Inseason Forecasting of Bristol Bay, Alaska, Sockeye Salmon (Oncorhynchus nerka) Abundance Using Bayesian Probability Theory

Abstract

A Bayesian probability model was developed for the Bristol Bay, Alaska, sockeye salmon (Oncorhynchus nerka) fishery to examine potential benefits of using a formal framework for incorporating several independent run size estimators into a single "best" estimate of total run size. To simulate performance of the technique, a hindcasting procedure was used for the years 1980–87. The Bayesian composite forecast was always more accurate than the least accurate individual forecast and was sometimes more accurate than the most accurate individual forecast. Since forecast accuracy for each independent method varied greatly both among and within years, the Bayesian method avoided the difficult problem of selecting the most accurate individual forecast and allowed forecasts to be easily revised each day of the season as new information became available. The documentation provided by our method should make it easier to evaluate and improve inseason run size estimation procedures and should allow for smoother transitions when management staff changes occur.