INSTITUTE

OF
HYDROLOGY

OFPHYSICALLY.BASED
ONTHEROLE
D I S T R I B U TI iE{ O
DDELLING
HYDROLOGY

by
K , l B E V EaNn d P E O ' C O N N E L L

ASSTRACT

of hydrotoglcal
modeuinq is
Ihe rationale
of various
allscused, and the capabllitLes
of
noalels are conrsideled.
The potential
physicallfbased
dlstrlbuted
nodelling
in
hyaLology 1s noted sith lefereoce to speciflc
areas, atrd the current status of
applicatic'n
nodelling
assessed.
The role of
distrlbuted
Europ6en (sHE) in
the Systlme sydrologique
developlng a qeneral methodology for
physj.cally-based
distrj.buced
catchnent
rnodell.lng is aieflned-

REPORT
NO81
August 1982

CONTENTS
Page
ON THE RATIONALE OF IIYDROIGICEI

!'IODEITING

ON TITE CHOICE CF A I.IATHE}IATICAI, IIODEL

ON TIIE NEED FOR PHYSICAITIY-BASED

IN HYDROI,OCY
3.1
3.2
3.3
3.4

DISTRIBUIED

MODEI,S

catclment
changea
q)actally
variable
inputs
and outputs
tlte movenent of pollulants
anil sealtunel!
Prediction
of the hy&ofogicaf
responses

ungauqed catctments
Scllre further
benefits
based nodels

of alistributed

ON TAE CI'RRENT STATUS OF DISTRIBUIED

CATCIIUEIIT IIYDROIOGY

A BRIEF

'7
'1

a
of

physically-

!|ODELI,IIG

OF
IO

INTRODUCTION TO THE SIIE MODE]-

TOWARDS A GENERAL MEIITODOI.OGY EOR THE DISTRIBINED

MODEI,IING OF CATCTUBET SYDROI,oGY
6.L
6.2
6.3
6.4
6.5

Problen
definiti.on
Paraseter
definition
Uodel ca1.i-bration/validation
analysis
Sensitivity
Flanning
scenario
evaluatj.on

EIJ:|URE PROSPECTS

I9
20
22
23
23

25

Acknowl edgeEents

REEERENCES

27

" ~ o w a d a y smodel b u i l d i n g has become a generously s u p p r t e d

indoor sportn
Von Bertal anff y,

ON TEIE RATIONALE

(1P

I 968

RPDROLOGICAL MODELLING

Mathematical modelling in hydrology can be j u s t i f i e d from two

standpofnts: (a) the n e c e a a i t y to gain a better understanding of the
land phase of the hydrological cycle, and (b) t h e necessity for models
t o be used as aids t o decislon-makfng in the planning and management
o f water resources. From the former, reeearcharfented standpoint, a
rfgoroua approach t o river basin modelling would seek t o formulate,
couple and solve the equations of consemation of s M ~ S ,energy and
momentuu describing the movement of water over and through the soil,
I n etream channels and in aquifers; however, lacking the computational
facilit iea to implement such an approach, research hydrologists have
Invariably adopted a less rigorous approach where a model has been
required t o provide simulations of flow showing good agreement with
observed flows, achieved through some model fitting process. In the
past, this haa been achieved with models of a quasi-physical/semiempirical nature which have nor recesaarlly increased the understanding of the systems being modelled, nor have the model parameters
had a sound physical basis. Only in recent years has it proved
possfble t o adopt the more rigorous physics-based approach, thus
providing an opportunity t o increase our understanding of
catchment/river basin behaviour.

Besidea their r o l e as research tools, hydrological models have

came t o p l a y an increasingly important role as decisionlaaking a i d e in
water resources planning and management. For example, in evaluating
reservoir yield, s t r e d l o w records are rarely long enough t o allow
reliable estimates of yield to be obtained; longer records of rainfall
are frequently available and a rainfall-runoff model can be used t o
extend the streamflow record, thus providing more information for .
reservoir y i e l d evaluation. The increasing use of telemetry in the
short-term management of water resource s ys terns now means that
hydrological models can be employed as real-tfme flow forecasting
t o o l s , while water quality forecasting, also dependent on hydrological
model)Sng, is also a developing area of application.
The models employed in the above applications are t y p i c a l l y
lumped with either a quasi-physical/semi-apitical or black-box
structure, and thus requfre historical records of streamflow, as well
as of meteorological variables, for their calibration. Hence, they
all share the common attribute that if the hydrological response of a
catchment or river basin changes, for example, due t o man's
activitiee, then there ie no reliable means available of a l t e r i n g the
model parameters to reflect such changes, since the parameters are not
p h y s i c a l l y baaed +
Thfs report is concerned with physics-based distributed models of
catchment/river basin behaviour which, m the one hand, offer the
prospect of gaining a better understanding of rfver basin hydrology

and, on the other hand, can occupy important roles in a range of

applfcations t o be deffned in Sections 3 and 4 of this report. There
is an f n c t e a s i n g amount of concern worldwide about the effects of
development and man-made changes on both the water quantity, water
quality and erosion/sediment traneport regimes of river basins, and it
I s in t h i s area, where such models can provide a r e l i a b l e phyeieal
basis for assessing the effects of such changes, that these models
promise t o be of greatest practical use, Until now, their uee has
been rather fragmented, with a tendency towards the development of
models limited t o only one or tw processes and specffic t o a single
applfcation. It is hoped in t h i s report t o present a f i r s t approach
towards a general me thodology for d i e trlbut ed modelling in hydrology.
This w i l l be discussed with particular reference t o SHE, the Systzme
Hydrologique Europben. SHB has been developed as a general physfcally
based distributed modelling system by the Danish tIydraulics Institute,
SOGREAH (France) and the Institute of Hydrology (UK).
The structure
of the model that l o the basis of SHE is described in Section 5 .

"In p r a c t i c e , if the lumped systems models can serve our purpose,

then why should we Bother to i n v e s t i g a t e the more difficult distributed system for any other reason that the satisfaction of our
intellectual curiosity?"
2

Ven Te Chow,

I967

ON THE CHOICE OF A MATHEMATICAL MODFX

There is a stark contrast between the complexity of hydrological

r e a l i t y and the important and often pressing practicality of making
management decisions based on limited knowledge of that r e a l i t y ,
Knowledge of the real system will be embodied in a collection of
perceptions of the system u i t h a l l i t s complex web of variability in
space and time. These perceptions w i l l be essentially qualitative in
nature and will be based on previous personal experience of observed
behadour as well as that recorded in the hydrological literature. We
may call this accumulated understanding a perceptual map of the
eystem.
There is an important distinction between this personal
perceptual map and what actually goes on in the real world, a
distinction which implfes that the perceptual map is both inherently
subjective and capable of continued improvement.

However, decision-making processes require methematical models

that are rigorously d e f i n e d and capable of producing quantitative
estimates of the behavfour of the real world system* The d e f i n f t i o n
of such a model w i l l necessitate considerable simplification of the
perceptual map, and i n particular an important change of scale.
Qualitatively we may distinguish changes in hydrological processes and
variables a t very small space and time scales, whereas for management
decisions we often require bulk eatfmates (even i f in s s p a t i a l l y
distributed sense) o f the behaviour of the system. It is important t o
note that the simplication in the deffnition of a mathematical model
and also computational constraints must involve the introduction of
further subjectivity. A direct consequence is t h a t :
(i)

more than one model or model parameter set may give equally good
results ;

(ii)

different models may rats more or less highly in different

application^;

( i l l ) complex models may not necessarily g i v e better results than

simpler ones.

There are many ways of classf fying hydrological models (see

Clarke, 1973) but for present purposes we may distinguish models that
treat a catchment as a - s p a t i a l l y bariable system (distributed models)
or only fn terms of average quantities over the catchment area (lumped
models).
Lumped models rely primarily on comparison between observed
and simulated catchment outflowe for calibration of the model
parameters, and may in fact be based s o l e l y on the techniques of
systems anafysia in relating input t o output without reference t o the
i n t e r n a l mechanisms of the catchment (eg the constrained l i n e a r
systems model of Todinf and Wallis, 1977, or the transfer function
models of Box and Jenkins, 1970.

There I s a large number of lumped models which represent

hydrological processes by different functional forms defined
intuitively. These models are commonly tefmed 'conceptual ' models in
hydrology although all models muet be coneide red as conceptual
cons truc tione. In that the parameters of lumped 'conceptual' models
must be averages over the catchment area, the derivation of such
parameters must then be dependent on curve fitting techiques to match
simulated predictions wf th obeervations. This require& a period of
historical data to be available for model calibration and implies that
physical Interpretation of parameter values must then be made with
c a r e * A classic mdel.of t h i s type is the Stanford Watershed Mdel
(Crawford and Linsley, 1962) and its derivatives. However, the
poasibilitles of subjectivity in the definition of such models are
evidenced by the plethora of modela of this type (see for instance
Fleming, 1975).

A major category of dietributed models is that in which the

models are physically-based in the Bense of being defined in t e r m of
Physically-based models
theore ticaliy-acceptable continuum equat ione
are thue descriptive in the sense that the main o b j e c t i v e is t o
represent the behaviour of a physical process, often with the object
of gaining a better understanding of that process. InseinctiveTy, it
would be expected that physically-based models would provide
sirnulatione that are in some way better than other models In that they
must reflect more closely our perceptual map of the real world. It
must be stressed, however, that the distinctions between types of
model dram here is not clear c u t . In particular, the mathematical
definition of physically-based models f a often such that analytical
solutions t o the equations cannot be found and resort must be had to
approximate numerical solutions based on a finite difference or finite
element discretieation of the space and time dimensions. Such modele,
in t h i a case, also involve a degree of 'lumping' of catchment
processes at the model grid ecale, a scale that will o f t e n be larger
(because of computational limitations) than a scale ehsracteriatie of
the operation of the hydrological processes. Secondly,
physically-based models are usually based on process 'laws1 that are
essentially empirically based (eg Aarcy's Law or the uniform flow
equa tionr auch as the Manning or Darc y-We1 sbach equations).
There
remains therefore considerable scope for subjectivfty in the
d e f i n i t i o n of physically-based d i s t r i b u t e d models. However, the
important dff ference from the l m p e d 'conceptual' modele 1s that the
relationships used may be validated by process measurements in t h e
f i e l d , and improved over time as a result of experiments in the same
way a8 other scientific laws. Thuq physically-based distributed
models have parameters that are in principle meahrable In the f i e l d ,

There is, in a d d i t t o n , a class of semi-distributed models of

varying degrees of theoretical acceptability that bridge the gap
between the lumped models and the f i l l y distributed ph;sicalli-baaed
models. These modela use conceptual functional relationships for
different hydrological processes a p p l i e d t o a number of relatively
homogeneous subareas of the catchment treated aa lumped units. They
are essentially an extension of lumped 'cooceptualt models t o tho
semi-distrfbuted case and are t o some extent subject to similar
limitations. While some of these models can be a p p l i e d on the basis
of field measured parameters alone (eg Beven and Kirkby, 1979) it I s
more usual for the models, which remain relatively simple, t o be
calibrated by o p t i m i s a t i o n against an observed discharge record (eg

