Livro - Linear Theory of Hydrologic Systems (Dooge, 1988)

N
321
/
4y
"?-y7
LINEAR THEORY
OF HYDROLOGIC SYSTEMS
Technical Bulletin No. 1468

o
r-i
Agricultural Research Service

UNITED STATES DEPARTMENT OF AGRICULTURE
413771
LINEAR THEORY
OF HYDROLOGIC SYSTEMS
Technical Bulletin No. 1468
Agricultural Research Service

UNITED STATES DEPARTMENT OF AGRICULTURE
Washington, D.C.
Issued October 1973
For sale by the Superintendent of Documents, U.S. Government Printing Office

Washington, D.C. 20402Price $2.80
Stock Number 0100-02747
PREFACE
This publication is a shortened version of lectures given by Professor J. C. I.
Dooge, Department of Civil Engineering, University College, Dublin, Ireland,
in August 1967 at the Department of Agricultural Engineering, University of
Maryland, under the sponsorship of the Agricultural Research Service, U.S.
Department of Agriculture. Professor Dooge is a world authority on hydrologic systems, which are basic to computations for successfully planning the
best use of soil and water resources in agricultural watersheds.
The original course consisted of 18 lectures supplemented by problem
sessions and seminars; however, this pubUcation is confined to the first 10
lectures, which dealt with the general principles of the linear theory of deterministic hydrologie systems. Some important material, originally dealt
with in later lectures, has been included in summary form in lectures 7, 8, 9,
and 10 of this publication.
Trade names are used in this publication solely for

the purpose of providing specific information. Mention of a trade name does not constitute a guarantee or warranty of the product by the United States
Department of Agriculture or an endorsement by
the Department over other products not mentioned.
IV
ACKNOWLEDGMENTS
The ideas outlined in these lectures reflected the contribution of a verylarge number of research workers to the development of parametric hydrology.
Where possible my indebtedness has been clearly acknowledged in the cited
references. Apologies are offered for any failure to acknowledge such indebtedness. Apart from pubHshed papers and formal discussions, I have
benefited greatly from many stimulating informal discussions with colleagues.
In this connection I must particularly acknowledge the influence on my
thinking and my work of Eamonn Nash, Terence O'Donnell, and Dirk
Kraijenhoff. I mention these three among my colleagues because I have continually turned to them* over the years for stimulation or for comments on
work in progress.
The material of the lectures which follows was a development in more
comprehensive form of material which was the subject of seminars given at
Imperial College London, the University of New South Wales, and The
University of Wageningen. The initiative for giving the lectures came from
Heggie Holtan who was a continual source of encouragement. In the detailed
planning of the lectures, I received invaluable assistance from Don Brakensiek.
CONTENTS
Introduction
1
LECTURE 1: HYDROLOGIC SYSTEMS
3
The Systems Approach
3
Linear Time-Invariant Systems
17
Identification and Simulation.
29
Problems on Hydrologie Systems
39
Literature Cited
40
LECTURE 2: REVIEW OF PHYSICAL HYDROLOGY
43
Precipitation
44
Evaporation and Transpiration
45
Infiltration and Percolation
47
Ground Water Flow
48
Overland Flow and Channel Flow
50
Problems on Physical Hydrology
53
Literature Cited
54
LECTURE 3: REVIEW OF MATHEMATICS
58
Orthogonal Polynomials and Functions
58
Fourier and Laplace Transforms
62
Differential Equations
65
Matrices
67
Numerical Methods
70
Problems on Mathematics
71
Literature Cited
73
LECTURE 4: CLASSICAL METHODS OF RUNOFF PREDICTION 75
The Outflow Hydrograph
75
The Rational Method
79
Unit Hydrograph Concepts
84
Separation of Base Flow
89
Analysis of Complex Storms
93
Problems on Classical Methods
97
Literature Cited
98
LECTURE 5: IDENTIFICATION BASED ON CONTINUOUS
DATA
102
Transform Methods of Identification
102
Analysis by Fourier Series
105
Analysis by Fourier and Laplace Transforms
108
Moments and Cumulants
113
Laguerre Analysis of Systems
117
Time-Series Analysis
121
Problems on Identification Based on Continuous Data
124
Literature Cited
125
vi
LECTURE 6: IDENTIFICATION BASED ON DISCRETE DATA 127

Basic System Equations
127
Matrix Methods of Identification
130
Discrete Transform Methods
133
Time-Series Analysis of Discrete Data
138
Computational Methods
141
Problems on Discrete Systems Identification
145
Literature Cited
146
LECTURE 7: SIMULATION OF HYDROLOGIC SYSTEMS
148
Basic Ideas in Simulation
148
Regression Models
151
Digital Simulation
160
Optimization
169
Analogs and Physical Models
175
Problems on Simulation
184
Literature Cited
185
LECTURE 8: SYNTHETIC UNIT HYDROGRAPHS
190
Types of Synthetic Unit Hydrographs

190
Time-Area Methods
193
Empirical Expressions for Unit Hydrograph Parameters
_
200
Empirical Shapes for the Unit Hydrograph
206
Conceptual Models of the Unit Hydrograph
209
Comparison of Methods
216
Problems on Synthetic Unit Hydrographs
224
Literature Cited
227
LECTURE 9: MATHEMATICAL SIMULATION OF SURFACE
FLOW
232
Overland Flow
232
Unsteady Flow in Open Channels
243
Problems on Surface Flow
261
Literature Cited
262
LECTURE 10: CONCEPTUAL MODELS OF SUBSURFACE FLOW 267
Movement of Soil Moisture
267
Unsteady Movement of Soil Moisture
272
Comparison of Infiltration Formulas
278
Basic Equations of Ground Water Flow
283
Problems on Subsurface Flow
291
Literature Cited
293
List of Symbols
296
APPENDIX TABLES
300
AUTHOR INDEX
316
SUBJECT INDEX
323
vu
^'LINEAR THEORY
OF^YDROLOGIC SYSTEMS,,
By James C. I. Dooge^^
INTRODUCTION
These lectures were designed to introduce participants to the theory of
deterministic hydrologie systems. In recent years, this theory has been named
''parametric hydrology," but is also known as "dynamic hydrology" or
"deterministic hydrology." One object of the course was to make the participants aware of certain theories and techniques rather than to give them a
perfect knowledge of the theory or a complete mastery of the techniques.
Attention was directed to the essential unity underlying the many methods
that have appeared in the hydrologie literature as seemingly unrelated to
one another. Another aim of the course was to reformulate established concepts and techniques in terms of a general systems approach and thus to
extend their usefulness.
This publication follows the organization of the original course and is divided into lectures. Lecture 1, which is far longer than any other, consists of
a preview of the subject matter of the whole course. This is followed by two
review lectures, one on physical hydrology and the other on the mathematics
required for the study of deterministic hydrologie systems. Lectures 4, 5, and
6 deal essentially with the problem of the identification of deterministic
hydrologie systems and, thus, with the analysis of the behavior of a given
system. The next four lectures7 through 10deal with synthesis rather
than analysis. In them, the question of simulating the behavior of natural
hydrologie systems is discussed.
The original lectures were built around more than 100 figures, which were
included with the handout material for the course, and also projected during
the lectures. In this publication, many of these figures have been incorporated
into the text. The handout material also contained a number of problems
and a large number of references for each lecture. These were not confined
to what would have been directly necessary for a short, 2-week course. Rather
1 Formerly, Professor and Head, Department of Civil Engineering, University College,
Cork, Ireland (1958-70); since then. Professor and Head, Departriient of Civil Engineering,
University College, Dublin.
1
2 TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

they were chosen so that the participants could, after the completion of the
course, go more deeply into any part of the subject which was of particular
interest to them. These problems and references are included in this publication and appear at the end of each lecture. So as to facilitate further study
of individual aspects of the subject, some important references have been
repeated in the various lectures rather than cross-referenced from one lecture
to another.
LECTURE 1:
HYDROLOGIC SYSTEMS
The Systems Approach
What is a system?
Before starting to discuss hydrologie systems, it is well to be clear about
what we mean in this context by a system. There are, of course, almost as
many definitions of a system as there are books on the subject of systems
analysis and systems synthesis. It is worthwhile to review a few of these
definitions before arriving at a working definition which will serve our purpose.
The firgt definition by Stafford Beer {6y, an expert on management and
cybernetics, merely defines a system as ''Anything that consists of parts
connected together.'' This includes the essence of what a system is. It is something that consists of parts; there are separate parts in it, and they are connected together in some way. Of course, this does not bring us very far because
philosophers will tell us that everything which is created, everything which
changes, consists of parts. While it is true to say that everything is a system,
this does not help us very much to build up a consistent theory of hydrologie
systems.
A second definition is that given by MacFarlane (30) in his book on
"Engineering Systems Analysis'' in which he defines a system as ''An ordered
arrangement of physical or abstract objects." Here, the notion of some sort
of order enters the picture; the system is put together in accordance with
some sort of plan. Also we have the idea that there are two types of systemsa
physical or real system and an abstract one.
A third definition by Ackoff {2), who was a pioneer in operations research,
states that a system is, "Any entity, conceptual or physical, which consists
of interdependent parts." Again we get the idea that the system can be conceptual or physical and that the system consists of interdependent parts.
The fourth definition, by Drenick (19), stresses the manner of operation of
a system rather than its structure: "A device which accepts one or more inputs
and generates from them one or more outputs." This concept of a system, as
that which links inputs and outputs, is common in the literature. Further
definitions and descriptions of the systems approach in other disciplines are
found in works by Bellman (7), Doebelin (Jf4), Draper and others (18),
Ellis and Ludwig (î), Koenig and Blackwell (^4), Lee {27), Lynch and
Truxal (29), Paynter (37), Stark U3) and Tustin (U)Having considered a large number of definitions of a system, I decided to
2 Italic numbers in parentheses refer to Literature Cited at the end of each lecture.
3
accept as adequate for the present purpose, the definition that, "A system is
any structure, device, scheme, or procedure, real or abstract, that interrelates
in a given time reference, an input, cause, or stimulus, of matter, energy, or
information, and an output, effect, or response, of information, energy, or
matter.'' This definition includes the concepts contained in the definitions
given above. The emphasis is on the function of the systemthat it interrelates, in some time reference, an input and an output. In mechanics, we
tend to talk about inputs and outputs; physicists and philosophers often speak
of causes and effects; workers in the biological sciences talk of stimuli and
responses. These are merely alternative words for the same two concepts.
Reference to an input does not restrict the concept to a single input. The
input could consist of a whole group of inputs so that we would have an input
vector rather than a input variable. In some cases, the input could be completely distributed in space and thus represented by a function of both space
and time.
The definition refers to inputs (and outputs) as consisting of matter, energy,
or information. In some systems, both the input and output would consist of
material of some sort; in others, attention would be concentrated on the
input and output of energy; while in other systems, the concern would be with
the input and output of information. There is no need, however, for the input
and the output to be alike. It is perfectly possible to have a system in which an
input of matter will produce an output of information or vice versa. That
there is no necessity for the natures of the input and output to be the same
has been emphasized in the definition by using the reverse order to describe
the natures of the input and the output. The essence of a systemwhich can
be real or abstractîs that it interrelates two things.
Concept of system operation
In dealing with problems in applied science, our concern is to predict the
output from the system we are interested in. Figure 1-1 shows the three
elements that together determine what this output will be. In the classical
approach, certain assumptions are made about the nature of the system and
the physical laws governing its behavior; these are then combined with the
input to predict the output. To apply this classical procedure, it is necessary
to know the physical laws or to be able to make reasonable assumptions about
them. It is also necessary to be able to describe the structure of the system
and to specify the input. A distinction is made here between the nature of the
system itself and the physical laws of its operation. The nature of the system
refers only to its inherent structure, that is, to the nature of the components
of the system and the way in which these components are connected.
In hydrology, as in many other areas, the classical approach tends to
breakdown either because, on the one hand, the physical laws are impossible
to determine or too complex to apply, or, on the other hand, the geometry of
LINEAR THEORY OF HYDROLOGIC SYSTEMS
DESCRIPTION
OF SYSTEM
FIGURE
OUTPUT
1-1.Factors affecting output.
the system is too complex or the lack of homogeneity too great to enable us to
apply classical methods to the prediction of the behavior of the system. In
the systems approach, an attempt is made to evade the problems raised by the
complexity of the physics, the complexity of the structure of the system, and
the complexity of the input.
Figure 1-2 shows the essential nature of the systems approach to the
problem. In figure 1-2, the elements of figure 1-1 are rearranged, and the
concept of system operations is introduced. In the systems approach, the
complexities arising from the physical laws involved and from the structure
of the system being studied are combined into the single concept of the system
operation of this particular system. If either the nature of the system or the
physical laws are changed, then the systems operation will be changed. These
effects are showTi in the vertical relationships in figure 1-2. In dealing with
one particular system, however, we can use this combined concept of system
operation as being the element which accepts the input and converts it into
an output.
Thus, in the systems approach, attention is concentrated on the horizontal
relationship in figure 1-2. In systems analysis, we are concerned only with the
way in which the system converts input to output. If we can describe this
TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE
LAWS
SYSTEM
OPCRATIONi
INPUT
OUTPUT
NATURE
OF
SYSTEM
FIGURE
1-2.The concept of system operation.
system operation, we are not concerned in any way with the nature of the
systemwith the components of that system, their connection with one
another, or with the phj^sical laws which are involved. The systems approach
is an overall one and does not concern itself with details which may or may
not be important and which, in any case, may not be known.
The concern of the systems approach with overall behavior rather than
details can be exempUfied by the unit hydrograph approach to predicting
storm runoff. In this approach, precipitation excess is taken as the input and
the direct storm runoff as the output. The operation of the whole watershed
system in converting precipitation excess to direct storm runoff is summarized
in the form of the unit hydrograph. We are not concerned with arguments
about whether there is, or is not, such a thing as interflow, nor with arguments
as to whether overland flow actually occurs; and if it does, what the friction
factor is. We may overlook our ignorance of the physical laws actually determining the processes in various parts of the hydrologie cycle. We may ignore
the problem of trying to describe the complex watershed with which we are
dealing; we do not have to survey the whole watershed by taking cross sections
on every stream as we would have to do if we wanted to solve the problems
by classical hydraulics. Instead we assume that all the complex geometry in
the watershed and all the complex physics in the hydrologie cycle is described
for that particular watershed (but of course for that one only) by the unit
hydrograph. The systems approach is basically a generalization of this standard
technique that has been used in applied hydrology for many years. The
essential feature is that in dealing with the analysis of a particular system,
attention is concentrated on the three horizontal elements in figure 1-2.
This does not mean that the structure of the system or the physical laws
can be completely ignored. If our problem is one of synthesis or simulation,
rather than analysis, it is necessary to consider the vertical elements in figure
1-2. Again we have an analogy with the unit hydrograph technique in applied
hydrology. If we have no records of input and output (that is, of precipitation
excess and of storm runoff) for a watershed, it is necessary to use synthetic
unit hydrograph procedures. This is done in applied hydrology by correlating
the parameters of the unit hydrograph with the catchment characteristics.
In this way, the effect of the structure on the system operation is taken into
account. Because the physics does not change from watershed to watershed,
it might be thought that no assumptions are made about the physics of the
problem in synthetic unit hydrograph procedures. This is not so. The whole
unit hydrograph process of superimposing unit hydrographs and blocks of
precipitation excess depends on the superposition principle, which will only
apply, as we shall see later, if the system we are dealing with is linear. Therefore, unit hydrograph procedures make the fundamental assumption that
the physical laws governing direct surface runoff can be represented as
operating in some linear fashion.
In the above example, the details of the operation of a system were ignored
because they were too complex to be understood. In other cases, the details
are ignored because they are not important. Again we can take an example
from classical hydrology. The problem of routing a flow through an open
channel can be solved by writing down the equation of continuity and the
dynamic equation and proceeding to solve the problem for the given data by
the methods of open channel hydraulics. Even with large, high-speed computers, the solution for the case of a nonuniform channel is extremely difficult.
The solution proceeds step-by-step down the reach and marches out stepby-step in time. In practice, the detailed results for the discharge and depth at
every point along the channel are not required since all we usually wish to
know is the hydrograph at the downstream end. Whether we use the method
of characteristics, an explicit finite difference scheme, or an implicit finite
difference scheme, difliiculties of one sort or another arise in the numerical
solution of this problem. Most of the information which we have gained with
such labor is of little interest to us as applied hydrologists. More than 30 years
ago, hydrologists dodged these difficulties by introducing the idea of hydrologie routing, that is, the idea of treating the whole reach as a unit, trying to
link up the relationship between the upstream discharge and the downstream
discharge without bothering with what went on in between.
Systems terminology
As in every other discipline, a terminology has grown up in systems analysis
and systems engineering. The meaning of the more important concepts and
terms must be clear before we can understand what is written in the hterature
concerning the systems approach.
A complex system may be divided into subsystems, each of which can be
identified as having a distinct input-output linkage. A system or a subsystem
may also be divided into components, each of which is an input-output element, which is not further subdivided for the purpose of the study in hand.
Thus, a system is composed of subsystems, and the subsystems themselves
consist of components.
Reference is frequently made to the state of a system. This is a very general
concept. Any change in any variables of the system produces a change of
state. If all of the state variables are completely known, then the state of the
system is known. Perhaps it is easiest to look at this in hydrologie terms. If
we knew exactly where all the water in a watershed washow much of it was
on the surface, how much of it in each soil horizon, and how much of it in each
channelwe would know the hydrologie state of the watershed. The state
of a system may be determined in various ways. In some systems, it is determined historically, that is, the previous history of the system determines its
present condition. In other cases, the state of the system is determined by
some external factor which has not been included in the system under examination. In still other cases, the state of the system is stochastically determined
or else assumed to be stochastically determined, that is, determined by a
random factor.
A system is said to have a zero memory, a finite memory, or an infinite
memory. The memory is the length of time in the past over which the input
affects the present state. If a system has a zero memory, then its state and its
output depend only on the present input. If it has an infinite memory, the
state and the output will depend on the whole past history of the system. In
a system with a finite memory, its behavior, its state, and its output depend
only on the history of the system for a previous length of time equal to the
memory.
The distinction between Hnear and nonlinear is of vital importance in
systems theory as it is in classical mechanics. The analysis and synthesis of
hnear systems can draw on the immense storehouse of linear mathematics
for techniques. The special properties of linear systems will be dealt with in
detail later. For the moment, it will sufice to say that a linear system is one
that has the property of superposition and a nonUnear system is one that
does not have this property.
Another important distinction is between time-variant and time-invariant
systems. A time-invariant system is one whose input-output relationship does
not depend on the time at which the input is applied. Most hydrologie systems
are actually time-variant; there are seasonal variations throughout the year
and a variation of solar activity throughout the day. Nevertheless, the advantages of assuming the systems to be time-invariant is such that these real
variations are usually neglected in practice.
It is necessary to distinguish between continuous and discrete systems, and
also among continuous, discrete, and quantized systems. Whereas hydrologie
systems are continuous, the inputs and outputs may be available in either
continuous, discrete, or quantized form. A system is said to be continuous
when the operation of the system takes place continuously. A system is said
to be discrete when it changes its state at discrete intervals of time. An input
or an output of a system is said to be continuous when the values of it are
either known continuously or can be sampled so frequently as to provide a
virtually continuous record. An input or an output is said to be discrete if the
value is only known or can only be sampled at finite time intervals. An input
or an output is said to be quantized when the value only changes at certain
discrete intervals of time and holds a constant value between these intervals.
Many records of rainfall, which are only known in terms of the volume during
certain intervals of time, are in effect quantized records.
We can talk of the input and output variables and the parameters of the
system as being either lumped or distributed. A lumped variable or parameter
is one whose variation in space is either nonexistent or has been ignored. Thus,
the average rainfall over a watershed, which is used as the input in many
hydrologie studies, is a lumped input. Where the variation in one or more
space dimensions is taken into account, the parameter is a distributed one.
Either the parameters of a system itself or the inputs or outputs can be lumped.
The behavior of lumped systems is governed by ordinary differential equations
with time as the independent variable. The behavior of distributed systems is
governed by partial differential equations.
A distinction is also made between deterministic and probabilistic systems.
In a deterministic system, the same input will always produce the same
output. The input to a deterministic system may be either itself deterministic
or stochastic. A probabilistic system is one which contains one or more elements in which the relationship between input and output is statistical rather
than deterministic. The present lectures are mainly concerned with deterministic systems.
The distinction is sometimes made between natural systems and devised
systems. The essential feature of natural systems is that though the inputs
and outputs and other state variables are measurable, they are not controllable.
In a devised system, for example, an electronic system, the input may be both
controllable and measurable.
Other descriptions of systems are that they are either simple or complex.
Complex in this context usually means systems with feedback built into them.
Some systems have negative feedbacks built into them to produce stability
and others are designed for ultrastability, that is, to be stable even against
10
unanticipated changes in the external environment. Beyond feedback we have

adaptive systems which learn from their past history and improve their
performance.
A causal system is one in which an output cannot occur earlier than the
corresponding input. In other words, the effects cannot precede the cause.
In electrical engineering, the limitation to causal systems is sometimes
abandoned to achieve certain results. All of the systems dealt with in hydrology are causal systems. Simulation systems are also referred to as being
realizable. This has much the same meaning as causal insofar as it means that
the system is nonanticipative in its operation.
A further important property of systems is their stability. A stable system
is one in which if the input is bounded, then the output is similarly bounded.
In hydrology, virtually all our systems are stable and extremely stable. In
most cases, when the input to a hydrologie system is bounded, the bound on
the output is considerably less than that on the input.
Basic problems involving systems
We have already seen that a system is essentially something which interrelates an input and an output. Thus, from an overall viewpoint there are
three elements to be consideredthe input, the system operation, and the
output. This general relationship can be represented either by a rectangular
box, in which the system H converts the input x{t) into the output y(t).
Alternatively, it may be represented by the general mathematical relationship :
y{t)=^h{t)Mt)
(1)
where h{t) is a mathematical function characterizing the system operation

and xl/issi symbol denoting that the function h(t) and the input function x{t)
are combined in some way to produce the output function y{t). If the operation of the system can be described in any way, then we are concerned with
the interrelation of three functionsthe input function, the system operation
function, and the output function.
If we have derived a mathematical representation of the operation of the
system and we know the input, then the problem of finding the output is a
problem of prediction. In terms of the unit hj^drograph approach, the problem
is to determine the storm runoff knowing the unit hydrograph and the given
or assumed effective rainfall.
If, however, we do not know the unit hydrograph, it is necessary to derive
it from the past records. This is the problem of finding a function describing
the system operation knowing the input and output; it may be described as
the problem of system identification. The problem of system identification is
much more difficult than the problem of output prediction. It is important to
realize what we mean by system identification. We cannot identify the system
uniquely in the same way as we might identify someone from their fingerprints.
11
Rather can we identify the behavior of the system much the same way as a
criminal might be identified by his modus operandi. All that system identification tells us is the overall nature of the systems operation and not any details
of the nature of the system itself.
The various problems that can arise are shown on figure 1-3. If we have
a given system, then the problem is one of analysis, as in the case of the
structural engineer who is faced with the analysis of a given design. There are
three elements in the system relationship; hence there are three types of
problems in analysis with which we must concern ourselves. In each of these
situations, the problem is to find one of the elements when given the other two.
The third problem of analysis is detection. This occurs when, knowing how
our system operates and knowing the output, we wish to know what is the
input. This is the problem of signal detection and the problem inherent in all
instrumentation. In hydrology, as in many other fields of engineering, this
particular problem has been widely ignored. The engineer has been too content
to assume that his instruments are perfect, that is to assume that the input
to an instrument is correctly given by the output recorded by the instrument.
It is only in recent years that there has been any study of hydrologie instruments from a systems viewpoint. The problem of signal detection, or signal
identification, is mathematically the same as the problem of system identification and, therefore, also substantially more difficult than the problem of
output prediction.
The problem of prediction is that of working out the interrelationship of the
two functions h{t) and x{t) shown on the right-hand side in equation 1. The
PROBLEMS ARISING WITH SYSTEMS
TYPE OF P^&SM
INPUT
mW^ uawr
?
Prediction
Analysis Identification
Identification
Synthesis (Simulation)
FIGURE
y
y
??
1-3.Classification of systems problems.
12
problem of system identification or signal detection is that of unscrambling

one of the components on the right-hand side of the equation. This involves a
problem of inversion, which is inherently difficult.
Besides the problems of analysis, we have also the problems of synthesis.
This corresponds to the problem of the structural engineer who has to design
a structure as well as know how to analyze it. In hydrology, we do not design
watersheds, except possibly in urban hydrology, but even here we do not design
them from a hydrologie viewpoint. We do, however, attempt to simulate
complex hydrologie systems by simpler models, and this is essentially a
problem of synthesis. The problem of synthesis is to devise a system which
will convert a known input to a known output within certain limits of accuracy. It involves the selection of a model and the testing of the operation of
this model by analysis. This is even more diflicult than the problem of identification, and hence the double question mark in figure 1-3.
A scientific approach to the analysis and synthesis of systems must rest on
a firm mathematical foundation. In the following lectures, the mathematical
techniques used at present in parametric hydrology are introduced and their
application described. Those interested in studying more deeply the mathematics of system behavior can do so in books by Aseltine (5), Zadeh and
deSoer (46), DeRusso and others (5), Gupta (^3), and Wymore (45).
Hydrologie systems
Although we have already referred to certain isolated problems in hydrology, it is well to consider the hydrologie cycle as a whole before considering
the various hydrologie subsystems. Figure 1-4 shows a diagram of the hydrologie cycle by Ackerman and others (1),
Similar diagrams can be found in any standard textbook. These diagrams
can be compared for such qualities as artistic merit and draftsmanship, but
what do they mean from a systems point of view? Those who use the systems
approach are known to have an aversion to such diagrams and to insist on
drawing everything in terms of neat rectangles. These austere rectangular
boxes do not even have the color of modern abstract art to save them from
criticism. From their appearance one would deduce that they show much less
information than the figures such as that shown in figure 1-4. Actually, this
is not so. Figure 1-5 is a systems representation or block diagram of the
hydrologie cycle and is based on figure 1-4. Actually there are less assumptions in the block diagram of figure 1-5 than in the representation in figure 1-4.
The whole hydrologie cycle is a closed system in the sense that the water
circulating in the system always remains within the system. The whole system
is driven by the excess of incoming radiation over outgoing radiation, and the
movement of wâter through the hydrologie cycle is only possible because of
this source of energy. In figure 1-5, the system represented by the hydrologie
cycle has been divided into subsystems. Thus we have the atmospheric sub-
13
system, the subsystem represented by the surface of the ground, the subsurface subsystem or unsaturated phase, the ground water subsystem or
saturated phase, the channel network subsystem, and the ocean subsystem.
Each of these subsystems will contain individual components, but for the
purpose of an overall analysis and overall discussion, these components have
all been lumped into one subsystem. The hydrologie cycle shown in figure 1-5
is a system in which the inputs and outputs are material. Water in one of its
vQeep
percolation
FIG.
1-4.Representation of the hydrologie cycle.
14
^_J
Rb
i_ L
Wi
ATMOSPHERE
r* ATMOSPHERE
J"l.
Wr
SURFACE
I
^ I
CHANNEL R.O.
NETWORK
SOIL
OCEAN
T.F.
XT
I
_i
GROUND WATER
Qg
?l
I
LITHOSPHERE
FIGURE
.-.-l = ^=L.l.
LITHOSPHERE
1-5.Block diagram of the hydrologie cycle.
phases either moves through the cycle or is stored in some part of the cycle at
all times. Figure 1-6 shows a representation of the hydrologie cycle developed
by Kulandaiswamy.* The latter figure is similar to the systems representation
used by electrical engineers.
Neither classical hydrology nor systems hydrology deals with the hydrologie
cycle as a whole. Hydrology leaves the atmosphere to the meteorologists, the
lithosphre to the geologists, and the seas to the oceanographer. The resulting
subsystem is shown in figure 1-7. In outlining this subsystem we have cut
across certain lines of water transport and, consequently, the system is no
longer a closed one. These lines of water transportprecipitation, evaporation, transpiration, and runoffare now either inputs or outputs to our new
3 KULANDAISWAMY, V. C. A BASIC STUDY OF THE RAINFALL EXCESS-SURFACE RUNOFF
RELATIONSHIP IN A BASIN SYSTEM. Ph.D. thesis, Univ. of Illinois. 1964 [Available as Publication No. 64-12535.] from University Microfilms, Inc. P.O. Box 1346, Ann Arbor, Mich.
48106
15
system. Whereas precipitation is clearly an input and runoff an output, it is

not always easy to decide whether evaporation and transpiration are inputs
or outputs. One reasonable standpoint is to consider potential evaporation as
an input and actual evaporation as an output.
The system shown in figure 1-7 is clearly a lumped system. But this does
not involve any more assumptions than are made by classical hydrologists
when they consider the individual basin, whether it be a parking lot, an
experimental plot, or a natural watershed. These are all basinsthey are all
systems, which convert a certain hydrologie input into a hydrologie output.
It is possible to divide up the system and subsystems shown in figure 1-7
into components. Thus, we could divide the soil into various layers, or divide
the ground water into two ground water components, one of which is shallow
and subject to transpiration, and the other of which is so deep that no ground
water loss can occur through transpiration.
The distinction shown in figure 1-7 between overland flow, interflow, and
ground water flow is not generally made in applied hydrology because it is
virtually impossible to separate the three types. Instead, applied hydrologists
distinguish between surface flow and base flow and use a model of the hydrologie cycle something like that shown in figure 1-8. The precipitation is
divided into (1) precipitation excess and (2) infiltration and other losses.
The precipitation excess produces direct storm runoff. The infiltration replenishes soil storage which is drawn down upon by transpiration. Any excess
infiltration after soil moisture storage is satisfied forms recharge to ground
water, which eventually emerges as base flow. The presence of the threshold
in the soil storage phase of the system makes it impossible to treat the whole
system as linear, even where the evaporation and transpiration are completely
known. The development of the unit hydrograph theory as a linear relationship between precipitation excess and storm runoff avoided this difficulty by
fi
Infiltration
itiof^lI
1 Infiltration
InfiltrotinL
JL\
Subsurface inflow
|C
\ Q\ to channel
EvoDoration
Ground water
Qg Inflow
channel
*'
VSlnflQWt?,!
V-x channel I
c
-
Retention
J<.
Interceotion
^+1 +
Rainfall loss
Rainfall (R)^
Rainfall
V_/Rai
excess
FIGURE
Qs_
Surface inflow
to channel
1-6.Kulandaiswamy's block diagram.
I Runoff
to basin
16
T. 1
i"
^t
SURFACE
Qo
Fj '
\'
Q'
SOIL
CHANNEL R.O. ^
NETWORK
i
R
f
tf^Qm
IMH \AIATCD
v^l wyv/I^L^' wf^-% 1 t-rv
Qg
FiGUBE 1-7.The catchment as a system.
DIRECT
STORM
RESPONSE
I
I
SOIL
MOISTURE
GROUND
WATER
ET
FIGURE
1-8.The simplified catchment modeL
17
the elimination of the base flow and the infiltration. It is the existence of this
thresholdrather than the difference in response time between the surface
response and the ground water responsethat necessitates the separation.
In applied hydrology, the full model shown in figure 1-8 is not used. In
practice, the base flow is separated from the total hydrograph in some arbitrary fashion, and the precipitation excess is then taken so as to be equal in
value to the storm runoff. On the other hand, in soil moisture accounting the
threshold effect inherent in soil moisture storage is taken into account. It is
only recently that studies have taken both phases into account. Also, it is only
recently that the systems techniques developed for surface water have been
appHed to the problems of ground water response, notably by Kraijenhoff
van de Leur (25).
If we wish to consider the whole system shown in either figure 1-7 or figure
1-8, then we are of necessity dealing with a nonlinear system. This brings in
all the difficulties of nonUnear mathematics. It is not surprising, therefore,
that the concentration has been on the individual elements shown in figure
1-8. Over the past 35 years, unit hydrograph techniques have been developed
for dealing with the direct response in runoff and these techniques are all
based on the assumption of linear behavior. Similarly, drainage engineers
dealing with the saturated zone have used linearized equations, though it was
not until very recently that it was realized this would enable systems methods
to be used without further loss of generality (25). The unsaturated phase
involving soil moisture storage remains the most difficult part of the hydrologie cycle to handle. Not only does a threshold exist, but there is a feedback mechanism because the state of the soil moisture determines the amount
of infiltration. It is in the unsaturated phase that the greatest diflculties will
be encountered and that the greatest amount of work needs to be done.
The systems approach has been fruitful in many other disciplines. Such
work as has been done on parts of the hydrologie cycle has been encouraging.
There is every reason to believe that the application of the systems approach
to the whole hydrologie cycle will produce a coherent theory of hydrologie
systems, which can form the basis for an applied hydrology with a scientific
basis. The development over the past 15 years can be followed in the references
cited at the end of this lecture. General surveys of the problem from varying
points of view have been given by Paynter (36), Amorocho and Hart (3),
Kraijenhoff van de Leur (26), Nash (33), and Dooge (17).
Linear Time-Invariant Systems

The essence of linearity is the principle of superposition, which may be
described as follows (an arrow signifies that a particular input to the system
results in a particular output) :
Ixi{t)-^yi{t) and 0:2(0^2/2(OJ
18
then the system is said to be Unear if:

Xi{t)+X2{t)-^yi{t)+y2{t)
This principle includes the principle of homogeneity in the special case in
which X\ = X2.
The principle of superposition, of course, is not confined to the addition of
only two inputs. Any number of inputs can be added together as long as the
principle holds; the output will be the sum of the individual corresponding
outputs. Since integration is a limiting form of summation, if the input to the
system can be expressed as the integral of any function/(), then the corresponding output can be obtained by integrating the output due to an input
The system linearity defined by the principle of superposition must be distinguished from the existence of a general linear (that is, a straight line)
functional relationship between input and output. It can easily be verified that
if the input to a system is x and the output \sy = ax+hj the system is not linear.
A system is said to be time-invariant w^hen its parameters do not change
with time. For such a sj^stem, the form of the output depends only on the form
of the input and not on the time at which the input is applied. Thus, if
x{t)^y{t)
then for a time-invariant system:
x{t+T)-^y{t+T)
where r is a time shift which may be either positive or negative.
In hydrology, the assumptions of linearity and time-invariance are not
valid, but nevertheless have been used for a long time in applied hydrology
because of the simplification they introduce. The ability to predict the output
from a hydrologie system is based on past records of input and output. By the
assumption of time-invariance, it is possible to predict an output for a given
input if that particular input has already occurred at some time during the
period of record. Without the assumption of time-invariance this would not
be possible. The further assumption of linearity allows the prediction to be
made even though the shape of input in which we are interested has not
occurred in the past. This is done by (1) breaking down the past input and
the input being considered into basic elements of standard shape but varying
volume, (2) decomposing the past output so as to obtain the output due to a
characteristic input element of standard volume, (3) using the latter result to
predict the output due to the individual characteristic elements of the input
being considered, and (4) superimposing the outputs from these individual
characteristic elements to obtain the total output. This is the basis of the
unit hydrograph procedure, which deals with the storm runoff for a unit
period.
19
The problems of systems analysis and synthesis are also greatlj'' simplified
if the input and output of a system are assumed to be lumped. In a lumped
system with a single input and a single output, the behavior of the system
would be governed by an ordinary differential equation. For the system with
several inputs and several outputs, the behavior of the system would be
described by a set of differential equations. If the inputs and outputs are not
lumped, then the system behavior must be described by partial differential
equations. Since partial differential equations are much more difficult to
handle than ordinary differential equations, there are distinct advantages in
using lumped inputs and outputs in the first attempt to formulate a theory
of system behavior.
The assumptions of linearity and time-invariance are also reflected in the
type of differential equations which would describe the behavior of the system.
Thus a lumped linear system would correspond to an ordinary linear differential equation. If the system were also time-invariant, then the differential
equation would be an ordinary differential equation with constant coefficients.
The fact that ordinary differential equations with constant coefficients are
far easier to handle than any other type indicates the advantages of making
the assumptions of lumping linearity and time-invariance in the handling
of system operations.
The assumption of linearity helps us greatly with the problem of predirection. If a complex input can be described in terms of a set of simple
characteristic functions and the output corresponding to each of these characteristic functions is known, then the output due to the complex input can be
obtained by superposition. This question has been well discussed by Sievert
(40). It is, of course, possible to expand an arbitrary function in a great
variety of ways. Thus, we could expand the function in terms of a power
series:
x{t)^Co + Cit + C2t'+
(2)
or in terms of an exponential series:

x{t) =Co+Cie-'+C2e-2+
(3)
The trouble with such series is that in the case of a function which is given
numerically, it is difficult to determine the values of the coefficients in the
expansion with good accuracy. If, however, we expand x{t) in terms of a set
of functions/(O :
X(t) =Co/o(0 +Ci/i(0 +C2/2(0 +
(4)
where the functions /(O are orthogonal (see ''Orthogonal Polynomials and
Functions,^' lecture 3) then the property of orthogonality can be used to find
the coefficients d relatively easily and with good accuracy.
In choosing between the orthogonal functions available it is, of course,
convenient if the orthogonal series used to fit a given x{t) is as short as
20
possible. Consequently, one set of orthogonal functions may be preferable

to another set because of the nature of the input. If the input is expanded in
terms of a set of orthogonal functions/(i) in accordance with equation 4 and
the output corresponding to each of these orthogonal functions is given by:
Mt)-^9t)
(5)
then the output from the system due to the input x{t) is given by:
y(t) =CQgo(t) +cigi{t) +C2g2{t) +
(6)
where the values of the respective coefficients in equations 4 and 6 are equal*
It is also convenient if the output corresponding to the typical orthogonal
function is simple in form. Thus, the choice of a convenient set of orthogonal
functions for representing the input, output, and response function depends
both on the nature of the input and the nature of the system.
Electrical engineers deal with lightly damped systems in which the inputs
are usually sinusoidal. Consequently, Fourier methods of analysis are of great
utility in electrical engineering, since the sine and cosine functions are orthogonal to one another and are of the same general form as the inputs and
outputs. Consequently, the Fourier methods were the first to be developed in
systems analysis. The various developments of Fourier methodsthe Fourier
integral for dealing with transients and the Laplace transform for dealing with
unstable systemsare natural developments. These well-established techniques can be found in standard texts such as Gardner and Barnes (22).
In hydrology, however, the systems are not lightly damped and the responses are not oscillatory in nature. Instead, we have systems that are very
heavily damped. It would, therefore, be foolish to take over from the electrical
engineer the techniques he has developed for his particular problems without
close examination of their relevance to hydrologie systems.
Continuous forms of the convolution equation
The derivation of the fundamental equation for system operation of a linear
system depends on the use of the concepts of an impulse function and the
impulse response. The impulse functionor Dirac delta functionîs really
a pseudofunction or distribution which is usually defined as having the
properties:
5(-o)=0,
{t-to) dt = l
when
tp^to
(7)
(8)
/_
The delta function is usually visualized as the limiting form of a pulse of some
particular shape as the duration of the pulse goes to zero. The more correct
21
mathematical definition of the delta function:
x(t)= f
{t-T)x(T) dr
(9)
is actually more directly useful for our purpose here. Siebert (40) has pointed
out that equation 9 is a special limiting form of equation 4, in which/(i) is
replaced by the orthogonal delta function o(r) and the orthogonal coefficients d are given by X(T) .
The impulse response of a system, h{t)j is defined as the output from the
system when the input takes the form of an impulse or delta function, that is,
if x{t) =0(0, then y{t) =h(t). If the system is linear, the impulse response
gives as complete a description of the system behavior as is needed. In surface
water hydrology, the lUH is the impulse response of the catchment.
The two concepts given above can be used to derive a convenient mathematical formulation of system operation for a lumped linear time-invariant
system. If the impulse response of the system is h(t)j then we have:
d{t)^h{t)
For a time-invariant system
{t-T)-^h{t-T)
For a linear system
X(T)
{t-T)-^x{T)h{t-T)
Any arbitrary input x{t) can be considered as being made up from an

infinite number of delta functions as indicated by equation 9 above. Since the
operation of integration is linear, the output from such an input x{t) will be
given by integrating the weighted output x(T)h(tT) corresponding to the
individual delta functions 5(r) :
x{t)-^ f
h{t'-r)x{T) dr
(10)
Thus for an input x{t) and an output y{t), we have the relationship:
2/(0=
Ht-T)x{T) dr
(11a)
'00
The right-hand side of this equation represents the well-known mathematical

operation of convolution, which is often represented by an asterisk so that
we can write :
y{t)=h{tyx{t)
(lib)
Thus, the completely general relationship indicated in equation 1 has been
replaced by the definite convolution relationship represented by equation 11
for a lumped linear time-invariant system. As long as we confine our attention
to such systems, we will be concerned with the solution of equation 11. The
22
problem of prediction now becomes the solution of equation 11 for known

values of x(t) and h{t) and, hence, represents only the operations of multiplication and summation which are inherent in convolution.
The twin problems of system identificationthe determination of h(t)
and of signal identification^the determination of x{t)are now seen to
involve the solution of an integral equation which is, of necessity, a much more
difficult mathematical problem than that of convoluting two known functions.
The problem of synthesis is now seen to be that of devising a simulation
system whose impulse response will, to a sufficient degree of approximation,,
represent the function h (t) which is required. The impulse response in equation
11 is the lUH used in hydrology. In other disciplines, it is variously referred to as
an impulse response or a characteristic response or a weighting function; in
mathematics it is referred to as a kernel function, a Greenes function, or an
influence function.
Though we are largely concerned with lumped linear time-invariant systems,
it is instructive to consider briey the more general forms of the mathematical
relationship between input and output w^hen these assumptions are relaxed.
If instead of a single input, we had a number of lumped inputs, then the
relationship would be as follows:
2/(0 = 2 r Xi{r)hi{t-T) dr
(12)
=1 '-00
An equation of the above type would apply to the case where the rainfall was
measured at several points in the catchment and the values of Xi(t) represented
the individual rainfall records. In such a case, hi{t) would represent the
contribution from the portion of the catchment area corresponding to the
i^^ rain gage to the flow^ not at the outlet from that subcatchment but at the
outlet from the whole catchment. The solution of the identification problem
in this case would involve the solution of a set of simultaneous integral equations. If the rainfall were taken as completely distributed over the catchment
area, then the equation for the outflow at the end of the area wôuld be given
by:
y {t)=
x{T,a)h{t-T,a-a)dT da
(13)
In a system which has a lumped input and is linear but time varying, then
the impulse response h{t,T) is a function of both the elapsed time t and the
time r at which the impulse of input is applied to the system. Thus we have
the relationships :
(t)^h{tfi)
d{t-T)-^h{t,T)
x(T){t-r)-^x(T)h{t,r)
23
Using the property of linearity, we would have for an input x{t)j the output
given by:
x{t)-^
x{r)h(t,r) dr
(14)
so that the system operation is defined by:

2/(0= r x{r)Ht,T)dT
(15)
'00
Since equations 11, 12, 13, and 15 are superposition integrals, they apply
only to linear systems. There is no corresponding general formulas for the case
where the system is nonlinear, but special formulas can be developed when the
system is assumed to belong to a particular class of nonlinear systems.
If we make the assumptions of lumped inputs and outputs, linearity, and
time-invariance, we have the general superposition integral given by equation
11a. Since in this equation, r is a dummy variable of integration, we can
replace it by r in which case the superposition integral becomes:
y(t) =
h{T)x{t-T) dr
(lie)
Equations 11a and lie are equally valid formulations of the relationship
among the input, the system operation, and the output.
The limits of the superposition integral can be modified if we make the
further assumption that the systems being considered are causal, that is, that
the output cannot occur before the input. Since the impulse response h{t) is
the response to a delta function at time = 0, the impulse response function
will be zero for a negative argument. Thus for causal systems, equation 11a
can be written as :
2/(0= /' x{T)h{t-T) dr
(16a)
00
and equation lie can be written as:
2/(0= r Hr)x(t-T) dr
(16b)
'o
If the system has a finite memory, or if the input has existed for only a
finite time, then the limits will be further modified. If the length of the memory
is n, then the impulse response will be zero for arguments greater than n and
equation 16a may be modified to read:
2/(0= /
' i-n
x{T)hit-T) dr
(17a)
24
and equation 16b will be modified to read:

y{t)= f h{T)x{t-T) dr
(17b)
The equations given above will hold for the case where the input has occurred for an infinite time in the past. For an isolated input, it is convenient
to take the time zero at the start of input. In this case, the value of the input
x{t) will be zero for negative argument. For an isolated input to a system with
infinite memory, equation 16a will be modified to:
2/(0= ( x(T)h{t-T) dr
(18a)
2/(0= ( h{r)x{t-T)dT
(18b)
and equation 16b to:
For an isolated input to a system with finite memory the limits of integration
in equation 17 will also be modified so that the range of integration will not
exceed , but in practice, it is more convenient in this case to retain the limits
and record the zero values. Equation 18 is the normal form of the convolution
equation which is dealt with in parametric hydrology. Except in special
circumstances, which will be noted, it is the form used in the present lectures.
Classical systems analysis as developed by electrical engineers has grown
up around frequency analysis, which is essentially the analysis of periodic
phenomena. Care must be taken if these techniques are to be used in the analysis of hydrologie systems. Such techniques can only be used if the sj^stem
under review has a finite memory. In such a case, if the input is of length M
and the memory of length TV, then the length of the output will be P where:
P = M+N
In hydrologie terms, M is the duration of rainfall excess; N is the base length
of the lUH, and P is the duration of surface runoff. Since, in the case of a
single storm event, everything that we are interested in is contained between
zero time and P, we could assume the whole phenomena as periodic with a
period JT, provided that T is equal to or greater than P. This would mean
that both the input and the output would be assumed to be repeated at the
chosen interval T, Since these would be repeated inputs and not isolated
inputs, we would not be entitled to set the limits of the convolution integral
at zero and t as in equation 18. If the memory were finite and equal to iV,
25
the convolution equation would then become:

y{tkT)= f
x(TkT)h(t-T) dr
(19a)
x{t-T^kT)h{r) dr
(19b)
'^ t-N
y{tkT)= /
where
x{tdzkT)=0
for
M<t<T
(19c)
and
T>P = M+N
(19d)
Discrete forms of convolution equation

The form of the convolution equation given above as equation 18 is for the
case where both the input and the output are continuously defined. If either
the input or the output is given in quantized or discrete form rather than
continuous form, the convolution equation must be modified accordingly.
In the classical unit hydrograph procedures, the rainfall is frequently given
as a histogram, that is, in quantized form. In such a case, we deal not with an
lUH, but with the finite period unit hydrograph introduced in 1932 by
Sherman (39). A histogram input with an interval D can be defined either in
terms of the histogram ordinates x{t), where t is the actual time elapsed, or
in terms of the histogram areas X{aD), where o- is the number of intervals
elapsed before the beginning of the interval in question. The latter is more
convenient and is used below. The histogram of input can be expressed in
terms of the volumes of input X((TD) in successive standard periods as follows:
x(t)= x{(TD)PD{t-aD)
(20)
where
Pz>(i-crZ))= ^ for
aD<t<{(T+l)D
(21a)
and
PD (t-aD)
=0
for other values of t
(21b)
Equation 21 is in effect the equation for a rectangular pulse of duration D and

unit volume. Note that the volume of such a pulse is D and not unity.
Having replaced the delta function by the square pulse, we now replace
the impulse response h{t) by the pulse response hnit) which is defined as
being the output from the system when the input is given by the rectangular
pulse defined in equation 21. Thus, we have:
PD{t)->hD(t)
26
For a time-invariant system:

PD{t-(TD)-^hD{t-(TD)
For a linear system :
X{aD) 'hD{t-(TD)-^X{aD) *hD{t-aD)
Since summation is a linear process, we can write the output due to the input
defined by equation 21 as:
X(t)-^ "if X{aD)hD{t-aD)
(22)
so that the relationship between input and output for the system is given:
2/(0= i, X{<7D)hD{t-<TD)
(23a)
(r=00
which corresponds to equation 11a for continuous input. As in equation lie,

this equation can be written in the alternative form:
2/(0= i, X{t-aD)hD{cTD)
(23b)
ffssz00
As in the continuous case, the limits of summation will be affected by the

further assumptions of causality, finite memory, or zero input for zero time.
In particular, for a causal system with an infinite memory, we have for
isolated input:
y{t)= "Z Xi,TD)ho{t-aD)
(24a)
<TD=0
<TD=t
y(t)= J2 X{t-aD)hDaD)
(24b)
<TD=0
Equation 24 is the convolution equation for a finite period unit hydrograph.

Both the unit hydrograph and the output are defined continuously even
though the input is defined in quantized form being constant over each interval
of length D.
In some of the early unit hydrograph work, both the input of rainfall excess
and the output of storm runoff were represented by volumes over a fixed
interval. The convolution equation for this case would be:
Y(sD) = X{aD)dD(sD-<7D)
(25)
27
where both s and a are discrete variables and do is the distribution graph for
the interval length D for the particular catchment. The distribution graph
dD represents the proportion of the inflow during a standard interval which
runs off in successive standard intervals.
In some cases, the input and output are only sampled and, thus, are only
available in the form of functions known at discrete moments of time. In this
case, the convolution equation would take the form :
y(sD)= i, X((TD)hD{sD-(TD)
(26a)
which can be written without ambiguity as:

y{s)= Z((7)/iz>(s"cr)
(26b)
O-ss00
Here again both s and a are discrete variables and ho is the finite period unit
hydrograph. For a causal system with an isolated input this, of course, can
be written as :
y(sD) = X{aD)hD{sD-cTD)
(27a)
<T=0
or
y{s)= T.X{a)hn{s-a)
(27b)
<r=0
where yis), X(o-), and hnisa) represent the ordinates of the output, the
input, and the finite period unit hydrograph, respectively, at standard intervals D.
Equation 27 can also be written in the alternative form:
yi= ^Xjhi-j
y=o
(28a)
yi= YjXi-jhj
(28b)
3=0
In the above equation, x has been used to represent the volumes of input which
appear as X in equation 27. This is done in the interest of simplifying matrix
equations which are developed later.
When written out in full, equation 28b has the familiar form given in textbooks on classical hydrology which is given below for an input lasting for five
28
units of time and a system memory length of three units of time :

yo = hoxo
(29a)
yi = hiXo+hQXi
(29b)
y2 = h2Xo+hiXi+hoX2
(29c)
t/3 = hzXo+h2Xi+hiX2+hoX3
(29d)
yA = h3Xi+h2X2+hiX3+hoXA
(29e)
2/5 = hsX2+h2X3+hiX4^
(29f )
2/6 = /3Z^3+/i3a^4
(29g)
yi^hzXA
(29h)
The above set of simultaneous equations can be written in the matrix form:
{2/}p+l.l=CX]^l.n+l{/l}n+l.l
(30)
Where the matrix of inputs which has p+1 rows and n+1 columns is given
below:
Xo
Xi
Xo
Xm
Xfn1
XiXo
(31)
Xo
x,
P+l,n+l
An alternative matrix formulation of the discrete case is :

{y}p+l,l=ZH']j^l,m+l{x}m+l,l
(32)
where the H matrix is made up from the h vector in the same way as matrix
31 and has p+1 rows and m+1 columns.
Equations 27, 28, 29, and 30 are merely alternative ways of formulating the
relationship between the volume of input and the rate of output. Where the
input is defined strictly as a discrete function, it is necessary to adjust the
equations. Thus equation 27b for the relation between input volume and
29
output rate would be replaced by:

<r8
(33)
2/(s)= â;((T)i>(s(r)Z)
Note that as D approaches zero, equation 33 approaches the form of the

continuous convolution equation 11a.
Identification and Simulation

Classical unit Hydrograph methods
The problem of identification is the characterization of the system response
from a given record of input and output. In hydrologie terms, the problem is
to derive the unit hydrograph from a given record of precipitation excess and
storm runoff. The classical method of solving this problem was by trial and
error. Though it has nothing of the systems approach about it, this method
has been illustrated in a systems fashion in figure 1-9. In the classical approach,
some form of the unit hydrograph, that is, the impulse response, or pulse
response, is assumed and applied to the given rainfall excess. The prediction
of the output for this assumed unit hydrograph is merely a matter of simple
multiplication and addition. The output based on the assumed unit hydrograph
is then compared with the actual output and a decision made as to whether
the fit is close enough.
If the fit is judged to be sufficiently close, then the assumed unit hydrograph
is accepted. Otherwise, the assumed unit hydrograph is modified and the
procedure repeated until an exception fit is found.
While the above procedure may be acceptable as an ad hoc method of
getting a specific answer to one particular problem, it cannot be accepted as
deserving of the name of scientific hydrology unless both the criterion of
acceptable fit and the rule for modifying the trial unit hydrograph are objectively defined. The technique of optimization by eye has been widely used,
modify h
accept h
FIGURE
1-9.Identification by trial and error.
30
not only in unit hydrograph studies but also in many other branches of
hydrology. The supposedly learned journals on scientific hydrology abound
with papers in which two curves are said to be a sufficiently accurate approximation of one another or in which a curve is said to represent some plotted
data to a reasonable degree of accuracy. In a rational science, it should be
possible for a second worker to use another scientist's data and reach exactly
the same conclusion. The systems approach in hydrology attempts to achieve
this latter objectivity instead of the subjectivity inherent in many of the
methods in use today.
Figure 1-10 is a systems representation of the Collins {10) method of deriving the unit hydrograph. This is an iterative method and one which is a
distinct improvement on the trial-and-error approach. In Collins' method the
assumed unit hydrograph is not applied to the whole precipitation excess
record, but only to all the rainfall volumes other than the maximum. The
resulting estimated runoff, therefore, represents the runoff due to rainfall in
all periods except the period of maximum rainfall. When this estimate is
subtracted from the actual runoff, the difference gives an estimate of the
runoff due to the rainfall in the unit period of maximum precipitation excess.
When divided by the appropriate volume of precipitation excess in the period,
this runoff due to maximum rainfall gives a new estimate of the unit hydrograph, and the whole process is then repeated. Except for unusual conditions, the iterative procedure is convergent. If the unit hydrograph is constrained to be causal, that is, to have zero ordinates for negative time, then
the effect of the ColUns' procedure is to concentrate any error in the matching
of the runoff hydrograph into the portion of that hydrograph due to rainfall
before the period of maximum rainfall.
Transform methods of system identification
Parametric hydrology has concerned itself with the development of such
objective methods as the impulse response or the rectangular pulse response
for determining the unit hydrograph. These methods will be discussed in
O
FIGURE
1-10.Identification by iteration (Collins' method).
31
detail in later lectures, but a brief preview is in order at this point. The
methods used can be grouped into two general classes, one of which may be
referred to as transform methods and the second as correlation methods.
Figure 1-11 shows the general approach of the transform methods to the
problem. In these methods, the known input and the known output are transformed in some fashion. These transformed inputs and transformed outputs
are then used to determine the transform of the impulse response or the
rectangular pulse response. If the transformed response can be inverted, the
actual impulse response or pulse response will then be known in the original
time domain. A complete transform method of identification therefore, contains three elements: (1) The transformation of the input and the output;
(2) the use of a linkage equation, which defines the transform of the system
response in terms of the transform of the input and the output; and (3) an
inversion of the transformed system response to get the system response as a
function of time.
The most widely used transform method in systems analysis is the Laplace
transform. In this method, the Laplace transform of the input and the output
are found. The Laplace transform of the impulse responsewhich is given
the special name of the system functionis then found by dividing the
Laplace transform of the output by the Laplace transform of the input. The
system operation is thus described in the transform plane, but most hydrologie situations will be described numerically rather than functionally. To
determine the impulse response as a function of time involves the difficult
problem of the numerical inversion of the Laplace transform.
In 1952, Paynter {36) applied the method of systems analysis based on the
Laplace transform to various problems in hydraulic engineering. He was
largely concerned with problems of water hammer and turbine governing,
but in part III of his paper, he dealt with the problem of flood routing. Unfortunately, for the development of systems hydrology, Paynter's ideas were
not followed up at the time.
In 1959, Nash {31) ^ then working in the HydrauUc Research Station in
Great Britain, attempted to describe the lUH in terms of its statistical
moments. He showed that for a linear time-invariant system, the moment of
the input, the impulse response, and the output are connected by the
equation
MR{y) = Z (^)M,{h)Mj,.,{x)
(33)
where MR{y) is the R^^ moment of the function y{t). Moments may be taken
either about the time origin or about the respective centers of the individual
functions. This is essentially a transform approach since the moments of a
function are a transform of it, and Nash's theorem of moments, given above
as equation 33, is the linkage equation between the transformed input, the
32
SYSTEM
H
-^T(H)-^
T(P)FIGURE
*^Q
T(Q)
1-11.Identification by transformation.
transformed output, and the transformed systems response. The problem of

inversion (finding the form of a function given its moments) is again an
extremely difficult one and can be shown to be equivalent to the problem of
numerically inverting a Laplace transform.
Next, O'Donnell {34) applied harmonic analysis to the problem discussed
by Nash. O'DonnelPs approach was to find the Fourier coeflScients of the
system response. The method depends on the fact that the terms of a Fourier
series are orthogonal. The response function is known (to a degree of accuracy
depending on the length of the series) once the Fourier coefliicients for the
function are known. Thus, the harmonic analysis method used by O'Donnell
does not encounter any difficulty in the inversion procedure. Because Fourier
analysis is concerned with periodic functions, the method can, however, only
be applied to systems with finite memory.
In 1964, Levi and Valdes {28), working in Mexico, appUed the Fourier
transform to the problem of systems identification in hydrology. In the same
year, Diskin^ took up Paynter^s work and applied the Laplace transform in
more detail to the study of unit hydrographs.
In 1965, Dooge {16) suggested the use of Laguerre coefficients rather than
harmonic coefficients for the analysis of heavily damped systems, such as are
encountered in hydrology. This method was developed because Dooge felt
the method of harmonic analysis, which depends on sine curves as its basic
elements, was not entirely suitable in hydrology where many functions were
of a dead beat type rather than an oscillatory one. It was thought that if an
orthogonal method could be derived in which the elements of the series were
of much the same form as the gamma distribution (which had proved so useful
* DiSHKIN, M.
A BASIC STUDY OF THE LINEARITY OF THE RAINFALL-RUNOFF PROCESS
IN WATERSHEDS. Ph.D. Thcsis, Univ. 111. Urbana. 1964. [Xerox copy available by purchase
from University Microfilms, Inc., P.O. Box 1346, Ann Arbor, Mich. 48106 as Publication
No. 64-8375.]
33
in applied hydrology), that the number of terms required to represent a given

response function would be less than in the harmonic method.
The above methods of systems identification will be discussed in greater
detail in lecture 5. Meanwhile, it is only necessary to note that they are all
objective methods of system identification.
Correlation methods of system identification
The second group of objective methods of system identification consists of
methods based on least squares correlation. The method of least squares was
applied to the derivation of unit hydrographs by Snyder (P) in 1955 and also
developed independently in AustraUa by Body (8) in 1959. Body published
in detail the matrix operations involved and the adaption of the method for
digital computers. Snyder (42) published the matrix formulation of the
method in 1961.
The set of equations represented in equation 29 comprises (p+1) equations
in (n+1) unknown values of h and, consequently, is overdetermined. In
theory, any group of (n+l) equations could be selected from the (p+1)
equations available to solve the equations for the values of the unknown
ordinates {hi) of the unit hydrograph. In practice, of course, the data are not
exact, and, consequently, no unique mathematical solution exists which would
be vaUd for all inputs. If the first (n+l) equations are chosen and the equations solved by forward substitution, the ordinates of the unit hydrograph
may become unstable and unrealistic. The procedure introduced by Snyder
and Body is to use all the equations and the least squares criterion to produce
the optimum values of the unknown ordinates of the unit hydrograph. The
matrix form of the unit hydrograph equations is given by equation 30 :
{y}p+l=iX^lH-l,n+l{h}n+l,l
(30)
The least squares formulation of the problem is given by:
mvi.îC2/]p4-i.i=mvi.p+imî.n+i{}n+i.i
(34)
Since the product of the transposed matrix Z^ and the original matrix X is
necessarily square, this product can be inverted, and the vector of unknown
unit hydrograph ordinates can be written as,
{hUi,i={XJX:}}-^\:Xj{y}
(35)
This procedure is shown diagrammatically in figure 1-12. The record of input

is used to determine the input matrix, and this is then multiplied by its
transpose. The output vector is also multiplied by the transpose of the input
matrix, and these two products are used to determine the optimum unit
hydrograph, which is then accepted as an estimate of the true unit hydrograph.
The method of time-series analysis, also shown in figure 1-12, can be classed
as a correlation method. If the record is a continuous one, or a discrete record
34
{0}
Hô}
{ôpt}
<^pp (K)FIGURE
^ pQ (K)
1-12.Identification by correlation.
existing for infinite time, it is not possible to apply the least squares method,
since the matrices become extremely large and impossible to invert. In the
case of an inflow which is not isolated, it is also impossible to use the method of
Laplace transforms or Fourier transforms since the function may not behave
at infinity in accordance with the requirements of mathematical theory.
However, a long-time series can be transformed and described in terms of its
autocorrelation function. The autocorrelation function of a time series is
defined as the limit :
<l>xx{k) = - J^ x{i)x{i+k)
(36)
^ i=-p
where n = 2p+l is the number of data points as p tends to infinity.

Where two time series are known (for example, an input and an output),
we can determine their cross-correlation coefficient which is defined as the
limit :
1 ^
<t>xy{k)= - 23 x{i)y{i+k)
(37)
as p tends to infinity. If we have a causal, linear, time-invariant system, it
35
can be shown that the optimum impulse response in the least squares sense
is given by :
3=oQ
<t>xv{k) = X^ hopt{ j)(t>xx{k-j) ;
when
k>0
(38)
which is a discrete Wiener-Hopf equation. We started off with the ordinary

convolution equation (equation 26) and ended up with another convolution
equation. However, equation 38 connects the cross correlation of x and y.
If the input is isolated, no advantage has been gained, and it can be shown
that equation 38 is equivalent to the least squares procedure of Snyder and
Body, though more complicated. If, on the other hand, we have an infinitely
long time series which we are continuously sampling, then the problem has
been reduced to manageable form. The time series approach is currently being
developed at the Massachusetts Institute of Technology under Eagleson (W),
and work is also being done by Bayazit^ of the University of Ankara.
Methods of simulation
Even if we could completely solve the problem of identification, this would
only enable us to predict the future outputs from an individual system. Complete identification would not help us in any way to predict the output from a
system of the same class for which records of input and output were not
available, or to study the effect of variations in the parameters of similar
systems on their outputs. Furthermore, the identification of nonlinear systems
is extremely difficult, and, in such cases, it is natural to turn to simulation
rather than identification as the basis of a prediction. It is important to remember that we are still interested in the overall performance of the system rather
than the details. We are looking for a rehable predictor rather than a photographic reproduction when we seek a model to simulate our system. The
model system used to simulate an actual system may be either abstract or
real. According to Chorafas (9), ^'Simulation is simply a working analogy.
Analogy means similarity of properties or relations without identity.'^ A
model may be defined as being a system which can reproduce some, but not
all of the properties of the prototype.
Figure 1-13 shows the division of methods of simulation into three broad
groups. It is intended as a basis for discussion rather than a strict classification.
In this tentative classification, the problem of simulation is looked upon as
5
BAYAZIT, M.
INSTANTANEOUS UNIT
HYDROGRAPH DERIVATION BY SPECTRAL ANALYSIS
AND ITS NUMERICAL APPLICATION. CENTO Symposium on Hydrology and Water Resource
Devlpmt. Ankara. 1966.
36
PROBLEM
CONCEPTUAL
MODEL
SEMI-DIRECT
SIMULATION
DIRECT
SIMULATION
SCALE MODEL
SPECIAL ANALOG
NETWORK ANALYZER
HELE-SHAW
FIGURE
INDIRECT
SIMULATION
DESK CALCULATOR
DIGITAL COMPUTER
DIFFERENTIAL
ANALYZER
1-13.Methods of simulation.
being a two-stage problem. First, we take the actual field problem and abstract
from it a conceptual model of the problem. This conceptual model might be
very simple or it might be extremely complex. In other words, we might do a
lot of the work at this stage or very little. The next step is to attempt to derive
quantitative results from the conceptual model. The method in which this is
done often depends on the extent to which the conceptual model has been
developed. The two stages shown on the figure represent the two problems of
abstraction and of completion.
If the conceptual model has not been developed to any great extent, it will
probably be necessary to use a direct method of simulation to get quantitative
answers. An example from hydraulic engineering may be used to illustrate
this.
In the design of a hydraulic structure, the conditions may be so complex
that all we can say of our conceptual model is that we believe gravity forces to
be dominant in the problem. We could then decide to build a model which was
geometrically similar in some respect to the prototype and which would be
designed according to model laws based on the Froude number. Such a
hydraulic model would be a direct simulation of the problem and would be a
close imitation of the prototype. It would be possible to recognize the different
parts of the prototype in the model. On the other hand, a problem in the
hydraulics of open channels might be solved by developing a much more
complete conceptual model. This model would be based on the geometry of
37
the situation, the equation of continuity, and the dynamic equation of unsteady open channel flow.
The finite difference equations incorporating physical assumptions and the
geometry of prototype would constitute an abstract model of the actual
problem. If it were completely specified, the actual computations could be
done on a general purpose computer of some type; thus, we might use a desk
calculator, a digital computer, or a differential analyzer. In this type of
indirect simulation, it would not be possible to identify visually any part of
the prototype in the model. The physical model can only solve one particular
ad hoc problem, but does not require a great deal of work at the conceptual
phase. On the other hand, the indirect simulation on a computer of some type
can solve a very wide variety of problems, but the amount of work done in
setting up the problem, i.e., constructing the conceptual model is often very
great. In between these two we have methods of semidirect simulation in which
we can construct a model which will solve particular types of problems.
Examples of these are netwôrk analyzers and Hele-Shaw models.
Simulation in hydrology
Figure 1-14 shows the Stanford Watershed Model Mark II used to simulate
the land phase of the hydrologie cycle. Though the Mark II model is shown
here, the Stanford model has since been developed to the Mark IV (11) and
Mark V stage in which the performance of the model has been improved at a
PRECIPITATION
EVAPORATION
^ DEPLETION
RUNOFF FROM IMPERVIOUS AREAS
UPPER ZONE STORAGE

INTERCEPTION AND
DEPRESSION STORAGE
DIRECT RUNOFF
DEPLETION
LOWER ZONE
STORAGE
DEPLETION
GROUND WATER h^ TRANSLATION a

STORAGE
ROUTING STORAGE
SUBSURFACE
GROUND WATER
FLOW
FIGURE
SURFACE RUNOFF
INTERFLOW
TRANSLATION 8
ROUTING STORA
HOURLY OR
DAILY STREAMFLOW
1-14.Stanford Watershed Model Mark II.
38
cost of extra complexity. Figure 1-14, however, shows the main features of
the Stanford model. The model as shown is essentially a conceptual model and
represents the first, or conceptual phase, of the simulation process as described
above. It is a flow diagram representing the main features of the simulation
model, and it must be supplemented by operational rules for determining the
amount of moisture movement from one component to another.
The computation, which is the second part of the simulation process, is
carried out on a digital computer. The model shown in figure 1-14 could, of
course, be computed by any other means, but the digital computer is the most
convenient method. There have been many other instances of the simulation
of subsystems or components in the hydrologie cycle and the solution on a
digital computer.
Dawdy and O^Donnell {12) pioneered the systematic study of objective
techniques for the optimization of parameters of simulation models. This key
question is discussed in later lectures.
Numerous attempts have been made to simulate the direct storm runoff
from a watershed by a conceptual model, which would be simple in form but
would have essentially the same operation as the watershed under study.
Many of these conceptual models involve some simple arrangement of linear
storage elements only, or else a simple arrangement of linear storage elements
and linear channels (15). In most cases, the behavior of these conceptual
models is predicted by analytical methods; however, any method for final
computation may be used.
Figure 1-15 shows the analog simulation of a linear storage element as given
by Shen (38). In this case, the analog element is not a direct analog of a
catchment element, but an analog simulation of a conceptual element for use
where an arrangement of conceptual elements has been synthesized to simulate
the action of the watershed. Figure 1-16 shows the simulation of a linear
channel also by Shen. A linear channel is purely a conceptual element because
no one has ever seen one and no one ever will. The analog units shown in
figures 1-15 and 1-16 are direct analog simulations of the conceptual elements,
but it is also possible to have indirect analog simulations in which the mathematical equation for the conceptual element is written down and then an
E,
R
AAAr
>
Î
FIGURE
1-15.Direct analog of linear reservoir.
E2
-o
0.8C
E,
39
0.8C
I/6R
3.6C
FIGURE
1-16.Direct analog of linear channel.
analog unit assembled in which each of the mathematical operators is simulated and appropriately connected.
This is the end of a review of the development of parametric hydrology and
a preview of the material to be covered in the present course. The development
of the subject has been going in many scattered directions since 1932, but in
recent years it has gathered pace and is beginning to settle into a consistent
body of knowledge. No matter what our problem, no matter what types of
models we seek to use, we face essentially the two difficult problems of system
identification and system simulation. Our present knowledge is such that
identification can only be carried out with some degree of success if we make
the assumptions of linearity and time-invariance. We need not be restricted in
simulation because we can build in the nonlinearity and time-variance into
our model and predict the operation of the resulting nonlinear system in some
fashion. Nevertheless, if we wish to simulate objectively, or indeed efficiently,
it is desirable that the nonlinearities be reduced to a minimum and that if
possible the nonUnearity be confined to one part of the model, while the
remaining subsystems and their components are linear in action.
Problems on Hydrologie Systems

1. The following terms are commonly used in hydrology:
Unit hydrograph
S-hydrograph
Instantaneous unit hydrograph (lUH)
In each case, write down the corresponding terms used in other disciplines to
denote the same concept.
40
2. The Muskingum method is commonly used in flood routing. Describe

this method and distinguish between the separate problems of prediction,
identification, and simulation.
3. Describe a part of the hydrologie cycle with which you are familiar; use
the nomenclature of the systems approach. Show the relationship of this part
of the hydrologie cycle to the other parts of the cycle by means of a simple
sketch. By means of a second sketch indicate how this part of the cycle might
be considered as consisting of a number of subsystems.
4. For the part of the hydrologie cycle described in question 3, list one or
more classical methods used in applied hydrology. Do these methods make the
assumptions of linearity or time-invariance? Describe the methods using
systems nomenclature.
5. For some particular part of the hydrologie cycle, give examples of the
use of simulation in hydrologie forecasting.
Literature Cited
1) AcKERMAN, W. C, COLMAN, E. A., and OGROSKY,
1955. FROM OCEAN TO SKY TO LAND TO OCEAN.
H. O.
U.S. Dcpt. AgF. Yearbook 1955:
41-51, illus.
2) AcKOFF, R. L.
1962. SCIENTIFIC METHODS. 464 pp. New York.
3) AMOROCHO, J., and HART, W. E.
1964.
A CRITIQUE OF CURRENT METHODS IN HYDROLOGIC SYSTEMS INVESTIGATIONS.
Amer. Geophys. Union Trans. 45(2): 307-321.

4)
and ORLOB, G. T.
1961. NON LINEAR ANALYSIS OF HYDROLOGIC SYSTEMS. Water Resources Center,
Univ. Calif. (Berkeley) Contrib. 40, 147 pp., illus.
5) AsELTiNE, JOHN A.
1958. TRANSFORM METHODS IN LINEAR SYSTMES ANALYSIS. 300 pp., iUuS. NeW
York.
6) BEER, S.
1959. CYBERNETICS AND MANAGEMENT. 214 pp., iUuS. NeW York.
7) BELLMAN, R. E.
1961. ADAPTIVE CONTROL PROCESSES: A GUIDED TOUR. 255 pp. Princeton, N.J.
8) BODY, D. N.
1959.
FLOOD
ESTIMATION:
DIGITAL COMPUTER.
illus.
UNIT
GRAPH
PROCEDURES UTILIZING A HIGH-SPEED
Water Res. Found. Austral. (Sydney) Bui. 4, 41 pp.,
9) CHORAFAS, D. N.
1965. SYSTEMS AND SIMULATION. 503 pp. NeW York.
0) COLLINS, W. T.
1939. RUNOFF DISTRIBUTION GRAPHS FROM PRECIPITATION OCCURRING IN M:ORE
THAN ONE TIME UNIT. Civ. Engin. 9(9): 559-561.
1) CRAWFORD, N. H., and LINSLEY, R. K.
1966.
DIGITAL SIMULATION IN HYDROLOGY: STANFORD WATERSHED MODEL IV.
Tech. Rpt. 39, 210 pp. Dept. Civ. Engin. Stanford University, Calif.

(12) DAWDY, D. R., and O'DONNELL, T.
1965. MATHEMATICAL MODELS OF CATCHMENT BEHAVIOUR.
Jour. Hydraulics Div. 91(HY4): 123-139.
(13) DERUSSO, P. M., ROY, R. J., and CLOSE, C. M.
1965.
STATE
VARIABLES FOR ENGINEERS.
41
Amer. SoC. Civ. Ellgin.,
608 pp. New York.
(14) DOEBELIN, E. O.
1966.
MEASUREMENT SYSTEMS: APPLICATION AND DESIGN.
743 pp. NeW York.
(15) DOOGE, J. C. I.
1959.
A GENERAL THEORY OF THE UNIT HYDROGRAPH.
JoUI. GeophyS. ReS. 64(2):
241-256.
(16)
1965.
ANALYSIS OF LINEAR SYSTEMS BY MEANS OF LAGUERRE FUNCTIONS.
SoC.
for Indus, and Appl. Math., Jour. Control, Ser. A, 2(3): 396-408. Philadelphia.
(17)
1967.
THE HYDROLOGIC CYCLE AS A CLOSED SYSTEM.

Bull., XIII, Ann. 1: 58-68. Gentbrugge.
(18) DRAPER, C. S., MCKAY, W., and LEES, S.
1952.
INSTRUMENT ENGINEERING.
Intematl. Assoc. Sci. Hydrol.
3 V. NeW York.
(19) DRENICK, R. F.
1965.
AN APPRAISAL OF THE STATUS AND FUTURE OF SYSTEM THEORY. PrOC.

