Using an artificial neural network to patternize long-term fisheries data from South Korea

Abstract

This research used a Self-Organizing Map (SOM) to patternize long-term fisheries data of 30 species from South Korea. A spectrum of catch amounts from fish assemblage data over 48 years was successfully clustered and visualized on a two-dimensional map. Temporal variation in fish data was explored using a SOM. Five yearly clusters identified different time periods from 1954–1961, 1962–1973, 1974–1982, 1983–1996, and 1997–2001. These different periods reflected environmental and economic forcings on fish catch in Korea. Specific fish species were dominantly related with different time periods. Association of collected fish species was additionally patternized on a SOM. Characteristics of catch data, such as overall abundance and increasing pattern, were identified using a SOM. This artificial neural network demonstrated a powerful capacity to deal with the large amounts of fish catch data with various external forcings and heterogeneity of sampling over a long time period.