Mein et al, 1974; Knapp e t al, 1975; h e and Iklleur, 1976; Solomon
and Gupta, 1977; Boyd et al, 1979).
These models do, nonetheless,
allow additional. hydrological, soil and vegetation data t o be
incorporated into the callbratton procesa.
The above classification of hydrological models has been based
mainly on a consideration of model structure. A further
classification may be made into determinietic and stochastic models.
This depends largely on whether the model parameters and or model
outputs are considered to be t o tally error-f ree (deterministic) ox
s u b j e c t to error (stochastic)
lumped models that are d e t e r m i n i s t i c
in concept (eg the unft hydrograph) often haw an i m p l i c i t stochastic
structure (see Clarke, 1973). Fox the case of physically-based models
that are deterministic in operation and haw parameters that can be
measured in the f i e l d , the classification w i l l depend on whether these
measurements are without error (a deterministic model) or subject to
error (a stochastic model). In the case where the required parameters
are measured without error, a atochastlc classification for the model
map still result i f , when made1 predictions are compared with
obsemationa, errors are observed. If the errore are due t o the fact
t h a t the model is not a perfect facsimile of the real world, then it
should be clasaed as etochastfc. I f , however, the only source of
error i s due to measurement errore in the model inputs then the source
of stochasticfty lies essentfslly outside the model which should then
be classified as deterministic. In practfce of course it is
impossible t o separate sources of error, but i t w i l l generally be true
that most models, including those that are deterministic in operation,
will have an implied stochastic component. This s t r i c t definition i s
often relaxed so that models are referred to a8 deterministic if they
are determinietic in operation alone. Underlying the uee of
detenninfstic models is an assumption that the resolvable
(deterministic) component of a process dominates the unresolvable
(stochastic) part to the extent that a deterministic simulation can
provide meaningful forecasts of system behaviour.
Given the wide range of hydrological models a v a i l a b l e , the choice
o f a model for a particular problem is never a simple one, and will
inevitably be based on economic constraints and personal preferences>
as well ae purely hydrological considerations. Data availability is
o f t e n a crucial factor in such a decision. However, any general
criteria for model choice should be based on matching the requirements
of a management problem w i t h the complexity of the model used.
Modelling hydrological systems is an a c t i v i t y that is, in rnany
applications, governed by a law of dfmfnishing returns* Very s i m p l e
models can of ten provfde a useful, if not neceesatily adequate,
representation of the system and the benefits that derive from
additional complexity may not warrant the corresponding increase in
costs.
Thus, f f a problem requirea a model to extend catchment
dfscharge records from a longer rainfall record, i t may be s u f f i c i e n t
to use a lumped model and probably would not ma&
sense t o use a
complex distributed model demanding far greater resources. On the
other hand, i f i t is necessary t o p r e d i c t the e f f e c t s of localised
land use change then it is necessary t o use a distributed
physically-based model in which the changes in catchment
characteristics can be d i r e c t l y r e f l e c t e d i n changes in the model
parameters

There are further reasons for chooeing physically-based

distributed hydrological models in preference t o lumped models. In
particular, distributed models allow much readily available d a t a t o be
incorporated d i r e c t l y into the model. For example, one of the moat
important character is tice determining the hydrologic response of a
catchment must be the nature of the catchment topography. Yet,
topography is very o f t e n completely ignored I n the formulation of
lumped 'conceptualv models. In add5 t i o n there is ample evidence t h a t ,
due t o topography and the variability of soils and vegetation, the
response of catchments is non-linear and spatially variable, both
within and between storm periods (see review by t l e w l e t t , 1974; Freeze,
I974 and Dunne, 1978). Such variability w i l l be reflected In the
predictions of a distributed and physically-based model. In addition,
physically based models may overcome eome of the problems posed by the
prediction of extreme events that are n o t represented in h i s t o r f c a l
r a i n f a l l or discharge records. The forecasts of lumped models in
eases that are beyond the range of the records used for calibration,
must be open t o d0ub.t. Providing that the assumptions underlying a
physically-based model are not invalidated under extreme conditions,
forecaate of extreme response should be theoretically more
acceptable. It is expected that the sound theoretical basis of rheos
models w i l l be the p r i n c i p a l reason for a more extensive application
of distributed models.

"Nearby is the graceful l o o p of an old dry creek bed, The new

creek bed is ditched straight as a ruler; it has k e n runcurledf
by the county engineer t o hurry the runoff. On the h i l l in
the background are contour& s t r i p crops; they have been
'curled' by the erosion engineer to r e t a r d the runoff.
The
water must be confused by so much advice".
~ l d L
o eopld
The Sand County Almanac,

3.

ON THE NEED FOB PHYSICALLY BASED DISTRIBUTED MODELS IN HYDROLOGY

There are four (related) areaa for whfch physfcally-based

s p a t i a l l y distributed hydrological models can f u l f i l the needs of
practical applications: catchment changes; spatially variable inputs
and outputs; the movement of pollutants and sediment through a
catchment and forecasting the hydrological reaponae of ungauged
catchents, In each case alternative methods exist and these will be
d i s c u e s d below i n relation t o thc distributed modelling approach. In
a d d i t i o n , some further benefits of physically-based models are
diacueaed.

3.1

Catchment changes

This category covers a l l foma of hydrologically effective

changes in catchment characteristics, includfng both natural and
man-made changes (but not the external problem of changes in natural
inputs t o a catchment). Such changes include land use changes due t o
the effects of forest management, urbanisation, forest clearance for
agricultural purposes, the effectr of forest fires, e t c . Such changes
are almoet invariably localised and piecemeal, such that catchment
change is a distributed problem. The parameters of a physically-based
d i s t r i b u t e d model have direct physical interpretation, and their range
can be established reasonably well on the basis of field and
laboratory investigations. It follows that knowledge of parameter
values can be d i r e c t l y related to catchment c h a r a c t e r i s t i c s such as
s o i l type and land use and auch knowledge should be traneferable,
wlthfn reason, from area t o area. n u s it should be possible t o
p r e d i c t the effects of catchment changes prior to data becoming
.
available. In addition the localised nature of such catchment changes
can be e a s i l y incorporated into the spatially distributed model
structure

There are other methods available for predicting catchment

changes, in particular usfng lumped 'conceptual' models. The
representation of s p e c i f i c hydrologic processes by functional oms in
such modefa hae led t o the interpretation of t h e parameter values as
quasi-physical quantfties. This h a s resulted in much unguarded
optimism ae exemplified by the following quotations:

'A model is not only an excellent exploratory tool but also a

direct aid t o creative thinking. Potentially I t can provide
answers regarding the sensitivity of the total system t o changes
in certain parameters and the amount of change needed t o achieve
a desired result' (Rfley e t a l e , 1968, p . 2 7 ) ;
'Any a c t f v i t y of man which may change the watershed can be
measured or p r e d i c t e d and again related to the model parameters

used, and a new flow sequence can be simulated. The planner will
then have a quantitative estimate of marginal hydrologic change
to uae, an estimate of a degree of precision never before
poseiblet (James, 1972, p . 305);

' Other changes such as vegetation manipulation, forest fires,

soil conservation techniques s t c can be p r e d i c t e d if data for
c a l i b r a t i o n are available. Where adequate rainfall and
streamflow exist ordinary calibration techniques w i l l define the
amount of the change and the mechanism through which i t works '
(Hydrocamp, 1971, p. 4-2).
It must be stressed that these authors overstate the case for
lumped tconceptualt models. It is dangerous t o interpret the degree
of mathematical equivalence with observed catchment hehaviour achieved
by such models in terms of physical significance, particularly in the
analysis of catchment changes. Lumped parameter values should not be
taken as having sighificance outside t h e model structure (Nash and
S u t c l i f f e , 1970), and the transfer of parametric information over time
or space to different model applfcatione muet be considered extremely

hazardous.
3.2

Spatially variable input8 and outputs

In t h a t lumped catchment models can deal wly with quantities

averaged over the catchment area, i t is necessary to use a distributed
catchment model when there is significant spatial variability in
catchment inputs and outputs. Particular examples are the movement o f
rainetonns over a catchment (Surkan, 1974, Urlson e t al, 1 9 7 9 ) and
localised river and groundwater abstract i o n s and recharge. Such
spatial variability is handled e a s i l y by distributed models, provided
that suitable input d a t a are available. It may be that technical
developments such aa r a i n f a l l meaeurement by radar and other remote
senaing techiques w i l l provide such data in a form directly usable by
distributed models.
3.3

The movement of w l l u t a n t e and sediment

Water provides a supply, transport and dispersion mechanism for

pollutants and sediment, and since most water quality problems are
distributed in nature, distributed models ate therefore most suitable
to provide the basic hydrological input t o predictions of water
quality (see for instance, Bresler, 1973; Brazil e t al,, 1977; Nelson,
1978; Selim, 1978).
Again there are alternative methods for making
such predict ions, Water q k l i t y modelling can be approached using
systeme analysis input/output techniques provided that data are
available for calibration of a model. Indeed the complexity of water
quality interactions makes this an attractive alternative when such
d a t a are available, provided that the sys tern is relatively atat lonary
over both calibration and forecasting periods, Lumped 'conceptual'
models have also been used t o make quality predictions. An enhanced
v e r s i o n of the Stanford Watershed Model has been used to predict the
movement o C rad l o a c t i v e aerosols through a catchment (Huff and Kruger ,
1967).
It would seem that this case of non-point pollution, in which
the quality variable may be consfdered as uniform over an area, 1s
much more amenable t o lumped modelling than the case of spatially
variable inputs. In fact it i a difficult t o see how localised Bources
of p o l l u t a n t s or sediment can be successfully handled by spatially
averaged models except I n t h e caae of one or two recurring sources.

3,4

Predictton of t h e hydrological responses of ungauged catchments

The possibility of using physically-based distributed models for

predicting the responses of ungauged catchments arf ses d i r e c t l y from
t h e p h y s i c a l significance of the model parameters. Sfnce it is
p o s a i b l e to e s t a b l i s h reasonable ranges Eor the parameters on the
b a s i s of intensive, short-term field investigations, such models can
be used t o generate at least approximate hydrological predicrions
without the benefit of long concurrent records of precipitation and
streamflow for calibration. I n fact any information on the
characterlstiee and hydrology of a catchment that is readily available
can usually be incorporated in the modellfng process.
Most alternative methods for making predictions on ungauged
catchments rely on the use of statistical regression of the model
parameters against catchment characteristics using parameters
calibrated on gauged catchments. The regression equations may then be
used to p r e d i c t parameter values for ungauggd catchments. Such
analyses have been made both for unit hydrograph models (NERC, 2975)
and more complex conceptual models (Ross, 1970; James, 1972; Jarboe
and Haan, 1 9 7 4 ) . I n view of some of the d i f f f cult ies of calibrating
conceptual models (see for example, Ibbitt, 1972) there are obvious
dangers in this approach which neceaearily only crudely approximates
the causes of variability between catchments by the use of surrogate
variables in the regression equations.
Same further benefits of distributed physfcally-based m o d e l s
D i s t r i b u t e d models of water Flow can provide information for a
variety of linked problems, some of which, such as estimating
consolidation or subsidence, may be directly coupled to the solutton
of the flow problem (see for instance, Corapcioglu and R tutsaert ,
1977; Naraslmhan and Witherspoon, 1978). Other problems nay be less
d l r e c t l y linked, for example the prediction of pore pressures in slope
stability s t u d i e s (eg Hodge and Freeze, 1 9 7 7 ) and the water quality
problems discussed above. Other interactions w i t h d e t e r m i n i s t i c
hydraulic models may be expected t o develop.
The use of physically baaed models focusses attention on the
inadequacy of the hydrological data upon which many water resources
management d e c i s i o n s are based. B y considering the uncertainties in
the parametric data explicitly, estimates of the resultant uncertalnty
in the p r e d i c t e d hydrology can be made. A t present this process is a t
a relatively crude level in applications of complex catchment models,
generally involving a sensitivity analysis d t h i n an expected range of
estimated parameter values. However, present s t u d i e s i n v o l v i n g one or
two processes (see Section 4 ) show that the available techniques may
be expected t o become considerably more sophlst lcated. This type of
sensitivity analysis should a i d the management process in both
providing estimates of the uncertalnty in hydrological predict ions and
in suggesting how additional expenditure in the field detenninatlon of
parameter values might he used most b e n e f i c t a l l y . This may become one
of the most important uses of physically-based simulation models.

