Th-301A-4
Expected Future Performance of Salmon Abundance Forecast Models: Evaluation of Competing Models with and without Environmental Effects

Thursday, August 21, 2014: 9:20 AM
301A (Centre des congrès de Québec // Québec City Convention Centre)
Michael O'Farrell , Fisheries Ecology Division, Southwest Fisheries Science Center, National Marine Fisheries Service, Santa Cruz, CA
Arliss Winship , Biogeography Branch, Center for Coastal Monitoring and Assessment, National Centers for Coastal Ocean Science, National Ocean Service, Silver Spring, MD
William Satterthwaite , Fisheries Ecology Division, Southwest Fisheries Science Center, National Marine Fisheries Service, Santa Cruz, CA
Brian Wells , Fisheries Ecology Division, Southwest Fisheries Science Center, National Marine Fisheries Service, Santa Cruz, CA
Michael Mohr , Fisheries Ecology Division, Southwest Fisheries Science Center, National Marine Fisheries Service, Santa Cruz, CA
The management of Pacific salmon fisheries relies heavily on abundance forecasts, and there is continued interest in their improvement. Using Sacramento River fall Chinook salmon as a case study, we evaluated the scope for improving the current forecast approach that relates the Sacramento Index (SI; an index of adult age 3-5 ocean abundance) to jack (estimated age 2) spawning escapement from the previous year. Alternative models added effects of density dependence, local environmental conditions, the abundance of the previous cohort, and trends or autocorrelation in the jack-to-SI relationship.  Forecast performance was assessed with two cross-validation frameworks allowing evaluation of bias, accuracy, the ability of the models to track trends in the SI, and the potential for forecast errors to be of sufficient magnitude to cause management errors. Several models achieved higher accuracy than the current model, but no single model performed best across all criteria, and substantial forecast error remained across all approaches considered. Models incorporating trends or temporal autocorrelation in the jack-to-SI relationship demonstrated potential for modest forecast improvements with relatively little additional model complexity.