costs
and the maintenance costs caused by an excessive number of pump switches. Again,
other objectives like water quality and security of supply due to interruptions in
pumping could also be taken into account in the multiobjective framework.
Using multiobjective analysis, the decision-makers can better assess the tradeoffs
between different objectives. Although by using this approach they cannot identify a
solution that is clearly the best, they can discover a set of good (near optimal) solu-
tions, have reasonable grounds to make sensible decisions, and avoid those alterna-
tives that are clearly poor.
REFERENCES
Alperovits, G., and Shamir, U., (1977). “Design of Optimal Water Distribution Systems.” Water Resources
Research, 13(6), 885.
Austin, and Robinson. (1976). “Cost Optimization of Rural Water Systems.” Journal of Hydraulics Divi-
sion, ASCE, 102(8), 1119.
Bellman, R. E. (1957). Dynamic Programming. Princeton University Press.
Bhave, P. R. (1980). “Selecting Pipe Sizes in Network Optimization by LP.” Journal of Hydraulics Divi-
sion, ASCE, 105(HY7), 1019.
Bhave, P. R. (1983). “Optimization of Gravity Fed Water Distribution Systems.” Journal of Environmental
Engineering, 104(4), 799.
Boccelli, D. L., Tryby, M. E., Uber, J. G., Rossman, L. A., Zierolf, M. L., and Polycarpou, M. M. (1998).
“Optimal Scheduling of Booster Disinfection in Water Distribution Systems.” Journal of Water
Brion, L. M., and Mays L. W. (1991). “Methodology for Optimal Operation of Pumping Stations in Water
Distribution Systems.” Journal of Hydraulic Engineering, ASCE, 117(11), 1551.
Camp, T. R. (1939). “Economic Pipe Sizes for Water Distribution Systems.” Transactions of the American
Society of Civil Engineers, 104, 190.
Cohon, J. L. (1978). Multi-objective Programming and Planning. Academic Press, New York, New York.
Constans, S., Brémond, B., and Morel, P. (2000). “Using Linear Programs to Optimize the Chlorine Con-
centrations in Water Distribution Networks.” Proceedings of the ASCE Joint Conference on Water
Resources Engineering and Water Resources Planning and Management, Minneapolis, Minnesota.
Coulbeck, B. (1984). “Optimization of Water Networks.” Transactions of Institute of Measurements and