' A s Scientists w e are i n t r i g u e d by the possibility of pssembling

o u r concepts and b i t s of knowledge into a n e a t package to show
t h a t we do, a f t e r a l l , understand our science and i t s complex

interred a t e d phenomena *

W M Kohler, J969.

4.

ON THE CURRENT STATUS OF DISTRIBUTED MODELLING OF CATCHMENT

HYDROLOGY

The development and application of d i s t r i b u t e d modelling in

hydrology has been very fragmented. There have been numerous papers
on the theoretical aspects of modelling various hydrologic processes
independently using both analytical methoda for simple cases and
approximate nmerical methods of increasing sophistication. There is
a much smaller literature on modela involving more than one process
and on the application of distributed model8 to real world problems.
It is beyond the scope of this report t o provide a full review of the
development of distributed model8 and ue shall concentrate on some
general features of distributed modelling with approprfate
illustrations from a mall nmber of specific s t u d i e s .

First, consider the catchment as a f u l l y distributed system,

continuously variable over time and 3 space dimensions. I t is
possible to write down general partial differential equations for the
process of mass and energy transfer within the catchment continuum,
together with bundaxy conditions for those equations, based on
physically realistic assumptions compatible with current kmwledge of
hydrological processes. These equations comprise the general
distributed model of catchment hydrology (see for example the
d i s c u s s i o n s of Freeze and Harlan, 1969; and Freeze, 1978)* The
complete system of non-linear equations is impossible t o solve
analytically for any case of practical interest and resort m u s t then
be made to approximate numerical solutions. Numerical solutions can
only be calculated for a finite nmber of points (nodes) in space and
time, and the complexities of the system are such that computing
conetraints will limit that number of nodes, A consfderable amount of
research, mostly concentrated on studies of s f n g l e processes, has been
aimed at developing solution methode that are sufficiently accurate
but remain efficient in t e r m of computing requirements and t h e
dlscretisation of the catchment Into nodes.
The e a r l i e s t solution technique, and atill the most commonly
used, is the f i n i t e difference method pioneered by Richardson (1910).
Finite difference techniques were f i r s t applied to groundwater flows
by Shaw and Southwell (19411, unsaturated f l o w by Klute ( 1 9 5 2 ) and
channel f l o w s by Stoker (1957). Finite difference methods have been
considerably improved and a wide range of methods is now a p p l i e d q u i t e
routinely t o steady s t a t e and transient problems in groundwater and
channel routing s t u d i e s . More recently, however, several other
methods have been,used in hydrology including f i n i t e element methods
(Pfnder and Gray, 1977), integrated finite difference methods
(Narasfmhan and Witherspoon, 1977) and the methods of characteristics
(Amein, 1966; Wlggert and Wylie, 1976). The applCcation of a
3-dimensional complete catchment model solved by f i n i t e difference
methods has been demonst rated on llmi ted hypothe tical problems by
Freeze (1971, 1972a, 1972b).

The stage has now been reached where the solution method8 are
sufficiently well advanced to give satisfactory results for a wide
range of problems, providing that the problem of i n t e r e a t l a -11
posed in terms of the specification of the boundary condition8 and
data input t o the model. Many studies have shown that these
d i s t r i b u t e d models can reproduce the behaviour of hydrological systems
as measured i n the laboratory (eg Ragan, 1966; Tang and Skaggs 1977;
Haverkamp e t a l , 1977; Luthin e t al., 1975; Mu~ik, 1974) and in t9ls
field h e i n and Fang, 1970; Bresler e t al., 1979; Feddes et al., 1976;
Pikul e t al., 1974; Smith and Woolhiser, 1971, although fn complex
field situations I t may be d i f f i c u l t t o obtafa completely satisfactory
eimulations ( Stephenson and Freeze, 1974).

Now that the basic solution techniques of distributed modelling

are well understood there has been a considerable nmber of studies
e s s e n t i a l l y filling i n points of detail, There have, for instance,
been several compaxlsons of ntrmerical solution techniques (eg Price,
1974; Trescott and Laraon, 1977; Fhavertcamp e t a l . , 1977; Hayhoe;
1978).
Other uorks have attempted t o increase the generality of the
mathematical formulation, often a t the expense of additional
complexity. Ekamples axe the inclusion of compaction coefficients
(Freeze, 1971; Naraeimhan and Witherspoon, 1976), non-Darcy type f l o w
i n ground water flow models (Valker, 19691, inclusion of hysteresis i n
unsaturated flow models (Uhfsler and Klute, 1965; Hoa e t al., 1977;
Gieaal e t al., 1973; Lees and Watson, 1975) and alternative flow
relationships for upland channel flows (Beven, 1979b)
Further
s t u d i e s have examined the effects of using different discretfsations
(Hayhoe, 1978; Rushton and Tomlinson, 19771, of different methods of
calculating the equation coefficients (Appel, 1976; Haverkamp e t al.,
19791, specifying the initial and boundary conditions (Rushton and
Weddesburn, 1973; bbrris, 1979; Zaradny, 1978) and stability criteria
(eg Young, 1971; S t r e l k o f f , 1970).
Methods for reducing the computing
requirements of distributed models have also been developed. Some of
these methods rely on formulating the solution m e t h d t o reduce the
amount of storage required (eg Prickett and b n n q u i s t , 1973; Fread,
1971) and others on decoupliag the f u l l y 3-dimensional problem into a
number of one or two dimensional componente, for example the surface
flow models of Kibler and Woolhiaer (1970); Engman and Rogowaki (1974)
and Ross e t al. (1979); the two dimensional h i l l s l o p e elements of
Beven (1977); the models of multilayer groundwater systems ( e g
Bredehoft and Finder, 1970); and the structure of the SHE model
described in Section 5 below. This decoupling must introudce some
additional inaccuracy, particularly f n models such as the kinematic
cascade where the complexities of the topography of natural h i l l slopes
are represented by 'equivalent' rectangular planes (see l a n e and
Woolhiser, 1 9 7 4 ) . However, in models such as the SHE model (see
Figure 1) the loss of accuracy compared w i t h an equivalent
3-dimensional model may be small and a finer d i s c r e t i s a t i o n may be
handled on current computers, so a s t o allow applicatfon of a complete
catchment model t o real catchments. It must be added that w i t h all
approximate solution methods the need for discretisation may itself
generate problems in the application t o real world problems, since the
grid s c a l e of the model must often neceaearily be larger than the
scale at which the hydrologic processes operate. The problem of
defining parameter values a t t h e grid scale w i l l be considered further
in Section 4.

There has also been recent interest in d i s t r i b u t e d models vfewed

as stochastfc models i n the s e n s e d e f i n e d i n Section 2.
All the
m o d e l s descrfbed above are deterministic in operation and assume t h a t
the true values of the parameters and boundary conditions of the model
are known without error. This will rarely, if ever, be true in
a p p l i c a t i o n s to real situations and the effect of allowing the
parameter values and boundary conditions to be defined stochastically
with a non-zero variance has been the e u b j e c t of severs1 s t u d i e s i n
groundwater flow (eg Bibby and Sunada, 1971; Freeze, 1975; Gelhar,
1974; Sagar and Kisiel, 1972; Tang and Finder, 1977; Sagar, 1978; Bakr
e t al., 1978; Dagan, 1 9 7 9 ) ; overland and channel flow (Chiu, 1968;
Machado and OfDonnell, 1979); and a complete hillslope model (Freeze
1980). It is also worth noting that f i e l d observations, on which
parameter estimates are based and comparisons w i t h simulated results
made, are also subject to sampling and measurement errors. Some
studies have shown that the error variance associated with such
measurements may be considerable ( s e e for instance Hills and Reynolds,
1969, and Kiesling et al., 1977) and that a i n g l e measurements may not
give good estimates -of the spatial average of a variable. T h i s
certainly has a bearing on the specification of parameter values at
the g r i d scale. This problem w i l l be discussed further in Section 6
but tw p o i n t s seem worth making here.

First, estimates of the variance of parameter values have

generally been made on the basis of tntensive spatial sampling
programs (eg the log normal distribution of hydraulic c o n d u c t i v i t i e s
found by Willardson and h r s t , 1965; Nielsen e t al., 1973; Rogowski,
1972). It is commonly assumed i n studies of the stochastic nature of
p h y s i c a l l y based distributed models that these measured s p a t i a l
distributions represent the point sampling distribution of that
parameter and that each measurement represents an independent sample
from the underlying population. This can only be true i f the s p a t i a l
autocorrelation of the parameter is small relative t o the dfstance
between samples. If spatial autocorrelation ie significant at the
model grid scale, then the spatial pattern of the measurements must be
taken into account in the determination of parameter values d e f i n e d a t
the grid scale. In this case, the sampling variance associated with
the g r i d square parameter values may be smaller than that based on the
assumption of independent masurements from a point sampling
distribution (see discussion in Bevcn, 1981)

A second, related, problem is that point meaeurements of

hydrological parameters are u s u a l l y based on experiments that are
small in scale relative t o practical model grid scales (from
laboratory cores to small plot experiments). The theory of
geostatistics suggests t h a t , for e t a t i c quantities, as the sampling
volume of a quantity ( t h e support volume or measurement scale)
increases, the sampling variance will decrease (Journel and
Huijbregts, 1978; Delhomme, 1979; Burgess and Webster, 1980). A
similar relationship between sampling variance and measurement scale
can be expected t o hold by analogy for dynamic systems.
Certainly,
theore tf cal experiments have shown that the calculated variance of the
solution output variables from distributed models arising from the
variance in parameter estimates, decreases as the number of space
dimensions considered increases (see Bakr et al,, 1978; Dagan, 19791,
reflecting the compensation for variability that might be expected in
the continuous flow system. It should be noted, however, that for
non-linear flow systems, the nature of t h i s compensation may depend on

the flow c o n d i t i ~ n,
s such that appropriate grid square parameter
values may vary over time. The defiaitian of grfd square parameter
values, while dependent on the measured point sampling distribution of
a parameter, muat be considered as a higher order problem, I n that
grid square values w i l l also reflect model structure, model g r i d ecale
and spatial structure in the measurements* This whole question of the
interactions between hydrological processes, measurement scale, model
structure and model parametera requires much further study.

Thie review har necessarily been brfef and incomplete but has
attempted t o illustrate the current status of both the technfquee and
underetanding of distributed modelling in catchment hydrology. There
have been very f e w attempts t o d e l cmpleta catchments in a truly
distributed manner except l a caaea where catchment response is
damfnaeed by surface f l o w . The methods of diatributed modelling are
now sufficiently well developed that data limitations generally
outwelgh the limitations of technique in applf cations to the real
world. It should then be posaible t o s p e c i f y a catchment model for
general use t o gain experience of applications t o practical prablema,
A first attempt a t such a model 1s the baais of SllE which is described
in the n e x t section.

"We must not omit to take notice of what w e see w i t h our eyes, t h a t
w a t e r n a t u r a l l y tends downwards, t h a t it cannot s u f f e r the a i r to
be anywherembeneath it
t h a t it loves to fill u p every concavity
into which it r u n s ; t h a t the mare you endeavour to force it, the
more obstinately it struggles against you, nor is i t ever satisfied
till it o b t a i n s t h e rest which it desires."
Alberti (1404-14721

...

5.

