Accounting for Climate Variability in Forecasting Pacific Salmon in Data-Limited Situations

Accounting for Climate Variability in Forecasting Pacific Salmon in Data-Limited Situations

S. Kalei Shotwell, Milo D. Adkison, and Dana H. Hanselman

Accounting for Climate Variability in Forecasting Pacific Salmon in Data-Limited SituationsThis is part of Fisheries Assessment and Management in Data-Limited Situations
To download the free PDF [1.1 MB], please enter:


Poor understanding of the major sources of environmental influence on Pacific salmon precludes quantitative forecasting in data-limited situations. Since 1997, low numbers of returning chum salmon to the Kuskokwim and Yukon rivers of Alaska have resulted in low harvests and significant negative economic and social impacts to rural residents of the region. The causes of these recent declines are unclear; however, poor ocean conditions for survival are thought to be important. No formal forecast has been available for this region, as estimates of the population size necessary to derive a quantitative forecast of returns were lacking. We recently generated abundance and escapement estimates for these two river systems. Our objectives in this study were to describe the spawner-recruit dynamics of this system and to identify important environment-recruit relationships. We explored a set of variables with plausible mechanistic relationships to five biological processes: freshwater survival, early marine survival, early marine predation, open ocean survival, and open-ocean competition. We winnowed variables in these life history categories through an exploratory phase, and then used formal model selection procedures on those remaining variables under restricted combination scenarios. Our best models implied strong environmental effects and explained 89% and 81% of the variability in the data in the Kuskokwim and Yukon respectively. Cross validation estimates of forecast error were much smaller for models containing environmental covariates, confirming their utility. We also performed stepwise variable selection on the same set of reduced predictor variables. Results were similar to the previous models, yet identified the most influential life history stage rather than the broad scope of the restricted category-based models. We recommend managers use both forms of model selection and concentrate future research efforts on processes that confirm the mechanisms implied by the best predictor variables. We caution managers to be conservative when applying these models to management decisions and to consider simulation analyses to incorporate uncertainty in the reported estimates. The procedures developed here are applicable to other data-limited salmon systems.

Item details