Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
ABSTRACT
Diskin, M.H. and Simon, E., 1977. A procedure for the selection of objective functions for
hydrologic simulation models. J. Hydrol., 34: 129--149.
Hydrologic simulation models are calibrated by comparing observed data with data
generated by ;he models. The comparison is made in an optimization procedure using an
objective function adopted for that purpose and a set of data which is a subset of all data
available or observable. The choise of the set of data and of the objective function to be
used for any given model is a subjective decision which influences the values of the model
parameters and the performance of the model. The set of data used should be comparable
to the engineering application for which the model is intended. The objective function can
be chosen by a procedure outlined in the article in which a number of possible functions
are considered and compared with reference to one or more engineering applications.
INTRODUCTION
R e c e n t y e a r s h a v e w i t n e s s e d a large i n c r e a s e in t h e n u m b e r of c o n c e p t u a l
h y d r o l o g i c m o d e l s p r o p o s e d or a d o p t e d f o r t h e s i m u l a t i o n o f t h e b e h a v i o r o f
w a t e r s h e d s . T h e s i m u l a t i o n m o d e l s c o n s i d e r e d h e r e i n are t h o s e b a s e d on w a t e r
balance equations applied at discrete time intervals to the elements of the
m o d e l s as well as t o t h e e n t i r e m o d e l . M o d e l s d e s c r i b e d in t h e t e c h n i c a l litera-
t u r e r a n g e f r o m t h e v e r y c o m p l i c a t e d c o m p r e h e n s i v e S t a n f o r d m o d e l (Craw-
f o r d a n d L i n s l e y , 1 9 6 6 ) to t h e v e r y s i m p l e s p e c i a l p u r p o s e l i n e a r r e g r e s s i o n
m o d e l ( D i s k i n , 1 9 7 0 b , D i s k i n e t al., 1 9 7 3 ) . T h e p o t e n t i a l user o f h y d r o l o g i c
m o d e l s m a y be c o n f u s e d b y t h e w i d e c h o i c e o f m o d e l s a v a i l a b l e a n d t h e
n e c e s s i t y t o c h o o s e b e t w e e n v a r i o u s m o d e l s all o f w h i c h a p p a r e n t l y p r o d u c e
e q u a l l y g o o d results. S o m e g u i d e l i n e s are t h u s n e e d e d f o r t h e b e n e f i t o f t h e
users o f h y d r o l o g i c m o d e l s . T h e s e will be u s e f u l also t o t h e h y d r o l o g i s t w h o
m a y wish t o p r o p o s e or d e v e l o p y e t a n o t h e r m o d e l .
The basic rule for the selection of a hydrologic model is, according to
Dawdy (1969), to adopt the simplest model compatible with the uses for
which the model is intended. Except for some obvious simple cases (Diskin
et al., 1973) the development of a model to meet a given set of conditions
and demands is not a simple task. The above rule is, however, useful for
practical applications only if some yardstick or measure is available for
assessing the degree to which the model meets the requirements of its users.
To avoid bias, it is advisable to develop the measure for the utility of the
model with reference to a number of objective functions and not only the
one used in the optimization procedure.
Following Box and Jenkins (1970), Diskin (1970a), Dooge (1972), James
(1972), and Diskin et al. (1973), the following procedure may be formulated
for the selection of a hydrologic model.
(a) Define the problems to be solved by the model or the range of applica-
tions for the model.
(b) Define the processes which take place in the watershed, their possible
effect on the applications intended for the model, and the constraints on
these processes.
(c) Formulate the various procedures proposed or available for the solution
of the problems to be solved.
(d) Select the type of model likely to meet the required applications, and
define the extent of the phenomena to be represented by the model.
(e) Define the structure of the model and the equations representing its
various components.
(f) Select the time increments used for representing the input and o u t p u t
data and for carrying out the computations.
(g) Select an objective function to be used in the optimization scheme.
(h) Calibrate the model with the aid of past records using a suitable opti-
mization procedure and the objective function adopted.
(i) Examine the performance of the model in the light of its intended uses.
(j) Refine the model if necessary. This may include introducing changes in
the structure of the model and in the time increments used.
(k) Apply the model to the problem or problems for which it was intended
and evaluate the results produced.
Steps (e)--(j) may be repeated a number of times in the light of results ob-
tained in the final stages of the analysis.
It should be pointed out at this stage that the selection of an objective
function for the optimization procedure is in itself a subjective decision
which influences the optimal values of the model parameters. Thus the opti-
mal set of parameters is optimal only in the context of the objective function
selected.
