Joint use of multiple environmental assessment models by a Bayesian meta-model: the Baltic salmon case

Abstract An approach to construct a meta-modeling framework using Bayesian calculus is presented. It allows the inclusion of several—both causal and empirical—models in a Bayesian belief network that is capable of real-time updating of uncertainties in the system. The approach can be used for diagnosis and forecasting purposes. It is illustrated with the assessment problem of the threatened, wild salmon stocks in the Baltic Sea. As usual, the institutions responsible for the management are prone to accept well-known standard assessment models. In this case, the Virtual Population Analysis approach and a set of regression models have been used. A debate on possible superiority of one or more of these models over the others is met at each annual international negotiation among stock assessment experts. The need has arisen to develop a methodology that would allow analytic fusion of these models to learn from the properties of the models and from the problem setting in general, as well as to facilitate probabilistic predictions that would utilize as much of the available information as possible. Moreover, the probabilistic approach is very compatible with and supportive to the use of the precautionary principle (risk-averse attitude). The belief network approach has shown appropriateness for handling of this type of problem in environmental and resource management.