Symposium on System Theory. 380 pp. New York.
(20) EAGLESON, P. S., MEJIA-R, R., and MARCH, F.
1966. COMPUTATION OF OPTIMUM REALIZABLE UNIT HYDROGRAPHS. Water Resources Res. 2(4): 755-764.
(21) ELLIS, D. O., and LUDWIG, F. J.
1962. SYSTEMS PHILOSOPHY. 387 pp. New York.
(22) GARDNER, M. F., and BARNES, J. L.
1942. TRANSIENTS IN LINEAR SYSTEMS.

illus. New York.
(23)
V. 1 : lumped coustant systems.
389 pp.,
TRANSFORM AND STATE VARIABLE METHOD IN LINEAR SYSTEMS.
426 PP.
GUPTA, SOMESHWAR C.
1966.
New York.
(24) KoENiG, H. E., and BLACKWELL, W. A.
1961. ELECTRO-MECHANICAL SYSTEM THEORY.
(25)
KRAIJENHOFF VAN DE LEUR,
504 pp. NeW York.
D. A.
1958.
A STUDY OF NON-STEADY GROUND WATER FLOW WITH SPECIAL REFERENCE

TO A RESERVOIR COEFFICIENT. Ingenieur 70(19): 87.
1966.
RUNOFF MODELS WITH LINEAR ELEMENTS. Part III, recent trends in hydrograph synthesis. TNO^ Proc. and Inform. Note 13. The Hague. ,
(26)
(27) LEE, Y. W.
1960. STATISTICAL THEORY OF COMMUNICATIONS.
(28) LEVI, E., and VALDES, R.
1964.
509 pp., iUus. New York.
A METHOD FOR DIRECT ANALYSIS OF HYDROGRAPHS.
Jour. Hydrol. 2(2):
182-190.
TNO stands for "Organisatie Toegepast Natuurwetenschappelijk Onderzoek" [Netherlands Organization for Applied Scientific Research]. The abbreviation TNO is invariably
used, and the full title does not appear even on publications of the organization.
42
(29) LYNCH, W. A., and TRUXAL, J. G.

1961. INTRODUCTORY SYSTEM ANALYSIS. 445 pp., illus. NeW Yoik.
(30) MACFARLANE, A. G. J.
1964. ENGINEERING SYSTEMS ANALYSIS. 272 pp., illus. Reading, Mass.
(31) NASH, J. E.
1959. SYSTEMATIC DETERMINATION OF UNIT HYDROGRAPH PARAMETERS. JoUF.
Geophys. Res. 64(1): 111-115.
(32) -^
1960. A UNIT HYDROGRAPH STUDY WITH PARTICULAR REFERENCE TO BRITISH
CATCHMENTS. Inst. Civ. Engin. Proc. 17: 249-282.
(33)
1967.
THE ROLE OF PARAMETRIC HYDROLOGY. Inst. Water Engin. Jour. 21(5).

Dublin.
(34) O'DONNELL, T.
1960. INSTANTANEOUS UNIT HYDROGRAPH DERIVATION BY HARMONIC ANALYSIS.
Internatl. Assoc. Sei. Hydrol. Pub. 51: 546-557. Gentbrugge.
(35)
1966. METHODS OF COMPUTATION IN HYDROGRAPH ANALYSIS AND SYNTHESIS. Part
III, recent trends in hydrograph synthesis. TNO Proc. and Inform.
Note 13. The Hague.
(36) PAYNTER, H. M.
1952. METHODS AND RESULTS FROM M.I.T. STUDIES IN UNSTEADY FLOW. Boston
See. Civ. Engin. Jour. 39(2): 120.
(37)
1960. ANALYSIS AND DESIGN OF ENGINEERING SYSTEMS. Cambridge, MaSS.
(38) SHEN, J.
1965. USE OF ANALOG MODELS IN THE ANALYSIS OF FLOOD RUNOFF. U.S. Geol.
Survey Prof. Paper 506-A, 24 pp.
(39) SHERMAN, L. K.
1932. STREAM FLOW FROM RAINFALL BY THE UNIT-GRAPH METHOD. Engin. NeWS
Rec, 108: 501-505.
(40) SlEBERT, W. M.
1961. SIGNALS IN LINEAR TIME-INVARIANT SYSTEMS. In E. H. Baghdady, ed.,
Communication System Theory. Ch. 3: 33-57. New York.
(41) SNYDER, W. M.
1955. HYDROGRAPH ANALYSIS BY THE METHOD OF LEAST SQUARES. Amer. SoC.
Civ. Engin., Jour. Hydraulics Div. 81: 1-25.
(42)
1961.
MATRIX OPERATION IN HYDROGRAPH COMPUTATIONS.

Knoxville.
TVA ReS. Paper No. 1.
(43) STARK, L.
1968. NEUROLOGICAL CONTROL SYSTEMS. Studics in Bioengineering, 428 pp.
New York.
(44) TusTiN, A.
1957. MECHANISM OF ECONOMIC SYSTEMS. Ed. 2. 191 pp. England.
(45) WYMORE, A. WAYNE
1967. A MATHEMATICAL THEORY IN SYSTEMS ENGINEERING: THE ELEMENTS. 353
pp. New York.
(46) ZADEH, L. A., and DESOER, C. A.
1963. LINEAR SYSTEM THEORY. 628 pp. New York.
LECTURE 2:
REVIEW OF PHYSICAL HYDROLOGY
Lecture 2 is a review of physical hydrology. It might be wondered why we
bother with a review of physical hydrology since it has already been stated
that the essence of the systems approach is to ignore the details of the physics
involved. The systems approach was described as being an attempt to get
around the complex geometry and the complex physics of the hydrologie
system. If we were solely concerned with problems of identification, this attitude of ignoring the details of the system would be a reasonable one. We can
identify a system (that is, find an expression for its impulse response) without
any knowledge of physical hydrology at all. In 1965, Dooge {12) developed
a method of system identification based on the use of Laguerre functions,
which he thought might be appropriate in hydrologie problems. However, the
first application of the new method was in the problem of determining residence times in chemical engineering. This was possible because the method
was merely a method of system identification, and such methods are not by
any means tied to the hardware of the particular system being analyzed.
If, on the other hand, we are going to simulate a hydrologie system, our
knowledge of physical hydrology will be of greater importance. Such a knowledge is useful in model building because the closer we simulate the physical
reality, the better our model will be. If we build a model that is in conflict
with the physical realities, then we can hardly expect to get very good results
from such a model. The present review, therefore, will be a brief summary
of physical hydrology from the point of view of its possible use in the simulation of hydrologie systems by models of various types.
Our quantitative knowledge of physical hydrology is summarized in the
various formulas which are available in the literature. These formulas are
themselves models of the physical process which they are taken to represent.
The factors that are included in a formula and the relation between them all
involve simplifying assumptions concerning the relevant physical processes.
This lecture deals with the various parts of the hydrologie cycle in turn and
discusses some typical concepts and formulas. These are dealt with in more
detail in general reference works such as those by Linsley, Khler, and Paulhus
{82)] Chow {6)\ Soil Conservation Service {50)] and Eagleson {13), The
problem of measuring the various hydrologie quantities is discussed in publications by the World Meteorological Organization {58), the International
Association of Scientific Hydrology {25), and by Corbett {9). Some important
books and papers containing further information on physical hydrology are
included among those listed at the end of this lecture.
43
44
Precipitation
In most of our work on hydrologie systems, precipitation is taken as an
input. Consequently, we are usually not worried about the processes of hydrometeorology {13y 16y 4i)' Our main problems are those concerned with measurement and with the sampling that is inherent in any system of measurement. However, the question of snowmelt, which is on the borderline between
meteorology and the land phase of hydrology, is of interest to us. If we wish
to include snowmelt in a simulation model, then we must either know something or assume something concerning the physical processes involved {14y
52). If some of the components of our simulation models seem somewhat
crude, we may take some consolation from the fact that most of the physical
equations and formulas used in applied hydrology are equally crude. Thus the
daily snowmelt in inches is frequently computed by a formula like the following:
M=0.06(T,ean-24)
(1)
where the daily snowmelt in inches {M) is related only to the mean daily
temperature in degrees Farenheit {T). The additional effects of wind velocity
and precipitation can be allowed for by using a formula of the following type:
M= (0.029+0.0084 7+0.007 P,) (Tn.ean-32) +0.09
(2)
In the first equation, the snowmelt is related only to the mean temperature,
which is a crude way of relating the energy required to melt the snow to the
energy available from radiation. In the second formula, radiation, convection,
and conduction have all been taken into account. A more complex equation
proposed by Light {31) is derived from an eddy-conductivity equation based
on the analysis of atmospheric turbulence, and expresses the rate of snowmelt
Das:
^ = 80
Rmlog,(a/2:o)
( ft^loge(6/2:0)
(y.1 A<^vT^-{e-^W)
I
L
v\
where
D snowmelt in centimeters per second
p = density of air
fco=von Karman's coefficient (0.38)
a = elevation of anemometer in centimeters
00 = roughness parameter (0.25 cm.)
h = elevation of hygrothermograph in centimeters
u wind velocity at anemometer level in centimeters per second
Cp = specific heat of air (0.24)
r=air temperature, in degrees Centigrade, at hygrothermograph level
e=vapor pressure of air in millibars
p = atmospheric pressure in millibars
(3)
45
When we recall the nonhomogeneous nature of a watershed and the variations

in the factors involved, we become somewhat doubtful of the advantage of
using very complex equations.
Evaporation and Transpiration

Total evaporation has been defined as including water lost by evaporation
from water surfaces, moist soil and snow, together with water lost by transpiration from vegetation, in the building of plant tissue and through interception (30). The concepts involved and formulas used have been reviewed
elsewhere {18, 17, U, 45, 53, 57).
The classical formula for evaporation from open waters was that given by
Dalton {10) :
= C(6,-ea)
(4)
which related the rate of evaporation E^ from a water surface to the vapor
pressure deficit (e,e). Since then, many more complex formulas have been
derived. The Dalton formula is the simplest formula based on the mass transport approach to evaporation. If allowance is made for the windspeed, V, we
get an empirical formula of the form:
E,= {a+hV){e^-ea)
(5)
If the variation of wind with height is taken into account, more complex
formulas are obtained. Typical of these is the equation by Thornthwaite and
Holzman {^6), which is based on the logarithmic wind law and is :
'
_ 133.3(72-Fi) (61-62)
(T-459.4) log.(/i2Ai)2
,^v
Still more complex formulas have been derived, and these were evaluated
in the comprehensive Lake Hefner study {17). The study of the evaporation
of Lake Hefner was a comprehensive operation lasting several years, but
after a detailed study of the various formulas and a most careful measurement
of conditions, it was concluded that the best equation for predicting evaporation from Lake Hefner would be of the form:
^0-0.00177 7 (e^.-ea)
(7)
which is of the same form as the empirical formulas used 50 years ago and
only one step better than Dalton's original formula of 150 years ago.
An alternative approach to the subject of evaporation is the use of the
energy budget. This can be summarized in the formula:
^^=
PL(1+)
^'^
The numerator in equation 8 gives the amount of energy available for the
46
transfer of both moisture and sensible heat from the water to the air in contact
with it. It is given by the incoming shortwave radiation from the sun (Q^)
plus the energy advected into the body of water (Qa) minus the total energy
losses due to the combination of reflected shortwave radiation (Qr), longwave
back radiation (Q&), and increased energy storage in the body of water (Qw).
To express the evaporation in terms of the amount of moisture transported,
it is necessary to divide this net energy by the product of the density (p)
and the latent heat of vaporization (L) corrected by means of the Bowen's
(4) ratio (R) to allow for the transfer of sensible heat.
In 1948, Penman (39) combined the two ideas of mass transport and energy
budget to produce a combination formula which enables us to estimate the
evaporation from readily available climatic data. His basic formula is:
E.+ iA/y)H
^'- 1+(A/T)
^'^
where Ea is a measure of the aerodynamic evaporation or the evaporation

from a mass transport point of view and H is Si measure of the net energy
required for evaporation. Penman and others have refined this approach in
the past 20 years (5J.)
In the case of transpiration, we also have a wide variety of formulas of
different degrees of complexity. In many of them, a figure for cumulative
degreedays above a certain base temperature is used as a crude estimate of
the energy. Thus, we have the Hedke formula (18), which was developed for
use in irrigation work:
ET=^k (T-To)
(10)
where the cumulative value of degree-days is used as a measure of the energy
required for total transpiration. Blaney and Criddle (3) developed a number
of formulas of the same general type. In 1948, Thornthwaite (47) developed
the following empirical formula:
^
^ ^ /10T\
ET=hQ{)
(11)
which enables the monthly transpiration (ET) to be calculated from climatic

data. In the formula, T is the monthly mean temperature and 7 is a temperature efficiency index which depends on the 12 monthly mean values of temperature. The exponent a is a function of /. The result obtained must be corrected for latitude and season to allow for the variation in the hours of
sunshine. Penman (4-0) derived a formula for transpiration similar to his
evaporation formula; it is written as:
l+(A/7)
^'^'
In the above equation, the aerodynamic evaporation Ea is modified because
47
of change in the roughness factor of vegetation compared with open water

and the energy term HT is shghtly changed because of the modification in the
heat exchange occurring between the vegetation and the air. Penman found
that in practice an extremely good estimate could be obtained by applying
a coefficient to the estimate for open water evaporation :
ET=f'E,
(13)
The following formula by Turc {^9) has been widely used in studies of water
balance in Africa by French hydrologists :
(14)
ET-=
where P is precipitation and L is a temperature index.

We have, thus, a variety of formulas for evaporation and transpiration, all
of which have a physical foundation to a lesser or greater extent. The final
formulas are, however, all empirical and represent a simplification of the very
complex physics involved. In incorporating them into a simulation of the
hydrologie cycle or part of it, we are at liberty to choose the particular formula
that suits our purpose best.
Infiltration and Percolation

The soil phase in the hydrologie cycle involves the phenomena of infiltration or the entry of water through the surface of the soil, its downward percolation through the unsaturated zone and its storage in that zone.
Information on infiltration may be obtained from the results of tests with
infiltrometers, from the analysis of hydrographs from plot experiments and
from the derivation of basin indices for complete watersheds. As in the case
of other phenomena in the hydrologie cycle, a number of empirical formulas
for infiltration are available. Kostiakov {29) proposed as an empirical formula
for the amount of infiltration {F) during the period of high-rate infiltration :
F = htî^
(15a)
which is equivalent to an infiltration rate (/) of:
Horton {2Ii) proposed an exponential formula for infiltration which has

been widely used :
f=fc+(fo-fc)e-''
(16a)
48
The corresponding formula for the amount of infiltration is:

^=/c-+(^) (l-e-*0
(16b)
Philip (4^) analyzed the problem of infiltration using the principles of soil
physics and developed a series solution which can be approximated by:
F=^S'''+At
(17a)
or in terms of the infiltration rate :
/=+^
(17b)
The first term of Philip's equation is seen to be identical with the Kostiakov
equation derived empirically 25 years earlier.
Hoi tan (21) used the relationship :
f-fc = a{S-Fy
(18a)
For a value of n = 2 the equation for the infiltration rate can be written as :
/=/c-sec2 iV'^c{tc-t)2
(18b)
All of these formulas are empirical or have empirical coefficients and thus
may be considered as attempts to simulate the actual phenomena. Even if one
takes the full equation due to Phihp, to which equation 17 is an approximation, it is still a simulation of the process taking part in nature. This is because
Philip's full equation is based on the assumption that there is a perfectly
uniform soil, perfectly graded with no roots or root holes and no worms living
in the soil. For such an ideaUzed case, Philip's full equation is the most accurate of all the formulas given (except for very long elapsed times), but the
question arises whether, in view of the uncertainties in the field, it is worth
using anything more than a simple equation. We can never get away from
simulation, and it is quite fruitless to argue about one equation being approximate and another one accurate. They all involve various degrees of approximation, and our choice is a free one. The balance is one between the need for
simplicity on the one hand and for accuracy on the other.
Ground Water Flow
The physical assumptions underlying these formulas are given in such
references as (^, 7, ii, 33, 48j and 56). For one-dimensional flow in the saturated zone (that is, for the Dupuit assumptions), Darcy's Law takes the
form:
dh
=-kh-
(19)
where q is the flow per unit area, k is the hydraulic conductivity, and h is
49
the depth of saturated flow. The equation of continuity for the same conditions takes the form:
da
ax
dh
at
7^+/-=i(a;,0
(20)
where/ is the specific yield and i(Xjt) is the rate of recharge at the watertable
surface. Combination of these two equations gives:
dx L
], .Bh
+/-=(a;,)
dx J
at
(21)
Equation 21 is nonhnear, but in classical ground water hydraulics the equation

is linearized in one of two ways. Either we write:
-^^r^2+/^
dx^
dt = ^'(^')
(22)
or else we write:
2 dx^
2h
dt
^ ' ^
Ground water hydrologists have generally solved their problems on the

basis of such linear equations, which suggests that the application of linear
systems theory might be fruitful in this particular field. In view of the long
use of linear methods in hydrology, it is remarkable that a general linear approach has not been used in ground water hydraulics except recently and then
to a limited extent. Once the original nonlinear equations have been linearized,
all of linear mathematics and all of linear systems theory are available for the
solution of our problems.
Hydrologists frequently assume that the recession curve for base flow is
given by :
<4)
Q = Qoexp(^--j
(24)
This represents a more restrictive assumption than the simple one of linearity.
Equation 24 not only assumes that the ground water action is linear, but that
it acts as a single linear reservoir. Having made this assumption with regard
to the recession, there is no reason why the same assumption should not be
made in regard to the recharge of ground water and the whole ground water
system modeled by a single linear reservoir. In general, however, one can
assume linearity without restricting oneself to a single linear reservoir. If the
system is assumed to be linear, it is perfectly possible to derive a ground water
unit hydrograph just as is done for direct storm runoff.
50
Overland Flow and Channel Flow

In the case of overland flow and open channel flow, we can write down the
equation of continuity (including lateral inflow if necessary) and the dynamic
equation. By using the classical methods of open channel hydraulics, we can,
in theory at any rate, solve these equations for any particular case. Even
with high-speed digital computers, the solution of such cases, even for simple
geometry, is by no means an easy matter. Whether we use a characteristic
solution or some method of finite differences based on a rectangular network,
the computational problems are quite severe. In hydrology, the complexity of
these problems has been avoided by using approximate methods of solution,
most of which retain the continuity equation but replace the dynamic equation by some approximate relationship. This is to say that an applied hydrologist, when faced with the problem of overland flow or flood movement
in rivers, has replaced the field situation by a simplified model (^0, 59).
The fundamental problem of overland flow can be quite simply stated.
Rain falls vertically on the plane surface at the upstream end of which is
either a divide or a vertical boundary as shown in figure 2-1. If the supply
rate is constant, then for equilibrium conditions there will be a definite profile
of steady overland flow. Even this steady flow problem is not an easy one to
solve precisely. We do not know the friction laws operating in such a flow, or
the effects of lateral inflow on the velocity distribution, or what the effect
would be if infiltration were occurring simultaneously.
In tackling the hydrology of overland flow, we wish to know far more than
the profile of steady state flow. What is required is the hydrograph of nonsteady flow, which occurs due to any change in input conditions (for example,
the relatively simple case of a steady input of rain starting from initially dry
conditions) and also the nature of the recession from the steady state after the
Lateral inflow, r(x,t)
1111111111111111
-Ûpstream boundary
and dividing point
FIGURE
2-1.Overland flow.
51
cessation of input. If the process were linear, one of these results would be
sufficient to determine the hydrograph for any pattern of rainfall input. However, the phenomenon is nonhnear, and, thus, the principle of superposition
cannot be used. Every shape of input becomes a separate case and must be
handled on its own. Our interest is concentrated on three cases: (1) the rising
hydrograph for a constant input and initially dry conditions, (2) the recession
from steady outflow conditions after the cessation of input, and (3) the transition from one steady state to another when there are two different constant
supply rates in successive intervals of time.
One approximation to the overland flow problem assumes that there is, at
all times, a definite power relationship between the outflow at the downstream
end and the average detention on the surface. A large number of experiments
during the 1930^s indicated that if the equilibrium runoff were plotted against
the average equilibrium detention (that is, the storage at equilibrium divided
by the surface area) for a given experimental plot, the relationship could be
approximated by a straight line on log-log paper. This relationship apphed to
the condition when steady flow had been attained and storage was no longer
changing, that is, to the steady state solution. Horton (23) assumed that this
power relationship would hold throughout the unsteady flow phase and used
this assumption as the basis of the solution for the particular case where the
discharge was proportional to the square of the average detention.
The general assumption of a power relationship between discharge per unit
area (q) and detention or storage per unit area may be written as:
q = as'
(25)
This equation in fact replaces the full dynamic equation and is combined with
the continuity equation :
.e-.= ^
(26)
to solve the problem. Equations 25 and 26 can be combined to give:

ailcq^s^ J 1-q/q^
The integral on the right-hand side of equation 27 can be solved explicitly

for c= 1 (the linear case) and also for c = 2, 3, and 4. By suitable transformations, it can also be solved for c = | andforc = |. Horton solved the equation
for c = 2, obtaining the result :
- = tanA2(ai/V^2)
(28)
This equation has since been used for solving the overland flow problem and
designing airport installations (51). Izzard (26) carried out a series of notable
52
experiments on overland flow and proposed the use of a dimensionless rising

hydrograph and dimensionless recession hydrograph, corresponding to the
solution of equation 27 for c = 3. Because the integral in equation 27 is of
exactly the same form as the Bkhmeteff {!) varied flow function, tables of
the latter function can be used to solve equation 27 and hence the problem
of the overland flow hydrograph for any value of c which is tabulated. The
above class of solutions may be referred to as the Horton-Izzard solution. It
is not the only solution to the problem of overland flow and is given here only
as an example. The kinematic wave method has also been applied to the problem of overland flow. Both approaches are discussed in more detail in lecture 9.
Hydrologie flood routing represents an early application of the systems approach to a hydrologie problem. The full problem of flood movement in rivers
is complex, and in any case the details of the flow between the upstream and
downstream ends of the reach under examination are not of great interest.
When conditions in the whole reach are lumped, the continuity equation
becomes :
I-Q = ^^
(29)
This equation is used in all flood routing methods and is combined with some
special equation, which replaces the dynamic equation.
Among the well-established flood routing methods is the lag and route
method which assumes :
S(t)=K^Q(t+^^
(30)
that is, that the storage in the reach is proportional to the outflow taken at
some fixed time later than the time at which the storage is measured. In
another well-established routing method, the Muskingum method, the storage
is taken as being proportional to weighted values of the inflow and the outflow:
S{t)=K\:xI{t) + (l^x)Q(tn
(31)
Among other important flood routing methods is the use of the diffusion
analogy, which was introduced by Hayami about 1950 (19). This approach
was dealt with by Henderson (20). More recently, we have had the KalininMilyukov (28) method which is now widely used in Eastern Europe. This
latter method is based on the division of the reach into a number of characteristic lengths and the treatment of each of these lengths as a linear storage
element. Routing through the whole reach thus consists of routing through a
cascade of linear storage elements, and the impulse response function is the
gamma distribution. Though the gamma distribution was used by Nash (36),
53
Gray (16) ^ Reich (^5), and a number of others to represent the unit hydrograph, it was not appHed to flood routing until this was proposed by Kahnin
and Milyukov.
It is of interest that the above methods of routing floods through an open
channel are all linear methods, thus all are linear models of the actual process.
The whole subject of linear routing in open channels is discussed in lecture 9.
This brief review of physical hydrology is intended to give examples of the
formulas which summarize our quantitative knowledge of physical hydrology
and which are used in practice. In our best efforts at physical hydrology, we
still make many assumptions that are, in truth, simulations. In many of these
cases, the assumption of linearity has already been made. When such an assumption has been made, the attitude in parametric hydrology is to make the
most of the assumption.
Problems on Physical Hydrology

1. List a number of alternative definitions given for the physical phenomena involved in one particular part of the hydrologie cycle. Discuss these
definitions critically, and then list them in what you would consider to be
their order of merit.
2. Briefly describe the methods used for measuring the physical phenomena involved in some particular part of the hydrologie cycle. Discuss these
techniques critically, stating their advantages, disadvantages, and possible
improvements. How does the method of measurement used affect the definition of the physical phenomenon involved? What criteria could be used to
determine a suitable observation network for the particular phenomena
involved?
3. State what physical principles are involved in one particular part of
the hydrologie cycle. What physical formulas can be written down describing
the physical phenomena of this part of the cycle? What physical phenomena
are ignored in the formulas cited?
4. Mhat empirical formulas are used in hydrology in connection with the
phenomena discussed in question 3? What is the relationship between these
empirical formulas and any physical formulas available? What is the range of
validity of the empirical formula? What is the accuracy of the empirical
formulas?
5. What in your opinion are the most serious gaps in our knowledge of
physical hydrology? How important are these gaps from the point of view of
applied hydrology? Outline a research program which you think might help
to bridge an important gap in our knowledge of this subject, and give a rough
estimate of the cost and manpower involved.
54
Literature Cited
(1) BAKHMETEFF, B. A.
1932. HYDRAULICS OF OPEN CHANNELS.
(2) BEAR, J., ZASLAVSKY, D., and IRMAY, S.
329 pp., illus.
New York.
1968.
PHYSICAL PRINCIPLES OF WATER PERCOLATION AND SEEPAGE. UNESCO

Arid Zone Res. XXIX. Paris. 465 pp., illus.
(3) BLANEY, H. S., and CRIDDLE, W. D.
1950. DETERMINING WATER REQUIREMENTS IN IRRIGATED AREAS FROM CLIMATOLOGICAL AND IRRIGATIONAL DATA. U.S. Dept. AgF. SCS-TP-96, 48 pp.
August.
(4) Bowen, I. S.
1926.
THE RATIO OF HEAT LOSSES BY CONDUCTION AND BY EVAPORATION FROM
ANY WATER SURFACE.
Phys. Rev., Ser. 2, 27: 779-787.
(5) CHILDS, E. C.
1957. THE PHYSICS OF LAND DRAINAGE. In J. N. Luthiu, ed.. Drainage of Agricultural Lands. Agronomy 7: 1-78.
(6) CHOW, VEN TE, ed.
1964.
HANDBOOK OF APPLIED HYDROLOGY.
1418 pp., iUuS.
NeW York.
(7) COLORADO STATE UNIVERSITY.
1963.
SYMPOSIUM ON TRANSIENT GROUNDWATER HYDRAULICS.

Fort Collins.
Civ. Engin. Sect.
(8) COMMITTEE FOR HYDROLOGIC RESEARCH, TNO.
1964.
HYDRAULICS OF STEADY FLOW TO WELLS. TNO Proc. and Inform. Note 10.
The Hague.
(9) CoRBETT, D. M., and PRICE, C. H.
1943. STREAM-GAUGING PROCEDURE. U.S. Geol. Survey Water-Supply Paper 888,
243 pp. (Reprinted 1962.)
(10) DALTON, JOHN.
1802.
EXPERIMENTAL ESSAYS ON THE CONSTITUTION OF MIXED GASES; ON THE
FORCE OF STREAM OR VAPOR FROM WATER.

Phil. Soc. Mem., v. 5, pt. 2: 536-602.
Manchester (England) Lit. &
(11) DEWIEST, J. M. R.
1966. GEOHYDROLOGY. 366 pp., iUus. New York.
1965. ANALYSIS OF LINEAR SYSTEMS BY MEANS OF LAGUERRE FUNCTIONS.
SoC.
Indus, and Appl. Math., Jour. Control 2(3): 396-408.

(13) EAGLESON, P. S.
1969. DYNAMIC HYDROLOGY.
462 pp.
New York.
(14) GARSTKA, W. U.
1964. SNOW AND SNOW SURVEY. In Ven Te Chow, ed.. Handbook of Applied
Hydrology. Sec. 10, 57 pp., illus. New York.
(15) OILMAN, C. S.
1964. RAINFALL. In Ven Te Chow, ed.. Handbook of Applied Hydrology. Sec.
9, 69 pp., illus. New York.
(16) GRAY, D. M.
1962.
DERIVATION OF HYDROGRAPHS FOR SMALL WATERSHEDS FROM MEASURABLE
PHYSICAL CHARACTERISTICS. lowa State Univ., Agr. and Home Econ.

Expt. Sta. Res. Bui. 506, pp. 514-570. Ames.
(17) HARBECK, G. E., DENNIS, P. E., KENNON, F. W., and others.
1954. WATER LOSS INVESTIGATIONS: LAKE HEFNER STUDIES. U.S. Geol. Survey
Water Supply Paper 269. [Previously published as U.S. Geol. Survey
Circ. 229, 1952, as part of U.S. Navy Electron. Lab. Rpt. 237.]

(18)
55
T., chairman.
CONSUMPTIVE USE OF WATER IN IRRIGATION: PROGRESS REPORT OF THE
DUTY OF WATER COMMITTEE OF THE IRRIGATION DIVISION. Amer. SoC.
Civ. Engin. Trans. 94: 1349-1399.
HARDING, S.
1930.
(19) HAYAMI, S.
1951. ON THE PROPAGATION OF FLOOD WAVES.
Res. Inst., Kyoto University, Japan.
(20) HENDERSON, F. M.
Bul. 1: 1-16.
Disaster Prev.
1966. OPEN CHANNEL FLOW. 522 pp., illus. New York.

(21) HOLTAN, H. N.
1961c A CONCEPT FOR INFILTRATION ESTIMATES IN WATERSHED ENGINEERING.
U.S. Dept. Agr. Agr. Res. Serv. ARS 41-51, 25 pp.

(22) HoRTON, R. E.
1935. SURFACE RUNOFF PHENOMENA: PART I, ANALYSIS OF THE HYDROGRAPH.
Horten Hydrol. Lab. Pub. 101, 73 pp.
Ann Arbor, Mich.
(23)
1938.
THE INTERPRETATION AND APPLICATION OF RUNOFF PLOT EXPERIMENTS

WITH REFERENCE TO SOIL EROSION PROBLEMS. Soil Sci. SoC. Amer. PrOC.
3: 340-349.
(24)
1940.
AN APPROACH TOWARD A PHYSICAL INTERPRETATION OF INFILTRATION CAPACITY. Soil Sei. Amer. Proc. 5: 399-417.
(25) INTERNATIONAL ASSOCIATION OF SCIENTIFIC HYDROLOGY.

1965.
SYMPOSIUM ON DESIGN OF HYDROLOGICAL NETWORKS
and 68, Gentbrugge.

(26) IZZARD, C. F.
1946. HYDRAULICS OF RUNOFF FROM DEVELOPED SURFACES.
(Washington, D.C.) Proc. 26: 129-146.
(QUEBEC).
Pubs. 67
Highway Res. Bd.
(27)
1950.
TENTATIVE RESULTS ON CAPACITY OF CURB OPENING INLETS. In Surface

Drainage. Highway Res. Bd Res. Rpt. 11-B, Washington, D.C.
(28) KALININ, G. P., and MILUYKOV, P. I.
1957.
on RASCHETE NEUSTANOVIVSHEGOSYA DVIZHENIYA VODY V OTKRYTYKH

RUSLAKH [on the computation of unsteady flow in open channels.] Met.
i. Gidrologiya Zhuzurnal, 10: 10-18. Leningrad.
(29) KosTiAKOV, A. N.
1932.
[on the dynamics of the coefficients of water percolation in soils.] 6th

Comn., Internatl. Soc. for Soil Sei. Pt. A, pp. 15-21. Groningen and
Moscow.
(30) LANGBEIN, W. B., and ISERI, K. T.
1960.
GENERAL INTRODUCTION AND HYDROLOGIC DEFINITIONS. MANUAL OF HYDROLOGY: PART 1. GENERAL SURFACE WATER TECHNIQUES. U.S. Geol.
Survey Water-Supply Paper 1541-A, 29 pp.

(31) LIGHT, P.
1941. ANALYSIS OF HIGH RATES OF SNOWMELTiNG. Amer. Geophys. Union Trans.,
pt. 1: 195-205.
(32) LiNSLEY, R. K., KoHLER, M. A., and PAULHUS, J. L. H.
1949. APPLIED HYDROLOGY. 689 pp., illus. New York.
(33) LuTHiN, J. N., ed.
1957. DRAINAGE OF AGRICULTURAL LANDS.
Wis.
Agronomy 7, 620 pp., illus.
Madison,
56
(34) MEYER, O H.
1940. ANALYSIS OP RUNOFF CHARACTERISTICS. Amer. Soc. Civ. Engin. Trans.
105: 86.
(35) MusGRAVE, G. W., and HOLTAN, H. N.
1964. INFILTRATION. In Ven Te Chow, ed., Handbook of applied hydrology.
Sec. 12: 30 pp., illus. New York
(36) NASH, J. E.
1958. THE FORM OF THE INSTANTANEOUS UNIT HYDROGRAPH. General Assembly
of Toronto, Internatl. Pub 42, Compt Rend 3: 114^118. Gentbrugge.
(37)
1959. A NOTE ON THE MUSKINGUM FLOOD ROUTING METHODS. Jour. GeophyS.
Res. 64(8): 1053.
(38) OVERTON, D. E.
1964.
MATHEMATICAL REFINEMENT OF AN INFILTRATION EQUATION FOR WATERSHED
ENGINEERING.
(39) PENMAN, H. L.
1948.
U.S. Dept. Agr. Agr. Res. Serv. ARS 41-99.
11 pp.
NATURAL EVAPORATION FROM OPEN WATER, BARE SOIL AND GRASS.
Roy.
Soc. London Proc, Ser. A 193(1032): 120-145.

(40)
1963.
VEGETATION AND HYDROLOGY. Commonwealth Bur. Soils Tech. Comm. 53,

124 pp. Commonwealth Agr. Bur., Farnham Royal, England.
(41) PETTERSSEN, S VERRE.

1958. INTRODUCTION TO METEOROLOGY.
(42) PHILIP, J. R.
1957.
Ed. 2, 327 pp.
New York.
THE THEORY OF INFILTRATION: ITHE INFILTRATION EQUATION AND ITS
SOLUTION. Soil Sei. 83(5): 345-357.

(43) REICH, B. M.
1962. DESIGN HYDROGRAPHS FOR VERY SMALL WATERSHEDS FROM RAINFALL.
CER62BMR41, Civ. Engin. Sec, Colo. State Univ., 57 pp. Fort Collins.
(44) SELLERS, W. D.
1965.
PHYSICAL CLIMATOLOGY.
272 pp.
Chicago.
(45) SLATYER, R. O.
1967. PLANT-WATER RELATIONSHIP. 366 pp., illus. New York.
(46) THORNTHWAITE, C. W., and HOLZMAN, B.
1939. THE DETERMINATION OF EVAPORATION FROM LAND AND WATER SURFACES.
Monthly Weather Rev. 67(1): 4-11.
(47)
1948.
AN APPROACH TOWARD A RATIONAL CLASSIFICATION OF CLIMATE.
GeOg.
Rev. 38: 55-94.

(48) ToDD, D. K.
1964. GROUND WATER. In Ven Te Chow, ed., Handbook of Applied Hydrology.
Sec. 13, 55 pp., illus.
(49) TURC, L.
1954.
LE BILAN
D'EAU
L'COULEMENT
SOLS.
[THE
RELATIONS ENTRE LES PRECIPITATIONS,
DISTRIBUTION
OF
WATER
IN
SOILS,
BETWEEN PRECIPITATIONS, EVAPORATION AND FLOW].
L'EVAP
ET
RELATIONSHIPS
Ann. Agron. 5: 491-
596.
(50) U.S. DEPARTMENT OF AGRICULTURE.
1964. HYDROLOGY. Sec. IV, ses Nati. Engin. Handbook.
Washington, D.C.
(51) U.S. DEPARTMENT OF THE ARMY, CORP OF ENGINEERS.

1955. (A) SURFACE AND (B) SUBSURFACE DRAINAGE FACILITIES FOR AIRFIELDS.
Part 13, ch. 1 and 2. In Engineering Manual. Washington, D.C,
57
(52)
1956.
SNOW HYDROLOGY: SUMMARY REPORT OF THE SNOW INVESTIGATIONS.

pp., illus. Portland, Oreg.
437
(53) UNIVERSITY OF WAGENINGEN.

1956. INFORMAL MEETING ON PHYSICS AND AGRICULTURE. Netherlands Jour.
Agr. Sei. 4(1).
(54) VAN BAVEL, C. H. M.
1966. THE THREE-PHASE DOMAIN IN HYDROLOGY. Symposium on Water in the
unsaturated zone. Internatl. Union Geod. and Geophys. Proc. Wageningen.
(55)
1966.
POTENTIAL EVAPORATION : THE COMBINATION CONCEPT AND ITS EXPERIMENTAL
VERIFICATION. Water Resources Res. 2(3): 455-467.

(56) VAN SCHILFGAARDE, JAN
1965. TRANSIENT DESIGN OF DRAINAGE SYSTEMS. Amer. Soc. Civ. Engin. Proc,
Irrig, and Drain. Div. Jour. 91(IR3): 9-22.
(57) VEIHMEYER, F. J.
1964. EVAPOTRANSPIRATION. In Veu Te Chow, ed., Handbook of Applied Hydrology. Sec. 11, 38 pp., illus. New York.
(58) WORLD METEOROLOGICAL ORGANIZATION.
1965. GUIDE TO HYDROMETEOROLOGICAL PRACTICES. WMO-NO. 168, TP 82.
Sec. of the World Met. Organ. Geneva.
(59) YEVDJEVICH, V. M.
1964.
BIBLIOGRAPHY AND DISCUSSION OF FLOOD ROUTING METHODS AND UNSTEADY
FLOW IN CHANNELS.
U.S. Geol. Survey Water-Supply Paper 1690, 235 pp.
LECTURE 3:
REVIEW OF MATHEMATICS
If we are to develop objective methods for the identification and simulation
of hydrologie systems, sooner or later we find ourselves involved in mathematics and sometimes unfamiliar mathematics at that. The purpose of lecture
3 is to review some topics in mathematics that have been found useful in
parametric hydrology. The individual topics will appear again in subsequent
lectures when these mathematical techniques are drawn on as required. There
is no necessity to attempt to master completely the mathematics reviewed
in the present lecture.
In parametric hydrology, as in all engineering, the best strategy for the
apphed scientist is to make himself generally aware of what mathematical
tools are available but not to attempt to master them until he needs a particular piece of mathematics to solve a particular problem. Some of you may be
more interested than others in particular aspects of the mathematical foundations of parametric hydrology or in its computational aspects. Those interested
in such topics might find it useful to go through the references at the end of
this lecture in regard to the particular topic of interest and to work steadily
through the problems referring to that particular topic. Those who are not
interested in either analytical or computational mathematics need not worry
unduly about this aspect of our subject, but can accept the pragmatic view
that the techniques discussed here are well-founded and practicable. Books
which the author has found useful in respect of more than one mathematical
topic of interest in systems analysis are those by Gullemin (<?), Raven (20),
Korn and Korn (12) y and Abramowitz and Stegun (1),
Orthogonal Polynomials and Functions

The following set of functions
9o(t),gi{t),
g^(t),
gn{t),
is said to be orthogonal on the interval a<t<h with respect to the positive

weighting function w{t) if:
/ w(t)gm(t)gn{t)dt = 0,my^n
(la)
''a
[ w(t)gn(t)gn{t)=yn
'a
(lb)
where the standardization factor (jn) is a constant depending only on the

58
59
value of n. These two equations can be combined as follows:

/&
/ W{t)gm{t)gn{t)=yn-mn
'a
(Ic)
where mn is the Kronecker delta, which is equal to 1 when m equals n, but

zero otherwise.
If a function is expanded in terms of a complete set of orthogonal functions
as defined above:
A;=oo
/()= T,c,g,it)
(2)
then the property of orthogonahty can be used to show that the coefficient
(Ck) in the expansion is uniquely determined by:
c, = l [ w{t)gn{t)f(t) dt
(3)
If each of the functions gk{t) is so written that the factor of standardization

7fc is incorporated into the function itself, the set of functions is said to be
normalized as well as orthogonal. In a similar fashion, the weighting function
w(t) can for convenience be incorporated into the function gk(t).
At some time or other, most engineers come in contact with Fourier series,
which are the basic classical orthogonal functions in engineering mathematics.
The vast majority of functions in engineering analysis and synthesis can be
represented by an expansion of the form:
f{t) = J ao+ 22 (ak cos kt+bk sin kt)
(4)
It can be shown (8) that sines and cosines are orthogonal over a range of
length 2^ with respect to the weighting function 1 and with a standardization
factor Tras follows:
/a+27r
cos(m) COS (nt) dt = IT'mn
(5a)
sin(mO sm(nt)dt=zLTr'mn
(5b)
cos(mO sm{nt)dt = 0
(5c)
-a+27r
' n.
ô+27r
Because the terms of the Fourier series have this property of orthogonality.
60
the coefficients au and hu in equation 4 can be evaluated from:

1 r
(^k = f(t) cos{kt) dt
(6a)
1 r fit) sm(kt) dt
(6b)
h=-
From a systems viewpoint, the significance of equation 4 is that the function is decomposed into a number of elementary signals, each of which is
sinusoidal in form. For mathematical manipulation, it is frequently more
convenient to write the expansion given in equation 4 as a complex Fourier
series :
00
/(0= Z Ckexpiikt)
(7)
A:=-oo
For this exponential form of the Fourier series, the property of orthogonality
is expressed as:
.a+2T
/.
expli{mn)t2'dt = 2'jr'mn
(8)
where mn is the Kronecker delta, that is, is equal to 1 when m = n, but zero
otherwise.
We can determine the complex coefficients in equation 7 as:
1 r
Cifc = y expi-ikt)f{t) dt
(9)
If the function being expanded is a real function, then the coefficients ak and
bk in equations 4 and 6 will be real, whereas the coefficient Ck in equations 7
and 9 will be complex. The relationships between the coefficients are given by:
Ck = }^(ak-ibk)
(lOa)
c^k='i4{ak+ihk)
(10b)
Though Fourier series are widely used in systems engineering, they are not
the only types of orthogonal functions which are of use. There are three
classical cases of orthogonal polynomials. These are (1) the Legendre polynomials, which are orthogonal on a finite interval with respect to a unit
weighting function; (2) the Laguerre polynomials, which are orthogonal on a
semi-infinite interval with respect to the weighting function exp{~t); and
(3) the Hermite polynomials, which are orthogonal on an interval infinite in
both directions with respect to a weighting function exp( t^). Of these, only
61
the Laguerre functions have been used in parametric hydrology. Their definition can be written as:
/
exp{-t)Lm{t)Ln{t)dt = mn
(Ha)
It can be shown that the polynomials satisfying the above relationship are
given by:
in)=i: (-!)*(*")
(12)
By incorporating the weighting factor in the Laguerre polynomial, we can

define a Laguerre function <)n as:
<i)n(0=exp(-^V(0
*n
(13a)
p-tl2fk
= Z (-i)*a) -rr
(^3^)
which will obey the simple relationship:

(t>m(t)<l>n{t)dt=mn
(Hb)
which is an alternative form of equation Ha.

It can be seen from equation 13b that a Laguerre function can be expressed
as a series of gamma distributions with integral exponents. Therefore, any
function can conveniently be expanded through the medium of Laguerre
functions in terms of a series of gamma distributions with integral exponents.
This is of interest because of the use of the gamma distribution (not necessarily with an integral exponent) to simulate system responses in hydrology.
So far we have been talking about functions whose arguments are continuous and which are orthogonal under the operation of integration. In hydrology,
our data are frequently defined only at certain discrete points or as averages
over certain intervals so that the data are not available in continuous form.
Under these circumstances, it is necessary to use discrete rather than continuous mathematics. Unfortunately, most engineers are trained in continuous
mathematics and find some difficulty in going over to the discrete analogs of
the continuous formulas and methods. Instead of defining orthogonal functions as in equation 1, we can define discrete functions to be orthogonal if:
=6
X W{s)gm{s)gn{s) =nmn
8=a
where s is a discrete variable.
(14)
62
The Fourier approach can be apphed to a discrete set of equally spaced

data as well as to continuous data (see 'Time Series Analysis of Discrete
Data/^ lecture 6). The method of harmonic analysis or trigonometrical interpolation is based on the orthogonaUty under summation of the sines and
cosines of (2Il/N'ks). Apart from the special case of harmonic analysis, discrete orthogonal functions are not discussed to any great extent in the mathematical literature.
If an attempt is made to apply Laguerre functions to discrete data, it is
found that the Laguerre functions are not orthogonal under summation. It
was found that the discrete analog of the Laguerre function defined by equation 13b was:
fn(s) = {y2)^^+-+^y^ f (-i)Hna)
(15)
The polynomial in equation 15 is a special case of the Meixner polynomials.

Comparison of equation 15 with the corresponding equation 13 for the continuous case reveals a number of significant differences. The weighting function exp( /2) in equation 13 has been replaced by the weighting function
(3^)^/2 in equation 15, and the polynomial term t^/k\ in equation 13 has been
replaced by (I). If allowance is made for the difference in the operations in
the continuous and discrete cases, these terms are seen to correspond. Thus,
exp(t) may be defined as the function which differentiates into itself; similarly, the function 2' is a function which forward differences into itself. The
differentiation of t^/kl gives t^-'^/(kl) ! while the forward differencing of (I)
gives (;kî).
Further information concerning Fourier series and orthogonal polynomials
can be found in the references at the end of the lecture; notably in Guillemin
(8), Hamming (9), Hildebrand (10) ^ Lanczos (I4.), and Rainville (17).

Fourier and Laplace transforms have a number of applications in the linear
theory of hydrologie systems. They are useful in the solution of linear equations in dealing with the operation of linear systems and particularly in analyzing the transient behavior of systems. In addition, when the moments of
functions are used to characterize the functional relations between the input
and output of a system, Fourier and Laplace transforms can be used to determine the moments of given functions.
Transformation of the original function is made to simplify the mathematical procedure. On first attempting to master the techniques of Fourier and
Laplace transforms, the engineer may think that very little simplification is
63
achieved. However, once mastered, the techniques are extremely useful,

particularly since the Laplace transforms and, to a lesser extent, the Fourier
transforms are tabulated like logarithms or trigonometrical functions. By
using the Laplace transform, it is possible to transform an ordinary linear
differential equation with constant coefficients into an algebraic equation
which is far easier to solve. It is also possible to convert a partial differential
equation into an ordinary differential equation, again achieving a tremendous
simpHfication in the type of problem to be solved. Of course, these simpHfications are made at the cost of having to understand Laplace transforms.
The Fourier transform is particularly useful in the analysis of the transient
behavior of stable systems. From one point of view, the Fourier transform
may be looked on as a limiting form of a Fourier series. The latter apply to
functions that are periodic outside the interval of integration and consist of
an infinite series in which each term refers to a definite discrete frequency. If
the interval of integration is increased indefinitely, the series will be replaced
by an integral as follows :
f(t) = r F() expiit) do3

27r J_^
(16)
which corresponds to equation 7 with F() corresponding to c/b, with integration replacing summation, and with the term arising from the standardization
constant (211) appearing in the equation of the series instead of appearing in
the equation for calculating the coefficients. Just as the coefficients Ck in
equation 7 can be obtained from equation 9, so the function F() can be
obtained from:
F(c^) = r f(t) exp(-icoO dt
(17)
It would be equally permissible to introduce the standardization constant 211

in equation 17 and omit it from equation 16, or even to introduce the square
root of the factor into each of the equations.
Instead of looking on equation 17 as a limiting form of equation 9, it is
possible to consider it merely as the equation defining the transformation of
f{t) from the time domain to the frequency domain. Equations 16 and 17
have the advantage that, unhke equations 7, 8, and 9, they are not confined
to periodic phenomena. This advantage, however, is offset by the fact that
whereas equation 7 enables us to evaluate the function to a high degree of
accuracy by knowing the values of Ck, equation 16, which represents the inversion of the Fourier transform, is not by any means as easy to handle.
If the system we are examining is not stable, or if the functions involved
do not fulfill certain other conditions, then the Fourier transform not of f{t)
itself, but oif(t)e-'^y where c is a real number. Making this change gives us
64
the bilateral Laplace transform of the function:

FB{s)=\Bf{t)']
(18a)
= 7^[6-y(0]
(18b)
= [
(18c)
e-''f(t)e-^'^''dt
= f f{t)e-^'dt
(18d)
As ordinarily used, the Laplace transform is only defined between zero and
plus infinity, and virtually all tables are for this unilateral Laplace transform. In this form we have:
^(5)=XC/(0]
(19a)
= j f(t)e-^'dt
(19b)
'o
Equation 19b is the Laplace transform equivalent of equation 17 above.

The Laplace transform can be inverted to give the original function in the
same way as equation 16 by using the expression:
1
i.C+100
/(0=/
F{s)e'^ds
(20)
Again equation 20 is diflScult to solve, but must be used unless the function
F(s) can be found in a set of Laplace transform tables. Numerical inversion
of the Laplace transform is quite diflScult and involves the use of orthogonal
functions to represent the Laplace transform and the inversion of these functions term by term.
For discrete functions, the Laplace transform must be replaced by the
Z-transform. This can be written as:
ZUinT)2 = \UinT)(i-nT)2
(21a)
n=oo
= 'ZfinT)e-^'
(21b)
= Z/(nr)z-
(21c)
n=0
where
e'^'^Z
(21d)
This discrete transform has properties similar to those of the Laplace transform and has been tabulated.
Further information on transforms and their use in systems analysis can be
65
found in the following. Aseltine {) Doetsch (4), Jury (11), and Papoulis
(16). Extensive tables of transforms are available in Erdelyi {5) and Roberts
and Kaufman {21 ).
Differential Equations
Ordinary differential equations are differential equations in a single variable. If we are dealing with a lumped system, with lumped inputs and outputs,
then we will have only ordinary differential equations to handle in which the
single variable will be time. Ordinary differential equations are classified in
respect to their order and degree. The order of a differential equation is the
order of the highest derivative present in the equation. The degree of the
equation is the power to which the highest derivative is raised. A linear equation must of necessity be of the first degree because otherwise there would be
an essential nonlinearity and the principle of superposition would not apply.
In a linear differential equation, all the derivatives in the equation must be
to the first power and their coefficients must not be functions of the dependent
variable. Thus, if we have an ordinary differential equationor system of
ordinary differential equationswhich describes the dependent variable (y)
and its derivatives with respect to the independent variable (t) as functions
of the independent variable (t), then there is no restriction on the order of the
derivatives but each derivative must appear only to the first power, and, in
addition, the coefficients of the derivatives cannot be functions of y but may
be functions of t. The general form of such an equation is:
ao(t) 5+
^-^S+
an{t)y = xit)
(22)
If the coefficients are neither functions of y nor of t but are constants, then we
have an ordinary differential equation with constant coefficients given by:
d^v
^+
d^v
an-k -^+
any = x(t)
(23a)
Equation 22 could represent the operation of a lumped linear system, but for
equation 23 to represent the operation of a system, the system would have to
be both linear and time-invariant.
Since our starting point in systems analysis is the study of lumped, linear,
time-invariant systems, we will first be concerned in our analyses with the
solution of linear ordinary differential equations with constant coefficients
such as equation 23a. An alternative form for the latter equation is:
D^y+aiD^-'y+
aiD^-^y+
any = x{t)
(23b)
where D is the differential operator. This may also be written as:

p{D)=x(t)
(23c)
66
An equation such as 23 with a function of t on the right-hand side is said

to be nonhomogeneous and is more difficult to solve than a homogeneous
equation where the right-hand side is zero.
In accordance with the principle of solving simple problems first, the first
step is to look at the homogeneous equation :
p(D)y = 0
(24)
and postpone solution of the full nonhomogeneous equation until a solution

of the homogeneous equation has been found. The classical method of solving
this equation is to assume that the solution is made up of terms of the form:
y = C'exp(st)
(25a)
Any value of s which satisfies :

p{s)=0
(25b)
where p(s) is the same polynomial as p{D) in equation 24, will give a solution
of equation 24. If the original equation is of the n*^ order, then there will be
n roots, real or complex, for equation 25a. Consequently, the general solution
of equation 24, which is known as the complementary function, is given by:
n
y= ^Ckexp(skt)
(25c)
k=l
Real values of Sk give rise to exponential terms and complex values of Sk to

sinusoidal terms. In hydrologie systems which are heavily damped, the roots
are usually negative and real so that the solution consists of a series of exponentials with negative arguments. The n unknown constants Ck are obtained
from the boundary conditions.
Having solved the homogeneous equation, we now move on to the problem
of solving the nonhomogeneous equation. If a particular solution of the nonhomogeneous equation can be found:
y = yp(t)
(26)
then the complete solution of the nonhomogeneous equation is given by :

n
y = yp(i)+ J2ckexp(skt)
(27)
;=l
in which the first term of a particular integral will satisfy the right-hand side
of the equation, and the second term or complementary function will satisfy
the boundary conditions.
The solution of ordinary differential equations, such as equation 23, can be
greatly facilitated by the use of the Laplace transform. By taking the Laplace
transform of the equation and using the rules for the Laplace transform of a
derivative, we obtain an algebraic equation for the variable s in which the
67
boundary conditions are automatically incorporated. If the function on the

right-hand side of the equation is simple, its Laplace transform may be included. If not, it may be replaced by a delta function and the solution for this
case obtained. The solution for the actual right-hand side is then obtained by
convoluting the function on the right-hand side of the original equation with
the solution obtained by using a delta function. If the system is a complex
one, there may be derivatives on the right-hand side of the equation, and the
use of the delta function may require some caution and a mastery of its
manipulation.
If the system has distributed rather than lumped characteristics, then its
operation will be described by a partial differential equation. Most of the
partial differential equations encountered in engineering analysis are of the
second order. For one space dimension, the general second order homogeneous
linear equation with constant coefficients is given by:
a
dx^
\-b
dxdt
\-c- = d\-e-+jy
df
dx
u
{ô)
The first thing to determine about a partial differential equation is whether

it is hyperbolic, parabohc, or elliptic in form. This depends on whether the
discriminate62_4acisrespectively greater than, equal to, or less than zero.
Hyperbolic and parabolic partial differential equations correspond to problems of propagation (in both directions respectively), whereas elliptic differential equations represent the way in which the condition around the boundary
effects the interior of a space. The appropriate types of boundary conditions
are different for the three different types of equations.
Further details on the subject of differential equations and their solution
can be found in references by Lambe and Tranter {13), Fox (7), and Sneddon
{22).
Matrices
Matrices are essentially mathematical shorthand for representing arrays of
elements. A matrix is an array or table of numbers. Thus, we define the matrix
A as:
Cl21 0^22
(2n
A=
_(^ml ^m2 Clmn_
This matrix, which has m rows and n columns, is referred to as an m by n

matrix.
Matrix algebra tells us what rules should be used to manipulate such arrays
68
of numbers. If a matrix C is composed of elements each of which is given by

adding the corresponding elements of matrix A and matrix B, that is:
Cij = aij+hij
(29a)
then matrix C is said to be the sum of the two matrices A and B, and we write:
C = A+B = BÂ
(29b)
Matrix multipHcation is defined as the result of the operation:

CÂ'B
(30a)
where the elements of C are defined as:

Crt= ^rsh.t
(30b)
that is to say, the element at the intersection of the r*^ row and the t^^ column
in the C matrix is obtained by multiplying, term by term, the r*^ row of the
A matrix by the *^ column of the B matrix and summing these products.
This definition implies that matrix A has the same number of columns as
matrix B has rows. It must be remembered that matrix multipHcation is in
general noncommutative, that is:
A'B9^B'A
(30c)
A certain amount of nomenclature must be learned in order to be able to

use matrix algebra. A square matrix with the number 1 on all points of the
principal diagonal (that is, the one from top left to bottom right) and zero
on all the off-diagonal points is known as the unit matrix. It serves the same
function as the number 1 in ordinary algebra; it can be verified that multiplication of a matrix by the unit matrix gives the original matrix. A diagonal
matrix is one in which the elements on the main diagonal are nonzero, but all
the other elements are zero. An upper triangular matrix may have nonzero
elements on the principal diagonal and above, but only zero elements below
the main diagonal; similarly, a lower triangular matrix has nonzero elements
elements in the principal diagonal and below it, but only zeros above the
diagonal. The transpose A^ of a matrix A is the matrix which is obtained
from it by replacing each row by the corresponding column and vice versa.
The inverse of a matrix A-^ is the matrix which when multiplied by the original matrix A gives the unit matrix /, that is:
A'A-^ = A-^'A=I
(31)
A matrix will only possess an inverse if it is square and nonsingular, that is,
if its determinant is not equal to zero. The transpose of the inverse of a matrix
is referred to as the reciprocal matrix. A matrix is said to be orthogonal if its
inverse is equal to its transpose, that is:
A'' = A-^
(32a)
69
which is equivalent to:

(32b)
and to :
(32c)
The individual rows and columns of a matrix may be considered as row
vectors. Thus, the row vector which consists of a single row is really a matrix
of size 1 by n, whereas the column vector which consists of a single column is
a matrix of size m by 1. Two compatible vectors can be combined to give
either an inner product or an outer product. This is illustrated next for a
vector and its transpose.
The transpose of a row vector will be a column vector and vice versa. Consider a vector a which has n rows and one column; its transpose a^ will have
one row and n columns. If we premultiply a by a^ we obtain:
a2
aâ=[ai, 2
(33a)
c^n]
(33b)
= ai2+a22+.
(33c)
so that the result of the multiplication is a one by one matrix, that is, a scalar.
This is known as the inner product. The outer product is obtained by postmultiplying a by a^:
ai
2
[iz
an]
(34a)
Since this is the product of an nXl matrix and a iXn matrix, the result is
Siii nXn matrix as follows:
ai^ aia2
aâi a'?
cuan
a2an
(34b)
a-a^ =
ntti anC2
CirT
70
The comparison of equations 33 and 34 is a good illustration of the fact that

the multiplication of vectors is not commutative.
A set of simultaneous equations is represented in matrix form by:
Ax = b
(35)
where A is the matrix of coefficients, x is the vector of unknowns, and h is the

vector of the right-hand sides of the simultaneous equations. If it is required
to solve the problem for different sets of values on the right-hand side, the
most convenient method is to obtain the inverse of the coefficient matrix and
write the solution as:
x = A-%
(36)
A matrix only has an inverse if it is square and nonsingular; therefore,

equation 34 can only be written if the coefficient matrix is square. This is
nothing more than the old criterion that the number of equations must be
equal to the number of unknowns in order to obtain a direct solution. If,
however, only one set of equations is being solved, there are more efficient
computational routines. From the point of view of actual computation, a
matrix may be nonsingular but may still give rise to difficulty because the
equations are ill-conditioned and the matrix is almost singular so that the
numerical results may be unreliable. Special computer programs are available
for the inversion of matrices and for the solution of simultaneous equations.
Further information on matrices and their use is to be found in publications by Guillemin (<5), Raven (^0), Bickley and Thompson (5), and Wade
Numerical Methods
Because we deal with data and numbers rather than functions, the systems
hydrologist must have a firm grasp of numerical methods. Because of the
complexity of the systems with which he deals, most of his problems will require a solution on a digital computer. The various stages of the solution of
a problem using a computer may be grouped as follows :
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Problem identification
Mathematical description
Numerical analysis
Computer program
Program checkout
Production runs
Interpretation
It is outside the scope of these lectures to discuss these various stages. Nevertheless, those interested will be able to follow up any particular topic in the
references by Hamming (9), Hildebrand (10) ^ McCracken and Dorn (5),
71
Ralston (18), and Ralston and Wilf (19). In addition, a number of the problems at the end of this lecture and at the end of other lectures in the series
will give practice in the solution of problems involving numerical methods.
Problems on Mathematics
Orthogonal polynomials and functions
1. Find the Fourier cosine and Fourier sine coefficients for the expansion
of a number of the functions of a continuous variable given in Appendix
table 1. From these determine the Fourier exponential.
2. Find the coefficients for the expansion of a number of the functions
shown in Appendix table 1 in terms of Laguerre functions. Compare the results with those obtained in question 1 and comment on the difference.
3. Find the harmonic coefficients for the expansion of a number of the
functions of a discrete variable shown in Appendix table 2. What is the difference between the expansion of a function of a continuous variable by means
of a truncated Fourier series and the expansion of the same function by the
harmonic analysis of the function of a discrete variable obtained by sampling
the function of a continuous variable at the same number of points?
4. Determine the harmonic coefficients for the discrete set of values obtained by sampling one or more of the functions of a continuous variable
given in Appendix table 1. What is the effect of the frequency?
5. In the case of a function which is zero outside a certain Umited range,
what is the relationship between the Fourier exponential coefficient and the
moments of the function about the time origin?
6. Find the Fourier transform or Laplace transform of a number of the
functions given in Appendix in table I.
7. Show that the explicit form given in either the Laguerre or Hermite
polynomials is identical to the Rodriguez form.
8. The impulse response of a given system may be represented by function
11 in Appendix table 1 and the input to the system may be represented by
function 12 in Appendix table 1. Find the output from the system (1) by a
direct convolution and (2) by means of the Laplace transform.
9. If the r*^ moment of a function about the origin is given by
^
(n-1)!
and the function is zero for negative time, find the function.
10. Use the Z-transform to find the function obtained when a Meixner
polynomial of degree m is convoluted with a Meixner polynomial of degree n.
72
Differential equations
11. A number of unequal linear reservoirs are cascaded, that is, the outflow from one is the inflow to the next. Write the differential equation for the
outflow from the last reservoir in terms of the inflow to the first reservoir
and the storage constants of the individual reservoirs. What is the form of
the solution to this general equation? What is the form of the result if the
linear reservoirs are all equal?
12. The following equation is the impulse response of a given linear
system.
h{t) = (Cit+C2t'+Czt') .exp(-0
Draw two alternative arrangements of equal linear storage elements of unit
storage delay time which would have the same impulse response as the given
system. Then derive the differential equation for the response y{t) of the
given system to any given inflow x(t).
13. Find the solution of the following equation
'jt+ihh'
Does the result hold for all values of n? What is the relationship between
this result and the result obtained in question 11 forn equal linear reservoirs?
14. Solve the linear wave equation for a semi-infinite channel for zero
initial conditions and a given condition at the upstream end. What would be
the solution if only the first-order terms on the right-hand side of the equation
were retained? What would be the solution if only the second-order terms on
the left-hand side of the equation were retained? What type of flow is represented by these two solutions?
15. If in the linear wave equation the value of b and c are zero or of such
magnitude that the second and third terms can be neglected, what form does
the equation take, and what is the solution for the boundary conditions given
in problem 14? How does the form of this solution differ from the solutions
found in problem 14?
Matrix methods
16. Write out the set of simultaneous equations relating the ordinates of
the outflow hydrograph to the ordinates of the input hydrograph and the unit
hydrograph. Express this set of equations in matrix form in two alternative
ways. Give the matrix formulation of the direct solution and the least squares
solution for the unit hydrograph ordinates.
17. If the volumes of effective rainfall are given by function 6 in Appendix
table 2 and the ordinates of the unit hydrograph by function 8 in Appendix
table 2, use the matrix formulation to write down the ordinates of the outflow
73
hydrograph. Rework the problem with the volume of the unit hydrograph
made equal to unity.
18. What maximum runoff would be predicted for the effective rain and
the unit hydrograph shown in Appendix table 3.
19. The input to a linear system is given by function 3 on Appendix table
2 and the output from the system is given by function 4 in Appendix table 2.
Find the pulse response of the system by means of matrix inversion.
20. If the output of the system in problem 19 was taken as function 5 in
Appendix table 2, find the pulse response indicated by this output both by
the ordinary matrix method and by the least squares method.
Numerical methods
21. List several methods for numerical quadrature of a given function.
Draw a flow diagram for the application of one of these methods to the quadrature of one of the continuous functions on Appendix table 1, using either a
desk calculator or a digital computer. Give reasons for chosing the particular
quadrature method.
22. Develop a flow chart for a general computer program for determining
the coefficients in any orthogonal expansion of any given function. Write the
computer program for a section of the flow chart.
23. Write a computer program for matrix inversion and apply it to the
solution of problem 19.
24. Develop a flow chart for the derivation of a unit hydrograph from
records of total rainfall and total runoff. Write one section of the computer
program.
25. Discuss the methods available for the numerical solution of the linear
wave equation. Write out the finite difference scheme for solving the equation
by one of these methods and discuss how the boundary conditions would be
handled. What problems would you expect to encounter in computing according to the chosen method?
Literature Cited
(1) ABRAMOWITZ, M., and STEGUN, I. A., ed.
1964.
HANDBOOK OF MATHEMATICAL FUNCTIONS. U.S. Nati. Bur. Standards,

Appl. Math. Series 55, 1046 pp.
(2) ASELTINE, J. A.
1958. TRANSFORM METHODS IN LINEAR SYSTEMS ANALYSIS. 300 pp., illus. NcW
York.
(3) BiCKLEY, W. G., and THOMPSON, W. A.
1964. MATRICES: THEIR MEANING AND MANIPULATION. 168 pp. New York.
(4) DOETSCH, G.
1961. GUIDE TO THE APPLICATIONS OF LAPLACE TRANSFORMS. 255 pp. NeW York.
(5) ERDELYI, A., ed.

1954.
TABLES OF INTEGRAL TRANSFORMS.
V. 1.
NeW York.
74
(6) FoDOR, G.
1965. LAPLACE TRANSFORMS IN ENGINEERING.
(7) Fox, L.
Hungarian Acad. Sei.
Budapest.
1962. NUMERICAL SOLUTION OF ORDINARY AND PARTIAL DIFFERENTIAL EQUATIONS.

509 pp. Oxford, England.
(8) GULLEMIN, E. A.
1949.
(9)
THE MATHEMATICS OF CIRCUIT ANALYSIS.
590 pp., iUus.
Cambridge, Mass.
HAMMING, R. W.
1962. NUMERICAL METHODS FOR SCIENTISTS AND ENGINEERS. 411 pp. NeW York.
(10) HiLDEBRAND, F. B.
1956. INTRODUCTION TO NUMERICAL ANALYSIS. 511 pp. NeW York.
(11) JURY, E. I.
1964. THEORY AND APPLICATION OF THE Z-TRANSFORM METHOD. 330 pp. NeW
York.
(12) KoRN, G. A., and KORN, T. M.
1961. MATHEMATICAL HANDBOOK FOR SCIENTISTS AND ENGINEERS. 943 pp., iUuS.
New York.
(13)
and TRANTER, C. J.
DIFFERENTIAL EQUATIONS
London.
LAMBE, C. G.,
1961.
(14)
LANCZOS, C.
(15)
MCCRACKEN, D. D.,
(16)
PAPOULIS, A.
1957.
1964.
1962.
RALSTON,
(19)
RALSTON,
1965.
1960.
ENGINEERS
AND
APPLIED ANALYSIS. 539 pp. Pitman, London.

and DORN, W. S.
NUMERICAL METHODS AND FORTRAN PROGRAMMING.
THE FOURIER INTEGRAL AND ITS APPLICATION.
(17) RAINVILLE, E. D.
1960. SPECIAL FUNCTIONS.
(18)
FOR
365 pp.
SCIENTISTS.
457 pp.
327 pp.
372
pp.
NeW York.
NeW York.
New York.
A.
A FIRST COURSE IN NUMERICAL ANALYSIS.
578 pp.
A., and WILF, H. S.

MATHEMATICAL METHODS FOR DIGITAL COMPUTERS.
New York.
V. 1, 293 pp.
NeW
York.
(20) RAVEN, F. H.
1966.
(21)
MATHEMATICS OF ENGINEERING SYSTEMS. 524 pp. NeW York.

and KAUFMAN, H.
TABLE OF LAPLACE TRANSFORMS. 365 pp., illus. Philadelphia.
ROBERTS, G. E.,
1966.
(22) SNEDDON, I. A.
1957.
(23)
ELEMENTS OF PARTIAL DIFFERENTIAL EQUATIONS.
327 pp.
NeW York.
WADE, T. L.
1951. THE ALGEBRA OF VECTORS AND MATRICES.
189 pp.
Reading, Mass.
LECTURE 4:
CLASSICAL METHODS OF RUNOFF PREDICTION
The Outflow Hydrograph
The purpose of lecture 4 is to review the classical methods of runoff prediction as used by applied hydrologists and to reformulate these methods in
systems terms. Most of the methods were derived during the golden age of
classical hydrology between 1930 and 1945. When some of these methods are
looked, at from a systematic point of view, the assumptions stand out more
clearly, and both the limitations and the full range of applicability of the
methods are revealed. In many cases, the scope of the methods is considerably
wider than would appear from the classical formulation of the method.
Classical hydrology paid a great deal of attention to the runoff hydrograph
in an effort to determine how it could be predicted. Figure 4-1 is taken from
a contribution by Hoyt (18). He talks of five phases in the runoff cycle;
four of these are illustrated in figures 4-1 to 4-4. The first phase relates to the
end of a dry period when the streamow is relatively low, most of it being
supplied by base flow (Qb) from ground water storage. During this phase,
the soil moisture will have been reduced by evaporation (E) and transpiration (T) so that a substantial field moisture deficit will exist. If the dry period
has been very long, the rate of transpiration may be severely reduced below
the potential rate due to the drying out of the soil and the lowering of the
water table. Phase 2 of Hoyt^s runoff cycle relates to an initial period of rain
and is shown on figure 4-2. If the initial rain (P) is light, the amount infiltrated (F) will not be sufficient to make up the field moisture deficit and
hence, no recharge to ground water () will occur. During this second phase,
a portion of the rain will be intercepted by vegetation (F) or stored in depression storage (D).
T
Xi
/v?::-.::/;. R.-.;-. :.-.
FIGURE
4-1.Phase of low streamflow.

