258 Methods to rank and classify severe weather outbreaks

Monday, 24 January 2011
Washington State Convention Center
Chad M. Shafer, University of Oklahoma, Norman, OK ; and C. A. Doswell III
Manuscript (1.8 MB)

A multivariate linear-weighted index is proposed to rank and classify severe weather outbreaks. Initially, various types of archived severe weather reports from the period 1960-2006 were collected. Numerous nonmeteorological artifacts were found in the data, and efforts to reduce these were required. The top 30 days of each year, based on the total number of severe hail, wind, and tornado reports, were considered. These days then were ranked based on the meteorological and societal significance of the events by varying the weights for each of the report types used in the index. Findings suggested that the rankings generally agreed with subjective notions of these events. Three basic categories of outbreak severity were identified: major outbreaks, intermediate outbreaks, and marginal events. The rankings of the major and marginal events were relatively resistant to modifications of the weights, whereas the intermediate outbreak rankings were not so resistant. However, using cluster analysis techniques, classifying outbreak days based on the nature of reports was robust to changes in the weights. Five types of outbreaks were identified: major tornado, hail-dominant, wind-dominant, and mixed-mode days as well as days with substantial geographic scatter to the reports or multiple clusters of reports that were widely separated.

As geographic scatter initially was accounted for by using a latitude-longitude box for the median 50% of the latitudes and longitudes of the reports, days with large geographic scatter were ranked lower than days with a large number of reports over a relatively small, single region. However, days with multiple clusters of reports were counted as one event and were penalized by the original index technique. This is not desirable if distinct synoptic-scale systems were responsible for these separate clusters. A modification to the initial technique was developed using kernel density estimation, which allows for events to be considered by clusters of reports on a given day. The new method excludes events with large geographic scatter effectively, and includes multiple clusters of reports on a given day as separate events, as desired.

A final limitation was the arbitrary designation of events as 24-h periods. A technique is proposed to follow clusters of reports in short, adjoining time periods for consideration of outbreaks as single events with no time constraints. The technique allows for identification of multi-day outbreaks associated with individual synoptic-scale systems. This work will be used in future studies to study a large number of these events.

Supplementary URL: http://www.ejssm.org/ojs/index.php/ejssm/issue/view/19

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner