 |
 |
 |
 |
 |
 |
Methods & Models |
 |
Methods |
 |
 |
|
|
A mathematical models is the heart of any decision support system; the problem every
mathematical modeler has to face: The more realistic a model, the more complex its
behaviour - much to the users' distress.
In contrast, scientist only trust models that closely reproduce reality. IMPULSE wants
to resolve this conflict by using innovative methods:
» model aggregation
» data assimilation
Model aggregation - how to say more with less
It hardly makes sense to simplify a complex model by leaving out details that appear
to be unimportant: Relevant mechanisms or co-effects might be overlooked.
We must rather rethink our traditional approaches:
How can we say more with less?
Especially two concepts form the basis of model aggregation:
» By introducing effective parameters models with complex structures
(e.g. a large ecosystem model) are reduced to simpler models, that still fit the
specified phenomenon.
» Models with complicated boundary conditions can use simple
boundary condition without loss of precision, if coupled with an appropriate
nonlinear data model, e.g. neural networks. In this way the resolution of a hydrodynamic
model of the North Sea can be reduced from 500m to 20km.
Data assimilation - how to make a more reliable forecast
The forecast of a model is anything but precise: Too many inaccuracies can distort the forecast.
Therefore, it is important to continuously verify the result of a model:
If required model parameters and state variables need to be changed, in order to improve the forecast.
This can be acchived by comparing existing data with model results.
Data assimilation not only improves model forecasts but also helps us to learn about
the major weaknesses of a model.
|
|
|
 |
|
|