Control, 6(5), 271.
Coulbeck, B., and Sterling, M. (1978). “Optimal Control of Water Distribution Systems.” Proceedings of
the Institute of Electrical Engineers, 125, 1039.
Cowan, J. (1971). “Checking Trunk Main Designs for Cost-Effectiveness.” Water and Water Engineering, 385.
Cullinane, M. J., Lansey, K., and Mays L.W. (1992). “Optimization of Availability-based Design of Water
Distribution Networks.” Journal of Hydraulic Engineering, 118(3), 420.
Cunha, M. D., and Sousa, J. (1999). “Water Distribution Network Design Optimization: Simulated Anneal-
ing Approach.” Journal of Water Resources Planning and Management, ASCE, 125(4), 215.
Dancs, L. (1977). “Sizing Force Mains for Economy.” Water and Sewage Works.
References 681
Dandy, G. C., Simpson, A. R., and Murphy, L. J. (1996). “An Improved Genetic Algorithm for Pipe Net-
work Optimization.” Water Resources Research, 32(2), 449.
Deb, A. K. (1973). “Least Cost Design of Water Main System in Series.” Journal of the Environmental
Engineering Division, 99(2), 405.
Deb, A. K. (1976). “Optimization of Water Distribution Network Systems.” Journal of the Environmental
Engineering Division, 102(4), 837.
deNeufville, R., Schaake, J., and Stafford, J. J. (1971). “Systems Analysis of Water Distribution Networks.”
Journal of Sanitary Engineering Division, ASCE, 97(6), 825.
Dorigo, M., and Di Caro, G. (1999). “The Ant Colony Optimization Meta-Heuristic.” New Methods in
Optimization, Corne, D., Dorigo, M., and Glover, F., eds., McGraw-Hill.
Duan, N. L., Mays, L. W., Lansey, K. E. (1990). “Optimal Reliability-Based Design of Pumping and Dis-
tribution Systems.” Journal of Hydraulic Engineering, ASCE, 116(2), 249.
Fanni, A., Liberatore, S., Sechi, G. M., Soro, M. and Zuddas, P. (2000). “Optimization of Water Distribu-
tion Systems by a Tabu Search Metaheuristic.” Proceedings of the 7th IFORMS Computing Society
Conference, Cancun, Mexico.
Finney, R. L., Weir, M. D., and Giordano, F. R. (2001). Thomas’ Calculus. Addison Wesley Longman,
Tenth Edition, Boston, Massachusetts.
Fonseca, C. M., and Fleming, P. J. (1997). “Multi-objective Optimization.” Handbook of Evolutionary
Computation, Bäck, In T., Fogel, D. B., and Michalewicz, Z., eds., Institute of Physics Publishing and
Oxford University Press.
Glover, F. (1986). “Future Paths for Integer Programming and Links to Artificial Intelligence.” Computers
and Operations Research, 13, 533.
Goldberg, D. E. (1989). Genetic Algorithms in Search, Optimization, and Machine Learning. Addison-
Wesley.
Goldman, F. E. (1998). The Application of Simulated Annealing for Operation of Water Distribution Sys-
tems. Ph.D. dissertation, Arizona State University.
Goldman, F. E., and Mays, L. W. (1999). “The Application of Simulated Annealing to the Optimal Opera-
tion of Water Systems.” Proceedings of the 26th ASCE Water Resources Planning & Management Con-
ference, American Society of Civil Engineers, Tempe, Arizona.
Halhal, D., Walters, G. A., Savic, D. A., and Ouazar, D. (1997). “Water Network Rehabilitation with a Struc-
tured Messy Genetic Algorithm.” Journal of Water Resources Planning and Management, 123(3), 137.
Hillier, F. S., and Lieberman, G. J. (1995). Introduction to Operations Research. McGraw-Hill, New York,
New York.
Holland, J. H. (1975). Adaptation in Natural and Artificial Systems. MIT Press.
Jacoby, S. L. S. (1968). “Design of Optimal Hydraulic Networks.” Journal of Hydraulics Division, ASCE,
94(3), 641.
Jowitt, P. W., Garrett, R., Cook, S., and Germanopoulos, G. (1988). “Real-time Forecasting and Control of
Water Distribution.” Computer Applictions in Water Supply, Coulbeck, B., and Orr, C. H., eds.,
Research Studies Press Ltd., Leicester, United Kingdom.
Karmeli, D., Gadish, Y., and Meyer, S. (1968). “Design of Optimal Water Distribution Networks.” Journal
of Pipeline Division, ASCE, 94(1), 1.
Kettler, A. J., and Goulter, I. C. (1983). “Reliability Considerations in the Least Cost Design of Looped
Water Distribution Systems.” Proceedings of the International Symposium on Urban Hydrology,
Hydraulics, and Sediment Control, University of Kentucky, Lexington, Kentucky.
Kirkpatrick, S., Gelatt, C. D., and Vecchi, M. P. (1983). “Optimization by Simulated Annealing.” Science,
220, 671.
Lam, C. F. (1973). “Discrete Gradient Optimization of Water Systems.” Journal of Hydraulics Division,
ASCE, 99(6), 863.
Lansey, K. E. (1988). “Water Distribution Network Calibration Considering Multiple Loading Conditions.”
Proceedings of 15th Annual ASCE Water Resources Conference, American Society of Civil Engineers, 150.
Lansey, K. E., and Awumah, K. (1994). “Optimal Pump Operation Considering Pump Switches.” Journal
of Water Resources Planning and Management, ASCE, 120(1), 17.
Lansey, K. E., and Mays, L. W. (1989). “Optimal Design of Water Distribution Systems.” Journal of Water
Lasdon, L. S., and Waren, A. D. (1982). GRG2 User’s Guide. Department of General Business, University
of Texas, Austin.
Liang, T. (1971). “Design of Conduits System by Dynamic Programming.” Journal of Hydraulics Division,
ASCE, 97(3), 383.
Lingireddy, S., and Ormsbee, L. E. (1999). “Optimal Network Calibration Model Based on Genetic Algo-
rithms.” Proceedings of the ASCE Annual Conference of Water Resources Planning and Management,
Tempe, Arizona.
Loganathan, G. V., Greene, J., and Ahn, T. (1995). “Design Heuristic for Globally Minimum Cost Water
Distribution Systems.” Journal of Water Resources Planning and Management, 121(2), 182.
Maier, H. R., Simpson, A. R., Foong, W. K., Phang, K. Y., Seah, H. Y., and Tan, C. L. (2001). “Ant Colony
Optimization for the Design of Water Distribution Systems.” Proceedings of the World Water and Envi-
ronmental Resources Congress, Orlando, Florida.
Meier, R. W., and Barkdoll, B. D. (2000). “Sampling Design for Network Model Calibration Using Genetic
Algorithms.” Journal of Water Resources Planning and Management, ASCE, 126(4), 245.
Metropolis, N., Rosenbluth, A. W., Rosenbluth, M. N., Teller, A. H., and Teller, E. (1953). “Equation of
State Calculations by Fast Computing Machines.” Journal of Chemical Physics, 21(6), 1087.
Mitchell, M. (1999). An Introduction to Genetic Algorithms. The MIT Press, Cambridge, Massachusetts.
Morgan, D. R., and Goulter, I. C. (1985). “Optimal Urban Water Distribution Design.” Water Resources
Research, 21(5), 642.
Nelder, J. A., and Mead, R. (1965). “A Simplex Method for Function Minimization.” Computer Journal, 7, 308.
Ormsbee, L. E. and Chase, D. V. (1988). “Optimal Pump Operation Using Nonlinear Programming.” Pro-
ceedings of 15th Annual ASCE Water Resources Conference, 158.
Ormsbee, L. E., and Contractor, D. N. (1981). “Optimization of Hydraulic Networks.” Proceedings of the
International Symposium on Urban Hydrology, Hydraulics, and Sediment Control, Lexington, Kentucky.
Ormsbee, L. E., and Lingireddy, S. L. (1995). “Nonlinear Heuristic for Pump Operation.” Journal of Water
Ormsbee, L. E., Walski, T. M., Chase, D. V., and Sharp, W. (1989). “Methodology for Improving Pump
Operational Efficiency.” Journal of Water Resources Planning and Management, ASCE, 15(2), 148.
Quindry, G. E., Brill, E. D., and Liebman, J. C. (1981). “Optimization of Looped Water Distribution Sys-
tems.” Journal of Environmental Engineering Division, ASCE, 107(4), 665.
Ribeiro, L. and da Conceicao Cunha, M. (2000). “Tabu Search Algorithms for Water Network Distribution
Systems Optimization.” Proceedings of the Decision Making in Urban and Civil Engineering Confer-
ence, Lyon, France.
Sabel, M. H., and Helwig, O. J. (1985). “Cost Effective Operation of Urban Water Supply System Using
Dynamic Programming.” Water Resources Bulletin, 21(1), 75.
Sakarya, A. B. A., and Mays, L. W. (2000). “Optimal Operation of Water Distribution Pumps Considering
Water Quality.” Journal of Water Resources Planning and Management, 126(4), 210.
References 683
Salcedo, D., and Weiss, A. O. (1972). “Solution of Water Resources Problems with Economies of Scale by
Linear Programming.” Water Resources Bulletin, 8, 546.
Savic, D. A., and Walters, G. A. (1995). “Genetic Algorithm Techniques for Calibrating Network Models.”
Report No. 95/12, Centre For Systems And Control Engineering, School of Engineering, University of
Exeter, Exeter, United Kingdom, 41.
Savic, D. A., and Walters G. A. (1997). “Genetic Algorithms for Least-Cost Design of Water Distribution
Networks.” Journal of Water Resources Planning and Management, ASCE, 123(2), 67.
Schwefel, H. P. (1981). Numerical Optimisation of Computer Models. Wiley.
Shamir, U. (1974). “Optimal Design and Operation of Water Distribution Systems.” Water Resources
Research, 10(1), 27.
Shamir, U., and Howard, C. D. D. (1979). “An Analytic Approach to Scheduling Pipe Size Improvements.”
Simpson, A. R., Dandy, G. C., and Murphy, L. J. (1994). “Genetic Algorithms Compared to Other Techniques
for Pipe Optimization.” Journal of Water Resources Planning and Management, ASCE, 120(4), 423.
Simpson, A. R., Maier, H. R., Foong, W. K., Phang, K. Y., Seah, H. Y., and Tan, C. L. (2001). “Selection of
Parameters for Ant Colony Optimisation Applied to the Optimal Design of Water Distribution Sys-
tems.” Proceedings of the International Congress on Modelling and Simulation MODSIM, Canberra,
Australia.
Stathis, J. A., and Loganathan, G. V. (1999). “Analysis of Pressure-Dependent Leakage in Water Distribu-
tion Systems.” Proceedings of the Water Resources Planning & Management Conference, American
Society of Civil Engineers, Tempe, Arizona.
Sterling, M. J. H., and Coulbeck, B. (1975). “A Dynamic Programming Solution to Optimization of Pump-
ing Costs.” Proceedings of Institute of Civil Engineers, 59(2), 813.
Swamee, P. K., and Khanna, P. (1974). “Equivalent Pipe Methods and Optimizing Water Networks—Facts
and Fallacies.” Journal of Environmental Engineering Division, ASCE, 100(EE1), 93.
Van Zyl, J. E. (2001). A Methodology for Improved Operational Optimization of Water Distribution Sys-
tems. Ph. D. Thesis, University of Exeter, United Kingdom.
Van Zyl, J. E., Savic, D. A., and Walters, G. A. (2001). “A Hybrid Method for Operational Optimization
of Water Distribution Systems.” Water Software Systems: Theory and Applications, Vol. 2, Ulanicki,
B., Coulbeck, B., and Rance, J. P., eds., Research Studies Press, Baldock, Hertfordshire, United
Kingdom, 89.
Wagner, H. M. (1975). Principles of Operations Research. Prentice-Hall, Inc., Englewood Cliffs, New Jersey.
Walski, T. M. (1982). “Economic Analysis for Rehabilitation of Water Mains.” Journal of Water Resources
Walski, T. M. (1984). Analysis of Water Distribution Systems. Van Nostrand Reinhold, New York.
Walski, T. M. (1985). “Cleaning and Lining vs. Parallel Mains.” Journal of Water Resources Planning and
Walski, T. M. and Pelliccia, A. (1982). “Economic Analysis of Water Main Breaks.” Journal of the Ameri-
can Water Works Association, 74(3), 140.
Walters, G. A., and Lohbeck. (1993). “Optimal Layout of Tree Networks Using Genetic Algorithms.” Engi-
neering Optimization, 22(1), 27.
Watanatada, T. (1973). “Least Cost Design of Water Distribution Systems.” Journal of the Hydraulics Divi-
sion, ASCE, 99(9), 1497.
Wu, Z. Y., Boulos, P. F., Orr, C. H., and Ro, J. J. (2001). “Rehabilitation of Water Distribution System
Using Genetic Algorithms.” Journal of the American Water Works Association, 93(11), 74.
Wu, Z. Y., and Simpson, A. R. (2001). “Competent Genetic Algorithm Optimization of Water Distribution
Systems.” Journal of Computing in Civil Engineering, ASCE, 15(2), 89.
Wu, Z. Y, Walski, T. M., Mankowski, R., Herrin G., Gurierri, R. and Tryby, M. (2002b). “Calibrating Water
Distribution Model Via Genetic Algorithms.” Proceedings of the AWWA Information Management and
Technology Conference, American Water Works Association, Kansas City, Missouri.
Wu, Z. Y, Walski, T. M., Mankowski, R., Tryby, M., Herrin, G., and Hartell, W. (2002a). “Optimal Capac-
ity of Water Distribution Systems.” Proceedings of the 1st Annual Enviromental and Water Resources
Systems Analysis (EWRSA) Symposium, Roanoke, Virginia.
Yeh, W. G. (1985). “Reservoir Management and Operations Models: a State-of-the-Art Review.” Water
Resources Research, 21(12), 1797.
ADVANCED WATER
AND MANAGEMENT
Authors
Thomas M. Walski
Donald V. Chase
Dragan A. Savic
Walter Grayman
Stephen Beckwith
Edmundo Koelle
Peer Review Board


