You are on page 1of 7

GUEST EDITORIAL

OF CARTS AND HORSES IN HYDROLOGIC wrong-more often it is totally absent. One is reminded here
MODELING of a comment physicist Niels Bohr is supposed to have made
to one of his students: "This is not right ... This is not even
Historical Perspective wrong!"
Hydrologic modeling is a relatively new concept and some- Hydrological discoveries of the old school invariably were
times only a more impressive label for what used to be called the product of extensive personal involvement of its authors
"methods of hydrological computations" until the 1960s. in practical hydrological work related to water resource engi-
Thus, one would search in vain for "hydrologic models" in neering projects. Their papers have the unmistakable imprint
such applied-hydrology classics as Linsley et a!. (1949, 1958). of mature professionals going seriously about their business
Even the famous Handbook of Applied Hydrology (Chow and, in the process, getting ideas on how the current practice
Downloaded from ascelibrary.org by University of Aberdeen, Bedford Road on 03/21/15. Copyright ASCE. For personal use only; all rights reserved.

1964) mentions the term "hydrologic model" in only one of their profession could be improved. One can see that these
fourth-order heading in the context of statistical and probabi- people were not playing games.
listic data analysis-indeed, the only sentence explicitly re- On the other hand, this is exactly the impression one gets
ferring to hydrologic models that I could find in this from reading papers of the academic hydrologic modelers of
monumental volume is this: "Hydrologic models considered the new generation. While most of them have engineering de-
here are mathematical formulations to simulate natural hydro- grees, they have never practiced engineering and, while pos-
logic phenomena which are considered as processes or as sys- turing as hydrologists, they have, for the most part, never se-
tems." riously studied or practiced hydrology either. As I once
And that is what they are or at least what they are meant characterized a typical academic hydrologic modeler: "Sus-
to be: mathematical models of hydrologic processes or sys- pended between a technology he does not practice and a sci-
tems. ence for which he has not been trained, his 'research' is nat-
The "upgrading" of "computational methods" to "math- urally guided to performing elaborate pirouettes on the high
ematical models" was brought about by the advent of the wire of techniques connecting the distant poles and holding
computer and the consequent "mathematization" of many him in place" (Klemes 1988a).
kinds of analytical and technical work. This produced a burst Having come to hydrology after practicing water resources
of activity in mathematically literate university departments engineering for more than a decade, during which I gradually
such as engineering: computer-based mathematical modeling realized how little hydrology I knew and how important its
made attractive to them previously uninteresting empirical dis- knowledge was, I was eager to drink from the fountain of
ciplines that had low mathematical content and relied heavily hydrologic knowledge I thought must be hidden behind the
on manual execution of various numerical and graphical tech- intimidating jargon and algebra-all those convolutions, La-
niques. Such disciplines had evolved from practical needs of guerre functions, fractional and other noises, and so On. Alas,
various professions, remained largely within their domain of there was no fountain, only sterile desert! After a dozen years
interest, and, in the academic environment, were relegated to of exposure to practical hydrological problems and participa-
the fringes of scientifically more respectable fields. tion in neck-breaking decisions on multimillion investments
Hydrology was, of course, a prime example of such empir- made with the knowledge of the large hydrological uncertain-
ical disciplines, having evolved chiefly from hydraulic engi- ties involved, I was literally insulted when I realized that all
neering and its need for hydrology-related design parameters what this high-power prestidigitation, posturing as "scientific
and operational characteristics of various water resource hydrology," can teach me is how to fit a line to a few points,
projects. Here I am deliberately using the term "discipline" extrapolate a curve by the most abstruse and esoteric means
because even today hydrology is only slowly asserting itself available, and keep calibrating (i.e., fudging) an arbitrary sys-
as a science in its own right. These topics were discussed in tem with an excessive number of the degrees of freedom until
greater detail elsewhere (Klemes 1986, 1988). it fits some hydrologic record.
For hydrology as a science, the invasion of mathematical I soon made my displeasure known, deploring' 'the empha-
modeling was nothing short of a disaster. It has retarded rather sis .,. on the fitting of various preconceived mathematical
than advanced the development of hydrology because, with models to empirical data rather than on arriving at a proper
very few exceptions, it focussed all efforts on polishing the model from the physical nature of the process itself," warning
mathematical and computational aspects of methods and tech- that "such approach can hardly contribute to hydrological
niques, leaving the understanding of the substance at the 1930s knowledge," drawing attention "to the fact that inferences
level, where it had been brought by the old guard of profes- about physical features of a process, based on operational
sionals like Hazen, Sherman, Horton, Theis, to name a few. models, can be not only inaccurate but grossly misleading"
Even this is an overly optimistic assessment of the actual sit- (Klemes 1974), and pointing out the various malignant facets
uation. In reality, while this limited and fragmented hydrolog- of the misguided practices ever since (Klemes 1978, 1982,
ical knowledge did exist, it was often deliberately ignored and 1986, 1987, 1988a, 1991, 1994, 1995). However, if this per-
avoided by the new generation of modelers. Out of enthusi- sistence looks like a paranoic obsession, I may say in my de-
astic naivety or calculated self-interest, they flaunted their hy- fense that, except for the 1974 "Hurst phenomenon" paper,
drological ignorance as "absence of bias" and believed (some all the other papers cited were written based on invitations.
still do) that the mathematical rigor of models and the close- This editorial is an elaboration of the argument sketched in
ness of their fit to empirical data are the supreme guarantors (Klemes 1995). It was written at the suggestion of the editor
of scientific objectivity and the key to true and reliable hydro- of this journal, who has assured me that the message still is
logic understanding. by no means obsolete.
The contrast between the new and the old could not be
sharper. The focus of the old guard was the substance and its Diagnosing the Problem
hydrology was usually right, though sometimes the mathe- There are many causes for the regrettable state in which the
matics was not. The focus of the new generation of modelers mainstream of hydrologic modeling has been meandering for
has been the form: the mathematics is usually correct and it the past 30 years or so. They fall into two broad categories
would be a compliment to say the hydrology is sometimes that can be labeled as "scientific" and "socioeconomic."
JOURNAL OF HYDROLOGIC ENGINEERING / APRIL 1997/43

