Environmental Engineering Reference
In-Depth Information
validity of the model therefore is to estimate how well the given ecological
context conforms to what the chosen modelling approach actually can capture.
Hence, three questions need to be considered when starting to assess the quality of
a computer model (1) what are the underlying assumptions of the model and (2)
how do these assumptions relate to the model results of the specific case, and (3)
do the boundary conditions hold?
The importance can be well illustrated with ordinary differential equation (ODE)
models (Chap. 6). Characteristic for ODEs are functional relations of a limited
number of variables. When using these relations within an ecological context, in
most implementations it is assumed that heterogeneous structures and activities can
be functionally represented in form of homogeneous variables. The degree to which
this assumption holds is very important for the validity of the results. ODEs are
frequently used to describe population growth or density. Starting from an
Eulerian
perspective (see Turchin 1998), for a population study a point in space can be
focussed and the fluxes or the numbers of dispersing population members are
counted for this point. The counts (f.i., the bypassing individuals of the population)
are summarized and averaged over the inspected time interval. In physics the
density of particles in a liquid may be measured this way. A question that needs
to be considered is how reasonable it is in a particular application to assume that
population density can be measured with such an approach? This crucially depends
on species characteristics and behaviour. What might be operational for certain
unicellular organisms in an homogeneous medium may be inappropriate for colo-
nial birds - with many situations in between where the degree of adequacy is an
issue to be carefully examined.
When we have assessed these questions we can proceed to step 2: in what
way do the specific assumptions interfere with the model outcomes? What we
have at hand is a formal abstraction, not the biological objects themselves. That
is, the model does not give us information about the area in which the population
was recorded, specific properties of the individuals, their possible interactions,
spatial heterogeneities in the environment or temporal/seasonal differences -
unless specifically implemented. Model representation focuses, e.g. only on the
temporal changes of the sheer numbers as they were recorded in the point of
interest.
The third aspect refers to the problem delimitation. In constructing the model,
decisions have been made as to which aspects to include, and which ones to leave
out of consideration. In a specific application or repetition of a situation it may well
be the case that unexpected external influences play a role which were not consid-
ered during development. In most of the practical cases where model predictions do
not hold, this implicit
ceteris paribus
condition was not valid (see Sect. 23.4.1).
In all further steps of the assessment of specific model qualities, e.g. how it fits
to empirical data and situations, the answers of the questions about assumptions
and interactions will be of great value to judge the reliability limits of a devel-
oped model. Biological relevance and plausibility are crucial aspects in model
evaluation.