A P P E N D I X
E
SCADA Basics
A Supervisory Control and Data Acquisition (SCADA) system is a widely distributed

computerized system primarily used to remotely control and monitor the condition of
field-based assets from a central location. Field-based assets include wells, pump sta-
tions, valves, treatment plants, tanks, and reservoirs.
For a water distribution network, the common objectives of a SCADA system are to
do the following:
• Monitor the system
• Obtain control over the system and ensure that required performance is
always achieved
• Reduce operational staffing levels through automation or by operating a sys-
tem from a single central location
• Store data on the behavior of a system and therefore achieve full compliance
with mandatory reporting requirements for any regulatory agency
• Provide information on the performance of the system and establish effective
asset management procedures for the system
• Establish efficient operation of the system by minimizing the need for rou-
tine visits to remote sites and potentially reduce power consumption during
pumping operations through operational optimization
• Provide a control system that will enable operating objectives to be set and
achieved
• Provide an alarm system that will allow faults to be diagnosed from a central
point, thus allowing field repair trips to be made by suitably qualified staff to
correct the given fault condition and to avoid incidents that may be damaging
to the environment
686 SCADA Basics Appendix E
E.1 COMPONENTS OF A SCADA SYSTEM

SCADA encompasses the transfer of data between a SCADA central host computer
and a number of remote sites (Remote Terminal Units or RTUs), and the central host
and the operator terminals. Figure E.1 shows a generic SCADA system that employs
some form of data multiplexing (MUXs) between the central host and the RTUs.
These multiplexers serve to route data to and from a number of RTUs on a local net-
work, while using one or very few physical links on a Wide Area Network (WAN)
backbone to pass data back to the central host computer.
Figure E.1
Generic SCADA
system network
SCADA systems consist of

