Draft Paper for Comment: Decreased Proportion of Tropical Cyclone Landfalls in the United States

December 28th, 2006

Posted by: Roger Pielke, Jr.

Below you will find a short, draft paper on the decreasing proportion of U.S. hurricane landfalls to total North Atlantic hurricanes from 1851-2006 that I will soon submit for publication in a peer-reviewed journal. I am not worried about pre-publication on the web and the policies of the big journals as Nature has already declined to send it out for review, suggesting that the specialty journals are a more appropriate venue. I have shared earlier versions of the paper with a range of different scientists inside (and outside of) the tropical cyclone research community, and I thank those who have so far responded for their helpful suggestions. I welcome any comments readers here may have as well.

Decreased Proportions of Tropical Cyclone Landfalls in the United States: Data Artifact, Blind Luck, Natural Variability, and/or Global Warming?

Roger A. Pielke, Jr.
28 December 2006


This short note is motivated by several recent studies that examine Atlantic tropical cyclone statistics over the past century and a half finding a significant upward trend (Holland and Webster, 2007; Mann and Emanuel, 2006). Both studies attribute an observed increase in North Atlantic (NATL) tropical cyclone activity to anthropogenic causes. This paper uses a simple approach to examine trends in U.S. hurricane landfalls 1851-2006. The annual number of U.S. hurricane landfalls has remained remarkably constant over this period exhibiting no trend. However, out of the total NATL storm activity, the proportion of landfalling storms making landfall in the United States has exhibited a marked decrease. These trends raise important but heretofore largely unexamined research questions about tropical cyclone landfall theory, data, and analyses.

North Atlantic Tropical Cyclone Data

Two datasets in this analysis are kept by the U.S. National Oceanic and Atmospheric Administration. The first dataset includes all tropical cyclones of at least tropical storm strength (i.e., maximum winds >17 meters/second) in the NATL 1851-2006 (1). The second dataset includes all such storms that affected the U.S. coastline at hurricane strength (i.e., maximum winds of >33 meters/second) (2). It is widely accepted that the number of landfalling storms is among the most accurate metrics available for tropical cyclones in the Atlantic (Landsea 2005, Emanuel 2005).

Issues about data quality are at the center of a vigorous and scientifically productive debate within the tropical cyclone research community (e.g., Landsea et al. 2006, Kossin et al. in press). This paper proceeds under the same assumptions on data quality used by Holland and Webster (2007),

Before 1945 the overall statistic of total tropical cyclones numbers contains useful information . . . In summary, we consider that the veracity of the NATL tropical storm data base is sufficient to enable the broad brush analysis that we undertake in this study. Prior to 1945 we concentrate on the total number of tropical cyclones, irrespective of intensity.

And also Mann and Emanuel (2006),

A reasonably reliable record of annual North Atlantic tropical cyclone counts is thus available back into the late nineteenth century.

The point here is not that these conclusions about data quality are necessarily correct, but to argue that if one accepts the analyses in these studies predicated on the fidelity of long-term counts of NATL tropical cyclones, then it follows logically that one also has to accept the analysis presented here using the exact same data.


Figure 1 shows the total number of United States hurricane landfalls 1851-2006.


There is no trend in the data. This finding is generally accepted, and is supported by the lack of trend in normalized U.S. hurricane damage over the past century (Pielke, et al. under review). Figure 2 shows the percentage of total storms that affected the United States as hurricanes from 1851-2006 (3).


There is a decrease in the proportion of storms striking the United States from any starting point prior to about 1970. This decrease increases in statistical significance as the record is extended further into the past. A linear trend line is presented on the graph for both the raw data and the 9-year moving average (4). Apparent in Figure 2 starting in 1944, which is considered to be the most accurate data period in the Atlantic (Neumann et al. 1999, Landsea et al. 1999), is a decrease in the proportion of storms striking the coast over this time period as well (statistics discussed below). Holland and Webster (2007) identify three climatic regimes, 1905-1930, 1931-1994, and 1995-2005. The proportion of landfalling storms for each of the three eras indicates a steady decline, respectively at 31.8%, 17.0%, and 14.6%.

Figure 3 shows for 1950-2006 the relationship of the percentage of landfalling storms and NATL August-September-October (ASO) sea surface temperatures (5).