A B RIEF f NTRODUCTION TO THE SHE MODEL

The SHE: model that is the basis of the s y s t & e Hydrologlque

~ u r o ~ g eis
n a physically-based dis trlbuted model, determinir tic in
operation, that incorporates components for all the m a j o r hydrological
processes. The model has been developed from the non-linear partial
differential equations of flow for the processes of overland and
channel flow, unsaturated and saturated subsurf ace flow, solved by
f i n i t e difference methods. The model is c m p l e t e d by point snow melt,
interception and evapotranspiration components. While conriderable
effort has been made t o ensure that the solutions for each component
are accurate and efficient, the structure of the model represents a
compromise between the restrictions of computing and data requirements
of the model and the need to represent the complexity of real
catchments. A description of SHE is given by Jonch-Clausen (1979);
only a brief smmary is provided here.

It is not yet economically viable t o produce en operational model

that is f u l l y 3-dimensional in space and which will also allow the
required accuracy of dfscretisation in both horizontal and vertical
planes. On the assumption that, in the unsaturated zone, vertical
flow is, on most slopes, far more important than lateral flow the SHE
model has been rationally simplified such that independent
one-dfmensional unsaturated flow component8 of variable depth are used
to link a two-dimensional saturated eubaurface flow component and a
two-dimensional surface f l o w component (Figure 1 ) . It is planned to
use up t o 2000 grid points in the horizontal and 30 in the vertical t o
allow adequate definition of a catchment areal unft.
The application of a model of this type requires considerable
i n p u t s of parametric and exogenous d a t a , including parameter values
that change over time, for example through the growing Peason of a
crop. Such data will not always be readily available and considerable
flexibility has been built Into the model t o allow different modes of
operatlon demanding different levels of d a t a availability, It is
stressed that parameter values are i n p r i n c i p l e measurable in the
field and it is hoped that the general availability of a model of t h i s
t y p e , w i l l instigate more widespread measurement of the data required,
i f not on a routine basis, then a t least as part of the application of
the model t o a specific project.

The interceptfon/evapotranspiration component of the model

de tetmines the total evapotranspiration and net rainfall from

meteorological input data. The interception process is modelled using

a variant of the Rutter model (Rutter e t al., 1975) and has canopy
storage and drainage parameters that can be estimated by experiment
(see for instance Rutter e t al., 1 9 7 5 ; Gash and Morton, 1978).
There
are several modes for the evapotranspiration component .of which the
most complex is the full Penman-Monteith equation (Monteith, 1965)

F
R A I N . SNOW

mPRECIPITATION

1
-FIGURE
.a

Schematic representation of
t h e s t r u c t u r e of t h e SHE model

- DIMLNSIOHAL

SATURATED

FLOW GROUNDWATER MODEL

RECTANGULAR GRID)

which requires aerodynamic and canopy resistance parameters (see for

instance Szeicz and Long, 1969; SzeCcz e t al, 1969; Gash and Stewart,
2 9 7 5 ; Tan and Black, 1976).
The interception/evapotransplration
component i n t e r a c t s directly with the r o o t zone, which 1 s the upper
part of the unsaturated zone component. The extraction of moisture
for transpiration from the root zone is d i s t r i b u t e d accordfng t o the
vertical distribution of roots. Vegetation characteristics and
meteorological i n p u t s t o 'the model can vary from grid square t o g r i d
square. A f u l l description and sensitivity analysis of this part of
the SHE model is g i v e n in Beven (1979a).

The overland and channel Flow components represent the surface

runoff and river f l o w processes. They are m o d e l l e d by finite
d i f f e r e n c e solutions to t h e simplified S a i n t Venant equations (with
inertia terms neglected).
In the modelling of overland flow the s l o p e
of the water surface is assumed t o be parallel t o the ground surface
(kinematic assumption) but for channel f l o w , a water surface s l o p e
term is included in the formulation so that backwater e f f e c t s can be
modelled (see Preissman and Zaoui , 1979). The overland and channel
flow components require parametric d a t a on channel dimension and flow
resistance functions for each g r i d square.
The unsaturated flow component uses a finite difference s o l u t i o n
of the Richard's equation in a multi-layered s o i l i n c l u d i n g the root
zone. An interacttve solution technique has been a d o p t e d t o cope wlth
the important non-lineartties of the flow equation. The unsaturated
flow component datermines rates of infiltration end recharge in the
coupling with the surface flow and saturated groundwater flow
components. A particular problem associated w t h the structure of SHE
is the l i n k between vertical flow i n the unsatu ated zone and
horizontal flow in the saturated zone. A s a t i s f q t o r y approach to
modelling this transition has been developed (Abbqt et al., 1979).
The unsaturated zone solution requires information on the soil
moisture and hydraulic conductfvity characteristics, but theae may
vary for each soil layer and w i t h soil type in d i f f e r e n t grid
squares. The combined unsaturated zonelevapotranspiratfon components
have been tested by Jensen and Jonch-Clausen (1981).

The saturated groundwater flow component is, in the f i t s t version

of the SHE model, restricted t o a single layer, unconfined aquiEer
w i t h direct links t o any surface water b o d i e s . The model is based on
an alternating d i r e c t i o n i m p l i c i t finite d i f f e r e n c e s o l u t i o n to the
Boussinesq equation. Thfs formula t i o n must ree t f i c t application of
t h e model at p r e s e n t , but it fs envisaged that this component will be
extended t o cope with multt-layered, confined/unconfined aquifers in
the future. The saturated zone model requlres data on the extent and
depth of the aquifer together with conductlvfty and storage
c o e f f i c i e n t parameters for each g r i d square.
The snow melt component can also he used i n s e v e r a l modes
dependent upon data availability. The simpler modes are based on
Degree-day and Energy Balance predictions of melt rates, with routing
through the snowpack based on the empirical model of Anderson (1968).
The most complex form is represented by the f i r a t attempt t o model
both the energy and mass flux within a snowpack takLng changes in t h e
structure of the pack i n t o account. F i n i t e difference methods are
used to o b t a l n simultaneous solutions t o the flow of heat and w a t e r i n
the pack (see ~ l o r r i sand Godfrey, 1979; Morris 1982).
Snowmelt runoff
may he generated e i t h e r at the surface or at the base of the pack.

Factore affecting the accumulation and melting of snow will be

reflected in the difference in meteoro~oglcal inputs and parameter
values between grid squares.

It is stresaed that a l l the model components, while being

phyeically-based as far as possible, rely on empirical equations or
simplifying assumptions that are approximations t o the complexities
observed in the f i e l d , It is important t o recognise t h a t , while such
approximations must be limited in v a l i d i t y t o some extent, they are
consistent with present quantifiable knowledge of the physical
processes of catchment hydrology, and may be expected t o be improved
over time. The modular form of the structure of the SHE model
facilitates such changes. Perhaps the most important part of the SHE
model is the model ' frame' , known aa the Chef d'orchestre, that links
the model components and orders the s o l u t i o n of the components in
time. It is the f l e x i b f l i t y of the frame that allows a general
methodology of distributed modelling t o be outlined, as is discussed
in Section 6 .

"The deterministic simulation approach espoused here is potentially

superior to any empirical model t h a t requires less d a t a . If the
technique can be shown to have p r a c t i c a l v a l u e , it w i l l encourage
increased measurement of the necessary data."
Allan Freeze, 1971
6.

TOWARDS A GENERAL METHODOLOGY FOR THE DISTRTBUTED MODELLING

OF CATCHHEXT HYDROLOGY

The formulation of the SHE model o u t l i n e d i n the previous section

incorporates elements d e f i n e d on the b a s i s of the available
(quantifiable) physical knowledge of the catchment system. A t t h i s
stage I t has been necessary t o make some simplifying assumptions such
as the exclusion of hysteresis effects and lateral unsaturated flows,
but providing that the model is not used outsfde the range of
c o n d i t i o n s for which its assumptione remain reasonably valid, it is
likely that the ef Eects of such s i m p l l f ications will be small in
relation t o the l i m i t a t i o n s imposed by the data requirements of the
model, Thus the SHE model should provide a suitable b a s i s for a f i r s t
approach to a general methodology for distributed modelling (Beven e t
al, 1980). However the physical acceptability of the model is only
one aspect of a methodology for operational applications. In general,
( I)

the model must be physically relevant t o the system that it

represents ;

(2)

the model should incorporate the information from as much r e a d i l y

a v a i l a b l e hydrological and other c a t c h e n t data as possible;

(3)

the model should be a b l e t o airnulate a wide range of problems,

both transient and steady s t a t e , for s i n g l e processes and
combinations of processes up t o the level of a complete catchment
system;

(4)

the model should be able t o cope adequately with the difference

in response times of different processes and with the time
variant rates of change due t o d i f f e r e n t meteorological
conditions;

(5)

the model should be able t o cope adequately with non-stationarity

in the model parameters due, for example, t o seasonal effects on
the vegetation of the catchment ;

(6)

t h e m o d e l should be able t o cope with dynamic changes in the

component structure of the model for example when the unsaturated
zone disappears as the water t a b l e r i s e s t o the surface in
varf able source areas

The f i r s t two criteria should be s a t i s f i e d by any distributed

model for which both the undertying theory and the characteristtcs of
the solution techniques are acceptable in a given application. It haa
been suggested that the SHE model will s a t i s f y these c r i t e r i a over a
wide range of hydrological and geographfcal conditions. In additton a
considerable part of the e f f o r t expended fn the development o f SHE hae
been concerned wlth satisfying the last four criterfa in the
specification of the model 'frame' (the Chef d'0rchestye).

The model frame is used to control the input of the catchment

specification and the nature of the solution that is required,
reconciling the dynamic time step requirements for the individual
components and then ordering the solutions for the components w i t h the
correct transfer of v a l i d internal boundary data between components.
The d e t a i l e d specification of the frame necessitates the achievement
of a satisfactory compromise between maintaining s c f e n t i f l c rigour and
numerical efficiency. It is obviously neceesary that the transfer of
internal boundary data be handled so as to minimise inaccuracies in
the maintenance of mass balance, and that the control of time steps be
handled t o maintain s t a b i l i t y in the component solutions. The first
version of the frame is already operational with some restrictions on
the types of solutions that can be handled, but is expected t o be
revfsed and extended wet time as experience i n different situations
is incorporated.

Given t h e availability of a model l i k e SHE with a frame s u i t a b l e

for general application, there are several stages in the application
to a specific project, a s follows:

(4)
(5)

Problem definition
Parameter definition
Model calibrationlvalidati o n
Sensitivity analysfs
Scenario evaluation

6.1

Problem definition

(1)
(2)
(3)

Perhaps a t t h i s stage, i t is unnecessary t o raise the question of

whether a distributed model is really required t o solve a particular
management problem wfth its additional complexity, data requirements
and computational expense. However, while there will be projects f o r
which the choice will be obvious (see discussion in Sections 2 and 31,
in many situations it will be much less clear c u t . In such situations
it is necessary that the theoretical advantages of distributed models
be demonstrated in applications t o f feld problems so as t o overcome
the economic advantage of alternatfve lumped models. A t present. i t
can be expected that the use of d i s t r i b u t e d models will be confined to
projecte s p e c f f i c a l l y requfring d i s t r i b u t e d simulatLons as discussed
in Section 3.