• One or more field data interface devices usually called Remote Stations,
Remote Terminal Units (RTUs), or Programmable Logic Controllers (PLCs),
which interface to field sensing devices and local control switchboxes and
valve actuators.
• A communications system used to transfer data between field data interface
devices and control units and the computers in the SCADA central host. The
system can be radio, telephone, cable, satellite, and so on, or any combina-
tion of these.
Section E.1 Components of a SCADA System 687
• A central host computer server or servers (sometimes called a SCADA Cen-

ter, master station, master terminal unit, or MTU).
• A communications system to support the use of operator workstations that

may be geographically remote from the central host computer.
• A collection of standard and/or custom software [sometimes called Human

Machine Interface (HMI) software or Man Machine Interface (MMI) soft-
ware] systems used to provide the SCADA central host and operator termi-
nal application, support the communications system, and monitor and
control remotely located field data interface devices.
Each of the preceding components is described in the following subsections.
Field Data Interface Devices

Field data interface devices form the “eyes and ears” of a SCADA system. Devices
such as reservoir level meters, water flow meters, valve position transmitters, temper-
ature transmitters, power consumption meters, and pressure meters all provide infor-
mation that can tell an experienced operator how well a water distribution system is
performing. In addition, equipment such as electric valve actuators, motor control
switchboards, and electronic chemical dosing facilities can be used to form the
“hands” of the SCADA system and assist in automating the process of distributing
water.
However, before any automation or remote monitoring can be achieved, the informa-
tion that is passed to and from the field data interface devices must be converted to a
form that is compatible with the language of the SCADA system. To achieve this,
some form of electronic field data interface is required.
Remote Terminal Units (RTUs), also know as Remote Telemetry Units, provide this
interface. RTUs are primarily used to convert electronic signals received from (or
required by) field devices into (or from) the language (known as the communication
protocol) used to transmit the data over a communication channel. RTUs appear in the
field as a box in a switchboard with electrical signal wires running to field devices and
a cable link to a communication channel interface, such as a radio (see Figure E.2).
The instructions for the automation of field data interface devices, such as pump con-
trol logic, are usually stored locally. This is largely due to the limited bandwidth typi-
cal of communications links between the SCADA central host computer and the field
data interface devices. Such instructions are traditionally held within local electronic
devices known as Programmable Logic Controllers (PLCs), which have in the past
been physically separate from RTUs (see Figure E.3). PLCs connect directly to field
data interface devices and incorporate programmed intelligence in the form of logical
procedures that will be executed in the event of certain field conditions. However,
many water systems with SCADA systems have no PLCs. In this case, the local con-
trol logic is held within the RTU or in relay logic in the local switchboard.
Figure E.2
RTU cabinet with
RTU (center), radio
(top center), and field
wiring terminations
(left)
Figure E.3
Programmable Logic
Controller (PLC)
performing local
control functions,
physically separated,
but wired to a nearby
Remote Terminal Unit
(RTU)
PLCs have their origins in the automation industry and therefore are often used in
manufacturing and process plant applications. The need for PLCs to connect to com-
munication channels was not great in these applications, as they often were only
required to replace traditional relay logic systems or pneumatic controllers. SCADA
systems, on the other hand, have origins in early telemetry applications, where it was
only necessary to know basic information from a remote source. The RTUs connected
to these systems had no need for control programming because the local control algo-
rithm was held in the relay switching logic.
As PLCs were used more often to replace relay switching logic control systems,
telemetry was used more and more with PLCs at the remote sites. It became desirable
to influence the program within the PLC through the use of a remote signal. This is in
effect the “Supervisory Control” part of the acronym SCADA. Where only a simple
local control program was required, it became possible to store this program within
the RTU and perform the control within that device. At the same time, traditional
PLCs included communications modules that would allow PLCs to report the state of
the control program to a computer plugged into the PLC or to a remote computer via a
telephone line. PLC and RTU manufacturers therefore compete for the same market.
As a result of these developments, the line between PLCs and RTUs has blurred and
the terminology is virtually interchangeable. For the sake of simplicity, the term RTU
will be used to refer to a remote field data interface device; however, such a device
could include automation programming that traditionally would have been classified
as a PLC.
Field Data Communications System

The field data communications system is intended to provide the means by which data
can be transferred between the central host computer servers and the field-based
RTUs.
Bandwidth. An important property of a communications channel is its capacity to

carry data. The term bandwidth is used to describe this capacity. Originally, the term
bandwidth applied to the width, in Hertz, of an analog channel. For example, a tele-
phony voice channel that occupies the nominal band 0.3 to 3.4 kHz has a bandwidth
of 3.1 kHz and a radio channel that occupies the spectrum from 929.88875 to
929.8875 MHz has a channel bandwidth of 12.5 kHz. With digital transmission, the
term bandwidth has been extended to include the data transmission rate in bits per
second (bps).
SCADA Communications Availability and Protocols. The availability

imposed by the communications infrastructure is an important aspect of the SCADA
system. Because SCADA systems are typically deployed over large geographical
areas, links to remote SCADA outstations from the central host computer are often
multilayered, meaning that there may be several physical and logical paths through
which data must be routed before it reaches the intended destination. Such “long-
haul” links may impose a heavy financial consideration on the type of communication
systems used and the bandwidth used on those links.
As a result of cost constraints, SCADA communications links generally offer less

