Pointers for the Future - Environmental Modelling: Finding Simplicity in Complexity - page 438

Reductionist approaches come about because of our
difficulties in conceptualizing processes beyond scales
relating to human experience, as noted above. But they
frequently defeat the object, in that for environmental
systems we are interested in explaining the whole.
Simple models that can illustrate emergent behaviour are
useful exploratory tools and can illustrate the dominant
controls on different parts of the environmental system.
They can help explain chaotic and complex behaviour
in a way that linearized models cannot, as pointed out
by Favis-Mortlock. Millington et al ., Perry and Bond,
Fisher and Mazzoleni et al . demonstrate how the use
of individual-based models can provide explanations of
emergent behaviour in this respect.
Mulligan and Wainwright (Chapter 2) question
whether there are different levels of emergent behaviour
and whether or not they form hierarchies that might allow
us to simplify the modelling process between different
scales. Where different forms of complex system come
together, we may need to deal with one as a stochastic
variable to assist analysis, as demonstrated by Thornes.

reinventing the wheel. Both Thornes and Haraldsson and
Sverdrup question the wider usefulness of models other
than to modellers. This viewpoint is essentially derived
from the 'understanding' perspective. As illustrated by
Mulligan and by Engelen, some models may be designed
specifically for use, even if they might lack the most
powerful levels of explanation. Haraldsson and Sverdrup
define 'survivor' models as those which remain in use
after some time by the wider community. It is not neces-
sarily the case that such models are always the best - they
may simply be the easiest to apply, or the cheapest, or the
ones that fit a particular pattern of explanatory fashion,
or the ones that are most prolifically written about - but
it may be appropriate to investigate which models others
are using before embarking on a new study.
Some of the more advanced models may be considered
as 'game-playing tools' in order to develop further
understanding. In some senses, this approach is similar
to the heuristic method put forward by Thornes. It is
important to retain a sense of fun in our investigations,
not least so that our ideas do not become dulled and we
fail to see alternatives.

27.1.4 Howshouldwemodel?
Haraldsson and Sverdrup suggest that if modelling is
being carried out to develop our understanding of the
environmental system, we should build our own models
rather than simply apply a readily available model. The
ready-made model may be inappropriate to our specific
application, and it may be difficult a priori to assess the
extent to which this may be so. The opposite viewpoint
is expounded by Wright and Hargreaves, who suggest
that (at least for complex CFD code), we are better
off applying tried-and-trusted code, perhaps even from
a commercial source (but see the discussion on Open
Source initiatives below). Even so, it is important for
the underlying concepts to be thoroughly understood to
avoid the occurrence of problems further down the line
of the modelling process.
An appropriate answer to this question is, as is often
the case, something of a compromise between these two
extremes. In particular, the purpose of the models is an
important consideration, as noted by Perry and Bond.
Similarly, Twery and Weiskittel note that in policy appli-
cations as well as in others, different questions may be
most appropriately answered by different forms of model.
This is clearly a case of horses for courses! While the
development of more available and efficient modelling
frameworks and toolkits means that it is increasingly
easy to develop our own models, we need to beware of

27.1.5 Modellingmethodology

We have discussed modelling methodology in Chapter 2,
so only provide a few brief points here. Parameterization
is often a function of the scales at which environmental
models operate (see also the discussion above). In one
sense, it can be considered as an emergent property of the
way a system operates at a smaller spatial and/or temporal
scale. Yet parameterization is often paid too scant a regard
in the application of models. The sophistication of our
models is often much greater than the capacity of our data
collection efforts to parameterize them, though efforts are
abound to produce 'self-parameterizing' models that are
better suited to application by non-modellers since they
are delivered with at least a basic set of parameteriza-
tion data for application anywhere (see Chapter 20 by
Mulligan). Similarly, Haraldsson and Sverdrup note that
calibration is too often used to force a result that tells
us nothing about the system (and everything about the
modeller's preconceptions - see also Young and Leedal).
If calibration is employed, there should always be an
attempt to assess whether it is reasonable, otherwise the
whole exercise is virtually pointless
Visualization can be an important role of the modelling
process, as illustrated by Engelen, Twery and Weiskittel,
Mulligan and Wright and Hargreaves. But we should not
be misled by the presentation of graphical results that may