J. Hydrol. Eng. 1997.2:43-49.


Both have their historic roots, which have been briefly alluded played, believed in? The doyen of hydrology, James Dooge,
to earlier. My intent here is to examine, in more detail, the proclaimed quarter of a century ago: "Models are to be used,
scientific roots of the situation, touching on the other category not to be believed in!" (Dooge 1972). Until the advent of the
only in passing. "modeling revolution" in the 1960s, it was clear that empir-
I see the main causes of the scientific sterility of mainstream ical hydrologic models were meant to be used for applied pur-
hydrologic modeling in the lack of clarity regarding three fac- poses, not for the narcissistic self-gratification of modelers,
tors: (l) the difference between theoretical and empirical mod- which later (increasingly) seemed to be the case and prompted
els; (2) the difference between the notions "better model" and Dooge to make his admonition.
"better fit of a model"; and (3) the meaning of the notion When this is clear, it becomes obvious that a theoretical
"theoretical." Let's examine each of them in more detail. model should ask questions while an empirical model should
summarize answers already available and summarize them
Theoretical and Empirical Models simply and effectively, without spurious ornamentation.
Downloaded from ascelibrary.org by University of Aberdeen, Bedford Road on 03/21/15. Copyright ASCE. For personal use only; all rights reserved.

An empirical model may, of course (and usually does) rep-


The difference between these two kinds of models derives resent the first step in the development of a theoretical model:
from the difference between empirical and theoretical knowl-
it can inspire a question about the prototype that can be fol-
edge that represent two hierarchical, qualitatively different, lowed up, thus starting an iterative learning loop-called the
levels of scientific understanding. To put it simply, empirical scientific method-which is the basic tool for the construction
knowledge is circumscribed by the statement of facts, while of theoretical models. On the other hand, when all relevant
theoretical knowledge includes their interpretation and expla- questions have been answered (a very unlikely situation in
nation. Empirical knowledge tells us what happened, theoret-
practice), the final theoretical model can be regarded as an
ical knowledge tells us why. At each of these two levels, the
exhaustive empirical model since the sum total of answers and
knowledge is often effectively and usefully summarized in explanations it contains (explicitly or implicitly) do represent
mathematical form or mathematical models-and there the
the known facts (indeed, some philosophers hold that all
trouble often lies: the mathematical formalism, same in both knowledge is ultimately empirical, since even the theoretical
cases, may mask the difference in their substance. causal laws are, in the end, only facts repeatedly confirmed by
The point is that an equation or model does not know what experience).
it is being used to describe in a particular case, what physical
It can happen that a theoretical model will end up having
or other entities its symbols are meant to represent, whether the same form as the initial empirical model, the only differ-
its form is merely an ad hoc choice that happens to fit some ence being that the form has been justified by deeper knowl-
observed, i.e., empirical, pattern or whether it has been arrived edge. For example, an empirical unit hydrograph model may
at by consideration of some deeper aspects, which lend it a well come out as an appropriate theoretical model if the basin
more general validity; its algebraic form does not contain in- is confirmed, by other independent investigations, to function
formation about the range of its applicability, about the limits as a linear system.
beyond which its symbols can no longer represent the physical The most essential difference between theoretical and em-
variables whose names they carry. An equation does not know pirical models is the asymmetry in their inferential potential,
whether its variables represent anything real at all and it cannot which is the consequence of their hierarchical relationship: an
be made responsible for predicting nonsense, which it some- adequate form of an empirical model can be deduced from a
times may. These all are land mines scattered across the field valid theoretical model but the reverse is not true-the form
of mathematical modeling, and hydrological models seem to of an adequate theoretical model cannot be deduced from a
be exceptionally prone to being crippled by them. valid empirical model. This asymmetry emerges very clearly
Detection and avoidance of these mines is easier if one steps when we examine the difference between a better model and
back and examines the process by which knowledge is ad- a better fit of a model.
vanced, how it interacts with what is loosely called practice,
and why we may be striving for an explanation in one case Better Model and Better Fit of a Model
and be content with description in another.
At any point in history, certain facts are known as well as Model is generally defined as a representation of some pro-
understood, some are known and understood poorly or not at totype. This representation is never perfect and exhaustive-
all, and there are many facts yet unknown. From the known if it were, the model would not be the prototype's model, but
facts, some are found useful for practical purposes and are its duplicate. It is thus better to say that model is a simplified
exploited to better the standard of living. As a general rule, representation of a prototype, one that preserves some of its
the facts that are well understood tend to be more useful than characteristics but omits others. A "better model" can be de-
those understood less. The Darwinist may argue that the va- fined only relative to the features of the prototype that are
lidity of this rule is the very reason why the human mind has important for a satisfactory function of the model; the more
evolved such a strong urge for understanding and explanation of such features the model represents the better it is.
because this urge, called curiosity, gives humans a distinct A hydrologic model is supposed to represent mathematically
"selective advantage." Selective advantage or not (when one the results of transformations that specified water inputs un-
considers how many Pandora's boxes with potentially destruc- dergo during their passage through a river basin or its parts.
tive contents curiosity has opened up, it sometimes looks more It follows that the model will be the better, the more faithfully
like a selective disadvantage), curiosity is deeply ingrained in it represents the essential features of the dynamics of the pro-
our genes and it is the driving force behind theoretical models. cesses that cause these transformations. This in turn requires
On the other hand, there are many facts, which, though not that the dynamics be known to the modeler, and the only way
well understood, have great practical utility such as they are; he or she can acquire such knowledge is to consult the pro-
this drives the development of empirical models that describe, totype. Thus the general scheme for making a better model is
organize, and mobilize these facts to facilitate action. to apply the scientific method: construct a model based on the
This is the first thing that should be clear to the prospective available observations (data) and the existing theoretical
modeler: empirical models do not, and are not meant to, ex- knowledge (which may well be zero) or an initial hypothesis
plain what is behind data, but merely describe their observed about the dynamics and, if the results do not agree with the
patterns. But is the description itself the purpose? Is it the empirical evidence, go back to the prototype, check the dy-
beauty of the pattern that is to be captured, perfected, dis- namics through experiments or new observations, modify the
44/ JOURNAL OF HYDROLOGIC ENGINEERING / APRIL 1997

