Pointers for the Future - Environmental Modelling: Finding Simplicity in Complexity - page 440

and remote sensing measurements appropriately when
using them to parameterize or evaluate a model.
Models may require unmeasureable properties, such
as the complex three-dimensional structure of the sub-
surface (see Mulligan and Wainwright, Chapter 11) or
properties that are measurable but not at the 'effec-
tive' scale required by the models. All robust models
require some form of observational paradox, in that we
perturb the system in order to collect data for param-
eters or testing. The integration of the modelling and
field-work programmes can help to reduce the impacts
of these perturbations. It could be questioned (Engelen
and also Millington et al .) whether the development of
databases from local informants for integrated models
incorporating human behaviour might cause behavioural
modifications and such models always to be one-step
behind reality. It is also important to support models
with sufficient observations. Brazier illustrates the case
that some relationships built into models may only be
based on a small number of measurements. Without an
understanding of this limitation, it is impossible to know
where to focus the research effort to provide improve-
ments when models fail.

Bond. Scale is implicitly built into all our model repre-
sentations and into our field measurements. Applications
of data measured at one scale to a model that is integrated
at another may lead to completely misleading results.
Further work is required on how to make measurements
at scales that are appropriate, both for parameterization
and model evaluation and how to make the scales of
modelling converge with those of measurement.

27.1.9 Modellingandpolicy

As noted above, Twery and Weiskittel point out that
models used in an applied context can often be relatively
simple because relatively straightforward predictions are
required. Supporting this argument, Nearing suggests
that different approaches are appropriate in modelling
for (soil-) conservation issues, whereas Thornes demon-
strates how simple concepts can give relative directions of
change that can be used in management decisions about
land degradation. Yet there is a contradiction here in that
Brazier suggests that environmental managers require
'accuracy' in prediction and Mulligan notes that 'pol-
icy makers require simple, definitive answers', whereas
Goldstein et al . (Chapter 26) say that 'we may often be
able to tolerate probabilistic forecasts'. However, Twery
and Weiskittel, and Cloke et al . (Chapter 25) also note
that there are problems of dealing with uncertainty in an
applied context (as with all other contexts). Visualization
may be an appropriate means of dealing with the latter
problem (Engelen) but there are serious issues of how
uncertainty is conceptualized by different groups. Is an
incorrect but definit(iv)e result better than a result that
will be perceived as vague (or evasive) when it is presented
with a large margin of error? Certainly, there needs to be
more communication about what is possible in terms of
prediction (see above), even if there is no clear answer to
this question at present.
Engelen and also Mulligan demonstrate another need
for simplicity in policy-based approaches, in that speed
in producing results can have a significant impact on
the uptake of a particular model. Simplicity in individual
components may lead to models being better able to
deal with integrated analysis. However, Brazier cautions
against the hidden complexity of many models, in that a
GUI may hide a vast database of hidden parameterizations
and assumptions. Schneider (1997) highlights the same
issue within the context of integrated assessment models
(IAMs) for the impact of global climate change. In the
multidisciplinary issues of climate change and land-use
change impacts that are the mainstay of research in

27.1.8.1 Data and empirical models
Empirically based approaches may be found even in
process-based models, as illustrated by Haraldsson and
Sverdrup and by Nearing. Very few environmental mod-
els contain no empirical content (for example, some CFD
applications: Wright and Hargreaves), and it is important
to be aware of this limitation. Given that process-based
modelling is designed to provide an improvement in
terms of portability issues (cf. Grayson et al ., 1992), this
limitation is significant. There will always be some limit
to portability and, unless we remember this, we may end
up unnecessarily rejecting the process model (rather than
the hidden parameterization). Inductive approaches may
mean that empirical approaches are very useful for defin-
ing the appropriate model structure from data, at least
at a particular level of aggregation. Twery and Weiskittel
also note that empirical models may provide an adequate
level of process representation for certain applications.

27.1.8.2 Data and scale

Zhang, Drake and Wainwright illustrate how 'scale' can
be a thorny issue in that different researchers perceive
it to be a completely different question. Certainly, deal-
ing with scale is a nontrivial process that requires quite
sophisticated analysis, as they illustrated, as did Perry and