ON THE NECESSITY OF USING IMPRECISE PROBABILITIES FOR MODELING ENVIRONMENTAL SYSTEMS

One of the most important problems in the application of methods of pa rametric statistics to environmental systems is the impossibility of v erifying the assumptions on probability distributions (e.g. the assump tion of normally distributed measurements is usual but hardly exactly true). If Bayesian techniques are applied, the knowledge of probabilit y distributions is even worse, because also vague prior knowledge (typ ical in modelling environmental systems) must be formulated in the for m of (precise) prior probability distributions of model parameters or model structures. These two examples demonstrate the necessity of usin g imprecise probabilities in order to avoid arbitrariness in the choic e of probability distributions. In spite of these well-known problems, imprecise probabilities are rarely used in environmental systems anal ysis and prediction. In order to motivate a change of this situation, this paper briefly reviews various techniques for the formulation of i mprecise probabilities, and it demonstrates the advantages of using im precise probabilities in a Bayesian context (for prior distributions a nd for measurement distributions) with a simple didactical example. (C ) 1997 IAWQ. Published by Elsevier Science Ltd.