A Multi-Class Classification Scheme for Severe Weather Outbreaks

Discrimination of severe weather outbreak type has been primarily limited to considering outbreaks likely to produce tornadoes or non-tornadic severe weather. Previous work has demonstrated skill in this discrimination using composite analysis of major outbreaks of each type. A limitation to this work is the lack of consideration of the types of non-tornadic severe weather; namely high wind and large hail events. As an exploratory study, a support-vector machine (SVM) multi-class classification scheme was developed to discriminate between tornado outbreaks, outbreaks of major wind events, and outbreaks of major hail events. Composites of 20 of each type of outbreak were formulated using base-state meteorological variables (e.g. temperature, pressure, humidity, and u and v wind components) from the NCEP/NCAR reanalysis dataset and kernel principal component analysis and were used as training data for the SVM algorithm. Multiple experiments were conducted to identify the optimal parameters for the SVM classification. A withheld set of 10 events of each outbreak type were used to test the scheme. As an additional test, the composites were simulating using the Weather and Research Forecasting model and subjectively assessed for the outbreak type suggested by the model. Future work will consider additional events and an objective discrimination algorithm based on WRF simulations of these events.