J. Hydrol. Eng. 1997.2:43-49.


model on the basis of the new knowledge, and so on. This
process of improvement of the model is schematically illus-
trated in Fig. l(a).
In theory, this method of making a model better applies
equally to theoretical and empirical models because results of
a better theoretical model can, in principle, be summarized in
the form of a better empirical model. However, to paraphrase
George Orwell, in practice it applies "more equally" to the-
oretical models than to empirical ones. This is so because the
overriding purpose of an empirical model is to summarize (the
results of) what is known and give it to a client who may
already be waiting impatiently to use it in his project. So there
Downloaded from ascelibrary.org by University of Aberdeen, Bedford Road on 03/21/15. Copyright ASCE. For personal use only; all rights reserved.

is a strong disincentive to go back to the prototype (collect


more data, do experiments and perhaps theoretical research)
even if the result is not fully satisfactory and as long as the
client "can live with it" (after all, it could be an open-ended
exercise-it could take years, there is no guarantee of success YES

ND~
and the need for the project is here and now!). And so, as
Rogers (1983) once matter-of-factly summarized the typical
situation, the scientist or engineer yields to the various admin-
istrative and legal requirements and tries "to do the best he
can under the situation."
In practice, "doing the best" is improving the fit of the
model to the available data. It does not require going back to YES
the prototype and gather new information-all it needs is to
have one run the model in a "do loop" a few times and adjust ~
a parameter here and there to reduce some formal error func-
tion to improve the goodness of fit. This process is schemat-
ically illustrated in Fig. I(b).
BETTER
On comparing Figs. l(a) and l(b), we see that the first pro-
cess tries to make the model better with respect to the proto-
MODEL
type while the second does so only with respect to itself: the FIG. 2. General Model-Building Procedure
first tries to improve the representation of the substance, the
second of the form. Since this form is determined by the pat- substantially-there is nothing more to be learned about the
tern of the original data, which, in the second case, is the only prototype, no matter how many times the loop is repeated. It
information about the substance that ever enters the picture, is the inviolable axiom of the modeling process that the pro-
there is no way the second process can improve the model totype is the horse and the model is the cart it is pulling; thus
the basic message of this essay is that one cannot get ahead
(a) (b) by putting the cart before the horse.
Unfortunately for the advancement of hydrology, this is ex-
actly what the mainstream of hydrological modeling has been
doing for the last 30 years-pretending that a meticulous pol-
ishing of the fits of the various empirical models will trans-
form them into sound theoretical models! No wonder the result
has been stagnation, since, as my friend and colleague (one of
the best hydrologists I have met), Fred Morton, used to say:
"They are spinning the wheels of a car stuck in the mud."
It has to be emphasized that some amount of polishing is
legitimate and usually necessary in every kind of modeling as
it is, for instance, in every kind of writing: it is a rare occasion
that nothing can be improved on a first draft. Thus there is
usually secondary fitting, or calibration, loop within the pri-
mary modeling loop shown in Fig. l(a). Accordingly, under
the magnifying glass, Fig. I(a) would usually appear as shown
in Fig. 2, which illustrates the general process of model build-
ing.
YES YES
~ ~ Theoretical-What Does It Mean?
There is a widespread misconception about the meaning of
BETTER BETTER the notion •'theoretical," which reinforces the misconceptions
about modeling discussed earlier. It derives from the fascina-
tion with mathematics. This is a very common phenomenon
FIT OF A among nonmathematicians, who often view mathematics as
the purest and highest theory whose pronouncements have the
final authority about everything in this world and are unchal-
MODEL MODEL lengeable by definition because, if the mathematics is rigorous,
FIG. 1. Difference between Procedures Aimed at: (a) Improve- they are exact. Thus a curve drawn through a few points, if it
ment of a Model; (b) Improvement of the Fit of a Model can be expressed by a mathematical equation, automatically
JOURNAL OF HYDROLOGIC ENGINEERING / APRIL 1997/45