bandwidth and lower reliability than that offered by communications backbones com-
monly used on a process plant, where there are few geographical constraints and high-
speed fiber optic LAN infrastructures may be employed. Process plant communica-
tion backbones may exhibit 99.9 percent availability (less than 9 hours outage per
year) and bit error rates of better than 10-9 (1 error in every 109 bits). Comparatively on
SCADA links, where a combination of data radio, telephone line, and satellite link
technology may be employed, the availability may be as low as 99.0 percent (an aver-
age of approximately 90 hours outage per year) and bit error rates of 10-6 (1 error in
every 106 bits) or greater.
The availability discrepancy can be attributed to the fact that the multilayer SCADA
links traverse a greater number of media conversion and data routing ports as com-
pared to high-speed optic fiber LAN backbones. Therefore, there are many single
points of failure in a diverse SCADA communications network. Communications out-
ages typically result from equipment and power supply failures and human interfer-
ence. Better availability is possible through the use of redundant communications
paths to outstations; however, such designs can contribute significantly to the cost of a
communications system and therefore may not be financially viable if the communi-
cations link is not crucial to operational security.
SCADA communication protocols are designed specifically for the reduced reliability
communications links typically employed with SCADA systems and to provide
secure transmission of data, guaranteeing reliable delivery of data to the intended des-
tination in most circumstances. The protocols employ error detection and message
retry techniques usually by receive/transmit handshaking established through the use
of “headers and footers” attached to the raw data under transmission. Such extra
information introduces an overhead to the transmission of data, resulting in a trade off
between speed of data transmission and the reliability of the communications link. As
a result, the speed of data communications associated with SCADA is regularly
slower than that typical of a communications backbone commonly used on a process
plant, office, or factory floor application. Not only do the latter boast links using
media such as hard wired optic fiber, which usually allows higher speeds of data
transmission than available over radio links, but the communications protocols associ-
ated with SCADA systems introduce data transmission overheads which further slow
the rate of data transfer. Where media such as low bit error rate optic fiber cabling is
available, simple communications protocols may be used which do not require sub-
stantial transmission overheads.
Users of SCADA systems and the resulting data do not need to be aware of the com-
munications protocols used. In fact, the protocols should be transparent to the user.
However, it is important to understand that with the use of communications links such
as radio, there is a possibility, albeit small, that communication errors will occur.
For example, a control command could be sent to the wrong destination. SCADA sys-
tems often request confirmation from an operator to confirm that a control action is
required. This approach provides some level of protection against a control message
being sent to the wrong destination. It is much more likely, however, that the operator
has made the error and the control action has been requested of the incorrect outsta-
tion. The control confirmation check gives the operator another chance to select the
correct outstation for control.
An example of a SCADA communications protocol includes DNP 3.0 (Distributed

Network Protocol), a vendor independent protocol that incorporates multiple layers of
error detection and correction and allows a select/confirm regime for control actions.
Modbus is another widely used protocol for SCADA, but it does not offer the same
level of data transmission security as DNP 3.0. There are also a large variety of proto-
cols that are proprietary to individual SCADA vendors and offer capabilities similar
to those described in this section.
Common Communications Media. The following communications media

are common:
• Licensed radio links (UHF and VHF)
• Unlicensed “spread spectrum” radio links
• Public switched telephone networks
• Mobile telephony
• Microwave
• Cable TV networks
• Dedicated satellite links
• Dedicated cable, including fiber optics (for very short distance communica-
tion)
• Corporate WAN computer communications systems
For highly critical sites, it is not uncommon for combinations of these different media
to be used to ensure high reliability of communications to the site. Selection of the
preferred communications media depends on several important factors:
• The remoteness of the field equipment site
• The required reliability of the communications media (primarily determined
by the perceived operational importance of the remote site)
• Availability of communications options
• Cost of each option for the particular application
• Availability of power (power company, battery, solar, or other)
The communications systems used for SCADA are often split into two distinct parts:
a Wide Area Network backbone (WAN) and numerous Local Area Networks (LANs).
The interface between the two parts is commonly achieved through some form of
multiplexing.
Wide Area Network Backbone. The WAN connects the central host com-
puter to the multiplexers. It may comprise cable, radio, or satellite data communica-
tions links depending on the geographic distribution of the SCADA system. The
WAN links are generally full duplex (they provide simultaneous data transmission in
both directions) and may be configured in a star or loop topology.
The star and loop topologies employ dedicated point-to-point communications links
between multiplexers. The star configuration (as shown in Figure E.1) does not pro-
vide WAN redundancy. The loop configuration (see Figure E.4) links adjacent multi-
plexers and provides alternative communications paths for redundancy, therefore
providing higher reliability. Looped WANs require data traffic routers, and links must
be dimensioned to carry all WAN traffic.
Figure E.4
Looped WAN
configuration
In some cases, a WAN is not needed. An example is a simple SCADA system where
all RTUs are connected directly to the central host computer via a single multi-drop
communications link. These systems therefore effectively only contain an RTU local
area network.
Multiplexers. Generally, some form of multiplexing is required to connect a
WAN backbone to a local network of RTUs. Multiplexers allow different data streams
to share a single data link, as shown in Figure E.5. Multiplexers combine communica-
tions paths to and from many RTUs into a single bit stream, usually using time divi-
sion multiplexing (TDM) or other such bit stream manipulation techniques. The
multiplexers must be able to combine the traffic to and from tens or sometimes hun-
dreds of RTUs for transmission over the SCADA WAN.
Figure E.5
Basic data multiplexer
configuration
A simple form of multiplexer is to use a data traffic router together with a point-to-
multipoint radio, as shown in Figure E.6. In this diagram, LR refers to Local Radio,
PMR to Point to Multipoint Radio, and ROUT to data router.
Figure E.6
Data router with
point-to-multipoint
radio
The multiplexer may itself be a SCADA processing device that manages the local net-
work and not only combines the data, but also reduces the amount of data that must be
interchanged with the central host. The SCADA system may employ a tree network
with multiple hierarchical levels of multiplexer processors, as shown in Figure E.7.
Local Networks. Local networks connect the RTUs to the multiplexers, or

directly to the SCADA central host computer if there is no need for a WAN connec-
tion. Like the WAN, the local network may comprise cable, radio, or satellite data
communications links depending on the geographic distribution of the SCADA sys-
tem. The links may be private or rented from a telephone company.
Figure E.7
SCADA network with
multiple MUX levels
A common local network configuration is based on point-to-multipoint radio. The

radio links are generally half duplex or simplex, both of which allow transmission in
only one direction at any time. Half duplex links use different frequencies in each
direction, and simplex networks use a single frequency. In the configuration shown in
Figure E.7, the links are configured in a star topology. The local network may also be
a LAN or multidrop circuit.
Most local networks use a logical bus topology. In the bus topology, all stations share
a common transmission medium and some form of network access protocol must be
employed. Such protocols include ordered polling of each RTU, token passing, and
data packet collision detection and avoidance mechanisms.
Communications Protocols. Communications protocols define the method by
which data is transmitted along a communication link. As long as the transmitting
device follows a predefined set of rules for sending the data, the device at the receiv-
ing end is able to unravel the signal into meaningful data. For example, a protocol will
define information such as the length of time that each data packet is sent, the size of
the signal, and the required destination for the data.
An open system is one that allows for communications between different types of
devices (as supplied by different vendors, for example). Proprietary systems are by
definition closed and allow communications only between devices of the same type
(as supplied by a single vendor, for example).
Open systems avoid the disadvantages associated with using proprietary systems,
such as complete dependence on a single vendor and lack of information on how the
protocol functions. However, in order to realize the benefits of open systems, detailed
communications protocol standards are required to specify all aspects of the intercon-
nection between computers and other devices.
The Central Host Computer

