Attack of the 50-foot Tornado

I happened on a blog entry about the recent tornado victims which mentioned, in passing, that global warming has been suggested as a cause of an increase in the number of tornadoes. This surprised me, because I’ve never heard of such a connection before. Unfortunately, the blog entry provided no reference and didn’t seem to be written by the scientifically savvy, so there was no real clue about the source of the claim, nor was it credible based only on the blog entry. Nonetheless, it raised the question whether there’s been a change in tornado frequency during the modern global warming era. So I tracked down some statistics on tornadoes in the U.S., and did some exploring.

The most basic statistic is the number of tornadoes per year:

There’s no doubt that the number of recorded tornadoes has increased sharply during the time that records have been maintained. But in large part this is a selection effect: as population has increased and technology improved, our ability to record the presence of tornadoes has likewise improved. Hence there may not actually be more tornadoes, we may simply have noticed more of them. In fact, considerable research has indicated that tornado detection and reporting have improved over the years, so even though the increase in the number of recorded tornadoes is strong, there’s doubt about whether or not there’s been a real increase in tornado occurence in the U.S. It’s also worth noting that since about 1990 (when technology reached an advanced stage and U.S. population was significantly higher than 1950), there’s no indication of a trend in the total number of recorded tornadoes.

There is, however, an interesting event in the annual count of U.S. tornadoes. The year 2004 saw a record number, some 1,819 tornadoes. This count is 3 standard deviations above the trend line, while no other single year has a count exceeding two standard deviations above the trend line. Therefore there’s strong evidence that 2004 was an anomalous year, with dramatically more tornadoes — even accounting for the trend (possibly due to selection effects) — than any other in the observational record since 1950.

In part the large number of tornadoes in 2004 is due to the very large number during May of that year:

For the month of May, both 2003 and 2004 are anomalous years. 2004 is nearly 3 standard deviations above the trend line, and 2003 more than 3 standard deviations above the trend line, making both of them exceptional. This is further evidence that the most recent decade has actually seen more tornadoes than previous decades. Still, with only a little less than 6 decades of reliable data, it’s hard to be sure.

Tornadoes can occur during all months of the year, but May turns out to be the month during which they’re most frequent:

The warming during spring increases the chance that warm and cold air can meet, leading to the kind of storms that spawn large numbers of tornadoes.

It occured to me that I might gain insight by examining the number of large tornadoes over time; larger tornadoes are far less likely to be missed (and therefore undercounted) than small ones. One measure of a tornado’s size is its strength on the Fujita scale. But research reveals that Fujita scale estimates are subjective, have changed over time, and are possibly related to a tornado’s destruction rather than simply its size or power. Hence researchers have expressed doubt over the usefulness of trend estimates in the numbers of tornadoes falling into the various Fujita scale classes. Fortunately, the data I found also indicates a tornado’s width, i.e., the width of the tornado’s base. This is an imperfect but less subjective estimate, and is not related to the destruction caused by a tornado, so it might make a suitable separator between larger and smaller tornadoes. The largest tornado ever recorded occured in Nebraska in 2004, with a base estimated at 4400 yards wide (a two-and-a-half mile wide tornado!). So, I counted the number of tornadoes each year with a base 50 feet or larger, 100 feet, 200 feet, 300 feet, 600 feet, and 1000 feet or larger. From base widths larger than 0 feet (all tornadoes) to 600 feet, here are the annual counts:

The first thing I notice is that for 1998, the number of 50-foot wide tornadoes is the same as the total number. This is, frankly, hard to believe; it’s much more likely that width statistics for small tornadoes were poorly recorded that year, and that tornadoes with no width estimate were arbitrarily assigned a width of 50 feet. In fact it turns out that no tornado in 1998 has a recorded width less than 30 yards, or 90 feet, although two tornadoes that year in Minnesota are recorded as over a mile wide. It seems overwhelmingly likely that the 30-yard width is not an actual occurence, but a default value assigned to tornadoes of unknown width. So, it won’t do to analyze the number of 50-foot tornadoes, since that number is clearly unreliable for 1998 and possibly for other years as well.

But the largest tornadoes are more likely to be correctly represented in the records. So I looked at the number of tornadoes at least 600 feet wide, and at least 1000 feet wide. Tornadoes with a base which is two or three football fields across (or larger) aren’t likely to be missed, especially since such large ones tend to last longer and travel futher, increasing the likelihood of detection. Here are the annual counts for 600-foot-wide-or-larger and 1000-foot-wide-or-larger tornadoes:

There’s a clear change in the number of recorded tornadoes this large over time. The number of 600-foot tornadoes increased sharply in 1995, and the number of 1000-foot tornadoes in 1996. The 8 years with the most 600-foot tornadoes have all occured since 1995, and 8 of the 9 years with the most 1000-foot tornadoes since 1996. The change isn’t just a visual impression, it’s strongly statistically significant:

Also interesting is the fact that neither 600-foot nor 1000-foot tornadoes show any trend prior to 1995/1996. This indicates that their detection is not strongly affected by observational limitations, and lends credence to the idea that the strong trend in the total count of all tornadoes prior to 1990 actually is a selection effect rather than a real trend. I find very strong evidence that the number of large tornadoes increased, rather dramatically in fact, around 1995/1996. The size of the increase and the lack of a prior trend support the idea that it’s a real rather than observational change.

Why the change in tornado counts? One suggestion I found in the literature is that global warming has increased water temperatures, leading to more humidity, especially from the Gulf of Mexico, which provides fuel for the storms which spawn tornadoes. But I haven’t seen solid evidence of this, only speculation. It certainly makes sense that warmer air temperatures and greater humidity could lead to more, and larger, tornadoes. But until I see more evidence, I’ll continue to consider this speculation. In the meantime, one hopes that the increased number of violent tornadoes in the U.S. is only a temporary phenomenon.

UPDATE

Harold Brooks comments that the definition of, and measuring procedures for, width data which is the basis for some of this analysis, has changed, particularly in 1994 (just before the noted changes in the numbers of “large” tornadoes in the data). Hence the shifts noted here are likely a result of changes in data procedures, rather than actual changes in the frequency of large tornadoes.

The fact that tornado data are inconsistent throughout the last half-century is bad news. But on that basis, there doesn’t seem to be evidence of a notable change in tornado properties in the U.S., and that’s good news!