On this time scale at least there is no significant relationship, though perhaps a more comprehensive statistical analysis that goes beyond the present focus may yet reveal a relationship (cf. Jagger et al. 2007).


The data presented here suggests some interesting possibilities which cannot be resolved based on the data presented here.

First, it is possible that the decreasing trend in U.S. landfalls as a percentage of total storms is a statistical artifact resulting from an undercount of historical storms. The number of landfalling storms is certainly known with much higher certainty than the total number of storms. Thus, an undercount of the total number of storms would artificially increase landfalling storms as a percentage of total storms. This first interpretation is consistent with a view that the NATL record is most accurate since 1970, and that inaccuracies exist prior to that time. This view on data quality would also be inconsistent with those who argue that the NATL record is complete since 1944 (cf. Landsea et al. 1999). If the decreasing trend is an artifact of the datasets, then it would obviously call into question any analysis of trends based on the total number of storms over thuis time period.

Second, it is possible that the decreasing trend in U.S. landfalls as a percentage of total storms is (at least in part) a reflection of the actual behavior of the climate system. That is, if one accepts the trends reported by Holland and Webster (2007) and Mann and Emanuel (2006), then logically, it follows that one also has to accept the trends reported here. In this case, if NATL tropical cyclone activity has indeed increased, then there has also been a significant decrease in the proportion of storms that strike the U.S. coast.

Under this second possibility there is then the question of attribution. Figure 3 indicates that, at least since 1950, there is in fact no obvious relationship between NATL ASO SST and the proportion of total storms that make landfall. Given this finding, if SSTs are indeed the principle factor explaining increasing tropical cyclone activity, as some studies have suggested (e.g., Holland and Webster (2007), Mann and Emanuel (2006), Hoyos et al. (2006)) and SSTs are largely forced by greenhouse gases (Santer et al. 2006), then it follows that explaining the decreasing proportion of landfalling storms requires further investigations of climate processes beyond trends in SSTs. In other words, if global warming is increasing the incidence of NATL tropical cyclones, then the same dataset that indicates this result also suggests that global warming, and/or some other factors, are acting to diminish the proportion of total storms that strike the United States.

Emanuel (2005) raises the possibility that simple randomness might explain the lack of a trend in landfalling storm intensities in the presence of such a trend in the entire NATL. This may very well be the case (Pielke 2005); however, Emanuel’s point was made with respect to an integrated index of power dissipation, for which the landfalling component was only about 1% of the total NATL data. In the case of total storms, about 20% of the total storms make landfall over the entire dataset. It is therefore quite unlikely that randomness alone explains these results.

This conclusion is unavoidable if one accepts the results of Holland and Webster (2007) that “data errors cannot explain the sharp, high amplitude transitions between the climatic regimes in the North Atlantic, each with an increase of around 50% in cyclone and hurricane numbers.” With an increase of 50% in total hurricane numbers in two transitions among three climate regimes, and landfalling storms representing approximately 20% of the long-term total, it is statistically improbable that these changes would not manifest themselves in increased landfalls, all else being equal. Consider that that 31.8% of all storms that made landfall during 1905-1930 (Holland and Webster’s first NATL TC climate regime). Under a binomial distribution the probability of subsequently observing 131 or fewer landfalls out of 788 in the period 1931-2006 (i.e., the actual observations) at the earlier period’s probability of landfall is less than 0.00001. Thus, it necessarily must be true that either the data is flawed or there are real changes in the landfall characteristics of NATL climatology.

More recently, from 1944 to 1974 17.9% (31 years, 55/307) of total NATL storms made landfall in the U.S., and from 1975-2006 (31 years, 51/360) 14.2% of total NATL storms made landfall in the U.S.. The probability of observing 51 or less landfalls in the second period at the earlier period’s landfall rate is 0.054. Thus, there is reason to believe that even in the most recent half century where storm counts have been assumed to be most accurate there are either data problems or real changes have occurred in the climatology of NATL tropical cyclone landfalls.


This short note has revisited trends in U.S. landfalling hurricanes both as annual totals and as a proportion of total North Atlantic tropical cyclone activity from 1851-2006. The data indicate that there are no trends in landfall numbers but a marked decrease in the proportion of storms that make landfall. There are several possibilities for this decrease.

•One explanation is that earlier data fails to accurately represent the total actual number of tropical cyclones, thereby artificially increasing landfalling storms as a proportion of the total.