75
76
FIGURE
4-2.Phase of initial rainfall.
The third phase, shown in figure 4-3, is associated with the continuation of
rain for some time. If this occurs, the storage in the surface depressions (D)
will be satisfied and overland flow (Qo) will occur; similarly, if the infiltration
into the soil is sufficient to fill the soil moisture storage (S), then recharge
(R) to the ground water will occur. The streamflow will rise relatively rapidly
due to overland flow (Qo) and any return of interflow (Qi) to the stream.
Subsequently, there will be a more gradual increase in streamflow due to
outflow from the ground water reservoir (Q^), which is being recharged by
gravitational soil water (). When the general conditions are favorable to
rainfall, there is a high relative humidity and both evaporation and transpiration tend to be reduced. In the analysis of conditions during prolonged rainfall, evaporation and transpiration are frequently neglected. In this third
I t IM 1 1 M i i I IN 1
FIGURE
4-3.Phase of prolonged rainfall.
77
phase, the rapid rise of the level in the stream channels due to overland flow
and interflow may result in increased bank storage, that is, a recharge of the
ground water close to the stream as a result of the increased streamflow.
Hoyt characterized the fourth phase as a period when rainfall has continued
sufiiciently long and with sufficient intensity so that all available natural
storage has been satisfied. This condition rarely occurs in natural watersheds
of any appreciable size. In the case of small watersheds, both urban and rural,
the storage may be satisfied and the condition reached where the rate of runoff
is equal to the supply rate. This phase is not separately illustrated but is
similar to the third phase shown in figure 4-3.
The fifth phase described by Hoyt is illustrated in figure 4-4. It is the condition when the rain has ceased, but suflftcient time has not elapsed for channel
storage and surface retention to be depleted to the level at which they were
during the first phase. During this fifth phase, evaporation (E) and transpiration (T) may be considerable because the plentiful supply of moisture
allows evaporation to take place at almost the potential rate. Streamflow
will decline but only gradually as surface storage, channel storage, and ground
water storage are drawn upon in turn. This fifth phase is illustrated on the
last line of figure 4-4.
We might argue about the details of this particular picture of the runoff
cycle, but not about its general nature. How does this picture compare with
the systems view of the same phenomena? Can Hoyt^s approach interpreted
from a systems point of view? At first glance there seems little in common between the classical picture of figure 4-1 to 4-4 and the systems diagram shown
in figure 1-8 (p. 16). On closer examination, however, we realize that the two
can be related to one another. In figures 4-1 to 4-4, the channel storage and
the storage in the soil above the water table are shown pictorially; in figure
1-8 the same storages are represented by rectangular boxes. Hoyt^s classification and illustration of the phases of the runoff cycle are based on two inputs.
Qb
FIGURE
Qb
4-4.Phase of declining streamflow.
78
one of precipitation and the other of potential evaporation and transpiration;

these are also the essential inputs of figure 1-8.
Classical hydrology, as exemplified by Hoyt^s analysis of the runoff cycle,
makes the assumption that there is either rainfall and no transpiration, or
else transpiration and no rainfall. If this assumption is permitted in systems
hydrology, as in classical hydrology, then the task of the systems hydrologist
is greatly simplified. Instead of dealing with multiple inputs, it is possible to
deal with alternating inputs; thus we might consider the precipitation and
the potential evaporation for a given catchment as analogous to controls on
a storage tank operated in such a way that when one valve is open, the other
is shut and vice versa. While a complete model would have to take care of
simultaneous multiple inputs, the first approximation could follow the classification of Hoyt.
In the systems formulation, it is only necessary to use two phases. The first
phase would be the rainless period. The initial storage in the different parts
of the watershed would be determined by the previous history of the system.
The variation in that storage would be determined by the natural recession
of storage plus the effect of potential vapotranspiration on the soil moisture.
The second phase would be the rainy period. The initial condition would be
set by the history of the system during the previous rainless period in which
evaporation and transpiration would be neglected leaving precipitation as the
only input.
The decomposition of the total hydrograph into components is shown in
figure 4-5, which is based on a figure by Linsley, Khler, and Paulhus {25).
In figure 4-5, the total hydrograph has been drawn on semilog paper. The
ground water recession is taken to be exponential, thus giving a straight line
on this plot. The exponential recession is continued back from A to B, and
B is then joined to the start of the rise of the hydrograph. When the assumed
ground water flow is subtracted from the total hydrograph, the hydrograph
of surface runoff plus interflow plotted in figure 4-6 is obtained.
Again the straight line recession may be extended back from C to 2), and
the interflow separated out leaving the surface runoff. Thus, the total hydrograph has been divided into three elementsground water flow, interflow,
and surface runoffeach of which is plotted as a triangle on semilog paper.
This figure is reproduced here as an illustration of a particular concept of the
components of the hydrograph without any comment on the extent to which
it reflects the position in most natural hydrographs. Whether we approach the
problem of runoff prediction from a classical or a systems viewpoint, it is
necessary to make some assumptions as a basis for the runoff prediction. The
division of the runoff cycle into phases and the division of the runoff hydrograph into the three components described above are examples of such
assumptions.
79

100
.r-'TOTAL HYDROGRAPH
^6ROUDWATER"REFSON^
16
17
18
19
FIGURE
20
21
MARCH 1936
22
23
24
25
4-5.Hydrograph of total flow.
The Rational Method

During the latter part of the 19th century and earlier part of the 20th
century runoff was predicted in one of two ways. Most engineers used empirical formulas which were derived for particular cases and then applied to
other cases on the assumption that conditions were similar enough for the
predictions to be of some value. The second method used was that which has
come to be known as the '^rational method.'' In this review there is little need
to examine the empirical formulas as they were ad hoc models whose parameters were derived for one particular case and then used in a wider context.
The rational method, however, was essentially a procedure and, as its name
implies, was an attempt to approach the problem of runoff prediction rationally. The assumptions which it made were unduly restrictive but, nevertheless,
it is interesting to discuss this approach here as it was the one which lead
ultimately to the development of some important methods in classical and
modern hydrology.
Though the rational method is often dated from the papers of Kuichling
80
100 r
50
O
O
q.
3 10
L
O
er
<
SURFACE RUNOFF
PLUS INTERFLOW
20
21
MARCH 1936
22
23
24
25
FiGUEE 4-6.Surface runoff plus interflow.
(23) and Lloyd-Davies (26), the method is clearly outlined in a paper by

Mulvaney (28) presented to the Institution of Civil Engineers of Ireland in
1851. In this paper, Mulvany gives a clear exposition of the concept of the
time of concentration and its relation to the maximum runoff in the following
terms:
The first matter of importance to be ascertained in the case of a small or mountainy
catchment is the time which a flood requires to attain its maximum height, during the
continuance of a uniform rate of fall of rain. This may be assumed to be the time necessary for the rain which falls on the most remote portion of the catchment to travel to
the outlet, for it appears to me that the discharge must be greatest when the supply
from every portion of the catchment arrives simultaneously at the point of discharge
supposing, as above premised, the rate of supply to remain constant, and this length of
time being ascertained, we may assume that the discharge will be the greatest possible
under the circumstances of a fall of rain occurring, of the maximum uniform rate of
fall for that time.
Mulvany then cites the example of a catchment with a time of concentration of 3 hours. He points out that 1 inch of rain falling in 3 hours on such a
catchment would give a greater flow than 2 inches of rain falling in 24 hours.
He goes on to discuss the factors which affect the time of concentration as
81
follows:
This question of time as regards any catchment, must depend chiefly on the extent,
form and rate of inclination of its surface; and therefore one great object for investigation is the relation between these causes and their effect; so that, having ascertained
the extent, form and average inclination of any catchment, we may be able to determine
in the first place, the duration of constant rain required to produce a maximum discharge, and consequently to fix upon the maximum rate of rainfall applicable to the
case. It is evident that, as a space of time is reduced, the rate of maximum rate of rain
is increased.
Mulvany was concerned with the maximum rate of runoff and that he
assumed a constant rate of rainfall. The circumstances of the development of
the rational method have been described elsewhere by Dooge (11).
The original rational method which was used to predict the maximum runoff was modified in the 1920's to allow for nonuniform intensities of rainfall
during the storm and also to allow for irregularities in the shape of the catchment. The first proposal for adapting the classical rational method to take
account of variations of rainfall within the storm period appears to have been
that by Hawken and Ross {15,37). K few years later, a second variation was
introduced to overcome the defect in the original rational method thatin
certain irregular shapes of catchment encountered in the design of sewerage
schemesthe predicted discharge from a part of the catchment could be
greater than the predicted discharge from the whole of the catchment. The
first modification of this type appears to be that due to Reid {34) in 1926.
Methods of handling the nonuniform rainfall can also be studied in papers
by Rousculp {38), Coleman and Johnson {8), Judson {21), Ormsby {33),
Harte {14) i and Laurenson {25). The method of allowing for a higher runoff
from a partial area depends on the type of rainfall formula used. The methods
are described in papers by Riley {35), Escritt {13), and Munro {29). Some
of these methods for allowing for the nonuniformity of rainfall and irregularity of area are discussed in somewhat more detail in lecture 8, (see ^TimeArea Methods'Oj where they are related to the process of deriving synthetic
unit hydrographs. In both of these lines of development, use was made of a
time-area diagram, which indicates the distribution of the time of travel from
different parts of the catchment.
Figure 4-7 top shows a watershed on which have been drawn isochrones of
equal travel time. Thus, each point on the isochrone labeled r = 4 has a travel
time of 4 hours, that is, it takes 4 hours for water to travel from any point on
that isochrone to the outlet. If a detailed survey of the catchment is available,
the position of the isochrones can be estimated by making allowance for the
time of overland flow to a channel and then calculating the time of flow in the
channel by Manning's formula or by some similar method.
If the area of that part of the catchment whose time of travel is less than
or equal to a given value of r, is plotted against that value of r, we obtain a
time-area diagram as shown in figure 4-7, bottom left. According to the rational
82
TECHNICAL BULLETIN NO. 1468, U.S. DEPT OF AGRICULTURE
FIGURE
4-7.Toy. Isochromes of travel time. Bottom'. Left, time-area curve; right, timearea-concentration curve.
method, this diagram shows for any value of r, the area which will contribute
to the maximum flow at the outlet due to rainfall with a duration equal to r.
Often it is more convenient to use the time-area-concentration curve shown
on figure 4-7, bottom right. The latter is the derivative of the time-area curve,
and its base length is equal to the time of concentration (tc). The time-areaconcentration curve in the modified rational method corresponds to the
instantaneous unit hydrograph (lUH) in the unit hydrograph method (30).
In applying the modified rational method, the maximum rate of runoff was
obtained by superimposing the cumulative rainfall pattern (or the rainfall intensity pattern) and the time-area diagram (or the time-area-concentration
curve). To facilitate comparison, the time scales on the two diagrams were
made the same but with the time scale on the time-area diagram reading from
left to right and the time scale of the storm rainfall curve reading from right
to left. When the time-are-concentration curve and the rainfall intensity
curves were used, the maximum runoff was obtained by superimposing the
maximum rainfall intensity over the maximum of the time-area-concentration
83
curve, then multiplying corresponding ordinates of the two curves, and, finally,
summing these products to obtain the maximum runoff. It is easy, in the
hindsight of the systems approach, to interpret and describe this graphical
and numerical procedure as a convolution of rainfall intensity and the timearea-concentration curve. By sliding one curve laterally over the other, it was
possible, in the modified rational method, to obtain ordinates other than the
maximum and, with patience, to obtain sufficient points to define a complete
hydrograph of runoff (12).
In fact, we now realize that these methods developed in the 1920^s use the
time-area concentration curve as a synthetic unit hydrograph. Before the unit
hydrograph had been invented, engineers were deriving synthetic unit hydrographs (or synthetic -hydrographs) in the form of time-area-concentration
curves (or time-area diagrams) by using Manning's formula to estimate the
time of travel. Because such synthetic unit hydrographs were based purely on
translation and did not take account of storage effects (either in the sewerage
system or on the ground, in the soil, and in the channel network), it is not
surprising that when combined with the true rainfall intensity pattern, they
tended to overpredict the peak rate of runoff. It is worthwhile noting that in
the original rational method in which a uniform rainfall intensity is assumed,
the error due to assuming uniform rainfall intensity and the error due to
neglecting storage were opposite in sign. Thus, the predicted peak would not
be as great as in the modified rational method and might in fact be closer to
the true peak.
Those who used empirical formulas for the time of concentration were also
using a synthetic method; this time one based on empirical relationships between this particular parameter and the watershed characteristics. The rational method is still quite properly used in certain routine design problems
such as small roadway culverts.
The rational method may be considered as a parametric method in which
a simple model is used. The basic formula of the rational method is given by:
Qmax= C.2(c) -A
(1)
in which Qmax is the estimated peak discharge, C is a coefficient whose value

must be determined in some way, i{tc) is the intensity of rainfall of the chosen
frequency for a duration equal to the time of concentration (tc), and A is the
area of the catchment. In a recent publication, Nash (31) pointed out that
the rational method might have been developed on the basis of parameter
optimization. In this case, the data would have been examined to determine
the values of C and tc which gave the optimum fit in some defined sense. To
do so for a reliable set of data would be an interesting exercise.
Because these lectures are concerned with parametric hydrology, we have
only discussed the application of the rational method to the prediction of
individual storm events. Equation 1 can also be taken in a statistical sense
84
in which C represents the ratio of the peak rate of runoff of a given frequency
to the rainfall of the same frequency and a duration equal to the time of concentration. The use of the rational method in this way is outside the scope of
the present lectures, in which we are largely concerned with the rational
method as a forerunner of unit hydrograph procedures.
Unit Hydrograph Concepts

The unit hydrograph concept and its development was one of the highlights of the classical period of hydrology. Figure 4-8 reproduces figure 1 of
Sherman^s basic paper (40) published in 1932. In this figure, Sherman illustrated for the case of a triangular unit hydrograph the effect of rain during
successive standard periods in building up the shape of the surface runoff
hydrograph through the superposition of displaced triangular unit hydrographs, which combine to give the total runoff hydrograph. If the duration
of effective precipitation is greater than the base of the unit hydrograph, the
runoff becomes constant. For about 25 years, unit hydrograph methods were
widely used in applied hydrology without a recognition of the essential assumption involved, namely that the relationship between rainfall excess and
surface runoff was that of a linear time-invariant system.
It is instructive to quote a classical formulation of unit hydrograph pro-
,Maximum runoff rate

/
for continuous
uniform rainfall.
TIME
FIGURE
4-8.Superposition of unit hydrographs.
85
cedures and to compare this with the systems formulation of the same basic
idea. One of the best classical discussions of unit hydrograph procedures is
that given in ''Elements of Applied Hydrology'' by Johnstone and Cross (W).
They state the basic propositions of the unit hydrograph as follows:
We are now in a position to state the three basic propositions of unitgraph theory, all
of which refer solely to the surface-runoff hydrograph:
I. For a given drainage basin, the duration of surface runoff is essentially constant
for all uniform-intensity storms of the same length, regardless of differences in the
total volume of the surface runoff.
II. For a given drainage basin two uniform-intensity storms of the same length produce
different total volumes of surface runoff, then the rates of surface runoff at corresponding times t, after the beginning of two storms, are in the same proportion
to each other as the total volumes of the surface runoff.
III. The time distribution of surface runoff from a given storm period is independent
of concurrent runoff from antecedent storm periods.
The classical statement of unit hydrograph theory quoted above can be

summarized in six words: The system is linear and time-invariant. Proposition
I and proposition II together make up the property of proportionality. If,
the length of input remains constant but the volume of input increases, then
the base length of the outflow is not altered, but the ordinates of the outflow
are raised in proportion to the volume of input. Proposition III is the principle
of superposition, which allows us to decompose the input into separate parts
and then superimpose on one another the separate outputs to obtain the total
output.
The classical manner of stating the unit hydrograph concepts and properties was not questioned until about 1955. Nowadays, we make the assumption
that the watershed, in converting precipitation excess to direct storm runoff,
acts as a linear time-invariant system. It is interesting to note the comments
which Johnstone and Cross (20) make following their outlining of the three
basic propositions:
All these propositions are empirical. It is not possible to prove them mathematically.
In fact, it is a rather simple matter to demonstrate by rational hydraulic analysis that
not a single one of them is mathematically accurate. Fortunately, nature is not aware
of this.
In this regard our position has not changed. We are aware that the assumptions of linearity and time-invariance for a catchment system are not strictly
correct, but we are content to adopt them as long as they are useful. We can
look at the fundamental equations of open channel flow and show that they
are nonlinear; we can look at laboratory results which show that the runoff
from model watersheds is nonlinear; we can look at field results and demonstrate their nonlinearity. Nevertheless, we cling to the assumption of linear
operation. The reasons are that linear methods are relatively simple, are by
far the best-developed methods we have, and that the results obtained by
using these linear methods are acceptable for engineering purposes. We will
86
TECHNICAL BULLETIN NO. 1468, U.S. DEPT, OF AGRICULTURE
continue to use them until such time as workable nonlinear methods are developed and that are more accurate without being unduly complex or
costly.
The original unit hydrograph developed by Sherman was a continuous
hydrograph of runoff due to uniform rainfall in unit period. Later, Bernard
(4) introduced the idea of a distribution graph in which runoff is expressed,
usually as a percentage, in terms of volumes of runoff in standard periods.
Where the flow is subsequently routed through reservoir storage or channel
storage, it may be convenient to use a distribution graph rather than a unit
hydrograph.
The -hydrograph, or -curve, is defined as the hydrograph of surface
runoff produced by a continuous effective rainfall of 1 inch per hour. If the
unit hydrograph has been normalized to unit volume, then the D-hour unit
hydrograph corresponds to rain falling at a rate of l/D inches per hour for
D hours. For a rate of 1 inch per hour lasting for D hours, the ordinates of
the D-hour unit hydrograph have to be multiplied by D. In the S-hydrograph,
there are D inches in the first unit period of D hours, D inches in the second
unit period, D inches in the third unit period, and so on. The equation of the
S-hydrograph is, therefore given by:
S{t)=D^hD{t-iD)
for nD<t<{n+l)D
(2)
One of the big advances in classical unit hydrograph theory was the discovery that the -hydrograph could be used to convert a unit hydrograph
from one unit duration to another. Before this, it was necessary to find a
storm of the appropriate duration to derive the required unit hydrograph
from the data. If you wanted a 6-hour unit hydrograph, you had to find a
6-hour storm, or a storm whose duration was an even submultiple of 6 hours
so that the shorter unit hydrograph could be developed and then shifted and
superimposed to give a 6-hour unit hydrograph. Figure 4-9 shows the classical
diagram of the relationship between the S-hydrograph and the unit hydrograph. Once the -hydrograph has been obtained from any unit hydrograph,
a unit hydrograph of a new given period can be derived from it by displacing
the -curve by the required amount, subtracting the ordinates of the two
-curves, and normalizing the volume. This process can be represented by
the equation:
hD{t)=
S{t)-S{t-D)
(3)
As D becomes smaller and smaller, the process represented by the above

equation comes closer and closer to the definition of differentiation, and in
87
the limit we have :
Kit) =-18(1)2
(4)
The hydrograph defined by equation 4 is known as the instantaneous unit

hydrograph (lUH). It was developed in hydrology from hydrologie concepts
rather than from systems analysis where it was already known under a variety
of names, but most commonly as the impulse response (see pages 20 and 25,
lecture 1). The main motivation for its derivation in hydrology appears to
24
48
72
96
120
144
HOURS
FIGURE
4-9.Unit hydrograph and S-curve for 1,290-square-mile drainage area.
88
have been the need to simplify the treatment of synthetic unit hydrographs.
For a finite period unit hydrograph, the shape naturally depends on the unit
period (D), and it was discovered that for very short durations the changes
in shape were quite slight. Some workers in the field suggested going to the
limit and using an lUH, thus getting rid of the variable D.
Once the lUH is accurately known, any other finite period unit hydrograph
can be obtained through the -hydrograph. Indeed, the time-to-peak of a
finite period unit hydrograph of any given duration can be found directly
from the lUH. The peak of the finite period unit hydrograph, given by equation 3, is the time for which the above expression is a maximum. Since the
derivative of the -hydrograph is the lUH ho{t), then the condition for the
maximum ordinate of the finite period unit hydrograph hnit) is:
ho(t)-ho{t-D)=0
(5)
that is, the peak of the finite period unit hydrograph occurs at the time when
the ordinate of the lUH is equal to the ordinate at a time D earlier. The
ordinate of the finite period unit hydrograph at any time is given by the
integral expression:
hi ho{t)dt
"-
hD{t)=-
(6)
^ -^ t-D
Looked at from the viewpoint of classical hydrology, all of these results have
to be proved before we are convinced that the lUH can be used to derive any
other expression which we wish. From a systems viewpoint, we know from
our basic theory that for a linear time-invariant system, the impulse response
contains all the necessary information about the behavior of the system.
The process of deriving finite period unit hydrographs from an S-hydrograph is not as easy in practice as it appears on first sight. This is because the
>S-hydrograph may not be known continuously, but only at certain intervals
of time. If we start off with a unit hydrograph which is defined only for 6-hour
intervals, the derived AS-curve will be defined for the same intervals. We can
certainly try to derive the unit hydrograph for a period of 1 hour, 2 hours,
or 3 hours from this >S-curve, but the results may not have much meaning.
If there are inaccuracies in the original unit hydrograph, then there will
probably be oscillations in what would be a smooth -hydrograph. These
oscillations may lead to grossly erroneous ordinates in a second unit hydrograph derived from the /S-hydrograph. Though a smooth lUH will always
produce a smooth -hydrograph, there is no guarantee that the >S-hydrograph
derived from a smooth finite period unit hydrograph will itself be smooth.
Some of the problems at the end of this lecture are designed to show the pitfalls in this particular connection. Though hydrologists attribute oscillations
in >S-curves to measurement and other errors in the data, it is quite possible
LINEAR THEORY OF HYDROLOGC SYSTEMS
89
for oscillations to arise in iS-hydrograplis derived from synthetic finite period

unit hydrographs which appear physically reasonable.
Separation of Base Flow

The first step in analyzing an actual hydrograph is to separate the base
flow from the direct storm runoff. Hydrologie literature abounds with methods
for making this separation. The effect of different types of storm event on the
hydrograph are shown schematically in figure 4-10 which are due to Horton.
Figure 4-1OA shows the effect of an intense rainfall of short duration. Because of the high intensity there would be surface runoff, but due to the short
duration and consequent small volume, the field moisture deficit might not
be satisfied, and, thus, there would be no recharge to ground water. Under
these conditions, the base flow recession before and after the storm event
would follow the same general curve, and the response of the hydrograph
would consist of a sharp rise and sharp recession back to the same master
curve of base flow recession.
On the other hand, if we have prolonged rainfall of small intensity, we get
the condition shown in figure 4-105. In this case, the intensity does not exceed
the potential infiltration rate, and, thus, there is no surface runoff. However,
the rainfall is sufficiently prolonged to make up the field moisture deficiency
and to give a recharge to ground water shortage. The effect of this recharge is
to increase the amount of ground water outflow or baseflow, and the recession
curve is shifted as shown in a stylized fashion on figure 4-105. In this case,
the recession curve after the rainstorm has the same shape as the recession
before the rainstorm but is shifted in time. More usually, however, in storms
which are of consequence in hydrologie analysis, both of the above effects are
combined so that we get both the distinct peak and a measurable amount of
surface runoff on the one hand and a recharge of ground water giving a shift
in the master recession curve on the other. This mixed condition is shown in
figure 4-lOC. One of the first steps necessary in unit hydrograph analysis is
to separate out these two effects.
If during the analysis of a discharge hydrograph, we encounter a storm
event of the first type, as shown on figure 4-1OA, where there is no recharge
to ground water, then there is no problem in separating the surface runoff
from the baseflow. All that has to be done is to join up the line of recession
before and after the storm event and treat all flow above this single master
recession curve as surface runoff. In the second case, as shown on figure
4-1OC, where all the flow is base flow, no difficulty arises because this is not
a storm event from the point of view of surface runoff.
For a storm event giving rise to both surface runoff and ground water recharge, some method of separating the two must be applied if the unit hydro-
90
High Intensity, short duration

B
Low intensity, long duration
Mixed response
FIGURE
4-10.Hydrograph response to different types of storm events: A, Surface water

response; B, base flow response; C, combined response.
91
graph of direct runoff is to be derived. The applied hydrologist has quite a

wide choice of such separation techniques to draw from in the technical literature, but few of them are soundly based. In these methods, the base flow is
separated in some arbitrary fashion and then the total precipitation is adjusted so that the volume of effective precipitation is equal to the volume of
direct storm runoff. There is no attempt to link infiltrating rainfall with
ground water recharge and hence, with ground water outflow. Transition from
the recession curve before the storm event to the recession curve after the
storm event is usually taken as being of relatively little interest in applied
hydrology, but this is a grave error. In fact, the form of this recession gives us
the shape of the ground water unit hydrograph, a concept which has been
studiously ignored by applied hydrologists over the past 35 years. If a block
diagram is drawn of the procedure described above, it would show an open
loop between the infiltration into the soil and the ground water outflow. This
would indicate that these two quantities would have to be either separately
measured or else connected by a subsystem. In the systems formulation of
catchment response, this open loop is closed as shown in figure 1-8 (p. 16).
Most workers in applied hydrology are ready to accept that a good representation of the recession curve can be got by fitting a straight line to the
recession part of the hydrograph plotted on semilog paper. This is equivalent
to assuming that the ground water reservoir acts as a single linear reservoir.
Once this assumption has been made, the maximum benefit should be obtained
from it and the further assumption made that the ground water reservoir acts
as a single linear reservoir during the recharge as well as during recession.
Figure 4-11 shows the application of this approach. The total precipitation
is taken as being divided into precipitation excess and a constant rate of infiltration; this represents a </)-index approach rather than the use of a more
sophisticated infiltration equation. The first part of the infiltration will recharge ground water at a constant rate. The ground water hydrograph will be
as shown in figure 4-13. From A to B during the replenishment of field moisture deficit, the base flow will continue to decline as before and we will
have:
i-m]
Q = QAexp\-r--^)\
(7)
From B to C, the ground water reservoir will operate as a linear reservoir

being recharged at a uniform rate (Ro) and the outflow will be:
Q= (QB-RO) expF^^^^^J+Ao
(8)
After the cessation of rainfall and an allowance for time of travel through
the soil, the recharge to ground water will cease and the recession will be ex-
92
ponential as before:
Q = Qcexp|^^-^J
^ -
(9)
The above approach to ground water separation is rational insofar as it is

based on a definite model of ground water behavior. As such, it is superior
to purely empirical rules usually quoted.
There is little doubt that the actual separation of base now made in practice
in ad hoc hydrologie studies is superior to the separation that would be obtained by a blind application of the rules of thumb and empirical procedures
quoted in the textbooks. This is so because the hydrologist is usually familiar
with the particular watershed under examination. He modifies these empirical
rules to get a commonsense result based on his own sensitivity to hydrologie
behavior and his knowledge of the watershed. The trouble is, however, that
though the individual separation in ad hoc studies may be reasonably correct,
it makes the comparison of results between one watershed and another very
difficult when there is a large subjective element in the manner of separating
base flow from surface runoff.
In his study of 90 storm events on 48 British catchments, Nash (32) proposed a method of base flow separation which, though not founded on any
physical principle or model, had the great advantage of both being objective
A-B recharge of
soil moisture
B-C recharge of
groundwater
t
FIGURE
4-1 LSeparation of base flow.
93
and of affording some scope for investigating the effect of the assumption on
the results obtained. He proposed separating the base flow by drawing a
straight Hne from the start of the rising portion of the flood hydrograph to a
point on the recession such that the time between the end of the effective
rain and the point on the recession was equal to three times the lag between
the center of effective rain and the center of storm runoff. The point on the
recession to which the separation line was drawn could only be determined
by trial and error. In effect, Nash's method of separation gives an lUH whose
base length is three times its lag.
Analysis of Complex Storms

Leaving aside for the moment the difficulty of ensuring that the base flow
separation has been correctly made, we turn to a consideration of the problem
of deriving the shape of the unit hydrograph from the surface runoff hydrograph due to a complex pattern of effective precipitation. In the early unit
hydrograph studies in the 1930^s, the procedure was essentially one of trial
and error. This approach has already been referred to in lecture 1 and illustrated on figure 1-9. Without an objective criterion for the acceptance or
rejection of a trial unit hydrograph, the subjectivity of such an approach was
necessarily very high.
At the end of the 1930's, some less subjective methods were developed,
but these still did not have the objectivity required of a really scientific
method. In 1939, Collins (9) suggested an iterative method in which a trial
unit hydrograph is assumed and applied to all periods of rainfall except the
maximum. The trial surface runoff hydrograph thus generated is subtracted
from the total measured hydrograph to give a net runoff hydrograph, which
can be taken as the runoff due to the ignored maximum rainfall in a unit
period but which will also contain the errors in outflow due to errors in the
trial unit hydrograph. If this net hydrograph is then assumed to be the outflow
due only to the maximum rainfall in a unit period and ordinates are adjusted
by dividing by the volume of the maximum rainfall, we obtain a second approximation to the shape of the finite period unit hydrograph. This process is
repeated until there is no appreciable change in the ordinates of the trial unit
hydrograph. Another special method for determining the shape of the unit
hydrograph from a complex runoff unit hydrograph is the graphical method
described by Sherman (4^). If consistently applied without modification,
methods such as these could be ranked as objective methods of hydrograph
derivation since strict application of the method would always give the
same result. In practice, they were rarely objective since any anomalies in the
derived hydrographs were arbitrarily corrected by the investigator.
In the 1940's, the derivation of the unit hydrograph from complex storms
was based on the solution of the set of simultaneous equations giving the
ordinates of the finite period unit hydrograph (or volumes of the distribution
94
graph) and the rainfall volumes in each unit period. These equations, which
have already been given in lecture 1 may be written as:
yo = XQho
(10a)
yi^xiho+xohi
(10b)
2/2 = X2ho+xhi,+Xoh2
( 10c)
ym = Xmho + Xm-lhi+
(10m)
ym+i = Xmhi+
yp=
(lOn)
Xmhp-m
(lOp)
This set of equations can, of course, be summarized as :

k=i
yi= Y^Xkhi-k
(11)
In the above set of equations, the values of i/o, t/i,

yp are assumed to be
known, the values of a:o, Xi,
Xm are known, and the problem is to find
the values of /io, /ii,
hp^rn- From a mathematical viewpoint, this can be
done by solving the first equation for /lo; substituting this value in the second
equation and solving for hi] substituting for the value of /lo and hi in the third
equation and solving for h^] and so on until all the unknown values of h are
determined. In practice, the existence of errors in the values of the effective
precipitation x, or the direct runoff y, will produce errors in the ordinates of
the unit hydrograph h. The substitution of an inexact value of h^ in the second
equation will produce an error in /ii, and the substitution of these two erroneous values in the third equation will produce an error in /12. Under certain
circumstances, the error in the values of h, that is, in the ordinates of the
unit hydrograph, can grow rapidly and quite unreal values are obtained in
the solution for the later ordinates of the unit hydrograph.
Several methods have been proposed to overcome this disadvantage of the
above direct algebraic solution by forward substitution. One of these was the
method of least squares, whose use is mentioned by Linsley, Khler, and
Paulhus {25). The method was developed by Snyder {42) in the United
States and Body {5) in Australia and programed for the digital computer.
The least squares method of unit hydrograph derivation will be discussed in
greater detail in lecture 6. Another approach to this problem was that of
Barnes {3). In his approach, any oscillations occurring in the unit hydrograph
were eliminated by deriving the unit hydrograph in the reverse order. This is
95
in line with general experience in numerical methodsa calculation which is

unstable in one direction is usually stable if taken in the reverse direction.
Barnes further suggested that the estimated effective precipitation should be
adjusted until the unit hydrograph obtained in the forward and reverse directions was substantially the same.
Although the derivation of the unit hydrograph from the outflow hydrograph due to a complex storm (that is, the problem of identification) is a
difficult one to solve, the prediction of the flow hydrograph due to a complex
storm when the unit hydrograph is known is relatively easy. All that is required is the application of each of the volumes of effective precipitation in
a unit period to the known flnite period unit hydrograph. To obtain the outflow hydrograph, carefully locate each volume of effective precipitation in
time and then sum the results. In terms of the set of simultaneous equations
10a to lOp, the problem is simply to determine the left-hand side knowing all
the values of x and all the values of h.
Classical hydrology nearly always made use of a finite period unit hydrograph and, therefore, of the superposition of a finite (and usually small)
number of block rainfall events. Research workers who are interested in placing the classical unit hydrograph approach on a sounder theoretical basis
tended to use the lUH rather than a finite period unit hydrograph. The procedure for prediction is similar in this case except that summation is replaced
by integration. The relationship is shown on figure 4-12. In the upper part of
the figure, the rainfall falling between the time r+dr has been shown as
shaded. The volume of precipitation represented by this shaded area is x{T)dr,
If hit) is the lUH produced by a unit volume of precipitation excess falling in
an infinitesimal short time at = 0, then the shape of the hydrograph due to
the shaded area of precipitation will be the same as the shape of this lUH,
but the ordinates must be multipHed by x{T)dT, and the whole hydrograph
must be displaced along the time axis by an amount r. Each element of precipitation excess will produce a similar hydrograph.
Instead of concentrating on the effect of all times in the future of a given
element of precipitation excess, we can concentrate on the outflow at a given
time and examine how this is made up from contributions from precipitation
excess at all times in the past. As seen from figure 4-12, the contribution of the
shaded area of effective precipitation to the outflow at a time, t, will be:
by{t)=x{T)h{t-r) dr
(12)
Because all elementary areas of precipitation excess whose value of r is less

than t will contribute to the outflow at a time , we get for the outflow the
relationship :
y{t)= [
x{r)h{t-T) dr
(13)
'00
which is the familiar convolution relationship for a lumped linear time-invari-
96
1 /X(T)h(t-T)dT
FIGURE
4-12.Convolution of inflow with lUH.
ant causal system. The above derivation is inherent in the time-area version
of the rational method, or isochrone method, as this method is sometimes
known. The above physical demonstration parallels the purely mathematical
derivation of the convolution relationship given in lecture 1.
In the 1950^s, a number of research workers in hydrology, working independently of one another, began to grasp that unit hydrograph methods
represented the application in hydrology of systems techniques used in other
disciplines. An essential step forward here was the recognition that the unit
hydrograph method was merely the assumption that the watershed under
examination was converting effective precipitation to storm runoff in a linear
time-invariant fashion. The gradual development of the systems formulation
of hydrologie problems can be traced in publications by Larriev {^4) > Nash
(30) j Dooge (10) y Amorocho and Orlob (2), Kuchment (22), and Roche
(36). The changes brought about by this new viewpoint can be appreciated
97
if the references given are compared with the treatment of the corresponding
topics given in the number of textbooks on hydrology published in the 1940^s
by Meinzer (27), Foster (13), Johnstone and Cross (20), Wisler and Brater
(45) j Linsley, Khler, and Paulhus (25), and the American Society of Civil
Engineers (1).
All of the concepts and methods of the classical unit hydrograph approach
can be neatly formulated in systems nomenclature. The only necessary assumptions in the unit hydrograph approach are those of linearity and timeinvariance (10). Once these assumptions are made, the relation between the
input, the output, and the system response are given by the convolution
equation. Where the inputs and outputs are defined continuously, the convolution equation takes one of the continuous forms discussed on pages 28 to
35 of lecture 1. The various methods available for the solution of the continuous convolution equation are discussed in detail in lecture 5. If the input and
output data are only defined as discrete points, then the unit hydrograph
approach can be formulated in terms of the discrete forms of the convolution
equation discussed on pages 35 to 40 of lecture 1, and the methods of solution
used in these cases are discussed in detail in lecture 6.
Problems on Classical Methods

1. The time-area variations of the rational method enable the complete
hydrograph to be predicted for a given storm. What is the relationship between this method and the unit hydrograph method?
2. Table 3 in the Appendix shows the effective rainfall in inches and the
runoff in cubic feet per second for the Big Muddy River at Plumfield, 111.,
for April and May 1927. Derive the 24-hour unit hydrograph from these
figures.
3. The figures defined by function 9 in Appendix table 2 when reduced to
unit volume represent the ordinates at hourly intervals of a 2-hour unit hydrograph. (1) Determine the runoff if the volume of effective rain in successive 2-hour periods is given by function 6 in Appendix table 2. (2) Calculate
the ordinates of the S-curve and from these derive the ordinates of the 8-hour
unit hydrograph. (3) What would be the effect of ignoring the variation of
the intensity of effective rainfall in the given storm? (4) Derive the 1-hour
unit hydrograph.
4. Carry out the calculations indicated in question 3 for the case where
the 2-hour unit hydrograph is defined by a triangle of unit volume whose
ordinates at hourly intervals are in the proportion indicated by function 8 in
Appendix table 2. Comment on the results obtained.
5. Assume that the hydrograph of effective precipitation is given by function 12 on Appendix table 1 and the hydrograph of storm runoff by function
13 on Appendix table 1. Determine as accurately as possible the form of the
lUH.
98
6. List a number of conventional methods used for separating the base

flow and the storm runoff. Compare these methods critically, and give your
opinion as to the probable order of merit.
7. An effective rainfall lasting 2 days produces an outflow lasting 6 days.
If the daily volumes of outflow are distributed according to function 11 in
Appendix table 2, apply Barnes method to determine the distribution graph
for the catchment.
8. For the outflow given in problem 7, show that a second unit hydrograph
can be derived from the same outflow hydrograph. Is it possible to prove that
there are no further exact solutions except these two?
9. (1) For the output obtained in either question 3 or question 4, make a
small alteration in one or more ordinates of the output and then seek to derive
the unit hydrograph for the original input and the adjusted output. Compare
the resulting unit hydrograph with the original unit hydrograph. (2) For the
same example, make an adjustment in an ordinate of the input leaving the
output unaltered and again proceed to derive a unit hydrograph. Contrast
the effects of errors in the input and the output.
10. Derive a matrix formulation for the Collins^ method of deriving a unit
hydrograph.
Literature Cited
(1) AMERICAN SOCIETY OF CIVIL ENGINEERS.
1949. HYDROLOGY HANDBOOK. Manual of Engin. Pract. No. 28, 184 pp.
(2) AMOROCHO, J., and ORLOB, G. T.
1961. NONLINEAR ANALYSIS OF HYDROLOGic SYSTEMS. Water ResGUFces Center
Contrib. 40. 147 pp., illus. Univ. Calif., Berkeley.
(3) BARNES, B. S.
1959. CONSISTENCY IN UNIT-GRAPHS. Amer. Soc. Civ. Engin. Trans. Jour. Hydraulics Div. 85(HY8): 39-63.
(4) BERNARD, M. M.
1935. AN APPROACH TO DETERMINATE STREAMFLOW. Amer. Soc. Civ. Engin.
Trans. 100: 347.
(5) BODY, D. N.
1959.
FLOOD
ESTIMATION:
DIGITAL COMPUTER.
illus.
UNIT-GRAPH
PROCEDURES UTILIZING A HIGH-SPEED
Water Res. Found. Austral. (Sydney)
Bui. 4: 41 pp.,
(6) CHAMIER, G.
1897.
CAPACITIES REQUIRED FOR CULVERTS AND FLOOD-OPENINGS.
Inst. Civ.
Engin. Proc. 134: 313.

(7) CHOW, VEN TE.
1964. RUNOFF. In Ven Te Chow, ed.. Handbook of Applied Hydrology. Sec. 14:
54 pp. New York.
(8) COLEMAN, G. S., and JOHNSON, A.
1931. RAINFALL RUNOFF. Inst. Munie. County Engin. Jour. 58: 1403.
(9) COLLINS, W. T.
1939. RUNOFF DISTRIBUTION GRAPHS FROM PRECIPITATION OCCURRING IN MORE
THAN ONE TIME UNIT. Civ. Engin. 9(9): 559-561.

1959. A GENERAL THEORY OF THE UNIT HYDROGRAPH.
99
Jour. Geophys. Res. 64(2):
241-256.
(11)
^
1957.
THE RATIONAL METHOD FOR ESTIMATING FLOOD PEAKS.

311-374.
London.
(12) EscRiTT, L. B.
1950. SURFACE WATER SEWERAGE.
(13)
FOSTER, E. E.
(14)
HARTE, C.
1948.
1932.
(15)
RAINFALL AND RUNOFF.
200 pp., illus.
487 pp.
London.
New York.
A.
Discussion of RAINFALL-RUNOFF CALCULATIONS by M. T. M. Ormsby.
Inst. Munie. County Engin. Jour., v. 59, p. 978.
HAWKEN, W. H.,
1921.
Engineering 184:
and Ross, C. N.
THE CALCULATION OF FLOOD DISCHARGES BY USE OF A TIME-CONTOUR PLAN.
Inst. Engin. Austral. Jour. 2: 85-92.

(16) HoYT, W. G.
1942. AN
OUTLINE OF THE RUNOFF
CYCLE.
Pa. state Coll. Tech. Bui. 27, pp.
57-67.
(17)
and CADY, R. C.
STUDIES OF RELATIONS OF RAINFALL AND RUNOFF IN THE U.S.
Survey Water-Supply Paper 772, pp. 123-247.
HARROLD, L. L.,
1936.
U.S. Gcol.
(18)
1942.
(19)
THE RUNOFF CYCLE. In Meiuzer, O. E., ed., Physics of the Earth-IX:

Hydrology. Ch. Xld, pp. 507-513. New York.
JENS, S. W.
1948.
DRAINAGE OF AIRPORT SURFACESSOME BASIC DESIGN CONSIDERATIONS.

Amer. Soc. Civ. Engin. Trans. 113: 785-809.
(20) JoHNSTONE, D., and CROSS, W. P.

1949. ELEMENTS OF APPLIED HYDROLOGY.
276 pp.
(21) JUDSON, C. C.
1932. RUNOFF CALCULATIONS, A NEW METHOD.
NeW York.
lust. Munic. Couuty Engin. Jour.,
V. 59, p. 861.
(22) KUCHMENT, L. S.
1964. OBOBSHCHENNAYA FORMOOLA DLYA RASCHETA GIDROGRAFA STOKA [A GENERALIZED FORMULA FOR THE CALCULATION OF A DISCHARGE HYDROGRAPH.]
Cent. Forecasting Unit Trans. [Trudy], TsIP No. 113, p. 23. Moscow.
NNLST Translation No. 3355. Sept. 1966. Boston Spa, Yorkshire,
England.
(23) KuiCHLiNG, EMIL.
1889. THE RELATION BETWEEN THE RAINFALL AND THE DISCHARGE OF SEWERS
IN POPULOUS DISTRICTS. Amer. Soc. Civ. Engin. Trans. 20: 1-56.
(24)
LARRIEV, J.
1954.
CONTRIBUTION A
FLOODS].
L'ETUDE
DES CRUES [CONTRIBUTION TO THE STUDY OF
La. Houille Blanche 6: 725.

1949. APPLIED HYDROLOGY. 689 pp., iUus. New York.
(26)
LLOYD-DAVIES, D. E.
1906.
THE
ELIMINATION OF
STORM WATER FROM SEWERAGE SYSTEMS.
Engin. Proc. 164: 41-67.
London.
lust. Civ.

(27) MEINZER, O. E., ed.
1942. PHYSICS OF THE EARTHIX: HYDROLOGY.
712 pp., iUuS.
NcW York.
(28) MTJLVANY, T. J.
1850.
ON THE USE OF SELF REGISTERING RAIN AND FLOOD GATJGES.
Inst. Civ.
Engin. Proc. 4(2): 1-8. Dublin.

(29) MuNRO, C. H.
1956. BASIC THEORY OF URBAN STORM DRAINAGE DESIGN. Inst. Engin. Austral.
Jour. 28: 301.
(30) NASH, J. E.
1958. DETERMINING RUNOFF FROM RAINFALL. Inst. Civ. Engin. Proc. 10: 163184.
(31)
1967. THE ROLE OF PARAMETRIC HYDROLOGY. Inst. Water Engin. Jour. 21(5).
Dublin.
(32)
1960. UNIT HYDROGRAPHS STUDY WITH PARTICULAR REFERENCE TO BRITISH CATCHMENTS. Inst. Civ. Engin. Proc. 17: 249-282.
(33) ORMSBY, M. T. M.
1932.
RAINFALL AND RUNOFF CALCULATIONS.
Inst. Muuic. Couuty Engin. Jour.,
V. 59, p. 978.
(34) REID, JOHN.
1926.
THE ESTIMATION OF STORM-WATER DISCHARGE.

Engin. Jour., v. 53, p. 997.
(35) RiLEY, D. W.
1931.
(36)
Inst.
Muuic.
County
NOTES ON CALCULATING THE FLOW OF SURFACE WATER IN SEWERS.

Munie. County Engin. Jour., v. 58, p. 1,483.
lust.
ROCHE, M.
1965.
POINT DE VUE MATRICIEL SUR UN OPERATEUR LINEAIRE DE TRANSFORMATION

PLUIES-DEBITS [PRIMARY POINT OF VIEW ON A LINEAR OPERATOR FOR TRANSFORMATION OF RAINFALL INTENSITY]. Proc. Symposium OU Representative
and Experimental Areas (Budapest). Internatl. Assoc. Sei. Hydrol. Pub.
66(1): 234-246. Gentbrugge. (In French.)
(37) Ross, C. N.
1921.
THE CALCULATION OF FLOOD DISCHARGES BY THE USE OF A TIME CONTOUR

PLAN. Inst. Engin. Austral. Trans. 2: 85.
(38) RouscuLP, J. A.
1927.
RAINFALL-RUNOFF ANALYSIS IN STORM-SEWER DESIGN. Engin. News-Rec. 98:
270.
(39) SHERMAN, L. K.
1932.
STREAM FLOW FROM RAINFALL BY THE UNIT-GRAPH METHOD.
Engin. NeWR-
Rec. 108: 501-505.

(40)
1932.
THE RELATION OF HYDROGRAPHS OF RUNOFF TO SIZE AND CHARACTER OF
DRAINAGE BASINS.
Amer. Geophys. Union Trans. 13: 332-339.
(41)
1942.
(42)
THE UNIT HYDROGRAPH METHOD. In Meiuzer, O. E., ed.. Physics of the

EarthIX: Hydrology. Ch. Xle, pp. 514-525. New York.
SNYDER, W. M.
1955.
HYDROGRAPH ANALYSIS BY THE METHOD OF LEAST SQUARES.
Civ. Engin., Jour. Hydraulics Div. 81: 1-25.
AMER. SOC.

(43)
1961.
MATRIX OPERATIONS IN HYDROGRAPH COMPUTATIONS.

Res. Paper 1. Knoxville.
101
.
Tenn. Val. Authority
(44) WILLIAMS, G. R.
1950. HYDROLOGY. In H. Rouse, ed., Engineering Hydraulics.
illus. New York.
(45) WiSLER, C. O., and BRATER, E. F.
1949. HYDROLOGY. 419 pp., illus. New York.
Pp. 229-320,
LECTURE 5:
IDENTIFICATION BASED ON CONTINUOUS DATA
Transform Methods of Identification
Lecture 5 deals with the identification of hnear time-invariant systems
where the data are given in continuous form, that is, by functions of a continuous variable. Historically, unit hydrograph procedures were first developed
for discrete or quantized data and only later adapted to continuous data.
In a systematic approach, one can start either with continuous inputs and
outputs or with discrete inputs or outputs. Since most hydrologists are more
familiar with continuous mathematics than with discrete mathematics, the
present lextures deal with continuous data before going on to discrete data.
In lectures 5 and 6, we will be dealing only with the question of identification;
the problem of simulation will be dealt with in lectures 7, 8, 9, and 10.
In tackling the problem of system identification, we are trying to develop
objective methods for describing the way in which a particular system operates
on inputs in order to produce outputs. This descriptionwhich may be
expressed in graphical, numerical, or functional formwill reflect the general
operation of the system but will tell us nothing about the nature of the system,
about the nature of any of its components, or the way in which these components are put together. If we can obtain a description of the operation of the
system for some general class of inputs (and if our assumptions of linearity and
time-invariance are reasonable), then we will have little difi^culty in predicting the output from the system due to any input belonging to this general
class. If linearity holds, then we can use the principle of superposition to
predict the output from any shape of input; if time-invariance holds, we can
apply the description of the operation of the system obtained from past
records to a future time. These assumptions may appear unduly restrictive,
but the strategy of parametric hydrology is to master the special case of linear
time-invariant systems before relaxing these assumptions.
It was shown in lecture 1 that the assumptions of linearity and timeinvariance allow us to relate the input and output of a particular system by
the convolution relationship :
KO = r
x(T)h{t-T) dr
(la)
2/(0= f
x{t-r)h{r) dr
(lb)
or
where h{t) is the impulse response of the system and provides a complete
102
103
description of the operation of the system. If the system is a causal system,

then the relationship between input and output is given by:
2/(0= [
x{r)h{t-r)dT
(2a)
or
y{t)= [ x{t-T)HT)dT
(2b)
If, in addition to the system being causal, the input is isolated, then we can
write:
y{t)= [ x{r)h{t-T)dT
(3a)
y(r)= f x{t-r)h{r)dT
(3b)
or
provided the time origin is taken to be not later than the start of the input.
In these circumstances, the problem of system identification reduces to the
mathematical problem of determining the function h{t) when given the
functions x{t) and y{t) and the relationship indicated by equations 1, 2, or 3.
The approach to the solution of the identification problem by transform
methods was mentioned in lecture 1. In these methods, the input, output, and
impulse response, which are connected by the convolution relationship:
y{t)=x{t)*h{t)
(4)
are each subjected to the same transformation so that:

r(x)=T[x(0]
(5a)
T(y) = T\jy{t):\
(5b)
T{h)=Tlh{t):\
(5c)
and
These transfbtined functions are then connected by the relationship:
T{y)=T{x)XT{h)
(6)
where X is the operation in the transform domain, which corresponds to convolution in the time domain.
If equation 6which may be described as a linkage equation (18)is
simple in form, then the transform of the system response may be expressed
in terms of the transforms of the input and the output. This transform of the
impulse response can then be inverted, though sometimes only with great
difficulty, to obtain the system response in the time domain:
h{t)=T-'[T{h)l,
(7)
The general procedure is illustrated in figure 1-11. There are three separate
stages in the identification process: (1) the transformation of the input and
the output (equation 5), (2) the solution of the linkage equation (equation
6), and (3) the inversion to obtain the impulse response in the time domain
(equation 7). The efficacy of any transform method depends on the ease with
which these three operations may be carried out. Nearly all of the methods
proposed for the identification of hydrologie systems with continuous input
and output, where the input can be isolated, may be considered as transform
methods. These methods are discussed in detail later in this lecture, but at the
moment, it is only necessary to commend briefly on their relationship to
one another.
System identification based on Fourier series involves the expansion of both
the input and the output into a series of sine and cosine terms. In each case,
the coefficients in the Fourier series represent the transformation of the
respective function, and the determination of these Fourier coefficients
represents the step corresponding to equation 5 above. Because the sines and
cosines are orthogonal to one another, the Fourier coefficients for the input
and output can easily be obtained by integration. If a linkage equation can
be obtained corresponding to equation 6, then the Fourier coefficients of the
impulse response can be determined from the Fourier coefficients of the input
and the output {18). The solution of equation 7, that is, the inversion of the
transform, offers no difficulty because the impulse response in the time domain
can be reconstituted from its Fourier elements. Though the Fourier method is
largely appHed to periodic data, it can be appUed, in the case of an isolated
input, to a system with a finite memory by basing the analysis on the assumption that the input and the output are periodic with a period which is equal
to or greater than the length of the output.
The restriction to isolated inputs and finite memories can be relaxed by
using the Fourier integral or Fourier transform instead of Fourier series {20).
This was the line of development adopted by electrical engineers in dealing
with transient phenomena. The use of the Fourier transform, however, has the
disadvantage that the problem of inversion is much more difficult than in
Fourier series. If the Fourier coefficients of the impulse response are known,
then the impulse response itself is known in the time domain to an accuracy
depending on the number of Fourier terms. In contrast, the Fourier integral is
difficult to invert, particularly if it is only known numerically. In systems
analysis, the Fourier integral is usually replaced by the Laplace transform to
enable us to handle unstable systems or systems whose stability is in doubt.
105
The numerical inversion of the Laplace transform (6) is even more difficult
than numerical inversion of the Fourier integral.
The first transformation method used to analyze hydrologie data was the
method of moments proposed by Nash (17) in 1959. From a theoretical point
of view, the moments (and the cumulants which will be discussed later) can
be derived from the Fourier integral or the Laplace transform. Moments and
cumulants share with the Fourier integral and the Laplace transform the
advantage of a simple linkage equation coupled with the disadvantage of
difficulty of inversion.
Dooge (9) has proposed the use of Laguerre functions in place of Fourier
analysis. Laguerre analysis shares with the Fourier series the advantage of
orthogonality and with the Fourier transform the property of covering the
range from zero to infinity. However, Laguerre analysis has the disadvantage
of requiring a more complicated linkage equation, which makes the determination of the coefficients of the impulse response numerically less stable
than where the linkage equation consists of a single term.
Analysis by Fourier Series
The definition and properties of Fourier series and other orthogonal functions were discussed in lecture 3 (see pp. 86-93). In the present section, we are
concerned with the application of Fourier series to the identification of linear
time-invariant systems. For such a system, the input, impulse response, and
output are connected by the convolution relationship :
2/(0= /
x(T)h{t-T) dr
(7a)
If the system is causal, this relationship can be written as :

2/(0=
x{r)h{t-r)dT
(7b)
'00
and if the system is causal and has a finite memory M, then we have :
/i(0=
x{T)h{t-T) dr
(7c)
'' t-M
Where the input is periodic with a period T^ the output will be given by:
y(t+kT)=
x{r+kT)h{t-T) dr
(8)
If the period of the input (T) is greater than the sum of the duration of

input (iV) plus the memory length of the system (M), that is, if
T> (M+N)
(9a)
then the value of the output y will return to zero during each period. Equation
8 can be replaced by the equation for an isolated output due to an isolated
input :
y(t) = /'
x{T)h{t-T) dr
(9b)
'' t-M
which is seen to be identical with equation 7c. An isolated storm event can be
analyzed by Fourier methods provided the assumed period T is greater than
the duration of output, which is the condition given by equation 9a.
In the Fourier analysis of systems, we need to obtain the Fourier coefficients
of the output as a function of the Fourier coefficients of the input and the
Fourier coefficients of the impulse response. These coefficients appear in the
Fourier series expansion of the three functions :
^
/.m27r\
^(0= 2-/ cêxpC^-I
(10a)
Kt)= 2: Tnexp(t^
(10b)
y{t)= E Cpexp(^pj
(10c)
The exponential or complex form of the Fourier series has been used in the
above equations because the linkage equation between the respective coefficients and other properties take a particularly simple form in the complex
coefficients. Since h{t) is zero for values of t between t M and t=Tj equation
9b can also be written as:
y{t)=
x{T)h(t-T) dr
(9c)
By the property of orthogonality we have :
Cp= I f y(t) exp [-i- ^') dt
(11)
Substitution from equation 9c into equation 11 gives:
^p=|/exp(-.^')r x{r)h{t-r)dr dt
(12)
107
Reversing the order of integration gives :

Cp=j;f x{r) j\xv{-i'^^h{t-r)dtdr
(13)
which can also be written as :

^^= U'/^'^ exp(-^')/%xp(-^^)M-r)d.dr
(14)
Performing the inner integration With respect to {tr) gives:

^v= j
^W expi-^--JTpdr
(15a)
exp l-i ^Y'J dr
(15b)
or
Cp = 7p j
X{T)
so that on integration with respect to r we obtain :

Cp=r.7p-Cp
(16)
which is the required Hnkage equation between the Fourier coefficients of the
output Cp, the Fourier coefficients of the input Cp, and the Fourier coefficients
of the impulse response y p.
In practice, the linkage is not quite this simple, because for a real function
the exponential Fourier coefficients will be complex. Accordingly, it is preferable to wTite the output in terins of cosine coefficients (A^) arid sine coefficients {Bk), the input in terms of cosine coefficients {ak) and sine coefficients
{hk)i and the impulse response in terms of cosine coefficients ak and sine
coefficients ft. Because we have :
Ck^y2{Ak'-iBk)
(17a)
Ck=^y2{ak-ihk)
(17b)
yk = y2{oLk-ik)
(17c)
y2{Ak-iBk)=T'y2{ak-ihk)y2{oLk-ik)
(18a)
equation 16 can be wTitten as:
which the real part gives :

T
Ak=- ^ {akOLk-hkk)
(18b)
T
Bk= - {akk+hkOik)
(18c)
and the imaginary part :

In system identification, we need to express the coefficients of the impulse
response (a and ) in terms of the coefficients of the input (a and h) and the
coefficients of the output {A and B). These are obtained by solving equations
18b and 18c for ak and /?&, getting:
2
T
QkAk+bkBk
k^+hk^
,,^ .
2
1
kBkhkAk
ak^+bk"
.,^, .
Once the values of ak and j^ have been obtained, the form of the impulse
response is easily deterniined since it is given by:
, , X
. .
^ /
fc27r .
k2irt\
hit) =Hao+ E ( ik cos +k sm j
(20)
If only a limited number of coefficients are determined, the efect is that the
high frequency components neglected by the truncation are not included in
the impulse response. Because hydrologie systems are heavily damped, the
neglect of high frequency components does not give rise to appreciable error.
The linkage equation derived above is for Fourier coefficients defined in
terms of a continuous function. If the data were defined continuously, it
would be possible to compute these coefficients either by Gaussian quadrature
formula based on a very large number of equally spaced sample points. In
lecture 6, the same linkage equation is obtained for the pulse response where
the input and the output are defined discretely. In the latter case, the linkage
equation was derived and applied by O'Donnell (18) to actual data of surface
runoff.
Analysis by Fourier and Laplace Transforms

As mentioned in lecture 3, the Fourier and Laplace transform techniques
have been widely used in the analysis of nonperiodic phenomena (12j 20), In
these cases, a simple linkage equation can also be found. Most hydrologie
systems are inherently stable and, thus, could be analyzed by Fourier transforms; however, Laplace transforms are more widely treated in the engineering
and mathematical literature, and the tables of transforms are more extensive
{12, 23), In lecture 3, both techniques were mentioned and both will be discussed in this lecture.
The Fourier transforms of the input, output, and impulse response are
given by:
x()= r x{t) exp(io)t) dt
(21a)
LINEAR THEORY OF HYDIOLOGIC SYSTEMS
109
Yio>)= f
y(t) exp{-io>t) dt
(21b)
H{o>)= f
h(t) expi-iut) dt
(21c)
'00
For a linear time-invariant system, we have the relationship:

y(t)=^ r x{T)h{t-T) dr
(22)
' 00
It is necessary to find the linkage equation between the Fourier transform

of the output and the Fourier transforms of the input and the impulse response.
Substituting from equation 21 into equation 22, we obtain:
Y{)= r exp(~co) r x{T)h(t-r)drdt

'-oo
(23)
'00
Reversing the order of integration gives:
Y{)= r
exp{-it)h{t-T)dtdT
X{T)
'00
(24)
'00
Replacing t by (r) as a variable of integration in the inner integration and

rearranging exp (it) gives :
F(a))= r
X{T)
expi-ir)
'-oo
exp[-tco(-r)]/i(-r)d(-r) dr
(25)
'-00
and performing the inner integration gives:

7(0,) = r x(r) exp(~icoO-^(co) dr
(26a)
'00
= H{o)) f
X{T)
exp(-tW) dr
(26b)
'-00
Performing the remaining integration then gives the required relationship:

y(co)=ff(co).X(co)
(27)
As compared with analysis by Fourier series, the coefficients of the Fourier

series analysis are replaced by the continuous functions of the Fourier transform. If we allow for this difference, the linkage equation given by equation 27
is seen to be of the same form as the linkage equation for Fourier analysis given
by equation 16 above.
While the form of the relationship shown in equation 26 is suitable for
lio TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

analytical purposes, it is necessary for the purposes of calculation to separate
the real and imaginary part of the Fourier transform. Thus, we need to write:
Xic,)=XE{)+iXi()
(28a)
Y{) = YRi)+iYi()
(28b)
H{cc)=HR{)+iHi{)
(28c)
Substituting the expressions from equation 28 into equation 27 and equating

the real and imaginary parts, we obtain:
-HiMXii)
(29a)
Fi(co) =HR{)Xi{) +Hi{)XRi)
(29b)
7(CO) =HEMXE{C)
In the identification problem, we need to express the real and imaginary

parts of the Fourier transform of tlie impulse response in terms of the real and
imaginary parts of the Fourier transforms of the input and the output. These
are given by:
HRM=
Hi{)=
XR[03)^ + XI{0))^
XR{O})^+XI{O))^
(30a)
(30b)
In electrical engineering, it is unusual to express the Fourier transform of

the system in terms of the amplitude and the phase angle. In hydrologie
systems, the formulation of equation 30 is probably more convenient.
The determination of HR{O)) and Hi{) only specifies the impulse response
in the frequency domain. To find the description of the impulse response in
the time domain, it is necessary to invert the Fourier transform H{o)). This
is given by:
1 /*
h{t)=- /
211 J_^
HR{O))
cost-'Hi(o)) sinQd
(31a)
Because h(t) is real, we have:

HR{-^)=HR{^)
(31b)
Hi{-) = -Hi()
(31c)
and
so that we can write:
1 /*
h{t)=n ^0
IHR{O)) COS(COO
-Hi{) sinco] d
(31d)
If h{t) is causal, that is, if it is identically zero for all negative values of ,
111
we can also write:

2 /**
h(t)=HR{O))
n ^0
do)
(31e)
h{t) =- rHi{) siii(coO d
(31f)
COS(COO
or
n ^0
Equation 30 may be compared with equation 19. Again, the Fourier integral
approach is similar to the Fourier series approach except for the replacement
of summation by integration. The necessity to integrate suggests the possible
use of values of determined by the requirements of Gaussian quadrature.
For the bilateral Laplace transform, defined by:
FB{s)=Bfm
= r m expi-st) dt
(32)
'00
the development of the linkage follows exactly the same steps as in the Fourier
transform. However, in the more usual unilateral Laplace transform defined
by:
F{s)=Um
= r/(0 exp(-sO dt
(33)
'o
care must be taken with the Umits of integration.

For a linear time-invariant system for which the input is zero for negative
time, we have the relationship:
y(t)= r x{r)h{t-T) dr
'o
(34)
The Laplace transform of the output is given by:
Y{s) = r y{t) exp(-sO dt

= ( exp(-sO ( x{T)h{t-T)dT dt
(35a)
(35b)

Reversal of the order of integration gives:
/OO
F(s)= /
/.OO
X{T)
exY>{-st)h(t-T)dtdr
= /
a;(r) exp( sr) /
'o
'o
exp[s(r)]/i(r)d c/r
(36a)
(36b)
Because the systeih is causal, A(r) will be zero for any value of t less than r,
and, consequently, the lower limit of integration for the right-hand integral
can be set equal to r, thus giving us:
/OO
F(s)= /
/ 00
x(r) exp(-sr) /
expl-s{t-T)2h{t-T)dtdT
(37)
Changing the variable in the inner integration from to u = r, we obtain:

F(s)= /
a:(T) exp( sr) /
exp{ su)h{u)dudT
(38a)
Y{s)= xir)exp{-ST)'H{s)dT
(38b)
Y{s)=H{s)
(38c)
X{T)
expi-sr) dr
Y{s)=H{s)'X(s)
(38d)
Once again, the linkage equation has the same general form as in the case
of Fourier series and Fourier transform. Equation 38d only gives us the
Laplace transform of the impulse response or the system function as it is
sometimes called. The numerical inversion of a Laplace transform is extremely
diflcult. One of the most efficient ways of doing it is to expand the Laplace
transform in terms of a series of orthogonal polynomials and then invert this
series term by term (6). It would appear, however, that if the orthogonals
are going to be used for inversion, then we might as well start and base our
whole analysis on the use of orthogonals.
Both the Fourier transform method and the Laplace transform method
have been used for the identification of hydrologie systems. In 1952, Paynter
(21) suggested the use of Laplace transform methods for the study of both
hydraulic and hydrologie systems. Diskin^ determined the Laplace transforms
for a large number of storm events. The watersheds examined wêre between
1 DiSKIN, M. H.
A BASIC STUDY OF THE LINEARITY OF THE RAINFALL-RUNOFF PROCESS
IN WATERSHEDS. Ph.D. thesis, Univ. 111.
64-8375.]
1964.
[University Microfilms Publ. No.
113
30 square miles and 1,420 square miles in area and between four and 10 storms
were examined for each watershed. In the same year, Levi and Valdes (16)
discussed the application of Fourier transform techniques to the determination
of the lUH and appHed the method to the Tuxpan River in Mexico. More
recently, Blank and Delleur (7) used the Fourier transform approach in a
study of 1,059 hydrographs from 55 watersheds in Indiana.
Moments and Cumulants

The first transform method of identification applied to hydrologie data was
based on moments used by Nash (17) in 1959. In systems analysis, moments
are used in the same sense as in statistics. Thus, the *^ moment of a function,
which has been normahzed to unit area, about the point a, is defined as:
MR{f)= r mît-a^dt
(39)
'00
In particular, moments about the time origin are defined as :
UR'{)=
r m-t^'dt
(40)
and moments about the center of area are defined as :

UR{f)= r m{t-Ui'Ydt
(41)
' 00
The moments are related to the Fourier transform and the Laplace transform; in the theory of statistics, the Fourier transform is used in the form of a
characteristic function or a moment generating function. If we are dealing
with functions that are zero for negative time and are only interested in
moments about the origin, it is possible to perform all the operations necessary
with the ordinary Laplace transform. If, however, we wish to deal with the
moments about the center of area (or with functions which are not zero for
negative time), then it is necessary to use either the Fourier transform or the
bilateral Laplace transform. The following development is in terms of the
bilateral Laplace transform, which is defined by:
FB{S)=
fit) exp{-st) dt
(42)
If the above expression is differentiated with respect to s, we obtain:

d.
[FB (s) ] = - / /(<) < exp ( - St) dt
(43)

and if the differentiation is carried out R times:
-^[PB{S)^={-IY
m't^'exv{-st)dt
(44)
By setting s = 0 on both sides of the equation, we obtain:

^CFB(s)l=o=(-l)y
m-t^dt
(45a)
(45b)
= ("-1)^-C/'(/)
so that the *^ moment about the origin can be obtained from the Laplace
transform provided that the transform exists and can be differentiated R
times at s = 0.
The relationship between the moments of the input and the output and the
impulse can be obtained as follows. For a linear time-invariant system,
we have:
(46)
YB(S)=XB{S)'HB{S)
The *^ moment of the output about the origin = 0 is given by:

UR'{y) = {-l)îYB{snô
(47)
Substitution from equation 46 into equation 47 gives :

d^
Un'iy) = (-1)[Xfi(s) -HB{sn^
(48)
Using Leibnitz's formula for the continued differentiation of a product, we

have:
k==R
fjk
R-k
Un'iy)=(-1) L a)-j:,XB{sns-^-j^HB(sn^
ds^
A;=0
* ds^~^^
(49)
which gives the relationship between the moments about the origin as :
k=R
UR'V)
= Z (?) Uk'ix) Ur-k'{h)
(50)
It can be shown that if the moments of the normalized output are taken
around the point t = a, the moments of the normalized input around t = ), and
the moments of the normalized response about the point t = c, and if a = &+c,
then the relationships between the moments is the same form as equation 50.
In particular, if the moments are taken about the respective centers of area,
115
UE{y)= i:Ck)U,(x)UE.,{h)
(51)
we have:
which was the form of the theorem of moments for a Unear time-invariant
system pubhshed by Nash (17). For the special case of = l, equation 50
becomes:
C7/(2/) = f//(s) + C7/()
(52)
which expresses the fact that the lag of the output is equal to the lag of the
input plus the lag of the impulse response. For = 2 and = 3 in equation 51,
because 7i( ) =0, we have the special cases:
U2y) = U2x) + U2(h)
(53a)
U^{y)Ûz{x) + Uz{h)
(53b)
This special additive relationship does not hold for any higher moments.
Equations 50 and 51 represent linkage equations between the moments of
the output, the input, and the impulse response. Once the moments of the
input and the output have been determined, the corresponding moments of the
impulse response can also be determined. The final inversion of the latter can
only be made via the Fourier transform or Laplace transform. The problem of
moment inversion is to determine the nature of the function given the moments
of that function. If the Laplace transform of the function is consistent when
near zero, it may be expressed in terms of a Maclaurin series :
which can be written as :

Fis)=Z(-'i)''U,'{f).
(55)
Even if only a few moments are known, they give a certain amount of information about the Laplace transform near the origin and, therefore, of the
original function at relatively large values of time.
Moments are not the only set of parameters which may be used to describe
the response function; in some cases they are not the most convenient set.
Another set of useful parameters used in statistics are the cumulants or socalled semi variants (H)- These are defined as the set of parameters for which
the logarithm of the characteristic function (or Fourier transform) is the
generating function. All the cumulants except the first are unaffected by a
change of origin. In a similar manner to the moments, the cumulants can be
derived by continuous differentiation of the logarithm of the Fourier transform

or the Laplace transform. Thus, the cumulants may be defined by:
Knif) = (~l)^[logF,(5)l^
(56)
For a linear time-invariant system we have:

Y{s)=X{s)'H{s)
(38a)
and therefore:
logF(s) =logX{s) +\ogH{s)
(57)
Differentiating both sides of equation 57 -times, and setting s = 0, we obtain:

/7
d^
/7
ClogF(5)l^ = [logX(s)l^+[logff(s)],o
(58)
which is clearly equivalent to:

KR{y)=Kn{x)+KR{h)
(59)
thus, indicating that in the case of cumulants we get the simple additive
relationship of equation 59 for all orders of cumulant.
The simple form of the moments relationship in equation 52 and equation 53
is due to the fact that the first cumulant is equal to the first moment about the
origin and the second and third cumulants are equal to the second and third
moments about the center of area, respectively. The fourth cumulant is equal
to the fourth moment about the center of area minus three times the square
of the second moment about the center of area and is known in statistics as
excess kurtosis. The Gaussian distribution has a first cumulant which determines the position of the mean and a second cumulant which determines the
variation about the mean, but all cumulants above the second are zero. In
the gammâ distribution, which is widely used in hydrology, the *^ cumulant
takes the form
KB = n(R-l)\K^
(60)
where n and K are the parameters of the gamma distribution.

Nash (17) also introduced the idea of plotting dimensionless shape factors
derived from moments in order to compare the shape of derived unit hydrographs. He defined a dimensionless moment of order R as the *^ moment
about the center of area divided by the first moment about the origin raised
to the power of i?, that is,
^" = 0^
(61)
117
In dealing with the linear theory of open channel flow, Dooge^ found it more
convenient to define dimensionless shape factors in terms of the cumulants
rather than the moments. These are defined as :
and can be plotted against one another to compare different functions or

models with one another or to compare a model with the data which it is
attempting to simulate.
In the above discussion of moments and cumulants, it has been assumed,
as indicated earUer, that all the distributions involved have been normalized
to unit area. The use of normaUzed distributions is convenient both in theoretical investigations and in actual computations. If required, however,
corresponding relationships can be derived for the case where the input and
output have not been normalized.
Laguerre Analysis of Systems

It was noted previously that a Fourier analysis of systems had the advantage of orthogonaUty but the disadvantage that the method could only be used
for an isolated input to a system with finite memory. The success of the
method, however, would suggest that in systems with infinite memory an
alternative method of analysis which might be useful would be one based on
functions which are orthogonal over the whole range from zero to infinity
instead of only over a finite range. Because Laguerre polynomials are orthogonal over the range 0 to QO with respect to the weighting factor exp( 0,
this suggests the use of Laguerre functions defined by:
/() =exp (-0 g {-mi)m
(63)
as the basis of the systems analysis. Dooge (9) has suggested that these
functions may be more convenient than Fourier methods for heavily damped
systems because the Laguerre functions can be seen to be made up of gamma
distributions, a function which has been widely used to represent the damped
response typical of natural watersheds.
If Laguerre functions are to be used as the basis of system identification,
then it is necessary to express the input, output, and impulse response in
2 DODGE, J. C. I. LINEAR THEORY OF OPEN CHANNEL FLOW, HI: MOMENTS AND CUMUDept. Civ. Engin., University College, Cork, Ireland. 1967. (Unpublished
report.)
LANTS.
terms of Laguerre functions. Because the Laguerre functions are orthogonal

to one another this is easily done. There is no guarantee, however, that the
Laguerre functions would be convenient functions to use in the analysis of the
system. The first step necessary is to examine what the effect is of convoluting
one Laguerre function with another, that is:
git)= [ fm{r)fnit^r) dr
(64)
where/w(0 and/n(0 are Laguerre functions as defined by equation 63. The

right-hand side of equation 64 results from multiplying a power series of order
m by a power series of order n and then integrating, thus producing a power
series of order (m+n+1). The resulting power series could therefore consist
of {m+n+1) terms, each of which is a Laguerre function. In practice, all but
two of the terms drop out and only the last two terms remain, the result being:
^(0=/m+n(0-/m+n+l(0
(65)
For the Laguerre series analysis of a system, we proceed as before and

expand the input, impulse response, and output in terms of Laguerre functions :
W1=00
x{t)= a/(0
(66a)
h{t)= 2a/n(i)
(66b)
2/(0= ZA^fAt)
(66c)
p=0
Due to the property of orthogonality, these coefficients are given by:
am= x{t)fm{t) dt
(67a)
n= r h{t)fn{t)dt
(67b)
'0
^>= r y{t)h{t)dt
(67c)
The linkage equation can be derived as follows. Substituting for y{t) in

equation 67c, we obtain:
^P= r fp{t) [ x{T)h{t-T)dT dt
(68)
119
and substituting in this equation the expressions for x{t) and h{t) in equations
66a and 66b, we obtain:
00
t
^f mag00
71=00
Ap= / fp{t) / Jâ,U{r) Jânfnit-T)dTdt

'0
'0
(69)
m=0
Reversing the order of the summations and the integrations, we have:

m=oo
n=aoo
i.00
/i
Aj,= a 22 / fp(t) / U{r)n{t-r)dT dt

m=0
n=0
'0
(70)
'0
Using the result of equation 65, this becomes:

m=oo
w=oo
/.oo
4p=Z.E"/ /p(OC/m+n(0-/^n+l(0]<
(71)
Integrating with respect to t and using the orthogonahty relationship, we have :

m=Qo
wôo
Ap= 2^ um 2^ nEp.m+n5p,m+n+lJ
m=0
n=0
(72)
Performing the summation with respect to n results in:

m=oo
Ap= ZJ Cim[0Lp-m^ OL p-m-\]
(73a)
m=0
Since the Laguerre coeflScients of the impulse response are only defined for
nonnegative values of n, this can be written :
m=p
in=p1
Ap= 22 dmOtp-m
2 CLkOp-1-m
m0
(73b)
m=0
which can be readily shown to be equivalent to :

k=p
m=p
X^fc= ^map-m
A;=0
(74)
m=0
The problem of identification is to determine the values of an given the

values of Ap and am. This can be done by successive calculation of the values
of a in accordance with :
k=p
aoap= Âfc
;==0
m=pl
^ amOip-m
(75)
m=0
Once the values of an are determined, the impulse response is easily found in
terms of equation 66b.
The main purpose of using orthogonal functions is to determine conveniently

the coefficients in the expansions of the given input and output. It is possible
by means of Laguerre analj^sis to express the linkage in terms of gamma distributions rather than coefficients of Laguerre functions. The input and other
functions can be expanded in terms of gamma distribution as follows :
Because gamma distributions are not orthogonal to one another, it is not

possible to obtain the values of the coefficients, dr, directly, but they can be
expressed in terms of the corresponding Laguerre coefficients obtained from
equation 67a, or corresponding equation. The relationship between the two
sets of coefficients is given by :
mao
dr={-2y+' ' ('^)a,
(77)
m=r
The result obtained by convoluting two gamma distributions, one of order

m and the other of order n, is a gamma distribution of order m+n+1 as
indicated by equation 78 :
^'e-^/Hr/2)'" e-(^-^>/n^"-V2)"
J<^
2{m\)
/o
2(m!)
2(n!)
-t/2
'2(m!)(n!y
w+n+1
2~
./3(m+l,n+l)
(78b)
where im+l, n+l) is a beta function. Expressing the beta function in terms
of factorials gives :
g(^)= 2(m+n+l)!
or ^ in.
(^8c)
When the input function, output function, and impulse response functions
are all expanded in terms of gamma distributions, we have the relationship:
Z DrT^= "f j:^^^^^ "f /^^^
(79)
By comparing the terms on the two sides of this equation, we obtain the
Hnkage relationship :
:=p-1
/)p = 2. ; rffcVfc-i
(80)
Because, in the problem of identification, we need to express the values of b
121
in terms of the values of D, and we need the Hnkage equation in the form:
k=p
dodp = Dp+i ^dkp-k
(81a)
;=l
if the value of do is zero, then we can use :

k=p
didp-i = Dpî ^dkdp-k
(81b)
fc=2
It is not known whether use of the linkage equation in terms of gamma distributions is numerically more stable than direct use of the Laguerre coefficients.
If a function is to be expanded in terms of a Laguerre series, the length of
series required to reproduce the function to a given degree of accuracy will
depend on the time scale chosen for the Laguerre functions. In any given
function, it is possible to determine the optimum time scale for Laguerre
representation. For a time scale other than the optimum to reproduce the
function to the same accuracy, a longer series would be required. In system
identification, there would be different optimum time scales for the input and
the output. The problem of choosing the optimum time scale in this case is
currently under investigation.
Though Laguerre analysis has been applied to some discrete hydrologie field
data (for which it is not orthogonal and therefore not appropriate), it has only
been tested on synthetic hydrologie data of a continuous type (9, 10). The
method has, however, been applied to the anal3^sis of cascaded systems in
Chemical Engineering by Anderssen and White (Jf ).
Time-Series Analysis
If the interval between nonzero inputs is shorter than the memory of the
system, then the output will not return to zero and the techniques described
above will not be applicable. In such cases, the input and output can be
viewed as time series and can be described in terms of their autocorrelation
and cross correlation as is done in the case of time series in communication
theory {15).
The autocorrelation function may be defined by the limit :
(/>XX(T)=-/
1 '-r/2
x{t)x(t+T)dt
as
r->oo
(82)
as
T^co
(83)
and the cross correlation function by :

1 /-^/^
<t>xy{r)=-
x{t)y{t+r)dt
1 ^ -Til
If there were no errors in input or output, then any of the systems described

earlier in this lecture should, apart from errors in computation, predict the
impulse response perfectly. If however, there are errors in the data, that is,
noise on the system, then perfect prediction is not possible. Methods of time
series analysis have been proposed by Eagleson (11) and by Bayazit (3) as a
method of handling this problem of noise in the same way as is done in communications engineering.
For any assumed causal impulse response, the residual error in the output
Ordinate is given by:
r{t)=y{t)- f h{r)x{t-T)dT
(84)
The optimum linear response is one which minimizes the residual given by the
above equation in some sense. If the criterion is taken as one of least squares,
then the problem is to minimize the expression :
rTI2
1 f'^"'
EUiit)^ = -;f,
lr{t)Jdt
as
T-ôo
(85)
-Til
Insertion of the value of r{i) from equation 84 in equation 85 gives:

Eih{t)'] = \y{t)h(T)x{t-r)dT\ dt
1 J -Til V
J S
J
as
Tôo
(86)
The problem, therefore, reduces itself to finding the optimum value hopt(t)y
which, when used in equation 86, minimizes the expression E[h{t)']. Squaring
the expression between square brackets in equation 84 gives rise to three
terms as follows :
-y{t)y{t) dt
as
T-^^
2
f^
--^yiO / h{T)x{t-r)drdt
1
-/
JQ
h(Ti)x(t-Ti)dTi
(87a)
as
T^^
h{T2)x{t-T2)dT2dt
(87b)
as
T-ôo (87c)
Both the first and the third terms must be nonnegative because they are the
result of squaring the terms inside the square brackets in equation 84. The
first of the three terms, that is, that given by equation 87a, is clearly equal to
(t>yy(0). The reversal of the order of integration in equations 87b and 87c and
use of the definitions of the autocorrelation and cross correlation function
given by equations 82 and 83 reduce the second term to a single integral and
the third term from a triple to a double integral. The expression to be mini-
123
mized can now be written as :

-00
Elh{t)'] = (l>yy(0)-2 /
/OO
h{T)(l>xy{r)dT+ /
/OO
h{Ti)dTi /
h{T2)<t>xx{ri-T2) dr^
(88)
The minimization of the above expression is obtained by a manipulation of
the ordinates of the impulse response h{t) until the optimum causal impulse
response is obtained. If the optimum causal linear response is denoted by
hopt{t)y then any nonoptimum linear response can be denoted by:
h{t)=hoAt)+^'h'it)
(89)
where e is an arbitrary real nonnegative number and h'{t) is an arbitrary

causal function. If /iopt(0 is a true optimal, then we must have:
^[/i(0]=^[/iopt(0+i-^'(0]>^C/iopt(0]
(90)
where Elh{t)^ is the error criterion defined by equations 84 to 88.

