Developing Bayesian network models within a Risk Assessment framework

Abstract

The risk assessment framework is increasingly being applied to examine both human and non-human stressors on ecological systems. Risk-based decision-making aims to quantify the likelihood of a threat occurring, the consequences of this to an ecological system, process or value, and the associated uncertainty in the predictions. Until recently, the ability to predict changes in dynamic ecosystems due to stressors was limited by both the poor understanding of the drivers of ecological processes and structure, and the lack of modelling tools that could represent such complexity with associated uncertainties. However, the recent growth in the use of Bayesian network tools for ecological risk assessments has resulted in major advances in better understanding and managing ecosystems despite their inherent complexity. Bayesian networks have the advantage of being able to investigate the impacts of multiple stressors in complex environments, while explicitly acknowledging the associated uncertainties resulting from inherent variability and lack of knowledge of ecological systems within an adaptive framework. Bayesian networks have the flexibility to incorporate diverse knowledge systems, ranging from 'gut feel' to quantitative process-based or simulation models. In this paper, we discuss the relationships between the risk assessment framework and Bayesian network building process, and will illustrate the main concepts with a series of Bayesian network models.