Geoscience Reference
In-Depth Information
that none of the models give a good prediction but, as in any testing of a theory or model in science, this
is actually the most valuable and interesting result in learning how to represent the processes.
Again, we are ending with a focus on the value of observation to modelling rainfall-runoff processes
but in a way that demonstrates that there has been significant progress in the last 10 years in understanding
some of the issues involved. The models of everywhere concept would seem to be a useful and positive
way ahead. It should change hydrological modelling practice in the future.
12.8 Some Final Questions
This is likely to be the last topic I publish before retiring as a hydrological modeller. When I started my
research career, I had thought to understand the evolution of landscape (having been heavily influenced
by Mike Kirkby as an undergraduate student). But to do that I reasoned that it would be really important
to get the water flows right. So my PhD thesis was concerned with modelling water flows on hillslopes.
The physically based finite element model I developed was not a great success (the story is told in Beven,
2001) but the problems I encountered were the basis for the rest of a research career. It is clear from the
discussions in this topic, however, that not all the problems have been solved and it seems pertinent to
end with some questions that need to be addressed in the future.
In Chapter 9, I suggested a framework for developing a new generation of rainfall-runoff models and
in this chapter I have tried to convey some really positive ways in which rainfall-runoff modelling might
develop in the future. It is, however, difficult to predict the future of rainfall-runoff modelling. There
are no predictive techniques for methodological advances. Hydrological science is currently in a period
of gradual development typical of “normal” science in the sense of Thomas Kuhn. It was Kuhn who
suggested that science often progresses by “paradigm shifts” interspersed by periods of normal science.
Computer advances have certainly made it easier to prepare GIS databases of spatially distributed inputs,
to present space and time variable model outputs and to calibrate and carry out sensitivity analyses of
more complex models and parameter spaces, but there have been no major paradigm shifts in approach
unless we count a shift from purely deterministic simulation to stochastic formulations (but of much the
same model structures and without much in the way of additional supporting data).
If
such
a
paradigm
shift
is
to
be
achieved,
then
we
need
advances
that
respond
to
the
following questions:
•
Can we provide some guidance on the value of different types of data for hypothesis testing and
constraining the uncertainty in rainfall-runoff models, particularly where there might be little in the
way of discharge observations available?
•
What novel measurement techniques can be envisaged and implemented in the future that will be
effective in hypothesis testing and constraining the uncertainty in rainfall-runoff models?
•
What
new
process
representations
can
be
supported
by
the
observations
in
different
types
of catchment?
•
Can we make remote sensing a central part of defining and evaluating the dynamic behaviour of
rainfall-runoff modelling? The difficulty of sensing the subsurface and the need for an interpretative
model in deriving hydrologically useful information from remote-sensing images will generally limit
the utility of such data for rainfall-runoff modelling.
•
Can we find some ways of dealing with epistemic uncertainties in model definition, evaluation and
prediction that does not involve pretending that they are equivalent to statistical uncertainties? In
particular, can we find ways of distinguishing between information and disinformation (or some way
of eliminating the latter by improved observational methods) for hypothesis testing and constraining
the uncertainty in rainfall-runoff models?
