08 October 2012

New Paper: Normalized Tornado Damage in the United States: 1950-2011

I am a co-author on a new paper just accepted for publication in the journal Environmental Hazards titled "Normalized Tornado Damage in the United States: 1950-2011." My co-authors are Kevin Simmons (Austin College) and Daniel Sutter (Troy University). This blog post provides an FAQ on the paper, which the journal's editor tells us is expected to be formally published no later that first quarter 2013.

What is the title and abstract of this new paper?
Normalized Tornado Damage in the United States: 1950-2011

in press, Environmental Hazards

Kevin M. Simmons, Daniel Sutter and Roger Pielke, Jr.


In 2011, thunderstorms in the United States resulted in 550 deaths from tornadoes and more than $28 billion in property damage, according to data from the US National Oceanic and Atmospheric Administration, with the vast majority of economic losses resulting from tornadoes. This paper normalizes U.S. tornado damage from 1950 to 2011 using several methods. A normalization provides an estimate of the damage that would occur if past events occurred under a common base year's societal conditions. We normalize for changes in inflation and wealth at the national level and changes in population, income and housing units at the county level. Under several methods, there has been a sharp decline in tornado damage. This decline corresponds with a decline in the reported frequency of the most intense (and thus most damaging) tornadoes since 1950. However, quantification of trends in tornado incidence is made difficult due to discontinuities in the reporting of events over time. The normalized damage results are suggestive that some part of this decline may reflect actual changes in tornado incidence, beyond changes in reporting practices. In historical context 2011 stand out as one of the most damaging years of the past 61 and provide an indication that maximum damage levels have potential to increase should societal change lead to increasing exposure of wealth and property.
Why is this paper important?
This is the first paper to comprehensively "normalize" historical economic losses from US tornadoes. Normalization methods have been widely applied to phenomena around the would including US hurricanes and Australian bushfires. We use damage data collected by the National Oceanic and Atmospheric Administration, a US government agency which houses the Storm Prediction Center. We use three methods to adjust the data to a common base year (2011).The analysis allows us to say something about damage over time, in this case since 1950. The Figure below shows normalized damage 1950-2011 for one of our adjustment methods.
What do you find?
  • Overall we find a decrease in damages since 1950.
  • Even so 2011 was one of the 3 most costly years in our dataset.
  • Our dataset includes 56,457 tornadoes, of which 33,746 caused some recorded damage.
  • Since 1950, in round numbers, tornadoes resulted in about half the normalized damage as did hurricanes and twice that of earthquakes 
  • The strongest two categories of tornadoes (called EF4 and EF5) represent about 1% of all reported events but have caused almost 45% of all normalized damage.
  • The most damage per sqaure mile from1950-2011 has occurred in Massachusetts and Connecticut.
  • The most damage overall  from 1950-2011 has occurred in Texas and Alabama.
  • During the calendar year 80% of damage occurs January-June.
  • The most damaging months are April (31%), May (20%) and June (16%).
What about long-term trends in damage and tornado incidence?
Here is what we say in the paper:

 "The normalized results are also suggestive that the long-term decrease in reported tornado incidence may also have a component related to actual, secular changes in tornado incidence beyond reporting changes. To emphasize, we do not reach any conclusion here that stronger that “suggestive” and recommend that this possibility be subject to further research, which goes beyond the scope of this study.

On climate time scales there is no indication of increasing incidence of tornadoes, and the increases documented over the short (sub-climate) period 2000-2011 are strongly influenced by the large number of events documented in 2011. However, the decreased frequency of high damage events in recent decades as compared to previous decades is a notable feature in the time series and provides strong counter-evidence to claims found in the scientific literature that the atmospheric environment that spawns tornadoes has intensified leading to more intense events on climate time scales (e.g., Trenberth, 2012). Such claims are commonly found in the popular media and also in the insurance/reinsurance sector where they influence public opinion and decision making in business and government. The most recent review by the IPCC found no basis for claiming an increase (or decrease) in tornado incidence or intensity (IPCC, 2012)."
How can I get an advance copy?
If you would like a copy of the accepted paper (that is, before the final revisions and pre-proof) you can send an email request to me, rpielkejr@gmail.com.


  1. Roger,

    Now, see... what if the trendline for normalized tornado damage is declared to be "non-linear"? It would follow that we've only just started a serious increasing of normalized tornado damage-- yes? (Even more-so if the dataset is arbitrarily begun at, oh, 1980-ish)

    "I kill me", Alf

  2. Do building codes, which have (presumably) become more stringent over the years, have an effect on monetary damages?

    P.S. The labels on the horizontal axis of the graph are truncated.

  3. Roger is your paper the first to address "normalization" for tornacoes? It has been done for hurricanes but this is the first paper I've seen that addresses tornadoes.

  4. -2-HowardW

    Thanks, we do not see evidence of a building code signal in the tornado or hurricane data. One reason is that codes have not uniformly become more stringent and building styles have changed. For instance, with respect to hurricanes, the best performing buildings date to the first half of the 20th century.

