R.M. Peterson Milo D. Adkison 2000 Historically, forecast error for returns of sockeye salmon Oncorhynchus nerka to Bristol Bay, Alaska, has been large. Using cross-validation forecast error as our criterion, we selected forecast models for each of the nine principal Bristol Bay drainages. Competing forecast models included stock-recruitment relationships, environmental variables, prior returns of siblings, or combinations of these predictors. For most stocks, we found prior returns of siblings to be the best single predictor of returns; however, forecast accuracy was low even when multiple predictors were considered. For a typical drainage, an 80% confidence interval ranged from one half to double the point forecast. These confidence intervals appeared to be appropriately wide. application/pdf 10.1577/1548-8675(2000)020<0069:POBBAS>2.0.CO;2 en Taylor and Francis Predictability of Bristol Bay, Alaska, sockeye salmon returns one to four years in the future article