In studying the variability of significant tornado climatology in the United States, the goal is to model the variability of significant tornado occurrence in the country. This study looked at the regional and the national tornado variability in the United States.
In order to model variability, "raw" climatology data was first analyzed. Five different percentiles, the 10th, 25th, median, 75th and 90th, can be analyzed and from this, the number days for each gridpoint where the 90th percentile is greater than the maximum median for the year were plotted on a graph. The greater number of days this occurs on during the year, the higher the variability of significant tornadoes. Since the climatology data might not be entirely right, the beta distribution was used to model the tornado data by analyzing the distribution's two parameters, particularly parameter p, in the distribution. The number of days when p is greater than one, representing low variability, in a year was then plotted on a national map. The Komolgorov-Smirnoff test was run to test our beta distribution's results against the climatology data. A more sophisticated model using the beta distribution was used in order to study the variability of these tornadoes and the average number of tornado days per century was plotted.
The results from this study indicate that the Monte Carlo model using the beta distribution for the input is a good model of the significant tornado climatology in the country. The climatology data with the Monte Carlo model produced a small variance. The Komolgorov-Smirnoff test indicated that the beta distribution had good and bad fits and that there was no pattern to which places did well and which places did poorly. One of the most important results is that there is low variability over the Plains states and areas to the east have high variability. Therefore, the Plains states have a definite tornado season and significant tornadoes are likely almost all year for places east of the Plains.
Paper available upon request.