Guest essay by Alberto Z. Comendador
Temperatures in the USA have generally increased since the 1950s. Of course, anything that mostly increases over time will have a positive correlation with everything else that also increases; in the USA this includes tornado counts. And of course most correlations are utterly meaningless and devoid of causation. For a hilarious take on the issue, see spurious correlations.
The problem , as NOAA itself says, is that there is a bias: just because we observe more tornadoes than before, doesn’t mean their number has actually increased.
‘With increased National Doppler radar coverage, increasing population, and greater attention to tornado reporting, there has been an increase in the number of tornado reports over the past several decades. This can create a misleading appearance of an increasing trend in tornado frequency.’
They illustrate this by excluding the weakest tornadoes (the F-0 category), which is the one most affected by this observation bias. If one does so, there doesn’t seem to be a long-term increase.
From the NOAA Storm Prediction Center:
And if you look only at strong tornadoes, F3 and up, they’ve actually declined.
Today I want to look at the issue from another angle: changes in temperature. I have seldom seen this kind of analysis in climate and it seems to me temperature provides a good opportunity, because the yearly changes are so large compared to the long-term trend. By contrast, any observation bias has to be small. The population of states, coverage of Doppler radar, etc. are virtually identical from one year to the next.
NOAAs’ temperature data page shows this:
Temperature swings from one year to the next are sometimes in excess of 1ºC. In fact, that’s about as much as total warming since 1979.
Now, there are some aspects of climate for which yearly temperature swings are not very relevant; for instance, glacial melting depends mostly on the absolute temperature, not the change from the year before. Sea level rise depends both on water expansion (through heat accumulation) and additional water mass (through glacial melting). So it shouldn’t surprise us if we don’t see any correlation between temperature swings and sea level rise.
Still, I wanted to see if there’s something about tornadoes that makes them increase or decrease when temperatures go up. We’re always hearing about the 2ºC target, so I figure changes of over 1ºC should be enough to cause some change in the tornado count.
How I counted
NOAA’s storm database is here. As of this writing, at least the files for 2015 and 2016 have a new weird format that apparently removes the ‘event type’ column (and a lot of other things); I haven’t looked at all the files, for obvious reasons, but fortunately I downloaded the data a few months ago and those do indicate the event type (tornado, hail, etc).
So the files and R-Code I actually used are stored here. The link is only missing the data for 2016, which I got from Wikipedia. If anyone knows of a more ‘official’ answer I can plug it in – though honestly it wouldn’t change the results. In the Dropbox link you’ll also find my R-code and the temperatures for 1951-2016.
The 1951-2016 period has 66 years, but since we’re looking at changes, there are 65 ‘points’. And as you can guess looking at the chart…
…the correlation is 1.5%, which is to say effectively zero.
The problem with using changes in temperature is that, obviously, absolute temperature also plays a role. Perhaps going from 14ºC to 15ºC has no effect on tornadoes, but moving from 15ºC to 16ºC does cause an increase. To test this I checked only the years since 2000: an increase in temperature from that baseline leads to higher absolute values than one from earlier dates.
In that case the correlation is -33%. I don’t think this really means there is a negative relationship between tornadoes and temperature – it’s just that with so few points you will get skewed results.
What this doesn’t mean
This simply shows there is no correlation between tornado change and temperature change. Of course, in reality any change in temperature will affect tornadoes somehow; perhaps the changes that increase this count are offset by others that decrease it, or perhaps the influence of temperature is simply too small and gets lost in the noise of variability. To probe deeper, one could also look at trends depending on regional temperatures, or comparing day time with night time and so on.
For now that is beyond my means, in terms of both skill and free time. I hope you enjoyed this post – feel free to tinker with the data and code provided in the references.
NOAA/NCEI Storm event data: https://www1.ncdc.noaa.gov/pub/data/swdi/stormevents/csvfiles/
Data files and R-code used to make counts: https://www.dropbox.com/sh/lb69l5isa25guc2/AAC4-gYr3N6TVDxEGzBB8fJVa?dl=0
R graphical language (free): https://www.r-project.org/