Bayesian networks for modelling habitat suitability of an endangered species

Abstract

Bayesian networks (BNs) are simple graphical causal models that have been applied in a diverse range of fields. They were first conceived in the 1980s at the interface between artificial intelligence, expert systems and statistics, to deal with problems of reasoning and decision making under uncertainty. They have served many purposes including diagnosis, prediction, simulation and analysis, and are particularly suited to problems involving causality with inherent uncertainty. In environmental modelling, one of their strengths is in their ability to integrate various forms of knowledge across disciplines into a single modelling framework. In this paper we introduce a BN that was developed to model the habitat suitability of Astacopsis gouldi, the endangered giant freshwater crayfish in Tasmania. The BN was linked to GIS, thereby placing the model inputs and outputs in a spatial context. The modelling work is based on research by the Tasmanian forestry practices industry that developed a set of habitat mapping rules for the species that were translated into a habitat suitability map. The BN is used to represent current knowledge of Astacopsis habitat, however, unlike the previous mapping work, all causal relations are made explicit and transparent to users. The BN also allows management strategies to be tested, which better promotes system understanding, and its modular architecture will enable it to be integrated into a larger model or Decision Support System, making it more useful in a decision making context.