Environmental Engineering Reference
In-Depth Information
is inadequate, or it might be because of structural
inadequacies in the model (using the wrong
hypotheses). The important point is that recogni-
tion of model failure implies that the representa-
tion of a catchment will be improved. If the model
continues to produce predictions within accept-
able (or unavoidable) limits of uncertainty then no
improvements will be needed, even if it can never
be guaranteed that amodel is producing acceptable
results for the right reasons (Kirchner 2006),
suggesting that continued re-evaluation remains
necessary.
Once such 'models of everywhere' are available,
the learning process will result in place becoming
as important as model structure. Some model
structures might prove to be more appropriate in
some places; elsewhere, other structures might
be appropriate. Everywhere, the idiosyncrasies of
particular places will mean that the effective
parameters required to represent the processes in
those places might vary significantly. The most
important issue for the future of modelling in this
context will then be how to inform the learning
process.What types of datawill it bemost valuable
to collect ormake use of in deciding on appropriate
model structures as hypotheses and effective
parameter values? What new measurement
techniques might be important in informing the
learning process? How can uncertainty in the
predictions for individual places and different
scales of application best be constrained? In these
questions lies the future of distributed modelling
in hydrology and hydraulics.
particular catchments or river reaches, it is likely
that the representations of different places (in
terms of either effective parameter values or
hypotheses) might evolve quite differently. In
some cases (or for some purposes) it may be
possible to simplify model structures without
increasing prediction uncertainty, in other cases
more complexity may seem to be supported. The
concept is not a proposal for a single monolithic
model structure applied everywhere; it is for
a modelling platform that allows the representa-
tion of places to evolve, taking full account of
modelling uncertainties.
Such an evolution requires, however, that the
model predictions be tested locally. This could be
a purely qualitative evaluation based on local
stakeholder expertise (such as the involvement of
stakeholders in the development of a flood model
for Ryedale in Yorkshire by Oxford and Durham
Universities). Where quantitative hydrological
data are readily available, such as at established
gauging stations, then this can be seen as a process
of re-evaluation of model performance and/or
recalibration over time. This will be, however,
only a limited evaluation of the spatial predictions
of a distributed model. The same problems of
availability and commensurability of spatial data
arise as in model calibration. It might, therefore,
only be possible to implement such a learning
process for the distributed predictions by making
specific spatial surveys (such as incremental dis-
charge measurement campaigns), using remote-
sensing images (taking account of the uncertainty
in interpretation from image to hydrological vari-
ables) or post-event evaluations, such as flood
extent mapping after specific future events in
evaluating the predictions of hydraulic models.
As noted above (see 'Model rejection issues'),
we will learnmost in this learning process when it
is shown that the distributed predictions are
inadequate (and local stakeholders will be very
happy to point out obvious inadequacies). If the
model predictions can be shown to be inadequate
then it implies that some effort must be expended
infinding outwhy.Model failuremight be because
of inadequate characterization of the input uncer-
tainties, it might be because the model calibration
References
Abbott, M.B. and Minns, A.W. (1998) Computational
Hydraulics. Ashgate.
Abbott, M.B., Bathurst, J.C., Cunge, J.A., O'Connell, P.E.
and Rasmussen, J. (1986) An introduction to the
European Hydrological System - Systeme Hydrologi-
que Europ ´en, SHE. 1. History and philosophy of
a physically-based, distributed modelling system.
Journal of Hydrology, 87, 45-59.
Aronica, G., Hankin, B.G., Beven, K.J. (1998) Uncer-
tainty and equifinality in calibrating distributed
roughness coefficients in a flood propagation model
