ARTIFICIAL NEURAL NETWORKS TO FORECAST BIOMASS OF PACIFIC SARDINE ANDITS ENVIRONMENT

We tested the forecasting performance of artificial neural networks (A NNs) using several time series of environmental and biotic data pertai ning to the California Current (CC) neritic ecosystem. ANNs performed well predicting CC monthly 10-m depth temperature up to nine years in advance, using temperature recorded at Scripps Institution of Oceanogr aphy pier. Annual spawning biomass of Pacific sardine (Sardinops sagax caeruleus) was forecasted reasonably well one year in advance using t ime series of water temperature, wind speed cubed, egg and larval abun dance, commercial catch, and spawning biomass of northern anchovy (Eng raulis mordax) and Pacific sardine as predictors. We discuss our resul ts and focus on the philosophy and potential problems faced during ANN modelling.