Biophysical regions identification using an artificial neuronal network: A case study in the South Western Atlantic

A classification method based on an artificial neuronal network is used to identify biophysical regions in the South Western Atlantic (SWA). The method comprises a probabilistic version of the Kohonen’s self-organizing map and a Hierarchical Ascending Clustering algorithm. It objectively defines the optimal number of classes and the class boundaries. The method is applied to ocean surface data provided by satellite: chlorophyll-a, sea surface temperature and sea surface temperature gradient, first to means and then, in an attempt to examine seasonal variations, to monthly climatologies. Both results reflect the presence of the major circulation patterns and frontal positions in the SWA. The provinces retrieved using mean fields are compared to previous results and show a more accurate description of the SWA. The classification obtained with monthly climatologies offers the flexibility to include the time dimension; the boundaries of biophysical regions established are not fixed, but vary in time. Perspectives and limitations of the methodology are discussed.