On the relationship between preliminary and final tornado counts in the SPC database

Under the current protocol for producing the Storm Data listing of tornadoes, it can be up to three months between a tornado and the final listing of the event in the official database of tornadoes in the US. As a result, it is tempting to use the preliminary listing of tornadoes that is available almost immediately after the event as a proxy for the final listing. Changes from the preliminary to final listings take place for a number of reasons. In particular, the number of tornadoes can increase because of late reports being received and/or listed in a Local Storm Report (LSR) at a local forecast office. The number of tornadoes can go down from the preliminary to the final count mostly because of multiple reports of the same tornado. This latter situation is particularly likely for long-track tornadoes and extensive tornado events.

We have examined the preliminary and final tornado reports on a daily and monthly basis beginning in 1998 and, at the time of writing, ending in February 2008. Through February 2006, the first class of changes (increasing numbers from preliminary to final counts) led to a 9.2% increase in the final listing of tornadoes. For the 62 months with more 40 preliminary reports from January 1998-February 2006, the lowest ratio of final to preliminary reports was 0.852 (November 2002). Only 4 months were below 0.9 and 17 of the 62 (27%) showed a decrease from preliminary to final counts.

The situation changed dramatically starting in March 2006. From that month through February 2008, 20 individual months had more than 40 preliminary reports. The lowest ratio of final to preliminary was 0.598 (February 2007). Twelve months had values lower than the 1999-February 2006 minimum, 16 were below 0.9 and 17 below 1. Over the March 2006-February 2008 period, the number of tornadoes in the final log was 19.4% lower than the preliminary log.

We have fit regressions to the daily preliminary and final numbers. The slope of the regression fit for the 1998-February 2006 period is 0.923, while for the March 2006-February 2008 it was 0.701. (The y-intercept for both is ~0.5.)

The importance of this change can be seen in media coverage of the 2008 tornado season. While the final tornado count for January-February (232) was nearly equal to the record of 233 from 1999, it was much lower than the preliminary of 368. The preliminary count gave the impression that the 2008 season was vastly higher than anything previously seen, while the final count gave the impression that the year was historically very high, but was not astronomically high. Given that the final numbers were not available until the middle of May 2008, however, it was a challenge to understand and communicate the magnitude of the tornado season to date to the public.