Substitution from equation 89 into equation 88 and segregation of the terms
involving hopt(t) and h'{t) results in the equation:
E[/i(0] = ^[/iopt(0]-2e h'T)c>,y{T)dT
'o
+ 2e /
h'{Ti)dTi /
/iopt(T2)</)xx(ri r2)cir2
+ /
h'{ri)dTi /
/i'(T2)</)xx(ri-r2) dT2
(91)

E[h{t):\ = E\:hoAt)']-2eh+eU2
(92)
where the second term on the right-hand side of equation 92 corresponds to

the second and third terms on the right-hand side of equation (91), and the
third term on the right-hand side of equation 102 corresponds to the fourth
term on the right-hand side of equation 91.
For /iopt(0 to be a true optimum, it is necessary for the condition of equation
90 to hold and hence for :
(i.-'-i) <0
(93)
for any value of h'{t) and any nonnegative value of e. Because the fourth term
on the right-hand side of equation 91 is a perfect square, then h must also be a
perfect square, and thus for arbitrarily small values e the condition for

optimality reduces to :
/i<0
(94)
Il can be seen from an inspection of the second and third terms on the righthand side of equation 91b to be:
/"
/i= /
/oo
h'{T)4>^y{T)dT- /
/.oo
h'{Ti) /
Aopt(r2)0xx(Ti-r2)c?r2C?ri
(95)
If 7i as given above is either zero or negative, then the condition of equation

94, and hence of equation 90, is satisfied. If, however, /i is negative for a given
function h\t)y then it will according to equation 95 be positive if the sign of
h'{t) is changed. Consequently, unless 7i is zero, it is possible to find a function
h\t) which makes /i positive and therefore violates the condition of equation
94 for an arbitrarily small value of e. In these circumstances, the function
ho^t{t) would not be truly the optimum causal linear response. Accordingly,
the condition for the optimum response is :
/i = 0
(96)
Since the condition must hold for any arbitrary causal function h'{t), then
wê must have:
</>xy(ri) /
/iopt(r2)0xx(ri T2) (ir2 = 0
(97)
Because W{t) was defined as a causal function, the above relationship need
only hold for nonnegative values of n.
The condition represented by equation 97 is frequently referred to as the
Wiener-Hopf equation, and the solution of this equation gives the optimal
causal linear response for a system whose input and output are in the form of
continuous time series. It can be seen that determination of the optimum
linear response in the least squares sense depends not on the original functions
but only on the autocorrelation function of the input and the cross correlation
function between the input and the output.
Problems on Identification Based on Continuous Data

1. Find the Fourier coefficients for the functions given in table 1, of the
Appendix.
2. Find the Fourier transform and Laplace transform for the functions
chosen in problem 1.
3. Find the first 4 moments and cumulants for the functions chosen in
problem 1.
4. Find the Laguerre coeflScients for the functions chosen in problem 1.
125
5. The input and output to a linear time-invariant system are given on

Appendix table 1 by functions 12 and 13, respectively. Find the impulse
response of the system by some method of system identification suitable for
continuous data.
6. Find the impulse response for the data of problem 5. Use a different
method of system identification.
7. In Appendix table 1, the output from a linear time-invariant system is
given by function 16 and the input, by function 15. Find the impulse response
of the system by some method of system identification.
8. Find the impulse response by a second method of system identification for
the data of problem 7.
9. Compare the results of problems 5 and 6 or problems 7 and 8, and the
difficulties of the two methods used and give reasons for the differences found.
10. Write a general computer program for the identification of linear timeinvariant systems for which the data are available in continuous form.
Literature Cited
(1) ANDERSSEN, A. S., and WHITE, E. P.
1969.
THE ANALYSIS OF RESIDENCE OF TIME DISTRIBUTION MEASUREMENTS USING

LAGUERRE FUNCTIONS. Canad. Jour. Chem. Engin. 47: 288-294.
(2) ASELTINE, J. A.
1958. TRANSFORM METHODS IN LINEAR SYSTEMS ANALYSIS.
300 pp., illuS.
NeW
York.
(3) BAYAZIT, M.
1964. APPLICATION OF POWER SPECTRUM ANALYSIS TO UNIT HYDROGRAPH DERIVATION. Tech. Univ. Istanbul Bui. 17(1).
(4)
1965.
A NUMERICAL METHOD TO DETERMINE THE UNIT HYDROGRAPH.

Istanbul Bul. 18(1).
Tech. Univ.
1966.
INSTANTANEOUS UNIT HYDROGRAPH DERIVATION BY SPECTRAL ANALYSIS

AND ITS NUMERICAL APPLICATION. Central Treaty Organization Symposium on Hydrology and Water Resource Development Proceedings.
(5)
Ankara.
(6)
BELLMAN, R. E., KALABA, R. E.,
1966.
and
LOCKETT, J.
NUMERICAL INVERSION OF THE LAPLACE TRANSFORM.
249 pp., illus.
New
York.
(7) BLANK, D., and DELLEUR, J. W.
1968. A PROGRAM FOR ESTIMATING RUNOFF FOR INDIANA WATERSHEDS, PART 1.
LINEAR SYSTEMS ANALYSIS IN SURFACE HYDROLOGY AND ITS APPLICATION
TO INDIANA WATERSHEDS. Tech. Rpt. 4, Water Resources Res. Center,
Purdue University, Ind.
(8) DiSKiN, M. H.
1967. A LAPLACE TRANSFORM PROOF OF THE THEOREM OF MOMENTS FOR THE
INSTANTANEOUS UNIT HYDROGRAPH. Water Resources Res. 3(2): 385-388.
(9) DOOGE, J. C. I.
1965.
ANALYSIS OF LINEAR SYSTEMS BY MEANS OF LAGUERRE FUNCTIONS.
Indus, and Appl. Math., Jour. Control 2(3): 396-408.
SoC.

(10)
(N.D.)
(11)
LAGUERRE ANALYSIS OF SYNTHETIC DATA.

Cork, Ireland.
EAGLESON, P. S., MEJIA-R, R.,
1966.
THE
and
MARCH,
COMPUTATION OF OPTIMUM
Dept. Civ. Engin., Univ. Col.
F.
REALIZABLE UNIT HYDROGRAPHS.
Water
Resources Res. 2(4): 755-764.

(12) GARDNER, M. F., and BARNES, J. L.
1942. TRANSIENTS IN LINEAR SYSTEMS, V. 1: lumped constant systems. 389 pp.,
illus. New York.
(13) GUILLEMIN, E. A.
1949. MATHEMATICS OF CIRCUIT ANALYSIS. 590 pp., illus. Cambridge, Mass.
(14) KENDALL, M. G., and STUART, A.
1958. THE ADVANCED THEORY OF STATISTICS. V. 1, Ch. Ill: moments and cumulants. London.
(15) LEE, Y. W.
1960. STATISTICAL THEORY OF COMMUNICATION. 509 pp., illus. New York.
(16) LEVI, E., and VALDES, R.
1964.
A METHOD FOR DIRECT ANALYSIS OF HYDROGRAPHS.
Jour. Hydrol. 2(2):
182-190.
(17) NASH, J. E.
1959.
SYSTEMATIC DETERMINATION OF UNIT HYDROGRAPH PARAMETERS.
Jour.
Geophys. Res. 64(1): 111-115.

(18) O'DONNELL, T.
1960.
INSTANTANEOUS UNIT HYDROGRAPH DERIVATION BY HARMONIC ANALYSIS.
Internatl. Assoc. Sei. Hydrol. Pub. 51: 546-557.
(19)
1966.
METHODS OF COMPUTATION IN HYDROGRAPH ANALYSIS AND SYNTHESIS.

PART III, RECENT TRENDS IN HYDROGRAPH SYNTHESIS. TNO PrOC. and
Inform.
(20) PAPOULIS, A.
1962.
Note 13.
The Hague.
THE FOURIER INTEGRAL AND ITS APPLICATIONS.
327 pp.
NeW York.
(21) PAYNTER, H. M.
1952.
METHODS AND RESULTS FROM MIT STUDIES IN UNSTEADY FLOW.
BostOU
Soc. Civ. Engin. Jour. 39(2): 120-165.

(22) RESTREPO, J. C. O., and EAGLESON, P. S.
1965. OPTIMUM DISCRETE LINEAR HYDROLOGIC SYSTEMS WITH MULTIPLE INPUTS.
Mass. Inst. Technol., Hydrodyn. Lab. Rpt. 80.
(23) ROBERTS, G. E., and KAUFMAN, H.
1966. TABLE OF LAPLACE TRANSFORMS. 365 pp., illus. Philadelphia.
LECTURE 6:
IDENTIFICATION BASED ON DISCRETE DATA
Basic System Equations
Because hydrologie systems are continuous systems with continuously defined inputs and outputs, it might be thought that the methods of system
identification described in lecture 5 would be the most appropriate techniques
to use in identifying hydrologie systems. In many cases, however, hydrologie
data are only available in discrete or quantized form. A good deal of rainfall
data is only reported as hourly volumes, and the input in these cases is represented by a number of square pulses because all that we know are the volumes
(or the mean rates) of rainfall during each interval. Modern recording equipment is usually digital in form, but the frequency of sampling is so high that
the records could, if necessary, be treated mathematically as a continuous
record without appreciable error. Even where a continuous or virtually continuous record is available, it may be uneconomic to process the complete
record. In this case, the record will be sampled and the sample data processed
in some way. The data, though actually recorded in continuous form, must be
considered discrete data for the purpose of analysis.
If an attempt is made to analyze a square pulse by a series of continuously
defined orthogonal functions, serious difficulties of representation arise. Even
if a large number of terms is used in the series, the discontinuities at the beginning and the end of the pulse will not be faithfully reproduced and oscillations, known as Gihbs' oscillations^ will occur. In harmonic analysis, certain
mathematical techniques are available for the smoothing of these oscillations.
It seems preferable, however, to accept the discontinuous nature of the data,
and instead of looking for the impulse response of the system, to try and
identify the response of the system to a square pulse of standard length. In
effect, this means seeking the finite period unit hydrograph rather than the
lUH and, thus, returning to the basic concept used during the original development of unit hydrograph methods.
If we define the pulse response UD) as response of the system to an input
of unit volume occurring at a uniform rate for a period D, then as expressed
in equation 24a, lecture 1, the relationship between input and output is given
by:
y{t)= X{aD)hD{t-<TD)
(1)
(rD=0
127
where X{aD) is the volume of input in the interval from time aD to (a+l)D
and y{t) is the continuous output. In the above case, both the finite pulse
response and the output are continuously defined.
If the pulse response is only defined at certain intervals; then as expressed
in lecture 1, equation 27a, the relationship between input, output, and pulse
response is defined in terms of the discrete variables s and a:
ff=S
yisD)= 'X{cD)hD(sD-aD)
(2a)
<r=0

yis)= J2Xi<T)hD(s-cr)
(2b)
where the interval between ordinates is D. It is necessary for the interval at

which the impulse response and output are determined to be a submultiple
or a multiple of the unit period of input, D. Otherwise, interpolation will be
necessary before the ordinates making up the output are summed together.
If the output is taken in block form, as in Bernard's distribution graph, then
we have:
yis)= XX(<7)dj,s-(T)
(2c)
<r=0
where do represents the distribution graph, that is, the distribution of volume
of output for the unit period of input.
The above equations are for the case where the input is defined in terms of
volumes. If the input is defined in terms of discrete ordinates, then the equation corresponding to equation 2b would be:
(T=8
y{s)= J^xia)hDs-(T)D
(2(i)
<r0
for the discrete input x(s).

The convolution relationship of equation 2 can be written in matrix form.
(See lecture 1.)
{y]p+i,l=L^']p+l,n+l{h}n+l,l
(3)
where y is the vector of outputs, X is the matrix of inputs, and h is the vector
of the pulse response. The general structure of the matrix X formed from the
input vector x is indicated in lecture 1. A typical matrix of inputs is formed as
129
follows :
X=
xo
Xi
a;o
X2
Xi
XQ
Xz
X2
Xi
Xo
0^4
Xz
X2
Xi
X4
Xs
X2
0^4
0:3
0:4
(4a)
The above example shows the case where the input lasts for five unit periods
and the pulse response has four ordinates; it illustrates the general method of
forming the matrix which might be called the convolution form of matrix.
For the above case, the output has eight ordinates in accordance with the
general relationship p = n+m, where there are m blocks of input, n+1 ordinates of the pulse response, and p+1 ordinates of output. It is equally possible to leave the input as a vector and convert the impulse response into a
matrix:
{y}î.i=HJp+l,m+ll
im+1.1
(4b)
The form given in equation 4a is most convenient in the identification of the

pulse response; form 4b is most convenient where a derived pulse response
has been adjusted to eliminate anomalies and where it is required to ascertain
what corresponding adjustment in the input should be made.
This lecture is concerned with the various methods which might be employed to solve equations of the forms given above. In contrast to the case of
continuous data where the solution of an integral equation was called for, in
discrete or quantized data it is only necessary to solve a set of simultaneous
equations. Consequently, we would expect that matrix methods would be
applicable to the identification problem for discrete data. We would also expect that discrete versions of the various transform methods and of time
series analysis described in lecture 5 would also be available. These are discussed in the remainder of the lecture.
Matrix Methods of Identification

If the available input and output were completely free from error there
would be no difficulty in determining the ordinates of the pulse response or
finite period unit hydrograph. The set of simultaneous equations given in
matrix form in equation 4a can be written out as :
yo = Xoho
(5a)
yi = Xiho+Xohi
(5b)
y2=^X2ho+Xihi+xoh2
(5c)
y i = Xiho-\
(5i)
\-Xohi
2/m-l = Xm-lhQ + Xm-2hi+ +Xohm-l

y m = XmK-\
\-Xihm
2/p_l = Xmhn-1 + Xm-lhn

yp=
XrJln
(5m)
(5n)
(5p)
If the output and input are known, then all the values of the vector Xo, Xi^
X2
Xm and of the output vector 2/0,2/1,2/2
?/p-i, y^ are known. The
values of the unknown ordinates of the pulse response or unit hydrograph,
that is, Ao, /il, /i2
hn-ij hn can be determined successively from the set of
equations 5. Thus, equation 5a is used to obtain the value of ho; substitution
of this value in equation 5b enables us to calculate the value of hi; and so on
until all the unknown ordinates of h have been determined. Where there is no
error in the data, the values obtained by the solution of the first n equations
automatically satisfy the remaining equations. This method of solution by
forward substitution is equivalent to solving a subset of the equations 5a to
op in the form :
{^}(n4-l) = [^](n+l){/i}(^+l)
(6)
which indicates that only the first (n+1) values of output and the first
(n+1) rows of the X matrix are used. There are now the same number of
equations as unknowns, so that direct algebraic solution is possible. We also
note that the matrix of inputs X is now a square matrix and this can, therefore, be inverted. There is, of course, no necessity to use the first (n+1) rows
to form the (n+1) by (n+1) matrix. Any (n+1) rows could be used, but
if the first (n+1) rows or the last (n+1) rows are used, the matrix is triangular and therefore more easily inverted.
An alternative way of equating the number of equations and the number
of unknowns is to treat the unit,hydrograph or pulse response as if the num-
131
ber of its ordinales were equal to the number of ordinates of runoff. In this
case, the matrix equation becomes:
{2/}.+i=mp-fi{Mp+i
(7)
Again, the equations can be readily solved by direct algebraic methods. In

the absence of errors in the input or output data, only the first n-ordinates
will have significant values, and the ordinates of the unit hydrograph between
hn+i and hp will come out as identically zero.
The rule for calculating any ordinate of forward substitution (that is, the
use of the first (n+1) equation is:
XQ
/i,= ^^^-^li^^^^^^^^' for i>m
(8b)
Xo
which can be solved successively for i = 0, 1, 2,

n.
For backward substitution (that is, the use of the last (n+1) equations),
the corresponding formulas are :
hn-i^
for
%<m
(8c)
Xm
K-i = ^^"~ ^^'""'^"'-'+''^"-^' for i>ni
(8d)
Xm
which can be solved successively for i = 0, 1, 2

n.
In the absence of errors in the data and of errors of computation, it is immaterial which set of (n+1) equations are used to solve for the (n+1) unknown ordinates of the unit hydrograph. The direct solution by forward
substitution (or backward substitution) is, however, unreliable in practice
due to the presence of error. It can readily be shown by the use of synthetic
data that if errors occur in the measurement of the input or the output, unrealistic unit hydrograph ordinates are obtained in the solution. We are thus
faced with the problem of finding an optimum solution for the unit hydrograph using all the information available.
The matrix method based on least squares solves the problem in the form
of equation 4a. It assumes that the length of the unit hydrograph is known
by subtracting the length of the input from the length of the output and,
therefore, that the ordinates of the unit hydrograph between hn+i and hp are
zero. In the presence of error, therefore, we must restrain these particular
ordinates and distribute the error on the other ordinates.
The method of least squares is based on minimizing the sum of the squares

of the residuals between the actual output and the output predicted by using
any particular value of h. The residuals are given by the column vector:
{r}p+i,i=-- {2/}>+i,i-[^]p+i.n+i{^}n+i.i
= {y-Xh}pî,i
(9a)
(9b)
the sum of the squares of these residuals is most conveniently got by using
the inner product, that is, by multiplying r by its transpose r^. Taking the
inner product (see page 99, lecture 1) and using the rule for the transpose
of a product, we obtain:
2] r.^^r'^r^' iy^-h''X''^ly-Xh:\
(10)
On multiplying out the terms, we get:

E r\ = y''y-y'^Xh-h^X'^y+h^X^Xh
(11)
Since h and y are column vectors, their transposes will be row vectors, and,
consequently, the second and third terms on the right-hand side of equation
11 are scalar in form. Since a scalar transposes into itself, the two terms must
be equal so that we can write:
Z r\=^y'^y-2h^X^y+WX^Xh
(12)
The problem is to choose the ordinates of the response vector h so as to

minimize the expression given in equation 12. For ordinary functions, this can
be done by taking each ordinate in turn and setting the partial derivatives,
with respect to that particular ordinate equal to zero. However, the compression of vector notation may be used. Because the first term we are differentiating does not involve /i, the derivative for this term will be zero. Vector
differentiation of the second term on the right-hand side of equation 12 with
respect to h resembles the ordinary differentiation of the first power of a
function. Similarly, vector differentiation of the third term on the right-hand
side of equation 12 with respect to h resembles the ordinary differentiation of
the second power of a function. The result of differentiation with respect to h
can be readily verified. Setting the result equal to zero is given by:
-2X^y+2X^Xh==0
(13a)
or
X^Xh^X^y
(13b)
The vector h which satisfies equation 13b makes ^ TI^ a minimum. It is

therefore the best least squares solution to the original set of equations 5a to
5p. To solve equation 13b for /i, it is necessary to invert the matrix given by:
Z = X^X
(14a)
133
thus obtaining the solution

h = Z-'X^y
(14b)
Since the matrix formed by multiplying any matrix by its transpose is necessarily a square matrix, this inversion can be carried out.
Note that the sum of the squares of the residuals for the above solution is
not an absolute minimum. It is a minimum subject to the restraint that h has
a base length from zero to n+1, that is, that the values of the ordinates from
hn+i to hp are zero. The effect of other constraints and of errors generally will
be discussed later in the lecture.
Discrete Transform Methods

Transform methods are available for handling discrete data which correspond to the transform methods for continuous data discussed in lecture 5,
'Identification Based on Continuous Data." Thus the classical Fourier series
can be replaced by the finite Fourier series, which will reproduce exactly the
functions involved at the sampled points and can be used to interpolate
trigonometrically between these points. In place of the Laplace transform, we
have the Z-transform which was developed for use with sampled-data systems.
Dooge (written commun., 1966) has derived a discrete analog of the Laguerre
methods of analysis, but this has not yet been fully developed or published.
The method of harmonic analysis has been applied to hydrologie data by
O'Donnell (P, 10). If an output is specified at a number of equidistant discrete points, then it can be fitted exactly at chese points by a function of the
form :
y(s) =~+ E A, cos [k-^)+ g B, sm i^-^j
(15)
where n = 2p + l is the number of data points. Since there are only n-pieces
of information, it is impossible to derive more than ?i-meaningful coefficients
Ak and Bk for the data. Since sines and cosines are orthogonal under summation, the coefficients can be determined from:
A, = -Z 2/(8) cos (^
(16a)
s=0
Bk = - J^yis) sinii^-^\
)
n ,=0
(16b)
\ ^^ /
where k can take on the integral values 0, 1, 2
p-1, p. The above formu-
lation can also be expressed in the exponential form:

y{s)== XCkexpl
k=^-p
(17a)
n
Ck = - 22y(s) expi
(17b)
For a linear time-invariant system which is causal and has an isolated input, the discrete ordinates of the input, output, and pulse response are connected by:
y{sD)= ^X(<TD)hDsD-aD)
(18a)
where X represents the volume of input in successive unit periods of length,

D; ho represents the finite period unit hydrograph for the unit period (D)
defined at intervals equal to the unit period; and y represents the output defined at intervals equal to the unit period (D). For convenience, the unit
period can be taken as the unit of time and the relationship written as:
a=s
y(s)= J^X(a)hD{s-a)
(18b)
<r=0
If the input is of finite duration and the memory of the system is finite, then
we can use finite Fourier series in the same way as infinite Fourier series were
used in lecture 5. The development is analogous and will not be repeated in
detail. The discrete functions representing the input and the output are assumed to be periodic with a period equal to nD. Since the input is periodic, it
is necessary to write the relationship between input pulse response and output
as:
y(s)=
J2
X{(7)hD{s-a)
(18c)
The linkage equation can then be found in similar manner to that indicated
by equations 11 to 16, lecture 5.
By substituting equation 18c in equation 17c, reversing the order of summation, and using the orthogonality relationship twice, it can be shown that
in the discrete case, we have the linkage relationship :
Ck = nCk'ak
(19)
which is the same as the linkage equation given by equation 16 of lecture 5,

except for the fact that the symbol n is used for the period in the discrete
case, the symbol T for the period in the continuous case. For the expansion
in trigonometrical rather than exponential form, the linkage relationship
takes the form :
135
A, = l{a,a,-h,)
(20a)
B, = l{a,,-ha,)
(20b)
which correspond to equations 18b and 18c, respectively, in lecture 5.

The fact that the trigonometrical functions are orthogonal under both integration and summation results in the same linkage relationship for continuous
and discrete data. The coefficients appearing in equations 19 or 20 of the
present lecture are frequently referred to as harmonic coefficients and the
coefficients appearing in equation 18 of lecture 5 are referred to as Fourier
coefficients. The differences between the two cases should, however, be clearly
recognized. Firstly, the coefficients a^ and ft in equation 20 of this lecture,
define the finite period unit hydrograph Ifi^î) ; the corresponding coefficients
in equation 18, lecture 5, define the instantaneous unit hydrograph /io(0Secondly, the coefficients of equation 20 of this lecture are finite in number
because only as many coefficients as there are data points can be determined
altogether. For continuous functions, there is no limit to the number of coefficients which can be calculated if required. Thirdly, the coefficients of
equation 20 of this lecture, when substituted into the finite Fourier expansion,
define the pulse response IXB at discrete points which are equally spaced at
the unit interval, D; whereas, the coefficients derived in lecture 5 define the
lUH continuously.
If a function only exists at discrete points, or is only known at discrete
points, it is not possible to obtain its Laplace transform directly. Such a function can be considered as being defined by:
/(0= :f{t){t-nT)
(21)
n=0
where n is an integer and T is the interval between data points. If the Laplace
transform is now taken, we have:
[/(0]= /(nT)6^
(22)
n=0
It is customary to write:
Z= expisT)
(23)
and hence to write what is known as the Z-transform of the discrete variable
as:
i^(Z)=Z[/(nT)]= J2finT)Z--
(24)
The properties of the Z-transform (4) are similar to those of the Laplace
transform, in particular, for a linear time-invariant causal system given by:
<r=s
y{s)= J2x{a)h{s-a)
(25)
(7=0
We have the following simple relationship between the Z-transform of the

input:
YiZ)=HiZ)^X{Z)
(26)
If y{s) and x(s) are given numerically, it is possible to compute Y(Z) and
X{Z) and, hence, to determine the Z-transform of the pulse response:
^(Z)=g|
,27,
If H{Z) can be expanded in inverse powers of Z, the coefficients of the expansion will give the ordinates of h{s) since by definition:
H{Z) =h(0)+h{l)Z-'+h{2)Z-'+
(28)
In practice, however, it is likely that, as in the Laplace transform^ the Ztransform will not be easily inverted in practical cases where we have numerical data rather than a mathematical function.
The other transform method discussed in this lecture corresponds to the
Laguerre analysis of systems with continuous inputs and outputs. If an attempt is made to represent a square pulse by a series of Laguerre functions,
the discontinuity cannot be well represented even if the number of terms in
the expansion is quite high; for 50 terms, the oscillations will be of the order
of 25 percent of the height of the pulse. Accordingly, if it is wished to use
quantized data, the method of Laguerre analysis described in lecture 5 is no
longer adequate without modification. At first, it was hoped that the Laguerre
functions might be orthogonal under summation as well as integration, as
with trigonometric functions. Unfortunately, this did not prove to be the
case, and it was necessary to derive the discrete analog of the Laguerre
functions.
Though some books on numerical analysis mention that discrete analogs of
the classical continuous orthogonal polynomials do exist, no description of
these was available in any of the literature cited. Eventually, the form of
the polynomials orthogonal under -summation from zero to infinity with respect to a damping factor exp( s) was derived from first principles. It is
1 DOOGE, J. C. I. LINEAR THEORY OF OPEN CHANNEL FLOW, III: MOMENTS AND CUMULANTS. Dept. Civ. Engin., University College, Cork, Ireland. 1967. [Unpublished report.]
137
easier, however, in retrospect to derive the form of the discrete analog to the
Laguerre function by an analogy between discrete and continuous operations.
Integration in the continuous case becomes summation in the discrete case
and differentiation in the continuous cases is replaced by forward differencing
in the discrete case. The Laguerre polynomial^ is defined by:
in(0= Z(-l)*(*)*^,
(29a)
In the above equation, the continuous variable t occurs in the form t^/k\.
This function has the property that when differentiated with respect to , it
maintains the same form but with the order reduced by one degree. The
analogous discrete polynomial might be expected to be that function of the
discrete variable s which has the analogous discrete property, that is, the
function when forward differenced with respect to s maintains the same form
but with the order reduced by one degree. It can be verified that the form of
discrete function required is the binomial coefficient (l). Hence, we would
expect the discrete analog to the Laguerre polynomial to be of the form:
Mn{s)=T,i-lV{l)il)
(29b)
k=0
It can be shown that the above expression is a Meixner polynomial with a

value of 6 = 0 and c = 3^.
The Laguerre polynomials defined by equation 29a are orthogonal in the
range zero to infinity with respect to the weighting factor exp( ), that is.
e-'Lm{t)Ln{t)=mn
(30a)
If the Meixner polynomials are to be written in similar form, it is necessary

not only to replace the integration by a summation but also to find the discrete
analog of the weighting factor exp( 0. The reciprocal of the weighting function in the continuous case, exp(0, differentiates into itself. This suggests a
function which forward differentiates into itself as a discrete analog. It
may be readily shown that 2^ has such a property and consequently, 3^^ may
be tried as a weighting factor. When this is done one obtains the orthogonal
relationship :
E {y2yMUs)Mn{s) =2-^'bmn
' Op. cit. p. 88, lecture 1, and p. 173, lecture 5.
(30b)

The normalized orthogonal function in the continuous case is:
/n(0=e-^/^f (-1)^(;;)^
(31a)
and the normalized orthogonal function in the discrete case is:

/.() = (M)<+''+^>'^ E (-1)*(^)(^)
(31b)
A;=0
The function given in equation 31b may be described as a Meixner function

and its properties explored by the use of the Z-transform (6).
To derive a method of Meixner analysis, it is necessary to determine first
of all the effect of convoluting one Meixner function with another as follows:
g(s)= T.fmMfnis-a)
(32)
g{s)^^{2yi%^n{s)-Un+^{s)
(33)
It can be shown that :

Again the result is similar to that for Laguerre functions except for the scale
factor 2^'^. The linkage equation is derived in a similar manner as for Laguerre
functions and is given by:
k=p
k=pl
A^= ^ {2yi'a,a^_,- a,ap_x_i

k=0
(34)
k=0
which corresponds with equation 73 of lecture 5. For the identification problem, it is necessary to determine successively the values of the Meixner coefficients for the response functions. These are given by:
apao= J2(y2)^'-''-'^"Ak- Z aka,.k
;=0
(35)
k=0
The method of Meixner analysis is still under development and so far has
only been applied to synthetic data. There is some indication that it is not
as numerically stable as the Laguerre analysis of continuous data. In Meixner
analysis, as in Laguerre analysis, it is necessary to choose an appropriate time
scale to represent the functions involved by a relatively small number of
coefficients.
Time-Series Analysis of Discrete Data
The method of time series analysis developed by Wiener (7) and used in
the theory of communication has been applied to discrete hydrologie data by
the group working at the Massachusetts Institute of Technology under
139
Eagleson {3). The problem is to determine for a given set of discrete inputs
and outputs, the causal linear response which is optimal in the least squares
sense. It is necessary to define the discrete analogs of the autocorrelation and
cross correlations defined by equation 82 and 83 of lecture 5 for the continuous case. The autocorrelation function for a discrete variable f{s) is defined
as the limit :
0//W=- /(s)/(s+^)
as
p-ôo
(36a)
^ s=-p
where n = 2p+lis the number of data points. The cross correlation function
between two discrete variables/(s) and g{s) is defined as the limit:
0/.W=- ilf(s)g{s+k)
as
p-^co
(36b)
For a causal linear time-invariant system, we have :

y{s)=
ix{a)hn{s-a)
(37a)
(r=00
or
y{s)= J2X{s-a)hDM
(37b)
(7=0
where D is the interval between the equally spaced discrete or quantized

data (and consequently also the unit period of the finite period unit hydrograph or pulse response hois)), which will enable us to predict the output
with minimum error. The individual error prediction for the single ordinate
is given by :
ri = yi- J2Xii-a)hnM
(38)
(7=0
where i is the integer denoting the ordinate of output concerned. For a continuous record which has been sampled or of discrete or quantized data, we
wish to minimize the least squares error, that is,
E[/i(s)] = - (r,) 2 = minimum
(39a)
If this is to be done by manipulation of the ordinates of the response function,

then we have the condition :
Pi
fir-
\ Hfi'^ Z2r,;f^ = 0
hU) ,;
,^,
hij)
It is clear from equation 38 that for every value of j greater than 0:
(39b)
-= -X{i-j)
dhij)
(40)
SO that
dVi
(41)
-Ti
The criterion of equation 39 can now be written as:

(42)
The above relationship must hold for all values o j greater than 0. Reversing
the order of summation in the second term gives :
E X(i-j)y(i)-- ZhDia) Z X{i^j)X{i-a)=0
(43)
Using the definition of autocorrelation and cross correlation for discrete functions given by equations 36a and 36b above, equation 43 can be written as:
(44)
<t>xy{j)= ^hD((T)<l)xx(j(T)
provided that the value of ^ is zero or positive. This is the discrete form of the
Wiener-Hopf equation used by Eagleson (3) in the analysis of hydrologie
systems.
Where we are dealing with isolated inputs and systems of finite memory we
also have isolated outputs. In this case, the correlation method of analysis of
time series can be shown to be equivalent to the least squares method. As
shown in lecture 6, page 187, the least squares method arranges the input
and output data in the form:
X'^y = X^Xh
(45)
where the matrix X is of the form given by equation 4b of the present lecture,
that is, the input vector has been used to form a convolution-type matrix.
Accordingly, for the example given on page 39, lecture 1, and page 181, lecture
6, the left-hand side of equation 45 will be of the form:
2/0
Xo
Xi
a;2
:z^3
^4
Xo
Xi
X2
Xz
XA
2/3
Xo
aî
X2
Xz
X4
2/5
Xo
Xi
0^2
Xz
X4
L2/7J
2/1
2/2
X^y =
2/4
2/6
(46)
141
In the above example, the matrix X has four rows and eight colums, and
the column vector y has eight rows. The result of multiplying them together
will be to produce a column vector with four rows. If we follow the rules of
vector multiplication, these four rows will be given by:
Xoyo+Xiyi+X2y2+Xzy3+XAy^ = (l)xy{0)
(47a)
xoyi+Xiy2+X22/3+X32/4+x^ys = <?i>x2/ ( 1 )
(47b)
Xoy2+xiy^+0:22/4+X3/5+x^ye = 0xy ( 2 )
(47c)
xoys+Xiy 4+X2y5+x^ye+Xiy^ = 0X/ ( 3 )
(47d)
The left-hand side of equation 45 is therefore seen to be of the same form as

the left-hand side of equation 44.
Similarly, it can be shown that:
'<t>xx(0)
<^.x(-l)
<>xx(-2)
<..x(-3)'
<>xx(l)
<t>TziO)
4>U-l)
<t>.z{-2)
X^X =
(48)
_<#>ii(3)
4>xx{2)
4>xxW
4>xx(0)
which can be seen to be a convolution-type matrix formed from the autocorrelation coefficients of the input vector. Because the multiplication of this
convolution-type vector by the optimum response vector h is equivalent to
the convolution of the input autocorrelation coefficient vector and the response vector, the right-hand side of equation 45 is equivalent to the righthand side of equation 44.
In time series analysis and the use of equation 44, the number of equations
is not limited as in equation 47. If the system being investigated were a truly
linear system with a finite memory and the input and output data were free
from error, the cross correlation coefficients for values of j greater than the
memory of the system would be zero, and the number of equations would be
the same in both methods. If, howêver, the system is not truly linear, or if
there are errors in the data, the time series correlation method gives additional equations which can be included in the set of equations to be solved. In
the latter case, the number of equations to be solved wall be greater than the
number of ordinates required in the pulse response and, hence, restraints can
be introduced into the solution.
Computational Methods
In a truly linear system in which the input and output are given in the discrete form and can be determined without error, all of the methods of system
identification described above will give the same answer within the limits of
computational error. For this case of error-free data, the choice between
methods is merely one of the ease of computation, and there is no reason to go
beyond the direct solution of the simultaneous equations involved by the
method of forward substitution. For these ideal conditions, once the number
of equations solved corresponds to the length of memory of the system, all
the remaining equations will be automatically satisfied by the values for the
ordinates of the output response already found. If, however, there are errors
in the data, or if the system is not truly linear, then the values of the ordinates
obtained for the optimum linear response may vary according to the method
used. In this more general case, the choice between the methods depends not
only on the convenience of computation but also on the manner in which the
various methods handle errors in the data and linearize any nonlinear properties of the system under identification. Except for the basic case of solution
by forward substitution, all of the methods require the use of a high-speed
digital computer unless the problem is trivially small. The techniques used in
the actual computations involved in the different methods are described in
the Hterature cited at the end of this lecture.
Only the essential features need be mentioned here. In the least squares
method. Body (2) suggested that the data be loaded into the computer as a
single unit in the form:
[Z/i/],.^4-i
(49)
In the least squares method, the input consists of (m+1) ordinates of the
input and (p+1) ordinates of the output. A convenient way of organizing
the calculations is as follows. The input data can be used to compute the
elements of:
Z = X^Z
(49a)
which we have already seen to be the discrete autocorrelation coefficient of
the input. It is also necessary to calculate the elements of:
Tf = Z^y
(49b)
which are the cross correlation coefiiicients of the input and output. A standard
routine for matrix inversion can now be used to solve for the unknown finite
period unit hydrograph as follows :
h^Z-W
(51)
Z will be a square matrix of size (p m+l) and W will be a column vector

with (n+1) rows. The unknown pulse response h will also be a column vector
with (n+1) rows.
Once again, it must be emphasized that the least squares method involves
optimization subject to the restraint that the length of the response function
143
does not exceed the amount given by the difference between the length of the
output and the length of the input. The prediction of the output using the
finite period unit hydrograph, which is optimum in the least squares sense,
will not be as good as the prediction of the output by a unit hydrograph,
which is allowed to be of the same length as the output. However, the use of
the method of least squares reduces the tendency towards unrealistic negative
or wildly oscillating ordinates which may occur with the forward substitution
method where there are errors in the data. If it were certain that the system
were linear and that these U7irealistic values were solely due to errors in the
data, then there is a strong argument for introducing the restraints involved
in the least squares estimate. However, if negative ordinates result from the
attempt to represent a nonlinear system in a linear fashion, then the case for
rejecting the negative ordinates is not nearly as strong.
For computation, further restraints are sometimes introduced into the calculation. Thus, Body (2) made the assumption that from a certain point
onward the finite period unit hydrograph (pulse response) shows an exponential decline and made use of this assumption to reduce the amount of
computation required. Similarly, Newton and Vinyard (5), in their description of the Tennessee Valley Authority method, referred to the introduction
of the restriction that the ordinates of the pulse response may be replaced
over a number of intervals by a straight line, thus simplifying the numerical
computation at the cost of this restraint.
The sequence of computations is standard for any transform method based
on orthogonal polynomials (10). The first step is to read in the input and
output data and compute the coefficients of the input data (c) and of the
output data (Cn) for the particular orthogonal expansion assumed. The corresponding coefficients for the pulse response or finite period unit hydrograph
(jn) are then calculated from the linkage equation. These coefficients in the
expansion of the pulse response can then be used to find the actual ordinates
of the pulse response. In the harmonic analysis method applied to hydrologie
data by O'Donnell (9), the actual period of runoff, when divided into standard
intervals, provides a finite number of equidistant data points. The number of
harmonic coefficients derived is the same as the number of data points. In
Meixner analysis, however, the range is from zero to infinity so that account
is taken not only of the finite number of data points in the isolated output
record but also of the infinite number of points after the conclusion of output.
For perfect matching, it would be necessary to take an infinite number of
terms in the expansion of the pulse response. However, it is only necessary
to carry the series far enough so that the ordinates within the period of output
are determined sufficiently accurately, and errors in the zero ordinates after
the close of output in which we are interested may be ignored.
In time series analysis, the autocorrelation and cross correlation coefficients
of the input and output must first be determined. If we are dealing with an
isolated input to a linear system of limited memory, then the autocorrelation

and cross correlation functions will become zero beyond a certain point. If,
on the other hand, we are dealing wdth a continuous time series, it would be
a matter of decision as to the point at which these functions should be truncated. After the autocorrelation and cross correlation coefficients have been
determined, it is still necessary to solve the A^ iener-Hopf equations given by
equation 44. If all of the equations are used, then it will be possible to predict
the output closely but unrealistic ordinates may be obtained. Eagleson (3)
introduced the idea of solving the equations subject to the restraint that no
negative ordinates occurred. This necessitates a computation of a linear programing solution to the W iener-Hopf equations.
The whole subject of the comparison between the various methods for system identification in the presence of noise and of possible nonlinearity is one
requiring careful investigation. Several research workers are known to be
working on the problem at the moment, but none of the results have so far
been published. It may be instructive to consider briefly the effect of an error
on a simple case using synthetic data. The data used are those in problems
1 and 2 at the end of this lecture. If the values of the input and output from
a system are given as:
a: = 2, 6, 1
(52a)
2/= 0,4, 14,8, 1,0
(52b)
any of the methods described above can be used to show that the linear pulse
response for the system is given by :
/i = 0, 2, 1, 0
(53)
If however, the output were mistakenly given as :

y = 0,4, 17,8, 1,0
(54)
then, the estimates of the optimum linear pulse response would vary with
the method used. In this case, the method of direct forward substitution
would give the linear response as :
h = 0, 2, 2.5, -4.5, 12.75, -36
(55)
which is clearly unstable.

In using the method of least squares, it is necessary to decide the length of
the pulse response to determine the size of the input convolution matrix. The
output is given at six points and the input is given for three standard intervals. It could, therefore, be assumed that the pulse response, which is the
response due to input in one standard interval, would not exceed four intervals
in length. Assuming the pulse response (the finite period unit hydrograph) to
be four intervals long, the least squares method gives as the optimum pulse
145
/i=-0.1o, 2.31,0.82, 0.02
(56)
response:
This result is seen to give an unrealistic negative ordinate at the beginning of
the impulse response. If the restraint were inserted that this ordinate should
be zero, the result obtained would be:
/ = 0, 2.23, 0.76, 0.01
(57)
The latter result is seen to be not too different from the true pulse response
given by equation 53. However, even ordinates as small as the fourth ordinate
of 0.01 have an effect on the solution. If the fourth ordinate were constrained
to be zero, the result wôuld be:
/i = 0, 2.18, 0.82,0
(58)
It can be seen from the above series of results that the more information
concerning the realistic form of the pulse response that is fed into the computation, the closer the result will be to the true pulse response, which has been
masked by the error in the output. Similar variations in the result are obtained
in the correlation method if a certain number of Wiener-Hopf equations are
chosen or if additional restraints are placed on the problem.
Problems on Discrete Systems Identification

1. If the input in a system is given in Appendix table 2 by function 3 and
the output by function 4, find the unit pulse response of the system both by
the direct algebraic method and by the least squares solution.
2. For the data of problem 1, find the autocorrelation function of the input
and the cross correlation function between the input and the output. Write
down the set of discrete Wiener-Hopf equations for these particular data.
Verify that the solution obtained in problem 1 is a solution to the latter
equations.
3. Use the Z-transform to identify the pulse response of the system in
problem 1 for the given input and output.
4. Solve problem 1 by either harmonic analysis or Meixner series.
5. In problems 1 to 4, what is the effect on the solution if the output function correctly given by function 4 in Appendix table 2 is mistakenly taken as
function 5 in Appendix table 2?
6. In Appendix table 2, if the input to a linear time invariant system is
given by the discrete function 18 and the output by the discrete function 19,
find the pulse response of the system by using Meixner series.
7. Find the pulse response for the data of problem 6 by using harmonic
analysis.
8. Appendix table 4 gives the effective rain and the storm runoff for a
rainfall event on the Ashbrook catchment. Find the unit hydrograph for the
catchment by a matrix method.

9. Find the unit hydrograph for the data in Appendix table 4 by using
harmonic analysis.
10. Find the unit hydrograph for the data in Appendix table 4 by using
Meixner series.
11. Find the harmonic coefficients or the Meixner coefficients of the functions for a number of the discrete functions given in Appendix table 2.
12. Find the Z-transform of the function for a number of functions given
in Appendix table 2.
Literature Cited
(1)
BAYAZIT, M.
1966.
INSTANTANEOUS UNIT HYDROGRAPH DERIVATION BY SPECTRAL ANALYSIS AND
ITS NUMERICAL APPLICATION. Central Treaty Organization Symposium on

Hydrol. and Water Resource Devlpmt. Proc. Ankara.
(2)
BODY, D. N.
1959.
(3)
FLOOD ESTIMATION:
DIGITAL COMPUTER.
Sydney.
GRAPH PROCEDURES UTILIZING A HIGH-SPEED

Water Res. Found. Austral. Bui. 4. 41 pp., illus.
UNIT
and MARCH, F.
THE COMPUTATION OF OPTIMUM REALIZABLE UNIT HYDROGRAPHS. Water
(4) JURY, E. I.
1964. THEORY AND APPLICATION OF THE Z-TRANSFORM METHOD. 330 pp. NeW
EAGLESON, P. S., MEJIA-R, R.,
1966.
York.
(5) KULANDAISWAMY, V. C.
1964. A BASIC STUDY OF THE RAINFALL EXCESS-SURFACE RUNOFF RELATIONSHIP
IN A BASIN SYSTEM. Ph.D. thesis, 111. Univ. (University Microfilms
Publication No. 64-12535.)
(6) LEE, P.
1966. PROPERTIES OF MEIXNER POLYNOMIALS. Dept. Elect. Engin. Monash
University, Melbourne.
(7) LEE, Y. W.
1960. STATISTICAL THEORY OF COMMUNICATIONS. 509 pp., illus. New York.
(8) NEWTON, D. W., and VINYARD, J. W.
1967. COMPUTER-DETERMINED UNIT HYDROGRAPH FROM FLOODS. Amer. Soc. Civ.
Engin., Jour. Hydraulics Div. 93(HY5): 219-235.
(9) O'DONNELL, T.
1960. INSTANTANEOUS UNIT HYDROGRAPH DERIVATION BY HARMONIC ANALYSIS.
Internatl. Assoc. Sei. Hydrol. Pub. 51, pp. 546-557. Gentbrugge.
(10)
1966. METHODS OF COMPUTATION IN HYDROGRAPH ANALYSIS AND SYNTHESIS.
PART III, RECENT TRENDS IN HYDROGRAPH SYNTHESIS. TNO ProC. aild
Inform. Note 13, 39 pp. The Hague.
(11) RESTREPO, J. C. O., and EAGLESON, P. S.
1965. OPTIMUM DISCRETE LINEAR HYDROLOGIC SYSTEMS WITH MULTIPLE INPUTS.
Mass, Inst. TechnoL, Hydrodyn. Lab. Rpt. 80. Cambridge.

(12)
147
ROCHE, M.
1965.
POINT DE VUE MATRICIEL SUR UN OPERATEUR LINEAIRE DE TRANSFORMATION

PLUIES-DEBITS PRIMARY POINT OF VIEW ON A LINEAR OPERATOR FOR TRANS-
FORMATION OF RAINFALL INTENSITY]. Proc. Symposiui! on Representative

and Experimental Areas (Budapest). Internatl. Asoc. Sei. Hydrol. Pub.
66(1): 234-246.
(13)
SINGH, K. B.
1962.
A NON-LINEAR APPROACH TO THE INSTANTANEOUS UNIT HYDROGRAPH THEORY.
Ph.D. Thesis, 111. Univ. Urbana. [Xerox copy available by purchase

from University Microfilm, Inc., P.O. Box 1346, Ann Arbor, Mich., 48106,
Pub. 62-6231.]
(14)
SNYDER, W. M.
1955.
HYDROGRAPH ANALYSIS BY THE METHOD OF LEAST SQUARES.

Civ., Engin. Jour. Hydraulics Div. 81: 1-25.
Amer. SoC.
LECTURE 7:
SIMULATION OF HYDROLOGIC SYSTEMS
Basic Ideas in Simulation
Having spent three lectures on the problem of analysis, we now turn to the
question of synthesis or simulation. It will be recalled from lecture 1 (see
pages 5-7, 24, and 27) that simulation consists essentially of synthesizing
a system (abstract or real) which will operate on the given inputs so as to
produce an output which will approximate the output of the prototype system
within a given degree of accuracy. Chorafas {18) has defined simulation as:
^^Simulation is simply a working analogy. Analogy means similarity of properties as relations without identity."
A model or a simulation reproduces some but not all the characteristics of
the prototype. Ideally, we might expect the simulating system to reproduce
the behavior of the prototype system exactly, but to do this the simulating
system would have to be as complex as the prototype. It is necessary to fix
the accuracy required and to choose the features of the prototype system
operation which we hope to imitate. Any attempt at simulation is intimately
tied up with standards of accuracy and with a definition of objectives. Unless
we are explicit on these matters, our simulation will not be scientifically
respectable.
In many parts of hydrology, as in many parts of mechanics, we simulate
the action of the system in which we are interested by a set of mathematical
equations. Thus, we can simulate the physical problem of open channel flow
by the equation of continuity and the dynamic equation. Already, two successive simulations are involved. The finite difference algorithm may then be
simulated on a digital computer so that there is a further degree of removal
from the original physical problem. At each level of simulation there is a
danger that the simulating system will not correspond in some important respect to the system it is attempting to simulate. At each level, we must ensure
that our imitation is sufficiently accurate for our purpose.
In open channel flow, we must be satisfied with the validity of the equations
of continuity and momentum; we must be satisfied that our finite difference
scheme is stable, convergent and accurate; we must be satisfied that our computer program does not involve an undue buildup of round-off error; and so on.
In devising a simulating system, it is necessary to compromise between
simplicity in the model and accuracy of prediction of the prototype behavior.
A simple system may simulate a prototype system to a high degree of accuracy
without resembling that system. In network theory, it is quite easy to show
that certain systems are equivalent to one another though quite different in
structure. It must be remembered that synthetic systems used in simulation
are at best only operationally equivalent to the prototype system.
148
149
Simulation has long been used in hydrology to transfer results from one
watershed to another. This can readily be done if we can find a relationship
between the operational behavior of a watershed for which measurements are
available and the characteristics of that watershed. Thus all the methods for
obtaining a synthetic unit hydrograph used in applied hydrology are methods
of simulating the behavior of an ungaged watershed. Sophisticated methods
of simulation have been introduced into hydrology in recent years. Simulation
is used in stochastic hydrology, where long records of flow are synthesized
from a relatively short historical record and used to study the behavior of a
reservoir or a reservoir system. Complex water resource systems have been
simulated and the decisionmaking process included in the simulation {28,
38, 4S).
In these lectures, however, we are only concerned with the use of simulation
in parametric hydrology. Lecture 8 deals with the question of synthetic unit
hydrographs and lectures 9 and 10 with the mathematical simulation of hydrologie systems by means of mathematical functions and conceptual models.
Accordingly, these two topics will not be dealt with in the remainder of this
lecture. Instead, attention will be concentrated on the basic principles of
simulation and on the remaining types of simulation which can be useful in
hydrology. These may be grouped under the headings of regression models,
digital simulation, analog simulation, and physical models. Since the discussion
ranges over so wide a field, the concern will be with general principles and
basic ideas rather than the details of any particular method of simulation.
The emphasis will be on the essential similarities between the basic steps involved in the different methods.
No matter what the field of application, the type of problem involved, or
the type of simulation, the approach is essentially similar. It is necessary first
of all to decide what type of model is to be used to simulate the action of the
prototype. Having decided on this, it is necessary to choose the components
of the model and their interconnection. Once a trial model has been determined in this way, the ability of the model to simulate the prototype must be
verified on the basis of a record of inputs and outputs measured on the prototype system. For a physical model, this is done by verifying that the model
can predict the output of the prototype system to an acceptable degree of
accuracy for a past event for which records are available. If it is unable to do
so, the model must be modified until adequate simulation is obtained. For a
mathematical model, the verification may consist of applying the model to a
case for which there is a known solution, as a prelude to applying it to a case
in which no solution is known. For a mathematical model, it may be necessary
during the verification phase to modify the structure of the model or to change
the values of some of the parameters of the model to achieve a satisfactory
precision of prediction. In modern hydrology extensive use is made of parametric synthesis in w^hich a form of mathematical or conceptual model is
assumed and the optimal value of the parameter is determined. Parameter

values in a model are said to be optimal when these particular values result
in a prediction of the output which is a closer fit (in some defined sense) to
the output from the prototype than can be obtained with any other parameter
values for the same model.
There is a wide choice among types of models suitable for simulating deterministic hydrologie systems. We may decide to use a physical model or an
analog model; we may decide to use a conceptual model consisting of an arrangement of linear channels, linear or nonlinear reservoirs, thresholds or
feedbacks, and so forth; or we may decide to use a mathematical model to
represent the hydrologie system by a set of mathematical equations. Even
after deciding the general type of model to be used, there are still a number of
matters to be determined. For example, if we have decided to use an analog
technique, we must decide whether we are going to use (a) a direct analog
model in which various sections of the prototype will be modeled directly and
be more or less recognizable in the analog, or (b) a general-purpose analog
computer in which the mathematical behavior of the protype is simulated by
analog components representing specific mathematical operations. If, on the
other hand, we have decided to use a mathematical simulation, there is a
choice between regression models, representation of the system by a set of
differential equations, or representation of the system operation by a mathematical curve belonging to some particular family. In the case of a conceptual
model, it is necessary to decide whether the model is to be linear or nonlinear,
whether thresholds are to be included, what particular types of component
are to be used, and how they are to be connected together.
In some types of hydrologie simulation, it is usual to determine parameter
values on the basis of field measurements or of personal judgment. However,
the operation of this initial version of the model should be thoroughly verified
and the model parameters adjusted until satisfactory operation is obtained.
Only then can the model be safely used as the basis of prediction. For most
mathematical and conceptual models, the values of the parameters are not predetermined but are optimized on the basis of known input and output. There is
some excuse for a lack of objectivity in the optimization process when faced
with ad hoc problems in applied hydrology. Even in this case, however, optimization on an objective basis has the advantage that the results from this
individual study can be comined with others in a meaningful fashion. In
hydrologie research, there is no excuse for avoidable subjectivity. Nevertheless, the hydrologie literature is full of models justified by a single illustration
which shows that the predicted output closely resembles the actual output.
Such 'ôptimization by eye" is incapable of being integrated into a general
body of scientific knowledge and is unworthy of the name of scientific
hydrology.
We can borrow from statistics and numerical analysis a number of criteria
of optimization. These include the methods of moments, least squares, mini-
151
max error, and maximum likelihood. One such technique may be preferable in
one situation, and another in another situation. What is important is that the
method be objective, repeatable, and that it be clearly described in any reporting of the work.
Regression Models
Regression techniques {27^ 57^ 65, 69) are essentially a method of simulation. Their main value is in prediction rather than in the investigation of
causal linkages. Once the decision has been made to use a rgeression method,
it is necessary to decide what type of regression model will be used.
An example {11) of multiple linear regression, a method which has been
widely used in hydrology, is shown on figure 7-1. In this example, the following basic relationship is assumed:
QTâ{A)\S)^{Si)'{I)^{t)f{0)^
(la)
In this formulation, the peak annual ood (Qr), which is taken as the dependent variable, is assumed to be related to unknown powers of the various
watershed parameters (A, , St, I, t, and 0), which are taken as independent
variables. QT is the annual peak discharge in cubic feet per second for a recurrence interval of T years; A is the drainage area in square miles; & is the
main channel slope in feet per mile; St is a measure of the surface storage
area; / is the 24-hour rainfall in inches for a recurrence period of T years;
t is a measure of freezing conditions in midwinter; and 0 is an orographical
factor. If the relationship given by equation la is expressed in logarithmic
form as follows :
logQT = loga+h{\ogA)+c{logS)+d{\ogSt)
+e{logI)+f{logt)+g{logO)
(lb)
then the relationship is linear both in the new logarithmic variables and in
the unknown parameters. Consequently, the unknown parameters {a, 6, c,
d, e, /, and g) can be determined by the standard techniques of multiple
linear regression.
The assumption of the particular relationship given by equation la or
equation lb makes this approach just as much a model as if the variables
were fed into an analog computer. Indeed, to the hydrologist devoted to the
analog approach, the method of multiple linear regression models would appear somewhat as shown in figure 7-1. In this diagram, each of the dependent
variables is fed into a function generator which raises it to the designated
power. The resulting outputs are then multiplied together to give the dependent variable. For the parameters to be optimized either in the regression
equation or in the analog, the value of the exponents of the independent vari-
ables that give the best fit between the predicted peak flows and the observed
peak flows must be determined. The logarithmic transformation of the variables would be paralleled in the analog case by replacing the analog shown in
figure 7-1 by one in which the function generators would transform the independent variables logarithmically and the multiplier would be replaced by an
adder.
QT = a(A)'>(S)C(S,)<^(l)(t)^0)9
^-^
[led
g
0
log QT= loga+bdog A)+c(logS)

+ d(logSt)+e(logl)+f(logt)
+ g(logO)
FIGURE
7-1.Regression analysis.
[lb]
153
The standard regression technique takes as a criterion the minimization of

the sums of the squares of deviations of the predicted values of QT from the
measured values of Qr. The model can be evaluated by examining the value
of the square of the multiple correlation coefficient (^). For a perfect model,
W would be equal to 1 and the closer R^ approaches 1, the better the simulation of the prototype system. As in all cases of simulation, we must try to
reconcile the advantages of increased accuracy and the convenience of keeping
the model as simple as possible. If one of the listed watershed parameters is
dropped from equation 1which is equivalent to fixing the appropriate exponent as zerowe can view this as either simplification of the model or a
constraint on the parameter. If one or more parameters are held at zero, the
optimum values of the remaining parameters can be determined and the corresponding value of R^ calculated. If, subsequently, one of the previously
constrained parameters is allowed to enter into the optimization procedure, a
new set of parameter values will be obtained for all the variables, and the
value of W will be increased provided the variable which now enters the relationship has an influence on the dependent variable not accounted for by
the other variables in the relationship.
In the example of multiple linear regression quoted above, which is taken
from Benson's {11) study of 164 station records in New^ England, the correlation of the 10-year peak flow Qio with the area A alone gave a value of E? =
0.78. This figure may be crudely interpreted as indicating an efficiency of
simulation of 78 percent. If, instead of using a single input, the channel slope
{S) is also taken into account, the value of R increases to 0.889 which is a
distinct improvement. The inclusion of the orographie factor (0) with the
area and slope increases R to 0.932; the further addition of the storage parameter {St) results in a small increase giving 0.945; inclusion of the temperature factor {t) brings the value up to 0.959. Inclusion of the precipitation
factor (/) does not give any further improvement. The exact values of the
exponents were simplified, thus giving a more convenient formula without
appreciable loss of accuracy. The final regression equation with simplified
exponents was:
0^ = 4.52
(2)
This is an interesting and somewhat surprising illustration of w^hat may happen when using a regression model; though rainfall is an important physical
cause of the runoff, the inclusion of the rainfall factor does not improve the
accuracy of prediction based on the other factors and rainfall is not included
in the final equation. The fact that the rainfall parameter does not improve
the accuracy of prediction would suggest that it is highly correlated with the
watershed parameters already included in the model and hence, has no additional independent information to contribute.

If a linear regression does not give a good working model, the use of curvilinear regression ma}^ improve the situation. The commonest model used in
curvilinear regression is polynomial regression, which for the case of simple
regression (only one independent variable) takes the form:
y = a+bx+cx'^+dx^+
(3a)
This is clearly equivalent to the multiple linear regression equation

y = a+bxi+cx2+dx^+
(3b)
if each of the powers of x is considered as a separate variable. Though equation

3a expresses a nonlinear relationship between y and x^ the equation is linear
in the unknown parameters (a, 6, c, d, and so forth) and the standard methods
for estimating these parameters can be used. The above approach can be extended to cover multiple polynomial regression, w^hich is represented by the
equation:
y = Pi{xi)+p2{x2)+pîxz)
(3c)
where p{x) denotes a polynomial of x.
If the factors are thought to combine as products rather than as sums,
then an equation of the following form is appropriate :
y =h {Xi) '2 {X2) 'fz (xz)
/ (Xn)
(4)
Equation la on page 000 is a special case of equation 4, which is adopted because it can be readily transformed to the multiple linear regression form
shown in equation lb. If this particular model does not result in satisfactory
simulation, then another model must be tried. A model which is another
special case of equation 4 is described below for the case where there are three
independent variables.
As a first approximation, it may be assumed that the individual variation of
y with Xi can be represented by the linear relationship :
fi(xi) -ai+hiXi+
(5a)
and a similar linear relationship is assumed for the other two independent
variables :
f2(X2) =-a2 + b2X2+
'
(5b)
fz(xz) =-a^+hxz+
(5c)
The general relationship for these assumptions can be written:

y = aia2az
+6x020 3X1+ai62 3X2+aia2hxz
+61^20 30:10:2+aih2hzX2Xz+bia2bzXzXi
+ bib2bzXiX2Xz
( 5d )
155
The relationship between the dependent variable, y^ the three original independent variables (oî, x^^ Xz) and the four products formed from them can
now be analyzed {57).
If the factors do not act independently of one another, then a model of
joint regression:
y=f{xi,X2yXi
Xn)
(6a)
must be used. The very general form of equation 6a may be modified by

assuming that the variables act in groups so that we can write some such
equation as :
y=fi(xi, X2) +/2(:c3, X4) +
(6b)
Unless there is some apriori reason to suggest a particular relationship, joint

regression analysis is more conveniently handled by graphic than by algebraic
methods. For joint regression, it is frequently helpful to assume the model of
multiple linear regression, then plot the residual values of y against the independent variables and fit ^'contours'' if there is any indication of a joint relation. In this case, the joint regression term is added to the linear terms.
Multiple regression analysis makes the assumption that all the errors are
concentrated in the dependent variable and also that the so-called independent
variables are not correlated with one another. Violation of the latter assumption does not prevent the derivation of a regression relationship which can be
used as a prediction tool, but it renders meaningless the tests of significance
used in a regression analysis. In hydrology, due to the operation of geomorphological factors, the watershed parameters used as independent variables are
often very highly correlated among themselves. Multivariate analysis (33)
seeks to avoid these two problems by treating all the variables alike and by
performing component analysis to determine any truly independent grouping
of variables which may be present. Wong {66) applied component analysis to
the data for New England floods referred to on pages 000 to 000. Wong described how to isolate orthogonal components and showed that for the average
annual flood it was possible to produce a relationship based on two parameters
which was as accurate as the multiple linear regression equation based on five
parameters. Some other papers dealing with the application of multivariate
analysis to hydrology are included among the references at the end of this
lecture {U, 58, 59).
The method of coaxial correlation described by Linsley, Khler, and
Paulhus (4^) has been widely used in applied hydrology. Coaxial correlation
is essentially a graphic method of nonlinear regression and is suitable for the
solution of ad hoc problems. In some cases, the shape of curves used reflected
certain assumptions about the soil moisture accounting involved. In its original form, coaxial correlation was subject to the disadvantage that the process
was a subjective one and different workers would produce different diagrams
from the same set of data. Those experienced in the use of the method tended
to follow a fixed procedure and to produce coaxial correlation diagrams which

were similar in their general form.
If coaxial correlation is to be used as a tool in parametric hydrology, it
should explicitly involve the assumption of a given model of watershed behavior. This approach to coaxial correlation is reected in the wôrk of Becker
(7, 5, 10). Figure 7-2 is based on his papers and can be used to illustrate the
approach to coaxial correlation based on physical reasoning and the use of a
particular model. The diagram is intended to be used to estimate the basin
recharge following rainfall. As indicated by arrows in figure 7-2, progression
is from A to JB to C quadrants. Quadrant A is intended to give the relationship between potential basin recharge and initial moisture content, the latter
being represented by an antecedent precipitation index. The separate lines in
quadrant A represent week numbers and, therefore, different seasons of the
year. Quadrant B allows for the effect of rainfall duration (and hence of the
rate of infiltration and replenishment of soil moisture) on the basin recharge.
A: SOIL MOISTURE
QUADRANT
MWW
FIGURE
7-2.Coaxial correlation.
157
Quadrant C reflects the effect of the amount of rainfall on the actual basin
recharge.
Becker argues that because of the physical processes involved, there are
certain constraints on the shapes of the curves in the different quadrants. He
invokes a simplified model of the watershed behavior to determine the general
nature of these restraints. The higher the value of the antecedent precipitation, the less storage will be available in the watershed for recharge. If the
antecedent precipitation index approaches an infinite value, then all the rainfall must run off, and there can be no recharge to the basin no matter what
the value of the volume of rainfall or duration of rainfall. Becker argues from
this that all the lines in quadrants B and C must pass through the origin,
whereas in the most of the published literature (see for example 51) y the lines
in quadrant B are drawn as parallel lines, and those in quadrant C are drawn
as meeting at a point on the axis between quadrants A and D,
The next step in Becker^s procedure is to take account of the fact that most
present-day models of total catchment response assume the existence of a
threshold between soil moisture and direct runoff {39, 5Ji). A simple threshold
operates as follows. If the storm rainfall is less than the initial field moisture
deficit, there wall be no direct runoff, and all of the rainfall will be accounted
for as basin recharge. If, however, the storm rainfall is greater than the initial
field moisture deficit, then the soil moisture storage will reach its threshold
value and direct storm runoff will occur. For a simple threshold, the amount
of direct runoff will be equal to the volume of rainfall minus the volume of
the initial field moisture deficit.
If the duration of the rainfall is sufficiently long, the intensity of rainfall
will be less than any predetermined limiting infiltration rate. Becker argues
that for a finite amount of rain and a very long duration, the line in quadrant
C must consist of a limiting line which makes the ordinate between quadrants
B and C equal to the ordinate between quadrants C and D, together with a
vertical line, corresponding to the amount of rainfall. When the deficit given
on the ordinate between quadrants B and C is greater than the amount of
rainfall, there is no direct runoff; then the recharge is equal to the amount of
rainfall and is independent of the value of the initial deficit. In this case, the
value of recharge to be read on the ordinate between quadrant C and quadrant D is governed by the vertical line corresponding to the rainfall amount.
The inclined equal value linewhich acts as an upper limit to the series of
vertical linesgoverns the determination of the basin recharge for the case
wehere the rainfall is greater than the initial field moisture deficit; in this case,
the basin recharge given on the ordinate between quadrants B and D.
Becker recognizes that a quadrant C pattern of this type (a series of vertical
lines for different rainfall amounts and one line at 45) is essentially a simplified model based on a lumping of the characteristics of the catchment, which
assumes that the rainfall distribution and the distribution of field moisture
deficit are uniform throughout the catchment. If allowance is made for the

variation of field moisture deficit (or of rainfall) throughout the watershed,
then the curve in quadrant C for any given rainfall amount will show a smooth
transition rather than a sharp break from the sloping 45 line to the appropriate vertical line. If the rainfall is very high relative to the initial field moisture
deficit then the threshold will probably be exceeded in nearly all parts of the
catchment, and the basin recharge will closely approximate the initial deficit.
If, on the other hand, the rainfall were very small relative to the initial deficit,
then the threshold capacity would probably not be reached in any part of the
catchment, and thus the basin recharge would be equal to the amount of
rainfall. For intermediate ratio of rainfall to moisture deficit, a smooth transitional curve would be obtained as shown in quadrant C, figure 7-2. The
curves shown on the figure reflect the assumption of varying thresholds
through the watershed, that is, of a multicapacity accounting system {39y
50, 63).
For very long durations, the low intensity of rainfall will ensure that all of
the rainfall will infiltrate into the soil. For such cases, the duration will not
affect the recharge to the basin, and hence the line in quadrant B will be an
inclined straight line giving equal values on the ordinate between quadrant A
and quadrant B and the ordinate between quadrants B and C. For the same
amount of rain and a shorter duration, the rate of rainfall may exceed the
infiltration capacity of the soil at some time during the storm, and the full
amount of potential basin recharge may not be realized; consequently, water
failing to infiltrate the soil will contribute to direct storm runoff. During a
storm event, infiltration is limited to the duration of rainfall {TR) plus the
time for which overland flow persists after the rainfall ends (To). Becker assumes a limiting rate of infiltration into the moist soil (/m), and hence for a
given duration of rainfall the recharge cannot exceed the product of this infiltration rate plus the total duration of overland flow (that is, the sum of the
rainfall duration plus the time of overland flow after the cessation of rain),
which is assumed to be constant. This limitation on infiltration is reflected in
the horizontal lines in quadrant B for low durations of rainfall; these lines
represent the limiting recharge/W(T+To).
Where the rainfall intensity is less than the limiting infiltration rate for a
wet surface, the actual basin recharge depends on the rate at which moisture
in the soil profile is replenished. If the water infiltrating through the surface
is in excess of that required for soil moisture recharge, then interflow will
occur and contribute to direct storm runoff. Becker assumes that the rate of
soil moisture recharge is proportional to the soil moisture deficit, varying from
zero when the deficit is zero (that is, the soil is at field capacity) to the rate
of maximum infiltration into a dry soil when the deficit is equal to the field
moisture capacity. This assumption gives an exponential decline with time in
the rate of soil moisture recharge. Becker shows that for a constant rate of
infiltration into a dry soil (/max) and a constant value of field moisture ca-
159
pacity (NW) that the volume of recharge for a given duration (TR) is proportional to the volume of soil moisture recharge for the same amount of
rainfall and an infinite duration. Consequently, for all cases where the recharge is not limited by the rate of infiltration through a wet surface, the lines
in quadrant B (which reflect the effect of duration on recharge) will form a
ray of lines through the center of the coaxial system as shown in figure 7-2.
The general shape of the curves in quadrant A can be shown to be plausible
by means of arguments based on relatively simple assumptions. If the catchment were homogeneous, one would expect the lines in quadrant A to be
straight lines joining the value of the soil moisture deficit under wilting conditions on the two axes. If, however, the catchment is considered as being made
up of a number of areas with differing maximum field moisture deficits, then
the curve connecting the initial soil moisture (represented by the antecedent
precipitation index) on the axis between quadrants A and D with the maximum possible recharge on the axis between quadrants A and B would take a
general hyperbolic form. The existence of different curves for different seasons
of the year would be expected due to the effect on moisture accounting of
evaporation, transpiration, and consumptive use. Becker showed that by
drawing the coaxial correlation diagram as described above, results could be
obtained as good as (if not better than) with the more conventional form
usually recommended. His approach has the advantage that the pattern of
lines in his diagram and the position of some of the lines can be related to a
definite model and to physically reasonable catchment parameters. It would
be interesting to link up Becker's approach with some of the models which
have been suggested for simulating the entire watershed response discussed
later in this lecture. Becker (9) also includes a quadrant reflecting the effect
of ground water level on the relation among antecendent conditions, rainfall,
and basin recharge. This quadrant is not shown in figure 7-2.
In all types of regression analysislinear or nonlinear, numerical or graphical, multiple regression, or multivariateit must be remembered that the
choice of a particular model and that the computational procedure merely
represent a way of optimizing the parameters of this model. By optimizing
the parameters on the basis of an actual record, we enable the particular model
chosen to simulate the operation of the prototype as nearly as possible in
accordance with some chosen criterion. There is no guarantee, however, that
we have chosen well in choosing the particular method or model used. Far
too little has been done in the systematic exploration of the problem of choosing between models.
Regression models of all types share with models in general the feature that
they may predict the behavior of the prototype without resembling or revealing the nature of the prototype system. However, it is correct to say that the
more closely a model is based on the physical nature of the prototype system
the more likely is it to prove its worth as a general-purpose model (that is, to
predict the behavior of the prototype under a wide variety of conditions) and
facilitate the meaningful comparison of parameter values derived from using
the same general model to simulate different prototypes.
Digital Simulation
Except in the simplest cases, the regression models previously described
and the mathematical methods of simulation described in lecture 9 require
the use of digital computers. In such cases, the use of the computer is not
compulsory in the simulation of the hydrologie system, but rather it is the
most convenient method of computation. In the present section, we are not
concerned with the use of the computer for these purposes of calculation
only but rather with simulations of hydrologie systems involving techniques
which are only feasible on a digital computer, for example, systematic searching for optimum values of relatively large numbers of parameters.
In the present section, a description will be given of some typical simulations on the digital computer of the hydrologie processes involved in elements
of the hydrologie cycle and the simulation of the total response of the catchment. The highly important subject of the optimization techniques required
to obtain objective estimates of the best values of the parameters to be used
is left until the next section.
Digital simulation can be used to reproduce the behavior of any element in
the hydrologie cycle. In lecture 2, we discussed the empirical formulas used
in applied hydrology for estimating snowmelt. The simplest of these formulas
was:
ilf = 0.06(r.ean-24)
(6a)
which relates the daily snowmelt in open areas in inches M to the mean daily
temperature (Tm) in degrees Fahrenehit. A complex formula which has been
widely quoted is:
D=
801oge(a/2;o) loge{b/zo)
u 6pT+(e-611)
4231
(6b)
P J
which relates the snowmelt D to a number of micrometeorological factors.^

In contrast to the above formulas, figure 7-3 shows a digital simulation
model for snowmelt developed by Amorocho and Espildora (i). It can be
seen that this simulation for the snowmelt process is much more complex
even than equation 6b. In the simulation process, the inputs to the system
are the meteorological conditions and the initial condition of the snow. The
simulation model is shown in the form of the flow chart which can be readily
programed for a digital computer. Indeed it would be difficult to apply the
model shown in figure 7-3 without the use of a digital computer due to the
^ Lecture 2, pp. 44 and 45.
161
^^
\METE
METEOROLOGICAL INPUTS
COND. OF SNOW
\\m.
TqSTp
YES
SNOW
>
SNOW
INTERCEPTION
RAIN
INTERCEPTION
CALCULATION
OF PARAMETERS
OF NEW SNOW
DISTRIBUTION TO
BARE OR SNOW
COVERED SOIL
NO
PRECIP. )
10
I^NETSNO^
'VACCUM. J
/" RAIN ^
V TO SOIL J
f RAIN TO
\^ SNOWPACK
)-
HEAT
BUDGET
15
COMPACTION
OF SNOWPACK
13
14
REDISTRIBUTION OF
SNOW LAYERS
MELTING
PROCESSES
YES^ EXCESS WATER

-^^^ REACHES GROUND
DJSURFACE
COMPUTE
ABLATION
NO
yWATERSHEi
23
\ INPUT /'
"^s-
20
REDISTRIBUTION
OF
SNOW LAYERS
/^ FINAL V
"iiyPARAMETERS/^
^
FIGURE
7-3.Simulation of snowmelt.
21
large number of logical decisions as well as computations which are required.