In these eases ft is neceesary t o further decide whether

transient or steady s t a t e simulations f o r single processes or
combinations of processes are required. This w i l l be in part d e c l d e d
by the nature o f the problem and the availability of d a t a and
resources. It i s obvious that a s i n g l e process, steady s t a t e s o l u t i o n
w i l l be much cheaper but will not always provfde a l l the i n f o m a t t o n
required. The a i m of a model frame, such as that embodied in SHE,
must be t o allow flexibility in the choice of such o p t i o n s , so that L t
will certainly be p o s s i b l e t o proceed from a s i m p l e analysis to more
complex formulations as appropriate.
A final important p a r t of problem definition is t h e speciftcatlon
of appropriate model boundary c o n d i t i o n s . This may i n v o l v e
considerable effort, particularly in the case of subsurface flaw
s y s t e m s where boundaries may not be c l e a r l y d e f i n e d , or ~ h e r e
attention is focussed on a small part of the c a t c b e n t with
non-hydrological houndari es

6.2

Parameter definition

The definition of parameter values is of crucial importance t o

the applicatfon of distributed modelling, particularly when I t is
intended that such models should be used deterministically.
Distributed models require the specification of several parameters and
parametric functions at each individual grfd equare (see Table 1)
leading to large numbers over a catchment area. Temporal changes in
parameter values may further increase t h i s number. The problems of
defining a l l the individual parameter valuer by separate measurement
are immense and it is therefore necessary t o explore the p o s s i b i l i t y
of reducing the expense of parameter estimation. Fortunately, the
physical basis of those parameter values allows a degree of hope that
this may be possible, in that it should be poaeible to transfer
parameters measured in one location t o be repreaentative of similar
areas elsewhere. Thus n o t only may it be possible t o d e f i n e
parameters on the basis of vegetation or eoil type within a catchment
(obviously requiring a much smaller overall number of parameter
values) but it should also be p o e s i b l e t o tranofer information from
studfes outside the catchment of intereet. In addition, the

TABLE I

Typical parameters required for each g r i d element in a d i s t r f b u t e d

model of catchment hydrology (as taken from the SHE mode ll

Vegetation

Aerodynamic resistance parameter campy resistance parameter

(may be related to soil moisture in r w t zone}

Topography

M e a n altitude
Surface slope

Unsaturated
zone

Root d i s t r i b u t i o n w i t h depth
Hydraulic conductivity/soil moisture relationship f o r each
soil layer
Capillary tension/soil moisture relationship f o r each soil
layer

Saturated

Conductivity in direction of each axis

Storage c o e f f i c i e n t (may be r e l a t e d to unsaturated zone
solution

zone
Overland and

channe 1 flow
Snow m e l t

Surface roughness or channel dimensions and flow coefficients

as appropriate
Snow surface roughness
Parameters of hydraulic conductivity/snow water content/snow
density/grain s i z e relationship
Parameters of snow water tension/snow w a t e r content/snow
density/grain size relationship
Parameters of effective thermal conductivity/snow density/
grain size
Parameters of equilibrium temperature/snow water tension
relationship

application of remote sensing in hydrology will have beneficial

interactions with distributed modelling i n terms of providing
spatially distributed data and data handling procedures (see for
instance Jackson e t a1 1976; Bauer et al, 1978; Schmugge, 1978; Jordan
et al, 2978; Jackson and PkCuen, 1979).
However, it cannot be denied t h a t a sampling problem remains, as
demonstrated by the discussion of parameters as s tochs8tic variables
in S e c t i o n 4, T h i s has several implications. First, where more than
one estfmste of a parameter value exists, the arithmetic average may
not be the most appropriate estimate t o use in the simulation (see for
instance Freeze, 1975; Dagan, 1979). Secondly, even where a parameter
is easily measured within a l l grid squares, for example topography,
there may be no simple way of deffnfng the most appropriate value of
that parameter a t the grfd scale, except t o assme that some
characteristic average value w i l l suffice. The consequence is that
t h e parameters should be viewed as d e f i n e d by some (unknown)
distribution rather than as fixed values truly characteristic of an
area. The definition of parameter values should be associated w i t h at
least a realistic range of p o s a i b l e values, and the sensitivity of
simulations t o variations w i t h i n that range should be explored.
Ultimately, i t may be p o s o l b l e to reformulate the model equations t o
characterise the system at the g r i d s c a l e .
Given an appreciation of these problems of parameter definition
is important that the techniques of measurement used should, as far
as possible, correspond t o the etructure and scale of the model befng
used. Thus, measurements i n the f i e l d are more appropriate than
laboratory methods, and coarse s c a l e measurements (such as pumping
t e s t s or tube well methods for eaturated conductivity values) are
better than point measurements (for example core samples taken back to
the laboratory). The aim should be as far as p o s s i b l e to integrate
the detafled characteristics of hydrological processes and o b t a i n
parameter values characteristic of the overall response of a process.
It is likely then that a c l o s e r match between measured values and
appropriate grid scale values will be obtained, although the sampling
problem ( I n both space and tfme) w i l l never be eliminated. Vandenberg
(1977) explicf tly considers the effects -of parameter variability on'
the determination of transmissibility parameters by pumping t e s t .
It

The development of tracer techniques may be useful i n this

respect. As one example, the use of tracers i n upland channels has
shorn how the complete response of a sequence of pools, r i f f l e s and
falls can be fntegrated and used t o d e f i n e parameter values for
channel routing (Beven e t a l . , 1979) and lead t o new i n s i g h t s into t h e
nature of the bulk characteristics of f l o w , as they change with
discharge ( P i l g r i m 1976; Newson and Harrison, 1978)
S i m i l a r methods
can be used to study overland flow on complex vegetated surfaces
(Newson and Harrison, 1978). Tracers have also been used for some
time as a method f o r determining groundwater flow parameters though
without complete agreement as t o t h e Interpretation of results (see
f o r i n s t a n c e Kref t and Zuber , 1978)

A number of f i e l d methods are available for measuring the

unsaturated flow properties of s o i l s (see for example H i l l e l e t al.,
1 9 7 2 ; Jeppson e t al., 1975, Royer and Vachaud, 1975).
The parameters
of the interceprion and evapotranspiration component can be obtained

from net r a i n f a l l (see for instance Calder and Rosier, 1977) and
micrmeteoroLogica1 aeasurements (Stewart, 1977).
The neasurement of parameter values i n the f i e l d assumes a
particular importance for dtstributed physically-based modelling.
Every measurement serves t o h e l p d e f l n e a correct range for a g i v e n
parameter value and w i t h continued accumulation of i n f o m a t i o n may
allow specific data deficiencies to be shown up. Each measurement
also serves t o some e x t e n t to examine critically the physical b a s i s of
the model being applied. While it may be of l i t t l e importance to the
application of lumped models that the modeller should have s t u d i e d t h e
characteristics of a catchment other than j u s t its rainfall/df scharge
record, it I s far more important i n the a p p l i c a t f o n of
physf cally-based models.

6.3

Model calibration/valldation

The term model .calibration is here used specifically t o mean the

selective improvement of initial parameter estimates by a comparison
between observed and simulated hydrologf c a l varf a b l e s . Model
v a l i d a t i o n implies the acceptance of a model as an accurate simulator
of the real world system, which will also depend on a comparison of
observed and simulated behavfour. In the application of a distributed
model there may not always be the requisite historical data available
for model calibration or validation, A program of data collection
should usually therefore be implemented a t the same ti= as the
measurement of parameter values, although it is worth noting that
measurement and spatial sampling problems also pertain to the
collectfon of data to represent observed behavfaur, which can itself,
therefore, only be estimated w i t h some error.
The allowance of the p o s s i b i l f t y of calibration of the parameters
of distributed models implies that the estimated values may vary from
the 'true' model values and that these models are therefore
stochastic. However, the number of parameter values required by
distributed models poses a difficult calibration problem. Several
writers have pointed o u t t h a t the information content of a
rai nf all/discharge record alone must l i m i t the number of parameters
that can be satisfactorily estimated Ln the application of lumped
models. It is true that the calibration of a distributed model can
take much more information into account in the calibration process,
but it is not known how far this may overcome the problem of the
number of parameters t o be estimated,

The calibration or inverse problem has been the subject of

considerable study in groundwater modelling. Early solutions relied
on trial and error methods t o reduce the difference between observed
and calculated behaviour.
Subsequently more rigorous mathematical
methods Rave been developed (see for i n s t a n c e Garay e t al., 1976;
Guvanasen and V o l k e r , 1978; Smith and P i p e r , 1978) including methods
that try t o incorporate subjective InEotmation on the nature of the
system (Neuman, 1973).
Nethods have also been proposed for the
i d e n t I f i c a t l o n of reslstance functions in rivers by comparison of
observed and computed downstream discharges (Fread, 1975; Recker and
Yeh, 1973).

It i s clear, however, that the calibration problem increases In

difficulty with increasing complexity of the system, since due t o the
integrative nature of hydrological systems, the parameters may exhibit
inter-dependence such that changing different parameter values may
produce a similar effect. One study that simulated a complex
hillslope system at a w e l l Instrumented site (Stephenson and Freeze,
1 9 7 4 ) used t r i a l and error calibration and concluded:

"We recognise that our calfbration is less than perfect, but it

is probably representative of what can be attained when a f u l l y
deterministic mathematical model is a p p l i e d t o a f f e l d site u i t h
a f a i r l y complex, but as always imperfect, set of f i e l d
measurements" (Stephenson and Freeze, 1 9 7 4 , p 293).
Faced with this situation the success of the a p p l i c a t i o n of
physically-based distributed models must depend on defining parameter
valuea by a priori estimation of f i e l d measurement alone w i t h i n a
suf f i e i e n t l y close range t o the true model value so that the
calibration process assumes less significance and observed data can be
used to i d e n t i f y s p e c i f i c deficiencies in the m o d e l .
6.4

Sensitivity analysis

The need for an analysis of the s e n s i t i v i t y of model simulations

t o changes in parameter values takes on an increased importance in the

l i g h t of the parameter definition and model calibration problems

discussed above. The methods of stochastfc distributed modelling
discussed in Section 4 are d i r e c t l y applicable and can give confidence
limits around the predicted values of variables due t o variance in the
estimates of parameter values
However, these methods, and more
standard methods of sensitivity analysis (see for instance Vemuri e t
a l . , 1969; McCuen, 1973, 1974; Saxton, 2 9 7 5 ; Colman and De Coursey,
1976; Beven, 1979a) have only been applied t o single process systems
and extension to more complex syatems may not be easy. The use of
approximate sensitivity methode, in examining t h e e f f e c t s of changes
in certain parameter values w i t h i n their specified range, remains very
important for coupled systems.
The importance of sensitivity analysis in the application of
distributed models has been discussed in Section 3.5.
It is
re-emphasised that t h f s may prove t o be one of the greatest b e n e f i t s
of physically-based modelling for which estimates of parameter
uncertainty can be obtained independently of the model structure.
Sensftivfty analysfs allows estimates of the reliability of
determfnistic predictions to be made which, at a l a t e r stage in the
management d e c i s i o n process, should be incorporated into the
evaluation of risk associated with project designs

6.5

Plannina scenario evaluation

We shall now assume that a model has heen specified and been
successfuly validated accordfng t o some criterion of utility. Such a
model can then be used in a planning role to Lnvestigate, f o r example,
the ef feets of alternatfve development scenarios on the hydrological
regime of a river basin. A t t h i s stage t h e p h y s i c a l b a s i s of

distributed models and the possibility of s p e c i f y i n g parameters a

p r i o r i , at least wfthin some range, assumes particular significance
s i n c e the e f f e c t s of various changes t o the catchment or management
s t r a t e g i e s for the hydrologic system can be evaluated by changing the
model parameter values. The u t i l i t y of such predictions must be
evaluated by comparing the amount of p r e d i c t e d change wfth the
sensitivity of the predictions t o the estimated range of given
parameter values.

''TO

prophesy is e x t r e m e l y difficult

the f u t u r e

."

especially w i t h respect to

Chinese proverb.

7,

FUTURE PROSPECTS

It is l i k e l y that the future development of improved dietributed

modela will remain fragmented, with research aimed a t improving the
formulation and solution technique of modela of individual proceases
and the coupling between models of component processes. However, the
foundations of distributed d o e l l i n g are now sufficiently firmly laid
to enable general d i a t r i b u t d models, such as the SHE model described
above, t o be made available for more routine application. Successful
exploitation of these models will depend on a number of factors but
the main scientific criteria t o be satisfied are seen as follows.
(1) The development of techniques to enable both the efficient
estimation of parameters t o within sane hydrologically meaningful
level of accuracy, and the use of available data for calibration or
tuning of the model

The development of a sufficiently flexible frame structure t o

(2)
enable a wide range of problems to be tackled easily.
( 3 ) The addition of water quality (includfng sedfment) , components to
the hydrological model t o extend the range of potential problems that
can be tackled by SHE into an area that of ten re.qufres d i s t r i b u t e d
predictions. (Such additions are envisaged as a second stage of the
SHE project).

