Because of the complexity of nature, our models of environmental systems must be crude simplifications of reality. If such models are based on a representation of the most important constituents and on a well-parameterized description of the key processes in a system, they can often simulate key aspects of observed pattern. This makes them useful to assess the degree of our understanding of the system mechanisms, to document and communicate this knowledge, and to show our current knowledge of future system behaviour. The degree to which observed output can be represented by an environmental model may often be satisfying for an environmental scientist, who is aware of unavoidable model deficits. However, the presence of systematic deviations between model results and observations caused by these deficits make it very difficult to derive statistically meaningful uncertainty bounds to model predictions. Such uncertainty bounds would be relevant for use of such predictions in environmental management. Different strategies to deal with model bias in environmental modelling will be discussed and the attempt will be made to work out recommendations for their use. In particular, techniques discussed and illustrated with examples will contain statistical bias description, constructive methodologies for improving the deterministic core of models, and techniques to add adequate stochasticity to model descriptions.

Keywords: Systematic errors; Bias; Environmental modelling; Stochastic models

Biography: Peter Reichert did his Masters and PhD in theoretical physics at the University of Basel, Switzerland, before joining Eawag, the Swiss Federal Institute of Aquatic Science and Technology. He works in the areas of aquatic ecosystem model development, statistical problems of environmental modelling, and use of models and societal valuations for decision support in environmental management.