These logical decisions are represented by lozenge-shaped boxes on figure 7-3.
The first question is asked in decision box 0: Is there precipitation? If there is
not precipitation, we travel by a route through box 1 and then consider the
heat budget in box 15 which takes into account such vectors as incoming
radiation. If there has been precipitation in the form of rain, the amount of
interception is allowed for, and a distribution of the precipitation is made
between bare soil and snow-covered soil in box 5. The rain going directly to
the soil is an output of the model and an input to the watershed. The rain
falling on the snowpack is taken into account in a heat budget.
In this simulation, the snow cover is divided into layers which are treated
separately. The effect of heat, rainfall, or new snow on the existing snow
layers are all taken into account. Each box in the flow chart represents a
physical process; some of these processes are of a high complexity, and this is
reflected in the computational equations used in the step. Thus, box 16 in
figure 7-3 has to be expanded into a flow chart as complex as figure 7-3 itself.
Equations 6a and 6b and the flow chart shown in the figure are simulations
of certain physical processes. We can recognize the empirical equations of
classical physical hydrology as very simple models of these physical processes.
Due to the advent of the digital computer, we can now replace these simple
physical equations by simulation models involving both complex mathematical relationships and multiple decision processes. The simulation shown in
outline on figure 7-3 and described briefly above is only one possible model of
the snowmelt process, and other digital models have been developed and reported in the literature. One such simulation model by Anderson and Crawford forms part of the later versions of the Stanford watershed model (5, 23).
The other individual processes in the hydrologie cycle may also be simulated in this way. The contrast between the classical empirical equations and
more complex simulations based on the digital computer can also be illustrated
for transpiration. Thus we could estimate transpiration according to a combination-type formula such as that of Penman (49) :
^^=
1 + A/T
^'^
where the potential transpiration ET is estimated as a weighted average of the

aerodynamic factor Ea and of the heat budget factor HTJ the weighting factor
A/7 being a function of temperature. This equation may be contrasted with
the simulation of transpiration shown in figure 7-4, taken from a recent paper
in ^Water Resources Research'^ (67). This represents the simulation of
the action of a plant in removing moisture from the soil and transpiring it to
the atmosphere. While it is not likely that such a detailed simulation would
163
IK
.
o
U)
UJ
to e
- 5
_l
UJ
-> r.OT
< o
<1
U.
<
-^r-
I
M
S UJ
- w
CO 3
K>
^ ?: '
o >O <
85
1?
li
iS
to
w
S- P
li-
a: o
4
+
be required in hydrology, it does indicate the complexity underlying the

processes with which the hydrologist is concerned.
The full simulation of the process is more complex even than shown on
figure 7-4. On the lower right-hand side of figure 7-4, stomatal aperture control appears as an input factor. Figure 7-5 shows that this factor, which is an
input to the transpiration simulation, itself depends in a complex fashion on a
number of inputs. When we look at simulations such as these, we realize that
the complexity of a formula such as Penman's is negligible compared with the
complexity of the physical processes which it is intended to represent. It is
interesting, in the case of the two simulations for transpiration and stomatal
control, that the authors first show the diagram from a botanical viewpoint,
then from a more abstract system viewpoint, and finally in terms of system
functions of the different operations involved.
U,OHT
(-S
LIGHT
TRANSPIRATION
PROCESS
PHOTOSYNTHETIC
PROCESS
[CO.] g
PHOTOSYNTHETIC
ACTUATOR - Gpa
NbED
SELECTION
OF CONTROL
SWITCH
TL
PHOTOSYNTHETIC
CONTROL MECHANISM - Gpc
S \
INTE",>
1 ^ GRATOR
1
1
*k
f
1
^.
-VO/^
__J
+AFULL TURGOR Tgt
*i ^ ^\
.^^L,
-.__
ATg
.
-^
INTE- ,'L
C| GRATOR
1
METABOLIC
CONTROL
MECHANISM-Gmc
Em
METABOLIC
ACTUATOR - Gma
OSMOTIC
PRESSURE
'
E> V
1 [COjg
(CHeO)n
METABOLIC
PROCESS
1
1
MESOPHYLL 8
EPIDERMAL
CELLS
a FAF rFl 1
[0.1
/
1
i_
DIFFEREN- U
TIATOR
y
z^^
1
AT
GUARD
CELLS
COMPARATOR
T
TURGOR PRESSURE MECHANISM
NL
FIGURE
(STOMAlAL /^PERTUf?E
7-5.Simulation of stomatal control.
165
Other elements of the hydrologie cycle can be similarly treated. The processes involved in flowwhether overland, in open channels, through soils, or
from ground water reservoirscan be simulated by models of varying complexity. These phenomena lend themselves to relatively simple simulation by
overall mathematical equations or by conceptual models. These methods will
be discussed in lecture 9. If, however, we were not satisfied with the use of
bulk friction formulas and insisted on taking into account the fine details of
turbulence structure and viscous dissipation of energy, the simulation of
these processes would become extremely complex. For flood routing in natural
channels, one may choose among the simple methods of flood routing used in
applied hydrology, the relatively simple conceptual methods which have been
developed recently, and the solution of the problem in its full complexity on
a digital computer.
In all these cases, we are faced with the dilemma of using either a simple
model which is easy to manipulate and comprehend but which may be too
crude a simplification of the physical process or, on the other hand, a highly
complex model which may be difficult to develop and expensive to operate to
obtain further accuracy. No matter how complex our simulation model may
be, the odds are that it still will not mirror the true complexity of the physical
processes involved and hence not reflect the physical reality of the situation.
While this failure might worry the pure scientist seeking to determine the
nature of things, it is of little consequence to the engineering hydrologist
who seeks only for a technique which will be sufficiently accurate for his immediate purpose. The research hydrologist comes somewhere between these
two extremes. He seeks results and methods that are grounded on a general
body of knowledge and hence of wide application.
Digital simulation can also be used to model the total response of a watershed. Here again there is a choice between a simple model, which will of
necessity be crude, and a more complex model in which it may be difficult to
optimize the parameters owing to their number and their interaction. The
simplest model of total watershed operation which gives any semblance of
reproducing the behavior of a watershed was discussed earlier in lecture 1
(see fig. 1-8). In its simplest form, such a model might attempt to simulate
a watershed by assuming that (1) direct storm runoff could be obtained by
routing precipitation excess through a single element of linear storage (Ki) ;
(2) base flow, by routing recharge to ground water through another element
of linear storage of longer storage delay time (K2) ; (3) the division between
precipitation excess and infiltration, by use of a constant infiltration rate
(/c) ; and (4) the recharge to ground water, by assuming a threshold of field
moisture capacity (M). Even in this highly simplified form, four parameters
would be required to describe the operation of the model, one for each of the
four elements.
If we now seek to make the model more accurate or more realistic by a
detailed simulation of any of the elements, a number of additional parameters

will be introduced. The number of parameters quickly increases, and the
problem of objectively determining their optimum values can be handled
only on a digital computer. The determination of the values of these parameters for optimal representation of the prototype is the key problem in digital
simulation of total catchment response. We may insert values of the parameters based on field measurements made either in the watershed under study
or in a similar watershed; but it would be foolish to take such measured
values or any textbook values as other than indicators of the order of magnitude of the parameters required for simulation.
Once an attempt is made to simulate the operation of a watershed by a
specific model, a model parameter which is designed to correspond to some
single physical parameter in the field may, in fact, take on other functions in
the simulation process. If our desire is to understand in detail the physical
processes which are involved, we have no option but to seek the parameters
corresponding to these additional functions and synthesize a more complicated model. If, on the other hand, our only purpose is to reconstruct the
operation of the prototype and predict the outputs, then we should seek the
value of the parameters of our model which optimize its performance.
The best known work on the digital simulation of the total watershed is
that done at Stanford University {21, 22, 23, Ifi, 45). The Stanford model
Mark IV is shown on figure 7-6. The various versions of the Stanford model
are essentially compromises between the oversimplification of the fourparameter model shown on figure 1-8 and the uncontrollable complexity of a
model which would attempt to include everything we know about physical
hydrology. The inputs to the model are precipitation in the form of mean
hourly rainfall and vapotranspiration in the form of daily means. The outputs are streamflow in the form of: (1) summary tables of mean daily flow;
(2) hydrographs of all storms greater than a given base; and (3) some monthly
data, such as volume of interflow and actual vapotranspiration and initial
and final soil moisture conditions. Other outputs can be obtained on an optional basis. A feature of the model is the division of soil moisture storage into
upper zone storage from which vapotranspiration takes place at the potential
rate and lower zone storage from which vapotranspiration takes place at a
rate less than the potential rate when the upper zone storage is exhausted.
The routing of the various flowsoverland flow, interflow, ground water flow,
and channel flowis based largely on reservoir routing. The Mark IV model,
which is more complicated than earlier versions, allows for such features as
overland flow and snowmelt. There are 19 parameters in the model (excluding
snowmelt parameters) and four initial parameters for setting the values of the
various storage components. All but four of these parameters are estimated
from the records or from maps.
167
An alternative model of the total watershed response is that Dawdy and

O'Donnell (24), which is shown on figure 7-7. This model is somewhat simpler
than the Stanford model and was deliberately designed to be so. While the
Stanford model on figure 7-6 is drawn in block diagram form, the DawdyO'Donnell model is drawn in terms of tanks and overflows after the manner of
Sugawara (60) and other Japanese workers in the field. It would be a useful
exercise to attempt to redraw each of the models in the other form. Dawdy
and O'Donnell were primarily interested in investigating the problem of developing the most efficient techniques for optimizing the parameters of a
model, rather than in simulating any particular watershed. For this reason,
they first fixed ''correct'' values of the parameters of the model shown on
figure 7-7, generated the output due to a synthetic input and then, starting
from erroneous initial parameter values, tried to discover from the record of
input and output the predetermined ''correct'' values of the nine parameters.
The inputs to the Dawdy-O'Donnell model are precipitation and vapotranspiration. The output is the eventual total runoff including surface runoff and base flow. There are nine parameters in the model whose values are to
be optimized. R* is the depression storage which must be satisfied before
overland flow occurs. The operation of the linear channel storage is characterized by a single storage delay time, Ks. The Horton equation is used to model
the infiltration, thus accumulating three more parameters /o, /c, and k. Field
G*
FIGURE
7-7.Dawdy-O'Donnell model.
169
moisture capacity is taken as M* and acts as a threshold on recharge to

ground water. Further parameters introduced are a ground water capacity,
G*, which enables the model to simulate water logging under very wet conditions and a maximum rate of capillary rise, Cmax, to simulate the loss of water
from the ground water by capillary rise during very dry periods. The ground
water reservoir is assumed to act as a linear reservoir, thus giving the ninth
parameter. KG. This model will be used to illustrate the problem of parameter
optimization later in the lecture.
A number of other models of the total catchment response have been developed for various purposes and from various points of view. Among those
which are described in the literature are the Tennessee Valley Authority
model (61 y 62) and models developed in Australia {12) and Japan {46).
Because all models of this type will be compromises, they will be different
from one another. The only way in which they can be judged is the efficiency
with which they carry out their specific purpose. If models are constructed
for different purposes, then it is impossible to compare them. We should be
very careful of saying that one model is better than another unless we are
sure that the objectives of both models are the same or can be expressed in
common terms.
Optimization
Frequent reference has been made above to the optimization of the parameters of a simulation model. The present section deals with this problem of
parameter optimization. The output predicted by the simulation model will
vary with the value of each of the parameters in the model. If the efficiency
of the model in predicting the output of the prototype is defined in terms of
an objective criterion, then the optimal values of the model parameters are
those values which give the optimum value of this defined criterion of efficiency. The choice between models and the choice of synthesis must necessarily be subjective, but the optimal values of the parameters should be objectively determined. If this is done, we will know (in regard to the application of any particular model to any particular set of data) that the model is
operating at its highest efficiency and thus may be fairly compared with any
other model operating at its own peak efficiency for the same set of data.
Optimization is essentially a mathematical idea and is, in a sense, somewhat
at variance with human nature. In our ordinary decisions of life, we ^^satisfize"
rather than optimize. As soon as a certain level of satisfaction or performance
is obtained, human judgment is usually satisfied and does not wish to go to
complete optimization. In this context, the decision is a correct one because
the effort expended in going from a satisfactory solution to an optimal one
may be very great, and the resulting gain may be very small. Indeed, we make
the same decision in simulation when we decide to compromise on a model
of a certain degree of complexity. However, if we are using mathematical
methods and a digital computer to find values of parameters for our model,
then the effort to optimize may not be appreciably greater than that of
achieving a certain level of performance and the truly optimal solution has
the added advantage of being unique or virtually so. There are many cases
in applied hydrology and in hydrologie design in which the correct decision
is to halt the process once a certain level of accuracy has been obtained. In
scientific research, on the other hand, optimization is necessary to eliminate
as much subjectivity as possible from the result.
If we are going to optimize, we can only optimize with respect to some
criterion. Unless a specific criterion is invoked, it is not even possible to say
whether the optimum has been obtained. Some hydrologists are convinced
that they are sufficiently experienced to optimize by personal judgment or to
otpimize by eye; if they are explicit in this respect, nobody will be deceived,
but very often the subjectivity is implicit. Objective criteria are, however, to
be preferred.
If we have chosen a specific model, then the predicted estimated output is
a function of the input and of the parameters of that model. Thus, in the case
of a simple model with three parameters, we could write :
y(t)=cl>lx{t),a,h,c']
(8)
where x{t) is the input, a, 6, and c are the parameters of the model, and y{t)
is the output predicted by the model. The problem of optimization is to find
values of a, 6, and c so that the predicted values oy{t) are as close as possible
to the measured values of y{t) in some sense to be defined. The most common
criterion is that the sum of the squares of the differences between the predicted outputs and the actual outputs will be a minimum:
E{a, 6, c) = X] (~2/i)^ = minimum
(9)
As an alternative to using a least squares criterion, we could adopt the Chebyshev criterion of minimizing the maximum error:
E(a, 6, c) = max {yiyi) = minimum
(10)
In this case, we avoid the occurrence of one or two large deviations between
predicted output and measured output whose presence might be accepted in
the least squares criterion, since their effect could be smoothed out by a
faithful reproduction in the remainder of the record.
Another criterion which can be used is moment matching. We can say
that if a model has the same first n statistical moments as the prototype,
then the two systems are equivalent in some sense. Actually it can be proved
that if the moments of the two impulse responses are identical up to the n*^
moment, then the systems will give identical output for any input which is
171
a polynomial of the degree n or less. Consequently, a model system with a

given number of parameters will reproduce the behavior of the prototype for
polynomial inputs if the values of these parameters are determined by matching the appropriate number of moments of the model with those of the prototype:
R{y) =<l>{ct, b, c) =fJiR(y)
(11)
Where a large number of parameters are involved, the method of moment

matching is not suitable because higher order moments become unreliable due
to the distorting effect of errors in the tail of the function on the values of the
moments. However, the method of moments has the great value that in cases
where the moments of the model system can be expressed as a simple function
of the parameters of the model, then the parameters can be relatively easily
derived.
For criteria such as least squares or minimax error, direct derivation of the
parameters may be far from easy. In certain cases, it is possible to express the
criterion to be minimized as a function of the parameters. To differentiate
this function with respect to each parameter in turn, set all the results equal
to zero and solve the resulting simultaneous equations to find the optimal
value of the parameters. For any but the simplest model, it will probably be
simpler to optimize the parameters by using a systematic search technique to
find those parameter values which give the minimum value of the error function. Such a search technique gives rise to its own difficulties, which will be
discussed later in this section.
It is often necessary to decide whether we wish to put bounds on the values
of the parameters. Any model with which we attempt to simulate the prototype will be based to a greater or lesser degree on our assumptions about the
nature of the physical processes in the hydrologie system under investigation.
We are then faced with a dilemma if the optimized values of the parameter
of this model turn out to be physically unrealistic. For example, we might
seek to simulate direct storm runoff by a cascade of equal linear storage elements. Such a model has two parameters, the storage delay time (K) of the
individual elements and the number of equal elements (n).
An analysis of the data by moment matching might indicate that both n
and K are negative. Similarly, we migiit insert into a model of a watershed
the Horton infiltration equation and then find on optimizing the parameters
that the value of/c turns out to be 1,000 feet per second. Even though we are
interested in predicting the output and the unrealistic parameter values give
a good prediction, we are inclined to reject such values and put bounds on the
variation of the parameter. This is to bring a subjective element into our
simulation and to import knowledge from physical hydrology into parametric
hydrology. It may or may not be the right thing to do.
If the restriction of the parameter to realistic values does not increase the
error function much above its minimum value, then it is certainly permissible
to use the model with the restricted range of parameters. If, however, the error
function is greatly increased by refusing to allow the parameters to take on
unrealistic values, then this may be an indication that the model itself is at
fault and should be modified or replaced. One consequence of optimizing subject to restraint is that the mathematics (and the computation) become more
difficult. In an analytical solution, partial derivatives must be replaced by the
use of Lagrange multipliers. If the error function and restraints are not linear,
we may be involved in nonlinear programing which means serious computational problems. In a systematic search technique, the extra difficulty created
by the introduction of bounds on the parameters is not nearly as serious. It is
important, however, to remember that the imposition of a restraint always
results in some loss of optimality. Where the error function does not vary
sharply, then the effect may not be serious.
As in all computations, our final task is to interpret our results. In the
simulation of hydrologie systems, it is difficult to know how much meaning
should be attached to the optimal values of the parameters found. It is probably correct to say that the answer to this problem depends on the model
used. If the model is an extremely good representation of the prototype, then
there is a good chance that the parameters are of physical significance, and
there is likely to be a close connection between the values of these physical
parameters and the corresponding field parameters of the prototype. If, however, the model is much more simple than the prototype, then there is no
guarantee that the parameters will correspond to the real physical parameters
of the prototype. It may well be that a particular parameter in the model is
an amalgam of several parameters in the prototype, but there is no guarantee
of this. It may be dangerous to try and give a close physical meaning to some
of the parameters found by optimization. It is safer to consider these parameters as the parameters of best fit and be satisfied with a model w^hich does
what we require it to do, namely, predict the output within a given margin
of error.
The optimization of model parameters by a systematic search technique is
a powerful approach made possible by the use of digital computers. It is,
however, not quite as easy as it might at first appear. If you consider the
almost trivial case of a two-parameter model, then the problem of optimizing
these parameters subject to a least squares error criterion can be easily illustrated. We can imagine the two parameters a and h as measured along coordinate axes and the squares of the deviations between the predicted and
actual outputs as indicated by contours in the plane defined by these axes.
The problem of optimizing our parameters is then equivalent to searching
this relief map for the highest peak or the lowest valley, depending on the
way in which we pose the problem. We have to search until we get, not merely
a local optimum (maximum or minimum), but an absolute optimum. To
examine every point of the plane would be prohibitive even to this simple
173
example. In using a search technique, we have no guarantee that we will

find the true optimum.
The simplest method of searching is to start at some point on the boundary
and travel parallel to the other axis until the optimum on that line is obtained. The direction of search can then be changed to a direction at right
angles to that just traversed and the search continued until the optimum
along that line is obtained. Again the direction can be changed and the process
repeated until a point has been obtained, which is the optimum in its immediate neighborhood. There would be no guarantee, however, that it would
be an absolute optimum. This simple method of searching turns out on examination to be quite inefficient even for a small number of parameters, and
more sophisticated techniques have been developed (63^ 64). Some of these
are based on the steepest descent methods, which are considerably more
efficient than the univariate technique described above. However, once more
than a few parameters are involved, even the sophisticated gradient methods
become inefficient compared with a direct search technique.
We saw that the simplest model for a total watershed would involve at least
four parameters and that models now being developed and used have more
than 20 parameters. Even engineers trained in descriptive geometry would
find it hard to visualize the complexity of the searching technique in such a
multidimensional problem.
A direct search technique based on Rosenbroek's method {52) was used by
Dawdy and O'Donnell for the systematic optimization of the parameters of
the model shown in figure 7-7 {24). The search through the multidimensional
parameter space was made in a sequence of stages. In the first stage, an initial
set of parameter values was assumed, and searches were made along the
parameter axes. At the start of each subsequent stage, a new set of orthogonal
axes was chosen for the search, the best direction for the new search being
determined from the progress made in the previous stage. During each stage,
movements were made along the new axes subject to their producing an improvement in the objective function and following a specific set of rules about
the size and direction of movement along the axes. These rules also specified
when a stage should end and a new set of orthogonal axes begin. Progress
was usually rapid in the first five or six stages but tailed off thereafter. The
whole process was revitalized by starting a new round of stages with the latest
parameter values from the end of the previous round of stages but starting
again with the parameter axes as the orthogonal search directions.
It was easier to obtain reasonable approximations to the values of some of
the parameters than others. If the parameters were initially set with a large
error, some of them would be within a few percentage points of the true value
after a single round of stages, whereas others might show little improvement
after 20 rounds, and some might end up further from their true value. A
parameter can only be readily optimized if it strongly affects the output and
the effect can be isolated in some fashion. Consequently, if the particular

input for which the model is tested does not call a particular parameter into
play, the effect of this parameter cannot be isolated or its value determined
from that particular record. The parameter in question can take any value
over a wide range without affecting the objective function. Thus, the operation or even the existence of the parameter for maximum capillary rise in the
Dawdy-O'Donnell model could only become apparent if a record was available
containing a long dry spell. Since the parameters difficult to optimize are those
which do not affect the output for the particular input used, failure to find
their values will not affect the model as a predictor provided it is used only
for inputs which are, by and large, of a similar type to the input used for the
optimization of the parameters.
There is a great deal more work to be done before the comparison of simulation models for hydrologie systems can be put on a proper objective basis.
A model structure or a set of parameter values that predict efficiently for
one type of input and one type of criterion of fit may prove quite inefficient
for another set of input data or another criterion of prediction. We must be
clear at all times what criterion we are using and what type of output we are
trying to predict. Table 7-1 is taken from a paper by Dawdy and Thompson
{26) and illustrates the effect of the use of different criteria on the optimization process. An attempt was made to fit the model developed by Dawdy and
O'Donnell to data for the Arroyo Seco, near Pasadena. The first objective
taken was to minimize the sums of the squares of the logarithms of the ratio
of computed to observed monthly discharges. As shown in the upper half of
table 7-1, 482 trials resulted in the values of the objective function. The
criterion was then changed to one based on daily discharges rather than
TABLE
7-1.Resvlts of three optimizing runs during 1943-44 i'^ Arroyo Seco

Sum of squared deviations of logarithms
Optimizing
criterion
Months. .
Days
Peaks
Source: Dawdy and Thompson {25).
Try
Months
0
89
344
482
0
9
360
0
6
35
156
0.68
.35
.18
.17
.17
.18
.26
.73
.72
.67
1.61
Days
48.3
28.2
24.1
24.7
24.7
24.5
15.1
40.8
40.7
58.0
96.2
Peaks
0.63
.49
.50
.49
.49
.48
1.02
1.11
1.09
.35
.01
175
monthly discharges. The next 360 trials were made on this basis with the
results shown in the middle of table 7-1. The last 156 trials shown on the
bottom half of the table 7-1 were based on a criterion of peak discharges.
For each trial, the value of the objective function according to each criterion
was evaluated though only the objective criterion indicated was used to search
for the optimal values of the parameters.
We can see from table 7-1 whether an optimization based on months goes
anywhere near getting the degree of optimization which would be obtained
if we concentrated on daily discharges or on peaks. It can be seen that optimization based on monthly discharges gives values of the parameters which are
not too far from the optimal for daily discharges and that optimization based
on daily discharges gives values of the parameters that are not too far from
the optimal for monthly discharges. The differences, though serious enough,
or not enormous. However, when we compare the value of the objective function when the parameters are optimized on the basis of peaks with the value
when the parameters are optimized on the basis of their daily or monthly
discharges, we find a tremendous difference. The criterion for peak matching
can be reduced to 0.01 when optimization is based on the peaks themselves
but only reaches a value of 0.49 for optimization based on months and 1.02
for optimization based on daily discharges.
These results are extremely interesting when we consider that what is involved here is not a change of model but merely a change in choice of the
period of now, which is the basis of the optimization. The model is a relatively
complex one, and the parameters used all have definite physical implications.
Nevertheless, it is not capable of acting as a general-purpose model for peaks,
daily discharges, and monthly discharges. If we are only interested in one of
these at a time we can adjust our model parameters accordingly. It would
also be possible to define a weighted objective function which would take into
account each of these separate objectives in some fashion. The weighting of
the different objectives, however, would itself tend to be subjective.
There is a scope for a great deal of work in the field of digital simulation.
A part of this should be devoted to a systematic exploration of the subject
using both noise-free synthetic data and synthetic data with controlled error.
It should, for example, be possible to determine from the input and output
records of a system whether or not one or more thresholds occur in the system. At the same time, another part of the work should be concerned with
the simulation of field data and the further problems involved.
Analogs and Physical Models

The use of analogs and physical models comes within the scope of parametric
hydrology since these analogs and models are used to simulate the action of
the prototype systems. Analogs may be divided into two typesindirect

analogs which solve the mathematical equations thought to govern the phenomena, and direct analogs which attempt to simulate the physical behavior
of the prototypes by an analogous physical system. Though physical models
have been used for a long time in hydraulics, their use in simulating hydrologie
systems gives rise to a number of difficulties which have not as yet been
overcome.
As mentioned above, an indirect analog seeks to solve the mathematical
equations which themselves simulate the action of the prototype system. The
most widely used type of indirect analog is the indirect electronic analog,
also known as an analog computer or a differential analyzer. The actual solution of the problem involves the standard techniques common to the large
variety of problems for which the analog computer is suitable. In the use of
an indirect analog for hydrologie systems, the key element is the formulation
of the mathematical equations to be solved, or the synthesis of conceptual
models whose mathematical equations can easily be written down. This may
be illustrated for the case of a very simple conceptual model consisting of two
linear reservoirs in series. Actually, as will be seen later, this particular model
can be readily represented by a simple direct analog.
For the first linear reservoir, the inflow (/) and the outflow (Qi) are connected by the relationship :
I-Q.-K^f
(12)
where K is the storage delay time of the reservoir. If two such elements are
cascaded, that is, are placed in series so that the output from the first (Qi)
is the inflow to the second, we have for the operation of the second element the
relationship :
Qi-Q2 = iC^
at
(13)
where Q2 is the outflow from the second reservoir. Substitution of the value
of Qi from equation 13 into equation 12 gives:
or
d^Q2
dQ2
K^-^+2K^+Q, = I
dt^
dt
(14b)
or
X.^=-2xf-Q.+7
dt^
dt
(14c)
177
In setting up an indirect analog for any system, it is wise to follow a basic

step-by-step procedure (4). The first step in the basic procedure is to draw a
block diagram of the type shown in figure 7-8 for two linear reservoirs in
series represented by equation 14. The highest derivative in the differential
equation is assumed to be known, and blocks are inserted to integrate it to
obtain the lower order derivatives and the unknown function itself, as shown
in figure 7-8. The appropriate terms are then combined by elementary arithematical operations, also shown in the diagram by blocks, to construct the
highest derivative in accordance with the equation which is being simulated.
Thus, in our case, the first derivative is multiplied by 2K and both the dervate
and the unknown function Q are reversed in sign before being added to the
original inflow; the sum of these three components is then divided by K^ to
produce the second derivative.
Since the sealers, adders, and integrators in an analog circuit reverse the
sign of the voltage, the block diagram must next be modified to allow for the
change in sign; at the same time, the individual symbols for the various opera-
at*
-/a,
(f
2k
-1
\'
\}
-1
i\
1 ^
i_
!<*
FIGURE
7-8.Block diagram for indirect analog.
tions may be inserted as shown on figure 7-9. It may be possible at the same
time to take advantage of the possibility of combining several operations into
one block.
After the block diagram has been modified, an equation is written for each
block in the diagram, and the scale factor is determined for each variable in
the circuit. This scaling is necessary to avoid overloading any element in the
computing circuit. To do this, it is necessary to have some estimate of the
maximum value of each of the variables. The analog of the system can now
be redrawn as shown in figure 7-10 and is seen to require two integrators and
one operational amplifier together with the necessary potentiometers. The
indirect analog has the advantage of allowing an extremely rapid adjustment
of parameters and visual presentation of the comparison of the simulated
output and the actual output. It has great advantages for exploratory work
and could be used with advantage in hydrologie investigations. A team at
Utah State University has pioneered the simulation of the total watershed
response on an electronic computer. The Mark I model contained 46 operational amplifiers, three multipliers, two function generators, and 192 potentiometers. The Mark II model contains additions to the above components together with some nonlinear elements and arrangement for greater
flexibility of operation {51).
There are a variety of types of direct electrical analog. They may be classified as continuous direct analogs, discrete direct analogs, or combined direct
FIGURE
7-9.Modified block diagram.
J
FIGURE
at^
179
at
7-10.Indirect analog for two linear reservoirs in series.
analogs. They have been used widely in the field of ground water flow (5,
20, 56), but there have also been a number of applications to flow in the unsaturated zone {13, H, SI) and to flow in open channels {32, 38, 55). Two
well-known forms of continuous direct electrical analog are the electrolytic
tank and Teledeltos resistance paper. These analogs are used in studying the
flow through porous media by utilizing the similarity between the differential
equations governing flow through porous media and those governing the flow
of electrical currents through conductive materials. For exploratory studies,
a simple electrolytic tank or Teledeltos resistance paper (or sheets of some
other conductive material) may be used. In the case of the electrolytic tank,
more sophisticated and accurate work is possible in both two and three dimensions. The method can be applied to anisotropic media by means of scale
distortion. In the case of continuous analogs, every point in the analog simulates the corresponding point in the prototype.
Discrete direct analogs have been more widely used in hydrology than the
continuous type. Such discrete analogs are usually discretized only in respect
of the space dimension, and time is left as a continuous variable when unsteady flow cases are studied. Such a discretization is subject to the same types
of error as are involved in the representation of a differential equation by its
finite difference form.
For problems involving the steady flow of ground water, a complex prototype system can be simulated by a direct analog made up from resistances
only. These resistances may be set out in either a symmetrical or an asymmetrical network and may be applied to two-dimensional plane flow, axisymmetrical flow, or three-dimensional flow. For other types of electrical
analog, it is necessary to determine the scaling of the analog carefully.
Unsteady flow problems in porous media can be successfully studied by an
analog network containing both resistances and capacitors (R-C network).

In such analogs, the electrical resistances simulate the resistance to flow, and
the capacitors simulate the storage properties of the aquifer. A discrete direct
analog for the one-dimensional linear diffusion equation used to solve land
drainage problems is shown in figure 7-11. Two types of R-C network analogs
are used. Slow analogs (time constants of the order of 10 minutes) record the
solution of the problem on paper charts, while rapid or repetitive analogs
(time constants of the order of a tenth of a second) show the solution on an
oscilloscope.
Direct electrical analogs based on R-C networks have been applied to
other phases of the hydrologie cycle besides the ground water phase. Because
the flow through unsaturated porous media can be represented by a diffusion
type equation, it is possible to represent this phase of the hydrologie cycle by
a similar analog to that used for ground water flow. It can also be shown that
a diffusion model gives a very close approximation to the complete solution
of the linearized equations for unsteady flow in open channels. Consequently,
the same type of R-C analog network could be used in this case also. This
suggests the possibility of simulating the various subsystems of the hydrologie
cycle by the same type of network analog.
Many other types of direct discrete electrical analogs have been applied to
surface water hydrology. Some of these were attempts to simulate specific
models of the hydrologie process as in the case of the electrical analog of the
Muskingum (41) and Kalinin-Milyukov (37) methods of flood routing. Some
parts of the hydrologie cycle can be simulated by conceptual models consisting
of standard elements, such as distortionless linear channels and linear storage
elements. These elements can, in turn, be simulated by a direct electrical
analog and the operation of the prototype system studied in this way.
Figures 7-11 and 7-13 show three simple elements which could be used as
AAAAA
AAAA/^
FIGURE
7-11.Analog for unsteady ground water flow.
>
181
building blocks in a direct electrical analog of a hydrologie system or of a

conceptual model of that hydrologie system. The element in figure 7-11 represents the typical element used to simulate a diffusion-type equation already
referred to. These elements can be linked as shown and as mentioned above,
more than one phase of the hydrologie cycle might be simulated by the use of
such elements. Figure 7-12 shows the direct electrical analog of a linear storage element and the connection of two such elements in series. The latter
arrangement corresponds to the indirect analog for the same system shown on
figure 7-10. Comparison of the two analogs shows little similarity between
them. Figure 7-13 shows the direct analog circuit suggested by Shen {55)
for a distortionless linear channel. Such an element could be used as part of a
lag and route model or similar conceptual model.
Because any function can be expanded in terms of Laguerre functions, it
can be shown theoretically that any linear system can be represented by an
analog system consisting entirely of linear storage elements, though the analog
system might need to include a large number of such elements connected in
series and in parallel. If a particular system can be represented by a small
number of such elements, then a direct analog with elements as shown on
figure 7-12 can be constructed.
The basic types of direct electrical analogs described above can be adapted
to deal with special problems. It is possible to combine continuous and discrete
elements in the one analog. While the discussion given above is concentrated
on the simulation of linearized hydrologie systems, the techniques indicated
can be adpated to include nonlinear elements, though this naturally introduces
certain complexities and difficulties.
There are a number of other direct analogs besides electrical analogs, and
some of these have potential applications in simulating hydrologie systems.
The best known nonelectrical direct analog is the Hele-Shaw apparatus or
viscous flow analog, which is widely used in two-dimensional ground water
investigations. In this type of analog, a viscous liquid is allowed to flow between parallel plates whose distance apart is about 1 mm. Properly used, the
Hele-Shaw model can be a powerful scientific instrument and not just a piece
of demonstration apparatus. Vertical versions of the Hele-Shaw apparatus
can be used to study such problems as flow to a parallel drainage system,
AA/V^
FIGURE
!>
-ww
7-12.Direct analog two linear reservoirs.
Qg
182: TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE
4
5
2/3R
1/6 R
3.6 C
FIGURE
4
5
I/6R
^WV
3.6 C
7-13.Direct analog for distortionless linear channel.
while horizontal Hele-Shaw models can be used to study conditions in a largescale aquifer. Another analog with possible applications in the study of
ground water systems is the membrane analogy, which has been applied to
some problems of flow towards wells.
The fact that still other analogs are available for hydrologie systems is
illustrated by the recent development by Diskin (26) of a salt-concentration
analogy for flow from a watershed. It would be a grave pity if absorption with
the digital computer was to lead hydrologists to neglect the many useful tools
available in the form of analogs.
If the space between a pair of parallel plates is filled with sand, or glass
beads, we have not a Hele-Shaw apparatus but a sandbox or granular model.
Such a device is more correctly described as a physical model than an analog.
Many problems involving the flow in unsaturated and saturated porous media
can be studied on such a model (6). The effect of the capillary fringe is relatively larger on such a model than in the prototype, and this may give rise
to considerable difficulty.
In the case of unsaturated flow, there are difficulties in the problem of model
scaling, but recent work indicates that these problems are being overcome.
In one particular version of the granular model, the filling material is glass
beads or crushed glass, the walls are transparent, and the liquid used has the
same refractive index as that of the glass. This enables themovement of a dye
tracer to be followed with ease. Columns of glass beads are used by soil
physicists in the study of the problems of infiltration and percolation of water
in the unsaturated zone. These represent idealization of the actual movement
183
in the soil and as such are attempts to simulate the prototype soil system on
simplified physical models.
Problems on the boundary between hydrology and open channel hydraulics
can be studied by the use of the hydraulic models, for which some highly
developed and widely tested techniques are available. The model of the
Mississippi, operated by the Corps of Engineers, handles what is essentially
a hydrologie problemthe routing of oods on the Mississippi and its tributaries. The reduction in the time scale of the model compared with the prototype enables this model to be used for making predictions. Many proponents
of numerical computation object to the technique of introducing artificial
roughness to ensure verification of a hydraulic model. Such people seem to
forget that the digital simulation of the same problem uses a so-called roughness coefficient which is more a repository of unknown effects than a roughness
factor. In many digital simulations, the values of Manning's n are adjusted
both w4th stage and along the channel until the required downstream discharge is obtained, \\hether we simulate on the hydraulic model or on a
digital computer, verification is necessary if our work is to be worthwhile.
In either case, the devices used to ensure verification are not always logically
defensible.
An unusual model of the hydrologie system of Lake Hefner was tested in
a wind tunnel at Colorado State University (15). The model laws were investigated and the evaporation from the lake was successfully studied on a
small scale.
The final type of model to be considered is a physical model of an entire
watershed. If such models attempt to do more than solve purely hydraulic
problems on a laboratory scale, they run into a great number of difficulties^
{17, 29, SO). Research is now going on in a number of countries on the behavior
both of laboratory-size catchments and of highly instrumented outdoor
''model" catchments, ^hat has been reported so far tends to underline the
difficulties involved in this line of research. It may not be possible to use such
small-scale physical models as prediction tools until such time as wê understand the inherent ''self-similarities'' imposed on hydrologie systems by geomorphological processes. Nevertheless, the results from such experiments on
laboratory catchments carried out under controlled and repeatable conditions
will yield extremely useful data w^hich should lead to a better understanding
of hydrologie processes and of the manner in which response parameters vary
with system parameters. Data from such small-scale laboratory catchments,
w^hich would be intermediate between synthetic mathematical data and field
observations, should prove extremely useful for testing other methods of
simulation.
2 MAMISAO, J. P. DEVELOPMENT OF AN AGRICULTURAL WATERSHED BY SIMILTUDE.
Thesis, Iowa State CoL, Ames. 1952.
M.S.
Problems on Simulation
1. Discuss one particular type of simulation in terms of the phases of the
simulation process discussed on page 150.
2. Contrast the different methods of simulation from the point of view of
convenience, accuracy, and stability for some part of the hydrologie cycle with
which you are familiar.
3. Appendix table 5 shows some digital computer data for the linear response of the uniform open channel. Hopefully, this data will be used to develop a method of predicting the time-to-peak for values of lengths of 5, 50,
and 100 miles and slopes of 1, 20, and 50 feet per mile. What sort of regression
analysis would be suitable in this particular case, and how would you go about
applying it to this particular problem?
4. Describe how^ the problem posed in question 3 might be solved by
analog simulation.
5. Describe what would be necessary if the same problem were to be solved
by a series of flume experiments.
6. What criteria of fit are most commonly used in deriving empirical expressions to fit hydrologie data? What other criteria could also be used? Discuss the merits of the different criteria.
7. Compute the evaporation and potential transpiration by a number of
formulas for the data given in Appendix table 8. Under what conditions would
you expect each empirical formula to work best? Can you draw a diagram
illustrating the different assumptions made about the relationship between
actual and potential transpiration?
8. Discuss the relationship between a complex simulation of the snowmelt
process and a formula relating the rate of snowmelt to degree days.
9. Compare a number of the total catchment models which have been
proposed in the literature. What are their common elements and how do they
differ?
10. Discuss the method described in the literature to obtain the optimum
parameters for various models of (a) the unit hydrograph, (b) ground water
response, and (c) total catchment response. Discuss how these methods might
be improved, and estimate the optimum parameters for some example in
literature which, in your opinion, have not been optimized.
11. Derive a direct and an indirect analog representation for both the
Horton and the Philip equations for infiltration.
12. Draw up a classification of the various types of analog and physical
models used in hydrology.
185
Literature Cited
(1)
1966.
(2)
(3)
and ESPILDORA, B.
MATHEMATICAL SIMULATION OF THE SNOW MELTING PROCESSES. Water Sci.
and Engin. Paper 3001, 156 pp. University of California, Davis.
AMOROCHO, J.,
and HART, W. E.
1965. THE USE OF LABORATORY CATCHMENTS IN THE STUDY OF HYDROLOGIC SYSTEMS. Jour. Hydrol. 3(2): 106-123.
A., and CRAWFORD, N. H.
THE SYNTHESIS OF CONTINUOUS SNOW MELT RUNOFF HYDROGRAPHS ON A
DIGITAL COMPUTER. Dept. Civ. Engin. Tech. Rpt. 36, 103 pp. Stanford
ANDERSON, E.
1964.
University, Calif.
(4)
ASHLEY, J. R.
(5)
BEAR,
1963.
(6)
INTRODUCTION TO ANALOG COMPUTATION.
294 pp.
New York.
J., and SCHWARZ, J.

1966. ELECTRIC ANALOG FOR REGIONAL GROUNDWATER STUDIES. Underground
Water Storage Study. TAHAL Tech. Rpt. 22. Tel-Aviv.
ZASLAVSKY, D.,
1968.
and
IRMAY, S.
PHYSICAL PRINCIPLES OF WATER PERCOLATION AND SEEPAGE.
UNESCO
Arid Zone Res. XXIX, 465 pp., illus.

(7)
BECKER,
1966.
A.
ERGEBNISSE EINER UNTERSUCHUNG BER DIE STRUCKTUR DER KOAXIALEN
GRAPHISCHEN STARKREGENABFLUSSBEZIEHUNGEN [RESULTS OF AN INVESTIGATION REGARDING THE STRUCTURE OF THE COAXIAL GRAPHIC HEAVY
RAIN RUNOFF-RELATiONs]. Wasscrwirtschaft-Wassertech. 16(3): 121-130.
(8)
1966.
ON THE STRUCTURE OF COAXIAL GRAPHICAL RAINFALL-RUNOFF RELATIONS.

Internatl. Assoc. Sei. Hydrol. Bui. XI(2): 121-130.
(9)
1967.
THRESHOLD CONSIDERATIONS AND THEIR GENERAL IMPORTANCE FOR HYDROLOGIC SYSTEMS INVESTIGATION. Fort ColUns International Hydrology Symposium.
Compt. Rend. 1: 94-102.
(10)
1968.
Wasserwirtschaft-Wassertech. 18(1): 16-21.
(11)
BENSON, M. A.
1962. FACTORS
INFLUENCING THE OCCURRENCE OF FLOODS IN HUMID REGIONS OF

DIVERSE TERRAIN. U.S. Geol. Survey Water-Supply Paper 580-B, 30 pp.
(12)
BROUGHTON, W. C.
1965.
A NEW SIMULATION TECHNIQUE FOR ESTIMATING CATCHMENT YIELD. Water

Res. Lab. Rpt. 78, 87 pp., illus. University of New South Wales, Austral.
(13) BouwER, H.
1960. A STUDY OF FINAL INFILTRATION RATES FROM CYLINDER INFILTROMETERS
AND IRRIGATION FURROWS WITH AN ELECTRICAL RESISTANCE NETWORK.
Trans. 7th Cong. Internatl. Soc. Soil Sei. 1: 448-457.
(14)
and LITTLE, W. C.
1959. A UNIFYING NUMERICAL SOLUTION FOR TWO DIMENSIONAL STEADY FLOW
PROBLEMS IN POROUS MEDIA WITH AN ELECTRICAL RESISTANCE NETWORK.
Soil Sei. Soc. Amer. Proc. 23: 91-96.
186
(15) CERMAK, J. E., and KOLOSEUS, H. J.

1953, 1954.
LAKE HEFNER MODEL STUDIES OF WIND STRUCTURE AND EVAPORATION.
Final reports 1 and 2, November 1953 and July 1954.

Engin., Colorado A&M College, Fort Collins.
Dept. Civ.
(16) CHEMERENKO, E. P.
1966.
PRIMENENIYE ELEKTR0NN0I VICHISLITELNOI MASHINI VO BREMYA GIDROLICHESKI VICHESLENII [uSE OF ELECTRONIC COMPUTERS DURING HYDROLOGI-
CAL CALCULATIONS].
Meteorologiya i Gidrologiya Zhuzurnal.
Leningrad.
(17) CHERY, D. L.
1955.
CONSTRUCTION,
INSTRUMENTATION
PHYSICAL HYDROLOGic MODEL.

State Univ., Logan.
(18) CHORAFAS, D. N.
1965. SYSTEMS AND SIMULATION.
(19)
CHOW, VEN TE,
1964.
AND
PRELIMINARY
152 pp.
503 pp.
(22)
(23)
OF
New York.
ed.
APPLICATIONS OF ELECTRONIC COMPUTERS IN HYDROLOGY.

of Applied Hydrology. Sec. 29, 30 pp. New York.
(20) COLORADO STATE UNIVERSITY.

1963. SYMPOSIUM ON TRANSIENT GROUNDWATER HYDRAULICS.
(21) CRAWFORD, N. H., and LINSLEY, R. K.
1962.
VERIFICATION
Utah Water Res. Lab., Utah
THE SYNTHESIS OF CONTINUOUS STREAMFLOW
HlS Handbook
Fort Collins.
HYDROGRAPH ON A DIGITAL
COMPUTER. Dept. Civ. Engin., Tech. Rpt. 12, 121 pp. Stanford University, Calif.
and LINSLEY, R. K.
1963. CONCEPTUAL MODEL OF THE HYDROLOGIC CYCLE. Internatl. Assoc. Sci.
Hydrol. Pub. 62: 573-587. Gentbrugge.
and LINSLEY, R. K.
1966.
DIGITAL SIMULATION IN HYDROLOGY: STANFORD WATERSHED MODEL MARK IV.
Dept. Civ. Engin., Tech. Rpt. 39, 210 pp. Stanford University, Calif.
(24) DAWDY, D. R., and O'DONNELL, T.
1965. MATHEMATICAL MODELS OF CATCHMENT BEHAVIOR. Amer. SoC. Civ. Engin.,
(25)
and THOMPSON, T. H.
1967. DIGITAL COMPUTER SIMULATION IN HYDROLOGY. Amer. Water Works
Assoc. Jour. 59(6): 685-688.
(26) DisKiN, M. H.
1967.
A DISPERSION ANALOG MODEL FOR WATERSHED SYSTEMS. International

Hydrologie Symposium Proc, Compt. Rend. 1: 38-45, Fort Collins, Colo.
(27) EzEKiEL, M., and Fox, K. A.

1959.
METHODS OF CORRELATION AND REGRESSION ANALYSIS: LINEAR AND CURVILINEAR. Ed. 3, 522 pp., illus. New York.
(28) FlERING, M.
1967. STREAMFLOW SYNTHESIS.
139 pp.
BostoU.
(29) GRACE, R. A., and EAGLESON, P. S.

1965. SIMILARITY CRITERIA IN THE SURFACE RUNOFF PROCESS. Hydrodynamics
Lab. Rpt. 77, School of Engineering, Massachusetts Institute of Technology, Cambridge.
(30)
1966.
CONSTRUCTION AND USE OF A PHYSICAL MODEL OF THE RAINFALL-RUNOFF
PROCESS. Hydrodynamics Lab. Tech. Note 11. School of Engineering,

Massachusetts Institute of Technology, Cambridge.
(31)
187
and BOWERS, S. A.
NON-STEADY-STATE MOISTURE, TEMPERATURE, AND SOIL AIR PRESSURE APPROXIMATION WITH AN ELECTRIC SIMULATOR. Soil Science Soc. Amer.
HANKS, R. J.,
1960.
Proc. 24(4): 247-252.

(32)
A.
ANALOG MODELS FOR FLOOD CONTROL SYSTEMS.
HARDER, J.
1962.
Amer. Soc. Civ. Engin.,

(33)
HARMAN, H. H.
(34)
HARRIS, B., SHARP,
1960.
1961.
MODERN FACTOR ANALYSIS. 471 pp. CMcagO.

A. L., GIBBS, A. E., and OWEN, W. J.
AN IMPROVED STATISTICAL MODEL FOR EVALUATING PARAMETERS AFFECTING
WATER YIELDS OF BASINS. Jour. Geophys. Res. 66(10): 3319.
(35) HuGGiNS, L. F., and MONKE, E. J.

1966. THE MATHEMATICAL SIMULATION OF THE HYDROLOGY OF SMALL WATERSHEDS.
Tech. Rpt. 1, 130 pp. Water Resources Res. Center, Purdue University,
Lafayette, Ind.
1949. ELEMENTS OF APPLIED HYDROLOGY.
(37)
1960.
(38)
276 pp.
NeW York.
P., and LEVIN, A. G.

USE OF AN ELECTRONIC ANALOG COMPUTER IN THE FORECASTING OF RAIN
FLOODS. Meterologia i. Gidrologia Zhuzurnal., 12: 14-18. Leningrad.
KALININ, G.
A.
TEORIYA VEROYATNOSTNYKH PROTSESSOV V GIDROLOGII I REGULITOVANII
RECHNAGO STOKA [THEORY OF STOCHASTIC PROCESSES IN HYDROLOGY AND
RIVER RUNOFF REGULATION]. Translated into English by Israel Program
KARTVELISHVILLI, N.
1967.
for Scientific Translation, 1969.

(39) KoHLER, M. A., and RICHARDS, M.
1962. MULTI-CAPACITY BASIN ACCOUNTING FOR PREDICTING RUNOFF FROM STORM
PRECIPITATION. Jour. Geophys. Res. 67: 5187-5197.
(40) LiNSLEY, R. K., and CRAWFORD, N. H.
1960. COMPUTATION OF A SYNTHETIC STREAM FLOW RECORD ON A DIGITAL COMPUTER. Internatl. Assoc. Sei. Hydrol. Pub. 51: 526-538.
(41)
(42)
(43)
FosKETT, L. W., and KOHLER, M. A.

1948. ELECTRONIC DEVICE SPEEDS FLOOD-FORECASTING.
News-Rec.
and others.
DESIGN OF WATER RESOURCE SYSTEMS.
MAASS, A. A.,
1962.
(44)
Engin.
441(26): 64-66.
KOHLER, M. A., and PAULHUS, J. L. H.
620 pp.
Cambridge, Mass.
MTALAS, N. C, and REIHER, B. J.

1967. SOME COMMENTS ON THE USE OF FACTOR ANALYSES.
Water Resources Res.
3(1): 213-223.
(45)
NEGEV, M.
1967. MODEL
OF SEDIMENT
Rept. 76.
(46)
r^T?
rr. .
Dept. Civ. i^ngin. iecn.
Stanford University, Calif.
and KIWATA, H.
FLOOD SIMULATOR FOR THE RIVER KiTAKAMi.
OTABA, K., SHIPATANI, K.,
1965.
(47)
YIELD OF A CATCHMENT.
Simulation 4(2): 86-98.
PAYNTER, H. M.
1952.
METHODS AND RESULTS FROM THE MIT STUDIES IN UNSTEADY FLOW.

(48)
ed.
1955. A PALIMPSEST ON THE ELECTRONIC ANALOG ART.
270 pp.
BostOU.
BoStOH
188
(49) PENMAN, H. L.
1948.
NATURAL EVAPORATION FROM OPEN WATER, BARE SOIL, AND GRASS.
Roy.
Soc. London Proc. Ser. A 193: 120-145.

(50) RICHARDS, M., and STRAHL, J. A.
1967. ELEMENTS OF RIVER FORECASTING. Weather Bur. Tech Memo WBTM
HYDRO-4. Washington.
(51) RiLEY, J. P., CHADWICH, D. G., and Bagley, J. M.
1966. APPLICATION OF ELECTRONIC ANALOG COMPUTER TO SOLUTION OF HYDROLOGIC
AND RIVER BASIN PLANNING: UTAH SIMULATION MODEL II.
129 pp.
Utah
Water Resources Laboratory, Utah State University, Logan.

(52) ROSENBROCK, H. H.
1960.
AN AUTOMATIC METHOD FOR FINDING THE GREATEST OR LEAST VALUE OF A
FUNCTION.
(53) SCHULZE, F. E.
1966.
Computer Jour. 3(3): 175-184.
RAINFALL AND RAINFALL EXCESS.
PART III.
RECENT TRENDS IN HYDRO-
GRAPH SYNTHESIS. TNO Proc. and Inform. Note 13: 9-30. The Hague.
(54) SHARP, A. L., GIBBS, A. E., OWEN, W. J., and HARRIS, B.
1960. APPLICATION OF MULTIPLE REGRESSION APPROACH IN EVALUATING PARAMETERS AFFECTING WATER YIELDS OF RIVER BASINS. Jour. GeophyS ReS
64: 1273-1286.
(55) SHEN, J.
1956.
USE OF ANALOG MODELS FOR THE ANALYSIS OF FLOOD RUNOFF.
U.S. Geol.
Survey Prof. Paper 506-A, 24 pp.

(56) SKIBITZKE, H. E.
1963.
THE USE OF ANALOG COMPUTERS FOR STUDIES IN GROUNDWATER HYDROLOGY.
Inst. Water Engin. Jour. 17(3): 216-230.

(57) SNYDER, W. M.
1956.
London.
SOME LINEAR MODELS FOR MULTIPLE REGRESSION ANALYSIS IN HYDROLOGY.
Tenn. Val. Auth. Hydrol. Data Branch.
Knoxville.
(58)
1962.
(59)
SOME POSSIBILITIES FOR MULTIVARIATE ANALYSIS IN HYDROLOGIC STUDIES.
Jour. Geophys. Res. 67(2): 721-729.

and STALL, J. B.
1965.
MEN, MODELS, METHODS AND MACHINES IN HYDROLOGIC ANALYSIS.
Amer.
Soc. Civ. Engin., Jour. Hydraulics Div. 91(HY4): 85-99.

(60) SUGAWARA, M.
1961.
ON THE ANALYSIS OF RUNOFF STRUCTURES ABOUT SEVERAL JAPANESE RIVERS.
Jap. Jour. Geophys. 2(4): 1-76.

(61) TENNESSEE VALLEY AUTHORITY.
1963. A WATER YIELD MODEL FOR ANALYSIS OF MONTHLY RUNOFF DATA.
Area Devlpmt. Res. Paper 2, 78 pp. Knoxville, Tenn.
(62)
1964.
BRADSHAW CREEK-ELK RIVER.
Tributary
A PILOT STUDY IN AREA-STREAM FACTOR
CORRELATION. Tributary Area Devlpmt. Res. Paper 4. Knoxville, Tenn.

(63) WILDE, D. J.
1964. OPTIMUM SEEKING METHODS. 202 pp. Englewood Cliffs, N.J.
(64)
and BEIGHTLER, C. S.
1967. FOUNDATIONS OF OPTIMISATION. 480 pp. Englewood Cliffs, N.J.
(65) WILLIAMS, E. J.
1959. REGRESSION ANALYSIS. 214 pp. New York.
189
(66) WONG, S. T.
1963.
A MULTIVARIATE STATISTICAL MODEL FOR PREDICTING MEAN ANNUAL FLOOD
IN NEW ENGLAND. Assoc. Amer. Geog. Ann. 53: 298-311.

(67) Woo, D. C., BoERSMA, L., and STONE, L. N.
1966. DYNAMIC SIMULATION MODEL OF THE TRANSPIRATION PROCESS. Water
(68)
STONE, L. N., and BOERSMA, L.
1966. A CONCEPTUAL MODEL OF STOMATA CONTROL MECHANISMS. Water ReSOUrceS
Res. 2(1): 71-84.
1964.
REGRESSION AND CORRELATION ANALYSIS. Section 8, Part II: 24 pp.

Chow, Ven Te, ed., Handbook of Applied Hydrology. New York.
In
LECTURE 8:
SYNTHETIC UNIT HYDROGRAPHS
Lecture 8 is largely devoted to a discussion of synthetic unit hydrographs
as developed in classical hydrology, and then as modified with the emergence
of the systems approach. The lecture is intended to serve the same purpose
for simulation as lecture 4 on ''Classical Methods of Runoff Prediction'' was
intended to serve for analysis.
In both classical hydrology and parametric hydrology, simulation techniques
were first developed for surface water hydrology. Thus, in this lecture we will
be primarily concerned with the direct storm runoff and its relationship to
precipitation excess. The problem of synthesis is to devise a system which will
operate on an input, x{t), to reproduce the required output, y{t), to a given
degree of accuracy. The dream of the applied hydrologist is to be able to
forecast direct storm runoff from a catchment map; this means being able to
predict the unit hydrograph from a contoured map (preferably with information on soil types) where no records are available for the derivation of
a unit hydrograph.
Types of Synthetic Unit Hydrographs

The standard synthetic procedure has been to derive a series of unit hydrographs in some systematic fashion for watersheds with adequate records
and then to correlate these unit hydrographs in some way with the watershed
characteristics. These correlations are then used to predict the scale and shape
of the unit hydrograph for some watershed whose characteristics are known
but for which no records of outow are available.
In classical hydrology, synthetic unit hydrographs developed along two
main lines, both of which converged at the time of the emergence of parametric
hydrology. These two lines of development are shown in figure 8-1. The
methods at the left-hand side made the general assumption that each catchment had a unique unit hydrograph, and those at the right-hand side made the
general assumption that all unit hydrographs might be represented by a single
curve, or a family of curves, or a single equation.
The first line of development {16, U, 45) derived from the rational method.
(See lecture 4, pp. 75-101). About the year 1920 {54), the rational method
was modified to include the effect of nonuniform rainfall distribution by the
use and time-area curve or the time-area-concentration curve. This modification was, in effect, an attempt to synthesize the response of the watershed on
the basis of the characteristics which could be read from a map. By using a
contoured map and the Manning formula, it was possible to construct the
time-area-concentration curve or the time-area-curve. This was assumed to be
190
191
TIME AREA
METHODS
EMPIRICAL
CURVES
ROUTED
TIME-AREA
EMP RICAL
EQUATIONS
ROUTED .
TRIANGLE
FIGURE
CONCEPTUAL
MODELS
CASCADE OF
RESERVOIRS
8-1.Type of synthetic unit hydrograph.
the instantaneous unit hydrograph (lUH) (or the S-hydrograph) for the
watershed involved, though the unit hydrograph method was not to be
developed for another 10 years. Since, in each case, the time-area-concentration curve was built up from the information available for the particular
catchment, each unit hydrograph was unique. In the 1930's, Zoch (7i)and
afterwards Turner and Bourdoin {65) and Clark (P)assumed that the
response of the watershed would be given by routing the time-area-concentration curve through an element of linear storage. In this case also, each unit
hydrograph would be unique, but the variation between them would be
reduced and differences in watershed characteristics smoothed out to a greater
or lesser extent depending on the degree of damping introduced by the storage
routing.
On the other hand, the second line of development tended to ignore variations in watershed characteristics and in the unit hydrographs. Thus, we find
in the hydrologie literature a number of curves which are presented as giving
the unique shape of the unit hydrograph. One, by Commons {12) was published in 1942. Unique representations of unit hydrograph shape were also put
forward by Williams {69), the SCS {68), and others. These assumed, in effect,
that there is one shape for the unit hydrograph, though in most cases the scale
is still left free and the specified shape is given in terms of dimensionless discharges (for example, g/max) and dimensionless time (for example, ^peak).
Since the volume of the unit hydrograph is conventionally taken as unity,
there is only one parameter to be fixed to determine the unit hydrograph. All
that is required in this empirical curve approach is to know the time-to-peak,
or the peak rate of discharge, and then to use the standard shape of unit
hydrograph to determine the unit hydrograph for the watershed. This is in
distinct contrast to the time-area curve method where all the watershed
information must be used.
As further studies were made of synthetic unit hydrographs, it was realized
that a one-parameter method was not sufficiently flexible and that twoparameter methods were required for adequate representation. These would
require the use of a family of curves from which the unit hydrograph could be
taken. Since it is easier to represent a two-parameter model by an equation
rather than a family of curves, the natural development of this approach was
towards the suggestion of empirical equations which would represent all unit
hydrographs. It is remarkable that people working in many different countries
all turned towards the same empirical equation for the representation of the
unit hydrograph. The independence of this development is proved by the fact
that they expressed this single equation in different forms. The equation in
question was the two-parameter gamma distribution or Pearson type III
empirical distribution.
About 15 years ago, these apparently quite separate lines of development
started to approach one another. O'Kelly, Nash, and Farrell working in the
Irish Office of Public Works found that there was no essential loss in accuracy
if the routed time-area-concentration curve was replaced by a routed isosceles
triangle U9). In their early work, this group had followed the approach of
Clark (9) and laboriously developed a time-area-concentration curve for each
catchment and then routed through a linear storage in order to obtain the
lUH. If the individual time-area-concentration curves for natural watersheds
could be replaced by isosceles triangles without serious distortion of the
resulting unit hydrograph, this was an indication that the smoothing of the
linear reservoir was such that the individual variations in catchment characteristics were removed by routing. Thus, the line development which started
out by treating every unit hydrograph as unique had been modified so as to
represent each unit hydrograph by a two-parameter system, one-parameter
being needed to fix the base of the triangle ( T) and the other the storage delay
time (K) of the linear reservoir. A somewhat similar approach was adopted by
the ses though, in this case, the triangle was nonisosceles. In our modern
terminology, find the unit hydrograph by routing a triangular inflow through
a linear reservoir represents using a conceptual model for the lUH.
While this development was taking place among exponents of the routed
time-area curve approach, there was a similar development among those who
followed the tradition based on empirical curves and empirical equations.
About 10 or 15 years ago, Japanese hydrologists (56, 61, 62) attempted to
simulate the response of rivers by models consisting of one or two linear storage
elements. Following this line, Nash {46) suggested the two-parameter gamma
distribution as having the general shape required for the lUH and pointed out
that the gamma distribution could be considered as the impulse response for a
cascade of equal linear reservoirs. He suggested that the number of reservoirs
193
could be taken as nonintegral if necessary. In this way, the second tradition

also arrived at a conceptual model but in this case, a different one. The routed
triangle had two parameters and generated a family of curves with a particular
shape due to the nature of the model. Similarly, the cascade model had two
parametersthe number of reservoirs n and the storage delay time of each
Kand generated a family of shapes specific to this model.
In 1959, Dooge attempted to develop a general representation of the unit
hydrograph based on the Aluskingum method of routing. When this did not
prove satisfactory, the assumption was made that the translation and storage
elements in the watersheds could be separated and the action of the watershed
represented by linear distortionless channels and linear storage elements or
reservoirs {17), This represented a more general type of conceptual model
than the routed triangle or the cascade of reservoirs and, in fact, included the
two of them as special cases.
Once this stage had been reached, the way was open for attempts to represent the unit hydrograph by all types of conceptual models. It may be dangerous to think of these conceptual models as anything more than an attempt
to simulate the watershed. Dooge {17) was quite convinced that the linear
storage elements which were part of the proposed general model had a real
physical meaning. Now he is by no means so sure. It may be that a breakthrough in understanding the morphology of watersheds would in the future,
allow a close link to be established between the nature of the prototype and
the structure of the optimum simulating system. Meanwhile, it is safer to
think of these models merely as attempts to simulate and to judge them by
their performance in doing so.
Time-Area Methods
It is instructive to review the subject of synthetic unit hydrographs from
its origins in the time-area versions of the rational method which were in use
even before the unit hydrograph method was discovered. In this way, we can
compare the approaches of the modified rational method, classical empirical
unit hydrograph methods, and modern methods of parametric hydrology to the
same problem and to the various elements of that problem. With the hindsight
afforded to us by our knowledge of unit hydrograph methods and of the newer
methods of parametric hydrology, we can recognize the earlier methods used
as special cases of the later approach.
As mentioned in lecture 4, pages 79-84, the original rational method was
intended for predicting the maximum discharge from a catchment and was
not concerned with the prediction of the whole hydrograph (7, 34, 40, 4-3).
Later developments of the method allowed for variations in rainfall intensity
during the design storm and, in doing so, enabled a full hydrograph of runoff
to be developed if required {11,24, 26, 28, 30, 50, 54, 55). Other developments
allowed determination of the question of whether a storm centered over part

of the area might not give a greater peak runoff than one spread over the
whole of the catchment area {21, U, 52, 33). These variations of the rational
method are summarized as follows :
Rainfall Assumption
Authors
Type
Area
Intensity
Classical rational method.
full
uniform
Time-area
full
hypothetical
critical
typical
'Tangent' methods
partial
uniform
Mulvany (1850)
Kuichling(1889)
Chamier (1897)
Lloyd-Davies(1906)
Ross (1921)
Rousculp (1927)
Ormsby (1932)
Hart (1932)
Hawken(1921)
Judson (1932)
Coleman & Johnson (1931)
Laurenson (1932)
Jens (1948)
Reid(1926)
Riley (1931)
Escritt (1950)
Munro(1956)
In 1921, Ross {54) suggested that a hypothetical storm be derived from the
curve of rainfall intensity versus duration and then used in conjunction with
the time-area-concentration diagram to predict the maximum rate of runoff
and, if need be, the whole hydrograph. In an appendix to Ross' paper, Hawken
{24) suggested introducing a factor of safety by shuffling the unit periods of
rainfall into a critical pattern of storm, that is, one in which the most intense
rainfall would be centered over the maximum ordinate of the time-areaconcentration curve, the second most intense rainfall over the second highest
ordinate of the curve, and so on. While the methods proposed by Ross and
Hawken can give safe^^values for design, they would, of their nature, tend to
overestimate the peak rate of runoff. In 1931, Coleman and Johnson {11)
suggested that the pattern of the storm rainfall be based on typical storms for
the area under investigation.
Under certain conditions (which arise mostly in urban catchments) the
runoff estimated by the rational method for part of the area may exceed the
runoff estimated by the same method for the whole area. Special techniques
were developed where the rainfall intensity-duration relationship was assumed
to be of the form :
h+ct
(1)
195
where i is the average rainfall intensity; t the duration of rainfall; and a, fe,
and c are empirical coefficients. In such cases, the partial area giving the
greatest estimated runoff can be determined by drawing a tangent to the
time-area curve, thus giving rise to the name ^'tangent method'^ for such
techniques. Where the rainfall formula is of exponential form:
-^.
(2)
the critical partial area may be found by the use of a series of overlay curves
{21, 44). The critical area may also be found by locating the intersection of
the two curves given by the time-area-curve (scaled up by a factor of h) and
the product of the time-area-concentration curve and the time elapsed. These
time-area methods w^re widely applied in urban hydrology and, to a lesser
extent, in the hydrology of agricultural watersheds.
The time-area variations of the rational method (known in the Russian
literature as genetic or isochrone methods) were actually crude methods for
developing synthetic unit hydrographs. The hypothetical or typical storm was
plotted to the same scale as the time-area-concentration curve, but in one
case the time scale was plotted in a reverse direction. The two curves were
then superimposed, and the products of corresponding ordinates taken and
summed together to obtain the runoff at any given time. The runoff for any
particular time was obtained by superimposing the zero point of the reversed
rainfall-intensity curve on the point of the abscissa of the time-area-concentration curve corresponding to the required time. By shifting the two curves
relative to one another, enough points could be determined to give a representation of the whole hydrograph of runoff for the pattern of rainfall intensity
used. This, in effect, was a graphical method of carrjdng out the mathematical
process of convolution. The time-area-concentration curve in such methods
has the same function as the lUH in unit hydrograph procedures. Thus, the
time-area-concentration curve, however found, was in fact a synthetic unit
hydrograph.
If the time-area-concentration curve was based merely on an estimate of
the time of translation over the ground and in channels, then the results
obtained tended to overestimate the peak rate of discharge from the watershed.
This was only to be expected since the effects of surface storage, soil storage,
and channel storage are all ignored, and the time-area-concentration curve
wâs based purely on translation. In practice, design engineers soon developed
ways of avoiding the tedium of constructing a time-area-concentration curve
for each separate watershed. Where they were interested only in the peak rate
of runoff, they developed empirical formulas for the time of concentration
{tc) and for the coefficient of runoff (C) in the equation:
Q^C'i{Q'A
(3)
where Q is the peak discharge, A is the area of the catchment, and i is the
rainfall intensity for a duration equal to the time of concentration, /c, and for
the particular frequency of recurrence chosen for the design. Others derived
the time-area-concentration curve but, realizing that their values of runoff
were too high, used empirical values for the time of concentration to correct
the time scale of the time-area-concentration curve. This was possible because
the time of concentration is equal to the base length of the time-area-concentration curve, that is, the lUH.
In urban design, rules of thumb for estimating the time of concentration
wêre used. The time of concentration was usually taken by calculating the
time of travel in the sewer and adding to it an inlet time, which usually is
within the range from 5 to 30 minutes. In such urban catchments, the coefficient C in equation 3 depended largely on the amount of impervious area in
the catchment and was also affected by any storage in the system. A typical
empirical formula for the value of C was one which related the coefficient of
runoff (C) to the number of houses per acre {N) in the following way {21) :
C=\//10
(4)
The range of coefficients normally used for different types of urban areas
can be found in standard reference books such as the American Society of Civil
Engineers ^'Manual on the Design and Construction of Sanitary and Storm
Sewers'' {1). More sophisticated methods have been developed in recent
years for the design of storm water sewers, but the discussion of them is outside
the scope of this lecture.
For agricultural catchments, a commonly used formula for the time of
concentration is that of Kirpich (31 ) :
tc = 0m7Sl-]
\s)
(5)
where tc is the time of concentration in minutes, L the length of flow in feet,

and S is the ground slope. The coefficient of runoff C may be related to a
number of factors by:
C =1.00-(Cr+Cs+Ce)
(6)
where CT varies inversely with the slope and has values between 0.1 and 0.3;
Cs varies between 0.1 for a tight clay and 0.4 for sandy loam; and Cc varies
with the vegetal cover between 0.1 for cultivated land and 0.2 for woodlands.
These remarks on the rational formula are made not as an encouragement to
its use but as a background against which to judge the further development
of synthetic unit hydrograph methods.
As indicated on figure 8-1, the time-area methods were, for unit hydrograph
purposes, replaced by a method in which the time-area-concentration curve
was routed through a linear reservoir. Zoch (71) put forward a general physi-
197
cal theor}^ of streamflow based on the assumption that, at any time, the rate of
discharge was proportional to the amount of rainfall remaining within the soil
at that time. He analyzed runoff due to a uniform rainfall of finite duration
and obtained the equations for four separate segments of the hydrograph.
Zoch solved these equations for two simple casesa rectangular time-areaconcentration curve and a triangular time-area-concentration curve. He
pointed out that the extension to the general case would involve the integration of a function of the type :
<l){x) =w{x) exp(Za:)
(7)
where w{x) represents the time-area-concentration curve and K is a constant.

