There are good reasons to believe that global warming leads to more storminess, but the exact nature of that transition is unclear and hard to measure. Part of the reason for this difficulty is that a given type of storm may become more likely under certain conditions caused by climate change, while a different kind of storm may become less likely, with the “storminess” overall increasing but doing so indifferent ways across time. Also, the most severe, and thus possibly the most important, weather events are infrequent so it is difficult to see changes over time with any statistical confidence. I address many of these issues here and here.

Looking at the raw data, it is clear that there are “more tornadoes” over time in the US. Have a look at this graph:

At first glance, his graph makes it look like there are a lot more tornadoes, but there is a strong effect of observer error; earlier tornadoes were simply missed much of the time, so the big increase you see here, while it may reflect an underlying increase in number of tornadoes, is not reliable and cant’ be taken as evidence. However the later years shown here, from 1950-something to the 1990s, seems to show an increase that could be taken as meaningfull

However, when people speak of tornadoes they often show this graph as evidence that there are not more of them over time:

Looking only at this graph it looks like the number of tornadoes per year in the US is pretty variable but not increasing, as one would expect if global warming was causing more of them.

There is a problem with this graph, however. Actually, a couple of problems (other than those pointed out here). The main problem is that the most frequent tornadoes are left off this graph. If we look at F0 grade tornadoes, not included here, we see that they have actually increased in frequency over time. If we include ALL tornadoes, and not just the kinds that don’t seem to increase in frequency over time, we get this graph:

Compare the scales of the last two graphs. It turns out that the number of tornadoes at the smaller end of the scale goes up quite a bit. It might be hard to see. The upper graph goes up to 900, the lower graph goes up to 1900. So, if we add all the data instead of just select data, we get many hundreds more tornadoes per year.

The proportion of tornadoes that are F0 increases over time as shown here:

… and the overall distribution of tornadoes by strength changes over time as shown in this very cool graph:

As I point out here, one of the contributing factors to variation over time in tornado frequency is the fact that we have somewhat arbitrary boundaries in which we measure them. For instance, the US-Canada border provides an arbitrary line across our data set. By not counting all North American tornadoes the same way, we may be adding unnecessary variability to the data. To demonstrate this, have a look at this graph showing tornado frequency per year in France and Germany, two countries that are right next to each other:

This shows a few things. For one thing, they don’t have too many tornadoes in that part of the world. For another thing, there is an increase in overall frequency over time, and this is not because of lack of reporting. The reporting problem in the US is partly because the western and central states were relatively empty in the old days, and also more technology was available for spotting tornadoes later. But the European and US data have the same shape over a similar time span, but France and Germany do not have the missing observations owing to vast unoccupied (sort of) territories.

But the main thing I want to demonstrate with this graph is the fact that dividing a largish area of land up into arbitrary units can cause your data go go all flooey. Increased variability in data owing to partitioning is a well known phenomenon and this is what it looks like.

Another part of the problem is that the largest storms, which may be the most important ones, have a great deal of variation in their occurrence. Compare any of the graphs above of all tornadoes or all excluding the F0 tornadoes of this graph of just the largest storms:

Not only is there a lot of variation in numbers of tornadoes at the larger end of the scale, but I suspect there is a lot of variability among the tornadoes in each class in terms of overall energy represented. An F4 tornado that lasts five minutes compared to an F4 tornado that lasts 20 minutes are hugely different, but this is not reflected in this sort of data.

Here is a graph showing the amount of storm damagein adjusted dollars over time in the US (pink) with average temperature (blue). Clearly, the total amount of damage goes up, and probably for a number of reasons including there being more stuff to damage, but also, likely overall increases in storminess including hurricanes, tornadoes, severe thunderstorms, etc.

Here is another graph that shows something similar:

There are many who do not want to link increases in severe weather to global warming. They are probably wrong. Global warming seems to increase severe weather overall. The best way to deny this is to cherry pick the data by ignoring variability across space, leaving out entire categories of storms, or focusing on just some kinds of storms. I suspect the size and severity of tornadoes at the larger end is increasing now, but did not start increasing until recently; time will tell if this is right. But overall tornadoes are so variable across time and space that they are not a reliable canary, as it were. But overall storminess seems to be on the increase, in accordance with expectations from the basis physics of climate, under warming conditions.

Photo Credit: Vvillamon via Compfight cc