The use of neural networks for forecasting sockeye salmon Bristol Bay returns is proposed. In Phase I, this project will collect and characterize available data sources, train a neural network to forecast the sockeye salmon run, and compare the performance of the resulting system against existing techniques. Because neural networks are able to automatically learn nonlinear mappings from diverse data types, it is expected that neural networks will produce more accurate forecasts. In Phase II this project will expand its scope to include fish population modeling.
