The 2012 tornado season in the United States got off to a quick start with well-above average numbers in January, February, and March. Later, over 80 tornadoes occurred on 14 April. Since then, the number of tornadoes in the US has been unusually low. In order to understand how low, we need to look at the long-term history of tornado occurrence. The most reliable portion of the tornado data begins in 1954 but, even after that, we have to be careful in how we interpret it. Since the mid-1950s, the number of tornadoes reported has increased by an average of 14 per year. The increase has been almost entirely in the weakest tornadoes (F0) and is highly likely that the causes are non-meteorological. We can think of this increase in the same way we think of inflation in economics and estimate its impact by adjusting historical tornado counts to account for it. This process, and how it can be applied to part of the year, is discussed here.

That inflation-adjustment process allows us to look at historical data, but a problem still remains of how to look at recent reports. Preliminary, eyewitness reports of tornadoes are collected by local National Weather Service Forecast Offices and the offices then evaluate those reports and produce a list of “final” tornado reports. This process of evaluation takes a few months to complete, so it can be challenging to answer the question “how many tornadoes occurred” shortly after an event. Over the last several years, a simple relationship between the preliminary and final reports has been observed with the number of final reports being approximately 85% of the preliminary reports. As a result, when looking at the preliminary reports in recent months, we can a pretty good estimate of the final reports simply by multiplying the preliminary reports by 0.85.

Let’s look at how many tornadoes we would expect based on the inflation-adjusted tornado count and compare this year’s tornadoes to that long-term expectation. To emphasize the small number of tornadoes since the middle of April, we’ll start on 15 April and add up the number of tornadoes each year through the end of July. In the accompanying chart, we see the distribution of the accumulated number of inflation-adjusted tornadoes as we got from 15 April-31 July. The distribution is based on the period from 1954-2011. The maximum and minimum of any of those years are shown in blue (note that the year associated with the maximum and minimum can change from day to day along the way). The heavy black line is the median of the distribution, the gray lines are the 25th and 75th percentiles (half the years will be between them), and the dashed lines are the 10th and 90th percentiles (4 out of 5 years will be between them). For comparison, the estimated number of final tornado reports from 2012 are shown in red.

Through the end of May, the tornado count for the period from 2012 goes along at approximately the 10th percentile of the long-term distribution but, after that, falls well below the previous low. To put this into perspective, the estimated number of final reports from June for 2012 is approximately 100. The previous inflation-adjusted low for any previous June is 94 in 1988. (Remember that the blue line represents the fewest number of tornadoes from any of the 58 years from 1954-2011.) The median number of June tornadoes in 1954-2011 was approximately 270.

July was even more remarkable than June. Only 24 preliminary reports were received, leading to an expected number of final reports of a little over 20. The lowest number of inflation-adjusted tornado reports from 1954-2011 is 73 (1960). Even without inflation adjustment, the fewest number of tornadoes in any July in that time period is 42 (1960), emphasizing the extraordinary nature of this July. The median number of July tornado reports is about 150.

When we look at the whole period from 15 April-31 July, the median tornado count in the record is 850, compared to 2012, with a little under 300. The 850 represents almost 2/3 of the usual annual total of about 1300. One way of thinking about the late spring and early summer tornado season is that the atmosphere missed more than 40% of a typical year’s tornadoes in 3 1/2 months. Compared to 2003, the comparable period in 2012 had more than 900 fewer tornadoes. 2011 had the second highest number of tornadoes in this part of the year, so in the last two years, the US has experienced the extreme high end of the distribution of the number of tornadoes and the extreme low end of the distribution.

Few tornadoes have occurred to date in 2010, especially when compared to very active recent years such as 2008. The adjusted annual tornado trend shown below indicates that 2010, with 79 tornadoes (adjusted) through April 15, is rapidly approaching the minimum number of tornadoes expected through this time of year (about 75). While 2010 has not quite reached that adjusted minimum threshold yet, it looks like it may be only a matter of another day or two with few or no tornadoes.