He suggested the use of series approximation or numerical integration.
Horton (27) introduced the idea of the virtual channel inow graph. This
was an attempt to derive from the outflow hydrograph a simple form of inflow
hydrograph which when routed through a linear reservoir would give the
outflow graph. The start of the channel inflow was taken at the same time as the
start of channel outflow and the end of channel inflow at the time corresp,onding to the point of contraflexure on the recession limb of the outflow hydrograph.
This, in fact, represented the estimation of the time of concentration from
the outflow hydrograph. Because of the further assumption of routing through
a single storage element, the recession limb of the virtual channel inflow
graph had to pass through the peak of the outflow graph. The only remaining
condition was that the volume under the inflow and outflow hydrographs
should be the same.
Clark (9) suggested that the unit hydrograph for instantaneous rainfall
could be derived by routing the time-area-concentration curve through a single
element of linear storage. Physically, this is equivalent to Zoch's formulation,
but the equations are simplified by reducing the rainfall duration to zero and
replacing the numerical integration of the term in equation (9) with the reservoir routing procedure. The Zoch-Clark method clearly represented an
advance over the time-area or isochrone methods, which ignored storage
effects and only took account of variations in the time of translation to the
outlet. The allowance for storage throughout the catchment by a single reservoir at the outlet seems a highly simplifying assumption but, nevertheless, a
step in the right direction.
As mentioned earher in this lecture, O'Kelly and his coworkers (49) replaced
the time-area-concentration curve by an isosceles triangle and thus produced
the lUH by routing an isosceles triangle through a linear reservoir. This was,
in effect, a combination of the Zoch-Clark approach with Horton's virtual
channel inflow graph.
The methods of Zoch, Clark, and 0'Kelly only became synthetic unit
hydrograph methods in the real sense of the term when empirical relationships
between some of the parameters of the process and the catchment characteris-

tics were derived. An empirical relationship is required which correlates the
base of the time-area-concentration diagram (that is, the time of concentration) to catchment characteristics. Except in the case of O'Kelly's method,
some method is required for estimating the shape of the time-area-concentration diagram itself once the base length has been determined. Finally, a
method of estimating the storage factor (k) must be prescribed. In the case
of a gaged catchment, the value of the storage constant K can be estimated
from the recession of the hydrograph. In the absence of records of storm runoff
the latter method cannot be used.
Johnstone {29) using the Clark method derived relationships based on 19
catchments with areas between 25 and 1,624 sq. mi. in the Scotle and Sandusky
River basins. Johnstone proposed the following relationship for the base of the
time-area-concentration curve :
(8a)
'~ r^ \s)
where tc is the base of the time-area-concentration curve (that is, the time of
concentration) in hours, L is the length of the principal stream in the catchment in miles, S is the average slope of the main stream in feet per mile, and
r is a branching factor based on the stream pattern. Johnstone found that
there was little loss of accuracy in neglecting the branching factor and writing,
vO.5
i. = 5.o(;^y
(8b)
where the terms have the same meaning as in equation 8a. Johnstone also
derived an empirical expression for the storage delay time (K) which is the
ratio of storage to outflow for the linear reservoir through which the time-areaconcentration curve is routed. On the basis of the catchments studied by him,
he proposed the following empirical relationship for the storage delay time K:
^ = 1.5+90-^
(8c)
where A is the area of the catchment in square miles, L is the length of the
main stream in miles, and R is an overland slope factor in feet per mile estimated by placing a square grid over the contour map and counting the number
of intersections of contour lines and grid lines.
Eaton (19) did a similar correlation study for seven Tasmanian rivers with
catchment areas varying from 48 to 322 sq. mi. He estimated the base lengths
of the time-area-concentration diagram to be given by:
te = lM(^J
(9a)
where tc is the base of the time-area-concentration curve in hours, A is the
199
catchment area in square miles, L is the length of the main channel in miles,
and r is a branching factor varying between 1.0 and 2.0. Eaton's use of a
branching factor rather than a slope factor can be explained by the lack of
contour maps for the region studied. He found that for five of the seven basins
the storage constant K was adequately defined by :
/A2 \l/3
i^ = 1.2(-)
(9b)
where K is the storage delay time in hours and the catchment factors are
defined for equation 9a.
O'Kelly {^9) presented results for 10 catchments in Ireland ranging in area
from 56 to 366 sq. mi. In his paper, the results are reduced to a standard
catchment area of 100 sq. mi. by assuming a hydrologie time-scale factor
based on one-fourth root of the area and then expressed graphically as a
function of the overland slope.
In a discussion of O'Kelly's paper, Dooge {15) indicated that a logical
extension of the idea of a model catchment (based on Froude similarity)
would be to express the base of the isosceles triangle ( T) as :
T=a
(10a)
where T is the base length of the inow triangle in hours, A is the catchment
area in square miles, S is the slope in parts per 10,000, and a is an empirical
constant. For the values of T derived by O'Kelly (4^), the parameter (a)
varied from 10 at a slope of 10 in 10,000 to 14 at a slope of 500 in 10,000.
On a similar basis the values of K could be expressed as:
ii=6
(10b)
where K is the storage delay time of the linear reservoir in hours, h is an

empirical constant, and the other factors are as for equation 10a. For the
values of K derived by O'Kelly, h could be taken in equation 10b as varying
from 13 for a slope of 10 in 10,000 to 10 for a slope of 500 in 10,000. Dooge^
also derived the relationship :
^0.41
^ = 2-58^,
(11a)
based on a least squares analysis of O'Kelly's data and his estimated values
1 DOOGE, J. C. I. SYNTHETIC UNIT HYDROGRAPHS BASED ON TRIANGULAR INFLOW.
M.S. Thesis, Iowa State Univ., Ames. 1956.
200
of T. In equation lia, A is the catchment in square miles and S is the overland

slope in parts per 10,000. A least squares analysis of the values of K derived
by 0'Kelly gives:
i^= 100.5
A 0.23
00.70
<
(lib)
\^^^)
where K is the storage delay time in hours, A is the area in square miles, and
S is the slope in parts per 10,000.
Empirical Expressions for Unit Hydrograph Parameters

We now turn to a review of the empirical line of development of synthetic
unit hydrographs based on the representation of all unit hydrographs by a
single curve or a family of curves. The procedures based on this approach
follow a standard pattern in nearly all cases. A number of unit hydrograph
parameters are chosen as the basis for defining the unit hydrograph.
At the same time, a number of catchment characteristics are chosen which
are thought to have the strongest influence on the shape of the unit hydrograph. For a number of catchments with adequate records of rainfall and
runoff, unit hydrographs are derived and the values of the unit hydrograph
parameters determined. These are then correlated with the chosen catchment
characteristics. This correlation can then be applied to the catchment characteristics of a catchment without adequate runoff records in order to estimate
the parameters of the unit hydrograph for such a catchment. The latter
parameters are then used to derive the full unit hydrograph by using a
standard shape of unit hydrograph or by using additional relationships between the basic unit hydrograph parameters and other features of the unit
hydrograph.
In the time-area methods reviewed in the last section, we discussed first
the shape of the unit hydrograph (that is, the time-area-concentration curve
routed through a single linear reservoir) and after this the empirical relationships by means of which the catchment characteristics could be used to
estimate the two parameters required, that is, the base of the time-areaconcentration curve {tc) and the storage constant characterizing the linear
reservoir {K). In dealing with the second line of development, the order of
discussion will be reversed. In the present section, we will discuss the empirical
relationships between the unit hydrograph parameters and the catchment
characteristics, leaving until the next section the question of the shape of the
empirical synthetic unit hydrograph. In this review of empirical methods,
attention will be concentrated on the main lines of approach, which will be
illustrated by examples. No attempt will be made to list all methods or all
features of the methods mentioned. Those interested in the latter can read
201
details of procedures in the original papers that are referenced at the end of
this lecture.
As mentioned above, the first two steps are the choice of unit hydrograph
parameters and catchment characteristics. Three types of unit hydrograph
parameters are usedtime parameters, peak discharge parameters, and
recession parameters. There are a large number of time parameters used in
unit hydrograph studies, the most important of which are shown on figure 8-2.
In this illustration, D is used to denote the duration of precipitation excess,
which is assumed to occur at a uniform intensity over this unit period. Common
time parameters used to characterize the outow hydrograph are: the time of
rise (r), that is, the time from the beginning of runoff to the time of peak
discharge; the time of virtual inflow (T), that is, from the beginning of runoff
to the point of contraflexure on the recession limb of the outflow hydrograph;
and the total runoff time or base length of the unit hydrograph {B). The
common time parameters used to connect the precipitation excess and the
hydrograph of direct runoff are: the lag time (L), that is, the time from the
center of mass of precipitation excess to the center of mass of direct runoff;
the lag to peak time {tp), that is, the time from the center of mass of effective
rainfall to the peak of the hydrograph; and the time to peak {tj), that is, the
interval between the start of rain and the peak of the outflow hydrograph.
FIGURE
8-2.Unit hydrograph parameters.
One of the most important factors in surface water hydrology is the delay
imposed on the precipitation excess by the action of the catchment. If the
parameter representing this delay is to be useful for correlation studies, it
should, if possible, be independent of the intensity and duration of rainfall. In
the case of a linear systemand unit hydrograph theory assumes that the
system under study is linearthe time parameters are independent of the
intensity of precipitation excess, but only the lag time (ti) has the property
of being independent of both the intensity and the duration. Accordingly,
with the hindsight given by the systems approach, we can say that only the
lag time should be used as a duration parameter in unit hydrograph studies.
In regard to discharge parameters, the peak discharge (max) is almost
invariably used when such a parameter is required. Another parameter, which
can be estimated for a derived unit hydrograph, is the time parameter K,
which characterizes the recession of the unit hydrograph when this recession
is of declining exponential form. In such cases, the unit hydrograph may be
considered as having been routed through a linear reservoir whose storage
delay time is K. If the recession can be represented in this form, a logarithm
of the discharge plotted against time will give a straight line, and the value of
K can be estimated from the slope of this line. Alternatively, the value of K
may be determined at any point on the recession curve by dividing the
remaining outflow after that point by the ordinate of outflow at the point.
Other parameters used to characterize unit hydrographs are the values of
W-50 and W-75, which are the width of the unit hydrograph for ordinates at
50 percent and 75 percent of the peak, respectively.
Nash (46, 47, 4^) suggested the use of the statistical moments of the lUH
as the determining parameters of the unit hydrograph. The first moment 7/ is
equal to the lag of the lUH L. For higher moments, Nash suggested the use
of the dimensionless moment factors obtained by dividing the moment of
any order about the center of area by the first moment raised to a powder
corresponding to the order of the moments. Nash showed that the moments
of the unit hydrograph could be derived from the moments of the precipitation
excess and the moments of the direct runoff without the necessity of deriving
the unit hydrograph itself.
The second stage in the standard procedure is the choice of catchment
characteristics. As might be expected, all procedures involve a scale factor,
but a variety of scale factors is used. The simplest scale factor is to use the
area of the catchment itself {A). Others commonly used are the length of the
main channel or length of highest order stream (L) ; the length to the center
of area of the catchment {Lcc) ; or for small catchments, the length of overland
flow (Lo). Where only one catchment characteristic is used (in a one-parameter
model), the catchment characteristic used is alwâys a length or area parameter.
A review of synthetic unit hydrograph procedures reveals slope as the
second most frequently used catchment characteristic and, therefore, if the
203
applied hydrologists have chosen wisely, the second most important catchment characteristic. Since slope varies throughout a watershed, a standard
definition of some representative slope is required. The slope parameters
most often used are the average slope of the main channel or some average
slope of the ground surface. The measurement of average slope parameters
usually involves tedious computations {10, 60).
Although area or stream length and channel slope or ground slope have been
used almost universally, there is no agreement about the remaining parameters. The shape of the catchment must have some effect, but there is such a
variety of shape factors to choose fromform factors, circularity ratios,
elongation ratios, leminiscate ratios, and othersthat the lack of uniformity
is not surprising. Another factor which must affect the hydrograph is the
stream pattern. This may be represented by drainage density or stream frequency or some such parameter.
Although parameters representing mean characteristics must have a primary
influence, the variations in certain characteristics from part to part of the
watershed will give rise to secondary parameters, which may not be negligible.
Thus, having taken area and slope into account, the third most important
parameter may well be variation of length or of slope rather than shape or
drainage density. The choice of catchment characteristics for correlation with
unit hydrograph parameters will remain a subjective matter until we have a
deeper knowledge of the morphology of natural catchments. The latter is a
vital subject for modern hydrology. If we neglect the study of geomorphological processes to concentrate on mathematical manipulations which have no
physical foundations, then the whole progress of hydrology may be impeded.
Having decided on the unit hydrograph parameters and the catchment
characteristics, it is necessary to correlate the two. In most methods used in
classical hydrology, the correlation has been one of linear regression. It may
be that the use of factor analysis would reveal significant groupings of catchment characteristics. If the same or similar groupings appeared in a number
of different regional studies, the catchment parameter thus indicated could be
tentatively assumed to have general validity and could be used consistently
in a variety of studies. The use of such general parameters might disimprove
slightly the degree of correlation between unit hydrograph parameters and
catchments characteristics for each individual study, but it would make the
various studies comparable with one another and point the way towards
general laws of catchment behavior. It is uncertain, of course, whether the
extra insight gained would be worth the extra work involved in following this
particular line.
In the same year in which he published his classical paper on the unit
hydrograph (57), Sherman published another paper {58) in which he proposed
that for a catchment without records a unit hydrograph be transposed from a
catchment of similar characteristics but with all the time factors adjusted in

proportion to the square root of the ratio of the two areas. In the following
years, a large variety of synthetic models were suggested which involved
correlations between catchment characteristics on the one hand and the unit
hydrograph parameters (or in some cases selected ordinates of the unit
hydrograph) on the other.
The most important of these Avere those proposed by Bernard (4),
McCarthy,^ Snyder {59), :Morgan and Hullinghorst,^ Mitchell Ul), Taylor
and Schwartz {63), the Bureau of Reclamation {66), and the Corps of
Engineers {67). Probably the most widely used method for synthetic unit
hydrographs is that proposed by Snyder, which has since been adapted by
many workers for their own needs. This method will be discussed briefly
below and comparative details of the other methods can be read in the references indicated above, details of which are given at the end of this lecture.
Snyder's work {59) was based on data from 20 catchments in the
Appalachians. He took as the basic unit hydrograph parameter the lag time
to peak {ip) defined as the interval in hours between the center of rainfall
excess and the peak of the unit hydrograph and took as the basic catchment
characteristic the product of the length of the main channel in miles (L) and
the length from the outlet to the center of area of the catchment in miles
{Lea). He suggested that the unit hydrograph parameter and the catchment
parameter could be connected by :
tp = Ct{LLca)'-'
(12a)
Having determined the time to peak of the unit hydrograph, Snyder assumed
that the recession from peak to zero flow took 3 days. He derived the base
length of the unit hydrograph from the formula :
(-)
= 3(^l+^j
(12b)
where B is the base length in days and tp the time lag to peak in hours. Snyder
related the peak of his unit hydrograph to the lag to peak already determined
by the relation :
gmax = 640.^
tp
(12c)
where max is the unit hydrograph peak in cubic feet per second per square
2 MCCARTHY, G. T. THE UNIT HYDROGRAPH AND FLOOD ROUTING. U.S. Corps of
Engineers Office, Providence, R.I. 1939.
2 MORGAN, R., and HULLINGHORST, D. W. UNIT HYDROGRAPHS FOR GAUGED AND UNGAUGED WATERSHEDS. U.S. Corps of Engineers Office, Binghampton, N.Y. 1939. (Unpublished manuscript.)
205
mile, tp is the time to peak in hours, and Cp is a coefficient that takes account
of the flood wave storage effected in the catchment.
For the catchments which he studied in the Appalachians, Snyder found Ct
to vary between 1.8 and 2.2, and Cp to vary between 0.56 and 0.69. Snyder
used a standard duration of rainfall (D) such that:
tp = 5.5D
(12d)
and the peak of the unit hydrograph had to be adjusted for other rainfall
durations. In his original paper, Snyder (59) published a diagram for deriving
the 24-hour distribution graph, but this was not adopted by later workers who
used his basic method.
A number of subsequent workers used Snyder's form of relationship between the lag time to peak and the catchment length parameters. Linsley
(39) found the value Ct varied from 0.7 to 1.0 for catchments in the Sierra
Nevada. The Corps of Engineers (67) found values of the same parameter
varying from 0.4 in southern California to 8.0 for States bordering the Gulf of
Mexico and recommended that the value Ct be determined in a given case from
neighboring or similar catchments. The Corps of Engineers investigations
indicated that the value of Cp could vary from 0.31 in the Gulf of ^Mexico
States to 0.94 in southern California.
In general, the empirical methods for synthetic unit hydrographs tended to
adopt a correlation equation of the general type :
C(A) CiLLâ^
tiior tp or V) = ^ or
(13)
The values of the exponents and the coefficients varied as might be expected.
For example, ]Mitchell {4-1) in his study of 58 Illinois streams found the lag
rime in hours (II) could be related to the area in square miles (A) by:
L = 1.05(A)o.6
(14)
and that the slope did not improve the correlation substantially. This result
becomes understandable when we realize that, for the catchments studied by
Mitchell, the coefficient of correlation between area and slope wâs of the order
of 0.9.
Some of the synthetic unit hydrograph methods resemble Snyder's in that
there is only one correlation with catchment characteristics. If a fixed shape
of unit hydrograph is used, then the synthetic unit hydrograph method is a
one-parameter method. If, however, a further degree of freedom is introduced
by using a relationship between unit hydrograph parameters involving an
adjustable coefficient, as in the case of equation 12c, then the method will
become a two-parameter one. In other cases, such as the method proposed by
Taylor and Schwârtz (64), there are two independent correlations of unit

hydrograph parameters with catchment parameters and again in this case, we
have a two-parameter method for deriving a synthetic unit hydrograph.
Empirical Shapes for the Unit Hydrograph
When unit hydrograph parameters have been determined, it is still necessary
to derive the complete unit hydrograph. If the time-to-peak, the peak discharge, and the base length of the unit hydrograph are known, then we know
three points on the unit hydrograph, and a curve can be sketched in by trial
and error to pass through these three points and to have the requisite area. A
number of authors have suggested particular shapes of dimensionless unit
hydrographs or of S-curves which can be used to determine a complete unit
hydrograph or S-curve, once a single parameter has been determined. Examples of such standard shapes are those described by Langbein (35), Commons
(12), the Bureau of Reclamation (66), the SCS (68), Williams (69), and
Bender and Roberson (3). Since a single curve is used to represent all unit
hydrographs (or all unit hydrographs within a given region, or all unit hydrographs within a given range of watershed size), it is only necessary to
determine one parameter from the catchment characteristics to fix the scale of
the actual unit hydrograph.
If it is desired to introduce more flexibility into the empirical approach, it
would be necessary to develop a family of curves to represent the shape of the
unit hydrograph. In this case, it would be necessary to derive two unit hydrograph parameters from the catchment characteristics. However, if we wish
to synthesize unit hydrographs with two parameters, that is, with two degrees
of freedom, then it is more convenient to use an empirical equation rather than
empirical curves to represent the synthetic unit hydrographs.
The first suggestion of an empirical equation to fit the unit hydrograph
appears to have been made by Edson {20). He argued that the time area curve
for a catchment would have the general parabolic form:
A{t)o,t^
(15)
and that the valley storage acts as a reservoir so that the discharge with time
decreases exponentially :
Q{t)ê-'^
(16)
Edson argued that both effects operate throughout the hydrograph and
therefore that the combined effects could be written as :
Qiit)o:t-e-^t
(17)
which can be normalized and written as :
207
where Q is the discharge per unit area, t is the time, C is a constant depending
on both the volume of inflow and the units used, and a and h are the parameters determining the shape of the unit hydrograph. The reasoning used by
Edson {20) to arrive at equation 18 is faulty, because he uses ordinary multipUcation instead of convolution to represent the effect of storage on the timearea curve. Nevertheless, he arrived at a form of the lUH.
Some years later, Japanese workers in hydrology {56, 61, 62) based the
form of the lUH on a conceptual model consisting of Unear reservoirs and used
as its equation :
h{t) = {ao+ait+a2t^) exp{-\t)
(19)
Following this Nash {46) suggested the model of a cascade of equal linear
reservoirs which gave the equation of the unit hydrograph as :
, , ,
{t/k)^-' exp{t/k)
where /io(0 is the ordinate of the lUH, n is the number of reservoirs, and
K is the storage delay time of each of the reservoirs. Nash suggested that in
fitting equation 20 to unit hydrographs, the value of n need not necessarily be
taken as an integer. Gray {23), Wu {70), and Reich^ all used the same mathematical function to fit derived unit hydrographs and to synthesize further
hydrographs.
The function represented by equations 18 and 20 (which are obviously
equivalent) is variously known in the hydrological literature as the ''gamma
distribution" or ''Nash's model." It is the same as the Pearson Type III
empirical distribution used in statistics, which is commonly written in the
form:
f{x) = ZM+-j expi- j
-a<x<^
(21a)
0<x<a)
(21b)
or
i{x)^--x^-'e-^
1 (Aj
equation 21 is clearly equivalent to equations 18 and 20.

The shape or distribution represented by equation 20, or the equivalent
equation 18, is a two-parameter distribution, K (or h) being a scale factor and
n (or a) being a shape factor. Thus, for complete synthesis, it would be necessary to have two independent relationships between the two parameters of
the gamma distribution and two independent catchment characteristics.
^ REICH, B. M. DESIGN HYDROGRAPHS FOR VERY SMALL WATERSHEDS FROM RAINFALL.
Civil Engin. Sec, Colo. State Univ., 57 pp., illus. 1962.
Edson suggested the direct use of the parameters for correlation purposes,
and this has been done by some later workers. Nash preferred to use the first
moment about the origin (the lag) and the second moment about the center
for correlation. Since these moments can be expressed as very simple expressions involving the parameters n and K, the values determined by one type of
correlation can, in practice, easily be converted to the other. Examples are
given in the next section of the correlation of gamma distribution parameters
with catchment characteristics as derived by Nash {J/.S) and Wu (70). As
mentioned above, the gamma distribution has been widely used in hydrologie
studies.
The model developed by TVA (64-) uses an empirical equation which
essentially involves a time transformation of the gamma distribution. It is
given by :
q{t)=C'^-^
wl
exp(-6-)
(22a)
where
where a, h, and m are parameters. When m has the value of 1, equation 22

reduces to the form of equation 18. The transformed gamma distribution
given by equation 22 has been used in stochastic hydrology by Kritskii and
Menkel (32),
Other mathematical equations have been proposed for the representation
of the form of the unit hydrograph, but none of them have been tested as
widely as the gamma distribution. DeCoursey (14) has proposed the use of
the gamma distribution as far as the point of contraflexure on the falling leg of
the unit hydrograph and then the use of an exponential recession from that
point on. Brakensiek (6) has recently proposed the use of a unit hydrograph
of the form:
which can also be expressed as :

w X
1 exp[- VK/tl
'^'^=2KTi2n-2Ht/Ky
. , .
^^^^^
and can be shown to be equivalent to a Pearson Type V empirical distribution

wdth a square root transformation of the time scale. It has two parameters;
hence, the problems of fitting and correlation would be essentially the same as
for the gamma distribution.
209
Conceptual Models of the Unit Hydrograph

In the preceding sections, we have traced the development of synthetic
unit hydrographs along two different lines. We have seen, as outlined on figure
8-1, that the line of development based on the time-area diagram led to the
conceptual model of routing an isosceles triangle through a hnear reservoir and
that the line of development based on purely empirical relationships led to the
use of the gamma distribution, which Nash {48) showed to be equivalent to
the conceptual model of a cascade of equal linear reservoirs. Within recent
years, attention has been concentrated on the simulation of the direct response
of catchments by conceptual models.
For a conceptual model to be an adequate tool for synthesizing unit hydrographs, it must provide a convenient method for predicting the shape of
the unit hydrograph, and a relationship must also be established between the
basic parameter of the conceptual model and the catchment characteristics.
For any conceptual model, we can relate such unit hydrograph parameters as
the lag {II), the time to peak {Q, or the peak discharge (max) with the basic
parameters of the model. Hence, it should be possible to combine a conceptual
model with any of the empirical relationships between unit hydrograph
parameters and catchment characteristics (some of which were reviewed in an
earlier section), which have been derived independently of any conceptual
model. Because we are dealing with synthetic unit hydrographs in this lecture,
we will concentrate on conceptual models of hnearized systems but will
indicate, where appropriate, the way in which the approach can be extended
to cover the simulation of nonlinear systems.
The use of conceptual models is quite expHcit in a paper by Sugawara and
Alaruyama {63) published in 1956. Starting with the case of a river where the
unit hydrograph could be approximately represented by a negative exponential
function, the authors developed a conceptual reahzation of the system operation in the form of an open vessel filled with water. The water discharges
through a capillary tube at the bottom, thus giving a linear relationship
between outflow and storage in the vessel. They then attempted to model the
behavior of certain rivers by means of the sum of several exponential components, that is, by using several different vessels with different storage constants arranged in parallel and taking different proportions of.the inflow.
By placing the capillary at a level higher than the bottom of the vessel, the
threshold effect of initial storage satisfaction could be simulated. (Further)
conceptual elements used were vessels tapped by capillaries at a number of
points, which produced a segmented linear storage-discharge relation that
could approximate a nonhnear relationship and, hence, simulate a nonlinear
system.
Shortly afterwards, Nash {46) pubUshed his work suggesting the gamma
distribution as the appropriate equation for the lUH. He derived this equation
by considering the effect of routing a delta function through a cascade of

equal linear reservoirs. For n such reservoirs in series, the impulse response of
the cascade (that is, the discharge from the last reservoir for a -function input
to the first) takes the form of equation 20, where K is the storage delay time in
each reservoir and n is the number of reservoirs. Nash suggested that this
equation could be generahzed and n allowed to take nonintegral values.
Nash also gave heuristic arguments for beheving that for a cascade of unequal linear reservoirs, the shape would not differ greatly from that given by
equation 20. His arguments suggested that for a given value of the dimensionless second moment {m^), the dimensionless third moment (ms) for a cascade
of unequal linear reservoirs would lie between the value for a cascade of equal
linear reservoirs, that is, 2(m2)^ and the value for the combination of a linear
channel and a linear reservoir in series, that is, 2(m2)^^^. For m2 = 0 or 1, the
values of niz are seen to coincide, and it can be readily verified that for values
of m2 between 0 and 1 the lines corresponding to the two limiting cases enclose
a comparatively narrow^ region of the mz m^ plane.
In 1959, Dooge {17) attempted to produce a general conceptual model of
the unit hydrograph. The argument was made that since the unit hydrograph
only existed for a linear system or a linearized system, a general model of the
unit hydrograph could contain only linear elements. As mentioned previously,
w^hen we wish to simulate we must first make up our minds about the type of
simulation and then about the components of our model. In this case, it was
decided to use as components of the model only linear distortionless channels
and linear storage elements. In an actual watershed, the inflow at any point
travels through the system to the outlet and in doing so is subject to both
translation effects and storage or attenuation effects.
The assumption made in Dooge's conceptual model {17) was that these two
effects could be completely separated from one another. The effects of translation in different parts of the catchment were considered to be lumped together
and represented by linear channels, whereas the storage effects in the various
parts of the catchment were lumped together and represented by linear
reservoirs. Since the model is a linear one, we have the full advantage of superposition and the operations may be carried out in any order. Since linear
channels merely delay an inflow without distorting it, any number of hnear
channels can be connected together to form one linear channel. Similarly, the
order of the linear reservoirs in a cascade can be altered without affecting the
response of the system. Channels and reservoirs can also be interchanged
without affecting the response of the system.
The most general model developed was one in which the storage in different
parts of the watershed was concentrated so that the flow from any part of the
watershed could be simulated by a linear channel whose length corresponded
to the time of translation (or time of concentration for that point) and a
number of linear reservoirs whose storage time need not be equal. If the
assumption is now made that for every point along an isochrone (that is, for
211
equal translation time to the outlet) the cascade of reservoirs to be passed

through in reaching the outlet are the same, the equation of the unit hydrograph can be written as :
where ho(t) is the ordinate of the lUH, V is the volume of inflow, T is the time
of concentration of the whole watershed, W{T/T) is the time-area-concentration curve, {t) is an impulse function, Ki is a typical reservoir storage
delay time, D is the differential operator, and JI represents successive
multiplication.
If the assumption is now made that the cascade of unequal linear reservoirs
appropriate to a given isochrone can be replaced by a cascade of equal linear
reservoirs, then the unit hydrograph can be written as the convolution of the
time-area-concentration curve and a gamma distribution, as follows:
exp[-(/iC)](/iC)n-i
"\TJ
K{n-l)\
-<^
V
^^
where n is not a fixed value but varies with the value of t. The general model
represented by equation 25 is still extremely flexible. If n = 1 for all points on
the catchment, then the model reduces to the Zoch-Clark model of routing
the time-area diagram through a linear reservoir. If n is greater than 1 but the
same for all points in the catchment, then the model represents routing the
time-area diagram through a number of reservoirs all situated at the outlet.
If the time-area-concentration curve is itself a gamma distribution with the
time scale K, then the model given by equation 25 reduces to the Nash model
of a cascade of equal linear reservoirs with inflow at the upstream end.
A number of conceptual models have been developed by graduate students
working under Professor Ven Te Chow {8) at the University of Illinois. The
model parameters in these cases were correlated not only with the catchment
characteristics but also with the intensity of rainfall. The analysis was consequently one of a linearized system rather than a linear system. Such an
approach takes account of the nonlinear effects due to varying levels of input.
The model used by Singh^ consisted of translation to the outlet and then
successive routing through two linear reservoirs of different storage coeflficients. The response function for this model would be :
'io(o='(0*
-i/K2_^-t/Ki
K2 KI
(26)
^ SINGH, K. P. A NONLINEAR APPROACH TO THE INSTANTANEOUS UNIT HYDROGRAPH.

Ph.D. thesis, 111. Univ. 1962.

Disking used two cascades of linear reservoirs in parallel, the number of
reservoirs and the storage coefficients being different in the two cases. The
response function for this model is :
(27)
cascade, ni and n2 are the number of equal reservoirs in each cascade and Ki
and K2 are the respective storage delay times.
Kulandaiswamy^ used a model which can be described as a generalized
Muskingum model. As shown in lecture 2, pages 43-57, the essential assumption of the Muskingum method of flood routing is that the storage is a linear
function of the inflow and the outflow. When the expression for the storage is
inserted in the continuity equation, an equation for the system is obtained
linking inflow and outflow and their first derivatives. If the Muskingum
assumption is extended to make the storage a function not only of the inflow
and the outflow but also of their derivatives, then wê have what might be
called a generalized Muskingum model. If the coefficients of the terms in the
general relationship depend on either the inflow, or the outflow, or both, then
we have a generalized nonhnear Muskingum model.
Kulandaiswamy restricted his detailed analysis to the case of a linearized
system in which the derivatives of the outflow higher than the third and the
derivatives of the inflow higher than the second were ignored, thus giving as
a general equation :
where I is the inflow to the system and Q the outflow, and ai, a2, 3, ^1 and 62
are constants, which are parameters of the system. For a heavily damped
system, all the roots of the polynomial on the left-hand side of equation 13 wîll
be real and negative. If the system can be represented by a number of cascades
in parallel (without reverse flow), then the values of 61 and 62, in the form given
by Kulandaiswamy in equation 28, will be negative. If 61 and 62 are both equal
to zero, then the system reduces to a cascade of three linear reservoirs w^hose
delay times are given by roots of the polynomial on the left-hand side of the
equation. If the coefficient 61 in equation 28 is negative and the coefficient 62 is
zero, then the model will in general consist of two cascades in parallel, each
6 DiSKIN, M. H.
A BASIC STUDY OF THE LINEARITY OF RAINFALL-RUNOFF PROCESS IN
WATERSHEDS. Ph.D. thesis, 111. Univ. 1964.

7 KULANDAISWAMY, V. C. A BASIC STUDY OF THE RAINFALL EXCESS-SURFACE RUNOFF
RELATIONSHIP IN A BASIN SYSTEM. Ph.D. thesis, 111. Univ. Urbana. 1964.
213
made up from linear reservoirs whose delay times are given by the polynomial
on the left-hand side of equation 28. If both 61 and 62 are negative, then the
equation will represent a system of three cascades in parallel.
In classical hydrology, use is made of routing through a nonlinear reservoir
in which the storage is proportional to some power of the outflow. If the
outflow is controlled by a weir, the exponent in the storage equation would
be three-halves; whereas for an outflow controlled by a deep sluice, the power
would be one-half. Prasad {51) introduced a conceptual model in which the
storage was expressed as the sum of two terms, the first of which is related to
some power of the outflow (as for the nonlinear reservoir) and the second of
which involves the rate of change of outflow.
As with all types of models, it is necessary to find the optimal values of the
parameters of a conceptual model. This can be done by the method of least
squares, by minimax error, by matching of moments, or by a direct search
technique on a digital computer. To use the method of least squares, it would
be necessary to differentiate the equations for the lUH with respect to each
of the parameters in turn and solve the resulting simultaneous equations. This
may involve us in some complex mathematics. It is easy enough to differentiate the gamma distribution with respect to time to find its peak, but to
differentiate it with respect to n or ii soon leads us into an undergrowth of
unfamiHar mathematical functions. Where conceptual models have been
used, the criterion of fit has been that the model should match the two coordinates of the peak of the empirically derived hydrograph. In effect, such a
criterion means matching the model to the prototype at two points only
(the origin and the peak) and ignoring the information available in the
remainder of the hydrograph.
In practice, it has been found relatively easy to compute the moments of
most conceptual models. This suggests that matching by moments be used as
the criterion for determining the optimal values of the parameters. The
general formula for the R^^ moment of the impulse response of a linear reservoir
about the origin is given by :
U'R={R)\K^
(29)
and the general expression for its cumulant is :

kR={R-l)\K^
(30)
If linear storage elements are combined in series, then the cumulants of the
resulting cascade are obtained by adding together the corresponding cumulants of the individual reservoirs. If linear reservoirs are combined in parallel,
the moments of the resulting system about the origin can be obtained by
adding the individual moments about the origin. The moments and the
cumulants have the advantage that they take into account the complete unit
hydrograph, but for the higher moments there is the disadvantage that the
recession limb of the hydrograph makes a dominant contribution to the value

of the moment and errors in the recession may distort this value.
Where a time-area-concentration curve is represented by a geometrical
figure and routed through a linear reservoir, then the cumulants of the resulting conceptual model are obtained by adding the cumulants of the geometrical figure representing the time-area-concentration curve and the
cumulants of the linear reservoir. Thus, for the case of the routed isosceles
triangle, if the base of the triangle is given by T and the storage delay time of
the linear reservoir by K, the cumulants of the resulting model are as follows :
k,= Ui' = ^+K
(31a)
],^=U2 = +K'
24
(31b)
]^^=Uz = 2K'
(31c)
k,= U,-3{U2y = QK^-^

950
(31d)
If the respective moments (or cumulants) of the conceptual model are

equated to the derived moments (or cumulants) of an empirical hydrograph,
then the values of the parameters that are optimal in the sense of moment
matching can be evaluated.
As mentioned at the beginning of the section, a conceptual model can only
be used to synthesize an actual unit hydrograph if some rule is available for
predicting the values of the parameters of the conceptual model on the basis of
readily available catchment characteristics. Usually such rules are based on
the correlation of the model parameters with catchment characteristics for
unit hydrographs derived from catchments where records are available. If the
parameters of the conceptual models chosen for conceptual models are not
very stable, or if the optimal values of the parameters cannot be sharply
identified from the past records of input and output, then the correlations on
which is based the synthesis of the unit hydrograph for a ungaged catchment
will be unreliable.
There is a great deal to recommend the proposal by Nash (47) that the
moments be used as the basis of this correlation because the lower order
moments are more stable than such parameters as time-to-peak and peak
discharge. On the basis of 90 storms on 30 British catchments (whose area
varied from 4.8 to 859 square miles), Nash (48) derived the relationship:
i7/ = ^ = 27.6(-)
(32)
215
where Ui is the first moment or lag, A is the area in square miles, and S is the
overland slope in parts per ten thousand. Before adopting this relationship,
Nash had tried the regression of the first moment on various combinations of
nine catchments characteristics. The coefficient of multiple correlation {R)
for the relationship given in equation 32 was 0.90. When the dimensionless
second moment was correlated against the catchment characteristics, the
best result obtained was :
f/2
0.41
"^^^T^ZT
(33)
where m^ is the dimensionless second moment and L is the length of the longest
stream to the catchment boundary in miles. In this second regression, the
coefficient of multiple correlation {R) was 0.45.
Once the moments of the unit hydrograph have been determined and
estimated, the equations relating the moments of the conceptual model chosen
for synthesis to the basic model parameters can be solved for the values of
these parameters. In gamma distributions, the parameters can be determined
directly from the moments since we have :
K=
(34)
n=
(35)
nil
The parameter values derived in this way can then be used to generate the
particular gamma distribution which is used as a representation of the lUH
for the catchment being studied.
Wu {70) has reported on a synthetic method derived from the records for
21 watersheds in Indiana varying from 7 to 100 square miles. He correlated
the time-to-peak with catchment characteristics and found:
_ 31.42(A)^-o85
where tp is the time to peak in hours, A is the area in square miles, L is the
length of the main stream in miles, and S is the slope of the main stream in
parts per ten thousand. The other parameter which was correlated by Wu was
the recession constant C, for which he proposed:
780 (A) 0-937
~(L) 1-474(^)1.473
(37)
where K is the storage constant in hours, and the catchment characteristics
are as defined for equation 36. Because the time-to-peak for the gamma distribution is related to the parameters n and K by:
t^={n-l)K
(38)
there is no difficulty in deriving the value of n from equations 36 and 37 and

so generating the synthetic unit hydrograph.
Comparison of Methods
In this lecture, we have outlined a number of methods which can be used in
attempting to solve the problem of synthetic unit hydrographs, that is, the
problem of predicting the unit hydrograph for a watershed in which no records
of inflow and outflow are available. A large number of methods have been
proposed, some belonging to the category of time-area methods, some to the
category of empirical methods, some to the category of conceptual models.
The hydrologist faced with an immediate problem, but anxious to use as
objective a method as possible might well ask, '^How shall I choose between
these methods?'' To answer this question, it is necessary to compare the
methods belonging to each category and also to compare the different categories.
In time-area methods, we must decide whether to use the actual time-areaconcentration curve or a geometrical figure and whether to route through one
or more linear reservoirs. It would appear that the extreme tedium of deriving
a time-area-concentration curve is not justified by any appreciable increase
in accuracy in representing actual unit hydrographs and that the judicious
replacement of the time-area-concentration curve by a geometrical figure is
unobjectionable. Care must be taken, however, that a catchment of untypical
shape is not forced into the straight jacket of being represented by a geometrical figure whose shape is based on the general shape of other catchments
in the region. Once it has been decided to route a geometrical figure rather
than a derived area-concentration curve, the problem really reduces to one of
a conceptual model. The question of what figure to use and how many reservoirs to route through can be determined by the methods given below for
conceptual models.
The empirical curves used to represent the unit hydrograph (nearly all of
which are one-parameter models) can be compared by dimensionless plotting.
It is important to remember that for one-parameter curves only one parameter
is available to act as a scale factor. Thus, a comparison by plotting the ratio
of discharge to peak discharge against time over time-to-peak may not be
vaUd as the volume under the hydrograph may not be normalized to unit
volume.
Both theoretical considerations and practical results in the field indicate
that the lag (the time interval between the center of precipitation excess and
the center of direct storm runoff) is the most stable time parameter and the
217
one most highly correlated with the catchment characteristics usually used.
It is suggested, therefore, that any comparison of dimensionless unit hydrographs should be made by plotting:
where q is the ordinate of the unit hydrograph, h is the lag as defined above,
V is the volume of rainfall excess, and D is the duration of rainfall excess.
Similarly any dimensionless S-curve should be plotted as:
S{t)=cl>(j
(33)
In each case, 0 is an undefined function representing the standard shape

adopted.
A complete comparison of different methods of synthetic unit hydrograph
generation based on empirical curves must take into account the empirical
relationships with catchment characteristics used to determine the basic
unit hydrograph parameter or parameters. A number of comparisons have
been made but none of them were comprehensive. Dooge^ compared, in a
crude fashion, the shape obtained by routing an isosceles triangle through a
linear reservoir with the shape of the dimensionless unit hydrographs proposed
by Commons (12), WiUiams (69), and the Soil Conservation Service (68).
The comparison was made by plotting the ratio of the discharge ordinate to
maximum discharge (q/q^s.^) against the ratio of the time to the time-of-peak,
(t/tp). Because all the curves were constrained to go through common points
at the origin and the peak, no great differences were revealed.
Coulter (13) in a study of rural catchments in New South Wales, compared
the synthetic unit hydrographs generated for nine catchments by the methods
of Taylor and Schwartz (63), Clark and Johnstone (9, 29), Eaton (19),
OTvelly (49), and Morgan and Hulhnghorst.^ For a few of the catchments,
the peak flows predicted by the various methods were quite close to one
another, but other catchments showed a three- or fourfold variation.
Dooge (footnote 1) put forward the idea of comparing methods of synthetic
unit hydrograph generation on the basis of their predictions of the unit
hydrograph for one or more standard catchments. A standard catchment is
taken as being one in geomorphological equilibrium. Though all catchments
are not in equilibrium, it may be assumed that catchments out of equilibrium
are tending to equilibrium and tend so more rapidly the more they are out of
equilibrium. Once the size of the standard catchment was fixed at 100 square
miles, the remaining topographical characteristics were fixed on the basis of
8 See footnote 1, p. 199.
9 See footnote 3, p. 204.
geomorphological principles and published relationships. In this case, the

channel slope was taken as 100 feet per mile and the ground slope as 400 feet
per mile. Other characteristics were a drainage density of 1.25 and a length
of overland flow of 2,200 feet.
Morgan and Johnson {42) compared the relative accuracy and reliability
of the synthetic unit hydrograph methods proposed by Snyder (4^), the Soil
Conservation Service {68), Commons (^12), and :Mitchell {41). They applied
the methods to 12 drainage areas in Illinois ranging in size from 10 to 101
square miles. Again wide variations between the synthetic unit graphs and
between them and the actual unit graph were found. No method consistently
over- or underestimated the actual peak discharge. The highest methods of
peak discharge ranked in the following order: SCS, Commons, Mitchell, and
Snyder. There was little difference between the estimates by the SCS method
and the "Commons method. When an observed lag was used instead of a lag
estimated on the basis of catchment characteristics, the synthetic methods
gave much better results. Studies by Coulter {13) and by IMorgan and
Johnson {^2) are of interest because they test the general applicability of the
empirical relationships between unit hydrograph parameters and catchment
characteristics originally derived from regions which are widely separated
from one another.
The following tabulation shows the ability of a number of methods to
predict the lag of a standard catchment. When McCarthy's method (see
Author
McCarthy (1938)
Snyder (1938)
Mitchell (1948)
O'Kelly (1955)
Nash (1960)
Location
Connecticut
Appalachians
Illinois
Ireland
Britain
Lag in hours
16.2
16.5
15.4
14.0
15.9
footnote 2), which was based on a very few streams in Connecticut, was
apphed to the standard catchment of 100 square miles, a lag of 16.2 hours was
obtained; Snyder's method based on work in the Appalachians gave a lag of
16.5 hours; Mitchell's method based on watersheds in Illinois gave a lag of
15.4 hours; O'Kelly's method based on a number of catchments in Ireland,
gave 14.0 hours; and Nash's method based on catchments in Britain a lag of
15.9 hours. With the exception of the result by O'Kelly's method, these are all
remarkably close varying only from 15.4 to 16.5 hours, a difference of only
7 percent. In addition to this remarkable concordance, there is a reason why
the catchments studied by O'Kelly would be expected to have shorter lag
times for standard dimensions. O'Kelly was concerned with the problem of
designing arterial drainage schemes (main river improvement schemes) in
Ireland. Accordingly, his method was based on the characteristics of rivers for
which such schemes had been carried out. Under post-drainage conditions,
219
catchments are expected to show shorter lag times than the average. It would
appear that, while the dimensionless unit hydrographs, which were used in the
past is a purely empirical fashion, will be replaced by mathematical equations
or by conceptual models, the correlations of hydrograph parameters (particularly lag) with catchment characteristics developed in the classical synthetic
methods may still be useful.
There remains the problem of deciding how complex the mathematical
equation or conceptual model must be and how to choose between different
models (or equations) of equal complexity. Nash (47, 48) proposed a general
synthetic scheme along the lines shown in figure 8-3. As has been repeatedly
emphasized, the relationships on which the synthesis are based must be
derived from the analysis of a number of watersheds for which measurements
are available and which serve as a sample for the region. Nash suggested that
the moments of the lUH be derived from a set of sample catchments and these
moments correlated with one another and with the catchment characteristics
to determine the number of degrees of freedom inherent in the response of a
catchment when operating on precipitation excess to produce flood runoff.
This would enable us to determine the number of parameters needed in the
simulation system. He suggested that the dimensionless moments of the actual
MOMENTS
OF lUH
CATCHMENT
PARAMETERS
'
'
\'
lUH
PARAMETERS
BEST
MODEL
\r
lUH
FIGURE
8-3.Nash's synthetic scheme.
DEGREES OF
FREEDOM

responses (and the dimensionless moments of a number of conceptual model
systems with the required number of parameters) be plotted against one
another. On such a plot of niz against m2 or m4 against m2, a one-parameter
model would plot as a single point; a two-parameter model, as a single curve;
and a three-parameter model, as a family of curves. By comparing the curves
for the model system with the plot of points for actual catchments, the best
model could be chosen.
Once the correlations of lUH moments with catchment characteristics
have been determined and the model chosen, it is possible to synthesize the
unit hydrograph for a watershed for which no records are available. Firstly,
the moments of the lUH are determined by the regression equations using the
values of the catchment characteristics of the particular catchment. These
predicted moments can be equated to the expressions for the corresponding
moments of the model chosen, and the optimal values of the model parameters
for the particular catchment thus determined. Once the model and the optimal
values of the parameters are known, the complete lUH for the particular
watershed can be generated.
Nash (4S) applied his method to the data for 90 storms on 30 catchments in
Great Britain. Regression analysis gave a relationship between the first
moment (mi) and the catchment characteristics of area and overland slope
with a coefficient of multiple correlation of 0.90. A further regression of the
second moment (^2) with mi and the overland slope gave a coefficient with a
multiple correlation of 0.51. Though the latter result is statistically significant,
it does not give a good determination of the second moment and, hence, the
ability of the scheme to predict an unknown unit hydrograph is impaired.
When Nash plotted the moments of his actual responses against one another,
as shown on figure 8-4, they covered a region rather than fell along a single
line. In discussing Nash's paper, Dooge pointed out that the data and the
curve are not strictly comparable. The data were derived on the assumption
that the base of the unit hydrograph was three times its lag; whereas, the
base of the gamma distribution is infinite. A crude correction can be made
and the two made more comparable by truncating the gamma distribution
according to the method of base now separation given by Nash so that the
base of the truncated gamma distribution is three times its lag. When this is
done, the truncated gamma distribution plots as a line lying below the data
points shown on figure 8-4 and, thus, appears to approximate a limiting
form rather than an average form for the lUH's derived by Nash. Figure
8-4 also shows the comparison of the data with three models: (1) a channel
and reservoir in series (curve A); (2) a cascade of equal linear reservoirs with
an upstream inflow, that is, the gamma distribution (curve B) ; (3) and the
cascade of equal linear reservoirs with lateral inflow (curve C).
The general synthetic scheme proposed by Nash could, with advantage, be
modified to the scheme outlined on figure 8-5. It is suggested that, instead of
221
correlating the moments with the catchments characteristics, the moments be

correlated among themselves to determine the number of degrees of freedom.
Thus, in a two-parameter system, the third moment will be completely determined once the first and second moments are known; whereas in a threeparameter system, the fourth moment will be known once the first, second, and
third moments are known. If the moments are made dimensionless by using
the first moment as a scaling factor, then the criterion for a two-parameter
model is that the third dimensionless moment is completely determined by the
second dimensionless moment (or cumulant) ; the criterion for a three-parameter system is that the fourth dimensionless moment (or cumulant) is completely determined by the second and the third dimensionless moments.
In his discussion of Nash's paper, Dooge (18) calculated the coefficient of
multiple correlation of ms with m2 for Nash's data as 0.717. This indicated
that the variation in the third dimensionless moment (ms) was onl}^ 50 percent
accounted for by variations in the dimensionless second moment (m2) and,
hence, that the two-parameter model would not be highly efficient as a basis
for simulation. However, the coefficient of multiple correlation between the
FIGURE
8-4.Shape factor diagram for Nash's data.
CATCHMENT
PARAMETERS
i
MOMENTS
OF lUH
BEST
MODEL
lUH
PARAMETERS
DEGREES OF
FREEDOM
\f
lUH
FIGURE
8-5.Modified synthetic scheme.
dimensionless fourth moment (^4) and the two lower dimensionless moments
{mz and m^^) was found to be 0.93, indicating that the variance in m^ was
accounted for by the variance in the lower moments to the extent of almost
90 percent. Considering the basic nature of Nash's date (which were normal
river observations rather than research readings), this was a very high correlation and indicated that a three-parameter model would probably give as
satisfactory a simulation as the data warranted. The remainder of the modified
general synthetic scheme shown on figure 8-5 is the same as for Nash^s original
proposal shown on figure 8-3, except that the parameters of the lUH are
correlated directly with catchment parameters.
It must be stressed that what is required in the correlation for unit hydrograph synthesis is not necessarily a correlation with individual catchment
characteristics. To determine the three independent lUH parameters that
would be required for a three-parameter model, it is necessary to have three
independent catchment parameters which between them would account for
90 percent or more of the variation in the shape of the lUH. Each of these
parameters might be made up from a number of catchment characteristics
(such as area, slope, drainage density, and shape) in the same way as the
Froude number and the Reynolds' number are made up from a number of
hydraulic characteristics.
The determination of the significant grouping of catchment characteristics
223
into catchment parameters remains one of the great unsolved problems of

surface water hydrology. Factor analysis may help in the preliminary trial
grouping of catchment characteristics, but it is likely that the final significant
forms of the groupings will only emerge through a better understanding of
geomorphological processes.
The shape factor diagram in which dimensionless moments or cumulants
are plotted against one another is a most useful device for comparing alternative conceptual models of the same number of parameters and of comparing
conceptual models with actual data. Thus, figure 8-6 shows a comparison
2.or
A
B
Routed rectangle
Routed triangle
Cascade of reservoirs
FIGURE
&-6.Shape factor for models.

between three two-parameter conceptual models: (1) a routed rectangle, (2) a
routed triangle, and (3) a cascade of equal linear reservoirs. It can be seen
that each of these two-parameter conceptual models defines a line in the
SsS2 plane, where S3 is the dimensionless third cumulant (or moment) and
S2 is the dimensionless second cumulant (or moment). These lines plot relatively closely together on the diagram, thus explaining why all of these models
have been suggested as a basis for simulating the same type of prototype
system.
Figure 8-7 shows a comparison of a Pearson Type III distribution (gamma
distribution) with a Pearson Type V distribution and a Pearson Type I distribution. This diagram could be used to decide which Pearson distribution to
choose for fitting unit hydrographs or other response curves by taking the
distribution which lay closest on the shape factor diagram to the plotted
points corresponding to the unit hydrographs for a number of sample catchments.
Figure 8-8 shows a comparison between the time-transformed gamma
distribution and the ordinary gamma distribution. The case plotted is for a
value of m = 3^ and corresponds to the type of model used by TVA. If curves
were drawn for other values of m, it would be possible to see if the plotted
points from sample catchments all fell along one line. This would enable us to
use a two-parameter model based on the value of m corresponding to that line
or else to indicate whether the family of curves swept out the region of plotted
points, thus allowing us to use equation 22 as a three-parameter simulation
of the prototype system.
Problems on Synthetic Unit Hydrographs

1. Compare the values of the lag, time-to-peak, and peak discharge given
by four different synthetic unit hydrograph methods for the 100 square mile
catchment whose characteristics are listed on Appendix table 6.
2. Compare the values of the lag, time-to-peak, and peak discharge given
by four different synthetic unit hydrograph methods (two empirical and
two time area) for the catchment whose characteristics are given on Appendix
table 7.
3. Compare a number of standard unit hydrograph shapes by plotting
dimensionless ordinates against dimensionless time.
4. Compare a number of standard S-curves by plotting dimensionless
ordinates against dimensionless time.
5. Compare a number of standard unit hydrograph shapes on a plot of
dimensionless third moment versus dimensionless second moment.
6. Describe the various steps of one particular method for synthetic unit
hydrographs, and comment on the strong and weak points in the method.
225
2.0 h
TYPE I
rn|/
j. v2m|+|
f =k(l + ^) '(l-:^J
TYPE ni f = k(l + ^)'êxp("4r)

TYPE H
FIGURE
f = -^;n
r(p-l)
8-7.Shape factor for Pearson's empirical distributions.
3.0 r
h(t) =
m exp[-(t/Kr](t/K)""'
k
(n/m-1)1
/nl
k, = K V m
ko =K'
FIGURE
lil!
(^-0!(^-')!-(^ -0!
(ft-O!
8-8.Shape factors for time-transformed gamma distribution.
227
7. Describe the relationship between the generahzed JMuskingum formulation of linear catchment response with the general linear equation with
constant coefficients. Illustrate by simple example.
8. Describe the relationship between the general linear equation with
constant coefficients and the representation of the linear catchment response
in terms of equal linear storage elements. Illustrate with a simple example.
9. Describe the relationship between the general linear equation with
constant coefficients and the representation of the catchment by a small
number of unequal storage elements. Illustrate your answer by a simple
example.
10. Choose a special linear model with an infinite time base. Calculate the
effect of truncating this response curve to make the base finite and work out
the corrections to be made to the moments of the response curve.
11. Compare a number of two-parameter models by plotting dimensionless
peak discharge against dimensionless time-to-peak.
12. Compare a number of two-parameter models by plotting in terms of
dimensionless shape factors.
13. Devise a general synthetic scheme for linear catchment response. Work
out a flow diagram for the computations involved. Apply this flow diagram
to a set of reliable data.
Literature Cited
(1) AMERICAN SOCIETY OF CIVIL ENGINEERS.
1969. DESIGN AND CONSTRUCTION OF SANITARY AND STORM SEWERS. Manual and
Reports on Engineering Practices No. 37. 128 pp. New York.
(2) BARNES, B. S.
1952. UNIT-GRAPH PROCEDURES. U.S. Bur. Reclam. Washington, D.C.
(3) BENDER, D. L., and ROBERSON, J. A.
1961. THE USE OF A DIMENSIONLESS UNIT HYDROGRAPH TO DERIVE UNIT HYDROGRAPH FOR SOME PACIFIC NORTHWEST BASINS. Jour. Gcophys. Res. 66(2):
521-527.
(4) BERNARD, M. M.
1935. AN APPROACH TO DETERMINATE STREAMFLOW. Amer. Soc. Civ. Engin.
Trans. 100: 347.
(5) BETSON, ROGER P., and GREEN, RALPH F.
1968. ANALYTICALLY DERIVED UNITGRAPH AND RUNOFF. Amer. SoC. Civ. Engin.,
Jour. Hydraulics Div. 94(HY-6): 1489-1505.
(6) BRAKENSIEK, D. L.
1967. A TECHNIQUE FOR ANALYSIS OF RUNOFF HYDROGRAPHS. Jour. Hydrol.
5(1): 21-32.
(7) CHAMIER, G.
1897. CAPACITIES REQUIRED FOR CULVERTS AND FLOOD-OPENINGS. Inst. Civ.
Engin. Proc. 134: 313.
(8) CHOW, VEN TE.
1964. RUNOFF. Section 14: 54 pp.. In Chow, Ven TE, ed.. Handbook of Applied
Hydrology. New York.

(9) CLARK, C. O.
1945. STORAGE AND THE UNIT-HYDROGRAPH.
Amer. Soc. Civ. Engin. 110: 1416-
1446.
(10) CLARK, JOHN I.
1966. MORPHOMETRY FROM MAPS. 40 pp. In Dury, G. H., ed.. Essays in Geomorphology. London.
(11) CoLEMAN, G. S., and JOHNSON, A.
1931. RAINFALL RUNOFF. Inst. Munic. County Engin. Jour. 58: 1403.
(12) COMMONS, C. G.
1942. FLOOD HYDROGRAPHS. Civ. Engin. 12: 571-572.
(13) COULTER, C. G.
1961. FLOOD ESTIMATION FOR UNGAUGED RURAL CATCHMENTS. Water Res. Lab.
Rpt. 40. University of New South Wales, Sydney.
(14) DECOURSEY, D. G.
1966. A RUNOFF HYDROGRAPH EQUATION. U.S. Dept. Agr. Agr. Res. Serv. ARS
41-116, 23 pp.
1955. Discussion of the employment of unit-hydrographs to determine the flows of
Irish arterial drainage channels, by J. J. O'Kelly. Inst. Civ. Engin. (Ireland), Proc. V. 4, pt. II.
(16)
1957. THE RATIONAL METHOD FOR ESTIMATING FLOOD PEAKS. Engineering 184:
311-374. London.
(17)
1959. A GENERAL THEORY OF THE UNIT HYDROGRAPH. Jour. GeophyS. ReS.
64(2): 241-256.
(18)
1961. Discussion of a unit hydrograph study with particular reference to British
catchments, by J. E. Nash. Inst. Civ. Engin. (Ireland) Proc, v. 20: 467470.
(19) EATON, T. D.
1954.
THE DERIVATION AND SYNTHESIS OF THE UNIT-HYDROGRAPH WHEN RAINFALL
RECORDS ARE INADEQUATE. Hydrologic studies and investigations.

Engin. Austral. Jour. 26: 239-243.
Inst.
(20) EDSON, C. G.
1951.
PARAMETERS FOR RELATING UNIT HYDROGRAPHS TO WATERSHED CHARACTER-
ISTICS. Amer. Geophys. Union Trans. 32(4): 591-596.

(21) ESCRITT, L. P.
1950. SURFACE WATER SEWERAGE. 200 pp., illus. London.
(22) GRAY, D. M.
1961.
SYNTHETIC UNIT HYDROGRAPHS FOR SMALL WATERSHEDS.

Amer. SoC. Civ.
1962.
DERIVATION OF HYDROGRAPHS FOR SMALL WATERSHEDS FROM MEASURABLE

PHYSICAL CHARACTERISTICS. lowa State Univ., Agr. and Home Econ.
Expt. Sta. Res. Bui. 506: 514-570.
(23)
(24) HARTE, C. A.
1932. Discussion of rainfall-runoff calculations, by M. T. M. Ormsby.
Munic. County Engin. Jour. 59: 978.
(25) HAWKEN, W. H., and Ross, C. N.
1921.
Inst.
THE CALCULATION OF FLOOD DISCHARGES BY USE OF A TIME-CONTOUR PLAN.
Inst. Engin. Austral. Jour. 2: 85-92.
229
(26) HENDERSON, F. M.
1963. SOME PROPERTIES OF THE UNIT HYDROGRAPH. Jour. Geophys. Res. 68(16).
(27) HoRTON, R. E.
1941. VIRTUAL CHANNEL INFLOW GRAPHS. Amer. Geophys. Union Trans. 22:
811-820.
(28) JENS, S, W.
1948.
DRAINAGE OF AIRPORT SURFACESSOME BASIC DESIGN CONSIDERATIONS.

1949. ELEMENTS OF APPLIED HYDROLOGY. 276 pp., iUs. NeW York.
(30) JUDSON, C. C.
1932. RUNOFF CALCULATIONS, A NEW METHOD. Inst. Munic. County Engin. Jour.
59: 861.
(31) KiRPICH, Z. P.
1940. TIME OF CONCENTRATION OF SMALL AGRICULTURAL WATERSHEDS. Civ.
Engin. 10(6): 362.
(32) KRITSKII, S. N., and MENKEL, M. F.
1950. GIDROLOGICHESKIE OSNOVY RECHNOI GIDROTEKHNIKI [HYDROLOGICAL BASIS
OF RIVER HYDROTECHNics]. Izadatel'stvo AN SSR.
1964. OBOBSHCHENNAYA FORMOOLA DLYA RASCHETA GIDROGRAFA STOKA [A GENERALIZED FORMULA FOR THE CALCULATION OF THE DISCHARGE HYDROGRAPH].
Central Forecasting Unit Trans. (TsIP), 113: 23. Moscow. NLLST
RTS No. 3355, Boston Spa, Yorkshire, England. Sept. 1966.
(34) KuiCHLiNG, EMIL.
1889. THE RELATION BETWEEN THE RAINFALL AND THE DISCHARGE OF SEWERS
IN POPULOUS DISTRICTS. Amer. Soc. Civ. Engin. Trans. 20: 1-56.
(35) LANGBEIN, W. B.
1940. CHANNEL STORAGE IN UNIT HYDROGRAPH STUDIES. Amer. Geophys. Union
Trans. 21: 620-627.
(36) LATEN, N. R., REILLY, T. L., and MINOTTE, J. S.
1940. DISTRIBUTION GRAPHS AND FLOOD ROUTING IN THE UPPER OHIO RIVER BASIN.
(37) LAURENSON, E. M.
1932.
(38)
1960.
HYDROGRAPH sYNTHETis BY RUNOFF ROUTING.

University of New South Wales, Sydney.
Water Res. Lab. Rpt. 66.

(40) LLOYD-DAVIES, D. E.
1906. THE ELIMINATION OF STORM WATER FROM SEWERAGE SYSTEMS.
Engin. Proc. 164: 41-67. London.
(41) MITCHELL, W. D.
1948. UNIT HYDROGRAPHS IN ILLINOIS.
111. State Div. Waterways.
(42) MORGAN, P. E., and JOHNSON, S. M.

1962. ANALYSIS OF SYNTHETIC UNIT-GRAPH METHODS.
Inst. Civ.
Springfield.
Amer. Soc. Civ. Engin.,
(43) MULVANEY, T. J.
1850. ON THE USE OF SELF-REGISTERING RAIN AND FLOOD GAUGES.
Engin. (Ireland) Trans. 4(2): 1-8. Dublin.
Inst. Civ.

(44) MuNRO, C. H.
1956.
BASIC THEORY OF URBAN STORM DRAINAGE DESIGN.
Inst. Engin. Austral.
Jour. 28: 301.

(45) NASH, J. E.
1958. DETERMINING RUNOFF FROM RAINFALL.
Inst. Civ. Engin. (Ireland) Proc,
10: 163-184.
(46)
1958.
THE FORM OF THE INSTANTANEOUS UNIT HYDROGRAPH. General Assembly

of Toronto. Internatl. Assoc. Sei. Hydrol. (Gentbrugge) Pub. 42, Compt.
Rend. 3: 114-118.
(47)
1959.
SYSTEMATIC DETERMINATION OF UNIT HYDROGRAPH PARAMETERS.
JoUr.
Geophys. Res. 64(1): 111-115.

(48)
1960.
A UNIT
HYDROGRAPH
CATCHMENTS.
STUDY
WITH
PARTICULAR REFERENCE
TO
BRITISH
Inst. Civ. Engin. Proc. 17: 249-282.
(49) O'KELLY, J. J.
1955.
THE EMPLOYMENT OF UNIT-HYDROGRAPHS TO DETERMINE THE FLOWS OF

IRISH ARTERIAL DRAINAGE CHANNELS. Inst. Civ. Engin. (Ireland) Proc.
4(3): 365-412.
(50) ORMSBY, M. T. M.
1932. RAINFALL AND RUNOFF CALCULATIONS. Inst. Munic. County Engin. Jour.
59: 978.
(51) PRASAD, RAMANAND.
1967. A NONLINEAR HYDROLOGIC SYSTEM RESPONSE MODEL. Amer. SoC. Civ.
(52) RED, JOHN.
1926.
THE ESTIMATION OF STORM-WATER DISCHARGE.

Jour. 53: 997.
Inst. Munic. County Engin.
(53) RiLEY, D. W.
1931.
NOTES ON CALCULATING THE FLOW OF SURFACE WATER IN SEWERS.
lust.
Munic. County Engin. Jour. 58: 1483.

(54) Ross, C. N.
1921.
THE CALCULATION OF FLOOD DISCHARGES BY THE USE OF A TIME CONTOUR
PLAN. Inst. Engin. Austral. Trans. 2: 85.

(55) RouscuLP, J. A.
1927.
RAINFALL-RUNOFF ANALYSIS IN STORM-SEWER DESIGN.
Engin. NeWS-ReC.
98: 270.
(56) SATO, S., and MIKAWA, H.
1956. A METHOD OF ESTIMATING RUNOFF FROM RAINFALL. Proceedings of Regional
Technical Conference on Water Resources Development in Asia and the
Far East. U.N. Flood Control Ser. 9: 152-155.
(57) SHERMAN, L. K.
1932. STREAM FLOW FROM RAINFALL BY THE UNIT GRAPH METHOD. Engin. NewsRec. 108: 501-505.
(58)
1932.
THE RELATION OF HYDROGRAPHS OF RUNOFF TO SIZE AND CHARACTER OF
DRAINAGE BASINS.

(59) SNYDER, F. F.
1938. SYNTHETIC UNIT GRAPHS. Amer. Geophys. Union Trans. 19(1): 447-454.
231
(60) STRAHLER, A. N.
1964.
QUANTITATIVE GEOMORPHOLOGY OF DRAINAGE BASINS AND CHANNEL NETWORK. Sec. 4-II: 38 pp., In Chow, Ven Te, ed.. Handbook of Applied
Hydrology. New York.
(61) SUGAWARA, M.
1961. ON THE ANALYSIS OF RUNOFF STRUCTURE ABOUT SEVERAL JAPANESE RIVERS.
Jap. Jour. Geophys. 2(4): 1-76.
(62)
and MARUYAMA, F.
1957. A METHOD OF PREVISION OF THE RIVER DISCHARGE BY MEANS OF A RAINFALL
MODEL. Symposia Darcy, Internatl. Assoc. Sei. Hydrol. Pub. 42, 3: 71-76.
Dijon.
(63) TAYLOR, A. B., and SCHWARTZ, H. E.
1952. UNIT-HYDROGRAPH LAG AND PEAK FLOW RELATED TO BASIN CHARACTERISTICS. Amer. Geophys. Union Trans. 33(2): 235-246.
(64) TENNESSEE VALLEY AUTHORITY.
1964.
BRADSHAW CREEK-ELK RIVER: A PILOT STUDY IN AREA-STREAM FACTOR
Off. Tributary Area Devlpmt. Res. Paper 4. Knoxville.

(65) TURNER, H. M., and BURDOIN, A. J.
1941. THE FLOOD HYDROGRAPH. Boston Soc. Civ. Engin. Jour. 28(3): 232-256.
(66) U.S. BUREAU OF RECLAMATION.
1949. FLOOD ROUTING, CH. 6.10. In Flood Hydrology. Part 6 of water studies,
V. 4. U.S. Bur. Reclam. Manual. (Revised, 1951.)
(67) U.S. CORPS OF ENGINEERS.
1963. UNIT HYDROGRAPHS. PART 1: PRINCIPLES AND DETERMINATIONS. U.S.
Engineer Office, Baltimore.
(68) U.S. SOIL CONSERVATION SERVICE.
1964. HYDROLOGY, SECTION LV. SCS Nati. Engin. Handbook Washington D.C.
[This is a revision for use in watershed planning.]
(69) WILLIAMS, H. M.
1945. Discussion, MILITARY AIRFIELDS: DESIGN OF DRAINAGE FACILITIES, by G. A.
Hathaway. Amer. Soc. Civ. Engin. Trans. 110: 820-826.
(70) Wu, I-PAI.
1963. DESIGN HYDROGRAPHS FOR SMALL WATERSHEDS IN INDIANA. Amer. SoC. Civ.
(71) ZocH, R. T.
1934, 1936, 1937. ON THE RELATION BETWEEN RAINFALL AND STREAMFLOW. In
U.S. Department of Commerce, Monthly Weather Review. Part I, v. 62:
315-322; part II, v. 64: 105-121; part III, v. 65: 135-147.
CORRELATION.
LECTURE 9:
MATHEMATICAL SIMULATION OF SURFACE FLOW
As a result of the development of conceptual models in unit hydrograph
theory, there has been a tendency since the mid-1960's to propose the use of
conceptual models to represent specific elements of the hydrologie cycle other
than the overall direct response of a catchment. The principles of mathematical
physics can be applied to the investigation of various parts of the hydrologie
cycle, and the individual processes can be represented by a set of equations
and boundary conditions. To solve these equations, it is necessary to make
further simplifying assumptions, which accentuate the simulation nature of
the mathematical solutions obtained. The replacement of these simplified
mathematical expressions by conceptual models is in accordance with the
general systems approach, which considers each system in terms of a certain
number of interconnected elements and judges a system by its overall operation rather than the precise nature of the elements themselves. Conceptual
models are formulated on the basis of a simple arrangement of a relatively
small number of elements, each of which is itself simple in operation. The most
widely used conceptual elements are linear reservoirs and linear channels.
Though conceptual models were originally introduced as highly simplified
versions of the actual physical operations involved, they can also be looked
upon as mathematical abstractions whose only function is to simulate the
behavior of the physical systems being studied.
In the two final lectures, we discuss four segments of the hydrologie cycle
that to some extent lend themselves to mathematical simulation and to the
synthesis of conceptual models. The four segments involved are overland flow,
open channel flow, unsaturated flow in soils, and ground water flow.
Overland Flow
Overland flow is an interesting example of the appUcation of mathematical
simulation and the possibility of applying conceptual models to the solution
of a hydrologie problem. Overland flow has been studied analytically, in the
laboratory, and in the field. It occurs early in the runoff cycle, and the inherent nonlinearity of the process is not dampened in any wây. Hence, the
methods of linear analysis and synthesis are inadequate in this case, and the
general approach used in developing linear methods must be extended.
A physical picture of overland flow is shown in figure 9-1. The essential
problem to be solved is to determine the flow off the plane at the downstream
end for given physical conditions and a given pattern of lateral inflow along
the plane. The equation of continuity for the two-dimensional lateral inflow
232
233
LATERAL INFLOW, r(x,t)
I I i i I I j i i 1 i i 1 I i
_Upstream boundary
and dividing point
Downstream boundary
->-X
FIGURE
9-1.Overland flow (two-dimensional).
problem can be written as :

dx
(1)
dt
where
q = q{Xjt) =rate of overland flow per unit width
y = y {x,t) = depth of overland flow
and
r = r(Xjt) =rate of lateral inflow per unit area.
The dynamic equation for two-dimensional overland flow can be written as :
1 du
gdt
dy
dx
udu
gdx
q
gy^
where
u = u{Xyt) = velocity of overland flow
So = slope of plane
Sf = friction slope.
Though the continuity equation is linear in q and y, the dynamic equation is
highly nonlinear. It is possible by means of a high-speed digital computer to
obtain a numerical solution of equations 1 and 2 for any given set of boundary
conditions. This approach will be considered briefly later in this section. For
the present, however, we are concerned with the simpler approaches to the
particular problem, that is, with the attempt to find a simple mathematical
simulation or a simple conceptual model.

The classical problem of overland flow is to solve the particular case where
the lateral inflow is uniform along the plane and takes the form of a unit step
function :
r(x,t) = U(t)
(3)
There are several parts to the complete solution of this problem. Firstly, there
is the steady-state problem of determining the equilibrium profile when the
outflow at the downstream end of the plane is equal to the inflow over the
surface of the plane. Secondly, there is the problem of determining the rising
hydrograph of outflow before equilibrium for the special inflow case represented
by equation 3 above. If the problem were a linear one, the solution of this
second problem (that is, the determination of the step-function response)
would be sufficient to characterize the response of the system, and the outflow
hydrograph for any other inflow pattern could be derived from it. However,
since the problem is inherently nonlinear, the principle of superposition cannot
be applied, and each case of inflow must be treated on its merits. The third
basic problem is that of determining the recession from the equilibrium condition after the cessation of long continued inflow. The nature of the recession
when the inflow ceases before equilibrium is reached (that is, before the outflow builds up to a value equal to the inflow) must be investigated, and this
constitutes a fourth basic problem. The next step is to investigate the effect
of an inflow formed by the superposition of two or more step functions. Thus,
the fifth basic problem involves consideration of the case where there is a
sudden increase from one uniform rate of inflow to a second higher rate of
uniform inflow. The sixth case considered is that when a uniform rate of
inflow is suddenly changed to a second uniform rate of inflow which is smaller
than the first.
A few of the classical experimental results by Izzard (28) are shown on
figure 9-2. The top figure shows a rising hydrograph, a recession, a second
rising hydrograph, and a final recession. The second figure shows the effect
of changing the inflow rate from 1.83 to 3.55 in. per hr. (4.65 and 9.02 cm. per
hr., respectively) and the lower diagram shows the effect of changing the
inflow rate from 3.65 to 1.84 in. per hr. (9.27 and 4.67 cm. per hr., respectively) . Also, shown in the figure is a logarithmic plot of the detention storage
on the surface of the plane against the discharge at the downstream end.
The first approach to the solution of the overland flow problem in classical
hydrology to be considered is that based on the replacement of the dynamic
equation 2 by an assumed relationship between outflow and storage. Because
this method wâs first proposed by Horton (22) for overland flow on natural
catchments and subsequently used by Izzard (28) for paved surfaces, it may
be referred to as the Horton-Izzard approach. Hydrologists noted that when
the equilibrium runoff (that is, the equilibrium discharge at the downstream
end) of a number of experimental plots was plotted against the average surface
detention (or total surface detention) at equilibrium on log-log paper, the
235
^WW
N
en
/
O
z
<
\\
/
^*<'T^"?
v\\\\
RUN No. I3
Lo = 72 feet
.S =0.01
J
i = 1.83 and 3.55
;/
UJ
\
\
inches oer hour
/
/
\
\ \V
N
14 16 18
MINUTES
2z
<
20 22 24 26 28 30
32
14 16 18 20 22 24 26 28 30 32
MINUTES
FIGURE
9-2.Hydrograph of overland flow.

experimental points fell approximately along a straight line. An exact linear
relationship on logarithmic paper would indicate that the equilibrium outflow
at the downstream end and the equilibrium storage were connected as follows :
q{L,te)=qe = aSe'
(4)
where e was the discharge at the downstream end of the plane under equilibrium conditions, Se was the total surface storage at equilibrium conditions,
and a and c were parameters.
In the Horton-Izzard approach to the overland flow problem, the assumption is made that such a power relationship holds not only at equilibrium, but
also at any time during the rising hydrograph or during the recession. Using
this assumption we can write :
q{L,t)=qL = aS'^
(5)
where qi is the discharge at the downstream end at any time and >S is the
corresponding total storage on the surface of the plane of overland flow. The
equation of continuity, equation 2, can be written in the hydrological form as:
r'L-qL = -7at
(6)
which for our assumptions can be written as :

q,-aS' =
at
(7a)
or
<^t = j^^
(7b)
f d{S/Se)
,_ ,
The solution of equation 7 is :

^
i=
diq/qeY''
{ay'^{qeY^-'^f^ J l-{q/qe)
^JW^
Equation 8 can be solved analytically for values of c == 1 (linear), c = 2, c = 3, or

c = 4, and also for ratios of these values, that is, for c = ^^ or ^^.
Horton (22) solved the equation of the rising hydrograph due to a step
function input for the case of c = 2, which he described as ^^mixed flow'' since
the value c is intermediate between the value of ^^ for turbulent flow and the
237
value of 3 for laminar flow. Horton's solution may be written as :

- = tanhM-^)
(9a)
where
Ke = -
(9b)
Qe
Since the system is nonlinear, the time parameter Kg will depend on the
intensity of inflow. Horton gave an empirical expression for the equilibrium
storage per unit width and his equation for the rising hydrograph has been
used in the design of airport drainage since that time. Izzard (27) presented
the solution for the case of c = 3 (that is, for laminar flow) in the form of a
dimensionless rising hydrograph. Izzard used as his time parameter a time to
virtual equilibrium, which is exactly twice the time parameter used in equation
9b above. It is of interest that the integral in equation 8 is of the same form as
the Bakhmeteff varied flow function and, hence, tabulated values of the
varied flow function may be used to tabulate or draw the rising hydrograph
for any value of c for which it is tabulated. Typical rising hydrographs are
shown on figure 9-3.
For recession from equilibrium, the recharge in equation 6 becomes zero,
and the insertion of the value for q from equation 5 leads to the solution:
(r)
= l + (c-l)
(10)
where q is the ordinate of the recession curve and t is the time elapsed since the
cessation of inflow, that is, the time since the start of recession. Typical
recession curves as predicted by the Horton-Izzard model are shown in figure
9-3. The special case of equation 10 for the value of c = 3 was given by Izzard.
If the duration of inflow (D) is less than the time required to reach virtual
equilibrium, we get a partial recession from the value of the outflow (qn)
which has been reached at the end of the inflow. It can be easily shown that
this curve is the same shape as for recession from equilibrium except that the
recession flow enters the curve defined by equation 10 at the appropriate
value of qe/qD'
If there is a change to a new rate of uniform inflow during the rising hydrograph, two cases can occur. If the new rate of inflow is higher than the rate
of outflow when the change occurs, the same dimensionless rising hydrograph
can still be used, but since qe is equal to the inflow^ at equilibrium, the value of
q/qe will change as soon as the rate of inflow changes. If the new^ rate of inflow
is less than the outflow when the change occurs, the hydrograph will correspond
to the falling curve of the varied flow function. The latter function can be used
to determine the shape of such a falling hydrograph, w^hich will be of the type
shown in the bottom of figure 9-2.

