A neural network model for forecasting fish stock recruitment

Abstract

A neural network model is developed to forecast the recruiting biomass of fish. The west coast of Vancouver Island, British Columbia, Pacific herring (Clupea pallasi) stock is selected as an example application based on data compiled from long-term ecosystem research and stock assessment programs. A fuzzy logic decision procedure was used to evaluate all possible neural networks. The output from the two "optimal" networks was compared with the output from a multiple regression analysis and a standard Ricker climate stock-recruitment model. The performance of the neural network models in reproducing a 41-year time series was far superior (R2 between the fitted and observed recruitment is about 60-70%) to the multiple linear regression model (R2 = 0.29) and the Ricker climate stock-recruitment model (R2 = 42%). This pilot study demonstrates how artificial neural networks can be used to improve the accuracy of fishery stock forecasts and hence the management of the fishery resources by making the actual harvest rate (catch/stock biomass) closer to the target harvest rate (desired catch/stock biomass).