Geoscience Reference
In-Depth Information
data from digital terrain and other geographical information systems; rapid advances
continue to be reported in the literature.
Among the main advantages of distributed models one can note that they allow the
exploration of the consequences of various simplifying assumptions; as a result, they
can lead to a better understanding of the various pathways and of the interplay between
the main processes and related aspects of complex hydrologic systems in the real world.
They can also be useful in the prediction of outflow from headwater catchments, pro-
vided their parameters can be determined. But this requirement subsumes also one of
theirmain shortcomings. Ideally, the parameters should be determinedapriori, that is
independently from the model's performance. In many cases, however, this is impossible
and the parameters must be estimated by calibration. But then, distributed models tend to
contain so many parameters that it becomes practically impossible to estimate them all in
objective and physically consistent ways. Another major drawback is that the underlying
mathematical rigor of the parameterizations of the model components may instill in the
practitioner a confidence and a sense of realism about their performance, that they do not
deserve, on account of the many simplifications and uncertainties involved. As a result,
the limitations of such models may not be fully understood by uninitiated users and they
may be applied to situations for which they were not intended.
In contrast, the lumped models, whose computational scales are of the same order of
magnitude as the catchment scales, rely on fewer parameters, which are generally easier
to estimate from the available data. Therefore, they are easier to apply in basin outlet flow
simulations for prediction and forecasting purposes. Unfortunately, as the computational
scale increases, it becomes increasingly difficult to give a physical interpretation to these
parameters, in the sense of the processes described in Chapters 2-10. This means that it
is usually impossible to predict changes in these parameters, as the catchment undergoes
physical changes, such as those resulting from an evolving land use or changing climate.
Another drawback is that even when the catchment characteristics remain unchanged,
catchment-scale parameters are incapable of accommodating spatial variability of the
input (e.g. rainfall) and of the flow processes (e.g.infiltration and evaporation). Moreover,
it is impossible to use this approach to describe the detailed flow paths required in the
prediction of pollutant transport or erosion. In spite of all these shortcomings, the lumped
approach continues to be useful in the prediction of streamflow for certain operational
and design purposes. Specific implementations of this approach are further treated in
detail in Chapter 12.
Again, in closing thisreview, it should be understood that, although a classification
into distributed and lumped models is useful to bring some order in the multitude of
possible approaches, it is also somewhat artificial. Comparison of the different methods
treated in Chapters 5-10 has made it clear that the lumped kinematic approach is merely
the simplest extreme in a continuous range of complexity levels, which can be applied in
up-scaling the analysis from the finest resolution of the full space- and time-dependent
conservation equations of momentum, energy and mass to the coarsest resolution, that is
the scale of the catchment itself. However, the level of model complexity necessary for a
specific application isstill not well known; nor is it clear what scenarios warrant the use
of more complex models or under what conditionsadistributed model will consistently
