Applying uncertainty principles in environmental modelling

Abstract

The authors implemented a causal probability network (CPN) model using the HUGIN shell to represent part of the overall effect of acid precipitation on lakes. A tool for entering raw data into the CPN model, based on earlier rule-based modeling efforts by the RAISON project at Canada Centre for Inland Waters, Environment Canada, was developed. The network editor developed allows the water chemistry and flow data to be directly applied to calculate marginal distributions ready for input to HUGIN. The authors used the RAISON experience to draw preliminary conclusions concerning environmental models which have representation and propagation of uncertainties in hypotheses and outcomes. The authors show the utilization of Bayesian inference to the watershed aggregates defined for the earlier, rule-based model developed for acid rain. Naturally occurring and industrial causes of aquatic acidity are incorporated in a network with probabilities generated from observations. A sample CPN with four nodes is illustrated. It represents the relationship between water chemistry, color, sulfate load, and the percent reduction of acid neutralizing capacity in a lake.<>