Bayesian Network Tool and expert opinions used to make predictions of probabilities of non-target attack from introduced biological control agents

Biosafety risks of proposed biocontrol agents (BCA), are generally determined through laboratory host specificity tests (especially physiological host range of parasitoids). The outcome of interactions of proposed agents in the natural environment can differ from laboratory-based predictions. We proposed the adoption of a more comprehensive, ecologically-based, probabilistic risk assessment approach to BCA introductions. A Bayesian network can integrate information on probabilities and uncertainties of a BCA to disperse, establish in new habitats, interact with non-target species in these habitats, and even negatively impact the populations of at risk nontarget species. The model, Biocontrol Adverse Impact Probability Assessment (BAIPA), was developed as a structured decision-making tool to assist researchers and regulators. To test the model we ran a series of expert elicitation workshops and provided the experts with detailed summaries of the known biology and ecology of the Target pest, the BCA and one at risk non-target species. After reading and discussing what was known and what was uncertain or not known, each expert selected, using their best informed estimate, the probability of the state of each input variable. These were then averaged across experts and used to populate each Input Model Component. For each case study we considered two different habitats. We will present the model predictions for direct impact on nontarget populations caused by an invasive generalist parasitoid (e.g. Meteorus pulchricornis) and by a purposefully introduced specialist BCA (e.g. Cotesia urabae (Braconidae)).