Finding Simplicity in Complexity in Biogeochemical Modelling - Environmental Modelling: Finding Simplicity in Complexity - page 274

Thus, such models may be useful for simulations or
design work after careful calibration, but do not expect
them to explain processes. Without the calibration, these
models lack the ability to generate a priori predictions.
In modelling the focus is often on the output instead
of robustness and performance, which makes the 'cult
of success' very strong. The 'cult of success' implies that
the calculated output is expected to pass through all
observation points. If not, the user of a model might
soon hear 'The model is wrong ... ' or 'Well, it is obvious
that the model does not work ... '. On the other hand, if
it is obvious that the model was calibrated by adjusting
one or several parameters, critics will soon remark, 'You
can probably make the thing fit anything ... '. Resist
this! Resist the temptation to over-calibrate! Ignore such
comments - they are never made from insight! The fit
need not be perfect; the line need not go through all data
points. Performance should only be sufficient to give an
adequate answer. It is from the lack of perfect fit that
learning is gained.
We may classify existing models into categories,
depending on which scale they are applied to as well as
the degree of integration of multiple processes over those
scales. A list, not claiming to be comprehensive, would
include examples as follows.
Single-problem models include: Botkin et al . (1972);
Cosby et al . (1985); de Vries et al . (1989); Jansson
(1991); Chen (1993); Sverdrup and de Vries (1994);
Sverdrup and Warfinge (1995); Crote et al . (1997); Berge
and Jakobsen (1998); Kimmins et al . (1999); Wright
et al . (1998); Warfinge et al . (1998); Crote and Erhard
(1999); Kros (2002); Sverdrup and Stjernquist (2002).
Integrated multiple-system models include: Parton et al .
(1987); Mohren et al . (1993); Sykes et al . (1996); Kram
et al . (1999); Sverdrup and Stjernquist (2002). Higher
hierarchy models include: Alcamo et al . (1990); den
Elzen (1994): Gough et al . (1994); Belyazid et al . (2010);
Sverdrup et al . (2011).
Examples of such models listed according to the cate-
gories described and the problems that the models address
are given below:
•

- atmosphere models (EMEP);
- groundwater substance transport models (FREE-
QUE).

Integrated multiple-system models:
- nutrient cycle - population models;
- nutrient cycle - trophic cascade models;
- climate change - carbon cycle - vegetation models
(CENTURY, FORSKA, BIOME);
- climate change - acidification models (FORSAFE);
- forest - management - nutrient-cycle models (FOR-
GRO, FORSAFE, PnET).

•

Higher hierarchy models:
- decision-management-biogeochemical models
(RAINS, IMAGE, CASM);
- neural network models for complex societal systems
(Adaptive Learning Network).

•

In the higher hierarchy models, then, biogeochemistry
only forms a subsystem of the whole model system, and
the focus of the question may be totally outside the
biogeochemical domain, even when the biogeochemical
modules form essential parts of the interactive system (see
Figure 17.9).
Most biogeochemical models do not exist as computer
codes or have any fancy names or acronyms attached;
they exist as mental models and are mostly only recorded
in language on paper. Many computer models exist only
to produce a few published articles, only to disappear
when the originator gets a new job. The mental models
will survive but the code is lost.
Very few of these models are operable in reality or useful
to people other than those who built them. Very few of
the models have been user-adapted, generally no support
apparatus will be in existence. This problem is related to
the fact that the answers are not driven by the models
but rather that the questions determine which model is
required, and every new adaptation of the question will
demand a modification of the model or even a completely
new model. Just taking a model, getting its answers and
then searching for the appropriate question that might
fit the answer is a backward way of working. It will only
work well if the questions are highly standardized.
Some useful models for bulk soil chemistry have come
from acidification research and we will use these as
examples. They all had the requirement that they should
be regionally applicable, testable, have a priori capac-
ity and do not need calibration on essential parameters.
These requirements excluded all of the earlier existing
models for soil chemistry. Of course, there are as many
biogeochemical models as there are questions to ask or

Single-problem models:
- acidification models (SAFE, MAGIC, SMART);
- eutrophication models (MERLIN, MAGIC-WAND,
Vollenweider eutrophication model);
- forest models (JABOWA, FORECAST, TREEGRO,
FOREST-BGC/PnET, NuChem, FORSANA);
- simple cycle models (carbon, nitrogen, mercury,
magnesium, etc.) (GOUDRIAN, DECOMP, SOIL,
SOIL-N, COUP, PROFILE);