This graph shows how the number of tornadoes so far in 2010 in the United States is very near the adjusted minimum value expected. Additional information about the chart and the method used to adjust for tornado report inflation can be found here: http://www.spc.noaa.gov/wcm/adj.html

Does the scarcity of tornadoes so far in 2010 say anything about what we can expect for tornado activity in May? Let’s review the past 30 years of tornado counts and see. Below is a chart depicting the departure from the decade’s average number of tornadoes for two periods in each year (1980-2009). The first period is the departure from the decade’s average number of tornadoes, January through April (gold). The second period is the departure from the decade’s average number of May tornadoes (light blue). In the early years of this chart (1980 through 1983), this chart would suggest that, yes, we can say that an above (below) average start to the year through April is indicative of a  above (below) normal May. However, that forecast methodology falls apart in 1984, 1985, and 1986 when, in each of those years, an above normal January-April period was followed by a below normal May. In 1987 and 1988, there is a strong correspondence between a low number of tornadoes through April, and a slow (below normal) May. Then in 1989, an inverse relationship occurs when a below normal start to the year is followed by an above normal May.

With the exception 1995, there is a remarkable 13 year stretch from 1990 to 2002 when the start of the year could possibly be used as an indicator of the May to come. During this period there were nine years when normal to below normal tornado numbers in the January through April period were followed by normal to below normal tornado activity in May. There were only three years during this period (1991, 1998, and 1999) when an above normal start to the year continued into May. If you were looking at only those years you might conclude that in 2010 we will see below normal tornado activity into May.

Trouble starts again in 2003 and 2004 when low tornado counts through the early months of the year are followed by remarkably active Mays. In fact, May 2003 and May 2004 are the most anomalously active Mays of the 30-year period and both were preceded by some of the most anomalously inactive January through April periods! Some semblance of correspondence returns in 2005 (both periods below normal) before a reversal shows up in 2006 and 2007 with above normal January through April periods being followed by below normal Mays. One is entitled to be quite flummoxed by this point in the analysis but the conclusion should be that, at least in the most recent years, making a prediction about tornado activity in May, based on the character of the season to date through April, would be a tricky bet.

Last year saw an active start through April only to be followed by a somewhat quiescent May. That situation was addressed in this blog post.

Despite the fact that the correlation coefficient (r) for this entire time series is an abysmal 0.15, there is still hope (albeit diminishing) that we can say something about the predictability of the future based on what has happened in the past. So, one last chart for all of those folks hoping to capture a tornado during the upcoming VORTEX2 field program…

The correlation coefficient on this time series for tornadoes only occurring in Texas, Oklahoma, Kansas and Nebraska is barely improved over the previous comparison for the entire United States. It’s quite possible that charts used to evaluate the stock market would exhibit a similar lack of correlation. And, perhaps it’s fitting to think of this little experiment in the same light by using an oft-quoted line from a typical mutual fund prospectus: “Past performance is no guarantee of future results.”

Stay tuned. We’ll know the answer soon enough.

*******

There were no tornadoes reported in the United States in February 2010.  Assuming that no late reports are received, it will be the first time in the National Weather Service’s database that starts in 1950 that there has been a February without a tornado.  If we include Tom Grazulis’s database of F2 and stronger tornadoes, the last time it’s possible there wasn’t a February tornado was 1947.  The last tornado reported in the US was on 24 January, in north-central Tennessee.  The last calendar month without a tornado was January 2003.

What does this tell us about the rest of the 2010 tornado season?  Somewhere between a little and nothing at all.  Most years that have started out with few tornadoes have ended up average or below.  However, there have been big exceptions.  Most notably, in 2003, we started out with no tornadoes in the first 45 days of the year.  Even as late as 29 April, it was the slowest start in the database (after adjusting for report inflation, as discussed here.) By the 11th of May, however, 2003 was well above normal following a remarkably active week. So, even though it’s been a slow start to the season, people still need to be aware of the threats that may happen later on.

What does it tell us about long-term trends? Again, essentially nothing. The large-scale atmospheric pattern that persisted over the US for the month of February was unfavorable for tornadoes. There’s nothing in the scientific literature that provides information on any changes to expect with tornadoes in the future, so the no-tornado February can’t be interpreted in that light.