1.0
q/qe
q/qe
t/ke
FIGURE
9-3.Shapes of rising and recession hydrographs.
Looked at as a conceptual model, the Horton-Izzard solution clearly assumes

that the whole system can be lumped together and treated as a single nonlinear
reservoir whose outflow-storage relationship is given by equation 5. Even
though this conceptual model is extremely simple in form, the fact that it is
nonlinear makes it less easy to handle than some of the apparently complex
conceptual models used to simulate linear or linearized systems. Thus, the
impulse response for such a system no longer characterizes the system because
the output will also depend on the form and intensity of the input. The cumulants of the impulse can no longer be added to the cumulants of the input to
obtain the cumulants of the output. The solution for a step function input
cannot be used to obtain the output for a complex pattern of input.
239
The second simple solution proposed for the overland flow problem is the
kinematic wave solution. It also involves a power relationship between discharge and depth but, in this case, not a lumped relationship covering the
whole system, but a relationship between the discharge and the depth at each
point and, therefore, a distributed relationship. The basic assumption for the
kinematic wave solution is that all the terms of dynamic equation 2 are
negligible compared with the slope term and the friction term, so that we
have:
So-Sf = 0
(11a)
which can also be written as :
q{x,t)=q = hyo
(lib)
If friction is taken according to the Chezy formula, the value c will be %;

whereas, if it is taken according to the Manning formula, the value of c will
be î. For zero initial conditions and an equilibrium discharge e (equal to the
product of the constant supply rate r and the length of overland flow L), we
have the following solution for the rising hydrograph:
Lji)
t<k:
-= 7
(12a)
t>h:
-=1
(12b)
qe
\tk/
where
t -
- '^''^ -J'
fl2c)
In equation 12, h is the kinematic time parameter and ye is the depth of flow
obtained when the equilibrium discharge qe is substituted in equation lib.
The kinematic time parameter h should be distinguished from the time
parameter for the Horton-Izzard model Ke defined by equation 9b above and
from the time to equilibrium te used by Izzard. The rising hydrograph for the
kinematic wave solution is shown on the upper part of figure 9-4.
The recession from full equilibrium for the kinematic solution can be
shown to be :
(roc-r
-c-j
(13)
tk
which is also shown on the upper part of figure 9-4. Where the duration of
inflow (D) is less than the time of kinematic rise {i])^ the kinematic wave
solution gives a flat topped hydrograph, in which the flow is constant until it
meets the full recession curve as shown in the lower diagram on figure 9-4.
The kinematic wave solution has been applied to overland flow by Henderson
M,
/1
/ 1
/ 1
/
1
/
1
y^
1
q/q e
j<
M2
\
\
\
\
\
\.
\w
^\^^
Duration > t^
Duration < tj^

FIGURE
9-4.Kinematic wave solution.
and Wooding {21), and Wooding {60, 61, 62) and used to construct a model
of catchment response. They analyzed the problem and developed equations
for the rising hydrograph and falling hydrograph by arguments based on the
method of characteristics.
The numerical solution of the overland flow problem has been tackled by
Woolhiser and Liggett {63). They reduced the equations of continuity and
momentum to dimensionless forms by expressing the variables in terms of the
241
normal depth and velocity at the downstream end of the plane for the maximum discharge. When this is done (see reference 63 for details), equations
1 and 2 become :
^+^=1
du
1 dy
udu
(14)
^^ ,
\
y/
where
in which the superscripts denoting that the variables are dimensionless

variables have been omitted for convenience. There are only two parameters
in these equations, the Froude number for normal flow at maximum discharge
(F) and the parameter K defined by equation 15b, which reflects the length
and slope of the plane (or channel) as well as the normal flow variables.
Equations 14 and 15 were expressed in characteristic form and solved by a
finite difference technique. For high values of K, the slope and friction
dominated the flow and, as might be expected in these conditions, the solution
approximated the kinematic wave solution. For values of the parameter K
smaller than 10, the kinematic wave solution was found to be a poor approximation.
A typical rising hydrograph found by Woolhiser and Liggett (60) is shown
on figure 9-5. It would appear that in the early stages of the rising hydrograph,
the shape of the hydrograph approximates to the kinematic solution, whereas
in the later stages it approximates more to the Horton-Izzard solution. This
is not unexpected because in the early stages of the flow dq/dx would be
relatively small, thus approximating the kinematic solution for which dq/dx
is zero downstream of the characteristic which starts from the upstream end
of the plane at the start of inflow. In the later stages of the rising hydrograph,
the value of dq/dx would approach the rate of lateral inflow and the HortonIzzard solution, based on an empirical relationship which is a good approximation at equilibrium, might be expected to give better predictions than the
kinematic model.
In simulating overland flow, either as a hydrologie system or as a subsystem
of a watershed, consideration should be given to the type of flow involved.
The Stanford model incorporates a rising hydrograph for overland flow developed by Morgali and Linsley {4^). Their hydrograph is for a high value of
K and hence approximates very closely to the kinematic solution. For lower
values of K, this rising hydrograph would not necessarily be a good representation of overland flow.
1.01
1IIIII
0.2
0.4
FIGURE
0.6
II
0.8
t/t
!_.
1.0
1.2
1.4
1.6
9-5.Comparison of rising hydrograph.
Dooge (P, 10) recently proposed as a conceptual model for problems with
lateral inflow a cascade of equal reservoirs, either linear or nonlinear, with
intermediate inflow. For overland flow, these reservoirs would be nonlinear.
This conceptual model is what the author has referred to as uniform nonlinearity. In such cases it can be shown that the outflow hydrograph for
uniform inflow can be represented in dimensionless form bj^:
Qe
\to to /
(16a)
where q is the outflow, e the equilibrium outflow, t the time, to a characteristic

time which depends on the intensity of inflow, and D the duration of uniform
inflow. For a step function input, there is no duration to affect the issue and
the equation of the rising hydrograph can be written as :
Qe
\to/
(16b)
It can be shown that the outflow from a cascade of equal nonlinear reservoirs
is of the form indicated by equation 16b. Dooge (10) has shown that the
cumulative outflows measured by Amorocho and Orlob (2) for pulse inputs to
a laboratory catchment (which was nonlinear in behavior) can be plotted as a
single line when a characteristic time based on the intensity of inflow is used
243
for dimensionless plotting. In the same paper, Dooge showed that the wide
variations in the unit hydrographs derived by Minshall (^7) can be enormously reduced by the same type of plotting.
A comparison of equations 8 and 9 with equation 16b indicates that the
Horton-Izzard model belongs to the class of uniformly nonlinear models with
Ke as the characteristic time. Similarly, a comparison of equations 12 and 13
with equation 16 indicates that the kinematic model also belongs to this class
with the h as the characteristic time. As already pointed out, the HortonIzzard solution represents the special case of one nonlinear reservoir. The
kxiematic wave solution for the linear case, can be approximated by a cascade
of linear reservoirs in which the product of the number of reservoirs and the
individual storage delay time remains finite as the number of reservoirs tends
to infinity. From these considerations, it is plausible to suggest that it might
be possible to simulate satisfactorily the hydrographs generated by Woolhiser
and Liggettw^hich are intermediate between the kinematic solution and the
Horton-Izzard solution^by a cascade consisting of a finite number of equal
nonlinear storage elements.
Unsteady Flow in Open Channels

The problem of predicting the discharge hydrograph at a downstream point
on the basis of the hydraulic properties of the channel and a known discharge
at an upstream point is a classical problem in hydrology. The various methods
proposed for its solution can be reviewed in the recent bibliography by
Yevjevich {6Ji). The equation of continuity for unsteady flow in open channels
without lateral inflow is given by :
dQ dA
-^+ = 0
dx
dt
,
,
(17a)
where Q is the discharge and A the area of flow. The above is the form in
which the continuity equation at a section is written in open channel hydraulics. Hydrologists more frequently write the continuity equation in the
lumped form obtained by integrating equation 17a over a channel reach, thus
obtaining the hydrologie equation :
where I is the inflow to the reach; 0, the outflow from the reach; and S, the
storage in the reach at a given time. In open channel hydraulics, the dynamic
equation is written as :
dx
gdx
g dt
The corresponding equation in hydrology is the equation for the looped
244
rating curve:
Q = CAjR[So-f--;)
^ \
dx g dx g dt /
(18b
where R is the hydrauhc mean radius and C is the Chezy friction factor, which
may be evaluated from a friction formula such as the ]Manning equation.
Equations 17a and 18a reflect the hydraulic approach to the problem of unsteady flow in an open channel, while equations 17b and 18b reflect the
hydrologie approach to the same problem. These two separate approaches
have developed independently of one another. A systematic approach to the
problem, however, enables us to reconcile the two.
Various methods have been used for the solution of the hydraulic formulation of the problem of routing a flood down an open channel. Mathematical
methods can be used to find solutions for simplified versions of equations
17a and 18a.
If we wish to go beyond these idealized mathematical formulations, it is
necessary to use numerical methods. The recasting of the equations in terms
of characteristic variables facilitates such numerical solution. Even before the
advent of high-speed digital computers, numerical solutions were obtained in
this way. The advent of the computer, however, has greatly facilitated the
numerical solution of the problem. The method of characteristics is still used
in some numerical approaches to the problem; in others, either an explicit or
an implicit finite difference scheme using a rectangular network is used. In
explicit schemes, serious problems of stability may arise, whereas in implicit
schemes the storage capacity required to solve the resulting simultaneous
equations is a limiting factor.
In the hydrologie approach to the solution of the routing problem, the continuity equation 17b is retained and the dynamic equation 18b replaced by
some simplifying relation. The methods most commonly used in applied
hydrology for flood routing are the ]\Iuskingum method of ^NlcCarthy,^ the
lag and route method of Meyer (4^), the diffusion analogy of Hayami (18),
and the successive routing method of Kalinin and Milyukov {30). These are
all linear methods, and thus, in practice, the channel reach is assumed to be
linear with constant parameter values or else is linearized and a relationship
found between the parameter values and the level of either inflow or outflow.
Since we are interested in a solution at one particular downstream location,
we do not need to know conditions at all intermediate points. A systems
approach would therefore seem to be more appropriate than a complete
1 MCCARTHY, G. T. THE UNIT HYDROGRAPH AND FLOOD ROUTING. Unpublished paper

presented at the Conference of North Atlantic Div., U.S. Corps of Engineers, Providence,
R.I. 1939.
245
numerical solution, which generates unwanted solutions at intermediate

points. However, our present systems techniques in hydrology are such that if
we wish to use a systems approach we must confine ourselves to the linearized
version of the problem. Recent studies have been made involving the complete
linearized solution to the equations of continuity and momentum for twodimensional flow in a uniform channel, (jf^ 16). This approach, in fact,
applies to the flood routing problem the same assumptions that are made in
unit hydrograph theory for the more complex problem of catchment response.
It is remarkable that in the past 25 years, during which the unit hydrograph
approach has been widely used, no corresponding attempt has been made to
treat a channel as a linear system.
If we confine ourselves to the case of a semi-infinite uniform wide rectangular
chnanel, without lateral inflow, for which the friction effect can be represented
by the Chezy formula, we can write equations 17a and 18a as:
dx
dt
evueuieu^^_^
dx gdx g dt
CV
(20)
The boundary conditions to be satisfied are the initial conditions determined

by an initial uniform flow throughout the length of the channel and an upstream boundary condition determined by the inflow hydrograph at the
upstream end. Though the equation of continuity (equation 19) is linear in
q and y, the dynamic equation (equation 20) is highly nonlinear.
If we consider a small perturbation about the steady discharge qo, then wê
can write the following equation for the perturbation of discharge (g) from
this reference value qo:
(^2/o-.o^) ^.-2^0-- = 3,o-+2,>So-
(21)
in which the coefficients have been frozen at the values corresponding to the
reference discharge (go). The above linearization was proposed by Deymie
(7) who also derived the solution given in equation 29. The work of Deymie
and of Masse {45) published in 1938 was not followed up, and the linearization
given in equation 21 and the solution given in equation 29 were developed
independently by Dooge^ in 1965.
Strictly speaking, equation 21 is only valid for perturbations small enough
that the variation coefficients in the nonlinear equation is not sufficient to
2 DooGE, J. C. I. LINEAR ANALYSIS OF FLOW IN OPEN CHANNELS.
randum, Univ. Col., Cork, Ireland. 1965.
Unpublished memo-

affect the result. If we follow the unit hydrograph approach and ignore the
fact that large perturbations give rise to nonlinear behavior, we can apply
equation 21 to large perturbations and accept the solution of this linearized
equation as an approximation to the solution of the original nonlinear problem.
How good an approximation it will be can only be determined for a given case
by comparing this complete Unear solution with the complete nonlinear
solution. The fact that linear routing methods have been used in applied
hydrology would indicate that the effect of linearization cannot be so catastrophic as to make linear methods worthless. The complete solution of the
hnearized hydrauhc equation 21 has the advantage that it can be used as a
standard against which to measure the simple linear models used in applied
hydrology. Indeed, the latter can be considered as attempts to simulate the
complete linear solution.
Since equation 21 is linear, it is only necessary to determine the solution for
a delta function input. For any other inflow, it is only necessary to convolute
the impulse response with the actual inflow. For convenience, the impulse
response of a channel obtained from equation 21 will be referred to as the
linear channel response (LCR).
If the original independent variables (x,t) are replaced by the characteristic
directions (m and n) and the dependent variable (g) is replaced by a new
transformed dependent variable {z), the equation can be written in the more
compact form :
ah
dh
-hh = 0
dmdn
(22a)
where
g = 0exp(--rt-\-sx)
m=t
(22b)
(22c)
Ci
n=t
(22d)
C2
where Ci and C2 are the characteristic wave velocities and r and s are parameters
defined in terms of the channel parameters (11). Though equation 22a is more
compact in form than equation 21, it is no easier to solve since the simpler
form of the equation is counterbalanced by the fact that the boundary conditions are not as convenient when expressed in terms of m and n as the}^ are
when expressed in terms of x and t.
Any of the standard mathematical techniques can be used for the solution
of either form of the equation, but it is probably more convenient in each case
to use Laplace transform methods. When this is done in terms of x and t,
247
the Laplace transform of the impulse response or LCR is found to be^ as

follows :
H{s) =expl-x\/as^+bs+c+exs+fx']
(23)
where a, h, c, e, and/are parameters depending on the hydraulic characteristics
of the channel.
Since the Laplace transform of the LCR is of exponential form, the cumulants can be determined by repeated differentiation of the quantity inside the
square brackets in equation 23 and evaluated at s = 0. This process is complicated by the continual occurrence of indeterminate forms which have to be
evaluated by L'HopitaPs Rule. When this is done and the values of the
parameters a, h, c, e, and/ are substituted, it is possible to write the cumulants
as follows :
ki=W = -^
(24a)
--=3('-?)(li)(riy
The result for the lag given by equation 24a indicates that for the linearized
solution, the average rate of propagation of the flood wave is 1.5 times the
velocity corresponding to the reference discharge. This corresponds to the
value indicated by the Kleitz-Seddon Law (32, 55) for the celerity of a flood
wave in a wide rectangular channel with Chezy friction :
dO do
c = = ^ = M2/^''^ = 1.5u
(25)
The higher cumulants can be made dimensionless by dividing by the appropriate power of the lag.
It can be readily seen from equation 24 that the resulting dimensionless
cumulants or shape factors are functions of the Froude Number and the
^ DOOGE, J. C. I. LINEAR THEORY OF OPEN CHANNEL FLOW: I-COMPLETE LINEAR SOLUTION

OF ROUTING PROBLEM. Unpublished memorandum, Univ. Col., Cork, Ireland. 1967.

dimensionless length parameter (Sox/y) of the following form:
Sn = MF) (D)^-^ for = 2, 3
(26a)
where
D=^
(26b)
2/0
Consequently, even if wê were unable to invert the transformed function given
by equation 23, it would still be possible to determine the cumulants of the
solution and to plot the solution for any given value of F on a shape factor
diagram.
The inversion of equation 23 gives a solution in the original {x,t) coordinates
consisting of two terms :
q{x,t)=qi+q2
(27)
where qi represents the head of the wave and g2, the body of the wave. The
term representing the head of the wave is of the following form :
,.=5(^-^3,|exp( px)
(28a)
where
2-^ So
, ,
It can be seen that the head of the wâve moves downstream at the dynamic
wave speed Ci in the form of a delta function of exponentially declining volume.
The body of the wâve has the form :
2 =
--h(-)exp{-rt+sx)
\ci
C2/
Ult--)
\
Ci/
(29a)
where
/i[2/ia] = modified Bessel function
(29b)
-V('-~)('-:i)
(2)
U[t'] = unit step function
(29d)
and
ci and C2 are the dynamic wave velocities, and r, s, and h are parameters
depending on the hydraulic properties of the channel (11).
The shape of the body of the wave for F = 0.5 and various values of the
dimensionless length factor D are shown on figure 9-6. For short lengths, the
impulse response declines monotonically; for intermediate lengths, the impulse
response is a unimodal curve with an appreciable initial ordinate. For long
channels, the unimodal shape of response rises from an initial ordinate which
249
is practically zero and declines again to zero. For other values of the Froude
number (F), the same three shapes are obtained, though the values of the
dimensionless length parameters at which a change in shape occurs increases
with the Froude number.
Figure 9-6 is plotted in dimensionless termsqUiY versus ilUand hence,
gives the erroneous impression that the peak is increasing as the flood wave
moves downstream. This is due to the fact that the time of travel in a reach
(o) increases with distance. The variation of the downstream discharge with
length of channel is shown in real terms on figure 9-7. This shows the result of
-1
1
r
SHORT CHANNEL
D=I.O
0.8
I 0.6
I 0.4
o.2
T
INTERMEDIATE CHANNEL
D = 5.0
FIGURE
9-6.Shape of impulse response.
computations for a channel with the steady state rating curve :

Co = 502/0'^
(30)
and for an inflow given by:

7(0 =125-75 cos (^j
0 <96
(31)
which corresponds to the inflow used by Thomas (57) in his classical paper
on unsteady flow in open channels. The figure shows the modification of the
flood wave for distances up to 500 miles (805 km.).
For any linearization of the routing problem, it is necessary to choose a
value of the reference discharge (go) about which the discharge is perturbed.
Since this value of go is used to evaluate 2/0 from equation 30 and hence uo and
the coefficients in equation 21, it will naturally affect the result. The effect of
the choice of reference discharge on the outflow at 50 miles (80.5 km.) for an
inflow given by equation 31 is shown on figure 9-8.
The inflow varies from 50 cubic feet per second per foot (4.65 mVsec./m.)
width to 200 c.f.s. per foot (18.6 mVsec./m.) width and the reference discharge is taken at values of 50 (4.65 m^), 100 (9.3 m^), 150 (13.9 m^) and
200 (18.6 m^) c.f.s. per foot width. It can be seen from figure 9-8 that for
the smaller values of the reference discharge, the flood wave is displaced in
time and occurs later as would be expected from equation 24a. It is interesting
T'I'Ifr
-1IIIr
00 mites
BASFLOW
20
_.
I
I
24
I
48
I
72
FIGURE
I J
I
i
I
I
I
96
120
144
168
TIME (HOURS)
I
1
1
192
216
9-7.Variation of outflow with distance.
1
L240
264
251
200
24
48
FIGURE
72
96
TIME (HOURS)
120
144
9-8.Effect of reference discharge.
to note, however, that the shape of the flood wave for the various reference
discharges is very similar.
For a channel whose rating curve is given by equation 30, that is, one with
Chezy friction, the Froude number is independent of the depth of now and
hence, the value of the second cumulant given by equation 24b is independent
of the reference discharge. Since the second cumulant of the outflow is equal
to the second cumulant of the inflow plus the second cumulant of the LCR,
the second cumulant of the outflow will, for a case of Chezy friction, be
independent of the reference discharge chosen. The reference discharge will
affect the third and fourth cumulants, but these may be small compared to
the third and fourth cumulants of the inflow. In any case, the third and fourth
cumulants do not have as marked an effect on the shape as the first and second
cumulants.
Having obtained the complete linear solution of the hydraulic equations,
it is now possible to compare the various special linear hydrologie solutions
with it and determine the accuracy with which they can simulate the complete
solution and the range within which they apply. This was done by the method
of moment matching, which is most convenient in this connection. The

results were checked by the method of least squares. Nine linear models were
studied and may be grouped as shown below:
One-parameter models:
dynamic wave equation
kinematic wave equation
Two-parameter models:
diffusion analogy
Muskingum method
lag and route method
Kalinin-Milyukov method
Three-parameter models :
diffusion plus lag
multiple Muskingum method
three-parameter gamma distribution.
Of most interest are the two-parameter models which have been used as
practical channel routing methods in applied hydrology.
The complete linear solution is a three-parameter system. If expressed in
dimensionless form, the dimensionless discharge can be formulated as a function of a dimensionless time parameter, a dimensionless length parameter,
and the Froude number. It may appear pointless to attempt to simulate the
three-parameter complete linear solution by another three-parameter system
which, at best, will be an approximation to it. However, the complete Hnear
solution is complex in form and relatively difficult to compute; if it can be
approximated with a sufficient degree of accuracy by another three-parameter
system which is easier to comprehend and easier to compute, then the simulation may be more convenient than the use of the original mathematical
solution.
A one-parameter simulation will plot as single point on a shape factor diagram. Hence, it can hardly be expected to simulate a three-parameter system
which plots as a family of lines. Nevertheless, its ability to simulate flood
routing may be tested by comparing the first moment of the one-parameter
model with the first moment of the complete linear solution given by equation
24a. If the two terms on the right-hand side of equation 20 are neglected, that
is, the difference between bed slope and friction slope is assumed to be negfigible
compared to the other terms, then we obtain the classical linear wave solution.
For a delta function input at the upstream end, this solution is a delta function
traveling down the channel at a velocity equal to the dynamic wave speed
{ci = uo+ s/gyo). The problem is only properly posed for Froude numbers less
than one. For such cases, the dynamic wave speed is greater than 1.5 UQ, which
is the average speed of translation as given by equation 24a.
Alternatively, if all the terms on the left-hand side of equation 20or,
what is the same thing, the terms on the left-hand side of equation 21are
neglected, then we get the linear kinematic wave solution. This is equivalent
to assuming that the dynamic equation may be used in the simplified form
253
appropriate to steady uniform flow, that is, that the effects due to changing
depth and velocity are neghgible compared to the effects of slope and friction.
In this case, the solution is also a translation without distortion, but this time
at the speed 1.5io so that the linear kinematic wave solution is a one-parameter
model which has exactly the same lag as the complete solution.
Most of the flood routing methods used in applied hydrology are twoparameter models. If the second and third terms on the left-hand side of
equation 21 are expressed in terms of the second derivative with respect to
distance on the basis of the linear kinematic wave solution (which is a first
approximation to the solution), then the equation becomes:
3,*^+^^
dx
(32)
Uo ot
which is a parabolic equation in contrast to the original equation 21 which

was a hyperbolic equation.
The parabolic solution (or diffusion analogy, or convective-diffusion
solution) obtained from equation 32 may be shown to be identical to the complete solution for the special case of the Froude number equal to zero and
may also be shown to have the same first and second moments as the complete
solution for any value of F. While it is preferable to think of this solution as a
parabolic approximation to the complete solution, equation 32 may be considered as a convective-diffusion equation in which the '^convective velocity''
is given by :
a = 1.5io
(33a)
and the ^'hydrauhc diffusivity" is given by:
-('-T)
Hydraulic diffusivity must not be taken to mean that the physical process
involved is one of diffusion. For the parabohc solution (or diffusion analogy),
the linear channel response is given by:
The cumulants for this response can be determined from the general equation
for the R^^ cumulant which is:
fc={(l)(3)(5)
(2-3))( )
\ax /
(-1
\a/
(35)
Substitution of the value a from equation 33a and the value of D from
equation 33b in equation 35 gives expressions for the cumulants in terms comparable to those used in equation 24 on page 373. When this is done, it is seen

that the cumulants given by equation 35 are the same as those indicated by
equation 24 for the special case o F = 0.
The other special models used in applied hydrology can also be compared
to the complete linear solution. The Aluskingum method of flood routing is
based on the assumption that in a reach :
= K[X7+(1-Z)0]
(36a)
which can be combined with the continuity equation to give :

0+K(l-X)^=I-KX~
at
at
(36b)
The linear system represented by the above equation can be shown to have
the impulse response :
The delta function term in equation 37 indicates the possibility of the

occurrence of negative ordinates in the outflow unless the inflow is such as to
enable the contribution of the first term to the convoluted outflow to counteract the effect of the second term. The cumulants of the Muskingum solution
can be shown to be :
ki=Ui' = K
(38a)
k2=U2={l-2X)K^
(38b)
kz=Us' = 2{l-3X+3X')K'
(38c)
/c4=i/4-3(C/2)2 = 6(l-4X+6Z2-4X3)X4
(38d)
The parameters of the Muskingum model can be optimized by equating the

first and second cumulants given above to the first and second cumulants of
the complete linear solution. This results in the values:
^=L5;
(^^^)
and
-hK-f)()
In a uniform channel, we can determine the optimum values of the parameters for the Muskingum method by using equation 39 provided we know the
optimum reference discharge and the properties of the channel. For nonuniform channels, the first and second moments of the impulse response can
255
be got by subtracting the moments of the inflow from the corresponding

moments of the outflow: the value of K is equal to the first moment and X
can be obtained from equation 38b once K is known. This would seem to be a
more objective procedure than the attempt to transform a looped storage
curve to a straight line by taking trial values of X. It will be noted from
equation 39b that for certain short distances the value of X will be negative.
From the point of view of classical hydrology which views X as a measure of
the amount of wedge storage present, this appears physically unreasonable.
From the point of view of mathematical matching, the negative value of X
is the correct value to use.
The lag and route method {Jf6) assumes that the storage at any time may
be taken as proportional to the outflow which occurs after the elapse of a time
lag (r) so that we can write:
S{t)=K'0{t+T)
(40a)
which can be combined with the continuity equation to give:
O(0+i^-^Ô(+r)=7(0
(40b)
This model has the system response :

KT:
h(t)=0
(41a)
[-m]
(41b)
The cumulants of the lag and route model may be readily derived either
from its Laplace transform or by taking moments about the origin and using
them to find the cumulants. The values are:
R = l:
ki = K+T
(42a)
i^>l:
kR={R-l)\K^
(42b)
The values of K and r which are optimal in the moment matching sense can
be obtained by equating the first two moments of the response to the first
two moments for the complete linear solution. This results in the values :
-ri['-i('-?)(f;)]
As in the Muskingum model, one of the parameters may take on ''unrealistic''
values. This may happen since the value r given by equation 43b may be
negative for short lengths of channel. Again it must be emphasized that this
unrealistic parameter value gives the best fit according to the chosen criteria
and should be used if closeness of prediction is required. The element of
unreality lies in the choice of this particular model for short channel lengths
and the assumption that the crude hydrologie reasoning on which it is based
will result in the optimum parameters performing the same function as they
do in the crude model.
The use of successive routing through a characteristic reach was proposed
by Kalinin and Milyukov (30) in 1957. This is the same model as the cascade
model used to represent the unit hydrograph. It was proposed for channel
routing by Kalinin and Milyukov on the basis of a linearization of the unsteady flow equation. The impulse response function of the model is given by
the gamma distribution :
whose cumulants are given by :

KR = n(R-l)\K^
(45)
As has been pointed out in deaUng with conceptual models of the unit
hydrograph, though the conceptual model is based on the idea of a cascade in
which the value n would be integral, nonintegral values of n may be used to
fit the model to prototype data. The Kalinin-Milyukov model, like the other
models discussed in this section, can he used as a linearized model. The
parameters though taken as constamt m a given food event, or part of a
given flood event, can be varied with the intensity of inflow to allow for
nonlinear effects.
By matching the first and second moments given by equation 45 to the first
and second moments of the complete linear solution, the following optimal
values for the parameters K and n are obtained.
-K'-?)fe.)
The parameter K is the time-constant for a single linear reservoir of the
cascade. If the average rate of travel of the flood wave (which is 1.5?^) is
used to convert this characteristic time to a characteristic length we obtain :
257
which is identical with the formula for the characteristic length proposed by
Kalinin and Milyukov {30) except for the factor (l-F^/).
The shape factor diagram S3/-S2 for the complete linear solution and the
classical flood routing methods is shown on figure 9-9. The complete solution
plots as a family of parabolas, and the diffusion analogy coincides with the
curve for F = 0. From equation 26, it can be seen that the higher the dimensionless length (D), the lower will be the value of S2 and the other shape factors
and vice versa. Thus, we can deduce from figure 9-9 that for short lengths
(high S2), the various two-parameter models other than the diffusion analogy
would appear to be about equal in their ability to simulate the complete
linear solution. For long lengths (small S2), however, the Muskingum method
is seen to have a value of S3 approaching 0.5, whereas the complete linear
solution (for all Froude numbers) and the other models all have values of S3
approaching zero. We would deduce from this divergence that for long
lengths of channel, the Muskingum method would not simulate the outflow
hydrograph as well as the other methods. That this is so is shown by flgure
9-10, which gives the predicted outflow for the complete solution (for qo=150
c.f.s. or 4.25 ml per sec.) and the different models for the Thomas input
defined by equation 31 and a channel length of 500 miles (805 km.).
The parabolic method and the Kalinin-Milyukov (30) method predict discharges which are graphically indistinguishable from the complete linear
solution. The lag and route method predicts the travel time to a fair degree of
accuracy, but underestimates the degree of attenuation. The Muskingum
method is seen to predict negative ordinates for the first 60 hours and a peak
discharge which is about 20 percent too high and whose time-to-peak is about
50 percent too small. It can be verified that for the short channel lengths the
Muskingum method performs as satisfactorily as the other methods. The
complete failure of the Muskingum method for the case shown on figure 9-10
is due to the fact that the time-to-peak of the resulting hydrograph is greater
than the time of inflow, whereas the Muskingum outflow must decline as soon
as inflow stops. As a rule of thumb, this suggests that the Muskingum method
will fail if the lag of the channel reach is greater than about half the duration
of inflow.
The ability of the three-parameter models to simulate the complete linear
solution can be similarly analyzed.^ As might be expected, the three-parameter
models are better able to simulate the three-parameter complete solution.
Figure 9-11 shows a plotting on a S4 S2 shape factor diagram of the complete
linear solution for a Froude number of 0.5 and the lines for each of the threeparameter models for the same Froude number. The closeness of the lines on
4 HARLEY, B. M. LINEAR THEORY OF FLOOD ROUTING.

Univ. Ireland. 1967.
M.
Engin. Sei. Thesis. Nati.

the shape factor diagram suggests that the actual hydrographs would be very
similar. In fact, it is not possible to distinguish the solutions when plotted in
hydrograph form at an ordinary scale.
The manner of variation of the three parameters in each of the models
which result from the matching of the first three moments to the first three
FIGURE
9-9.Shape factor diagram.
120
144
168
TIME (HOURS)
FIGURE
192
216
259
240
264
288
9-10.Simulation by two-parameter models.
moments of the complete linear solutionshows some interesting features. In

the case of the diffusion plus lag model, a change in the length of channel
considered does not result in any change in the value of the convective velocity
(a) or the ''hydrauUc diffusivity'' (D), but the third parameter, the lag (r),
varies in order to maintain the optimum solution and is directly proportional
to the length of channel. In the case of the three-parameter gamma model, the
reservoir lag time K remains constant as in the two-parameter KalininMilyukov model, but both the number of reaches (n) and the lag of the linear
channel (r) vary directly with the length to maintain similarity with the
complete linear solution. In the case of the multiple Muskingum model, the
values of K and X are independent of the reach length and the complete linear
solution is matched by using a number of Muskingum reaches which is proportional to the length. The conclusions given above are developed on the basis of
long reaches of channel and might not hold for short reaches.
The general approach described above can also be appHed to a channel with
lateral inflow.^ Treatment of this case is outside the scope of these lectures. It
may be said, however, that the derivation of the complete linear solution in
lateral inflow is more complex than the one given above. It should be noted
that the linear response obtained is in fact the lUH for a uniform channel.
5 O'MEARA, W. LINEAR ROUTING OF LATERAL INFLOW IN UNIFORM OPEN CHANNELS.
M. Engin. Sei. Thesis, Dept. Civ. Engin., Univ. Coll., Cork, Ireland. 1968.
260
It is interesting to note that one of the models which is most successful in

simulating the complete solution, particularly for Froude numbers approaching
1, is the model consisting of a rectangle routed through a linear storage
element. In fact, this model is the Zoch-Clark model of routing the time-areaconcentration curve through a linear reservoir.
3.6
3 41-
Complete (LCR)
Lagged Diffusion /
3.2[ 3-Parameter Gamma
Musliingum
(Multiple Reacli)
2.8
2.4
S4
2.0
1.6
1.2
0.8-
0.4-
MJBUBBJH
FIGURE
9-11.Shape factor diagram for three-parameter model (F=0.5).
261
Problems on Surface Flow

1. Calculate the steady state profile for overland flow from a plane 80 feet
long at a slope of 1 in a 1,000 with Chezy coefficient of 100 ft.i^Vsec. and a
lateral inflow of 0.001 feet per second. Draw both the profile and the velocity
distribution along the length of the plane. How would the result be affected
by the neglect of the various terms in the basic dynamic equation?
2. Compare the Horton-Izzard solution and the kinematic wave solution.
What is the relationship between the time to equilibrium in the two methods?
3. Compare the various methods proposed for the numerical solution of
the equation for unsteady overland flow. Based on the different methods,
what difficulties in computation would you expect?
4. Determine the rising hydrograph and the falling hydrograph by the
Horton-Izzard method for the data given in Appendix table 12.
5. Determine the rising hydrograph and the recession hydrograph for the
kinematic wave solution for data in Appendix table 12.
6. Determine the rising hydrograph and the recession hydrograph by a
method of numerical computation for the data in Appendix table 12.
7. Fit a Horton-Izzard type solution to the data for the data in Appendix
table 13.
8. Fit a kinematic wave solution to the data in Appendix table 13.
9. A wide rectangular channel has a bottom slope. So, of 3 feet per mile
(0.57 m. per km), a length of 200 miles (322 km.), and Chezy friction with a
C of 50. Find the discharge hydrograph at the downstream end, using the
method of characteristics if the inflow per unit width is given by function 5 of
Appendix table 1.
10. Use a finite difference scheme, either implicit or explicit, to solve
problem 1.
11. Discuss the question of the stability of the solutions obtained by finite
difference methods for unsteady flow in open channels.
12. Find the linear channel response of the given channel for this particular
flood event from the given inflow and from the outflow computed in either
problem 1 or problem 2.
13. Find the linear channel response for the data of inflow and outflow
given in Appendix table 10.
14. Derive the form of the linear channel response for the following classical
methods of flood routing: lag and route, Muskingum method, KalininMilyukov method.
15. What basic physical assumptions are made for the three classical
methods of flood routing mentioned in problem 6.
16. For the inflow and outflow hydrographs given in Appendix table 10,
find the best value of the lag and the routing coefficient to handle this flood
event by the lag and route method. Draw the linear channel response for
these parameter values.

17. For the inflow and outflow hydrographs given by Appendix table 10,
find the values of K and X for handling this flood event by the Muskingum
method. Draw the linear channel response for these particular values.
18. For the inflow and outflow hydrograph given in Appendix table 10,
find the value of n and k to handle this flood event by the Kalinin-Milyukov
method. Draw the linear channel response for these parameter values.
19. Derive the expressions for the cumulants of the complete linear solution
given in equation 24, page 000.
20. It has been suggested that apart from the effect on lag, a change in the
reference discharge produces only a very small change in the shape of the
outflow hydrograph. Would you expect this change in shape to be greater
where the inflow is a gamma distribution or where the inflow is a cosine curve?
21. What other models, besides those mentioned in the lecture, might be
used to simulate the linear channel response? Indicate a one-parameter model,
a two-parameter model, and a three-parameter model which might have been
used. Calculate the cumulants of these models.
22. In this lecture, the moments and cumulants have been used as a
criterion of matching. Discuss the significance of this criterion, and indicate
what other criteria might have been used and what difference this would have
made to the computations.
23. Using function 5 on Appendix table 1 as the inflow, compute the outflow
hydrograph in a wide rectangular channel for different values of o, C, and L.
24. For the corresponding inflow and channel characteristics used in
problem 23, compute the parameters of a two-parameter simulation model and
generate the simulated hydrograph.
25. For the inflow pattern in the channel of problem 23, compute the
parameters of a three-parameter simulation model and compute the simulated
hydrograph.
26. From a series of results of problems 23, 24, and 25, draw up rough working rules for the circumstances under which each of the models are valid.
Literature Cited
(1) ABBOTT, M. B.
1966. AN INTRODUCTION TO THE METHOD OF CHARACTERISTICS.
(2) AMOROCHO, J. and ORLOB, G. T.
1961.
London.
NONLINEAR ANALYSIS OF HYDROLOGic SYSTEMS. Water Resources Center

Contrib. 40, 147 pp., illus. University of California, Berkeley.
(3) BEHLKE, C. E.
1957. THE MECHANICS OF OVERLAND FLOW.
Calif.
(4) BRAKENSIEK, D. L.
1966.
243 pp.
Ph.D. Thesis.
Stanford University,
HYDRODYNAMICS OF OVERLAND FLOW AND NONPRISMATIC CHANNELS.
Soc. Agr. Engin. Trans. 9(1): 119-122.
AMCR.
263
(5) CHEMERENKO, E. P.
1966.
PRIMENENIYE ELEKTRONNOI VICHISLITELNOI MASHINI VO BREMYA GIDROLICHESKI VICHESLENII [USE OF ELECTRONIC COMPUTERS DURING HYDROLOGICAL
CALCULATIONS].
Meteorologiya i Gidrologiya Zhuzurnal.
Leningrad.
(6) CHOW, VEN TE.

1959. OPEN-CHANNEL HYDRAULICS. 680 pp., illus. New York.
(7) DEYMIE, P.
1939. PROPAGATION D'UNE INTUMESCENCE ALLONGEE (PROBLEM AVAL) [PROPAGATION OF AN ELONGATED INTUMESCENCE]. PFOC. 5th Intematl. Cong.
Appl. Mech., pp. 537-544. New York.
(8) DOOGE, J. C. I.
1960. THE ROUTING OF GROUNDWATER RECHARGE THROUGH TYPICAL ELEMENTS OF
LINEAR STORAGE. Commission on Subterranean Waters, General Assembly
of Helsinki. Internatl. Assoc. Sei. Hydrol. Pub. 52: 286-300.
(9)
1967.
A NEW APPROACH TO NONLINEAR PROBLEMS IN SURFACE WATER HYDROLOGY.
Commission on Surface Waters, General Assembly of Berne.

Assoc. Sei. Hydrol. Pub. 76: 409-413.
(10)
(11)
Internatl.
CONCEPTUAL MODELS OF SURFACE RUNOFF. International Symposium on

Floods, Their Prediction and the Defence of the Soil. Accademia Nazionale
di Lincie. Rome. (In press.)
and HARLEY, B. M.
1967.
LINEAR ROUTING IN UNIFORM OPEN CHANNELS. Internatl. Hydrol. Symposium. V. 1: 57-63, Compt. Rend. Fort Collins, Colo.
(12) DRONKERS, J. J.
1964. TIDAL COMPUTATIONS IN RIVERS AND COASTAL WATERS. 518 pp. Amsterdam.
(13) ESCOFFIER, E. F.
1965. ROUTING FLOOD WAVES ON THE APALACHICOLA RIVER. Internatl. Assoc.
Hydraulic Res. Xlth Internatl. Cong. Paper 3.38. Leningrad.
(14) FoRSYTHE, G. E., and WASOW, W. R.
1960.
FINITE DIFFERENCE METHODS FOR PARTIAL DIFFERENTIAL EQUATIONS. 444
pp. New York.

(15) GiLCREST, B. R.
^ .
.
1950. FLOOD ROUTING. Chapter X, 42 pp. In Rouse, H. R., ed.. Engineering
Hydraulics. New York.
(16) HARLEY, B. M., and DOOGE, J. C. I.
1968. PROBLEMS IN SIMULATING AND EVALUATING VARIOUS METHODS OF LINEAR
FLOOD ROUTING USING A SMALL DIGITAL COMPUTER. Tucson International
Symposium on the use of analog and digital computers in hydrology.
Internatl. Assoc. Sei. Hydrol. Pub. 81: 417-425.
(17) HATHAWAY, G. A.
1945. MILITARY AIRFIELDS; A SYMPOSIUMDESIGN OF DRAINAGE FACILITIES.
(18) HAYAMI, S.
1951. ON THE PROPAGATION OF FLOOD WAVES.
Bui. 1: 1-16. Kyoto University, Japan.
.
-O
T *
Disaster Prevention Res. Inst.
(19) HENDERSON, F. IM.

.
.
rp
1963. FLOOD WAVES IN PRISMATIC CHANNELS. Amer. Soc. Civ. Engin. Trans.,
Jour. Hydraulics Div. 89(HY-4): 39-67.

(20)
1966. OPEN CHANNEL FLOW. 522 pp., illus. New York.
and WOODING, R. A.
1964. OVERLAND FLOW AND INTERFLOW FROM A LIMITED RAINFALL OF FINITE
DURATION. Jour. Geophys. Res. 69(8): 1531-1540.
(22) HoRTON, R. E.
1938. THE INTERPRETATION AND APPLICATION OF RUNOFF PLOT EXPERIMENTS
WITH REFERENCE TO SOIL EROSION PROBLEMS. Soil Sci. SoC. Amer. PrOC.
(21)
3: 340-349.
(23)
1940.
AN APPROACH TOWARD A PHYSICAL INTERPRETATION OF INFILTRATION CAPACITY. Soil Sei. Soc. Amer. Proc. 5: 399-417.
(24) HuGGiNS, L. F. and MONKE, E. J.
1966.
THE MATHEMATICAL SIMULATION OF THE HYDROLOGY OF SMALL WATERSHEDS.
Water Resources Res. Center Tech. Rpt. 1, 130 pp. Purdue University,
Lafayette, Ind.
(25) ISAACSON, E. J., STOKER, J. J. and TROESCH, B. A.
1958. NUMERICAL SOLUTION OF FLOW PROBLEMS IN RIVERS. Amer. SoC. Civ.
Engin., Jour. Hydraulics Div. 84(HY-5): part 1, 1-18.
(26) IWAGAKI, Y.
1951. THEORY OF FLOW ON ROAD SURFACES. Kyoto Uuiv. Col. Engin. Mem.
13(3): 139.
(27) IZZARD, C. F.
1944. THE SURFACE-PROFILE OF OVERLAND FLOW.
Amer. Geophys. Union Trans.
25(6): 959-968.
(28)
1946.
HYDRAULICS OF RUNOFF FROM DEVELOPED SURFACES. Highway Res. Bd.

(Washington, D.C.) Proc. 26: 129-146.
(29) KALININ, G. P., and KUCHMENT, L. S.
1963. THE USE OF INFLUENCE FUNCTIONS FOR COMPUTING A DISCHARGE HYDROGRAPH. Berkeley Symposium of Surface Waters. Internatl. Assoc. Sei.
Hydrol. Pub. 63: 77-89.
(30)
and MILYUKOV, P. I.
1957.
o RASCHETE NEUSTANOVIVSHEGOSYA DVIZHENIYA VODY V OTKRYTYKH

RUSLAKH [ON THE COMPUTATION OF UNSTEADY FLOW IN OPEN CHANNELS].
Meteorologiya i Gidrologiya Zhuzurnal 10: 10-18. Leningrad.

(31) KEULEGAN, G. H.
1944. SPATIALLY VARIABLE DISCHARGE OVER A SLOPING PLANE. Amer. GeophyS.
Union Trans. 25(6): 956-968.
(32)
KLEITZ, C.
1877.
SUR LA THEORIE DU NON-PERMANENT DES LIQUIDES ET SUR SON APPLICATION

A LA PROPAGATION DES CRUES DES RIVIERES [ON THE THEORY OF UNSTEADY
FLOW OF LIQUIDS AND ON ITS APPLICATION TO THE PROPAGATION OF RIVER

FLOODS].
Pont et Chausses Annales (France), sem. 2, No. 48: 133-196.
1964.
OBOBSHCHENNAYA FORMOOLA DLYA RASCHETA GIDROGRAFA STOKA [A GENERALIZED FORMULA FOR THE CALCULATION OF THE DISCHARGE HYDROGRAPH].
Central Forecasting Unit Trans. (TsIP) 133: 23. Moscow.

NLLST RTS 3355, Boston Spa, Yorkshire, Sept. 1966.
Translation
(34)
1965.
GIDROLOGICHESKIE PROGNOZI C PRIMENENIEN ELEKTRONSIK VICHISLITELNIK

MASHIN [HYDROLOGIC PREDICTION USING AN ELECTRONIC COMPUTER]. Central Forecasting Unit Trans. (TsIP) 141. Moscow.

(35) KULANDAISWAMY, V. C.
1966. A NOTE ON MUSKINGUM METHOD OF FLOOD KOUTING.
273-276.
(36) LAURENSON, E. M.
1959.
265
Jour. Hydrol. 4(3):
STORAGE ANALYSIS AND FLOOD ROUTING IN LONG RIVER REACHES.
JOUF.
Geophys. Res. 64: 2423-2431.

(37)
1962.
HYDROGRAPH SYNTHESIS BY RUNOFF ROUTING. Water Res. Lab. Rpt. 66.

University of New South Wales, Sydney.
(38) LIGGETT, J. A.
1959. UNSTEADY OPEN CHANNEL FLOW WITH LATERAL INFLOW. Dept. Civ. Engin.
Tech. Rpt. 2. Stanford University, Calif.
(39)
1961. GENERAL SOLUTION FOR OPEN CHANNEL PROFILES. Amer. SoC. Civ. Engin.,
Jour. Hydraulics Div. 87(HY6): 89.
(40)
and WOOLHISER, D. A.
1967. DIFFERENCE SOLUTION OF THE SHALLOW WATER EQUATIONS. Amer. SoC.
Civ. Engin., Jour. Engin. Mech. Div. 93(EM-2): 39-71.
(41) LIGHTHILL, M. J., and WHITHAM, G. B.
1955. ON KINEMATIC WAVES: PART IFLOOD MOVEMENT IN LONG RIVERS. Roy.
Soc. (London) Proc, Ser. A 229(1178): 281-316.
(42) LIN, PIN-NAM.
1952.
NUMERICAL ANALYSIS OF CONTINUOUS UNSTEADY FLOW IN OPEN CHANNELS.
Amer. Geophys. Union 33(2): 226-234.

(43) McCooL, D. K., GWINN, W. R. and REE, W. O.
1966. SPATIALLY VARIED STEADY FLOW IN A VEGETATED CHANNEL.
Agr. Engin. Trans. 9(3): 440-444.
Amer. SoC.
(44) MASSAU, J.
1900. MEMOIRE SUR L'INTEGRATION GRAPHIQUE DES EQUATIONS AUX DERIVEES
PARTIELLES [REPORT ON THE GRAPHIC INTEGRATION OF THE EQUATIONS OF
PARTIAL DERIVATIVES]. Aunals de l'association des Ingenieurs Sortis des
Ecoles Spciales de Gand, 23: 95-214. Translated by H. J. Putnam,
Rocky Mountain Hydraulics Lab., Allenspark, Colo.
(45) MASSE, P.
1939. RECHERCHES SUR LA THORIE DES EAUX COURANTES [RESEARCHES ON THE
THEORY OF WATER CURRENTS]. Proc. 5th Intematl. Cong. Appl. Mech.,
pp. 545-549. New York.
(46) MEYER, O. H.
1941. SIMPLIFIED FLOOD ROUTING.
Civ. Engin. 11(5): 306-307.
(47) MINSHALL, N. E.
1960. PREDICTING STORM RUNOFF FROM SMALL EXPERIMENTAL WATERSHEDS.
Amer. Soc. Civ. Engin., Jour. Hydraulics Div. 86(HY-8).
(48) MORGALI, J. R.
1963. HYDRAULIC BEHAVIOUR OF SMALL DRAINAGE BASINS.
Dept. Civ. Engin., Stanford University, Calif.
(49)
and LINSLEY, R. K.
1965. COMPUTER ANALYSIS OF OVERLAND FLOW.
Hydraulics Div. 91(HY-3): 81-100.
Tech. Rpt. 30, 115 pp.
Amer. Soc. Civ. Engin., Jour.
(50) NASH, J. E.
1959. A NOTE ON THE MUSKINGUM FLOOD ROUTING METHODS.
Res. 64(8): 1053.
Jour. GeophyS.

(51)
PAYNTER, H. M.
1952.
METHODS AND RESULTS FROM MIT STUDIES IN UNSTEADY FLOW.

Boston
(52) RAGAN, R. M.
1966.
LABORATORY EVALUATION OF A NUMERICAL FLOOD ROUTING TECHNIQUE FOR
CHANNELS SUBJECT TO LATERAL INFLOWS.

111-121.
Water Resources Res. 2(1):
(53) RiCHTMYER, R. D.
1957.
DIFFERENCE METHODS FOR INITIAL-VALUE PROBLEMS.
239 pp.
NeW York.
(54) SCHONFELD, J. C.
1951.
PROPAGATION OF TIDES AND SIMILAR WAVES.

Hague.
(55) SEDDON, A. J.
1900. RIVER HYDRAULICS.
Rijkswaterstaat.
The
(56) STOKER, J. J.
1957.
WATER WAVES:
APPLIED MATH.
THE MATH THEORY WITH APPLICATIONS.
567 pp.
V.
IV, PURE AND
New York.
(57) THOMAS, H. A., SR.

1934. THE HYDRAULICS OF FLOOD MOVEMENTS IN RIVERS. Carnegie Inst. Tech.
(Pittsburgh) Engin. Bul., pp. 1-70.
(58) VASILIEV, O. F. and others.
1964. A NUMERICAL METHOD FOR CALCULATING THE PROPAGATION OF LINEAR
WAVES IN OPEN CHANNELS. Akad. Nauk, Leningrad, Dok. 8(7).
(59) Woo, D. C. and BRATER, E. F.
1962. SPATIALLY VARIED FLOW FROM CONTROLLED RAINFALL. Amer. SoC. Civ.
Engin., Jour. Hydraulics Div. 88(HY-6): part 1, 31-56.
(60) WOODING, R. A.
1965a.
A HYDRAULIC MODEL FOR THE CATCHMENT-STREAM PROBLEM.

KINEMATIC WAVE THEORY. Jour. Hydrol. 3: 254-267.
PART i:
(61)
1965b. A HYDRAULIC MODEL FOR THE CATCHMENT-STREAM PROBLEM.
NUMERICAL SOLUTIONS. Jour. Hydrol. 3: 268-282.
PART II :
(62)
1966.
A HYDRAULIC MODEL FOR THE CATCHMENT-STREAM PROBLEM. PART III I

COMPARISON WITH RUNOFF OBSERVATIONS. Jour. Hydrol. 4: 21-37.
(63) WooLHisER, D. A., and LIGGETT, J. A.
1967.
UNSTEADY, ONE-DIMENSIONAL FLOW OVER A PLANETHE RISING HYDROGRAPH. Water Resources Res. 3(3): 753-771.
1964.
BIBLIOGRAPHY AND DISCUSSION OF FLOOD ROUTING METHODS AND UNSTEADY
FLOW IN CHANNELS. U.S. Geol. Survey Water-Supply Paper 1690, 235 pp.
(65) Yu, Y. S., and MCNOWN, J. S.
1964. RUNOFF FROM IMPERVIOUS SURFACES. Jour. Hydraulics Res. 2(1): 3.
LECTURE 10:
CONCEPTUAL MODELS OF SUBSURFACE FLOW
Lecture 9 dealt with mathematical simulation and conceptual models for
flow processes that constitute part of the direct catchment response to inflow.
It was concerned, therefore, with the simulation of processes contributing
to the formation of the unit hydrograph, which is required either as the direct
storm response or as part of the simulation of the total catchment response.
The remaining two subsystems shown on figure 7-6 are the soil-wâter response
system and the ground water response system.
The treatment of these two phases of subsurface flow is similar to that for
overland flow and channel flow in the last lecture. In each case, the basic
equations derived from the physics of unsaturated and saturated flow in
porous media will be given, together with an account of the more important
solutions based on simplified versions of the fundamental equations. A brief
description will then be given of how conceptual models may be used to
simulate these portions of the hydrologie cycle. It has been mentioned previously that the soil phase is the subsystem of the hydrologie cycle in which
the least systems work has been done. Only in recent years has any wôrk of
this type been done in regard to ground wâter flow. Consequently, the present
lecture will be largely concerned with a review of the theoretical and empirical
relationships which have been proposed and which are necessary as a background to the tackhng of the problem from a systems viewpoint.
Movement of Soil Moisture

' In considering the soil phase of the hydrologie cycle, wê are concerned with
the rate and amount of infiltration into the soil through surface entry, the
rate and amount of downward percolation from the surface to the water
table, the amount of soil moisture held in storage, and the rate and amount of
depletion of soil moisture storage either by evaporation at the surface of the
soil or by transpiration through plants. Infiltration is probably the most
important of these processes since it controls the extent to which total precipitation becomes effective as an input to the system representing the rapid
response of the catchment. Physical information on infiltration is available
from laboratory experiments, field results in infiltrometers, analysis of recorded
hydrographs, and the computation of watershed indicators of equivalent
rates of infiltration.
Any theory of infiltration must be grounded on the principles of soil physics
(i, 5). The water in unsaturated soil is held against gravity mainly by the
action of soil suction. The curve showing the relationship between soil suction
and soil water content is referred to as the moisture characteristic curve for
that particular soil. The soil moisture characteristic curve exhibits a hysteresis
267
effect for any given history of alternative wetting and drying. The moisture
characteristic curves in such cases can span the area between two Umiting
curves, one for drying and the other for wetting.
If we ignore the effect of temperature and osmotic pressure, the movement
of water will take place under the action of a potential difference in accordance
with a generalization of Darcy's Law:
V=-Kgrsid<j>
(1)
where V is the rate of flow per unit area, K is the hydraulic conductivity of
the soil (which is dependent on moisture content), and </> is the hydrauUc
head or potential. The potential is made up of pressure head and elevation:
<t>=-+z
7
= -S+z
(2a)
(2b)
where
S= is the soil suction
y
(2c)
and z is the elevation above a fixed datum.

In considering infiltration, percolation, and evaporation, we are largely
concerned with flow in a vertical direction. For vertical flow, equation 1
becomes :
V=-K~iS+z)
oz
(3a)
(9 Si
V = K--K
(3b)
or
If the soil suction (S) is assumed to be a single-valued function of the moisture

content (c), then we can define the hydrauhc diffusivity of the soil as:
B.-K'
(4a,
V=-Df-K
(4b)
and write equation 3 as :
which is th one-dimensional diffusion form of Darcy^s Law. Over a given

range of moisture content, the variation in D will be less than the variation
inK.
For unsteady flow in an unsaturated soil in a vertical direction, we have the
269
equation of continuity :
^+^ = 0
dz dt
(5)
where V is the rate of flow per unit area and c is the moisture content expressed
as a proportion of total volume. Combination of equations 1 and 5 gives us :
(K'-^)J-^
(6a)
^lU'-^yj^j^
(6b)
dz\
dzj
dt
or using equation 2b we get :

dz\
dz J
dz
dt
and with equation 4b we get :
(D'^)+'-^J-^
dz \ dzJ dz dt
(6C)
For infiltration into a very dry soil (or upward movement from the ground
water to a dry soil surface), the gradient of the soil suction will be very much
larger than the difference in elevation. Consequently, the last term (K) on the
right-hand side of equations 3b and 4b can be neglected compared with the
other two terms; similarly, the second terms on the left-hand side of equations
6b and 6c can be neglected. Omission of these terms corresponds to the assumption that the effect of gravity on water movement is negligible compared to
the effect of the gradient of soil moisture suction.
Equation 6 is a nonlinear parabolic equation since hydraulic conductivity
(K) and the hydraulic diffusivity (D) are functions of the moisture content
(c). Equation 6c has the same mathematical form as the concentrationdependent diffusion equation in mathematical physics and is the most convenient form for theoretical analysis.
A number of authors have suggested empirical relationships between the
unsaturated permeability {K) or the hydraulic diffusivity (D) on the one
hand, and the moisture content (c) or the soil suction (S) on the other. These
can be used in the place of purely empirical moisture characteristic curves to
predict water profiles and water movement; they could also be used as the
basis for conceptual models of the movement of moisture in the unsaturated
zone. Bear, Zaslavsky, and Irmay (1) suggested that unsaturated permeability can be related to saturated permeability by the equation:
i^sat
\nno/
where c is the moisture content, n is the total porosity, and no is the ineffective

or irreducible porosity. Gardner (13) suggested expressing unsaturated
permeability as a function of soil moisture suction by an equation of the form:
K=
6+5"
(8a)
K'sat
h+S"^
(8b)
where K is the unsaturated permeability, S is the soil moisture suction, m is a

parameter which has a value of approximately 2 for heavy soils and approximately 4 for sands, and a and h are empirical parameters. Gardner also used
an exponential relationship between unsaturated permeability and soil
suction :
K
Tr- = exp(-a)
(8c)
Some special cases of the relationship given in equation 8a had been suggested previously by Wind (^5) and by Remson and Fox (36). Gardner and
Mayhugh {14) have suggested the following relationship for the hydraulic
diffusivity:
D
= exp[a(c-fe)]
(9)
where Do is the value of the hydrauhc diffusivity for the moisture content
c = 6 and a and h are experimental parameters.
Under steady state conditions with no loss or gain of moisture to the
atmosphere, the soil moisture profile will be in equilibrium. The moisture in
the unsaturated zone is held above the water table against the pull of gravity
by the soil suction; the curvature of the interface between soil air and soil
water allows the soil water to be at a pressure less than atmospheric.
In a steady percolation rate {q) from the surface to the water table, we have :
.=i^[l-^]
(lOa)
or in terms of the hydraulic diffusivity:

5 = ^+0 ^
(10)
The level above the water table at which a particular moisture content occurs
271
can be determined from :

z= [
-Tp-dS
(11a)
z= /
-^.0= /
dc
(lib)
In a multilayered soil, the integration can be carried out separately in

each layer :
'
^^dS
^'
dc
(12a)
(12b)
Since it is the soil suction that is continuous across the boundary between
layers, there will be discontinuities at the boundaries if the computation is
done in terms of moisture content.
The steady upward movement of water from below the water table to
provide a steady rate of evaporation (e), gives rise to a similar formula, except
that in this case we have :
-(f-O
(13a)
e=-D--K
dz
(13b)
z= f' --.dS
(14a)
,= f
Jc
(14b)
and the solution is given by :
-J^.dc
K+e
If the water table is very close to the surface, there will only be a small
drying of the surface, and evaporation can occur at the potential rate. In the
case of a deep water table, however, the gradient necessary to move water up
from below the water table results in a high soil suction at the surface and,
consequently, a lower moisture content and a lower unsaturated permeability.
By using an empirical relationship between K and S, it is possible to integrate
equation 14a and so predict the soil profile for capillary rise {36, 37). A similar
calculation could be used to estimate transpiration by using a constant suction
at a given elevation to simulate root action.

For the conditions of deep water table and high evaporation rate, it can be
shown that there is a Umiting rate of evaporation which depends on the depth
of the ground water and the soil properties (13). Under some conditions, this
concept of a hmiting rate of evaporation, depending not on climatological
data but on soil properties and conditions, may be of importance for the
modeling of the soil phase of the total catchment response.
Gardner has shown that if the unsaturated permeability can be assumed to
have the relationship with soil suction given by equation 8a, then for any
given value of m, the limiting rate of evaporation would be given by:
eum
constant
where Zo is the depth of the water table below the surface. Evaporation may
take place at greater than this hmiting rate, but if it does the water being
supplied from soil moisture storage rather than ground water storage and the
soil moisture distribution is not that of a steady state. Schleusener and Corey
{39) have reported experimental results indicating the existence of a maximum
rather than a limiting rate of evaporation from the soil surface and suggest
that this phenomenon can be explained by the effect of hysteresis.
Unsteady Movement of Soil Moisture

In practice, the soil moisture rarely attains an equilibrium profile. Rather
than having a constant rainfall rate or a constant evaporation rate at the
surface for a long period, we have alternating precipitation and evaporation
resulting in continual changes in the moisture profile and the unsteady movement of water either upwards or downwards in the soil. As hydrologists, we
are largely concerned with conditions which occur when a dry soil is wetted
by precipitation at a higher rate than the average or when a wet soil is depleted
of its moisture content by an evaporation rate higher than average. As in the
case of steady downward percolation, or steady upward capillary rise, the
problems of upward and downward movement are essentially similar, and
techniques which work for one will be appropriate for the other. Due to
limitations of space, only the infiltration problem will be dealt with in the
present discussion.
It is important to distinguish between the infiltration capacity of the soil at
any particular time and the actual infiltration occurring at that time. Infiltration capacity is the maximum rate at which the soil in a given condition can
absorb water at the surface. If the rate of rainfall or the rate of snowmelt is
less than the infiltration capacity, then the actual infiltration will be less than
the infiltration capacity since the amount of moisture entering the soil cannot
exceed the amount available. A number of empirical formulas for infiltration
capacity have been proposed in the literature. Those by Kostiakov {23),
Horton {19), Holtan {17), and Overton {29) are discussed below. The theo-
273
retical formulas which are discussed are: the solution of the basic equation
on a constant diffusivity (4), the solution based for a constant profile (30),
and Philip's general solution for the ponded infiltration case (31).
The following notation will be used for both empirical and theoretical
formulas :
/=
/o =
fc =
fe=
F=
Fc =
Fp=
Fe=
Fce=
Fpe=
rate of infiltration capacity

initial rate of infiltration capacity
ultimate rate of infiltration capacity
rate of excess infiltration ( ffc)
volume of infiltration
ultimate volume of infiltration
volume of potential infiltration (Fc F)
volume of excess infiltration (Ffct)
final volume of excess infiltration (Fcfc't)
volume of potential excess infiltration {FceFe).
In the case of formulas for infiltration volume, the corresponding formula for
infiltration rate can be gotten by differentiation. In the case of formulas for
infiltration rate, the formula for infiltration volume can be gotten by integration. All of the above definitions refer to infiltration capacity. If it is necessary
to distinguish it, the actual infiltration rate can be designated by /A and the
actual volume of infiltration by FAAs mentioned above, attention will be confined to the more important
empirical equations found in the literature. In 1932, Kostiakov (23) proposed
the following formula for the initial high rate of infiltration:
(16)
where / is the rate of infiltration up to the time when the infiltration rate would
be equal to the saturated permeability of the soil. Horton (19) suggested the
following formula for the rate of infiltration capacity :
f-fc={fo-fc) exp{-kt)
fe=foeexp(-kt)
(17a)
(17b)
Holtan (17) suggested that the rate of excess infiltration in the early part
of a storm could be related to the volume of potential infiltration (Fp) by an
equation of the form :
f-fc==a{FpY
(18a)
fe = a{FpY
(18b)
or
Overton {29) showed that for a value of n = 2 in equation 18, the rate of

infiltration could be expressed as a function of time in the following form :
f=fc'êc'iVWcitc-tn
(19a)
tc = J\-tan-^(FcJj)
(19b)
where tc is given by :
jc
Jc/
and is the time taken for the infiltration capacity rate to fall to its final
value fc.
We turn now from empirical formulas based on the analysis of field observations to theoretical formulas based on the principles of soil physics. For a
soil whose moisture characteristics and unsaturated permeability (or hydraulic
diffusivity) are known, equation 6 for the unsteady vertical movement of
moisture in a soil can be solved by numerical methods (22, 31, 38). We are,
however, more concerned with simplified mathematical formulations of this
particular problem.
One of the simplest models of infiltration into a soil (and the subsequent
downwârd percolation of the wetting front) is that obtained if the hydraulic
diffusivity is taken as constant (4) and, in addition, either the hydraulic
conductivity taken as a constant or the effect of gravity neglected. In this
case, we have :
D =
dz" dt
(20a)
Instead of taking elevation (z) vertically upwârds from a datum, wê express
our equation in terms of the depth of percolation downward from the surface
(a:) so that we have:
D^J-^
dx''
(20b)
dt
For the problem of ponded infiltration into an infinitely deep soil, we have
the boundary conditions :
c = Co,
for
t = 0,x>0
(20c)
c = Cs,
for
t>0,x = 0
(20d)
where Co is the initial moisture content of the dry soil and Cs is the constant
moisture content at the surface (usually, but not necessarily, Csat). Equation
20b is a linear parabolic equation of the diffusion type and has a solution of
the general form :
-(f)
(21a)
which gives the moisture content (c) for a given depth of penetration (x) at a
275
given time (t). The depth of penetration for a given moisture content can be
written as :
a: = i/2.02(c)
(21b)
The total amount of infiltration up to a given time t is given by:
F= f X'dc+Ko't
(22a)
where Ko is the unsaturated permeability corresponding to the initial moisture

content Co. Substituting equation 21b into equation 22a we obtain:
F= r t''^4>2{c)'dc+K4
(22b)
-^ Co
This is clearly seen to give :

F = 1/2. [ 02(c) dc+Kot
(22c)
^ Co
which allows us to express the infiltration volume {F) as a function of time,

and of the initial and saturated moisture contents and the initial permeability :
F = cl>s{co,Cs)f^'+Kot
(23a)
It can be shown that 03 takes the form :
UD
03 = \/ (csat-co)
(23b)
Equation 23 is the infiltration capacity equation for the simple model of

constant diifusivity and constant permeability. On the basis of the definition
of hydraulic diffusivity in equation 4, this is equivalent to assuming that soil
suction is related to moisture content by:
S = | (Csat-C)
(24)
that is, that the soil suction is a linear function of the moisture content. It
should be noted that the two parameters in equation 23a both vary with initial
moisture content.
As an alternative to assuming constant diffusivity and constant permeability, we can make the assumption that the diffusivity is constant but that
the permeability is a linear function of the moisture content, that is, that
^=
-tVsat
(25a)
Csat
By inserting the relationship given by equation 25a in equation 4, wê obtain

a logarithmic relationship between soil suction and moisture content given by:
where
k^
(25c)
Csat
For the model of constant diffusivity and linear permeability, equation 6c

can be written as :
^ (PC
dc
dc
which is the same as equation 20 except for the addition of the '^convective^'
term. It is still a linear equation and is similar in form to both the parabolic
(that is, diffusion analogy) form of the linearized equation for unsteady flow
in an open channel, discussed in lecture 9, and to the linearized equation for
unsteady ground water movement to be discussed later in this lecture. For the
same boundary conditions as given in equations 20c and 20d, equation 26a
has the solution:
/ cCo\
,/ ^ / xkt\
,^
/kx\
^ / x+kt\
When converted from the form of equation 26b, which is appropriate to the
moisture profile, this solution gives for the rate of infiltration capacity:
Solutions to the problem of ponded infiltration have also been obtained by

assuming the hydraulic diffusivity to be a linear or an exponential function
of the moisture content (40).
In 1911, Green and Ampt (15) proposed a formula for infiltration into the
soil based on a model of uniform parallel capillary tubes. In fact, their approximate treatment is not dependent on this specific model but merely on the
assumption that the advancing moisture profile consists of two partsan
upper zone of higher moisture content (C2) separated from the original dry
soil (c = ci) by a sharp discontinuity (5, 30).
The rate of flow through the upper part of the soil may be written as:
V=K,.^-^^^^
(27a)
277
where 02 is the total head at the top of the column and is given by:
<t>2 = x+H
(27b)
where x is the depth of penetration of the higher moisture content and H is

the depth of ponding on the surface. <i)i is the total head immediately below
the discontinuous wetting front and is numerically equal to the suction (So)
Sit which air enters the soil medium. Consequently, equation 27a can now be
written as :
Since the upper part of the soil is assumed to have a constant mean moisture
content (02), we can also write:
V={c-cOat
(27d)
Combining equations 27c and 27d we have :
at
\C2Cl/ \
which integrated will give :
This equation has the disadvantage that it relates the depth of penetration
(x) to the time (t) in implicit form. However, it can be seen from equation 26c
that the rate of infiltration is extremely high for small values of x and approaches the value K2 for large values of x.
A more complete theory of infiltration allowing for concentration-dependent
diffusivity and for the gravity term has been developed by Philip (31). Philip
showed that the equation for the depth of penetration of given moisture
content can be represented by the series :
x = ai{c) 't'^''+a2(c)t+az{c)t'/^+
amic)t^^^+
(29)
and states that, for the range of t and of values of D -and K of interest to soil
scientists, the above series converges so rapidly that only a few terms are
required for an accurate solution. Equation 21b developed above for constant
diffusivity and constant permeability is seen to correspond to the first term
of equation 29.
As for the simpler model, the volume of infiltration can be obtained by
integrating the depth of penetration over the range of change in moisture

content. For the present model this gives :
/Csat
X'dc+Kot
(30a)
/Csat
[i{c)t''^+a<,{c)t+a,{c)t^f^+
'}dc+Kot
(30b)
CO
which converges except for very large values of t. Philip suggested that for
most practical purposes only the first two terms are required so that we can
write :
F = S.i/2+A.
(31a)
where S is called the sorptivity and is given by :
/Csat
ai(c) dc
(31b)
and the second parameter A is given by :

/Csat
a2(c) dc
(31c)
Co
In a series of papers, Philip {31, 32) discussed the imphcations of the

solution given by equation 30, the nature of the surface profile, the effect of
surface ponding, and other factors.
It must be emphasized that the solutions given above are all for one particular formulation of the infiltration problem. In every case, the analysis is
made on the basis of an infinitely deep soil profile with a uniform initial
moisture content, into which infiltration takes place as a result of the saturation of the surface. Such a stylized case wôuld have to be modified in several
respects before it would correspond closely to conditions in actual catchments.
In practice, the theoretical solution would be modified by the presence of a
water table at some finite depth, by the actual moisture distribution in the
profile at the instant that the surface is first saturated (which would depend
on the previous history of moisture distribution and movement in the profile),
by distinct layers in the soil profile which might give rise to interow, on the
possibility of shrinkage and swelling in the soil, and so on. Nevertheless, as in
so many other instances in hydrology, a simple model can be adopted to get a
feel of the phenomena under study and then be used as the basis of a more
complex model.
Comparison of Infiltration Formulas

It is interesting to compare with one another the mathematical equations
for ponded infiltration based on various simplifying assumptions and to relate
them to the empirical equations which have been suggested. This is done in
the present section for the theoretical and empirical equations mentioned
279
earlier. Finally, an attempt is made to relate the mathematical simulation of

infiltration to possible conceptual models of infiltration to explore the possibility of using conceptual models in the soil moisture phase of the hydrologie
cycle.
The first comparison made is for initial infiltration rates, that is, for the
form of the mathematical equations at small values of t. For the model based
on constant hydraulic diffusivity and constant saturated permeability, the
infiltration ratewhich can be obtained by differentiating equation 23is
given by :
/-(csat-Co) ^^^+K
(32a)
The infiltration rate is seen to vary inversely with the square root of the time
elapsed and to vary directly with the difference between saturated and actual
moisture content (that is, with the volume of pore space available).
The infiltration rate for the model based on constant hydraulic diffusivity
and a linear variation in unsaturated permeability with moisture content is
given in equation 26c. For small values of t, this equation can be expressed
(31) in the following form:
If only the first two terms are used, this becomes :

/=(c3.-c)^^+^+^
(320
It can be seen by comparing equations 32a and 32c that if the constant unsaturated permeability in the first model is taken as the mean value of the
initial and the saturated permeability, the infiltration rates will be identical
for those small values of the time in which the series within square brackets in
equation 32b can be adequately represented by the first two terms.
For the Green and Ampt model, the infiltration for small values of i and,
hence, small values of x can be obtained by neglecting 1 in the last term within
the brackets in equation 28a and then integrating to obtain:
x = J^^' (H+So) '2t
(32d)
By differentiating the latter equation and substituting the value in equation

27d, we obtain for the infiltration rate :
^^ lc2-Co){K2){So+H)
^^^V02
.ojy..2JKô^^^J
^32^^
An interesting comparison between the Green and Ampt model and the

model based on constant hydraulic diffusivity and constant unsaturated
permeability can be made as follows. If the concentration difference in equation 32a is taken under the square root sign, it will appear as squared. The
model of constant diffusivity and permeability in equation 24 implies a specific
relationship between soil suction and moisture content indicated by that
equation. Accordingly, one of the two equal factors of (CsatCo) under the
square root sign can be replaced by K.So/D so that we have for the infiltration
rate:
^^ j{C2-Co){K){So)
f'^
'
(^2
thus indicating a close similarity between the two models.

Finally, the behavior for small values of t of Philip's general solution for
ponded infiltration can be examined. PhiUp (32) suggested that for practical
purposes only the first two terms of equation 30 need be retained and that the
equation can be written in the form of equation 31a. The infiltration rate
corresponding to this equation is given by:
/=+^
(32g)
in which the parameter S is termed ^^the sorptivity.^'

All four models are thus seen to give closely similar solutions for the initial
period of infiltration and to correspond to the empirical equation proposed by
Kostiakov in equation 16, with the special value of 6 = 3^. From a systems
viewpoint, it would appear that the high infiltration rates at the start of a
storm could be represented by equation 32g with the sorptivity (S) and the
ultimate infiltration rate (A) as parameters to be determined.
A comparison can also be made between the behavior of the different models
at very large values of t. For the constant diffusivity and constant permeability model, the ultimate infiltration rate is given by the constant value of
the permeability K. For the model based on constant diffusivity and linear
variation of unsaturated permeability, the general solution given in equation
26c has the following form for large values of t (33) :
/j;)\3/2
^2\
For very large values of t the exponential term will render the first term on the
right-hand side of equation 33 negligible, and give as the ultimate value of the
infiltration rate the saturated permeability i^sat.
It is clear from the above discussion that all of the models are compatible
with the equation proposed for practical use by Philip and given in equation
32g. However, in linear models the indication is that the first term will be
281
proportional to the difference between the saturated moisture content and the
initial moisture content. Accordingly, it is suggested that a convenient formula
for use in the simulation of infiltration might be:
^â(c3at-Co)^^^
(34)
Using the form of equation 34 rather than equation 32g would enable us to
allow for the effect of varying initial moisture contents in the synthesis of a
catchment response. For any storm event, the initial moisture content would
be available from the soil moisture accounting.
Overton (29) has shown that a number of infiltration equations can be
derived by postulating a relationship between the rate of infiltration (or excess
infiltration) and the volume of either actual or potential infiltration (or excess
infiltration). Thus, we can write each of the models for infiltration in terms
of the variables listed. Thus, if we write:
^
we are using the assumption that the rate of infiltration is inversely proportional to volume of infiltration up to that time. Equation 35 can be readily
integrated to give :
F=\/2t
(36a)
or
/=^?.ri/2
(36b)
which is the Kostiakov formula for 6 = 3^. Similarly if wê write:

/,
Je = p
(37a)
f-f.-lF
W'W
that is
the solution can be show^n to be :
which is of the same form as the Green-Ampt solution.

If the rate of excess infiltration is taken as inversely proportional to the
volume of excess infiltration, as follows :

fe = y
(38a)
^~^' = 'F^t
^^^^)
or
then the solution is :

F=V2^t+fct
(38c)
which is Philip's equation 31a with:

S=\/2^
(38d)
A =fc
(38e)
It would be interesting to see if a rate-volume equation could be found that

would give additional terms in Philip's general solution.
^ If we relate the rate of infiltration to potential infiltration volume, the
simplest equation is :
f=ctFj,
(39a)
f=a(F,-F)
(39b)
f=fo-aF
(39c)
^ = -[l-exp(-aO]
a
(39d)
/=/o-exp(-aO
(39e)
or
or
which has the solution :
and
Assuming the relationship :
fe = aFp
(40a)
is equivalent to equation 39 since it reduces to :

f=fo-aF
(40b)
and, hence, it also has the solution:

/=/o-exp(-aO
(40c)
The more general relationship :

fe = aFpe
(41a)
or
f-fc = a{F,e-Fe)
(41b)
283
or
f-fc=fo-fc-a{F-U)
(41c)
f-fc=(fo-fc)exp{-at)
(41d)
has the solution :

which is the Horton equation. Finally the relationship :
fe = aF^'
(42a)
is the relationship proposed by 0verton himself which gives:
f=fc'êc'iV^c{tc-t)-]
(42b)
as given earlier in this section.

Apart from its intrinsic interest, the formulation of the infiltration as a
relationship between a rate of infiltration and a volume of actual or potential
infiltration would appear to have many advantages in the formulation and
computation of conceptual models of the soil moisture phase and the simulation of catchment response.
We are famihar with the concept of a linear reservoir as an element in which
the outflow is directly proportional to the storage in the reservoir. Equation
39a represents an element in which the inflow is proportional to the storage
deficit and, hence, might be considered as a special conceptual element to be
known as a linear absorber. The relationship indicated by equation 41a could
be considered as consisting of a linear absorber preceded by a constant rate of
overflow, which diverts moisture at the rate fc around the absorber and feeds
into the ground water reservoir even when the field moisture deficit is not
satisfied. By analogy, equation 35 might be considered as being represented
by a second type of conceptual element in which the inflow into it is inversely
proportional to the amount of inflow which has taken place. For want of a
better name, this might be referred to as a linear inverse absorber. Much
work remains to be done in this area, but there are indications of the track
to be followed.
Basic Equations of Ground Water Flow

Even though linear solutions have been widely used in ground water
hydraulics, until recently there has not been a deliberate treatment of ground
water response as a linear system. An assumption frequently made in applied
hydrology has been that the ground water reservoir acts as a single linear
storage element. This assumption is imphcit in the fitting of exponential
recession curves to hydrographs and the plotting of falling hydrographs on
semilog paper. Such a model is an extremely simple one, and we certainly have
available the techniques to go beyond it. If we wish to tackle the ground water
phase in the same way in which wê tackle the surface runoff phase, then we
should make the assumption that the ground water system is a linear system

and not that it is just a particular highly simplified, single-element, Hnear
system. If we do so, we have available all the techniques of linear analysis and
synthesis. We can derive a ^'ground water unit hydrograph^' provided we
know the ground water recharge and the ground wâter outflow.
From the point of view of synthesis, we can learn from what has been done
in the simulation of the surface runoff phase of the hydrologie cycle, but we
should beware of following too slavishly the approaches and the models
developed in that particular field. If we are to simulate successfully, we must
understand the physical hydrology of ground water flow and make use of
existing knowledge in this field so that our models can be as ''realistic'' as
possible. This section gives a very brief review of the basic equations of ground
water flow, and then discusses a linearized solution of a special case of ground
water flow and the possibiUty of simulating this solution by conceptual
models.
The basic equations of ground water flow are well-established and can be
studied in standard works such as Muskat {28), Polubarinova-Kochina {35),
Luthin {26), Harr {16), deWiest {10), and Bear, Zaslavsky, and Irmay {!).
Just as in open channel flow, we avoid the difficulties inherent in the analysis
of two-dimensional flow by reducing our problem to one based on the assumption of one-dimensional flow. With this assumption, the equation of continuity
for horizontal flow through soil in a saturated condition is:
^+/- = r(x,0
(43)
where q is the horizontal flow per unit width, h is the height of the water table,
/ is the drainable pore space (which is assumed to be constant), and r{x,t)
is the rate of recharge at the water table.
The assumption that the stream lines are all horizontal and the velocity
uniform with depth is known in ground water hydraulics as the DupuitForcheimer assumption. For these conditions, Darcy's equation:
V=-Kgrsid(l>
(44a)
yK~
(44b)
reduces to
where K is the hydraulic conductivity (usually assumed to be constant) and

gives us the following relationship between flow and height of water table:
^=-Kh-
(45)
Substitution from equation 45 into equation 43 gives us the differential
285
equation:
For a set of parallel drains or parallel trenches, which are a distance S apart
and which are subject to a constant rate of recharge at the water tables, the
equilibrium situation is given by:
d /, dh
(Khy)+r = 0
dx
(47a)
with the boundary conditions given by :

x=o
or
S,
h=d
(47b)
where d is the depth of water over the parallel drains or the depth of water in
the parallel trenches, whichever is appropriate. This nonlinear equation has
the solution :
Kh^+r Ix-^j =Kd'+Y
(48)
which is known as the elHpse equation.

