Geoscience Reference
In-Depth Information
by Beven (1989), Grayson
et al.
(1992b) and Beven (1996a). The latter paper summarises some of the
problems as follows (Beven, 1996a, p.273):
There is a continuing need for distributed predictions in hydrology but a primary theme of
the analysis presented here is that distributed modelling should be approached with some
circumspection. It has been shown that the process descriptions used in current models
may not be appropriate; that the use of effective grid scale parameter values may not always
be acceptable; that the appropriate effective parameter values may vary with grid scale;
that techniques for parameter estimation are often at inappropriate scales; and that there is
sufficient uncertainty in model structure and spatial discretisation in practical applications
that these models are very difficult (if not impossible) to validate.
It is directly followed by a response from Danish Hydraulics Institute SHE modellers (Refsgaard
et al.
,
1996, p.286). Their response concludes:
In our view the main justification for the distributed physically based codes are the demands
for prediction of effects of such human intervention as land use change, groundwater ab-
stractions, wetland management, irrigation and drainage and climate change as well as
for subsequent simulations of water quality and soil erosion. For these important purposes
we see no alternative to further enhancements of the distributed physically based mod-
elling codes, and we believe that the necessary codes in this respect will be much more
comprehensive and complex than the presently existing ones.
Similarly, Brian Ebel and Keith Loague have suggested that taking the distributed continuum approach to
the prediction of catchment processes is one way to avoid the “fog of equifinality”. While recognising the
ill-posedness of the catchment modelling problem, they suggest that ensuring that the model predictions
are consistent with both the physics of the processes and the internal state measurements in the catchment
is the only way to progress scientific hydrology and be sure that a model is predicting the right responses
for the right reasons (Ebel and Loague, 2006). How far this is possible in practical applications rather than
research sites remains to be seen. The justifications that underlie the development of physically based
distributed models are not in dispute. The need for prediction of the effects of distributed changes in a
catchment continue to increase (see Chapter 8). However, it is somewhat difficult to see how such advances
will be made unless new measurement techniques for effective parameter values or grid scale fluxes are
developed. There are theoretical problems about the physics of the process descriptions (Beven, 1989,
2001, 2006b) and practical numerical problems that will need to be overcome in the future development
of this type of model but the problem of parameter identification, particularly for the subsurface, will be
even greater.
Refsgaard
et al.
(2010) in their review of 30 years of SHE modelling come to somewhat similar
conclusions. They also point to the difficulty of defining process descriptions that reflect the complex-
ities and heterogeneities of the real system; the problem of defining effective parameter values within
a particular model structure; the potential for different model structures to produce similar predictions;
the importance of uncertainty estimation; and the need for new measurement techniques for model set-
up and evaluation. These issues will recur in the discussion of semi-distributed models in Chapter 6;
in considering the next generation of hydrological models in Chapter 9; and in the “models of every-
where” concepts of Chapter 12. They are fundamental to the future of hydrological science. The question
is whether the continuum differential equation approach that was outlined by the Freeze and Harlan
blueprint in 1968 is the best or only way of doing science in hydrology. As we will see, there may be
alternatives ....