The central host computer or master station is most often a single computer or a net-
work of computer servers that provide a man-machine operator interface to the
SCADA system. The computers process the information received from and sent to the
RTU sites and present it to human operators in a form that the operators can work
with. Operator terminals are connected to the central host computer by a computer
network so that the viewing screens and associated data can be displayed for the oper-
ators. Recent SCADA systems are able to offer high resolution computer graphics to
display a graphical user interface or mimic screen of the site or water supply network
in question. Figure E.8 shows the types of display screens offered by most systems.
Some examples include
• System overview pages displaying the entire water distribution system and
often summarizing SCADA sites that might be operating abnormally
• Site mimic screens for each individual RTU location showing up to the
minute site information and offering an interface to control items of equip-
ment at that site
• Alarm summary pages displaying current alarms, alarms that have been
acknowledged by an operator, and alarms that have returned to normal but
remain unacknowledged by an operator
• Trend screens enabling an operator to display the behavior of a particular

variable over time
Historically, SCADA vendors offered proprietary hardware, operating systems, and

software that was largely incompatible with other vendors’ SCADA systems. Expand-
ing the system required a further contract with the original SCADA vendor. Host
computer platforms characteristically employed UNIX-based architecture and the
host computer network was physically removed from any office computing domain.
However, with the increased use of the personal computer, computer networking has
become commonplace in the office and as a result, SCADA systems are now available
that can network with office-based personal computers. Indeed, many of today’s
SCADA systems can reside on computer servers that are identical to those servers and
computers used for traditional office applications. This has opened a range of possi-
bilities for the linking of SCADA systems to office-based applications such as GIS
systems, hydraulic modeling software, drawing management systems, work schedul-
ing systems, and information databases.
Figure E.8
The type of display
screen offered by
most SCADA systems
Operator Workstation Communications System

For water supply SCADA systems, several operators may require simultaneous access
to the SCADA central host computer to view the performance of the system. SCADA
systems are often designed to accommodate this requirement by including communi-
cations channels between the central host and the remote workstations accessed by
the operators.
Operator workstations are most often computer terminals that are networked with the
SCADA central host computer. The central host computer acts as a server for the
SCADA application and the operator terminals are clients that request and send infor-
mation to the central host computer based on the request and action of the operators.
The communications system in place between the central host computer and the oper-
ator terminals is a Local Area Network. SCADA LANs enable multiple users in a rel-
atively small geographical area to exchange files and messages, as well as access
shared resources, such as the central host computer.
Historically, SCADA LANs have been dedicated networks; however, with the
increased deployment of office LANs and Wide Area Networks (WANs) as a solution
for interoffice computer networking, there exists the possibility to integrate SCADA
LANs into everyday office computer networks.
The foremost advantage of this arrangement is that there is no need to invest in a sep-
arate computer network for SCADA operator terminals. In addition, there is an easy
path to integrating SCADA data with existing office applications, such as spread-
sheets, work management systems, data history databases, GIS systems, and water
distribution modeling systems. However, there are several disadvantages that should
be considered before integrating SCADA operator terminal LANs with office LANs:
• Corporate networks are often only supported during office hours while
SCADA LANs are most often required 24 hours per day, 7 days per week
• Communications links associated with SCADA may present a networking
security breach into the corporate computer network because some links
may bypass the office network’s usual security precautions
• During office hours, data traffic on the network associated with the corporate
network may seriously slow the networking of the SCADA operators
• SCADA network traffic generated during emergency operation procedures
may seriously slow the corporate computer network
• Linking the SCADA system with the office LAN provides ways for hackers
or terrorists to interfere with operation of the system
Software Systems
An important aspect of every SCADA system is the computer software used within
the system. The most obvious software component is the operator interface or MMI/
HMI (Man Machine Interface/Human Machine Interface) package; however, software
of some form pervades all levels of a SCADA system. Depending on the size and
nature of the SCADA application, software can be a significant cost item when devel-
oping, maintaining, and expanding a SCADA system. When software is well-defined,
designed, written, checked, and tested, a successful SCADA system will likely be
produced. Poor performances in any of these project phases will very easily cause a
SCADA project to fail.
Many SCADA systems employ commercial proprietary software upon which the
SCADA system is developed. The proprietary software often is configured for a spe-
cific hardware platform and may not interface with the software or hardware pro-
duced by competing vendors. A wide range of commercial off-the-shelf (COTS)
software products also are available, some of which may suit the required application.
COTS software usually is more flexible, and will interface with different types of
hardware and software. Generally, the focus of proprietary software is on processes
and control functionality, while COTS software emphasizes compatibility with a vari-
ety of equipment and instrumentation. It is therefore important to ensure that ade-
quate planning is undertaken to select the software systems appropriate to any new
SCADA system. Such software products are used within the following components of
a SCADA system:
• Central host computer operating system: Software used to control the

central host computer hardware. The software can be based on UNIX or
other popular operating systems.
• Operator terminal operating system: Software used to control the central

host computer hardware. The software is usually the same as the central host
computer operating system. This software, along with that for the central
host computer, usually contributes to the networking of the central host and
the operator terminals.
• Central host computer application: Software that handles the transmittal

and reception of data to and from the RTUs and the central host. The soft-
ware also provides the graphical user interface which offers site mimic
screens, alarm pages, trend pages, and control functions.
• Operator terminal application: Application that enables users to access

information available on the central host computer application. It is usually a
subset of the software used on the central host computers.
• Communications protocol drivers: Software that is usually based within

the central host and the RTUs, and is required to control the translation and
interpretation of the data between ends of the communications links in the
system. The protocol drivers prepare the data for use either at the field
devices or the central host end of the system.
• Communications network management software: Software required to

control the communications network and to allow the communications net-
works themselves to be monitored for performance and failures.
• RTU automation software: Software that allows engineering staff to con-

figure and maintain the application housed within the RTUs (or PLCs). Most
often this includes the local automation application and any data processing
tasks that are performed within the RTU.
The preceding software products provide the building blocks for the application-
specific software, which must be defined, designed, written, tested, and deployed for
each SCADA system.
E.2 DATA ACQUISITION MECHANISMS

