Statement from the Open Atmospheric Society on the paper ‘Peak tornado activity is occurring earlier in the heart of “Tornado Alley” ‘

oas_logo_423x423_xpar_bkgFrom The Open Atmospheric Society


Statement from the Open Atmospheric Society on the paper ‘Peak tornado activity is occurring earlier in the heart of “Tornado Alley” ‘ (DOI: 10.1002/2014GL061385)

The recent paper by Long and Stoy published September 10th, 2014 in Geophysical Research Letters and sent out as a press release⁴ by the American Geophysical Union makes claims that are not fully supported by the data.

The source of the data is from the NOAA Storm prediction Center (SPC) database of tornado reports. The OAS has no issue with the data, and believes it is correct and representative of actual tornado reports.

However, interpretation of the data by Long and Stoy may lead to a conclusion rooted in reporting bias.

Since the introduction of the nationwide WSR-88D Doppler radar network from 1992 to 1997, the ability to spot smaller scale tornadoes has increased dramatically, with warnings being issued based solely on radar signatures. Simmons and Sutter, 2004 stated:

The accuracy of tornado warnings has increased over the decades, and with the advent of Doppler weather radar, warnings can now generally be issued before a tornado actually touches down.

Further, in 2008, the Super Resolution upgrade² to the WSR-88D provided even better detection of the weakest tornadoes:

Super Resolution data should lead to increased tornado warning lead times. Simulations using Super Resolution data show that mesocyclone and tornado signatures can be detected at greater ranges than with legacy resolution data. In addition, other smaller scale features should be detectable in base products sooner or with greater reliability.

These improvements in technology, combined with greater awareness, greater saturation of news gathering, increased interest in both private and commercial storm spotting, plus an increased ability to rapidly report tornadoes has increased the count of weaker F0 and F1 Fujita Scale tornadoes in the last two decades.

In the paper Tornado Trends Over The Past Thirty Years³ they state:

The increase in reported tornado frequency during the early 1990s corresponds to the operational implementation of Doppler weather radars. Other non-meteorological factors that must be considered when looking at the increase in reported tornado frequency over the past 33 years are the advent of cellular telephones; the development of spotter networks by NWS offices, local emergency management officials, and local media; and population shifts. Changnon (1982) and Schaefer and Brooks (2000) both discuss these influences on tornado reporting. The growing “hobby” of tornado chasing has also contributed to the increasing number of reported tornadoes. The capability to easily photograph tornadoes with digital photography, camcorders, and even cell phone cameras not only provides documentation of many weak tornadoes, but also, on occasion, shows the presence of multiple tornadoes immediately adjacent to each other.

These improved detection and reporting reporting methods most certainly contribute to recent increases in tornado counts, and by extension earlier and more frequent tornado counts than would have been recorded 30 years ago.

The annual tornado trends chart is a result of the following methodology applied to the SPC observed tornado dataset by Harold Brooks, NSSL and Greg Carbin, SPC5.

An inflation adjustment was developed to reflect the improved detection. The SPC simple linear regression equation is fit to the 1954-2007 annual tornado totals. This equation is then used to compute the delta, or difference, between the original/observed annual tornado total and the smoothed, or “adjusted” annual total represented by the point on the linear trend line for that year.


Figure 1: Tornado count per year with linear observed trend not adjusted for inflation (better detection). (Brooks and Carbin SPC 2008)

“Using 2007 as the “baseline” year, we apply each year’s delta value from 1954 to 2007 to the linear trend value of 1283.3 for 2007. Each year is thereby adjusted, or standardized, to the 2007 annual tornado numbers. (Note that applying the delta of -185.3 to the 2007 adjusted value of 1283.3 results in the original/observed “baseline” total for 2007 of 1098 tornadoes.)
When these annual adjusted values are plotted, we see that the linear upward trend is removed from the data (see figure 2 below). Removal of this upward trend is desirable because the increase in tornado reports over the last 54 years is almost entirely due to secular trends such as population increase, increased tornado awareness, and more robust and advanced reporting networks. By removing the upward trend and making the broad assumption that 2007 represents something closer to reality for annual tornado numbers, we can attempt to answer the question, “what constitutes a normal year with respect to modern-day tornado reports?”

The answer becomes the adjusted average across the 54-year period, or 1283.3 tornadoes per year. This value is also the max trend line value at 2007 that was combined with the individual delta value for each year to adjust all annual totals in the data set.



Figure 2: Tornadoes adjusted to 2007 baseline (detrended) to account for inflation (better detection) (Brooks and Carbin 2008)

In a note on September 17 to the OAS, Mr. Carbin added:

“I have not de-trended any more total annual data since the 2007 analysis. Obviously, we have experienced a few very quiet years since the remarkable year of 2011. The key point in the statement above is that we are expecting 2007 to represent something close to an average annual total. If we use the (E)F1 and stronger annual counts for the past 61 years, the annual mean is below 1000 tornadoes and there is a distinct upward trend in the annual counts). However, is we limit the annual EF1 tornado counts to the past 25 years (1989-2013) then we get an annual mean closer to 1200, and essentially no trend (see figure 3 below). Whatever period is used, one significant characteristic of the annual counts since 2000 is the profound variability about the mean. Some of the greatest annual standard deviations (both positive and negative) have occurred within a little more than the past decade. This is perhaps the most puzzling aspect of the record.”




