Applications of Data Mining to Supercell Tornadogenesis

The ability to anticipate tornado development with greater accuracy has the potential to decrease adverse impacts to life, business, and property. For example, Simmons and Sutter (2009) found that the false alarm ratio for tornado warnings nationwide decreased from 1987-2004, coinciding with a 4-11% reduction in tornado-related fatalities. In order to improve forecast reliability, the National Severe Storms Laboratory is developing a warn-on-forecast (WoF) strategy, which is expected to increase the lead time of tornado warnings to 30 minutes (Stensrud et al 2010). WoF involves the use of algorithms to issue warnings based on automatic detection of favorable conditions for tornadic development in short-range numerical weather prediction models. The success of WoF hinges on the quality of the algorithms being applied as well as the accuracy of the models in use. Data mining techniques can be used to discover relationships between supercell features in hopes of improving understanding of the processes involved and clarifying which conditions within supercells are important for tornadogenesis. This approach also has the advantage of allowing researchers to efficiently review data from a large sample set of simulated supercells rather than relying entirely upon a relatively limited set of observed events. A data mining algorithm will be applied to a preliminary sample set of simulations to discover relationships between vortex strength and longevity and sounding character. Each simulation is created using the CM1 atmospheric model, initialized with a unique input sounding based on Weisman and Klemp (1980). This sample set will focus on wind variations in the hodograph and will hold surface mixing ratio constant in the soundings.