An artificial neural network (ANN) approach for predicting stock abundance of the Barents Sea capelin (Mallotus villosus) is presented. The method is based on training an ANN with a genetic algorithm using input data of ecological importance to the capelin stock. Stock abundance for the coming year is estimated using the trained ANN with the current set of ecological data. Time series of data on cod, herring and capelin abundance, and average weight of capelin for the period 1974-1999 are used to train the ANN. The model was tested for its ability to predict capelin abundance in single years, using the remaining time series for training. The resulting predictions correspond well to observations, and the ANN method gives higher predictive ability than a simple fisheries assessment model. The strength of the ANN method is its ability to predict changes in natural mortality and growth. However, the network is unable to predict the population crashes that took place in the mid 1980s and early 1990s without prior training to similar scenarios. This illustrates the importance of having the full range of ecological interactions represented in the training set. Since the method is simple and relies on data collected by most fisheries institutes, it could easily be applied in calculating stock prognoses.
All data in the Integrated Marine Information System (IMIS) is subject to the VLIZ privacy policy