Figure 3: EF1+ Annual Count not detrended for inflation due to improved detection (Carbin 2014)


The NOAA SPC had shown in 2013 that there was no trend in strong tornados – EF3+ (see figure 4).


Figure 4; US Annual Count of Strong to Violent Tornadoes (F3+), 1954-2012 (Carbin, SPC)

We also note, in the AGU press release for the Long and Stoy paper, it was mentioned:

“If we take Nebraska out [of the data], it is nearly a two-week shift earlier,” noted John Long, a research scientist in the Department of Land Resources and Environmental Sciences at Montana State University in Bozeman, Montana, and lead author of the new paper. For tornadoes rated above F0, the lowest rung on the original Fujita scale of tornado strength, the shift is also close to 14 days, according to a preliminary analysis by Long and his colleagues that’s not included in the new paper.

It is the opinion of The OAS that this sort of methodology to remove a portion of a dataset to cite a result is unsupportable and without justification. Climatic scale detection of earlier onset of tornado activity cannot be dependent upon removal of a portion of the dataset. Other analyses of the same data set by the primary investigator for tornado climatology show that there are no trends in frequency or intensity of tornadoes, and by extension, suggest that the claims made by Long and Stoy are little more than artifacts of statistical methodology and an increase in the ability to spot, report, and categorize tornadoes that would have gone unnoticed and unreported thirty years ago.


For more information on The Open Atmospheric Society, or to become a member, see

[Added, should have been in original release] Prepared by OAS members Joe D’Aleo and Anthony Watts. Assistance from NOAA SPC Greg Carbin is gratefully acknowledged. For further information, comments, or other issues, write to contact “at” theoas dot org or use the contact form at


1. Tornado Warnings: How Doppler Radar, False Alarms, and Tornado Watches Affect Casualties, Economic and Societal Impacts of Tornadoes 2011, pp 117-171 DOI 10.1007/978-1-935704-02-7_4

2. WSR-88D Build 10/Super Resolution Level II FAQs, NOAA Radar Operations center:

3. Tornado Trends Over The Past Thirty Years by Daniel McCarthy and Joseph Schaefer of the NOAA/NWS/NCEP/Storm Prediction Center, Norman, Oklahoma

4. Tornadoes occurring earlier in “Tornado Alley”, AGU Newsroom 16 September 2014,

5. Inflation Adjusted Annual Tornado Running Total Trends by Greg Carbin, NOAA Storm Prediction Center2008

  1. #1 by Harold Brooks on September 19, 2014 - 11:17 am

    Why would a change in the overall number of reports shift when they occur during the year? The technique finds the mean of the date of occurrence and the standard deviation of the date of occurrence, so that the total number of tornadoes for each 10-year period is irrelevant to the date calculations (see Figure 5). If you’re trying to say that weak tornadoes, which have increased in reports occur preferentially earlier in the year than strong tornadoes, that’s not physically likely.

    As far as the Nebraska comment is concerned, they give the result for Nebraska in the paper (earlier by 4 days, p-value of the change is 0.03). The comment in the press release is based off of Table 1, which shows that the peak in TX, OK, and KS have all moved >10 days earlier. All the states show an earlier peak, but Nebraska is the smallest change.

    Your comments make me wonder whether you actually read the paper.

  2. #2 by Anthony Watts on September 19, 2014 - 5:05 pm

    Dear. Mr. Brooks, thanks for your comments. You asked:

    “The trend is then estimated for those calendar dates. Unless you have some reason to think that the increase in reports has preferentially happened earlier in the year, the increase in reports since 1954 has zero impact on their result. “

    What we see in the data is an overall increase in reporting across all calendar dates. The new warning systems in place have had a great effect on tornado lead times, and an increase in storm chaser teams “itching to go” at the first sign of potential severe weather, combined with improvements in convective outlooks, NEXRAD signature detection, and other NOAA based and private improvements in storm chasing most certainly have an impact.

    For example:

    There are now many more eyes, budgets, and reputations staked on catching the early season tornadoes where there was not 20 years ago. Plus, with the ability to interact with NOAA radar live on ceel phones, tablets, etc, the accuracy of spotting tornadoes has increased.

    While they claim there is no effect on tornado reporting increases having a bias (mentioned in the last paragraph of the introduction) they did no analysis that I can see that specifically excludes that possibility. In fact, they removed some late season data, that if it had remained, might have negated some of the early shift they claim.

    I don’t know how one can justify valid tornado reports as an “outlier” simply on a whim. Who decides if 4 STD is the limit? That seems like an arbitrary author choice. Either it is valid data or it isn’t.

    We found no evidence that contradicts an increase in early season reporting due to these external reporting factors mentioned above. We would expect to see this evidence across the data set, as you note about table 1. f you have evidence that these external reporting factors have had no effect on the early reporting increase, I welcome seeing it.

    The idea of removing Nebraska to make a point with layman readers who cannot get access to the paper should not go unchallenged. The layman certainly can’t, since the paper is not accessible unless you are a GRL subscriber.

    For reference, here is Table 1 from Long and Stoy 2014:

    It should be noted that the conclusion engages in quite a bit of wordsmithing:

    Preaching for preparedness is a good thing, preaching that even though no other metrics or consensus support a link to AGW but somehow this one paper does via some statistical and outlier removal claims that in my view have not been fully examined, leads me to believe this paper is more speculative than factual.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s