(4) The successul validation of the models (including the validation

of water quality components where appropriate) on a number of real
world problems, including problems posed in a water resource
management context where data may be expected to be limited, at least
prior to the application.

It should be noted that when the model is used specffically for

scenario evaluation a t the destgn phase of a management project, full
validat i o n can only take place following the implementation of
decisions taken on the basis of model simuLations, This last point
is, therefore, a particularly stringent criterion of success, since it
will t e s t the theore t i c a l concepts on which these physically-based
models rest,
The economics of distributed modelling in the future w i l l depend
on frnprovements i n modelling and parameter estimation techiquee, the
availability of improved computing techiques such as the more
widespread introduction of parallel processors, and improved methods
for handling large quantities of s p a t i a l d a t a in model specificatton,
calibration and validation. However, it seems likely a t t h i s stage
that future developments, rather than leading t o cheaper applications,
may enable the present generatfon of relatively crude d l s t r i h u t e d
models t o a t t a i n a higher level of sophistication. Indeed the v e r y

availability of the spatially distributed data required by t h i s type

of model may require a c r i t i c a l re-examination of the theoretical
constructs on which present modela are baaed. T h i s will be
particularly important in the area of sensitivity a n a l y s i s which is
seen as one of the most important functions of this type of model.
The limitations of the hydrological d a t a on which many management
decisions are based must be recognized, and should be taken i n t o
account in the decision-making process. P t e s e n t physically-based
models allow the effects of data uncertainty t o be explored using
crude parameter perturbation methods. In the future, improved model
formulations and improved knowledge af the spatial variability of
hydrological processee and parameters may allow this to be done in a
more rigorous and satisfactory m y .

ACKNOWLEDGEMENTS
We are grateful t o our colleagues on the SHE project a t the
Danish Hydraulic I n s t i t u t e , Sogreah and the Institute of Hydrology for
comments on an earlier draft of this report.

REFERTl{CES

Abbott, X.8., IferEn, R.I., Jenson, K.H., Jonch-Cleusen. T. aod

llavno, tr. 1979. Couplltr8 of ungaturated aod aaturated zone oodels.
Pepet predented et 18th IAHRcodgress, Cagllsrl, Ita1y.
Albertl, t.B., (1404-1472). Ten looLs on Archltecture,
J. Lont, Alec Tranrl Lrd, Lndon, 1955, 213-214.
lneln,

ll.,

1966, Str.enflor

r23-130,

tran6lEted by

routlng, geter Reourcesf,segrch, 2(1),

lneln, l'1., and Fs!8, C.S. 1970. Iipltctt

flood routl.nS tn naturet
chennels. J. Hydraultcs Dtv, ASCE,96 (Hy12), 248I-2500,
Aaderaod, 8.A., 1968. Develolmnt and testlng of snospac& ener8y
baLa.rce equatlotrs, Teter Reaoutcea n4aearch, 4(1), 19-37.
Appel, C.A. 1976. A note on coiputlnS flnlte dlfference lnterblock
tranatrl B6lvl tles , Water Reeourceanesearch, 12(3), 561.
DaLr, 4.4., ceLhar, L.W., Grtjahr, A.L. and l,taculllan, J.R., 1978t
Stochaatlc anelysls of apatlal vsrleblllty
ln subsurface flowe.
1.
Coupartaon of one-and-three dlonslonal flor.
llater Resourcea
Reaatch, 14(2), 263.
Dauer, ll.l., Illxson, lit.ll., Davt6, B.J. and Btherldge J.B. 1978. Area
eatloetlon of cropa by digltel enelys{s of Lndset deta,
PhotoSrsmetrlc EngineerlrB end Redote SenslnS, XLIV (8), 1033.
Becker, L. ard Yeh, W.tl.c, 1973. Identtftcetlod of rulttple reech
channel peradetere. llster ResourcesRBearch, 9<2r, 326,
Bertalenffy, L. von, 1960. C,eneral Systeoa theory, ceorge Bralzlller
Inc., N.Y ( page13).
Bevea, (.J. 1970. Ull.lslope hydtoSrphe by rbe finlte
Earth Surface Proceea, 2, 13-28,

elerent

nethod,

Beven, K.J. 1979a. A rensittvlty analysis of penoan-Hontelthactual

evapotten6plratlon estloetes, J, Hydrology, 44, f69-f90.
Beven, K.J. 1979b. 0n the generallsed klnematlc routing rethod, Iater
Resource6Reseach, 15(5), 123A-1242.
Bevefl, R. 1981. Coodentson 'A Btochastic-conceptual analysls of
ralnfell-rutroff
processes on a hlllslope, by R. Allan Freeze, Itarer
ResourcesResearch, 17(2), 43t-432,
Beven, (.J., ctlnan, l(. and ttesson, l,t.D. 1979. FIoo and flor' routlng
ln upla.d chennel8, Hydrologtcet Sctences Bultertn, 2t<3), 301-325.
Beven, K.J, and Ktrkby, M.J. 1979. A physicaUy-baBed vartabte
contrlbutlng area oodel of basln hydrology, Hyalrologlcal Sclences
Bulletln, 24(t>, 43.

2A

leven, K.J., warren, R, and Zaoul, J. 1960. SRE, TorardB e

nethodology for phyetcally-based' dlstrlbuted DodellLng tn
hydroloSy, h llydrolo8ical FotecestinS' IAHS hbltcetlon No 129.
Btbby, R, aod sudada, D.K. 1971. Statlstlcel errot analysls of a
nunericel oodel of cooflned groundwater flow' lo Stochastlc
Hydraullcs, c-L Chr.u(ed), Unlverstty of ?lttsbutgh, 591-612.
Box, c.P. erld Jenkl$s, C. . 1970. TlEe serlea Analy8l3, ForecaBtlng
and Control. Holden-Dey, Sao frenclaco.
Boyd, u.J,, Ptlgrto D.fl, and Cordery, I. 1979. A stota8e toutlng
Eodel based oo catchoent geooorphology. J. Hydrol. 42(314>' 2O9.
Brerll, L.8., SeoderB, T,C. ard l,loolhlser, f,.A. 1977. Klneostlc
paraoeteE estioatloo for lranaport of Pollutsote Ln overland f1os.
Hydrology SFipoaluD, Fort
Paper predented et 3rd Internetlonal
colllns, coLoredo,
Bredehoeft, J.D, and Plnder C.F, 1970. Dltltal anelyats of areal
f1.o[| ln nultl-squlfer
Sroundreter ayate$a: a quesl-thtee dlDenslonal
lfater
rodel,
RsourceaRsearch, 6(3), 883-888.
Bresler, E. 1973. Stoultsneous t.ansport of aolutee atrd nater undet
translent unaaturated flod condltloos.
lJetet Resoutceg Reaeerch,

9<4t,975.
Blelorel,
H. and laufer, A. 1979. 8leld teal of solutlon
Bresler,8.,
flos dodela 1.r a heterogeneoua trrtgated
cropped Bollr geler
Resourcea Research, 15(3), 645-652.
Burgesd, T.M. and Webstr, R. 1980. Opttoel
larlthdlc
BapptnS of sol1 ptopettlea.
II.
sctence, 31, 333-341.

lntetpolallon
and
8l.ock KriStng, J. Sotl

Celder, I.R. and Rosler, P.T.n. 1977. the de818n of large

plastlc-sheet
net ralnfell
8eu8e3, J. gydtol. 30, 403-405,
Chiu, C.L. 1968. Stoche8tlc open ch6nnel flor.,
Aner. Soc. Ctv. EnB., 94 (EM3), 811-822.

J. Eng. ltech. D1v.,

Chos, v.T. 1967. ceneral Report of Technlcel Seslon lib.1,

Proceedftgs Internatlonel
HydroloSy SFposl.uE, Fort Col1lna,
coloraalo, v2, 50-65 (page 58).
Clarke, R.T. 1973. l,lethenetlcel oode16 ln hydrology,
Dralnage Pepr 19, UN/!AO, RoBe.

Irrlg6tlon

and

Coloen, G. end D Cout8ey, D.G. f976,

Senstttvtty
and Eodel verlsnce
analysls applled to 6ooe eveporatlon and evapotr6nap{ratlon dodel,
I,later Reaourcee Reaeerch, f2(5), 873,
Corapcloglu, l,l.Y, and Brutssert, I'1. 1977. VlscoeleBtlc equlfer oodel
applted to subal.dettce due to purnplnS. lleter Re6outce6 R6eetch,
1 3 ( 3 ) , 5 9 7.

Crawford, N.H. end lln.ley, R.K. 1962. tte aynthest. of contlnuous

atrcelflon
hydrottaphs on e dltttal
cosputer.
Stenford Unlvelalty
Dapertnent of Clvll Eogttrecrtnt, Tcch. Repolt 12.
De8an, c. 1979, ltodels of grounduater flos ln statl6t1ca1lv
hotroSeneousporou! fotuations, Llter Resourcaa Regearch, iS(f),

/,2,

Delho@e, J.p. 1979. Spattel v.rtablltty

and uncerrahty ln
groundrst.r per66ters, Itetet R.loutc le.ealch, f:12), ZOfZaO.
Duane, T. 1978. fteld studles of htllslope flo, processea, Chepter 7
ln Etlldlope XydroloSy, U.J. Xlrlby (Bd), lrllcy.
EnSDen,E.T. end Ro8owskl, A.S. 1974. A perttal area nodel for stotu
flor sFtheBls.
lfater Resourc.a te.eerch, 10(3), 454.
Feddcs, R.A,, Kor.altt, p., tbnDsa, S.p. and Bresl.r, E. 1976. Fl.ntre
dlffreoce aid flalte eloetrt ahulatloa
of fleld seter uptak by
pleate. ItdroloSlcel ScteDcesBullrin, 21, 8f-9E.
FIe[1n8, c. 1975. Co6puter 6lEuletton technlques tn hydrology,
El6ev1er, NY.
!re6d, D.L. 1971. Dtacusslon of rlDpuclt flood routl.rlg ln natutal
channler by N locltr atrd C.S. Frtr8, J. EydrrulLcs DIv., her. Soc.
ctv. Bqs.,97 (na7), 1156-1159.
Fread, D.L. 1975, Co[putatlon of stage dlscherge telatlonshlpr
affected by unsteddy flon.,
W.t.r nsource. Eullettn, 11(2j,
2t3-22A.
Fleere, R.A. 1971, lttee-dt[nslotr!].
flor' ln a groundEter be6ln, g.ter
( pa8e 355)

trenaleot, seturated-unsaturated
neaourcea Rasearch, 7<2), 347-366

Freere, R.A. 1972a. Tte role of subsurfdce flow ln the leneratlon of

surfece runoff.
1, Be6eflor contrtbutlons to channel flow. tdeter
Reaource6Reseetch, 8, 609-623,
Fteeze, R.A. 1972b. The role of subsurfsce flo|. ln geoeratlog surface
runoff.
2. UpdtraaD soutce area6. water Reaources Resear;h, g(5),
1272.
Freze, R.A. 1974. Stornftoe generatlon.
Space phy8lcs, 12<4>, 627-547.

