By Paul Homewood
http://news.fsu.edu/More-FSU-News/Researchers-develop-model-to-correct-tornado-records
[Note Update Below]
Quantification of long term tornado trends has been hampered by the fact that many more tornadoes are reported today than was the case in the past. NOAA summarise this well:-
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. To better understand the variability and trend in tornado frequency in the U.S., the total number EF1 and stronger, as well as strong to violent tornadoes (EF3 to EF5 category on the Enhanced Fujita scale) can be analyzed. These are the tornadoes that would have likely been reported even during the decades before Doppler radar use became widespread and practices resulted in increasing tornado reports.
FSU Professor, James Elsner, has attempted to develop a model to determine the real trends, in the paper “The Decreasing Population Bias in Tornado Reports across the Central Plains”.
I have no particular comment to make about the model itself. WUWT has covered this issue well already.
I was intrigued, though, by the comment in the press release that “The model shows that it is likely that tornadoes are not occurring with greater frequency, but there is some evidence to suggest that tornadoes are, in fact, getting stronger”
Elsner reinforces this message with this quote:-
“The risk of violent tornadoes appears to be increasing. The tornadoes in Oklahoma City on May 31 and the 2011 tornadoes in Joplin, Mo., and Tuscaloosa, Ala., suggest that tornadoes may be getting stronger.”
We’ll leave aside the obvious fallacy that a few isolated events constitute some sort of trend, and look at what the facts tell us.
But first, a bit of detective work.
A Statistical Model for Tornado Intensity
A close study of the Elsner paper does not appear to reveal any reference at all to “tornadoes getting stronger”. (The paper is here – if anyone finds such a reference, let me know!) So where does he get this evidence from?
It appears to be based on another paper he has written this year, A Statistical Model for Tornado Intensity. There is already, of course, a well established system of measuring the severity of tornadoes, the Enhanced Fujita Scale, which uses damage surveys to estimate wind speeds.
Elsner, however, has taken a different approach, and uses a combination of tornado path length and width to estimate wind speeds. (The logic being that the strongest tornadoes tend to have the longest/widest paths).
This approach has a very fundamental drawback – the reliability of historical data relating to length and width. In Elsner’s provisional work, he seems to have used 1985 as his start point, but his paper now uses 1994 instead. Elsner told to me in May that “the data had problems before the mid 1990’s”, a fact which is confirmed in his paper.
And we also know already that, in 1994, NWS changed their recordings of width from “mean” to “maximum”. (Even then, discrepancies in the data where path width has been underestimated have been found more recently, such as this one in 1997 – see here).
But, even assuming that the data since 1994 is accurate, we are still only left with a trend of just 18 years. Is Elsner seriously suggesting that reliable trends can be calculated over such a short period of time?
Nevertheless, his paper concludes
More work needs to be done to understand the upward trends in path length and width. The increases lead to an apparent increase in tornado intensity across all EF categories.
EF Scale
So, how does analysis using the traditional EF Scale compare with Elsner’s study? Does it confirm his results?
Let’s start by looking at NOAA’s graph of stronger EF3+ tornadoes.
http://www1.ncdc.noaa.gov/pub/data/cmb/images/tornado/clim/EF3-EF5.png
Certainly, 2011 sticks out , but it is absolutely clear that the last decade is unexceptional. It is also clear that there were many more strong tornadoes in the 1970’s, and to a lesser extent in the 1950’s.
This analysis surely casts huge doubt on Elsner’s conclusions, based as they are on such a short period.
But what about the strongest tornadoes, the EF-4’s and EF- 5’s? If Elsner’s theory is correct, surely we should be seeing an increasing frequency of these?
I have plotted below the numbers of these tornadoes since 1970’ using the data from NOAA’s Storm Prediction Centre.
http://www.spc.noaa.gov/wcm/#data
The situation is quite clear on EF-4’s – the trend has significantly declined, and even 2011 had far fewer events than 1974 – 17 v 29.