    We'll get that figure tidied up, thanks for the close look!

  5. -3-PapZu

    Thanks, Brooks and Doswell applied normalization methods to selected large events, but ours is the first comprehensive time series.

  6. -4- Roger
    Thanks for the explanation.

    By the way, if you're editing the figure, you might want to add "US" to the title. As a figure in the paper, that's unnecessary. But as a free-standing graph, the additional context is helpful.

  7. Roger,

    Saw a quote from you today in the front page USA Today article about the claimed increases in disaster losses. You are fighting the good fight on this, but many in the media and reinsurance industry (esp. Munich Re!) still refuse to listen about no observable trends in normalized disaster losses. Apparently, they just don't like that narrative and prefer to make up their own. :)

  8. Good morning, Dr. Pielke!

    Several things very quickly:

    I just finished reading your guest commentary in the Denver Post.

    I have been following the arguments over climate change for many years (full disclosure: I am a Skeptic, who follows the incoming science very closely).
    I am a writer / artist, who sometimes posts articles on climate change to the art site which I am a member of (Deviantart). I write based upon what I am learning. Many times, I do not have the time to lay things out as I would wish.

    What I would like to ask, in that regard, is this: Might I obtain your permission to re-post your Denver Post piece, along with the above article here, to the site in question?

    I've been asking this question of a number of scientists / Editorialists out there, including Mr Watts, Pointman, Dr Crockford of Polar Bear Science, and others to come. I would, of course, link back to your site, and the site of the original article in the Denver Post.

    It is my goal to get Accurate information out, on a site upon which True Believers of AGW, run rampant...

  9. So you might say that "April is the cruellest month".

    (T. S. Eliot, "The Waste Land")

  10. The published paper plus data can be found here:

  11. There is some evidence to suggest strong tornadoes are less frequent. I don't think it's compelling evidence and, as a result, I wouldn't phrase things the way Roger has in some venues. When I give talks, I talk about the raw observations and why I think there's strong evidence that they don't tell the complete story. If I have to summarize my position in a sound bite, it's we can't tell if there are more or fewer strong tornadoes, but there's stronger evidence for increased variability in recent years (work in progress.)

  12. Roger

    what do you make of Judy Curry's point that "the undercounting of tornadoes is a problem prior to 1990 (worse undercounting as you go back in time); hence a decrease in recent hurricanes relative to historical values cannot be attributed to undercounting"

  13. -12-@ReinerGrundmann

    Thanks (I assume that you mean tornadoes) ... in our paper we looked at multiple, independent datasets and found a consistent story -- no evidence for increasing tornadoes and a hint of support for an actual decrease.

    This is much the same as what Harold Brooks says in -11-.

    I'm actually pretty surprised that people are making a big deal about tornadoes, as there are not very many (zero?) scientists who would actually claim to see a trend in tornado incidence or impacts. This is also the view of the IPCC.


  14. In a recent post at another blog a fellow states that your method of determining intensity and incidence is suspect. Given that you are unable to tie a trend to physical mechanisms it is his opinion that your conclusions may not be reliable.

    My question - why is normalized tornado damage a better proxy than tornado reports for identifying trends?

  15. -15-DGH

    Thanks. You ask: "why is normalized tornado damage a better proxy than tornado reports for identifying trends?"

    Damage data are better for looking at trends in ... damage. We use tornado reports as a consistency check in the paper. If reported tornado and normalized damage trends align, that gives us some degree of confidence in the fidelity of the normalization procedures.

    Because they do align, and show a downward trend, that give some evidence to suggest that the observed decrease in strong tornadoes may have an actual physical component, rather than just reflect reporting changes, as described in the paper.


  16. - RPJr - 16

    So Is it fair to say that someone who only considered the incidence data for extreme tornadoes might come to a different conclusion regarding trends than someone who used that data as a consistency check for normalized damages?

  17. -17-DGH

    QUESTION: "So Is it fair to say that someone who only considered the incidence data for extreme tornadoes might come to a different conclusion regarding trends than someone who used that data as a consistency check for normalized damages?"

    Anything is possible of course, but this seems unlikely given the literature. In our paper we cite a number of studies that have looked at tornado incidence data: Brooks & Doswell, 2001; Brooks & Dotzek, 2008; Doswell, Brooks, & Dotzek, 2008; Feuerstein, Dotzek, & Grieser, 2005; Verbout, Brooks, Leslie, & Schultz, 2006.

    None argue that extreme tornadoes have increased. Several argue that there are important inhomogeneities in the data, which can limit what can be said on climate time scales. We attempt to deal with this by breaking the data into three distinct periods for our analysis and segregating by reported strength, as you've seen in the paper. This allows for a more meaningful comparison with the normalized loss data, given the reporting changes.

    The tornado incidence data aligns well with the normalized loss data.