In this article the problem of the selection of an objective function is
examined and some recommendations are outlined for the method of choos-
ing the function to be used. The objective function is normally defined as a
function of the difference between computed and observed data during the
131
calibration period. In the present work the comparison is based n o t on all the
measured data b u t on a specified subset of them. The subset used in any given
case is such th at the objective f unc t i on becomes oriented towards a certain
engineering application. Moreover, in some cases the comparison may be
made n o t between the c o m p u t e d and observed values but between param-
eters derived from these.
OUTLINE OF STUDY
The three watersheds used are shown in Fig.1 and some o f their properties
are listed in Table I. These watersheds were chosen mainly because of the
availability of rainfall and r u n o f f data of acceptable quality but they repre-
sent also different conditions within the broad climatic classification of semi-
arid watersheds.
The Ekron watershed near Bet-Elazari is a 62-km 2 watershed located in the
low hills area o f the coastal plane 25 km south of Tel Aviv. Most of the water-
shed has alluvial soils which are cultivated. The Upper Sorek watershed near
Ein Karem reservoir is in the central mountain range. It has an area of 79 km ~
and it is located a bout 10 km NW of Jerusalem. A small part of the watershed
is actually within the builtup area of the city. Most of the rest of the area is
uncultivated and includes some w o o d e d hills. The soil in the watershed is
terra-rosa and brown soils of small thickness overlying karstic limestone for-
mations. Some farming is carried out in the valley and in narrow terraces on
the steep slopes of the watershed. The Kishon watershed upstream of the
Kefar Baruch reservoir is a fairly large watershed (468 km 2) located about 45
km SW o f Haifa. A large portion o f the watershed is within the flat area of
the valley of Yezre'el characterized by heavy clay soils. The rest of the area is
on the steep slopes o f the Efrayim range. The low parts of the watershed are
134
0 10 20 30 40 50 km
LAKE
TIBERIA:
25O HAIFA
200
~o TEL
so.
EKRON
JERUSALEM/
lo.__~o
f 100
I
Fig.1. L o c a t i o n m a p .
150
I
200
TABLE I
Watershed
of each watershed by the Thiessen method, and mean monthly values of daily
evapotranspiration for each watershed, which were in the range of 1.8--4.0
mm/d. These average values were derived from actual observations of daily
evaporation from Class-A pans multiplied by a correction factor of 0.78.
Because of the long rainless season the various storage elements in the models
were assumed to be empty at the beginning of each rainy season.
R u n o f f data used were of two kinds. One was actual daily runoff, measured
by the Hydrological Service of Israel and published in its Hydrological Year-
books. The other was synthetic daily runoff data produced by the Tahal
simulation model which is a local adaptation of the Stanford mark IV model
developed by the Water Planning for Israel Company (Tahal). The parameters
used in the Tahal model for generation of these synthetic data were derived
by an optimization procedure using actually measured daily r u n o f f data. The
reason for using the synthetic data instead of the measured values was the
desire to eliminate noise in the measured data. It was thought that the use of
noiseless data could give a better basis for the various comparisons. Compari-
sons and conclusions based on measured data were found, however, to produce
the same results as those based on synthetic data.
One of the problems considered in the study was the effect of the length
of the data series. For this purpose two series lengths were considered, a short
series of 4 years and a long series of 16 years duration. While most years selected
for each of these series in each watershed were consecutive, this was not so in all
cases. It was considered more important to include years with high, medium
and low r u n o f f rather than have consecutive years. This procedure was justi-
fied because the annual long rainless periods made each year's r u n o f f indepen-
dent of previous years. Results reported in the example given below are those
related to the short series of data. Similar findings were, however, obtained
with the long series of data.
136
EI r+
+ I ' ~I +FC
VIMI+-+~ ~L/4IAG'OX~ +AL
El P
+ I+v Ip~LL~ ~ '
I \
[ + R2
i
SM
XK __ : R 1
XC t__ R
t J
1+
Fig.2. Structure o f m o d e l 12. Fig.3. Structure o f m o d e l 15.
138
This value is also the initial value used as a starting poi nt for next day's com-
putation.
139
The four parameters of model 15 (SM, SA, XK and XN) have by definition
only positive (or rather non-negative) values, with the additional requirements
that SA < SM and XK ~< 1.0. The maximum value of XN is unlimited. In prac-
tical cases the value of XN is expected to be in the range of 0.2 ~< XN ~< 3.0.
The procedure for the selection of the most suitable objective function is
illustrated below with reference to model 15 using a data set identified in the
study as engineering application 1 for the Ekron watershed. To facilitate com-
parison between the observed and generated values this set of data included
runoff values for all days on which either the measured or the generated
values were not zero.