The position with EF-5’s is less clear, with a slightly increasing trend. However, this trend has been heavily influenced by the single year of 2011.
With an average of just one EF-5 tornado a year, the numbers are simply too sparse to reach any proper conclusion. It still remains the case that 1974 had more EF-5’s than 2011, and that the decadal total in the 1970’s was significantly higher than the latest 10 year period.
Conclusions
- The Elsner paper relies on far too short a period to draw any meaningful conclusions.
- It also totally ignores the evidence of the 1950’s and 1970’s, which clearly points to a decrease in stronger tornadoes over the longer term , not an increase.
- Elsner’s conclusions seem to be heavily influenced by the EF-5 tornadoes in 2011.
- The analysis relies totally on the accuracy of path width and length data. While this may now be reasonably accurate, past data is highly suspect.
- There are attempts in the paper to link the supposed increase in tornado intensity to “climate change”. Elsner goes further in an earlier presentation, saying
“Evidence points to growing frequency and intensity of extreme weather worldwide due to global warming, but an effort to detect changes in the intensity of tornadoes has yet to be made.
Here we show compelling evidence for a growing trend in the ferocity of strong tornadoes across the United States.
But as there has been no warming since 1997, how can any increase in intensity since then be due to global warming?
- The paper’s conclusions state that “the EF Scale is not adequate for analyzing tornado intensity”. I am not sure if the tornado experts at NOAA would agree.
There is no doubt that the decreasing frequency and severity of tornadoes is a problem for those who claim that global warming has brought an increase in extreme weather.
There is already a well established system for measuring the strength of tornadoes, and a well established database to back it up. Yet Elsner has attempted to replace this with a statistical model, based on a small amount of potentially unreliable data.
Is this a case of moulding the evidence to suit the theory?
Footnote
To avoid any ambiguity, in the Introduction to Elsner’s paper, he equates “tornado intensity” with “wind speed”.
Update
When I ran the above graph for EF-5 tornadoes, I had left the zero fields blank. This apparently gives a false trend line, so I have now rerun it with zeros. The result, as below, now shows the true position, which is a declining trend.
Thanks to Donald Mitchell and Scott Scarborough for pointing this out.
References
1) FSU Press Release
http://news.fsu.edu/More-FSU-News/Researchers-develop-model-to-correct-tornado-records
2) “The Decreasing Population Bias in Tornado Reports across the Central Plains” by James B. Elsner and Laura E. Michaels
http://journals.ametsoc.org/doi/abs/10.1175/WCAS-D-12-00040.1?journalCode=wcas
3) “A Statistical Model for Tornado Intensity” by James Elsner, Thomas Jagger & Ian Elsner.
http://myweb.fsu.edu/jelsner/TornadoWindSpeeds.pdf
4) NOAA Tornado Climatology Page
http://www.ncdc.noaa.gov/oa/climate/severeweather/tornadoes.html
5) Storm Prediction Centre Database
http://www.spc.noaa.gov/wcm/#data
I don’t get it. I graphed the “US Tornadoes – EF5 only” and got a slope of -0.011182, down instead of up. The data for each year is:
1970 2011
110170310100101100003110001121000000011006
I get an average value of 1.09756098 with a slope of -0.011182.
What am I doing wrong? It visually looks to me like the slope should be negative and my Excel says so too!
[It will be more clear with spaces between the data. mod]
Maybe you can educate me a bit, when tornado prone regions have higher density populations the damage will obviously appear a lot more devastating. I ask this because twenty years ago, Britain had a great storm, that flattened 1 million trees, and tornadoes are not unheard of in medieval times near the coast. They start as water spouts, one hit Australia a year or two ago and most probably would have not been noticed if homes were not built in the area it hit land. When I lived in Bermuda, we saw water spouts a lot, and we had a hurricane just clip us. I lived on Blue Hole Hill and had an 180 degree view of the causeway and Atlantic. Anyway I just wanted to ask as I can’t understand why people build in Tornado or Hurricane, Cyclone regions without reinforcing their homes to withstand the forces that they will encounter.