•A second explanation is that the trends in the total number of storms are in fact reflective of increasing Atlantic activity and the decreasing proportion that make landfall results from some yet unknown climate process that may or may not have a relationship to human activity. There is no obvious relationship between SSTs and landfall proportions.

•A third possibility, that the remarkable stability of landfall numbers over time is due to randomness, is highly unlikely simply for basic reasons of probability if one assumes that landfall proportions are constant over the long run.

Scientists are nonetheless in agreement that the coming decades will see landfall numbers that exceed the average from 1970-1994. Decision makers should take care not to overlook the possibility that future landfall rates may exceed that observed in the historical record, whether due to global warming, randomness, natural causes, or some combination. A lack of knowledge about the future means that surprises should be expected. Given the importance of landfalling storms to decision makers, a concluding recommendation is that the research community should place even greater attention to the challenging and important scientific questions of hurricane landfall climatology.


1. http://www.aoml.noaa.gov/hrd/Storm_pages/Atl/ATLdate.dat

2. http://www.aoml.noaa.gov/hrd/hurdat/ushurrlist18512005-gt.txt See the NOAA WWW page for a discussion of “affected the U.S. coastline.”

3. The use of a ratio based on total storms follows the similar use of a ratio by Holland and Webster (2007) to examine trends in storm intensity.

4. Note that a 9-year moving average is used simply because this is the smoothing used in Holland and Webster (2007).

5. NOAA’s Climate Prediction Center provides NATL SST indices by month from 1950 here: http://www.cpc.noaa.gov/data/indices/sstoi.atl.indices

References to be added

12 Responses to “Draft Paper for Comment: Decreased Proportion of Tropical Cyclone Landfalls in the United States”

  1. Roger Pielke, Jr. Says:

    Judy Curry discusses hurricane landfall patterns here:

    I respond here:

    Not only does that thread get into hurricane landfalls and the recent Holland/Webster paper (which we discussed a bit last week), but it also has a lengthy discussion of the so-called Republican “war on science”! ;-)

  2. 2
  3. jae Says:

    Roger: Fascinating article. How has everyone overlooked such a simple “trend.” (I wonder if others have looked and decided to ignore it…)

    A couple of minor things:

    You say: “If the decreasing trend is an artifact of the datasets, then it would obviously call into question the trends based on the total number of storms documented by any study which is.”

    I think you left something out here.

    Also, there’s a typo in the intro. in this sentence: “This paper uses a simple approach identical to examine trends in U.S. hurricane landfalls 1851-2006.”

  4. 3
  5. Roger Pielke, Jr. Says:

    Thanks much jae … fixed and updated!

  6. 4
  7. Steve Hemphill Says:

    Consider this cranial flatulation, but I’ve always wondered if there was a correlation between missing vessels and untracked hurricanes before we could track or even know about them…

    A large percentage of those losses could have happened in what they thought of as the “Bermuda Triangle”.

  8. 5
  9. Bob K Says:

    I’m posting this at climateaudit and feel it also has a bearing on this thread.

    10kt, 15kt, 20kt, 25kt, 30kt, 35kt, 40kt, 45kt, 50kt, 55kt, more, total
    1851-60, 0, 0, 0, 0, 0, 0, 69, 0, 147, 0, 685, 901
    1861-70, 0, 0, 0, 0, 0, 0, 54, 0, 140, 0, 827, 1021
    1871-80, 0, 0, 0, 0, 52, 0, 367, 0, 396, 0, 1252, 2067
    1881-90, 0, 0, 0, 4, 56, 167, 291, 143, 349, 67, 1268, 2345
    1891-00, 0, 0, 2, 47, 94, 326, 332, 223, 405, 116, 1463, 3008
    1901-10, 0, 0, 4, 74, 288, 498, 269, 181, 190, 86, 873, 2463
    1911-20, 0, 3, 29, 44, 114, 225, 153, 142, 98, 80, 667, 1555
    1921-30, 0, 2, 13, 33, 55, 370, 191, 116, 97, 56, 861, 1794
    1931-40, 0, 18, 51, 72, 78, 732, 418, 173, 149, 113, 971, 2775
    1941-50, 0, 18, 26, 105, 80, 360, 295, 230, 188, 109, 1094, 2505
    1951-60, 0, 27, 10, 301, 178, 327, 180, 229, 191, 103, 1241, 2787
    1961-70, 0, 11, 27, 475, 502, 193, 171, 178, 170, 137, 1419, 3283
    1971-80, 0, 20, 90, 342, 478, 210, 240, 218, 160, 136, 886, 2780
    1981-90 21, 47, 81, 302, 418, 281, 247, 244, 188, 178, 816, 2823
    1991-00, 6, 21, 164, 251, 475, 329, 308, 315, 264, 171, 1334, 3638
    2001-05, 34, 29, 139, 273, 378, 293, 212, 250, 221, 166, 816, 2811
    total, 61, 196, 636, 2323, 3246, 4311, 3797, 2642, 3353, 1518, 16473, 38556