J. Hydrol. Eng. 1997.2:43-49.


becomes a theoretical model and commands a degree of re- to such heights? I would not dare to ask them for fear that it
spect that would make it blush with embarrassment if it were may be regarded as an insult to their intelligence.
aware of it. The analogy is straightforward up to this last question. With
This then often leads to the practice-deplored, for in- no hesitation we can substitute hydrology for anthropology,
stance, by the illustrious statistician Box (l976)-that a le- water resources engineering for the clothing industry, and, say,
gitimate physical problem is redefined as a mathematical prob- a log-Pearson distribution model fitted to maximum annual
lem for which a theory exists, and then a solution of this flows for the normal model fitted to human heights. It holds
mathematical problem is automatically regarded as a valid the- even in the operational context since the log-Pearson distri-
oretical solution of the physical problem. This approach is bution model of annual flow maxima can be equally useful to
most dangerous when extrapolation of such redefined mathe- the engineer for the sizing of culverts and spillways as the
matical formulation is believed to be a valid theoretical ex- normal models for the distributions of sizes of various parts
trapolation of the behavior of the original physical variable. of the human body are to manufacturers of shoes and under-
Downloaded from ascelibrary.org by University of Aberdeen, Bedford Road on 03/21/15. Copyright ASCE. For personal use only; all rights reserved.

wear. In theory there is no reason for not extending the anal-


Nonhydrological Analogy ogy even into the scientific domain since we can ask the le-
gitimate question: "Would any hydrologist, not to mention a
My purpose here is, of course, to demonstrate the fallacy water resource engineer, seriously entertain an idea that there
and danger of the foregoing misconceptions in the context of can occur a Mississippi River flood on some three-mile-long
hydrologic modeling, but, before referring to hydrology, I will Rock Creek because the scientifically fitted log-Pearson' 'the-
first introduce a nonhydrological analogy to better sensitize the oretical distribution model" of the Rock Creek floods assigns
reader to the general essence of the problem. to it a definite nonzero probability?"
The value of analogies is in facilitating an outsider's view
of a problem known to us only from the inside, by transposing
it into a context outside our personal emotional attachments, Back to Hydrology
which often prevent us from seeing the obvious. It is best Unfortunately, here the analogy breaks down but, simulta-
suited for giving an overall view rather than a detailed one, a neously, becomes most revealing because, in hydrology, the
hint rather than a proof. To serve this purpose, an analogy answer to the last question is "Yes, very often they would."
must not be carried too far. I hope the following example, Those who do say "yes" to similar questions typically think
using an anthropological analogy, meets this criterion. of themselves as "stochastic hydrologists." Anthropology will
Anthropology studies man as an animal, including his in- be safe from the hydrological disease until some underwear
dividual and collective behavior and his physical features such manufacturer, on becoming professor, considers himself or
as longevity, body size, etc. Apart from their intrinsic interest herself a stochastic anthropologist.
in the study of the human animal, some anthropological data To be fair, in hydrology the absurdity of extrapolation of
serve to obtain "design parameters" in various industries; for empirically chosen mathematical formulas beyond the limits
example, longevity in the life-insurance business and body size of observation is sometimes not as transparent as in the pre-
in the clothing industry. ceding anthropological analogies. But this is the very reason
These users process anthropological data of interest by var- why hydrologists and professionals applying hydrology should
ious statistical techniques and fit them with various mathe- be doubly alert to the danger it represents and why they should
matical models to derive the design parameters needed, but be clear about its root cause-the failure to appreciate the
there is a clear understanding that their models have no in- difference between a descriptive model of empirical facts and
trinsic anthropological value. If, for example, an anthropolo- a theoretical model, which embodies the dynamics of the phys-
gist wanted to know more about the limits of human life span, ical process that has produced them. Needless to say, here the
he would certainly not turn to the formula used for compiling term theoretical implies a physical, i.e., hydrological, theory
mortality tables and refine the estimation of its parameters, of the process or system. To avoid ambiguity, I have often
examine its asymptotic behavior, use it for generating large used the term "physically based" instead of "theoretical."
synthetic samples, fit these with other formulas, etc. Rather, Failure to see the foregoing difference has a long tradition
he may seek advice from experts in physiology, geriatrics, nu- in hydrology and I certainly cannot claim to be the first to
trition, etc. have noticed it. Well, before I was born, Horton, for example,
As Feller (1957) observed, "We hesitate to admit that man felt it necessary to stress this matter and brought it to the
can grow 1000 years old, and yet current actuarial practice attention of the hydrologic community in the following words
admits no bounds to the possible duration of life. According (Horton 1931):
to formulas on which modern mortality tables are based the
proportion of men surviving 1000 years is of the order of The use of empirical generalizations under proper con-
magnitude of one in IO I036 [read 10 to power 1036]_a num-
A

