Sensitivity analysis of a model used in the management of ESA-listed salmonids: making sense of 10,000 parameters

Paul McElhany and Ashley Steel

NOAA Fisheries Northwest Fisheries Science Center, 2725 Montlake Blvd East, Seattle 98112. Paul.McElhany@noaa.gov

Managing ESA-listed Pacific salmon poses complex problems. The diversity of habitats required by salmon to complete their life-cycle demands that habitat protection and restoration strategies address a large number of natural and anthropogenic factors. The scientific challenge is to develop models that help managers address this complexity, without the models themselves becoming so complex that they are no longer useful decision-making tools. We evaluated the uncertainty associated with one complex model and present lessons learned from this project with respect to the evaluation of other complex ecological models and to the communication of model sensitivity analysis results to managers and policy-makers. The Ecosystem Diagnosis and Treatment model (EDT) uses habitat and fisheries information to predict the performance of Pacific salmonid populations in terms of abundance, productivity and diversity. The model has been widely used in Puget Sound and in the Columbia Basin to assist in setting recovery goals and to predict the consequence of proposed management actions. Despite this widespread use, we do not know how much confidence to place in the model’s predictions. The model has a very large number of input parameters, most of which are estimated with a high degree of uncertainty; currently, the model ignores this uncertainty. We have completed a large Monte Carlo sensitivity analysis of the EDT model to provide decision-makers with estimates of model precision. Example management questions raised by existing applications of EDT that can be better addressed by considering the results of our sensitivity analysis include, 1) Are EDT predictions precise enough to use for recovery goals?, 2) How much should we invest in habitat action prioritized by EDT?, 3) How much should additional effort should we invest in applying the EDT model? and 4) How should we allocate monitoring efforts to increase the precision of model predictions? These sensitivity analyses can provide quantitative information for answering these questions and improving model-based decision-making.