    Here’s mean wind speed for all readings and also for speeds of 30kt or more.
    All, 30kt+
    1851-60, 71.6, 71.6
    1861-70, 71.7, 71.7
    1871-80, 62.7, 62.7
    1881-90, 61.2, 61.3
    1891-00, 61.4, 62
    1901-10, 52.1, 53
    1911-20, 57.8, 59.6
    1921-30, 61.2, 62.2
    1931-40, 53.1, 54.8
    1941-50, 58.6, 60.9
    1951-60, 58.5, 63.2
    1961-70, 55.1, 60.8
    1971-80, 49.2, 54.1
    1981-90, 48.4, 53.3
    1991-00, 52.7, 56.9
    2001-05, 50.6, 56.5
    total, 56.3, 59.3
    Hmmm. Doesn’t seem to format well.
    The data could be saved as a .csv file and put in a spreadsheet for better viewing.

  10. 6
  11. Bob K Says:

    The comment part of my above post isn’t there. I guess my copy/paste from my editor didn’t include the beginning. Here it is, so people won’t wonder what that data is about.

    It is a breakout of wind speed readings from the unisys hurricane database for the Atlantic basin. The file I used is tracks.atl that ends with 2005.
    It appears that as time goes on, more low readings are being included.

    There appears to have been a shift in observational techniques around 1950 and again around 1980.

    I don’t know how this affects the statistical reliability, but thought it might have a bearing on the lower percentage of US landfalling storms. Might it be appropriate to only use obsevations above a certain wind speed threshhold? Say 30-35kts. Those are certainly the ones most likely to be recorded over long
    periods of time.

  12. 7
  13. dd Says:

    Hurricanes churn the water surface, decreasing the local SST, so the passage of a hurricane makes it less likely that another will cross the same track for a period of time. From a mathematical point of view, this should tend to restrict the number of hurricanes striking the US Atlantic and Gulf coasts, even as the total number of hurricanes increases.

    I am not sure how to test this concept, but one might begin by looking at the distribution of lengths of hurricanes (either time or distance will do), and noting if the average length decreases during busy hurricane years.

  14. 8
  15. Roger Pielke, Jr. Says:

    I just came across these two articles, which I’ll be citing in the revision:

    Solow, A. R., Moore, L. J., 2002: Testing for Trend in North Atlantic Hurricane Activity, 1900-98. Journal of Climate. 15, 3111-3114.

    The detection of a trend in hurricane activity in the North Atlantic basin has been restricted by the incompleteness of the record prior to 1946. In an earlier paper, the complete record of U.S. landfalling hurricanes was used to extend the period of analysis back to 1930. In this paper, a further extension is made back to 1900. In doing so, the assumption in the earlier paper of an exponential linear trend is relaxed and the trend is estimated nonparametrically. The results show no significant trend in basinwide hurricane activity over the period 1900–98.

    Solow A. R., and L. Moore, 2000: Testing for a trend in a partially incomplete hurricane record. J. Climate, 13, 3696–3699.

    The record of annual counts of basinwide North Atlantic hurricanes is incomplete prior to 1946. This has restricted efforts to identify a long-term trend in hurricane activity to the postwar period. In contrast, the complete record of U.S. landfalling hurricanes extends back to 1930 or earlier. Under the assumption that the proportion of basinwide hurricanes that make landfall is constant over time, it is possible to use the record of landfalling hurricanes to extend a test for trend in basinwide hurricane activity beyond the postwar period. This note describes and illustrates a method for doing this. The results suggest that there has been a significant reduction in basinwide hurricane activity over the period 1930–98.