Data acquisition within SCADA systems is accomplished first by the RTUs scanning
the field data interface devices connected to the RTU. The time to perform this task is
called the scanning interval and can be faster than two seconds. The central host com-
puter scans the RTUs (usually at a much slower rate) to access the data in a process
referred to as polling the RTUs. Some systems allow the RTU to transmit field values
and alarms to the central host without being polled by the central host. This mecha-
nism is known as unsolicited messaging. Systems that allow this mechanism usually
use it in combination with the process of polling the RTU to solicit information as to
the health of the RTU. Unsolicited messages are usually only transmitted when the
field data has deviated by a prespecified percentage, so as to minimize the use of the
communications channels, or when a suitably urgent alarm indicating some site
abnormality exists.
Section E.3 Processing of Data from the Field 699
Control actions that are performed by using the central host are generally treated as
data that are sent to the RTU. As such, any control actions by an operator logged into
the central host will initiate a communication link with the RTU to allow the control
command to be sent to the field data interface device under control. SCADA systems
usually employ several layers of checking mechanisms to ensure that the transmitted
command is received by the intended target.
E.3 PROCESSING OF DATA FROM THE FIELD

Data can be of three main types:
• Analog data (real numbers), which will be trended (placed in graphs)
• Digital data (on/off), which may have alarms attached to one state or the
other
• Pulse data (for example, counting revolutions of a meter) is analog data nor-
mally accumulated or counted. Such data are treated within the SCADA
operator terminal software displays as analog data and may be trended.
The primary interface to the operator from the operator terminal is a graphical user
interface (GUI) display that shows a representation of the plant or equipment in
graphical form. Live data are shown as graphical shapes (foreground) over a static
background. As the data changes in the field, the foreground is updated. For example,
a valve may be shown as open or closed, depending on the latest digital value from the
field. The most recent analog values are displayed on the screens as numerical values
or as some physical representation, such as the amount of filled color in a tank to rep-
resent water level. Alarms may be represented on a screen as a red flashing icon above
the relevant field device. The system may have many such displays, and the operator
can select from the relevant ones at any time.
Data from the field are processed to detect alarm conditions, and if an alarm is
present, it will be displayed on dedicated alarm lists on the application software run-
ning on the central host computer. Any abnormal conditions that are detected in the
field are registered at the central host as alarms, and operators are notified usually by
an audible alert and by visual signals on the operator terminal computers. Operators
can then investigate the cause of the alarm by using the SCADA system. Historical
records of each alarm and the name of the operator that acknowledged the alarm can
be held within a self-contained archive for later investigation or audit requirements.
Where variables in the field have been changing over time, the SCADA system usu-
ally offers a trending system whereby the behavior of a particular variable can be plot-
ted on a graphical user interface screen.
E.4 LEVELS OF CONTROL

SCADA communications systems may be deployed over a wide geographical area
and hence it can be expected that communications link availability and speed will be
lower than that typical for a link between a computer and a PLC via a hard-wired
Ethernet Local Area Network (LAN).
Computerized control systems that are commonly used at water or wastewater treat-
ment plants are an example of where high-speed Ethernet LANs are employed as the
communications backbone. The control systems are similar to SCADA systems but
are more closely related to those systems developed for manufacturing or factory
floor applications. These are often referred to as Distributed Control Systems (DCS).
They have similar functions to SCADA systems, but the field data interface devices
are usually located within a confined geographical area. The communications LAN
will normally exhibit up to 99.98 percent availability, with substantially greater band-
width than employed in a SCADA system. A DCS system also will usually employ a
significant amount of remote loop control, where the required value of a field variable
is calculated based on the feedback received from a measured variable in the field. In
the case of DCS systems, this calculation is often performed within the central host
computer. In comparison, loop control required to operate a remote pump station, for
example, will regularly be housed within a local loop control device which calculates
the required value of the field variable locally and therefore separately from the cen-
tral host computer.
For example, consider the example of a valve position based on the level of water in a
reservoir. Figure E.9 illustrates a common control problem.
Figure E.9
Valve position control
Consider in this case that the position of the valve depends on the level of water in the
reservoir. An operator may control the process by issuing a Set Point, which is the
desired level at which the tank should stay. Once the level deviates from the level set
point, the controller detects the deviation and sends a signal to the valve position actu-
ator to move the valve to reduce the error. The level of the tank is continuously moni-
tored to enable the controller to trim the valve position.
Within plant floor DCS systems, it is common for this controller to reside within the
central host computer. The communication system connecting the local RTU and the
central host is fast and reliable and it is convenient to house the bulk of the computing
power within a centralized location.
Section E.5 Handling of Data During SCADA Failures 701
By contrast, SCADA systems generally cover large geographic areas and rely on a
variety of communications systems that are normally less reliable than a LAN associ-
ated with a DCS. Loop control based in the central host computer is therefore less
desirable. Instead, the controller application is housed in the RTU. The SCADA oper-
ator is able to alter the tank level set point remotely and perhaps may be allowed to
manually drive the valve open and closed when the control loop is disabled. However,
the automatic control of the valve is most often resident in the RTU. If communica-
tion to the remote site is lost, it is desirable that the local automatic control system
continue to operate; therefore, the RTU is an autonomous unit which could control the
valve without constant direction from the central host computer.
Of course, there is always the temptation to allow a great percentage of the automa-
tion functions to be centralized within a SCADA system. This approach has many
advantages, most notably:
• Computing power can be centralized in an office environment, reducing the
cost of field devices which must be designed to operate in sometimes harsh
conditions
• Engineering staff are much more readily able to continuously improve and
update control programs, ensuring that there is a standardization of control
algorithms across the SCADA network
• Expensive redundancy system failure proofing can be located in a central
location
The important thing to consider is the reliability of the communication link between
the SCADA central host and the site. Where a critical control algorithm is required
and the controller must be located remotely, the communication link must be
designed to contribute effectively to the reliability of the entire system. The cost asso-
ciated with this requirement may be prohibitive enough to warrant placing the auto-
matic control function at the site.
SCADA allows for many options by which assets and devices at remote sites may be
controlled. It is not uncommon for modern SCADA systems to employ a combination
of the previously mentioned mechanisms.
E.5 HANDLING OF DATA DURING SCADA FAILURES

