Environmental Engineering Reference
In-Depth Information
23.2 Working with Model Results
Within the process of constructing a model, ecological processes usually have to be
quantified in extensive measurement activities, handbooks have to be consulted,
code has to be written, debugged and sometimes, the reasons for unexpected model
behaviour have to be analyzed, understood and eventually corrective action has to
be taken (Chap. 2). Once the primary part of the developmental work is done and
after achieving first results, the work enters a further stage of development: securing
the specific correctness of the model results and then, continuing to work with the
assessed model and its results. Now, we need to understand what the model results
mean, how robust and reliable they are. This process is frequently referred to as
validation
. Depending on the model complexity this part can be as demanding and
relevant as the primary model development itself.
Validation of differential equation-based models has a long history (see e.g. Power
1993). For other model types specific evaluation approaches exist, which are accom-
panied by a large body of literature on specific aspects of model analysis (e.g. Rykiel
1996; Klepper 1997; Sargent 1998; Jager and King 2004). On this basis, some general
steps of quality assurance for models can be deduced for most approaches (Jakeman
et al. 2006). In this context we discuss the principles of parameter identification,
sensitivity analysis and the process of model validation. Finally, we will give some
suggestions on how to scientifically communicate model structures and model results.
We explicitly do not emphasize the usage of the term verification which is occasion-
ally applied in the context of model evaluation (Oreskes et al. 1994; Mitro 2001). The
term verification derives from the Latin
verificare
which stands for
making true
.
Philosophers are very careful in using the term
truth
. Can a model represent the
truth? In itself it can be formally and mathematically correct which is the reason
that the term verification is sometimes used for ensuring a mathematical correct
formulation of the model (e.g. Oreskes et al. 1994). There is always a necessary
deviation between model representation and represented reality, between
explanans
(the statement that explains) and
explanandum
(the context that is explained).
Modellers as well as those who apply models must be aware that a formal construct
developed for a system representation and real things
as they are
, do not necessarily
coincide. They are not identical. A model can capture eventually some of nature's
interesting properties, but not on the basis of a true identity. The important point is to
find out to what degree a similarity can be expected. In our opinion, this is indicated
better in the term validation than in the term verification, though the latter sometimes
is favoured in the literature (e.g. Sargent 1998; Manson 2003).
23.3 Assuring Correctness of Model Results
The protocol for model analysis follows certain steps, in particular, the specification
for model parameters, their possible ranges, and the analysis of the model behaviour
under specified assumptions (Sect. 2.4 and Fig.
23.1
). These steps are based on
