To address this problem, we have applied a simple inflation adjustment by computing departures from a linear regression to the annual number of tornadoes in the Storm Prediction Center from 1954-2003. Reports have increased by 13 per year. Using the expected values from the linear regression as a baseline, we have analyzed a number of aspects of the record. First, big tornado days, those days that have many tornadoes in the US, can be identified as exceeding some threshold of the annual expected number, say 7% of the annual total. Sensitivity to the choice of the threshold will be discussed.
Second, we have produced inflation-adjusted annual progressions of tornado counts for each year, thus allowing us to compare years throughout the record, in terms of being above or below the expected value at any time during the year. This allows us to define parts of years that are above or below the expected value on a level basis.
Finally, the counts of tornadoes per day can be used to develop a statistical model of tornado seasons. The model consists of the probability of at least one tornado occurring on a particular day, and a number of tornadoes on a day, given that at least one occurs. Running the model for tens of thousands of years allows us to estimate parameters of the distribution of expected numbers of tornadoes for any segment of the year. For the median of the distribution, the statistical model is never more than 20 tornadoes away from the observed (1954-2003) median. Percentiles out to the 10th and 90th are modelled well, but the extremes are not, particularly the low tornado count years, thus providing limits on what can be said about the most extreme years.