Forecasting catches of Pacific salmon in commercial fisheries of southeast Alaska

Abstract

Data collections since 1911 and statistical methods from time series analysis are employed to forecast catches of pink, chum, coho, and sockeye salmon in Southeast Alaska. Knowledge of the spatial and temporal domains favored by Pacific salmon originating in Southeast Alaska is summarized to provide a basis for estimating environmental variation experienced by each species. Catches in northern, southern, and all of Southeast Alaska are forecast with univariate ARIMA, transfer function-noise (TFN), and vector ARMA models. Univariate models for catch in numbers and catch in weight yielded similar results for each species. Air and sea surface temperatures, freshwater discharge, and coastal upwelling enter TFN models for several species and areas. Environmental variables allow TFN models to explain a small amount of variation in the catches (average of 19%) above that explained by univariate models. Forecasts for most, but not all, species and areas are improved (average of 16%) by including environmental data in TFN models. Stock-recruit models with a parameter for density dependent mortality provide the best forecasts of pink salmon catch and are recommended for future forecasts. Winter air and sea surface temperatures enter stock-recruit models for pink salmon, and forecasts of catch and recruitment in northern and southern Southeast Alaska tend to oppose each other and cancel (1981-1985), which suggests that the salmon are caught in areas other than where they originated. Mean absolute percentage error (MAPE) for forecasts of pink salmon catch from stock-recruit models in Southeast Alaska, based on data for 1981-1990, is estimated at 49%, with first, second, and third quartiles of 10%, 23%, and 83%, respectively. Catches of Pacific salmon in Southeast Alaska are significantly correlated and are forecast jointly with good accuracy by vector ARMA models, except when effects believed to result from density dependent mortality are present in the data. Correlations indicate that coho salmon smolts might prey on young pink salmon. Also, recruitment of pink salmon in Southeast Alaska and British Columbia is correlated; regional environmental influences might thus affect catches in both areas. In Southeast Alaska, MAPE for forecasting coho and sockeye salmon catch with time series analysis is about 20%, and about 30% for chum salmon.