Toward Understanding Tornado Formation Through Spatiotemporal Data Mining

Tornadoes, which are one of the most feared natural phenomena, present a significant challenge to forecasters who strive to provide adequate warnings of the imminent danger. Forecasters recognize the general environmental conditions within which a tornadic thunderstorm, called a supercell thunderstorm, will form. They also recognize a supercell thunderstorm with its rotating updraft, or mesocyclone, when it appears on radar. However, only a minority of supercell storms produce tornadoes. There are no obvious clues within any of the routinely observed data to indicate which supercell storms are going to produce tornadoes and which ones are not. So to be on the safe side, forecasters issue a tornado warning whenever they detect on radar a supercell thunderstorm with a strengthening mesocyclone, which is the parent circulation within which tornadoes form. This approach results in the warning being issued an average 10 to 15 minutes before the appearance of a tornado, but a tornado appears only 20 to 30% of the time that a warning is issued, which results in a large percentage of false alarms (e.g., Simmons and Sutter, 2011). Surveys conducted by the National Weather Service following devastating U.S. tornadoes reveal that these false alarms are one of the factors contributing to desensitization on the part of the public concerning the need to adhere to warnings (e.g., NWS, 2009, 2011). Having heard many warnings when a tornado did not form, members of the public tend to ignore the warning and only take shelter if they see