Data-adaptive reduction of process-based models

Process-based models of environmental systems typically are very complex structures. This complexity arises from the attempt to describe the manifold natural processes and intricate biological interactions of environmental systems in mathematical terms. Because of their high number of state variables and parameters, the resulting complex models are difficult to calibrate and detailed model analysis is needed to extract the key governing processes within these complex structures. Acknowledging these problems, the present thesis aims at finding a method to reduce complex process-based models. The main objectives for the development of the new method are its general applicability, its automated execution and its ability to construct reduced models which are interpretable in terms of system-specific mechanisms. The Mapping-based Genetic Reduction technique (MAGER) proposed in this thesis is a data-adaptive black-box procedure based on Genetic Programming which can be applied to ordinary differential equation models. In the course of this thesis, the MAGER scheme is applied to three predator-prey and consumer-resource models of different dimensionality. It is found that even relatively simple models can be reduced further and the results show that a formal conformity of physical and biological oscillating systems exists. In addition, the reduction of the biological systems involves a change in description level. Instead of traditional density or traits variables, the new models incorporate descriptions of biological interactions which leads to the notion of Ecological Interaction Models (EIM) for this new model class. The uniformity of the results further points to the generality of the EIM descriptions. As the MAGER scheme only depends on time series data and ignores former model structures, it is suggested that the method can also be applied to measured data or models from other scientific disciplines which offers many possibilities for further studies.

Level thesis:degree_level

thesis.doctoral

Grantor dc:publisher

Universität Oldenburg

Year

2008

Author dc:creator

• Bernhardt, Knut

• [Keine Schlagwörter von Autor/in vergeben.]

Repository record source_url

http://oops.uni-oldenburg.de/721

OAI identifier oai:identifier

oai:oops.uni-oldenburg.de:721