Revles! of CeoDhyalcsaad

Freere, R,A. 1975. A stochadtlc-conceptusl enalysls of

one-dherslonel
groondsater flo, in non-unlfod hoeodeneoueDedla,
Ileter nBourcesRe.earch, lt(5), j25-74.
Freeze, R.A, 1978. l,tathenattcat uodels of htl161ope hydrology,
Chepter 6 tn Elllslope HydroloSy, tI.J. Kirkby (ed), WI1ey.
Freze, R.A. 1980. A stochedtlc-conceptual enalysls of ralnfell
ruaof f - procesaea on a hillslope,
Water Resource6 Reseerch, 16(2),
391-408.

lo

Freeze, R.A. and llarlen, R.L. 1969' BluePrlnt tor a physlcally-based

digttally siroulated, hydrolotl.c-resPonse rodel'. J' ltydtology' 9'
237-258.
Garay, 8.L., Hoei0es
' Y.Y. dnd Dab, P. 1975. D{strlbuted ParaDeter
ldotlflcatlon of gtoundweter systerlB by non-llneet eBtiEetlon' J.
t{ydrology, 30 <ll2\, 47'
cash, J.H,C. end Morton' A.J. 1978. tn aPpllcetlon of lhe &rtter
dodel to the estlEetlon of the lnterceptlon loes fld Thetford
forest, J. ltydrology,38' 49-58.
Ca6h, J.H,C. snd Stewalt' J.B' 1975. ltre averege surfece realstence
of a plne forest dertved frco Bolten ratlo nessuteoent6, Bouodary
I-eyer Xeteorology, 8' 453-464.
celhar, L.w. 1974. stocbastlc eoelysls of Phreetlc aqulfers, water
Re6ourcesReserch, 10(3), 539.
clesal, w., Rlnger, u., Strebel, o. 1973. Nunerlcal treetnent of the
unsaturated sater flou quetton. cooPerlaon of exPerlaentel and
lfater Reaources Reseerch, 91 174.
codputed reBults.
of dlstrlboted
cuvenaaetr, v. and volLer, R.E. 1978. Identtftcatton
par:rnecers ln groundsatet bealos, J. ftydroloSy' 36, 279-293.
Ilaverkeip, R., vauclln, M., Touns, J.t lJlerenga' P.J. and Vdcheud, C.
1977. A cooparlaon of nunerlcsl slouletlon oodels fot
Soll Sci. Soc. her. J. 41, ZA5.
one-dlneaslodel lnflltratloo,
llaverkadp, R. and veuclln, U. 1979. A nole on estltretlng fhlte
dlfference lnterblock hydteullc conducttvlty velues fot translent
unsatnreted flow ptobleo. Itater Resources83earch, 15(1), 181'
Hayhoe, tt.N. 1978. Study of the relatlve effl.cledcy of flnlte
dlfterence and Galerkln techlques for riode111n8 8ol1 tater
trensfer.
Water RegourcesReseerch, 14(1), 97'
ttewlett, J.D. 1974. co@eots on letter6 relatlng to 'Role of
subaurface floe ln genetatlnS 8ulfac runoff, and upatrea source
Areasr, by R. Ai.1enFreeze, lfater Re6ource6Rseerch, 10(3) 605-607.
Etllel, D., Krento6, V,D. and Styllenou, Y, 1972. Ptocedure and
lestlng of an lnternel dralnage Eethod for dessurlng aoil hydraullc
characteristlcs ln Bltu, Sotl Sctence, 114(5) r' 395.
Ellls, R.c. end ReFo1d6, s.C. 1969, Illuatratlon6 of soll oolgture
varlebll{ty ln elected aree6 end plots of different alzca. J.
Hydtology,8(I), 27-47.
Hoa, N,T., Ceudu, R. and Thlrtlot, C. 1977. Influence of the
hystereglc effect on tran8lent f1o$6 ln saturated - uflEeturated
poroue nedla. water BesourceaResearch, 13(6), 992.
Itodge, R.A,L. and Fleeze, R.A. 1977. crouddosler flon 6ysLeds aod
slope stabtllty.
canadlan ceotechnlcdl Journal, L4(4), 466-476.

Buff, D.D. and RtuSr, P. 1967. tte chelt.cal and phydlcel p3laDeter'
ln a hydrologlc tredspott trodel for aedloectlve aerogola, proc.
Int. Eydrology Sy p, 1, 128-135, Fort CoUtns, Colorado.
Ilydrocotrp Inc. 1971.
2nd Edttton.

ltSP Operatlon. lhnuet,

palo Alto,

Cal1forn16,

Ibbltt,
R.P, 1972. Etfects of taadd data ertorB on th Deradetr
vshes for s cooceptoel rodel, getcr naources Reeearch, 8, 7G7g,
Jackson, R.D. Regneto, R.J. and ldson, S.B. 1976. ttiilng of ground
truth equleltlon durlng tebote gaae3&aentof soll vater conten!,
Reuote SenslnS of Envlrotr'nent, 4(4>, 249-255.
Jackaon, T.J, strd lrbcuen, n.g. 1979. Accuracy of bprvlou6 drea
valuea eatlaeted usldg reDotely scdsed data, lfatct Rrsoure6
Bulletln, f5(2), 436.
Jele8, L.D. 1972, Eydrologlc oodel.llng, peraiter esthstlon
rrdtershed ch.racterlsrlcs, J, Eydrolo8y, 17(4), 283-308.

and

Jarboe, J.E. and Haaa, C.T. L974, Caltbratln8 e w.ter yteld rodel for
d[all utrgauged tratersheds, Water Reaources Research, 1O(2), 256.
Jen6en, K.T. end Jonch-CleuseE, t. 1981. Unsaturated flos and
evapotrea3plratlo!
lodelll!8
as a colpoDeat of thc Europeed
llydioloSlcal SyateE (SflE), paper prese[ted at the Internetlonsl
Synpoalrh on R.lnfell-Runoff
UodelllnS, Ulssts.ippl
State
Unlverslty, lby 1981,
Jepp8on, R.W., Rat.l.s, tJ.J., liauon, If.R. end Schrelber, D.L. 1975. Use
of exlsymetrlc
lnftltretlon
iodl arrd fleld data to deterolne the
hydro1ogl.c61 ptoDcrtles of eollr, t{.ter nesources Reseetch, l1(l),
t27 .
Jonch-cl.eusen, t. 1979. Sy.toe gydrologlque Europaen: A Short
De6crlptlon, Danlsh Eydraullc Instttute.
Jordad, D.C., cravea, D.ll. and llalletter, M.C. 197E. Use of nenual
photogremetrlc
deosltooetly lo 16nd cover cL6a6lflcatlon,
EnSlneerlnS and ReDote Seastng, XltV (8), 1053.
Jou?nel, A.e. aod Buljbregrs,
Acadetrlc Pregs.

Ar J. 1978.

lltntdg ceostatt3ttcd,

Ketoltrlg, T.C. Devldson, J.U., Weaks, D.L. and Morrldon, R,D. 1977.
Preclalon s.lth rhtch selected 601.1Phyalcal peredeters cao be
estldeted, Soll Sclerce, 124, 241,
Klble!,
D.F. and Itoolhlser,
D.A. 1970. lhe klntretlc
cascade as a
hydrologic
Dodel, Colorado Stete lrnlvetslty
ltydroloSy papers l|o 39,
Fott Colllns, Colorado.
Klute, A. 1952, A nuoerlcal dethod for solvlng the flow equatlon for
water ln uaaaturated nater1a1, Soll Sctence, 73 (2), 105-116.

32

(napp, R.lt., creen D.lf., Pog8e, E.c. and Stanford c. 1975.

Developoent and field lestlnS of e ba6ln hydroloty 3hu1ator.
Re6ourcesxeseerch, 11(6) 879-888.

lfater

Kreft. A. aod Zuber, A, 1978. Deterolnatloo of aqulfer paroretete by

a t!'o-r'ell pulsed dethod uslng radloective tracers, co@enta, J.
HydroloSy, 41, 17r-176.
Iane, L.J. and Woolhiser, D.A. 1974. Slspll.ftcatton of oatershed
J. Hydrology, 35'
Seooetry affectlnS slnulatlon of surfsce runoff.

173-r90.

Lee, M.T. eod Delleur, J.l{. 1976. A vallab1e source lloear oodel of
strearn netoolk.
the ralnfell
ruooff ptocegs based on the ,ater6hd
!,leter ResonrcesRsearch, 12(5), f029.
trees, s.J. end llatson, K.K. 1975. lhe ure of a dependentdonaln oodel
of hysteresls in nuoerlcal so1l ttater studleg. Irater Reaourcea
R e a e a r c h ,1 1 ( 6 ) , 9 4 3 .
Luthto, J.N., othun, A. end Taylor, G.S. 1975, Coupled
ssturated-uosetutetd ttenslent flol, 1n porous uedle: experldntel
end ttrEerlc oodel3. Weter RBourcesResearch 11(b), 973.
Uechado, D. and Orllonnell, T. f979. A stochastlc lnterpretation of e
luoped overlaod flon Dodel, Ld uodelltna Hydrologlc Ptocesses'
Edtted by H.J. ltore1-Seytou!, J.D. Seles, T.G. Sander3 eod R.E.
So1th, lf6ter Re6ourceaPubllcetlons, Llttletoo, CoLorado, 259-269.
t{ccuen, R.H. 1973. Tte role of ensltlvlty
oodluog, J. $ydrolo8y, 18,37-53.

analy3ls ln hydroloSlc

llccuen, R.H. 1974. A senslttvtty snd rror anelysla of ptocedurea

used foi e6tloatlng evaporetl.od. l,leter Re3oureaBullettn, 10(3),
486-498.
Hetn, R.6., laurenaon, E.l,l. end Uc ahon, T.A. 1974. A sl.Dple
non-llnear oodel for flood eetloallon.
J. Hydreull.cs Dtv. ASCE,
100, 1507.
Montetth, J.L. 1965. Eveporetl.onend Envlroiment, S',bp. Soc. Exp.
8to1o8y, 19, 205-234.
ltorrie, E,ll. 1979. The effect of lhe sEell-slope epp.oxloetlon end
lolrer boundery condltlona on soluliond of the Salnt-Venant Bquatlons
J . H y d r o l o S y ,4 0 ( 1 / 2 ) , 3 1 - 4 8 .
l{ortla, E.N. and Codtrey, J.C. 1979. Ihe Europeenllydtologlcal SysteD
enot' toutlne, ir S.C. Colbeck and l.{. Ray, (Eds), Proc. }todelllng of
Soot' Cover Ru$off, CRREL,Henover, lilee Herp6htre, 269.
tlot.ls, E.l{. 1982. Sensl.tlvtty of the Europeannydrologtcsl Systeo
snoe oodel.s, Proc. SyEp. on HydroloSlcat Aspects of Alptne and Htgh}lountain Areas, IAHS PubL. 138, 221-231.
t{uztk, I. 1974. Laboratory erprloente ltlth surface runoff,
Htdtaultcs Dlvtsloo, ASCE,100, l{Y, 501-5f6.

J.

33

l{er.allhan, l.N, end liltherspooo,

dlffetencc uethod for enaltslng
Reaourc.. &!earch, l2(L), 57.

P.A. 1976. Atr lntlgreted ftntt.

fluld flos ln poroua Eedta, galcr

Nereahhen, T.I. 6nd Wltherspoon, p.A. 1977. ttrEerlc.l Dodel for

setureted-unaatureted flor ln deforuble
poloua ledie, 1 Thory.
llater Rlloulces R.6e.rch, 13(3), 657.
Nale.lahrn, T.ll. Iillthcr.pooo, P.A. .!d Edrlrd., A,C. 1978. l(ro.rlcel
[ode1 for aaturlted-unaaturated floe ln dafordable roroua ladlr.
2. ft. .l8orttlE.
lfstr R..ourc.. R.leerch 14(2), 255.
[..h, J.E. .nd &rtc],lffe,
J.V. f970. i,.v.r floe
conceptu.l lodcls.
J. ntdroloSy, 10, 2E2,
l{aturll Envlron[ent
London 5 vo16.

