Identification of eddies from sea surface temperature maps with neural networks

Abstract

This paper describes a new algorithm for automatic detection of Mediterranean water eddies from sea surface temperature (SST) maps of the Atlantic Ocean. The proposed approach looks at satellite images as textural patterns of water temperatures. Given a point of the map, information on the surrounding temperature gradient field is extracted through a binary mask and organized as a numerical vector of gradient angles. An artificial neural network (ANN) is trained to recognize those textural patterns of gradient angles that characterize eddy structures. Tested over a range of different binary extraction masks, the proposed identification system achieves accurate and robust learning results. Results are also characterized by very low rates of detection of false positives. The latter is particularly important because eddies occupy only a small portion of the ocean map area. Experimental comparison is carried out using Laws' method for texture analysis. The results show that the proposed gradient‐based algorithm allows similar performances with a reduced design effort. The complexity of the proposed approach also compares favourably to other methods for identification of mesoscalar phenomena published in the literature. Given the satisfactory accuracy of the results obtained, the gradient‐based approach may be preferable to current techniques as it is simpler and more easily reconfigurable.