  16. 9
  17. Rich Horton Says:

    “Fascinating article. How has everyone overlooked such a simple “trend.” (I wonder if others have looked and decided to ignore it…)”

    Actually, I’m sure quite a few people noticed it. I was writing about it for my blog back in August and September of 2005.



    In response to a criticism of my work I made the following observation:

    “In an attempt to refute my work the following was stated:

    ‘Misleading Because he limits his analysis to Hurricans that hit the United States,” and not the sum total of ALL Atlantic hurricanes, many of which never strike the US.’

    I didn’t address this issue in my original post because I believed the answer to that question was self-evident, but maybe for many it isn’t. If one wants to compare data on hurricanes into the 19th century it makes sense to limit the discussion to only landfalling hurricanes. Why? For many reasons:

    A) Without benefit of satellite imagery many non-land falling hurricanes in the 19th century would have been unknown. Failing to take this into account would seriously skew results in favor of saying there were many more post 1950 hurricanes. A similar thing happens when you hear of a “record number of tornadoes” striking the U.S. Well, with the tremendous increase in the number dopplar radar sites finding storms that might have never been seen by human eyes (e.g. remote rural areas, or in the dead of night), that should come as no surprise. It certainly doesn’t indicate that the number of tornadoes are increasing.

    B) Logically, there is no reason to think that the incidence of U.S. landfall should change statistically speaking from the 19th century to today. In other words, if there are more hurricanes across the Atlantic basin as a whole, there should be a corresponding increase in the number of U.S. land-falling storms. Why one would wish to believe otherwise escapes me.”

    It is nice to see work that backs up the gist of what I was claiming. (Although I’m sure my humble efforts wouldn’t pass muster in many other respects.)

  18. 10
  19. Roger Pielke, Jr. Says:


    Thanks much. You are absolutely right that this issue has been noticed (e.g., see the Solow articles cited above). And your analysis looks right on to me. What is interesting is that a number of very recent studies continue to use storm count data in uncritical fashion.

    Happy new year!

  20. 11
  21. Al Sandrik Says:


    Your Note is very interesting and I just wanted to add some personal comments if I might.

    I have done a great deal of the research on north Florida and coastal Georgia historical landfalls, some of which was utilized by the North Atlantic HURDAT Reanalysis Project. While the landfall trend you point out is pronounced for the US Atlantic Seaboard in general it is even more conspicuous for some portions of the coast. For example, let’s look at the South Atlantic Bight, and in particular the apex of the Bight in coastal Georgia. Along this stretch of coast the second half of the 19th Century was active with five landfalls (1854, 1881, 1893, & [2] 1898) and three of those were major events (1854, 1893, 1898). If we look at the 19th century as a whole, one can even go further with several likely major hurricane landfalls during the first half of the 19th century (1804 [cat 4?], 1813 [cat 3?] & 1824 [cat 3?]), and I say likely as they are currently outside the HURDAT database. Chris Landsea and I have written an Electronic Tech Memo on Georgia Coastal Landfalls which can be reviewed at…


    Now let’s contrast that “active period” with a second 100 year period from 1900 through 1999 which happens to be the period of the Jarrell et al. analysis and a large portion of the HURDAT Database. During this similar time frame we have only four direct or indirect landfalls (1911, 1940, 1947, & David “1979”) with no major landfalls! So for the 100 year period from 1800-1899 this small 111 mile stretch of coastline likely experienced six (6) direct major landfalls with none for a similar 100 year from 1900-1999.

    If we were to try to make a determination of North Atlantic Hurricane activity based solely upon the Georgia coastal activity one would tend to believe that there were a significant number of 19th Century NA hurricanes but that trend had tailed off significantly during the 20th Century to the point were hurricanes do not poise a major threat. Of course no one would jump to that type of a conclusion based upon such a limited dataset (unless they live along the Georgia coast that is, as because I fight that battle daily). I’ve looked hard for just why the landfall incidence has changed so dramatically along the Georgia coast and have come up with some good ideas but nothing I can hang my hat on. Maybe the end of the Little Ice Age has altered landfall patterns in the Bight and local shelf water SST’s may also play a role, but who knows for sure. So while the trends you point out are evident for the CONUS its really interesting if you look at smaller portions of the coast.

  22. 12
  23. Roger Pielke, Jr. Says:

    Thanks much Al for this very interesting comment!