ditions is neither avoidable nor undesirable. It is, however,


ber with 1027 billions of zeros. This statement does not make important to recognize the nature of the physical processes
sense from a biological or sociological point of view." involved and their limitations in connection with the use of
Similarly, the distribution of human heights (and various statistical methods and empirical formulas. It may happen
body sizes) can usually be well fitted with the normal distri- that several different types of formulas will give excellent
bution model, and its use in the clothing industry for the sizing fit to the data within the limits of observation but it is im-
of products is very useful despite the absurdity of the distri- portant to use the formula which not only fits the data but
bution's infinite tails. Again, this model has no intrinsic an- which gives rational results for the limiting or boundary
thropologic value even if fitted to the empirical data using the conditions. Hydrologic literature contains many empirical
most refined and efficient methods. Within the range of the generalizations which if extended would lead to absur-
actual observations it describes the pattern of their distribution dities ... Rock Creek cannot produce a Mississippi River
quite well; but what can anthropology learn from it, for ex- flood-any more than a barnyard fowl can lay an ostrich
ample, about the limits of human heights? Would any anthro- egg....
pologist, not to mention an underwear manufacturer, seriously
entertain an idea that a human being can be 10 m (or -10 It is ironic that in the same era when Horton was elevated
cm, for that matter) tall because the scientifically fitted theo- to the position of "Patron Saint" of scientific hydrology, with
retical distribution model assigns a definite nonzero probability awards and medals bearing his name, hydrology embraced the
46/ JOURNAL OF HYDROLOGIC ENGINEERING / APRIL 1997

J. Hydrol. Eng. 1997.2:43-49.


modeling paradigm, which turned its back on the Hortonian (1) surface flow with lag time less than one month; (2) shallow
spirit and his specific warning that "[while] the best possible subsurface flow with lag time about two months; and (3) deep
use should be made of all the available data .,. [this] rule ground-water flow with lag time of about seven months. Of
must often be tempered according to the necessities of econ- course, only the first component was physically realistic. The
omy of labor and with due regard to the avoidance of metic- second had nothing to do with an shallow subsurface flow
ulous accuracy" (Horton 1931). Hydrological modelers often (which was negligible and all safely in the lake within a few
did the very opposite: the "necessities of economy of labor" days at most), but was in fact a part of the recession limb of
being drastically reduced by the computer, they not only the surface outflow hydrograph, the part of the precipitation
stopped paying "due regard to the avoidance of meticulous delayed by the lake routing effect. However, by far the most
accuracy," but also made the cultivation of it their main pre- interesting was the third component, which beautifully falsi-
occupation. And, as if to flaunt their disdain for the Hortonian fied the hydrologist's basic postulate and compromised the al-
spirit, hydrologic modelers have pursued this "meticulous ac- leged physical meaning of his model. There was absolutely no
Downloaded from ascelibrary.org by University of Aberdeen, Bedford Road on 03/21/15. Copyright ASCE. For personal use only; all rights reserved.

