This paper examines the use of a neural network to analyse and predict the winter catch, in the Joban-Boso Seas off the Pacific coast of central Japan, of young Japanese sardine (Sardinops melanostictus) representing the index of recruits in the sardine stock. The supervised learning paradigm, a three-layer network and a back-propagation algorithm were employed in constructing the neural net. A number of biological, hydrographic and climatic factors constituted an input vector, the output being the catch of young sardine. The association of sardine abundance with environmental factors was quantified in the form of the trained neural network, which indicated important associations with the Southern Oscillation Index, with patterns of the Kuroshio and the... |