20 Future Perspectives
447
In a recent series of papers, Grayson et al (1992) question the value of distributed parameter models to truly represent the processes if the fundamental algorithms and assumptions cannot be validated. In their conclusions they state "the
misperception that model complexity is positively correlated with confidence in
the results is exacerbated by the lack of full and frank discussion of a models capability and limitations and the reticence to publish poor results". They go on to
conclude that "the seductive attraction of the more complex models is their ability
to provide information about points within the catchment, but it is concluded that
the representations used in current process based models are often too crude to
enable accurate, a priori application to predictive problems".
In an even more recent paper, Jakeman and Hornberger (1993) question "what
limits the observed data place on the allowable complexity of rainfall-runoff models". They further state "conceptual and physically based models developed and
used for describing rainfall-runoff processes tend to be over parameterized. They
are no more useful for prediction than are simpler models whose parameters are
identifiable from available data" (Jakemanand Hornberger, 1993).
From these examples, one can conclude that the development of more complex
and sophisticated models and innovative analysis techniques has not resulted in an
overwhelmingly improved ability to predict runoff. Often the question is asked,
why is it that more complex and more physically based models do not give us
better runoff predictions? To the authors this question seems to be ill posed for the
following reasons:
It is by no means logical, that a more complex and more physically based
hydrological model will yield more accurate results than simpler structured
models. The complex models usually contain a higher number of parameters
and thus may cause more potential modeling errors and often they do not have
an adequate data base which is needed for achieving accurate results,
The reason for developing models of a higher complexity which are distributed in area, is usually not the desire for more accurate results, but the wish to
apply the model also under changing conditions, e.g. future landuse changes,
which is usually not possible with the more simple structured models,
Often the more complex structured hydrological models working on the basis
of high resolution in space and time are fed with input data (e.g. precipitation,
snow, radiation) with a very coarse resolution in space, and often also in time.
The model complexity is not capable of making up for insufficient resolution
in space and time of the input data.
The idea to develop a general hydrological model for the whole water cycle,
or parts of it, which functions optimal for all conditions is still utopian. At the
present stage of scientific hydrology we still have to tailor the model for the
purpose. In each case we have to decide, which model would be most suitable
447
In a recent series of papers, Grayson et al (1992) question the value of distributed parameter models to truly represent the processes if the fundamental algorithms and assumptions cannot be validated. In their conclusions they state "the
misperception that model complexity is positively correlated with confidence in
the results is exacerbated by the lack of full and frank discussion of a models capability and limitations and the reticence to publish poor results". They go on to
conclude that "the seductive attraction of the more complex models is their ability
to provide information about points within the catchment, but it is concluded that
the representations used in current process based models are often too crude to
enable accurate, a priori application to predictive problems".
In an even more recent paper, Jakeman and Hornberger (1993) question "what
limits the observed data place on the allowable complexity of rainfall-runoff models". They further state "conceptual and physically based models developed and
used for describing rainfall-runoff processes tend to be over parameterized. They
are no more useful for prediction than are simpler models whose parameters are
identifiable from available data" (Jakemanand Hornberger, 1993).
From these examples, one can conclude that the development of more complex
and sophisticated models and innovative analysis techniques has not resulted in an
overwhelmingly improved ability to predict runoff. Often the question is asked,
why is it that more complex and more physically based models do not give us
better runoff predictions? To the authors this question seems to be ill posed for the
following reasons:
It is by no means logical, that a more complex and more physically based
hydrological model will yield more accurate results than simpler structured
models. The complex models usually contain a higher number of parameters
and thus may cause more potential modeling errors and often they do not have
an adequate data base which is needed for achieving accurate results,
The reason for developing models of a higher complexity which are distributed in area, is usually not the desire for more accurate results, but the wish to
apply the model also under changing conditions, e.g. future landuse changes,
which is usually not possible with the more simple structured models,
Often the more complex structured hydrological models working on the basis
of high resolution in space and time are fed with input data (e.g. precipitation,
snow, radiation) with a very coarse resolution in space, and often also in time.
The model complexity is not capable of making up for insufficient resolution
in space and time of the input data.
The idea to develop a general hydrological model for the whole water cycle,
or parts of it, which functions optimal for all conditions is still utopian. At the
present stage of scientific hydrology we still have to tailor the model for the
purpose. In each case we have to decide, which model would be most suitable