curacy" most vigorously in the fitting of exactly those "sta- "seven-month delayed outflow," whether surface or ground
tistical and empirical formulas ... which if extended would water, all the carryover being completed within one or two
lead to absurdities." months, depending on whether precipitation occurred at the
I never cease to be amazed by how otherwise reasonable beginning or end of a calendar month. However, there were
people fall into the trap of believing that, by embellishing mild autumn rains, which usually occurred about seven months
purely empirical models with refined mathematical formal- after the main runoff event, the spring snowmelt. These rains
isms, they somehow can breeze a "hydrological soul" into caused a small bump on the outflow hydrograph, which the
them and transform them into "theoretical hydrological mod- model wrongly identified as runoff delayed by deep ground-
els" that make sense even when extended far beyond the range water seepage from the spring. I then proposed to my friend
of observation. In the past, I even used to test empirically the an exact test based on synthetic data simulated by an exact
validity of such beliefs when opportunities offered themselves, model of a hypothetical system of the prescribed form but my
which only reinforced my conviction that such beliefs are mis- offer was declined.
guided and futile. Two examples will illustrate my experience. Both of the foregoing models, by the way, are good empir-
Almost 20 years ago, a noted stochastic-hydrologist friend ical models and can be useful in practical applications where
asked me for a set of flood peak discharge data to test his new nothing more than a concise representation of known facts is
regional flood-frequency model; he explicitly requested that I required. However, it is the modeler's responsibility to know
give him no other information except the numbers. For a hy- that facts and not nonsense are being represented. I perceive
drologic modeler to tell me he wants no hydrological infor- avoidance or repudiation of this responsibility as misguided,
mation on the hydrologic prototype he is modeling has always whether it is in theoretical or empirical modeling.
been one of the best ways to stir my blood-and so I sent to Such models are meaningless if we want to gain hydrolog-
my friend a set of fabricated numbers. As expected, his model ical insight into what is not yet known-if we are interested,
duly produced from them a "theoretical(!) regional(!!) say, in the shape of the upper tail of the flood distribution of
flood(!!!) distribution." Since I still have the computer Horton's Rock Creek per se, in a way similar to an astron-
printout with the results in my files, I can reveal in confidence omer's interest in the distribution of frequencies in the spec-
that, for example, the 1O,OOO-year flood for the region is trum of the Aldebaran. What is there to be learned about this
2.5181. While the units are not known (as requested, I sup- tail if, based on some log-Pearson III best fit to a few observed
plied numbers only), the magnitude of the flood must be me- floods barely representing the body of the distribution, we pre-
ticulously accurate because the parameters of the theoretical sume the answer to be known? What are the hydrological rea-
distribution model are given to eight decimal places. To the sons for the belief that this model will faithfully describe the
credit of the investigator, I must say that he detected some shape of the tail all the way into the nebulous heights of
peculiarities in my data, namely that some of the computed 1O,OOO-year and million-year floods-as one distinguished
model parameters pointed to a "small homogenous region," American hydrology professor once insisted to me, dismissing
while others to a "large heterogenous region." I assured him my scepticism as a "complete disregard of the very founda-
he was right on both counts: the small region was my desk, tions of the theory of mathematical statistics?"
the large region was my imagination. What an irony! When Karl Pearson developed his system
About 15 years ago, another hydrologist friend tried to con- of frequency distributions in response to the appeal of his bi-
vert me to his belief (which he called basic postulate) that if ologist colleague (Weldon) for mathematical tools to help him
a model provides a really accurate approximation then its analyze his morphological measurements of shrimps and shore
structure and parameters must correspond to the physical prop- crabs, Pearson always consulted Weldon about the probable
erties of the prototype. "Hence," he wrote, "the structure and biological limits to which his curves could be reasonable ex-
parameters of well fitted equations can clarify the hidden phys- trapolated. Now, more than a century later, hydrologists are
ical structure of the natural process.... [to achieve this, it is] consulting the extrapolated Pearson's curves when they seek
only necessary to have an exact fit [based on] long data series answers regarding the probable hydrological limits of floods!
to meet strict statistical requirements.... " To prove his point, If the deterministic modelers should be tempted to applaud
he offered to infer the "hidden physical structure" of some, me for rightly debunking their stochastic counterparts, they
unknown to him, basin just by fitting his model to a reasonably should pause. One does not have to look any further than, for
long record of monthly precipitation and runoff. I supplied him example, the unit hydrograph, to see that they have little rea-
with such data from a small research basin in the Canadian son for complacency. Let's just recall how this useful empir-
shield, with the total area of 0.35 km 2 , very little soil, and with ical concept has been redefined as "unit response of a linear
its steep granite slopes circling a good-size lake whose outflow system," which was then almost beaten to death by all kinds
represented the total basin runoff. I made this choice in order of rigorous theory (of linear systems, of course, not of hydro-
to have a basin whose physical structure was clear-in this logical systems), including Fourier and Laplace transforms,
case, and for the requested monthly data, it was an almost Laguerre analysis, time-series analysis, matrix methods, mo-
perfect example of a single, slightly nonlinear, reservoir. ment matching, cumulants. etc. All this prestidigitation has
Based on a very good fit of his model, the hydrologist iden- contributed next to nothing to the understanding of hydrolog-
tified the following three physical components in the runoff: ical systems, which, as some rudimentary hydrological inves-
JOURNAL OF HYDROLOGIC ENGINEERING / APRIL 1997/47

J. Hydrol. Eng. 1997.2:43-49.


tigations will readily reveal, generally behave as nonlinear talents to it!" He said "Look, I suppose it is true that society
systems. Ironically, it took a theoretical probabilist to tell hy- may not be getting its money's worth out of my work. But
drologists as much (Moran 1975): " ... entirely different phys- these problems are very interesting to me, I simply love to do
ical models may lead to the same stochastic behavior ... This it. And, as long as society is willing to pay me for what I do
is the situation in studying rainfall runoff [where the response] I will keep doing it." I appreciated his sincerity and had to
may be highly nonlinear it is clear that nonlinearity is an agree that, though useless for the advancement of statistical
all pervading problem Thus many questions remain un- hydrology, his work had an undeniable practical value.
answered.' , Many Horses, other than hydrology, are pulling the cart of
A more recent example is provided by the current obsession hydrologic modeling. One of them was born out of the intrin-
to assess the hydrological impact of climate change using sce- sic difficulty of accomodating hydrological engineering (by
narios of theoretical possibilities. These are invariably con- which I mean the application of hydrological knowledge in
structed as combinations of all hypothesized changes in the solving engineering problems) in the academic environment.
Downloaded from ascelibrary.org by University of Aberdeen, Bedford Road on 03/21/15. Copyright ASCE. For personal use only; all rights reserved.