Different SCADA systems cope differently with a failure event. Some systems rely
primarily on the inherent redundancy of the SCADA system, and others may use
some form of storage mechanism to archive data that may be recovered once the
SCADA system has returned to normal operating capacity. These options are summa-
rized as follows:
• Storage of data in the RTUs: Some SCADA systems rely on the capacity
of the RTU to store data collected from the field under normal operation and
then periodically transmit that data as an unsolicited message or when polled
by the central host. In times of SCADA system failure, the capacity of the
RTU is used to archive information until a backup central host is brought
online or the original system has recovered.
• System redundancy: Most SCADA systems incorporate some form of

redundancy in their design, such as dual communications channels, dual
RTUs, or dual central host computers. Such systems may be designed for
such redundant equipment to be online (hot standby) to ensure a seamless
transfer upon SCADA system failure, or offline (cold standby) where the
backup mechanism must be manually brought online to operational capacity.
Most SCADA systems employ a combination of the preceding mechanisms to ensure
data continuity during failure events.
E.6 ERRORS AND ACCURACY ISSUES

As previously discussed, SCADA systems for water distribution systems generally
use low bandwidth communication channels. Data from the field may therefore need
to be compressed by the field devices before being transmitted to the central host to
avoid overtaxing the capacity of the communications media to transmit information.
The result may be that data from the field include some form of error that must be
considered before analysis.
Common sources of error include
• Compression algorithms employed by the RTU prior to transmitting time
series data to the central host
• Compression algorithms employed by any data archive software that may be
used to store old data from a SCADA system
• Interpolation by the trending system of a SCADA system.
Such errors are not immediately obvious from the data trends received from a
SCADA system and may appear as unexplained unusual behavior of an analog vari-
able that may lead a data user to mistakenly suspect an error with a measuring instru-
ment located in the field. Through understanding of the particular mechanisms
inherent in the data collection system, it is possible to explain the unexpected behav-
ior of a variable and compensate for the error within the data finally used for model
validation purposes. For more information, see Chapter 6.
REFERENCES
Barnes, M., and Mackay, S. (1992). Data Communications for Instrumentation and Control. Instrument
Data Communications (IDC), Australia.
Cisco Systems, Inc. (2000). CCIE Fundamentals: Network Design and Case Studies. Cisco Press, Second
Edition, Indianapolis, Indiana.
Citect version 5 User’s Guide (1998). CI Technologies Pty. Ltd., Pymble, Australia.
Haime, A. L. (1998). “Practical Guide to SCADA Communications.” SCADA at the Crossroads Confer-
ence Workshop, The Institution of Engineers Australia, Perth, Western Australia.
Williams, R. I. (1992). Handbook of SCADA Systems for the Oil and Gas Industries. Elsevier Advanced
Technology Limited, 1st Edition, Great Yarmouth, United Kingdom.

Water Distribution Modelling PDF

Uploaded by

Copyright:

Available Formats

Water Distribution Modelling PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Water Distribution Modelling PDF

Uploaded by

Copyright:

Available Formats

What are the main components of a water distribution system?

What are the main components of a water distribution system?

What are the main purposes of a clearwell in water treatment?

What are the main purposes of a clearwell in water treatment?

ADVANCED WATER

Peer Review Board

Click here to visit the Bentley Institute

Water distribution modeling is the latest technology in a process of advancement that

1.1 ANATOMY OF A WATER DISTRIBUTION SYSTEM

Sources of Potable Water

Customers of Potable Water

1.2 WHAT IS A WATER DISTRIBUTION SYSTEM

1.3 APPLICATIONS OF WATER DISTRIBUTION MODELS

Long-Range Master Planning

Fire Protection Studies

Water Quality Investigations

Emergency Response. Emergencies are a very real part of operating a water

1.4 THE MODELING PROCESS

1.5 A BRIEF HISTORY OF WATER DISTRIBUTION

Courtesy of the Ductile Iron Pipe Research Association

1738 — Bernoulli publishes Hydrodynamica. The Swiss Bernoulli family extends

1843 — St. Venant develops equations of motion. Several researchers, including

1845 — Darcy-Weisbach head loss equation developed. Julius Weisbach publishes

1883 — Laminar/turbulent flow distinction explained. While earlier engineers

1906 — Hazen-Williams equation developed. A. Hazen and G.S. Williams develop

1920s — Cement-mortar lining of water mains. Cement mortar lining of water

1986 — Introduction of Dynamic Water Quality Models. At the AWWA Distribu-

1990s — Privatization of water utilities. The privatization of water utilities

1991 — Water Quality Modeling in Distribution Systems Conference. The

1993 — Introduction of water quality modeling tool. Water quality modeling

2001 — Automated calibration. Automated calibration of distribution models

Peer Review Board

Click here to visit the Bentley Institute

Model-based simulation is a method for mathematically approximating the behavior

2.1 FLUID PROPERTIES

Density and Specific Weight

where J = fluid specific weight (M/L2/T2)

where W = shear stress (M/L/T2)

where Q = kinematic viscosity (L2/T)

where Ev = bulk modulus of elasticity (M/L/T2)

Although the assumption of incompressibility is justifiable under most conditions,

2.2 FLUID STATICS AND DYNAMICS

However, because of the magnitude of pressures occurring in distribution systems,

where P = pressure (M/L/T2)

Absolute Pressure and Gage Pressure. Pressure at a given point is due to

P abs = P gage + P atm (2.7)

where Pabs = absolute pressure (M/L/T2)

Pgage = 8.7 psi

Velocity and Flow Regime

where V = average fluid velocity (L/T)

where D = diameter (L)

where V = average fluid velocity (ft/s)

Based on experimental evidence gathered by Reynolds and dimensional analysis, a

where Re = Reynolds number

Table 2.1 Reynolds number for various flow regimes

Uniform Velocity Profile Laminar Profile

2.3 ENERGY CONCEPTS

where H = total head (L)

• Turbulence due to changes in streamlines through fittings and appurtenances