Least squares regression is quick and easy to do. Unfortunately, least squares is very vulnerable to deceptive results due to a few unusual events: that is outliers. That is the case here, See 1974 and 2011. Least Squares is exactly the wrong tool to determine a trend with this data. If that is all that the original paper or this response used to determine trends, throw the whole lot out, there’s nothing of value in the discussion.
Kendall’s Tau would be a believable way to determine a trend. I’m sure a statistician reading this could suggest others. I’d find the trend myself but I’ve other commitments that are keeping me busy and worn out for the next few days.
Concerning the EF5 tornado graph. I get a positive slope of the same magnitude as shown in the graph if I eliminate all zero years from the data. I think Mr. Homewood made a mistake. The years in which zero EF5 tornadoes occurred should not be removed from the data, but that is the only way that that data would show a positive slope. Someone prove me wrong. If I eliminate all zero years from the data I get a trend slope of 0.006054. Multiply that by the 42 years shown and you get a rise of 0.254 which is the rise in the trend line shown in the graph above.
Concerning the EF5 tornado graph. OK, Here is the data with spaces between it. Each datum is the number of EF5 tornadoes in each year shown in the graph in the article.
1970…………………………………………………………………………………………………….2011– Years
1 1 0 1 7 0 3 1 0 1 0 0 1 0 1 1 0 0 0 0 3 1 1 0 0 0 1 1 2 1 0 0 0 0 0 0 0 1 1 0 0 6 –Tornadoes/year
This yields a linear regression crossing the y axis at (a mistake earlier) 1.09756 tornadoes with a slope of -0.011182 tornadoes per year. That is a negative slop! Not a positive slope.
To get the slope shown in the graph in the article above you have to remove all the zero years in the data. That would look like this:
1 1 1 7 3 1 1 1 1 1 3 1 1 1 1 2 1 1 1 6
This data gives a linear regression crossing the y axis at 1.736842 tornadoes (looks to be consistent with the graph in the article) with a slope of 0.006054 tornadoes per year. This slope, when multiplied by the number of years shown in the graph (42) yields a rise of 0.254 tornadoes which is the rise shown. Some how Mr. Homewood used a function which ignores zero tornado years. This is not correct.
Also, Actually I made a mistake. If the EF5 tornado graph goes to 2012. I thought the data ended at 2011. But, if 2012 is in the data, it appears to be a zero tornado year. Mr. Homewoods results would not change because his calculation ignores zero tornado years in calculating the trend. But if 2012 is a zero EF5 Tornado year, then the slope is -0.013138 tornadoes/year with a Y axis crossing of 1.1262 tornadoes. This would yield a linear regression line with reduction in tornadoes over the 42 year period of 0.62836 tornadoes. This negative slope is 2.17 times as great as the positive slope shown in the EF5 tornado graph! Willis… is that significant?
1974 and 2011 were two analog periods of the 18.6 year long lunar declinational cycles apart, I would not expect to see another severe tornado season like them for another 37 years.
http://research.aerology.com/lunar-declinational-affects-on-tornado-production/
the analog years for this year (2013) are 1994, 1976, 1958, and 1940.
Just like the hurricanes Irene and Sandy were repeat occurrences of the analog years from 1938 and 1939 the last intense hurricanes that came into the NY and New Jersey area.
It is important to keep the Lat and Long and track length for all tornadoes clean from extraneous fake tornadoes adjustments, as that information is usable in forecasting the location probabilities for future tornado production.
Hail reports by crop damage insurance claims archived are a good (but overlooked) resource to use to forecast potential future hail damage claims, only if you keep the Lat and Long data with the number of occurrences per date, so filtering of the analogs by lunar declination can forecast the repeating cycles.
