Using Kernel Density Estimation to Identify, Rank, and Classify Severe Weather Outbreak Events

Abstract

A method for ranking severe weather outbreaks of any type using a linear-weighted multivariate scheme has been introduced recently. The results of using this ranking method indicated that the scheme was capable of identifying the most significant severe weather outbreaks. However, the inclusion of days in which numerous reports were widely dispersed across a large region, or in which multiple clusters of reports that were geographically widely separated, was problematic. Though the studies included a variable (the so-called middle-50% parameter) that was effective in identifying these cases, a new way was needed to account for these days in a manner that agrees with subjective perceptions of these events. A candidate scheme introduced here uses nonparametric kernel density estimation to identify clusters of severe weather reports associated with a single severe weather event. Clusters with relatively few reports or sparse coverage within the region associated with the event then can be excluded quite easily. This technique also allows for multiple, regionally-separated clusters of severe reports to be considered in one day. After identifying clusters of severe weather events from 1960-2008, the cases are ranked and classified in a way similar to past research, using multivariate linear-weighting and cluster analysis, respectively. Results suggest that the most significant severe weather outbreaks again are identified appropriately, and the cases could be classified as major tornado, hail-dominant, wind-dominant, and minor mixed-mode events.