The use of support vector machines (SVMs) to distinguish various types of data has been introduced only recently in the field of meteorology. SVMs can be a valuable tool to determine the ability of forecast or observed severe weather parameters to distinguish among various types of severe weather. Implications of such studies can be profound, as these investigations may provide operational forecasters more efficient and more accurate methodologies to assess the types of severe weather that may occur in certain synoptic- and subsynoptic-scale environments.

National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis data and North American Regional Reanalysis (NARR) data are analyzed for 50 tornado outbreaks and 50 primarily nontornadic outbreaks occurring during the spring and fall seasons. These reanalysis datasets are chosen to investigate the ability of severe weather parameters to discriminate between these outbreak types at various spatial scales. A number of severe weather parameters, with an emphasis on parameters known to be associated with the formation of mesocyclones and tornadoes, will be computed for these reanalysis data at the approximate valid times of the outbreaks.

SVMs will be used to assess the ability of the tested severe weather parameters, both individually and in combination with other parameters, to discriminate between the two outbreak types. These assessments will focus on the magnitudes and spatial structures of the severe weather parameters centered at the location of the outbreak. Various statistics, including the accuracy, probability of detection, false alarm ratio, and skill scores, will be presented for the severe weather parameters analyzed. The capability of severe weather parameters to distinguish outbreak types at various spatial scales will be discussed.