Multicriteria Bayesian analysis of lower trophic level uncertainties and value of research in Lake Erie.

摘要

Human activities have severely disrupted the Lake Erie ecosystem. Recent changes in the structure of the lower trophic level associated with exotic species invasions and reduced nutrient loading have created ecological uncertainties for phosphorus and fisheries management. Decisions that naively assume certainty may be different and suboptimal compared to choices that consider uncertainty. Here we illustrate how multiobjective Bayesian decision analysis can recognize the multiple goals of management in evaluations of the effect of ecological uncertainties on management and the value of information from ecological research. The Lake Erie Ecological Model was used to project the impacts of each combination of management actions and lower trophic level parameter values for use in a decision tree. Value judgments and subjective probabilities required by the decision analysis were provided by six Lake Erie fishery agency biologists. The analysis shows that explicitly considering lower trophic level uncertainties can alter decisions concerning phosphorus loading and fisheries management. Of the research projects considered, investigation of goby predation of zebra mussels (Dreissena sp.) and lakewide estimation of secondary production appear to have the greatest expected value for Lake Erie phosphorus and fisheries management.; Unfortunately, in practical decision analyses, the “curse of dimensionality” entails making simplifying assumptions that can introduce errors into estimates of evaluations. In order to assess the effects of simplifying assumptions on the evaluations, we consider eight different problem structures that differ in terms of the decision space (continuous versus discrete), decision set (just fisheries management versus fisheries and phosphorus management) and uncertainties (just lower trophic level uncertainty versus both lower trophic level and other model inputs). To reduce errors arising from discretization of the decision space, we implement a multidimensional cubic spline for interpolating the performance of alternatives between a few simulated points. Results show that the simplifying assumptions can yield different estimates of evaluations. However, the average benefits from the research projects considered, no matter which problem structure is considered, are potentially greater than their cost. We also find that changes in the weights assigned to management goals affects decisions and value of information more than do changes in probability judgments.