As stated above, the first step in the analysis was the calibration of the
model selected for each watershed, using daily non-zero r u n o f f data corre-
sponding to engineering application 1. The results consisted of the optimal
parameters P] for each of the objective function U1 used. The set of optimal
parameters for each objective function was derived by the pattern search
method of optimization described by Green (1970). The starting point of the
search, i.e., the initial values of the parameters, were the same for all objective
functions considered. At the conclusion of this stage of the analysis values of
Wjk were c o m p u t e d for all objective functions using the sets of optimal
parameters PJ obtained by the pattern search procedure. In some of the cases
it was discovered that only local minima were reached since some of the Wlk
values were lower than the minimum computed for the optimal set of param-
eters. In these cases the pattern search procedure was again employed to find
a new set of optimal parameters. The starting point in these cases were the
set of parameters --Pi which gave the lowest value of Wlk. New values of Wlk
were computed with the new set of optimal parameters and the whole process
was repeated if necessary.
An example of the results obtained at this stage is given in Tables II and
III. Table II lists the values of the parameters for each of the objective func-
tions listed in Appendix I. Also given are values of the mean yearly flow as
derived from the generated data using the corresponding sets of optimal param-
eters and observed rainfall data for the 4 years period used in the calibration stage.
The values listed should be compared to the observed mean value of
2.983.106 m 3. Table III gives the values of each of the various objective func-
tions c o m p u t e d with the aid of data generated by model 15 using each of the
sets of parameters listed in Table II.
The next stage in the analysis was to replace the actual values of Wlk by
ranks Rjk according to the magnitude of the items included in each group of
Wjk values identified by a given value of the index k. The table of ranks derived
from Table III is shown in Table IV for the Ekron data. In cases where two or
more entries in Table III have the same rounded value, the corresponding ranks
140
'FABLE II
Optimal parameters and mean annual flow for model 15 according to various objective
functions using short-period data for the Ekron watershed
j = 1 2 3 4 5 6 7 8 9 10 11 12
k = 1 0.176 0.371 0.178 0.217 0.182 0.385 0.185 0.514 0.182 0.226 0.291 0.225
2 119.3 83.2 106.8 97.7 146.5 83.3 146.8 89.9 146.4 135.8 102.6 100.7
3 0.333 0.462 0.330 0.360 0.351 0.471 0.352 0.554 0.351 0.380 0.409 0.368
4 0.205 0.247 0.203 0.193 0.215 0.250 0.215 0.297 0.215 0.219 0.271 0.219
5 0.039 0.036 0.050 0.011 0.0001 0.040 0.005 0.028 0.0005 0.089 0.159 0.061
6 1.326 1.015 1.227 1.149 1.526 1.014 1.530 1.057 1.526 1.492 1.252 1.170
7 0.026 0.055 0.026 0.032 0.026 0.057 0.026 0.076 0.026 0.033 0.043 0.033
8 1.633 1.344 1.580 1.536 1.839 1.347 1.854 0.942 1.839 2.123 1.702 1.307
9 0.067 0.239 0.100 0.119 0.002 0.252 0.016 0.146 0.0007 0.134 0.274 0.165
10 0.173 0.404 0.212 0.243 0.177 0.424 0.215 0.288 0.176 0.144 0.403 0.315
11 2.157 2.680 2.211 2.294 2.805 2.751 3.051 2.559 2.808 1.109 0..403 2.567
12 0.059 0.060 0.058 0.058 0.064 0.061 0.064 0.058 0.064 0.085 0.073 0.053
142
TABLE IV
it = 1 1 10 2 6 4 11 5 12 3 8 9 7
2 8 1 7 4 11 2 12 3 10 9 6 5
3 2 10 1 6 4 11 5 12 3 8 9 7
4 3 9 2 1 4 10 6 12 5 8 11 7
5 7 6 9 4 1 8 3 5 2 11 12 10
6 8 2 6 4 11 1 12 3 10 9 7 5
7 2 10 3 6 5 11 1 12 4 7 9 8
8 7 3 6 5 10 4 11 1 9 12 8 2
9 4 10 5 6 2 11 3 8 1 7 12 9
10 2 11 5 7 4 12 6 8 3 1 10 9
11 3 8 4 5 10 9 12 6 11 2 1 7
12 5 6 3 4 9 7 10 2 8 12 11 1
Tot~ 52 86 53 58 75 97 86 84 69 94 105 77
TABLE V
Merit numbers assigned to various objective functions using the short series of data
1 1 4 1 1 6 6 19
2 5 9 7 5 4 1 31
4 2 5 2 3 1 2 15
5 9 6 9 8 5 7 44
8 6 8 8 9 8 5 44
9 4 3 5 4 2 3 21
10 7 1 4 2 7 8 29
11 8 2 3 5 9 9 36
12 3 6 6 7 3 4 29
b e t w e e n t h e m a t h e m a t i c a l f o r m u l a t i o n of t h e o b j e c t i v e f u n c t i o n s and t h e
engineering a p p l i c a t i o n . T h e s o u r c e o f this link can be t r a c e d to t h e t y p e o f
d a t a selected for each o f the engineering a p p l i c a t i o n s used.