How to correct tornado ( or temperature) historical record “right”?
well you need to know what was the right hsitorical data…well …then you have to make assumptions…
how do you know that your assumptions were right…? You can’t…you re just making assumptions…
I grew up in the heart of tornado alley in the 1950s and remember the scary TV news video of towns ravaged by killer tornadoes – little towns like Udall, Kansas wiped from the Earth, in May, 1955. There was the great Super Outbreak in 1974 in the Midwest. It’s an unscientific assertion, but I have difficulty, despite recent nearby disasters, believing there is any trend toward more or stronger tornadoes, at least in the last 60 years or so,
If you look here, you will see that Central Oklahoma and areas in Dixie Alley, since 1950, have had a notable concentration of F5 / EF5 twisters. Basing any conclusions on events in the last 5 years is highly misleading.
http://www.spc.noaa.gov/faq/tornado/f5torns.html
Donald Mitchell
Needless to say, when I entered zeroes for the years in the original graph, it gave a slope of -0.0143
Thanks for the tip. I had left the zero fields blank. When I enter zeros, it does turn the slope negative.
Scott Scarborough
I don’t get it. I graphed the “US Tornadoes – EF5 only” and got a slope of -0.011182, down instead of up.
See my reply above to Donald Mitchell
. You are quite right, there is a negative trend.
I still believe, though, that the number of EF-5’s is so small over the period, that no meaningful conclusions can be drawn.
The more I look at the FSU press release, the more mystified I become.
The model shows that it is likely that tornadoes are not occurring with greater frequency, but there is some evidence to suggest that tornadoes are, in fact, getting stronger.
“The risk of violent tornadoes appears to be increasing,” Elsner said. “The tornadoes in Oklahoma City on May 31 and the 2011 tornadoes in Joplin, Mo., and Tuscaloosa, Ala., suggest that tornadoes may be getting stronger.”
I have reread the paper referred to in this press release, and there is absolutely no reference at all to tornadoes getting stronger, or tornado intensity at all.The paper even says
It should be noted that we make no attempt here to examine issues associated with quantifying
tornado intensity through damage estimates.
The abstract is here.
http://journals.ametsoc.org/doi/abs/10.1175/WCAS-D-12-00040.1?journalCode=wcas
I have read the full paper as well, but I understand there are copyright problems with reproducing it.
Which all rather raises the question, why did Elsner make the statements he did? He may believe that his earlier paper provided such evidence, but it was not true to claim that this model did.
Perhaps it was just a case of the FSU mixing up the facts. But I would suggest that Elsner needs to issue some sort of clarification himself.
Before he even started his study, when he merely announced that he would do it, I would have bet my life savings that his new model would conclude that tornado intensity was increasing. I’ve seen it over and over again with data that doesn’t fit the model. Someone comes up with excuses to “adjust” the data or a new way of looking at the data (a new index), and suddenly the data falls into line.
@Paul Homewood
Now that you realize you’ve screwed up the EF5 regression, you really ought to correct it. Your ultimate conclusion, that there simply aren’t enough EF5s from which to draw meaningful inferences in either direction, remains sound, but that’s no reason to leave an incorrect chart in this post.
[Thanks – this has now been updated]
The Monster
I have redone the graph and asked Anthony to post an update.
(The updated graph is also on my website below.)
http://notalotofpeopleknowthat.wordpress.com/2013/09/14/are-tornadoes-getting-stronger/
It seems readily apparent by the evidence and by Elsner’s own words that this is more of the same extreme weather drum beating they’ve been throwing at the wall for a decade now. The campaign’s evolution seems a hockey schtick in itself as the claims progress further and further from the realities “on the ground.” It’s a sight to behold and, in a way, inspirational. Cheers!
This is the right blog for anyone who hopes to understand this topic.
You understand so much its almost tough to argue with you (not that I personally
would want to…HaHa). You definitely put a fresh spin on a subject that has ben written about for a long time.
Excellent stuff, just wonderful!
Speed as rate of forward motion — does it trade off with rotational speed, or width?
“The risk of violent tornadoes seems to be increasing. Tornadoes in Oklahoma City on May 31 and the 2011 tornadoes in Joplin, Mo. and in Tuscaloosa, Alabama, they suggest that tornadoes may be becoming stronger.”