forcing variables, such as precipitation and temperature, which For example, a professor of surgery can (must!) do surgery on
are then fed into various deterministic conceptual models to real live people; a professor of electrical engineering can build,
generate the changed runoff. Each of these combinations is say, a real adaptive antenna in his laboratory; a mechanical
presented as theoretically possible, although some of them can engineering professor can build, say, an actual centrifugal liq-
be completely impossible because of their physical incompat- uid separator or a better mouse trap; a professor of hydraulic
ibility within the accepted theory of atmospheric physics. engineering can build, test, and meaningfully improve at least
Their claimed theoretical possibility of course relates to the a scale model of a spillway for a specific real dam; and so on.
possible combinations of independent random variables into But, if it is often difficult even for a hydraulic engineering
which the forcing variables have been redefined without the professor to satisfy the (no doubt very reasonable) Simons
slightest effort to find out whether this assumption is jus- (1992) postulate that "People posing as hydraulic engineers
tified-and often even in spite of clear evidence to the con- should, in fact, be hydraulic engineers!", it is even more dif-
trary. Still more disturbing and dangerous is that results of ficult for a hydrological engineering professor. There is vir-
such modeling are routinely presented as, say, "sensitivity of tually nothing else for him to do than polish methods for the
runoff to climate change," rather than as sensitivity of models computation of hydrology-related design parameters in gen-
to arbitrary changes in their forcing variables, which is what eral, since a professor normally is not responsible, in any real
they really are. engineering sense, for design parameters for an actual levee
The preceding examples give an idea about the substance on the Mississippi River or even for a culvert to pass a rock
of the modeling problem in hydrology but not about its per- creek under a local dirt road. And so, if he cannot, or is not
vasiveness and magnitude. Nobody has captured the latter motivated to, switch from the science of engineering compu-
facet better than Harvard's late Fiering (1976): tations to the science of hydrology and become a true hydrol-
ogist who is able and willing to harness the hydrology horse
Fascination with automatic computation has encouraged to pull his modeling cart, he is more or less doomed to "per-
a new set of mathematical formalisms simply because they form the elaborate pirouettes on the high wire of techniques"
now can be computed; we have not often enough asked I mentioned earlier. It is, then, in his best interest to muddle
ourselves whether they ought to be computed or whether the water and cultivate fuzziness in the understanding of hy-
they make any difference ... we build models to serve drologic modeling.
models to serve models to serve models, and with all the Collusion of other factors has helped push the cart of hy-
computation, accumulated truncation, roundoff, sloppy drologic modeling off the road: even a very sophisticated and
thinking, and sources of intellectual slippage, there is some impressive-looking hydrologic model is cheaper than main-
question as to how reliable are the final results. taining a modest observation program in the field. In the era
of budget cuts, this has made modeling an attractive alternative
It was during this era that the term "hydrologic modeling" even to agencies that used to do solid hydrologic work in the
acquired a new meaning and a rather unsavory aftertaste. past. It is easier and more fun to play with a computer than
to face the rigors of fieldwork, especially hydrologic field
Are Hydrologic Modelers Dumb? work, which usually is most intensive during the most adverse
conditions. It is faster to get a result by modeling then through
This question invariably comes to mind when one encoun- acquisition and analysis of more data, which suits managers
ters some of the primitive misconceptions and practices de- and politicians as well as staff scientists and professors to
scribed earlier. Of course they are not dumb! Just the oppo- whom it means more publications per unit time and thus an
site-they are too bright and too clever to be of any benefit easier passage of the hurdles of annual evaluations and other
to the science of hydrology. Some of the misconceptions de- paper-counting rituals. And it is more glamorous to polish
scribed have been discretely kept behind a smoke screen of a mathematical equations (even bad ones) in the office than
carefully cultivated fuzziness, which has proved to be enor- muddied boots (even good ones) in the field.
mously helpful in keeping many modelers in business. Be- And, to repeat, hydrologic modelers are no fools.
cause, as Fiering (1976) already observed twenty years ago,
"It seems clear that some models are pressed because profes- Conclusions
sors are promoted and consultants are consulted in proportion
to their generation of sophisticated, mathematically-oriented The philosopher, George Santayana, has said: "Those who
models. This is too bad." do not remember the past are condemned to repeat it." This
Some 10 years ago, during a coffee break at a conference editorial may help those hydrologists and engineers who are
on flood probability modeling and related issues, my conver- too young to remember themselves, or too busy to have no-
sation with one brilliant professor was along the following ticed, see what has led to the rather unenviable state of hy-
lines: "Tell me, what is driving you to do all these things? drologic modeling, and perhaps even provide some hints on
You are not stupid. You know that all this theorizing and math- how to prevent the fulfillment of Santayana's prophecy.
ematical legerdemain is sterile and has next to nothing to do It appears that the danger of its fulfillment in hydrology is
with the real flood probabilities, especially for the very high less acute today than it was, say, a decade ago because, faced
floods. You could do so much useful work if you put your with the possibility of large-scale detrimental (and potentially
48/ JOURNAL OF HYDROLOGIC ENGINEERING / APRIL 1997

