A Statistical Model for In-Season Forecasts of Sockeye Salmon (Oncorhynchus Nerka) Returns to the Bristol Bay Districts of Alaska

Abstract:

We developed a model for in-season age-specific forecasts of salmon returns using preseason return forecasts, age composition of in-season returns, cumulative in-season returns by fishing district, and age composition and an index of abundance from an in-season test fishery. We apply this method to the sockeye salmon (Oncorhynchus nerka) fishery in the Bristol Bay districts of Alaska. The model generates point estimates and Bayesian probability distributions for return numbers by age and river, and it provides an integrated framework for including all of the major data sources currently used in in-season forecasting. We evaluated model performance using early-season data from 1999–2001 and compared the effects of four information sets on forecast accuracy. The four information sets were as follows: I, district-specific inshore return data; II, inshore return data and test fishery data; III, inshore return data and preseason forecasts; IV, inshore return data, test fishery data, and preseason forecasts. Forecasts from information sets II, III, and IV were less biased than those from information set I. However, in terms of the forecast interval, forecasts from information set II were best because the 95% highest posterior density regions of forecasts from information set II covered the actual returns most frequently.