Pointers for the Future - Environmental Modelling: Finding Simplicity in Complexity - page 439

hide underlying weaknesses in the approach. (Keith Beven
often warns to be wary of modellers presenting 3D graph-
ics because it means they have generally spent more time
in producing them than the model results - though that is
less likely to be the case today than previously). However,
the communication and understanding of model results
is often as important as the results themselves - especially
when addressing a nonmodelling audience - and output
sophistication has to grow in line with model sophisti-
cation or we will not learn enough about model (and
system) behaviour from the process of modelling. As with
every aspect of our scientific approach, there should be
a transparency in what we do and how we present our
methods and results (see Haraldsson and Sverdrup). The
increasing move to open-sourcing models holds much
promise in this regard, though remains uncommon for
the most sophisticated and well established models.

relatively low cost, and indeed this is a major strength,
allowing detailed analysis. But it can only provide a partial
solution on its own (e.g. Klemes, 1986, 1997).

27.1.8 Data issues

The distinction between 'modellers' and 'field workers'
can lead to problems with the use of field data for model
testing. Data collected in particular ways may contain
hidden effective parameterizations that generally lead to
unnecessary calibrations and the propagation of further
error through the model system. Error enters throughout
the modelling process, where Baird, and Mulligan and
Wainwright (Chapters 10 and 11) note that our ability to
measure properties accurately means problems in terms
of parameterization and model testing. Although there is
general acceptance that parameters contain uncertainty,
it is generally assumed, albeit implicitly, that the data
used to test models is without error (but see Young and
Leedal). This assumption is completely illogical! Most
measurements themselves are models: a mercury ther-
mometer represents temperature change as the change in
volume of a thin tube of mercury, a pyranometer uses the
increase in temperature of an assumed perfect absorber
(black body) as a surrogate for the incident radiation
load. Moreover many measurements like this one are
taken at a point and assumed then to represent an area or
volume. The measurement is a measurement at the point
measured. Everywhere else it is a model (interpolation).
Nearing demonstrates that natural variability in measured
properties has significant implications for the best case
scenarios of model tests. We cannot hope to produce more
accurate results than the properties we measure or their
spatio-temporal variability (see discussion in Wainwright
et al ., 2000). The baby should remain with its bath water!
Complex models have significant data requirements,
as illustrated by Nearing for process-based soil-erosion
models. Yet there is often a reluctance to fund the work
required to collect the necessary data, especially in the
long term. The low-cost conundrum strikes again! But
without detailed spatial measurements and long time
series, we will be unable to evaluate model performance
beyond simply trivial levels. This lack makes the mod-
elling no more than an inexpensive research pastime that
may lead to a better system conceptualization but is held
short of its potential use in contributing to the solu-
tion of serious environmental problems. Large datasets
are becoming available via remote sensing and GIS inte-
gration of existing databases but, as Zhang, Drake and
Wainwright point out, there is a need to interpret field

27.1.6 Process

Models provide a means of addressing the link between
the observable and the underlying cause. The underly-
ing process-form debate is a critical one in ecology and
geomorphology. One of the main advantages of the mod-
elling approach is that it allows us to understand the
limitations of traditional forms of explanation. Interac-
tions of simple processes lead to the emergence of form.
Difficulties in interpretation arise because of inherent
nonlinearities due to scale effects in both process and
form (e.g. Zhang, Drake and Wainwright). Both process
and form possess elements of complexity (Mulligan and
Wainwright, Chapter 11), and it is not necessarily the case
that we need complexity in one to explain the complexity
in the other.

27.1.7 Modelling inan integratedmethodology

Despite isolationist claims (unfortunately from both
sides), modelling is not an activity that exists in isolation.
Field or laboratory work is often seen simply as a means of
testing model output - it is not surprising, then, that mod-
ellers are often perceived as aloof and ignorant! As we have
noted already, we always take our preconceptions into the
modelling process, so we should at least try to make them
informed preconceptions. In reality, there is a strong
loop between fieldwork that suggests new models, which
require new data for testing, which suggest further model
developments, and so on. Mulligan and Wainwright, in
Chapter 11, also demonstrate that there is an important
interaction between physical and numerical models in
the same way. Modelling is often promoted because of its