J. Hydrol. Eng. 1997.2:43-49.


irreversible) anthropogenic effects on hydrological and related Klemd, V. (1974b) "The Hurst phenomenon: a puzzle?" Water Resour.
systems, more people seem to be appreciating the late Hall's Res., 10(4), 675-688.
(1971) warning that: "While we may be able to take stop-gap Klemes, V. (1978). "Physically based stochastic hydrologic analysis."
Advances in hydroscience. Vol. 11, Academic Press, Inc., New York,
measures, ... we cannot serve society honestly until hydrology N.Y., 285-355.
is researched, understood and taught as a single system." Klemes, V. (1982). "Empirical and causal models in hydrology." Sci-
During my recent visits to some American and Australian entific basis of water resource management, Nat. Acad. of Sci., Wash-
universities and research facilities, I saw many more examples ington, D.C., 95-104.
of solid hydrological research and honest investigative mod- Klemes, V. (1986). "Dilettantism in hydrology-transition or destiny?"
eling than I was used to seeing on similar occasions, say, 15 Water Resour. Res., 22(9), 177S-188S.
Klemes, V. (1987). "Hydrological and engineering relevance of flood
years ago. Thus I can end this editorial on a note of cautious frequency analysis." Hydrologic frequency modeling, V. P. Singh, ed.,
optimism-after all, computer games have become old hat Reidel, Dordrecht, The Netherlands, 1-18.
and the generation that introduced them into hydrology has Klemes, V. (1988a). "A hydrological perspective." J. Hydro., Amster-
Downloaded from ascelibrary.org by University of Aberdeen, Bedford Road on 03/21/15. Copyright ASCE. For personal use only; all rights reserved.

now reached the retirement age. dam, The Netherlands, 100, 3- 28.
Klemes, V. (1988b). "Hydrology and water resources management: the
ACKNOWL;,EDGMENT burden of common roots." Proc., Vlth IWRA Congr. on Water Resour..
Vol. I, IWRA, 368-376.
I am obliged to Stephen J. Burges of the University of Washington for
Klemes, V. (1991). "The science of hydrology: where have we been?
bringing to my attention the 1931 paper of Robert Horton.
Where should we be going? What do hydrologists need to know?"
Proc.• Int. Symp. to Commemorate 25 Years of IHDIIHP, UNESCO,
APPENDIX. REFERENCES Paris, France, 41- 50.
Box, G. E. P. (1976). "Science and statistics." J. Am. Statistical Assn., Klemes, V. (1994). "Statistics and probability: wrong remedies for a con-
71(356),791-799. fused hydrologic modeller." Statistics for the environment 2: water
Chow, V. T. (ed.) (1964). Handbook of applied hydrology. McGraw-Hili related issues, V. Barnett and K. F. Turkffian, eds., John Wiley & Sons,
Book Co., Inc., New York, N.Y. Inc., London, England, 345-366.
Dooge, J. C. I. (1972). "Mathematical models of hydrological systems." Klemes, V. (1995). "Of carts and horses in probabilistic hydrologic mod-
Proc., Symp. on Modelling Techniques in Water Resour. Sys.. Vol. I, elling." Proc., Conf on Statistical and Bayesian Methods in Hydro-
Environment Canada, Ottawa, Canada, 171-189. logical Sci., UNESCO, Paris, France.
Feller, W. (1957). An introduction to probability theory and its applica- Linsley, R. K., Kohler, M. A., and Paulhus, J. L. H. (1949). Applied
tions. Vol. 1. John Wiley & Sons, Inc., New York, N.Y. hydrology. McGraw-Hili Book Co., Inc., New York, N.Y.
Fiering, M. B. (1976). "Synthetic hydrology: an assessment." Water re- Linsley, R. K., Kohler, M. A., and Paulhus, J. L. H. (1958). Hydrology
search, A. V. Kneese and S. C. Smith, eds., John Hopkins, Baltimore, for engineers. McGraw-Hili Book Co., Inc., New York, N.Y.
Md., 331-341. Moran, P. A. P. (1975). "The future of stochastic modeIling.' , Proc.• Int.
Hall, W. A. (1971). "Biological hydrological systems." Proc.• 3rd Int. Congr. of Mathematicians, 517-521.
Seminar for Hydro. Prof, Agric. Experiment Station and Nat. Sci. Rogers, P. P. (1983). "Book review." Eos, 64, 419.
Found. Simons, D. B. (1992). "Future trends and needs in hydraulics." J. Hydr.
Horton, R. E. (1931). "The field, scope and status of the science of Engrg., ASCE, 118(HYI2), 1607-1620.
hydrology." Trans. Am. Geophys. Union; Prac., 12th Annu. Meeting,
Nat. Res. Council, Washington, D.C., 189-202.
Klemes, V. (1974a). "Some problems in pure and applied stochastic hy- V. Klemes
drology." Proc.• Symp. on Statistical Hydrol., U.S. Dept. of Agr., Chief Hydrologist (Retired), Environment Canada
Washington, D.C., Miscellaneous PubI. No. 1275, 2-15. 3460 Fulton Road, Victoria, BC. V9C 3N2, Canada

JOURNAL OF HYDROLOGIC ENGINEERING / APRIL 1997/49

J. Hydrol. Eng. 1997.2:43-49.

You might also like