It must be remembered that equation 47 is based on the Dupuit-Forcheimer
assumptions and is only correct if the flow can be vahdly approximated by a
purely horizontal flow; if the drains or the trenches do not penetrate to the
impervious layer, or if the depth {d) is small, the assumption ceases to be
reasonable. The various solutions proposed for dealing with the problem as a
two-dimensional flow may be reviewed in Luthin (26) or in a review paper by
Kirkham {21 ). In our discussion of both steady and unsteady flow, we will be
content to take the Dupuit-Forcheimer assumptions and the solutions derived
from them as the basis of our discussion.
The problem of the recession of the water table after cessation of recharge
is an important one in drainage engineering and has been widely studied. A
recent review of work in this field has been given by van Schilfgaard (43). As
in other fields, the first attempt is to seek a linear solution. There are two ways
in which equation 46 can be linearized. In the first and more common linearization, the water table height inside the bracket in the first term of equation
46 is frozen at some parametric value (h) and then removed outside the second
differentiation with respect to x, thus giving:
Kh^^,+r{x,t)=f^^
dx^
ot
(49)
In the second form of linearization, h^ is used as the dependent variable

instead of h and an equivalent parametric value of h is used to adjust the term

on the right-hand side of equation 46:
2-^+.(x,0=^^^
(50)
Though the first hnearization given by equation 49 is the more common

form, the second one given by equation 50 has the advantage that for the
steady state it gives the eUipse equation of equation 48, which is the correct
nonhnear solution, whereas equation 49 gives a parabohc shape to the water
table for the steady state condition. Both equations 49 and 50 are parabolic in
form and can clearly be solved by the techniques which have proved successful
in the analysis of problems in heat flow and of ''diffusion-type'^ problems
Equation 49 for the initial condition of a level water table = /o was solved
by Glover^ (Also, see reference 12 by Dumm)who obtained:
h
Hho-d)
sin(n7ra:/>S)
nVKht\
where
h = elevation of water table above impervious layer
d= elevation above the impervious layer of water surface
in trench (or above drain)
ho= maximum elevation of water table
X = horizontal distance from trench or drain
S = spacing of trenches or drains
K= saturated permeabiUty of soil
t = time elapsed since start of recession
/= drainable pore space
Kraijenhoff (24) has pointed out that the soil and drainage characteristics
in equation 51a may be grouped together into one parameter, which he defined
as the reservoir coeflScientj:
^ = -2Th
(51b)
so that Glover's solution can be written as :
Kraijenhoff also pointed out that Glover's solution was the solution for a
1 GLOVER, R. E., and BITTINGER, M. W. SOURCE MATERIALS FOR A COURCE IN TRANSIENT
GROUND WATER HYDRAULICS. Colo. State Univ. 1959. [Mimeographed.]
287
finite volume of recharge in an infinitesimal time and consequently equation

51c represents the impulse response of the ground water system.
If we adopt the second linearization instead of the first, a similar equation
can be obtained, except that it will be in terms of h^ rather than h. The dififtculty about conflicting predictions of the shape of the water table profile does
not affect us in our study of the recession of outflow. The outflow to a drain
or a trench is given by :
Q = 2M.=o
(52a)
In the first linearization, q is given by:

q=_Kh-=-Kh~
dx
dx
(52b)
and in the second equation, q is given by :
so that in either equation we obtain for the discharge :

Q=
SKh{ho-d)
^
>^
^
2^
n=l,3
/
2^^
exp(-n2-.J
\
/^Qo^
(53a)
J/
If the initial height of instantaneous recharge (hod) is expressed in terms

of the volume of recharge, the drain spacing, and the drainable porosity of
the soil, we have:
SKhV ^
q = -^;^ Z.
SJo =1,3.
exp(-n'-^
(53b)
SO that for an instantaneous input of unit volume we have as the impulse

response :
Ht)=-'-
Z expf-n^-.)
(54)
Obviously as t becomes large, the first term in the infinite series will dominate,
and the outflow will approximate that from a single linear reservoir. For small
values of t, however, the other contributions cannot be neglected, and for a
value of t equal to zero, they are all equal and add up to an infinite initial
value oih{t).
The response function given in equation 54 has been normahzed to have
unit volume, and its moments can be shown to be:
W = î = ^-i
(55a)

rr ,
7ir^ .,
'" '^r'-^
(55b)
The shape factors for the Glover solution as given by equation 54 calculated
from equations 55a to 55c as :
7
52 = -=1.40
(55d)
124
3 = -^ = 3.54
(55e)
If moment matching were taken as a criterion, then simulation of the Glover

solution by a cascade of linear reservoirs (that is, by the gamma distribution)
would require a value of ?2 = 0.7 and a value of 7^ = 1.15.
Note that if equation 54 were plotted on semilog paper, the first term would
plot as a straight line and the other terms would only make contributions at
small values of t Following the lines of classical hydrology, we might be
inclined to interpret such a result as indicating that the first term was the true
baseflow and that the contributions due to the other terms represented residual
interow or surface runoff. If we took the straight line on the semilog plot as
the baseflow, we would in fact truncate the infinite series of equation 54 and
use only its first term in forming our implicit model. Such a procedure would
have the further defect that we would take the lag of the system as equal to the
reservoir coefficient j rather than the value given by equation 55a. The work
which Kraijenhoff has initiated in applying the systems approach to the
ground water phase is most important in so far as it indicates the hkelihood
of considerable progress if the techniques of parametric hydrology developed
for the surface water part of the cycle are applied to the ground water.
Even if we wish to persist with the model of a single linear reservoir (that
is, the first term only of the Glover-Kraijenhoff equation), then we can
extract more use from this assumption than is normally done. If we assume
that the recession for the ground water phase of our watershed is given by:
Q = Qo-exp^--^j
(56a)
then we are in fact assuming that the ground water reservoir acts as a single
linear reservoir where we have :
If such a system is subjected to recharge at a uniform rate (), then the
289
ground water outflow during recharge will be given by:

Q = R l-exp(-^ +Qo'exp(-^
(56c)
where the time origin is taken at the start of recharge. If the recharge ends
after a time D, then the ground water outflow at this time will be:
Q = R [l-exp (^_|^j+Qo.exp (|-|)
(56d)
Both before and after the recharge, the ground water outflow will follow the
master recession curve. The outflow given by equation 56d is the same as
would have been given if there had been an instantaneous increase in discharge at a time = 0 of an amount :
h(i)-']
Q = |êxp(^-j-lJ
(56e)
which would then recede along with the initial outflow. Assuming for the
moment that there were no thresholds in the system and that recharge were
taking place directly to ground water, equation 56e could be used together
with a plot of ground water outflow and a knowledge of the volume of recharge
to determine the rate and duration of recharge. Quite apart from this aspect
of analysis, equation 56c indicates that the separation between ground water
and direct storm runoff should be taken as a curve which is concave downwards rather than as a straight line.
The discussion given above for the recession of the water table deals with
horizontal flow overlying a horizontal impervious layer. The analysis can be
adapted to flow over an inclined impervious layer, but still retaining the
Dupuit-Forcheimer assumptions and the linearization of the equation. If the
slope of the impervious layer is taken as a, then equation 45 must be modified
to give:
q=-Kh(-a\
(57a)
where h is still the elevation of the water table above the impermeable layer,
which has a downward inclination of a to the horizontal. Similarly, equation
49 must be modified to give :
Kh'^-Ka'^+rix,t)=f'^
dx^
dx
at
(57b)
This is still a parabolic linear differential equation and resembles in form the
convective diffusion equation, which has already been encountered as a model
of unsteady flow in an open channel. Thus we see that the same model can be
used to simulate unsteady flow in an open channel, unsteady flow in the unsaturated zone, and unsteady flow in the saturated zone.
Other simple models can be devised for the recession of ground water flow.
Thus we could estimate outflow on the basis of a succession of steady states in
each of which the ellipse equation was assumed. The relationship between
discharge and the water table for the ellipse equation is given by :
Q = ^ {h'm..-d')
(58)
while the storage above the water level in the drains or trenches is given by:
7=/.^ {hra..-d)
(59)
Two successive values of /max could be taken and the difference in storage
computed from equation 59. The average rate of outflow could then be
approximated from equation 58 for a value of /max half way between the two
assumed. Division of the change of storage by this mean rate of outflow would
give an estimate of the time taken for the level to fall by the amount assumed.
For the combination of a very deep trench and a shallow rise of water table,
as follows :
^1
a
(60a)
equation 58 could be written as :

SK
Q=-Âo(/i>ax-d)
(60b)
Comparison of equation 60b with equation 59 indicates that storage is proportional to outflow so that we would in fact get an exponential recession with
the recession constant given by :
K==
=?
(60c)
For a steep rise in water table in a shallow trench, that is, for:
h
^1
a
(61a)
equation 58 becomes :
Q = -A^max
(61b)
291
and equation 59 becomes :

7='./l.ax
(61c)
SO that the outflow is proportional to the storage squared:

^ 64 72
Q = --T7
(61d)
For intermediate conditions, it should be possible to simulate the recession

with fair accuracy by treating the ground water system as a nonlinear reservoir
with the outflow proportional to some power of the storage:
Q = a{V)'
(62)
where c has a value between 1 and 2.

The above discussion was not only based on a simplified analysis of ground
water storage and flow, but it has also disregarded the Hnkage between ground
water flow and the other phases of the hydrologie cycle. Space precludes a
discussion of the work that has been done in this regard. However, the approach which has been outlined above can also be applied to such problems
as the recharge of bank storage due to an increase in channel flow and the
subsequent recession after the channel flow has diminished (5, 4^), and the
interaction of ground water with the unsaturated zone and with the atmosphere {3, 11).
Problems on Subsurface Flow

1. Look up in the literature empirical results for the values of the soil
suction and the unsaturated permeability for a number of soils of different
types. In each case, derive the hydraulic diffusivity from the data. Compare
the absolute values, the range of values, and the variability of each of these
soil moisture parameters according to the different types of soils.
2. Fit the empirical equations mentioned in the text to the data of the
last problem. Compare the ability of the different formulas to represent the
data.
3. For the soils for which you have obtained or derived data, tabulate or
graph the equilibrium moisture distribution for various depths of ground
water. Calculate the effect on this distribution of differing rates of percolation
to the ground water or evaporation from the ground water, assuming steady
state conditions.
4. Show that equation 15 for the limiting rate of evaporation from ground
water can be derived from the assumption of the relationship between con-
ductivity and soil suction in the form given by equation 8a. Derive the value
of the numerical constant in equation for n equal to 2, 3, and 4.
5. For the data used in problem 1, calculate the rate of evaporation for
different values of soil suction at the surface for three assumed depths of
ground water.
6. The text stated that for the assumption of constant diffusivity and
constant permeability, the moisture content during unsteady infiltration can
be represented as a function of x^/t. Find the form of this function. Use your
answer to find the value of the coefficient of the first term in the infiltration
equation for constant diffusivity for a number of different values of the
moisture content.
7. Derive an infiltration equation from a relationship between rate of
infiltration (or excess infiltration) and either actual or potential infiltration
(or excess infiltration) other than those mentioned in the text. Compare the
derived equation to the standard equations.
8. Compare a number of infiltration formulas. What are the assumptions
underlying the different formulas? How would you fit each of the formulas
to the data given in Appendix table 9?
9. Compare the solutions given in the literature for the steady outow of
ground water in equilibrium with a constant rate of rainfall or infiltration.
Compare the solutions for a given set of conditions, and discuss critically
which solutions you consider would be the most accurate.
10. Compare the solutions given in the literature for the recession of the
maximum water table level. Compare the assumptions made and the effect of
the assumptions on the solution. Which solution would you consider to be the
most accurate?
11. Using either a steady state solution or a water table recession solution
other than those treated in the text, derive an expression for the recession of
ground water outflow. Compare this solution with the solutions already
derived.
12. Compare the various solutions for the recession of ground water outflow.
Contrast the assumptions made and the effects of these assumptions on the
form of the solution and on its accuracy.
13. Express the one-dimensional unsteady equation for ground water
outflow in the appropriate finite difference form for setting up the problem for
solution by direct analog. Show that this formulation is equivalent to a series
of linear storage elements, each one causing backwater on the one before.
14. Show that the system of backwater storage elements derived in the
last problem can be represented by an equivalent simulation system of linear
storage elements without backwater.
15. Represent one of the unsteady state solutions by a model consisting of
linear storage elements.
293
16. List the various methods for the separation of base flow from the total
hydrograph which have been proposed in the hterature. Indicate the physical
justification, if any, for these various methods. Rank a few of the methods
which you think are most accurate in order of their probable accuracy.
Literature Cited
(1) BEAR, J., ZASLAVSKY, D., and IRMAY, S.
1968.
(2)
PHYSICAL PRINCIPLES OF WATER PERCOLATION AND SEEPAGE.

Arid Zone Res. XXIX. 465 pp., illus.
and JAEGER, J. C.
CONDUCTION OF HEAT IN SOLIDS.
CARSLAW, H. S.,
1947.
(3)
386 pp.
New York.
CHAPMAN, T. G.
1966.
EFFECTS OF GROUND WATER STORAGE AND FLOW IN THE WATER BALANCE.
Symposium on Water Resources Use and Management.

Melbourne.
(4)
UNESCO
Pp. 290-301.
CHILDS, E. C.
1936.
THE TRANSPORT OF WATER THROUGH HEAVY CLAY SOILS.
JoUr. Agr. Sci.
26: 114-141, 527-545.

(5)
1967.
SOIL MOISTURE THEORY. In Chow, Ven Te, ed., Advances in Hydroscience.

V. 4, pp. 73-117. New York.
1969.
AN INTRODUCTION TO THE PHYSICAL BASIS OF SOIL WATER PHENOMENA.

493 pp., illus. New York.
(6)
(7)
COLORADO STATE UNIVERSITY.
1963.
SYMPOSIUM ON TRANSIENT GROUNDWATER HYDRAULICS.

University, Fort Collins.
Colorado State
(8) COOPER, H. H., JR., and RORABAUGH, M. I.

1963. GROUNDWATER MOVEMENT AND BANK STORAGE DUE TO FLOOD STAGES IN
SURFACE STREAMS.
(9)
CRANK,
1956.
U.S. Geol. Survey Water-Supply Paper 1536-J, 366 pp.
J.
THE MATHEMATICS OF DIFFUSION.
(10) DEWIEST, J. M. R.
1965. GEOHYDROLOGY.
366 pp., illus.
347 pp.
Loudon.
New York.
1960.
THE ROUTING OF GROUNDWATER RECHARGE THROUGH TYPICAL ELEMENTS

OF LINEAR STORAGE. Commission OU Subterranean Waters, General Assembly of Helsinki, Internatl. Assoc. Sei. Hydrol. Pub. 52: 286-300.
(12) DuMM, L. D.
1954. DRAIN-SPACING FORMULA.
(13)
1958.
(14)
Agr. Engin. 35: 726-730.
GARDNER, W. R.
SOME STEADY STATE SOLUTIONS OF THE UNSATURATED MOISTURE FLOW

EQUATION WITH APPLICATIONS TO EVAPORATION FROM A WATER TABLE.
Soil Sei. 85(4): 228-232.

and MAYHUGH, M. S.
1958.
SOLUTIONS AND TESTS OF THE DIFFUSION EQUATION FOR THE MOVEMENT OF
WATER IN SOIL.
Soil Sci. Soc. Amer. Proc. 22(3): 197-201.

(15) GREEN, W. H., and AMPT, G. A.
1911.
(16)
STUDIES ON SOIL PHYSICS: IFLOW OF AIR AND WATER THROUGH SOILS.

Jour. Agr. Res. 4: 1-24.
M. E.
1962. GROUND WATER AND SEEPAGE.
HARR,
315 pp., illus.
New York.
(17) HOLTAN, H. N.
1961. A CONCEPT FOR INFILTRATION ESTIMATES IN WATERSHED ENGINEERING.
U.S. Dept. Agr. Agr. Res. Serv. ARS 41-51, 25 pp.
(18) HoRTON, R. E.
1938. THE INTERPRETATION AND APPLICATION OF RUNOFF PLOT EXPERIMENTS WITH
REFERENCE TO SOIL EROSION PROBLEMS. Soil Sci. Soc. Amer. Proc. 3:
340-349.
(19)
1940.
(20)
AN APPROACH TOWARD A PHYSICAL INTERPRETATION OF INFILTRATION CAPACITY. Soil Sei. SOG. Amer. Proc. 5: 399-417.
IRMAY, S.
1956.
EXTENSION OF DARCY LAW TO UNSATURATED FLOW THROUGH POROUS MEDIA.
Symposia Darcy, Internatl. Assoc. Sei. Hydrol. Pub. 41: 57-66.

(21) KiRKHAM, DON.
1966. STEADY-STATE THEORIES FOR DRAINAGE.
Irrig, and Drain. Div. 92(IR-1): 19-39.
(22)
Amer. Soc. Civ. Engin., Jour.
KLUTE, A.
1952.
SOME THEORETICAL ASPECTS OF THE FLOW OF WATER IN UNSATURATED
SOILS.
Soil Sei. Soc. Amer. Proc. 16(2): 144-150.
(23) KosTiAKOV, A. N.
1932. ON THE DYNAMICS OF THE COEFFICIENTS OF WATER PERCOLATION IN SOILS.
Sixth Commission, Internatl. Soc. Soil Sei., part A, pp. 15-21. Groningen
and Moscow.
(24)
A.
A STUDY OF NONSTEADY GROUND WATER FLOW WITH SPECIAL REFERENCE
TO A RESERVOIR COEFFICIENT. Ingenieur 70( 19): 87-94.
KRAIJENHOFF VAN DE LEUR, D.
1958.
(25)
1966.
RUNOFF MODELS WITH LINEAR ELEMENTS. PART II. RECENT TRENDS IN

HYDROGRAPH SYNTHESIS. TNO Proc. and Inform. Note 13. The Hague.
(26) LuTHiN, J. N., ed.

1957. DRAINAGE OF AGRICULTURAL LANDS.
Wis.
(27) MusGRAVE, G. W., and HOLTAN, H. N.
1964. INFILTRATION. Sec. 12, 30 pp.
Applied Hydrology. New York.
(28)
MUSKAT,
1937.
Agronomy 7, 620 pp.
Madison,
In Chow, Ven Te, ed.. Handbook of
M.
FLOW OF HOMOGENEOUS FLUIDS THROUGH POROUS MEDIA.
York.
763 pp.
NeW
(29) OVERTON, D. E.
1964. MATHEMATICAL REFINEMENT OF AN INFILTRATION EQUATION FOR WATERSHED ENGINEERING. U.S. Dept. Agr. Agr. Res. Serv. ARS 41-99, 11 pp.
295
(30) PHILIP, J. R.
1954.
AN INFILTRATION EQUATION WITH PHYSICAL SIGNIFICANCE.

Amer. Proc. 77(2): 153-157.
Soil Sci. Soc.
1957.
THEORY OF INFILTRATION. CH. 1, THE INFILTRATION EQUATION AND ITS

SOLUTION. Soil Sei. 83(5): 345-357.
1957.
THEORY OF INFILTRATION. CH. 4, SORPTIVITY AND ALGEBRAIC INFILTRATION

EQUATIONS. Soil Sei. 84(3): 257-264.
1968.
A LINEARISATION TECHNIQUE FOR THE STUDY OF INFILTRATION.
(31)
(32)
(33)
posium on Water in the Unsaturated Zone, Wageningen.

Assoc. Sci. Hydrol. Pub. 82, v, 1: 471-478.
Sym-
Bul. Internati.
(34)
1969.
THEORY OF INFILTRATION. In Chow, Ven Te, ed., Advances in Hydroscience. V. 5: 216-296. New York.
(35) POLUBARINOVA-KOCHINA, P. Y.
1952. TEORIYA DVIZHENIYA POCHVENNOI VODI THEORY OF GROUND WATER MOVEMENT]. English translation by J. M. R. deWiest. 613 pp. New Jersey.
(36) REMSON, I., and Fox, G. S.
1955.
CAPILLARY LOSSES FROM GROUNDWATER.
Amer. Geophys. Union Trans.
36(2): 304-310.
(37) RICHARDS, L. A.
1931.
CAPILLARY CONDUCTIVITY LIQUIDS THROUGH POROUS MEDIUMS.
PhysicS 1:
313-318.
(38) RUBIN, J.
1968.
NUMERICAL ANALYSIS OF PONDED RAINFALL INFILTRATION WATER IN THE

UNSATURATED ZONE. Symposium on Water in the Unsaturated Zone,
Wageningen. Bul. Internati. Assoc. Sci. Hydrol. Pub. 82, v. 1: 440-451.
(39) ScHLEUSENER, R. A., and

1959.
COREY,
A. T.
THE ROLE OF HYSTERESIS IN REDUCING EVAPORATION FROM SOILS IN
TACT WITH A WATER TABLE.
CON-
Jour. Gcophys. Res. 64: 469-475.
(40) SCOTT, E. J., and HANKS, R. J.

1962.
(41)
SOLUTION OF THE ONE-DIMENSIONAL DIFFUSION EQUATIONS FOR THE EXPONENTIAL AND LINEAR DIFFUSIVITY FUNCTIONS BY POWER SERIES APPLIED
TO MOISTURE FLOW IN SOILS. Soil Sci. 4(5): 314-321.
SMITH, W.
1967.
0.
INFILTRATION IN SANDS AND ITS RELATION TO GROUNDWATER RECHARGE.
Water Resources Res. 3(2): 539-555.

(42) ToDD, D. K.
1955.
GROUNDWATER FLOW IN RELATION TO A FLOODING STREAM.

Soc. Civ. Engin. 81: 628.
(43) VAN SCHILFGAARDE, J.

1965. TRANSIENT DESIGN OF DRAINAGE SYSTEMS.
Irrig, and Drain. Div. 91(IR-3): 9-22.
PrOC. Amer.
Amer. Soc. Civ. Eng^n., Jour.
(44) WERNER
(45) WIND, G. P.
1955.
THE FIELD EXPERIMENT CONCERNING CAPILLARY RISE OF MOISTURE IN A
HEAVY CLAY SOIL.
Netherlands Jour. Agr. Sci. 3: 60-69.
LIST OF SYMBOLS
The following list of symbols should be used as a guide to assist the reader
in recognizing commonly cited variables, rather than as an exhaustive hsting
of all symbols used. In some cases, the number of the lecture in which a symbol
is used in a particular sense is indicated in parentheses. This list does not
include symbols used once or twice in a particular sense and defined where so
used. Neither does it include symbols used to denote systems parameters or
parameters in formulas except where such parameters are the subject of
discussion.
A
A
Ak
superficial area of catchment.

cross-sectional area of channel (9).
a Fourier coefficient of output (5, 6).
B
Bk
base length of unit hydrograph (8).

a Fourier coefficient of output (5, 6).
C
C
C
Ck
capacitor in analog circuit (1,7).

coefficient of runoff (4, 8).
coefficient in Chezy formula (10).
coefficient in expansion of output.
D
D
D
D
D
unit period of rainfall excess.

differential operator (3).
dimensionless length factor (9).
hydraulic diffusivity (9, 10).
duration of recharge (10).
E
E( )
evaporation.
error criterion.
F
F
Fa
F(s)
F(co)
Fiz)
volume of infiltration capacity.

Froude number (9).
actual volume of infiltration.
Laplace transform oif{t).
Fourier transform oif{t).
Z-transform of/(Obilateral Laplace transform.
imaginary part of F{o)).
real part of F(co).
FB{S)
Fj(co)
FR{O))

[/]
H
H
H{s)
H{)
convolution matrix of system response.

available energy (2, 7).
depth of ponding (10).
system function, that is, Laplace transform oih{t).
Fourier transform of impulse response.
I
/
/i[ ]
rate of inflow to channel reach.

24-hour rainfall (7).
modified Bassel function (9).
K
storage delay time.
K
hydraulic conductivity (2, 10).
KRif). . ..R'^ cumulant of/(OL
Lea
Lo
Ln{t)
length of channel.
length to center of area (8).
length of overland flow.
Laguerre polynomial.
M
duration of continuous input (1, 5).
Mn{s).... Meixner polynomial (3, 6).
N
memory length of system.
P
P
Pd
Pe
precipitation.
duration of output (1,5).
unit pulse of duration D(l).
precipitation excess.
Q
Q'
Qb
Qg
Qi
Qmax
Qo
Qs
runoff, flow, outflow.

energy (2).
base flow.
ground water flow.
interflow.
peak discharge.
overland flow.
direct storm response (surface flow).
R
R
R
R^
RO
recharge to ground water.

resistance in analog circuit.
hydraulic radius (9).
multiple correlation coefficient (7).
runoff.
297

S
S
S
Sf
Sg
So
S{t)
storage in reach, storage.

slope.
soil suction (10).
friction slope (9).
ground slope.
channel slope.
S-curve, S-hydrograph.
T
T
T
T
Tm
transpiration (2, 7).

length of time series (5, 6).
period of repeated function (5, 6).
time of virtual inflow (8).
mean temperature (1, 7).
U{t)
unit step function.

. . . .R^^ moment oif{t) about center.
C/i( /)
R^^ moment oif{t) about origin.
U.H
unit hydrograph.
UR{)
V
V
velocity.
volume of runoff.
[X]
Xi
[X]^
X{s)
X(w)
matrix of input values.

volume of input in unit period.
transpose of X.
Laplace transform of x{t).
Fourier transform of x(t).
ak
Fourier coefficient of input.
bk
Fourier coefficient of input.
Ck
coefficient in expansion of input.
vapor pressure.
/
/
rate of infiltration capacity.

specific yield, that is, drainable pore space (2, 10).
actual rate of infiltration.
arbitrary function of time (3, 5).
n^^ order Laguerre function (3, 6).
n*^ order Meixner function.
/A
f{t)
fn(t)
fn{s)

gn{t)
gn{s)
function orthogonal under integration.

function orthogonal under summation.
h(t)
{h}
hnit)
hi
hopt
ho{t)
impulse response function.

vector if impulse response ordinates.
pulse response.
Ordinate of unit hydrograph.
optimum linear response.
instantaneous unit hydrograph (lUH).
i
i{x,t)
precipitation intensity.
rate of distributed inflow (2).
m
MR
time at start of final period of rainfall excess (m MD).

dimensionless moment.
length of finite-period unit hydrograph (n^N+D).
length of output for discrete-time {p = P).
q{x,t)
qe
Co
discharge per unit width.

equilibrium discharge.
reference discharge (9).
r{xit)
r{t)
[r]
rate of lateral inflow.

residual error (5).
vector of residual errors (6).
s
s
discrete time variable.

complex argument of Laplace transform.
shape factor, that is, dimensionless cumulant.
SR
t
tc
te
0.. .
tp
continuous time variable.

time of concentration.
time to equilibrium.
lag time.
characteristic response of flow (9).
time to peak.
u{x,t)
velocity of flow (9).
IL
299

X
X
x^
x{t)
{x}
x{sD)
y{t)
yi
{y]
y{sD)
y{Xjt).. .
distance along channel (9).

depth below surface (10).
Ordinate of input.
continuous input function.
vector of discrete inputs.
discrete input function.
continuous output function.
Ordinate of output.
vector of discrete outputs.
discrete output function.
. . depth of flow in open channel.
ak
k
dmn
Fourier coefficient of system response.

Fourier coefficient of system response.
Kronecker delta.
(f)
(t>xx{k)
<t>xx{r)
<t>xy{k)
<t>xy{.r)
hydraulic potential.
discrete autocorrelation function.
continuous autocorrelation function.
discrete cross-correlation function.
continuous cross-correlation function.
coefficient in expansion of system response.

R^^ moment.
discrete time variable.
continuous time variable.
argument of Fourier transform.
MA.
<TD
T
CO
APPENDIX TABLES
TABLE
Function
No.
1
2
1.Continuous functions
.. 1
0
.. /4
l-/4
0
.. t/b
2
3
2t
15
Range
0<<2
elsewhere
0<<2
2<<4
elsewhere
0<<2
2<<5

TABLE
1.Continuous functionsCon.
No.
Function
Range
125-75 cos ^)
-oo<<oo
125-75 cos )
0<<96
/mm+(/max-/min)sn(-)
0<<T
exp{-t/k)
0<<oo
9
10
XT )
(t/ky-êxp{-t/k)
Z
7777
{n l)lk
\\-expi-af^)
yl
^7
o!
0<<uO
-oo<<oo
0<<oo
11
--exp(-/2)
o!
0<<)
12
-h+-)exp(-/2)
0<<oo
13
^(l+?i)'^p(-i)
0<<oo
14
lO(l-Oexp(l-0
0<<1
15
^''P^^~^)~
^"^^^^
16
(a) ^(2-0 exp(9-0
0<<1
+(2+3+4)exp(l-0
-(4-Oe
e-^)
(b) ( )exp(9-0
+(ll-30-(4-0e
(122-130+335)
1^
-(t-8)3(152-4-2)
+(11-30
1<<8
301

TABLE
2.Discrete functions
No.
Function
1
2
3
4
5
6
7
8
9
10
11
(2,1)
(6,4)
(2, 6, 1)
(0, 4, 14, 8, 1,0)
(0, 4, 11, 8, 1,0)
(1,5,2,2)
(0, 2, 4, 2,0)
(0,3, 6,4,2,0)
(0, 2, 4, 3,2, 1,0)
(0, 1, 3, 3.5, 2, 0.5, 0)
(6, 22,33,26, 11,2)
12
125-75 cos 1
13..
14
15
16
17
18
19
/-A
Range
(i)
- 00 <s< oo
125-75 cos --
(s)
0<s<96
/min+(/max-/min)
Sm( -^ j)
-/min) Sin-y
()<S <T
{s/ky-êxp{-s/k)
-T^^
0<S<oo
()(y2y"
{{i)-h()Ky2y'^'^"
i(^)+2(p+(^)K)^)(+^>/2
o<s<oo
o<s<oo
0<s<oo
{n l)lk
LUX'Ay
0<s<oo
303

TABLE
Day
10
11
12
13
14
15
16
17
18
19
20
21
22
3.Daily rainfall and average daily flow, Big Muddy River,

Plumfield, III. (area 753 sq. mi.), April 1927^
Rain
Runoff
Inches
0.45
Percent
Effective rain
15
Inches
0.068
.25
15
.037
.16
.38
.25
.85
.15
1.22
.87
1.01
.51
.58
17
27
320
520
55
72
79
81
83
86
.027
.102
.080
.440
.083
.880
.690
.820
.423
.500
.05
.16
.04
.09
.03
80
81
81
82
82
.040
.130
.033
.074
.025
Unit graph
Cusecs
1,950
2,590
3,370
3,870
3,540
2,470
1,310
610
350
160
80
0
Runoff
Cusecs
(132)
(176)
(229)
(263)
(312)
(264)
(167)
(553)
(666)
(1,615)
2,116
4,201
6,188
8,580
10,229
11,481
10,924
9,375
7,355
5,196
3,500
2,377
1 SHERMAN, L. K. STREAM FLOW FROM RAINFALL BY THE UNIT-GRAPH METHOD.

News-Rec. 108: 501-505. 1932.
Engin.
304

TABLE
4.Data for Ashhrook Catchment
Date and time

(hours)
Effective rain
Storm runoff
Cusecs
Cusecs
March 26:
15
0
1,829
18
30
3,530
21
34
8,330
24
March 27:
3
6
9
12
15
18
21
24
March 28:
3
6
9
12
15
18
21
24
March 29:
3
6
9
12
15
18
21
24
980
1,320
1,290
1,280
1,160
1,040
910
790
680
580
480
390
320
280
240
210
180
155
135
115
100
85
70
65
60

TABLE
.Data for Ashbrook CatchmentCon.
Date and time

(hours)
March 30:
3
6
9
12
15
18
...
21
....
24
March 31:
3
6
Effective rain
Storm runoff
55
50
45
40
35
30
25
15
5
0
305

iCi o
TH
(N
i^
Tf
05 CO
?5
GO
CO
Ci O 00
(M
CO CD 05 CXJ
lO
11
i-H
CO
CO
^
o o ^
'~'
00
(M
Ci
Tt^
T1
(-5
CO
CO CO
O ^
O o
Oi
00
-.
O <N 05 O O O C<l
-
Oi
O
CO
OOOOOOOO
CO
^
1>
iO
o
o
o
o
CO
^
T^
i-H
o
o
o
'^ 00
CO !>
Oi 00 ^ CO CO
O r-H TH o o
CO CO lO c o
o o '-H o o
O O O c o
o o o o o
o o o o o
o o o o o
o o o o o
o o o o o
O) CO 00 CO
^
^
>0
00
^
00
T-H
^
05
lO
CO
CO
^
to
rtH
00
C005
C0COOiÔO5
l>
'^(MCO
^r^cOlC1100
COCO
OCVjTHiO
CO^COOOO^<M
COOOCOîOOi-HCOt^rHCMiOCOO
a)
t^ CO
i^
1^
(M i^ CO
CO r^ on
CO o t^
o CO
CO
lO CO O o
'"'
lO
t> CO
TtH l^
o
TtH CO i^ rt^
05 CO <N i^
1-H
Oi
00
05
,__!
CO
05
TtH
TtH
O)
CO
00 CO
r^
Oi
CO
iisiiiiiiiiiiii
5
t^
l>
CO
CO
5e
TfH
CO
Tt^
r^
CO
CO
y^
(N O
'"^
Th
f^
O
(M
c^ o
o
(M
O
o 00
o O)
CO
CO
lO
(M
Oi
^
CO
r^
CO
CO
O o
rH
^
o o
O O O O o o o o o
rH
rH
00 o
r^
on 1^ CO
05
r^
CO
o
rvl
O O o O
h-
r^
Tt^
1^
'^
ce CO
m
M^ on
r^,
^
<-)
Cf)
O' O O
'
C^05C<J05Câ50(N05(N(N05(NrH10(N(NTtH(NTtHT-HOi--lCÔ'--fCSl
OiCOC^rHOCOilOiCOrHCÔO-HCOTtlCOTH(MOO'-HCO(MOcO<MO
rHÔCOrH^CO'-HÔOrHTtHOOOfMOiOOCOOOCOO^lOOOt^lOOO
C0C^-H(MC0(M(MC0CS|THG0(MTHTHC0O'^0C0^05O^C0O'-^C0)
(MO5C0rH(MOrHCÔi
rH TH (M
TH Ttl
TH TH
-^
o
t^
rH
o
o
o
o
o
o
o
o
o 00 (N
(N o CO
(M 00
^
00
CO
00 Tt< O
Ci Oi 00
CO o
S
CO
Ci
CO
rH
C^ t^
CO rt^
CO TtH
CO
^ 00 CO
rti CO ^
i-H 1> TH
O O i-H
O O O
o o
o o o lO o o
c^ o
c^
Tt^
Gi
(M 05
lOOOÔOOÔ
lO o
lO o
(M
C^
(M
rH lO o ^
<
(V
rHTHrH000>Oi0îO000000000>0^000000
CM(N(NrHrHrHOOOOOOCOCOCOOOOOOO
rHT(THrttrtlTH
/^-1/v-irviSK,-^,-^^
(N
C^ (M C: 05 O
c^(^^cs(^^c^^cs^(Nc^c^rHrHrHrHT-l1-lrHr^rH(^^cq(^^(^^(^^(^^(^^(^^(^^
OOOOOOQOGOOOOOGOOOOOOOOOQOOOOOOOOOOOOOOOOOOOOOOOQOOOOO
IOLO)00OIOLOO>OCOCOCOCOCOCOCOCOCOCOCOCOCOCOCOCOCOCO
000000000<M(M(M(MC^<M(M(NC^COCOCDCOCOCOcOCOCO
(NC<^c^c^cqc^c^(^^<^^cocococococococooooooo*oooooo
o
rH(MCOThliOC01>.00050'-l(MCOTtliOCOI>.00050rH(MCO'^lOCO^
rHrHrHrHi-H,-HTH,-l,-H,-HC<|(MC^C^C(M(M(M

TABLE
.Characteristics of a standard 100square-mile basin}

Item
Area (square miles)

Channel slope (feet per mile). .
Ground slope
Tributary angle
Drainage density (miles per
square mile)
Length of overland flow
(feet)
Stream order
Bifurcation ratio
Length ratio
Length to center of area
(Lea) (miles)
Length of channel (L) (miles). .
Width of basin (W) (miles)...
A2
10
100
100
400
75
1,000
1-25
2,200
4-5
3
2.5
H
22
9
1 DOOGE, J. C. I. SYNTHETIC UNIT HYDROGRAPHS BASED

ON TRIANGULAR FLOW. M.S. Thesis. lowa State Univ.
June 1956.
2 These columns are to be filled in by the user or student
in working problems.
307

TABLE
7.Geomorphic parameters of a simulated

hasiv}
Characteristics of drainage pattern:

Designation
Unit watershed
Sub-sub watershed
Subwatershed
Total watershed
No.
385
49
7
1
Area
Length^
Square
miles
0.05
.35
2.75
21.35
Miles
0.33
.87
2.87
8.09
Channel sizes:
6 = 6.79^0 03
2 = 1.92^006
n=0.04
.5^ = 0.003155 A-o 20
Surface characteristics for computing overland flow:^

800 feet by 1,750 feet = 0.05 sq. mi.
Overland slope
=10.2 percent
L
=400 ft.
n
=0.2
Rising hydrograph by Morgali.
Recession linear.
Assumed conditions for channel routing:
(a) In first-order channels, translation of overland
flow to outlet of sub-subwatershed at equilibrium velocity.
(b) Initial flow, 4.9 cusecs/sq. mi.
Channel uniform between junction.
Numerical routing (rectangular grid).
^ MACHMEIER, R. E., and LARSON, C. L. THE EFFECT
OF RUNOFF SUPPLY RATE AND DURATION ON HYDROGRAPHS
FOR A MATHEMATICAL WATERSHED MODEL. Jour. Hydraulics Div., Amer. Soc. Agr. Engin. 1966.
2L~ô.568. tributary angle = 45.
3 MORGALI, J. R. HYDRAULIC BEHAVIOR OF SMALL
DRAINAGE BASINS. Tech. Rept. 30, Dept. Civ. Engin.,
Stanford Univ. Calif. 1963.

TABLE
Month
No.
1
2....
3....
4....
5....
..
. .
..
..
..
8.Data on evaporation^
Solar
radiation
Average
temperature
Average
vapor
pressure
Ly.
C.
mh.
15
20
17
24
31
734
748
532
761
625
April
April
March
June
August
309
Average
wind
4
4
8
6
22
M./sec.
2.1
1.8
1.6
1.8
1.3
Roughness
Cm,
0.001
.02
1.0
.7
3.0
1 VAN BAVEL, C. H. M. POTENTIAL EVAPORATION: THE COMBINATION CONCEPT AND ITS

EXPERIMENTAL VERIFICATION. Water Resources Res. 2(3) I 455-467. 1966.
TABLE
Time
(minutes)
0
10
20
30
40
50
60
70
80
90
100
110
120
9.Data on infiltration^
Precipitation
P-P^^'
Inches
Inches
.33
.67
1.00
1.33
1.67
2.00
2.33
2.67
3.00
3.33
3.67
4.00
.33
.63
.89
1.07
1.20
1.30
1.36
1.43
1.50
1.56
1.63
1.70
1 MusGRAVE, G. W., and HOLT AN, H. N. INFILIn Chow, Ven Te, ed.. Handbook of
Applied Hydrology. New York. 1964.
2 PPa = precipitation minus precipitated excess.
3 Average depth (D) =0.32.
TRATION.

TABLE
Day
1
2
3
4
5
10.Data on inflow and outflow^

Inflow
Outflow
Cubic feet per second Cubic feet per second

93
85
137
102
208
141
320
205
442
290
6
7
8
9
10
546
630
678
691
692
380
470
539
591
627
n
12
13
14
15
684
671
657
638
609
648
660
664
660
650
16
7
18
19
20
577
534
484
426
366
635
610
580
540
488
21
22
23
24
25
298
235
183
137
103
430
365
300
233
178
26
27
81
75
132
100
1 LAWLER, E. A. HYDROLOGY OF FLOW CONTROL. In

Chow, Ven Te, ed., Handbook of Applied Hydrology.
New York. 1964.
311

TABLE
Outflow
volume
Q (c.c.)
11.Data from experimental watershed^
Time
(seconds)
Outflow
volume
Q (c.c.)
Time
(seconds)
Outflow
volume
Time
(seconds)
Q (C.C.)
X = 1.75 cc./sec.
10
20
30
40
50
60
70
46.0
65.4
82.7
93.9
105.5
116.8
127.3
150
160
170
180
190
200
210
193.0
199.6
206.2
212.7
218.4
224.2
231.1
290
300
310
320
330
340
360
277.9
283.9
289.7
295.2
301.5
307.4
317.9
80
90
100
110
120
130
140
137.0
146.2
154.6
162.5
170.8
178.8
185.8
220
230
240
250
260
270
280
237.1
242.7
248.3
254.3
261.0
267.0
272.0
380
400
420
440
460
480
500
329.2
341.0
353.3
364.5
376.3
387.9
399.5
X-2.58 cc./sec.
10
20
30
40
50
36.6
53.5
66.5
77.1
86.9
110
120
130
140
150
130.8
135.8
141.6
147.3
150.7
210
220
230
240
250
179.4
184.2
188.8
193.2
197.5
60
70
80
90
100
95.3
104.2
111.4
117.8
124.3
160
170
180
190
200
156.2
161.3
166.5
170.5
174.6
260
270
280
290
300
202.3
206.3
210.3
214.3
218.3
See footnote at end of table.

TABLE
(Outflow
volume
Q (c.c.)
11.Data from experimental watershed}Con.
Time
(seconds)
Outflow
volume
Q (c.c.)
Time
(seconds)
Outflow
volume
Q (c.c.)
Time
(seconds)
X = 3.56 cc./sec.
10
20
30
40
50
31.3
43.8
52.6
60.7
68.1
200
210
220
230
240
134.5
137.9
141.3
144.8
148.0
390
400
420
440
460
194.8
197.7
203.2
208.9
214.7
60
70
80
90
100
110
120
130
140
150
74.4
80.4
85.7
90.8
95.5
100.0
104.3
108.5
112.6
116.5
250
260
270
280
290
300
310
320
330
340
151.5
154.4
157.5
160.6
163.7
166.7
170.1
173.1
176.4
179.4
480
500
520
540
560
580
600
620
640
660
220.4
226.1
231.7
237.3
242.8
248.2
253.9
259.5
265.1
270.6
160
170
180
190
120.2
124.1
127.5
130.9
350
360
370
380
182.5
185.5
188.7
191.8
680
700
720
740
276.2
282.1
287.9
293.8
^ AMOROCHO, J., and ORLOB, G. T. NONLINEAR ANALYSIS OF HYDROLOGIC SYSTEMS.

Water Resources Center, Contrib. 40, 130 pp. Univ. Calif., Berkeley. 1961.

TABLE
No.
1 IZZARD, C. F.
12.Data for overland flow^
Length
Slope
Feet
12
72
72
72
72
72
Feet per foot
1.
2.
3.
4.
5.
6.
313
0.0001
.005
.04
.02
.04
.04
Surface
Rain
Inches per hour
asphalt
crushed slate
....do
turf
....do
....do
3.65
3.67
3.66
1.89
3.60
1.89
HYDRAULICS OF RUNOFF FROM DEVELOPED SURFACES.
Bd. (Washington, D.C.) Proc. 26: 129-146.
TABLE
Highway ReS.
1946.
13.Experimental data for overland flow
[Rising hydrograph: 0-2 min., = 1.89 in./hr.; 2-7 min., t = 3.78 in./hr.; = 1.94]
Time
Runoff
Time
Runoff
Minutes
Inches per hour
Minutes
Inches per hour
.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
7.0
.015
.095
.32
.61
1.13
2.04
2.80
3.27
3.52
3.67
3.78
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
0
.022
.071
.139
.224
.326
.441
.570
.712
.866
1.029
1.198
1.367
1.529
1.674
1.793
1.880
1.934
1.957
(M
Q,
Pi
^
O
O
O
O
O
2^ O 0
I
T3
Ci
O
o
Ci
.S o
4J
O)
Pi
5e
oioooo
ooooo
o o
O^t^^
(MOOlO
1-Hr-HOO
oooo
O'-lCOOO
OO(M^00
rH^r-H^^
CO'îOOCO
0000CX)C5Oi
CO CO (M (M
lO lO 00 00
lOÔiCi
oooo
oooo
oooo
rH 00
CO^COCO'^
T-i00(MC^LO
OOCOCO'-H
ooooo
o t^ Oi ^
OOCOCiTtHOi
T^THT-IILO
ÔOOO
ooooo
(N(NOO
oooo
oooo
OOCOrH
COOCDOO
oooo
I000i00
rtÔ'-lTtî-i
OOOOiOiO
o iotooio
o rHTt^rHrt^
T-i (MTt^^t^
^
oooo
OrHCSCO
00000000
fOCO'^CO
1OCO00O5
oooo
CO
O
O
o
o
o
o
^ o ^
t
2
oi>oo
ooooo
I> TH o
OOi-H^-OCO
fO>OCD(NTH
OOO'-H'-H
-IOCOO
OT-IC^)-^
(M (N (M (N
03
00
O
T-l
005l>
COr-H^TIGO
OOOO^CO
O'-HOOO
oooo
i3
0500t>
O^^t^
^^^C^
OOOO
i3
(M
(N
OOOOT-H
CO
rH
CO
o
r-H
l>
lO
T^
LO
o
lO
88
C^ lO T^ rH
CO (N o CO
o ^ o o
oooo
CD
^
o c
o c
^ rH o ^
c
o o o
o
r^
00
o
(NOO-i'-<
Oi-HNt^l:^
'^lOOQ
^1>.0000
O-î-HiOO
050i^t^t^
OOrHrHi-^
i-Hi-HMCa
OOOOOOOO
o ^ CO
UO C5 o
Pi
3
TH
tc3
rH
Tti
CO
lO
TH
o o
ooooo
lO
(M
00
o
OOOrHi-iTti
COOOOO
(Mt^TtÔOM
i-iO<N(NCO
(M ^ (M C^ t^
O'-HOOO
lO O CO !> !>
(N t^ (M t^ C<1
O (M CO TtH O
00 00 00 00 05
o CO
TtH UO
o t^
CO
o
OO
o
o
05
o
o
CQ
Oi
o
o
o
o
T-^
hi
03
<
CO
OJ
0)
Pi
Pi
t
03
(M
CO
T^
CO
O
TH
O O
OOOOO
OOOOO
T-HT-Î-H^
t^
00
CO
O
1>
t^
(N
O
SSe*-*
O
t^
rH
O
"^ Oi rft rfi

00 rt^ 00 00
(N (N (N (N
8888
O O O O
O O O
GO
I
o o o
o CO o
"^ o ^
o
o o o o
(Z> CO <> Oi ^
rt rH ^ T-( Cq
o 1-t (N
05
1I 1-1 00 Tfl CO
00 T^ (N O CD
(M (M 1-^ rH O
O O O O O
O lO rH 00
rJH r}^ O 05
CO (M 1-H O
O O O O
O O O O
05 o rH (M '^
rH (M (M C<l (N
Oa5l>OCO
GSOSiOOOCO
O^COMCO
CO
S8
CO
00
lO
1-1
CO
1-1 1-1 1-1 (M <N

00 00 00 00 00
CO CO
O CO
lO O
CO CO
88
88
COrHCOr^lM
THCOt>.cO'^
(MT-<000
05 CO 1-H 1> rH
C3 CO Tt< iC <N
OO 00 00 00 05
88
C<1
0^ ^ D
O CO rH
8883
rH
^
o
1-1
lO
o
o
(N
^
rfH
o
(M
1-t
IQ
(N
rH
(N
o
o
CO
<M
CO
'-H
11 88
yi
rH
1-^
CO (N CO CO
O ^
o o
88
1-H
1
8
(M T< (N (N
o o
t-H
O O O O O
rti
315
316
AUTHOR INDEX^
Abbott, M. B., 262
Abramowitz, M., 58, 73
Ackerman, W. C, 12, 40
Ackoff, R. L., 3, 40
American Society of Civil Engineers, 97, 98, 196,
Amorocho, J., 17, 40, 96, 98, 150, 185, 242, 262
Ampt, G. A., 276, 279, 294
Anderssen, A., 121, 125
Anderson, E. A., 162, 185
Aseltine, J. A., 12, 40, 64, 73, 125
Ashley, J. R., 177, 185
Bagley, J. M., 179, 188
Bakhmeteff, B. A., 52, 54
Barnes, B. S., 94, 98, 227
Barnes, J. L., 20, 41^ 108, 126, 227
Bayazit, M., 35, 122, 125, 146
Bear, J., 48, 54, 179, 182, 185, 267, 269, 284, 293
Becker, A., 156-159, 185
Beer, S., 40
Behlke, C. E., 262
Beightler, C. S., 188
Bellman, R. E., 40, 105, 112, 125
Bender, D. L., 206, 227
Benson, M. A., 151, 153, 185
Bernard, M. M., 86, 98, 204, 227
Betson, R. P., 227
Bickley, W. G., 70, 73
Bittinger, M. W., 286
Blackwell, W. A., 3, 41
Blaney, H. F., 46, 54
Blank, D., 113, 125
Body, D. N., 33, 35, 40, 94, 98, 143, 146
Boersma, L., 189
Boughton, W. C, 169, 185
Bouwer, H., 179, 185
Bowen, I. S., 54
Bowers, S. A., 187
Brakensiek, D. L., 208, 227, 262
Brater, E. F., 97, 101, 266
Burdoin, A. J., 191, 231
Carlslaw, H. S., 286, 293
Cermak, J. E., 186
Chadwich, D. G., 188
Chamier, G., 98, 193, 227
Chapman, T. G., 291, 293
1 This index cites ail references to authors. An italic page number indicates that details
of the reference are to be found on that page.
317
Chemerenko, E. P., 186, 263

Chery, D. L., 183, 186
Childs, E. C, 54, 267, 273, 274, 276, 293
Chorafas, D. N., 35, 40, 148, 186
Chow, V. T., 43, 54, 98, 186, 211, 227, 263
Clark, C. O., 191, 192, 197, 203, 217, 228
Clark, J. I., 228
Close, C. M., 4^
Coleman, G. S., 81, 98, 193, 228
Collins, W. T., 30, 40, 93, 99
Colman, E. A., 40, 194
Colorado State University, 54, 179, 183, 186, 293
Committee for Hydrological Research, TNO, 54
Commons, C. G., 191, 206, 217, 228
Cooper, H. H., 291, 293
Corbett, D. M., 43, 54
Corey, A. T., 272, 295
Coulter, C. G., 217, 218, 228
Crank, J., 286, 293
Crawford, N. H., 40, 162, 166, 180, 185, 186, 187
Criddle, W. D., 46, 54
Cross, W. P., 85, 97, 99, 187, 229
Dalton, J., 54
Dawdy, D. R., 38, 4I, 168, 174, 186
DeCoursey, D. G., 208, 228
Delleur, J. W., 113, 125
DeRusso, P. M., 12, 4I
Desoer, C. A., 12, 42
DeWiest, J. M. R., 48, 54, 284, 293
Deymie, P., 245, 263
Diskin, M. H., 32, 112, 125, 182, 186, 212
Doebelin, E. O., 3, 4I
Doetsch, G., 64, 73
Dooge, J. C. I., 17, 32, 4I, 43, 54, 81, 96, 97, 99, 105, 117, 126, 136, 190, 193, 199, 210,
217, 221, 228, 242, 245, 246, 247, 248, 263, 291, 293
Dorn, W. S., 70, 74
Draper, C. S., 3, 4I
Drenick, R. F., 3, 4I
Dronkers, J. J., 263
Dumm, L. D., 286, 293
Eagleson, P. S., 35, 4I, 43, 44, 45, 54, 122, 126, 138, 140, 144, I46, 186
Eaton, T. D., 198, 217, 228
Edson, C. B., 206, 207, ^^5
Ellis, D. O., 3, 41
Erdelyi, A., 65, 73
Escoffier, E. F., 263
Escritt, L. B., 81, 83, 99, 194-196, 228
Espildora, B., 160, 185
Ezekiel, M., 151, 186
Farrell, D., 192
Fiering, M., 149, 186
Fodor, G., 74
Forsythe, G. E., 263
318
Foskett, L. W., 187

Foster, E. E., 97, 99
Fox, G. S., 270, 271, 295
Fox, L., 67, 74Fox, K. A., 151, 186,
Gardner, M. F., 20, 4^, 108, 126
Gardner, W. R., 270.3 293
Garstka, W. U., 44, 54
Gibbs, A. E., 187, 188
Gilcrest, B. R., 263
Oilman, C. S., 44, 54
Glover, R. E., 286, 288
Grace, R. A., 183, 186
Gray, D. M., 53, 54, 207, 228
Green, R. F., 227
Green, W. H., 276, 279, 280, 294
Guillemin, E. A., 58, 59, 62, 70, 74, 126
Gupta S. C, 12, 41
Gwinn, W. R., 265
Hamming, R. W., 62, 70, 74
Hanks, R. J., 179, 187, 276, 295
Harbeck, G. E., 45, 55
Harder, J. A., 179, 187
Harding, S. T., 55
Harley, B. M., 245, 246, 248, 257, 263
Harman, H. H., 155, 187
Hrr, M. E., 284, 294
Harris, B., 187, 188
Hart, W. E., 17, 40, 185
Harte, C. A., 81, 99, 193, 228
Hathaway, G. A., 263
Hawken, W. H., 81, 99, 194, 228
Hayami, S., 52, 55, 244, 263
Hedke, C. R., 46
Henderson, F. M., 50, 52, 55, 193, 229, 239, 263, 264
Hildebrand, F. B., 62, 70, 74
Holtan, H. N., 48, 55, 56, 272, 273, 294
Holzman, B., 45, 56
Horton, R. E., 47, 51, 55, 197, 229, 234, 236, 243, 264, 272, 273, 294
Hoyt, W. G., 75-78, 99
Huggins, L. F., 187, 264
Hullinghorst, D. W 204, 217
International Association of Scientific Hydrology, 55
Irmay, S., 48, 54, 185, 267, 269, 284, 293, 294
Iseri, K. T., 45, 55
Issacson, E. J., 264
Iwagaki, Y., 264
Izzard, C. F., 51, 55, 234, 237, 243, 264
Jaeger, J. C, 286, 293
Jens, S. W., 99, 193, 229
Johnson, A., 81, 98, 193, 194, 228
Johnson, S. M., 218, 229

Johnstone, D., 85, 97, 99, 187, 198, 217, 229
Judson, C. C, 81, 99, 193, 229
Jury, E. I., 64, 7^, lJf6
Kalaba, R. E., 125
Kalinin, G. P., 52, 53, 55, 180, 187, 2U, 256, 257, 26J^
Kartvelishvili, N. A., 149, 179, 186
Kaufman, H., 65, 74
Kendall, M. A., 115, 126
Keulegan, G. H., ^^4
Kirkham, D., 285, 29J^
Kirpisch, Z. P., 229
Kiwata, H., 187
Kleitz, C, 26Jf.
Klute, A., 274, 294
Koenig, H. E., 3, U
Khler, M. A., 43, 56, 78, 94, 97, 99, 155, 157, 158, 187, 229
Koloseus, H. J., 50
Korn, G. A., 58, 74
Korn, T. M., 58, 74
Kostiakov, A. N., 47, 55, 280, 281, 294
Kraijenhoff van de Leur, D. A., 17, U, 286, 288, 294
Kritskii, S. N., 208, 229
Kuchment, L. S., 96, 99, 229, 264, 265
Kuichling, E., 79, 99, 193, 229
Kulandaiswamy, V. C, 14, 146, 212, 265
Lambe, C. G., 67, 74
Lanczos, C, 62, 74
Langbein, W. B., 45, 55, 206, 229
Larriev, J., 96, 99
Laten, N. R., 229
Laurenson, E. M., 229, 265
Lee, P., 138, 146
Lee, Y. W., 3, 41, 121, 126, 146
Lees, S,, 41
Levi, E., 32, 41, 113, 126
Levin, A. G., 187
Liggett, J. A., 240, 241, 243, 265, 266
Light, P., 44, 55
Lighthill, M. J., 265
Lin, Pin-Nam, 265
Linsley, R. K., 40, 43, 56, 78, 94, 97, 99, 155, 166, 180, 186, 187, 205, 229, 241, 266
Little, W. C, 185
Lloyd-Davies, D. E., 80, 100, 193, 229
Lockett, J., 125
Ludwig, F. J., 3, 41
Luthin, J. N., 48, 56, 284, 285, 294
Lynch, W. A., 3, 42
Maass, A. A., 149, 187
McCarthy, G. T., 204, 244
McCool, D. K., 265
McCracken, D. D., 70, 74
MacFarlane, A. G. J., 3, 42
319
320
McKay, W., U
McNown, J. S., 266
Mamisao, J. P., 18S
March, F., U, 126, 1^6
Maruyama, F., 192, 207, 209, 231
Massau, J., 265
Masse, P., 245, 265
Mtalas, N. C, 155, 187
Mayhugh, M. S., 270, 293
Meinzer, O. E., 97, 100
Mejia-r, R., U, 126, 146
Menkel, M. F., 208, 229
Meyer, O. K., 56, 244, 265
Mikawa, H., 192, 207, 230
Milyukov, P. L, 244, 256, 257, 264
Minotte, J. S., 229
Minshall, N. E., 243, 265
Mitchell, W. D., 205, 218, 229
Monke, E. J., 187, 264
Morgali, J. R., 241, 265, 266
Morgan, P. E., 218, 229
Morgan, R., 204, 217, 218
Mulvany, T. J., 80, 100, 193, 229
Munro, C. H., 81, 100, 190, 194, 195, 230
Musgrave, C. W., 56, 294
Muskat, M., 284, 294
Nash, J. E., 17, 31, 42, 52, 56, 82, 83, 92, 96, 100, 105, 113, 115, 116, 126, 190, 192, 202,
207, 208, 209, 214, 219, 220, 230, 266
Negev, M., 166, 187
Newton, D. W., 143, 146
O'Donnell, T., 32, 38, î, 4^, 103, 104, 108, 126, 133, 143, 146, 168, 186
Ogrosky, H. O., 40
O'Kelly, J. J., 192, 199, 217, 230
O'Meara, W., 259
Orlob, G. T., 40, 96, 98, 242, 262
Ormsby, M. T. M., 81, 100, 193, 230
Otaba, K., 169, 187
Overton, D. E., 56, 272, 273, 281, 294
Owen, W. J., 187, 188
PapouUs, A., 64, 74, 104, 108, 126
Paulhus, J. L. H., 43, 56, 78, 94, 97, 99, 155, 187, 229
Paynter, H. M., 3, 17, 31, 4^, 112, 126, 187, 266
Penman, H. L., 46, 56, 162, 188
Petterssen, S., 44, 56
PhiHp, J. R., 48, 56, 273, 274, 276, 277, 278, 280, 294, ^95
Polubarinova-Kochina, P. Y., 284, 295
Prasad, R., 213, 230
Ragan, R. M., 266
Rainville, E. D., 62, 74
Rainville, E. D., 62, 74
Ralston, A., 71, 74
Raven, F, H., 58, 70, 74

Ree, W. O., 265
Reich, B. M., 53, 56, 207
Reid, J., 81, 100, 194, 230
Reiher, B. J., 187
Reilly, T. L., 229
Remson, I., 270, 271, 295
Restrepo, J. C. O., 126, 146
Richards, L. A., 271, 295
Richards, M., 158, 188
Richtmyer, R. D., 266
Riley, D. W., 81, 100, 194, 230
Riley, J. P., 178, 188
Roberson, J. A., 206, 227
Roberts, G. E., 65, 74, 108, 126
Roche, M., 96, 100, 147
Rorabaugh, M. I., 291, 293
Rosenbrock, H. H., 173, 188
Ross, C. N., 81, 99, 100, 190, 193, 194, 228, 230
Rousculp, J. A., 81, 100, 193, 230
Roy, R. J., 41
Rubin, J., 274, 295
Sato, S., 192, 207, 230
Schleusener, R. A., 272, 295
Schonfeld, J. C., 266
Schulze, F. E., 158, 188
Schwartz, H. E., 205, 217, 231
Schwarz, J., 185
Scott, E. J., 276, 295
Seddon, A. J., 247, 266
Sellers, W. D., 45, 56
Sharp, A. L., 157, 187, 188
Shen, J., 38, 4^, 179, 188
Sherman, L. K., 25, 4^, 84, 93, 100, 203, 230
Shipatani, K., 187
Siebert, W. M., 19, 21, 4^
Singh, K. B., 147, 211
Skibitzke, H. E., 179, 188
Slatyer, R. O., 45, 56
Smith, W. O., 295
Sneddon, I. A., 67, 74
Snyder, I. A., 67, 74
Snyder, F. F., 204, 205, 218, 230
Snyder, W. M., 33, 35, 4^, 94, 101, 147, 151, 155, 188
Stall, J. B., 155, 188
Stark-, L., 3, 4^
Stegun, I. A., 58, 73
Stoker, J. J., 264, 266
Stone, L. N., 189
Strahl, J. A., 188
Strahler, A. N., 203, 231
Stuart, A., 126
Sugawara, M., 168, 169, 188, 192, 207, 209, 231
321
322
Taylor, A. B., 205, 217, 231

Thomas, H. A., 250, 266
Thompson, T. H., 174, 186
Thompson, W. A., 70, 73
Thornthwaite, C. W., 45, 46, 56
TNOi
Todd, D. K., 48, 56, 291, 295
Tranter, C. J., 67, 74
Troesch, B. A., 264
Truxal, J. G., 3, 42
Turc, L., 47, 57
Turner, H. M., 191, 231
Tustin, A., 3, 42
Tennessee Valley Authority, 169, 188, 231
U.S. Bureau of Reclamation, 206, 231
U.S. Corp of Engineers, 57, 205, 231
U.S. Department of Agriculture, 57
U.S. Department of Agriculture, Soil Conservation Service, 191, 206, 217, 218, 231
University of Wageningen, 57
Valdes, R., 32, 41, 113, 126
van Bavel, C. H. M., 57
van Schilfgaarde, J., 48, 57, 285, 295
Vasiliev, O. F., 266
Veihmeyer, F. J., 45, 57
Vinyard, J. W., 143, I46
Wade, T. L., 70, 74
Wasow, W. R., 263
Whitam, G. B., 265
Wilde, D. J., 173, 188
Wilf, H. S., 71, 74
Williams, E. J., 151, 188
Wilhams, G. R., 101
Williams, H. M., 191, 206, 217, 231
Wind, G. P., 270, 295
Wisler, C. O., 97, 101
Wong, S. T., 189
Woo, D. C, 189, 266
Wooding, R. A., 240, 264, 266
Woolhiser, D. A., 240, 241, 243, 265, 266
World Meteorological Organization, 57
Wu, I-Pai, 207, 208, 215, 231
Wymore, A. W., 12, 42
Yevjevich, V. M., 50, 57, 151, 189, 243, 266
Yu, Y. S., 266
Zadeh, L. A., 12, 42
Zaslavsky, D., 48, 54, 185, 267, 269, 284, 293
Zoch, R. T., 191, 196, 197, 231
1 TNO stands for ''Organisatie Toegepaft Natuurwetenschappelijk Onderzoek" [Netherlands Organization for Applied Scientific Research]. The abbreviation TNO is invariably
used, and the full title does not appear even on publications of the organization.
SUBJECT INDEX
Analog, direct, 178-182
Analog, indirect, 176-178
Analog models, 150, 175-183, 184
Analysis, 10-11
Autocorrelation function, 34, 121, 139, 145
Backward substitution, 131
Bank storage, 291
Barnes method of unit hydrograph derivation, 94-95, 98
Base flow, 16, 49
Base flow separation, 89-93, 98, 289, 292
Catchment as a system, 16-17, 165
Catchment characteristics, 190, 201, 203-204, 209, 214, 222
Catchment model, 37, 165-169
Catchment parameters, 222-223
Catchment response, 165-169, 184, 227
Causal system, 10, 23, 26, 103
Coaxial correlation, 155-160
Collins' method of unit hydrograph derivation, 30, 93, 98
Component, 8, 149
Components of hydrograph, 78
Computational methods 141-144
Continuous convolution, 20-25
Conceptual model, 36-37, 192-193, 209-215, 232, 267, 283
Continuous input (or output), 8
Continuous system, 8
Convolution, continuous, 20-25, 82, 95-96, 102, 195
Convolution, discrete, 25-29, 128
Correlation methods of identication, 33-35
Cross-correlation coefficient, 34, 121, 139, 145
Cumulants, 115-117, 124, 214, 238, 247
Cumulants, dimensionless, 116
Curvilinear regression, 154-155
Darcy's Law, 48, 267
Dawdy-O'Donnell model, 38, 168
Detection of signal, 11
Deterministic hydrology, 1
Deterministic systems, 1, 9
Delta function, 20-21
Differential equations, ordinary, 19, 65-67, 72
Diffusion analogy (flood routing), 52, 253
Diffusion equation (unsaturated flow), 269
Digital simulation, 38, 160-168
Direct analog, 178-182
Direct storm runoff, 15, 165, 190
Discrete convolution, 25-29
Discrete input (or output), 9, 25
Discrete system, 9
Distributed parameters, 9
Distribution graph, 26, 85, 128
323
324
Dupuit-Forcheimer assumption, 48, 284, 285, 289

Dynamic hydrology, 1
Effective precipitation, 15, 91
Ellipse equation, 285
Error-sensitivity, 98, 144-145
Evaporation, 15, 45-46, 184
Factor analysis, 204, 223
Feedback, 9, 17
Flood routing, 7
Forward substitution, 33, 94, 130-131, 142, 144
Fourier coefficients, 60, 71, 104, 105, 107, 124, 133
Fourier methods, 20, 60, 62, 104-105
Fourier series, 59-60, 106-107
Fourier transform, 32, 63, 71, 104, 109-112, 124
Frequency analysis, 24
Gamma distribution, 52, 120, 192-193, 206-208, 210, 215
Geomorphology, 204, 217, 223-224
Gibbs oscillations, 127
Glover solution, 286
Granular model, 182
Green-Ampt infiltration formula, 276-277
Green's function, 22
Ground water flow, 48-49, 284-290, 292
Ground water hydrology, 1
Ground water reservoir coefficient, 286
Ground water response, 17, 91, 185, 283-290
Harmonic analysis, 32, 71, 133-134, 143, 145, 146
Hele-Shaw apparatus, 181
Histogram input, 25
Horton-Izzard solution, 51-52, 234-238, 261
HydrauUc diffusivity, 268, 270, 291
Hydrologie cycle, 12-14, 53, 184
Identification by trial and error, 29-30
Identification of system, 10, 22, 29-35, 40, 102-146
Impulse function, 21-22
Impulse response, 21-22, 102, 124
Indirect analog, 176-178
Infiltration, 15, 47-48, 89, 272-282
Infiltration- capacity, 272
Infiltration formulas, 272, 292
Influence function, 22
Input, 4, 9-10
Input, discrete, 25
Input, distributed, 22
Input, isolated, 24, 26, 102
Inputs, multiple, 22
Input, periodic, 24-25, 104
Input, quantized, 25-26
Input, sampled, 27
Instantaneous unit hydrograph (lUH), 22, 40, 81-82, 87, 95, 195-196
Inversion, 11
Inversion of transform, 103, 110, 112, 115
Isochrones, 81, 96, 195

Joint regression, 155
Kalinin-Milyukov method, 52, 256-257, 261
Kernel function, 22
Kinematic wave solution, 239, 261
Kleitz-Seddon Law, 247
Kronecker delta, 59, 60
Lag and route method, 52, 255-256, 261
Lag time, 202, 224
Laguerre analysis, 32, 43, 71, 117-121, 136
Laguerre coefficients, 118, 124
Laguerre functions, 61, 71, 105, 116
Laplace transform, 20, 63-64, 66-67, 71, 246-247
Laplace transform method of identification, 31, 104-105, 111-113, 124, 135
Least squares and time series analysis, 122, 140-141
Least squares method of identification, 33, 73, 94, 131-133, 145
Least squares optimization of parameters, 170
Limiting rate of evaporation, 272, 291
Linear channel response, 246-248, 261
Linear programing, 143
Linear theory, 1, 8, 16-18, 40, 49, 85, 283
Linearized system, 211, 245, 285-286
Linkage equation, 103
Linkage equation (Fourier and Laplace transforms), 109, 111, 135
Linkage equation (Fourier series analysis), 107, 134-135
Linkage equation (Laguerre analysis), 119, 120
Linkage equation (moments and cumulants), 114-115, 116
Lumped parameters, 9, 19
Lumped system, 9, 15, 19
Matrices, 67-70
Matrix form of convolution, 28-29, 128-129
Matrix methods, 72-73, 98
Meixner functions, 62, 71, 137, 143, 145-146
Memory, 8, 23, 26, 106
Model, 148, 149
Model, conceptual, 148
Moments, 62, 70, 71, 105, 113-115, 124
Moments as catchment characteristics, 31, 203
Moments, dimensionless, 116, 210, 220-221, 224
Moments, matching, 170-171, 213, 287
Morphology of catchments, 204
Multiple inputs, 22
Multiple Hnear regression, 151-153
Multivariable analysis, 155
Muskingum method, 40, 52, 254, 257, 261
Muskingum model (generalized), 212-213, 227
Nonhnear system, 17, 35, 85, 232
Nonlinear theory, 1, 8, 50-51
Numerical methods, 70, 73, 241, 244, 260-261
Open channel flow, 243-260
Optimization, 38, 169-174
Optimum linear response, 122
Ordinary differential equations, 19, 65-67
Orthogonal functions, 19, 58-64, 71
325

Output, 9-10
Overland flow, 50-52, 232-242
Parameter optimization, 83, 149, 150, 166, 169-174
Parameters, significance of, 172
Parametric hydrology, 1, 83
Partial differential equations, 67, 72
Pearson type I distribution, 224
Pearson type III, distribution, 191, 224
Pearson type V, 208, 224
Percolation, steady, 271
Philip's infiltration theory, 277-278
Physical scale models, 183
Polynomial regression, 154-155
Precipitation, 15, 44-45
Precipitation excess, 15, 89, 190
Prediction, 11, 21, 39, 95
Probabilistic systems, 9
Pulse, 25
Pulse response, 25, 73, 127, 142, 145
Quantized input (or output), 9, 25
Rational method, 79-84, 95, 190
Realizable system, 10
Recession, 78, 89-92, 237, 239, 261, 292
Reference discharge, 250-251
Regression models, 151-159, 184
Reservoir, Unear, 72, 176, 192, 207, 210, 215-231
Reservoir, nonlinear, 213-238
Response time, 15
Rising hydrograph, 236, 239, 261
Routing a flood, 7, 52-53, 244
Run-off hydrograph, 75-78
Saturated zone, 17
Scale models, 182
Shape factors, 116, 223, 227
S-hydrograph, 39, 86
Signal detection, 11
Simulation, 7, 20-38, 43, 148-189
Simulation, analog, 39
Simulation, digital, 38
Simultaneous equations, 93-94
Snyder's synthetic method, 205-206
Snowmelt simulation, 160-162, 185
Soil moisture characteristic curve, 268
Soil phase of cycle, 267-282
Soil storage, 17
Stable system, 10
Standard catchment, 217-218
Stanford model, 37-38, 166-167
State of a system, 8, 13-14
Stochastic systems, 149, 208
Storage elements, linear, 72, 165
Subsystem, 8
327
Superposition, 7, 8, 17-19, 82, 84, 102

Superposition, integral, 23
Surface water hydrology, 1
Synthesis, 7, 12, 22, 148
Synthetic scheme (general), 219, 220-221
Synthetic unit hydrograph, 82, 149, 190-231
System, definition of, 3-4
System identification, 10-11, 22, 29-35, 39
System operation, 4-7, 10
Systems approach, 2-7, 17, 39, 43, 77, 96
Systems terminology, 8, 40
Terminology, systems, 8, 40
Threshold in system, 15, 17, 158, 165, 175
Time-arep-concentration curve, 81-82, 192, 195, 197, 214
Time-area-diagram, 80-81
Time-area methods, 80-82, 95, 97, 190, 193-199
Time-invariance, 8-9, 19, 40, 85, 102
Time-invariant system, 8-9, 18
Time of concentration, 80
Time scale in Laguerre analysis, 120
Time series analysis, 33-35, 120, 138-141, 143-144
Time transformation, 208, 223
Transform methods of system identification, 30-33, 102-120, 133-138, 143
Transpiration, 15, 46-47, 162-164, 184
Truncation of unit hydrograph, 220, 227
Uniform nonUnearity, 242-243
Unit hydrograph, 6-7, 10, 15, 16, 39, 184
Unit hydrograph concepts, 84-88
Unit hydrograph derivation (classical), 29-30
Unit hydrograph, finite period, 25-26, 27, 73, 87, 95, 97, 127, 135
Unit hydrograph, instantaneous, 39, 83, 87, 93, 95, 97, 135
Unit hydrograph parameters, 202-203
Unrealistic ordinates, 143, 145
Unsaturated permeability, 269-270
Unsaturated zone, 1
Verification of model, 149
Virtual channel inflow, 197
Wave equation, linear, 72, 73
Wiener-Hopf equation, 124, 140, 144, 145-146
Z-transform, 64, 71, 125, 145-146
^ U.S. GOVERNMENT PRINTING OFFICE: 1973 0491-470

Livro - Linear Theory of Hydrologic Systems (Dooge, 1988)

Uploaded by

Copyright:

Available Formats

Livro - Linear Theory of Hydrologic Systems (Dooge, 1988)

Uploaded by

Document Information

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

Livro - Linear Theory of Hydrologic Systems (Dooge, 1988)

Uploaded by

Copyright:

Available Formats

N

Technical Bulletin No. 1468

Agricultural Research Service

Technical Bulletin No. 1468

Agricultural Research Service

Issued October 1973

For sale by the Superintendent of Documents, U.S. Government Printing Office

Trade names are used in this publication solely for

LECTURE 6: IDENTIFICATION BASED ON DISCRETE DATA 127

LECTURE 8: SYNTHETIC UNIT HYDROGRAPHS

Types of Synthetic Unit Hydrographs

2 TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

4 TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

LINEAR THEORY OF HYDROLOGIC SYSTEMS

1-1.Factors affecting output.

TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

1-2.The concept of system operation.

LINEAR THEORY OF HYDROLOGIC SYSTEMS

TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

LINEAR THEORY OF HYDROLOGIC SYSTEMS

TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

unanticipated changes in the external environment. Beyond feedback we have

where h{t) is a mathematical function characterizing the system operation

LINEAR THEORY OF HYDROLOGIC SYSTEMS

PROBLEMS ARISING WITH SYSTEMS

1-3.Classification of systems problems.

TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

problem of system identification or signal detection is that of unscrambling

LINEAR THEORY OF HYDROLOGIC SYSTEMS

1-4.Representation of the hydrologie cycle.

TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

1-5.Block diagram of the hydrologie cycle.

LINEAR THEORY OF HYDROLOGIC SYSTEMS

system. Whereas precipitation is clearly an input and runoff an output, it is

1-6.Kulandaiswamy's block diagram.

TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

FiGUBE 1-7.The catchment as a system.

1-8.The simplified catchment modeL

LINEAR THEORY OF HYDROLOGIC SYSTEMS

Linear Time-Invariant Systems

TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

then the system is said to be Unear if:

LINEAR THEORY OF HYDROLOGIC SYSTEMS

or in terms of an exponential series:

TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

possible. Consequently, one set of orthogonal functions may be preferable

LINEAR THEORY OF HYDROLOGIC SYSTEMS

mathematical definition of the delta function:

Any arbitrary input x{t) can be considered as being made up from an

The right-hand side of this equation represents the well-known mathematical

TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

problem of prediction now becomes the solution of equation 11 for known

LINEAR THEORY OF HYDROLOGIC SYSTEMS

so that the system operation is defined by:

2/(0= /' x{T)h{t-T) dr

and equation lie can be written as:

TECHNICAL BULLETIN NO. 1468, U.S. DEPT. OF AGRICULTURE

and equation 16b will be modified to read:

and equation 16b to:

LINEAR THEORY OF HYDROLOGIC SYSTEMS