Classification of Severe Storm Cells Using Support Vector Machines

Abstract

Meteorological volumetric radar data are used to detect thunderstorms, which are the cause of most severe summer weather. To characterize the thunderstorms, a set of products (derived from reflectivity values and certain heuristics) might be used. Using these derived products, we developed a pattern recognition system based on the support vector machine (SVM) approach to determine its applicability to the classification of a severe storm cell data set from Environment Canada. The SVM classifier is compared with two radial basis function (RBF) network classifiers. In one of them, the centers are found using fuzzy clustering. In the other, the centers are found using the orthogonal least squares approach. The criterion for comparison is the classification accuracy over a testing set. The results show that the SVM approach is the best of these approaches, in terms of accuracy, for the storm cell classification problem.