453
Ranking severe weather outbreaks using a multivariate index
Chad M. Shafer, University of Oklahoma, Norman, OK ; and C. A. Doswell III
Recent studies have suggested that mesoscale models are capable of distinguishing major tornado outbreaks from primarily nontornadic outbreaks several days in advance of the events. However, a large portion of significant severe weather outbreaks cannot be classified as either tornado outbreaks or nontornadic outbreaks. To determine the relative severity of any severe weather outbreak, a method will be presented using multiple variables to develop an index for ranking severe weather outbreaks from the period 1960 to 2006. Storm reports during this period contain several nonmeteorological artifacts, requiring detrending techniques to reduce the effects secular trends introduce. In general, the detrending techniques implemented in this study appear to reduce the secular trends in the data sufficiently.
The resulting rankings show three basic types of cases: major severe weather outbreaks, relatively minor severe weather outbreaks, and days with severe reports featuring a large degree of geographical scatter. The rankings for the major severe weather outbreaks and for the high-scatter outbreak days are relatively resistant to the variables used and their weights in developing several multivariate indices. However, the rankings for the relatively minor outbreaks are more sensitive to these changes. These sensitivities exist despite the similar index values calculated by varying the weights of the meteorological variables. Because of these sensitivities, the multivariate indices are converted to a vector with four components (tornado parameters, hail parameters, wind parameters, and variables quantifying geographic scatter). A k-means cluster analysis is conducted on these vectors. The outbreak days cluster into five groups: major tornado outbreaks, outbreaks with primarily hail reports, outbreaks with primarily wind reports, outbreaks with no preference to hail or wind, and outbreak days with large geographical scatter. The prediction of the four-dimensional vector representation of the multivariate indices may be a useful means of predicting the type and relative severity of future outbreaks of severe weather.
Poster Session , Applications of Artificial Intelligence Methods to Problems in Environmental Science
Wednesday, 20 January 2010, 2:30 PM-4:00 PM, Exhibit Hall B2
Next paper