Reaeerch counctl

1975. llood

forec.sttnt

throu8h

studleg Repoit,

Nelaon, R.l{, 1978. EveluatlnS the cnvlrolllcntel

coturequencc3of
Srosudvatar conta[r.oetlon, Wstcr Raaourcca Reretch, 14(3), 409-450.
Neuu6tr, S. 1973. Celtbretlon of dlatrtbutad paraoeter ground$st!
flon lodefu vtered e6 a firltlpla
objactlve declalon proce6s uoaler
ui|c.rt.lnty.
lfeter Re.ourccB RG...rch, 9(4), 1006-1021.
lle6on, N.D. atd Errrlaon, J.c. 1978. CtEnoal studles ln the
PlFltoon experhental c.tch!at!.
Illtituta
of Eydrology Raport
47, w.lllqford.
t{tel8et, 0.R,, ElSger, J.W. 6nd Erk, K.t. 1973. Sperlal vartebtllty
of fleld-oc.dured,
prop.rtlar,BtlSerdl'.,
.oLlirater
42(7), 2t5-260,
Ptkul, U.F., Rd6oo, L eod Strc.tr R.L. 1974. A $nertcel oodel
bssed on couplad one-dlnenslodal Rlchard! atd Dousslnesq equatlonrt
lfet.r R!.ource. R.aeerch, I0(2), 295.
PtlgrlB, D.H, 1976. lravel thes rtrd rron-llneertry of flood runolf
frcd traccr leaaureienta o[ e ldtll
ratctshed] llater Rcaouiced
Reseereh, 11(3), 467-{96.
Ptndet, C,F. lnd Crry, ll.c. 1977, Flnlta EleEent Slfirlatlon
Surface lnd Sobaurface EydroloSyl Acadeol.c Preaa, London.

ln

PrelaaEann, A. end Zaout, J. 1979. L oodule 'Ecouleflent de 6urface"

due SysllD. Itydrologlque Europ6er (SHE), Peper pre3ented et l8th
IAHRContrera, Cegltart, Itely.
Prlee, R,K. 1974. Coopsrtson of four nuierlcaL [ethods for flood
routlng.
J. llydreultca Dlvlslon, ASCE, too ([y7), 879.
Prlckett, T.A. and Lnoqulat, C.C. 1973. Aqulfer sfuulatlon rodel for
use ofl dllk supported snall co[puter systeas.
Illtrots
State l,l6ter
Survey, Urbana, Clrculer 114.
Regen, R.l{. 1966. l,aboratoty evalugtlon of a nleerlcal flood routlng
technlqne for chenneLs sublect to latarel lnflows.
lfater Resourcs
R e s e e r c h ,2 ( 1 ) , 1 1 f - 1 2 1 .

34

Rlley, J.P., Chsdtlck, D.C. end l3raelsen, E.R. 1968. watetehed

slouletlon by electronlc anelo8 co[puter, ln "The Use of Analo8 snd
Dlgttsl Cooputers ln flydrolo8y, IAIS hrbn. No E0' 25-37'
Rogotrskl, A.s. 1972. wetershed phystc6r sol.l varlebtllty
!,tater ResourceaReaeatch, 8(4), f015-f023.

crllerla,

Ro.r8, 8.8., Contractor, D.N. and Shanholz, v.D. 1979. A flnlte

eleBent uodel of overland and channel floe for a33esa1n8 the
hydrologtc lilpeet of lendnl.e ch.o8e, J. HydtoloSy, 41(1/2), lf.
odal: the correletlon of
Ro6s. c.A. 1970. lte stanford lfet.rshcd
piocedur l.lth
valuea
aelected
by
a
cooputerlaed
?er6oeter
of lhe eetetshed. Rcsesrch
oeaaurable physlcal cha.ecterlatlcs
Itnlversity of lcntucky,
Report 35, Tater 8seelch Instltute
Lexllr8!on.
Rlcherdson, L.F. 1910. lhe epproxldate arllhdetlcal aolotlon by
ftnlte dlfferencea of physl.csl probleDs Lnvolvlng dlfferentldl
equdtlooa $tth an dppllcetlon to lhe streasea ln s iesonty da!.
Phtl Treo3. Royel Soc. S.r16s A,210,307-357.
of hyatereala
rioyer, J.M. 6nd vachaud, c. 1975. Pteld dter[lnallon
i.n sol.l eeter cheracterl.stlcs, soll sclcnce soc. Aoer. Proc., 39(2)'
Ruahton, K.R. and Toolioson, L.!1. 1977. PrDlsslble treah spacltg ln
J. nydroloSy, 34'
equlfer probleo .olved by flolt dtfferenc.3.
63-76.
Rushtoo, K.R. end lreddsbutn, L.A. f973.
aquifer slDul.stlon. Groundnat.r ff(1),

Stertlng
37.

coodl.tlons for

Rutler, A.J., ltorton, A.J., Roblns, P.C. 1975. A ptedlctr.ve @del of

lnterceptlon ln forests II.
relnf6ll
Gneiellaetlon of lhe oodel
and co perlson wlth obsewetlons ln aooe conlferous erd hetduood
6isnda. J. Applled Eco1o8y, 12, 367-80.
sagsr, B. 1978, Gal.erkttt flnLte eleDent ploccdure for an6lysla8 floe
throuSh randoo Eedla, lreter Reaource3 Reseerch, 14(6), 1035.
Sager, B. and Klalel, C.C. 1972, LlEtts of dcterElnlslle
predtctsblll.ty
of oeturaled flor equetlon6, ln Proc. 2nd SyDp oo
Frrndeoentala of Trenaport Phenoaeoaln Porous lcdte, Guelph, Cenade,
rAgR. r, 194-205,
Sdxton, K.E. 1975. SenBttl.vtty analyaes of the comblnalloo
tranapltatlon equellon. Agttc. l.leteoroloty, 15, 343-353.
Schnug8e,T. 197E. Reiote senrlnS of surfece soll ooldture, J,
Applted lbtorolo8y, 17(10), 1549-1557.
Selt$, H.U. 1978. Ttensport of reactlve rolutes durtng trenslent
ungaturatd eatet flou in ultt-layered solk, Soll Sctence, 126(3),
t27 .
Shev, F.S. and Southlrell, R.V. 1941. Relaxdllon oethodr epplied to
Problens reletlng to the bercoletlo! of
englneerlnS probleBB, VII.
Proc' Royal soc. se;tes A, 178.
ffirae tnro,rltr porous oeterlels.

Solth, P.J. end Plper, B.S. 1.978, A tron-litrar optht..tlon

!.rhod
for th estl.latlon of equlfr p.r&!etcrs,
J. nydtology, 39, 255-2]t.
Strl.th, R.E. end lloolhlser, D.A. 1971. Ov.laDd fld
on an
toftltrettng
surfece, Water ReeourceaReseetch, 7(4), 999-913.
Solonon, S.L 6nd Cupra, S.K. 1977. Dlstrlbuted nulerl.ce1 oodel for
estloatlng t_unoff and 6edltrent dl.acharge of ungauSed rlvt!.
ltater
Re6ourcesRe6.arch, 13(3), 619.
Stephenson, C.R. end trreere, R.A, 1974. l,{etheEstlc.l dltruletlon of
subulfsce flor. contrlbutlons to 6ltor,Ec1t rotroff, Reynol,da Creek
I,leterahed, Idaho. Itetr Resoutces Reseerch, LO(2t, 284-29A.
Steeart, J.B. 1977. Eyaporatton fro! the rrat canopy of a plne foEestr
wster Reaource3Reaeerch, 12(b), 915.
Stoker, J.J. 1957. Weter Ueves, Intalaclenca press, tO4,
Strelkoff, T, 1970. NuDerlcel solutlon of the St Venant equ6tlons,
J. flydruallcs Dlvl.slon, ASCE,95 (8Y1) 223-251.
Surkan, A,J. 1974. StEulatlon of 6tor! veloclty effectg on flol. frou
dtstrlbuted chenoel ntrorks, Iteter lsouice6 Reaeich, lO(O), 1149.
Scelcz, G. Endrlodl, c. and Tejchlao, S. 1969. Aerodynaltc end
eurfac fectora ln evaporetlon, W.tcr nrsourcea Research, 5(2),
380-394.
Scelcz, c. end Long, I.F. 1969. Surfsce leslstsnce of clop csnoples,
llater ReaourcedReaearch, 5(3) r 622-630.
Tan, C.S. and Dlrck, T.A. 1976. F6ctor. effectlnS tte cenopy
reBlstance of a Douglas Flr forest, Eoundary-Iayer ltteorolo8y,
475-488,

lO,

TanS, D,E. eod Flnder, c.E. 1977. SlDulatloa of grouodeete. floe and
6a5a tfanaport utrder unceital.nty, Mvences h llatel Re6ou!c.3, I(l),
25-30.
Teng, Y.F. end Skeggs, R.l.l. 1977. hperloental evaluetlon of
theoretlcal aolutlong fot subsurface drdtnage end lrrigatlon.
Resourcednesearch 13(b), 957.

l,l6ter

Todini, E. and l,Iell{s, J. 1977. Ustng CLS for ddlly or lon8er pertod
ralnfall-ruooff
node111n8, 1n ltsthatrstical ltodela for Surf6ce l,leter
Hydrology (Eds. T.A. C{rlant, }tal.ooeU. and nallts, J.R.), rttey S
Son6, London, 1.49-168.
Treacott, P,C. end Larsor, S.p. 1977. Collpetlson of lteietlve
for solvlnS 2-d1Den6lona1groundvatei flow equatlons, lletet
Resource6Resedrch, 13(1) , 125,

oodI6

Vandeoberg,A. 1977. Puop testlnS ln heteroSeneonsaqulfers, J.

Itydrology, 34, 45-62.

venurl, v., DrscuP, J.A.' ErdEann, R.C. atld vetrurt' N' !969'
and lts
analy6ls nethod of sysieo ldentttlcatlo$
Sensltlvity
Research,
5(2)'
polentlal
ln hydrologlc teeeelchr l{ater Re6ources

34r-349.
Volker, R.E. 1969' Non-ltnedr f1o!, ln porous oedla by flntte
eleBents. J. Hydreullcs Dlvtslon' ASCE'95 ([Y6)' 2093'21f4'
whlsler, F.D. and (lute' A. 1965. The nnoetlcal ane1ys13of
lnflltratlon
considerlng hysteres{E lnto a vertlcal 6011 cohidn at
Sotl Scl. Soc. ,{oer. Proc., 29' 489-494'
equiltbrtud under Srsvtty'
wlgSert, D.C, and wylte' E,B. 1976. llulerlcel ptedlctlone of
2-dloenslonal trenslent Sroudwatr floe by the oethod of
IJeter ReaoulceaRaaearch, 12(5) ' 971.
charectetl stlca.
llll1ard6oo, L.S. and l{tltst r R'L. 1965. Sauple size estliates ln
petueabtltly studle.' J. IrrlSatlon and Dtalne8e Dlvl8lon, 91 (lRf)'
1-9.
wll3on, C.8., valde3, J.B. snd RodrlSuz-Iturbe, I. 1979' On the
on storn nunoff.
of ralnfall
lnfluence of the gpsttel dlstrlbutlon
IJater Reaoulces Rese.rch, f5(2), 321.
Youog, c.i. 1971. converSeocc crlterls
Ifater Resources Reoeerch, 7(3), 536.

for ltretive

zereilny, ff, 1978. Boundary coodltlonr tn dodelltng

So11 Sclence, 125(2), 75.
un.atursted sotls.

toutlnS oodels,
Iratet flo!, Ln