Environmental Engineering Reference
In-Depth Information
selection of species composition of the zooplankton, phytoplankton and macro-
phyte community, respectively, were analysed. A large variety of approaches
were discussed in this context, see M
uller and Leupelt (1997) for an overview of
goal functions. From different perspectives, several features could be identified,
which were possible to be brought together in a theoretic ecosystem pattern
(Jørgensen 2002). Furthermore, biomass has been applied in two cases (Stra-
skraba 1979; Radtke and Straskraba 1980), exergy in 21 cases (Zhang et al.
2010). To test and exemplify some of the approaches, the Lake Glumsø model
was employed. The overall results supported the hypothesis of the application of
maximum exergy as a useful goal function as an emergent property of ecosys-
tem dynamics (Jørgensen 2002).
€
19.3.4 Equation Discovery
In the early phases of development of the Glumsø model a great deal of work was
put into the search for appropriate equations. During this period the search was
carried out largely because of the limited powers of computers at that time. In order
to save computating time, the most simple equation was looked for that was able to
simulate observations within acceptable accuracy. A more complex equation would
not only be more costly in computing time but would also likely involve more
parameters and eventually introduce a higher uncertainty in the model (cf. Costanza
and Sklar 1985). In some of the versions of the Glumsø model up to 7-8 different
expressions of various temperature dependencies were tested on process equations
for their efficiency, i.e. contribution in improving the precision of model predictions.
But does a model need to be entirely the work of a modeller - or could a
computer programme also be used in model development? At least computer
algorithms,
machine learning
techniques (see Chap. 14), can help in model con-
struction if it is to identify the most reasonable structure. If a large set of equations
exists that lead to nearly comparable results, a computer-based testing of alterna-
tives can be helpful. The approach is to some extent comparable with parameter
identification (see Chap. 23), where the value of a parameter is changed as long as it
fits an optimization criterion. Here, the number of equations and their algebraic
structure is varied and then the simulation results compared with the data. This is
what the work of Atanasova et al. (2006, 2008), Todorovski and Dzeroski (2006),
Todorovski et al. (1998) and Vladusic et al. (2006) describe using the Lake Glumsø
data as an example.
19.4 Overall Contribution
Over the decades, the focus of interest in ecological modelling has successively
changed - this went in line with alterations in the environmental situation. In the
late 1970s,
the environmental
impact of excessive nutrient
input
leading to