CONCLUSIONS
T h e d e v e l o p m e n t of h y d r o l o g i c s i m u l a t i o n m o d e l s requires a f e w m a j o r
decisions on t h e p a r t o f the h y d r o l o g i s t w h o is engaged in this task. T h e s e
include t h e d e f i n i t i o n o f t h e s t r u c t u r e o f the m o d e l and t h e s e q u e n c e o f c o m -
p u t a t i o n s used t o c o n v e r t the rainfall i n p u t into r u n o f f , the a d o p t i o n o f a
basic t i m e i n c r e m e n t f o r c a r r y i n g o u t the c o m p u t a t i o n , and the selection and
d e f i n i t i o n o f t h e o b j e c t i v e f u n c t i o n for t h e o p t i m i z a t i o n p r o c e d u r e a d o p t e d
f o r t h e c a l i b r a t i o n o f the m o d e l . Each o f these decisions, has an i m p o r t a n t
e f f e c t on t h e results o b t a i n e d with t h e m o d e l . T h e decisions are, h o w e v e r ,
m a d e in m a n y cases o n the basis o f i n t u i t i o n and e x p e r i e n c e o f t h e h y d r o l o g i s t
and are t h u s n o t free f r o m subjective c o n s i d e r a t i o n s .
T h e a i m of t h e p r e s e n t s t u d y was to investigate the e f f e c t s o f t h e o b j e c t i v e
f u n c t i o n s e l e c t i o n and t o arrive if possible at s o m e r e c o m m e n d a t i o n s or guide-
lines f o r this selection w h i c h will r e d u c e t h e a p p a r e n t s u b j e c t i v i t y involved.
T h e results o b t a i n e d are p e r f o r c e l i m i t e d b y the t y p e o f d a t a and m o d e l s in-
c l u d e d in the s t u d y b u t it is e x p e c t e d t h a t similar studies using o t h e r t y p e s
o f d a t a or o t h e r m o d e l s will yield similar conclusions.
T h e m a i n c o n c l u s i o n o f the s t u d y is a d e m o n s t r a t i o n o f t h e i n a d e q u a c y o f
a single o b j e c t i v e f u n c t i o n to serve as a universal t o o l f o r the o p t i m i z a t i o n o f
h y d r o l o g i c models. It has b e e n d e m o n s t r a t e d t h a t t h e r e is a definite link
b e t w e e n t h e m a t h e m a t i c a l f o r m u l a t i o n o f an o b j e c t i v e f u n c t i o n a n d the t y p e
o f e n g i n e e r i n g a p p l i c a t i o n for w h i c h the m o d e l is used. It a p p e a r s t h a t b e t t e r
results are o b t a i n e d if the o b j e c t i v e f u n c t i o n is selected a c c o r d i n g to t h e
144
engineering application for which the results will be used. For example, ob-
jective functions U1, U3 and UT, all based on the sum of squared deviations,
produce the best results for each of engineering applications 1, 3 and 4. These
three applications refer respectively to all daily non-zero r u n o f f values, to the
m a x i m u m m o n t h l y values of daily runoff, and to all daily r u n o f f values
higher than 0.5.106 m 3. The same objective functions give fairly poor results
if used for other engineering applications such as application 5 referring to
all daily r u n o f f values lower than 0.5.106 m 3 or to application 6 which is
based on the flow duration curve. Other combinations of objective functions
and engineering applications which produce good results are identified by the
merit numbers listed in Table V where a low number signifies a desireable
combination. The user of a simulation model should thus calibrate his model
with such an objective function that is best capable of generating the type of
data that interests him.
Another important outcome of the study is the desireability of considering
more than one objective function in the optimization procedure for a given
model and a given engineering application. There are two reasons for this
recommendation. One is that the best function for the given application is
not known in advance. The second reason is that an optimization procedure
utilizing a number of objective functions (in prallel) is less likely to lead to a
local minimum for the one objective function adopted.
Considering the optimal sets of parameters obtained by the various objec-
tive functions (Table II), it appears that the ranges of values obtained for any
parameter are relatively small. The same observation was repeated in similar
tables, not reproduced in this article, prepared by using other watersheds and
other models. These results point to the great sensitivity of the models con-
sidered to the values of the parameters since relatively small changes in the
values of the parameters lead to large change in the values of the objective
functions. Another conclusion drawn from Table II is the ineffectiveness of
using the mean annual flow as a criterion for the selection of model param-
eters. It appears that results very near to the observed value may be obtained
with almost any set of parameters derived from the various objective functions
considered.
Each of the objective functions considered in this article may also be used
as a measure of model effectiveness. In this role the proximity of the com-
puted value of this function to its ideal value is used as an additional criterion
for the selection of the best objective function for a given application.
The results presented herein do not lead to a recommendation that any one
objective function is the best for all cases. The procedures developed contribute,
however, to the reduction of the subjectivity involved in choosing an objective
function for a given case. It is thus possible to choose an objective function
that will be significant to the engineering application considered and to arrive
at an optimal set of parameter values that is nearer to the global optimum
than the set obtained with other procedures.
145
ACKNOWLEDGEMENTS
The paper is based on material prepared for and included in a D.Sc. Thesis
written by E. Simon (1974), with M.H. Diskin acting as research supervisor.
Partial financial support for the research project was obtained from the
Research Fund of the Faculty of Civil Engineering, Technion -- Israel Institute
of Technology, from the Israel Hydrological Service and from a research grant
of the I.B.M. Company. Assistance in developing and running the c o m p u t e r
programs was received from the C o m p u t e r Center of the Technion and from
personnel o f the D e p a r t m e n t of C om put e r Sciences in the Technion.
APPENDIX I
APPENDIX II
APPENDIX III
REFERENCES
Box, G.E. and Jenkins, G.M., 1970. Time Series Analysis, Forecasting and Control. Holden
Day, San Francisco, Calif., 553 pp. (see especially Ch. 1, pp. 1--19).
Crawford, N.H. and Linsley, R.K., 1966. Digital simulation in hydrology; Stanford water-
shed model IV. Dep. Cir. Eng., Standord Univ., Stanford, Calif., Tech. Rep. No. 39.
Dawdy, D.R., 1969. Considerations involved in evaluating mathematical modeling of urban
hydrologic systems. U.S. Geol. Surv., Water Supply Pap. No. 1591-D.
Dawdy, D.R. and Thompson, T.H., 1967. Digital computer simulation in hydrology. J. Am.
Water Works Assoc., 59(6): 685--688.
Dawdy, D.R., Lichty, R.W. and Bergmann, J.M., 1972. A rainfall--runoff simulation model
for estimation of flood peaks for small drainage basins. U.S. Geol. Surv. Prof. Pap.,
506-B.
Diskin, M..H., 1970a. Objectives and techniques of watershed modeling. Proc. ARS--SCS
Workshop on Watershed Modeling, Tucson, Ariz., pp. 9.1--9.35.
Diskin, M.H., 1970b. Definition and uses of the linear regression model. Water Resour.
Res., 6(6): 1668--1673.
Diskin, M.H., Buras, N. and Zamir, S., 1973. Application of a simple hydrologic model for
rainfall--runoff relations of the Dalton watershed. Water Resour. Res., 9(4): 927--936.
Dooge, J.C.I., 1972. Mathematical models of hydrologic systems. Proc. Int. Symp. on
Mathematical Modeling Techniques in Water Resources Systems, pp. 171--189.
Green, R.F., 1970. Optimization by the pattern search method. TVA, Division of Water
Control Planning, Knoxville, Tenn., Res. Pap., No.7.
James, W., 1972. Developing simulation models. Water Resour. Res., 8(6): 1590--1592.
149
Lichty, R.W., Dawdy, D.R. and Bergmann, J.M., 1968. Rainfall--runoff model for small
basin flood hydrograph simulation. Proc. Symp. on The Use of Analog and Digital
Computers in Hydrology, Tucson, Ariz. IASH Publ. No. 81, pp. 356--367.
Simon, E., 1974. A basic study of conceptual models for the hydrologic behavior of
watersheds and their engineering applications. D.Sc. Thesis, Technion -- Israel Institute
of Technology, Haifa. (In Hebrew, with English summary.)
Simon, E. and Diskin, M.H., 1975. Objective function formulation and their effect on
hydrologic simulation models. Faculty of Civil Engineering, Technion -- Israel Institute
of Technology, Haifa, Publ. No. 216.
Yevjevich, V., 1972. Probability and statistics in hydrology. Water Resour. Publ., Fort
Collins, Colo., 302 pp.