Holland and Webster Figure 1

Holland and Webster stated the following:

Figure 1 shows a strong statistically significant trend since the 1970s similar to that found by Hoyos et al. (2006) and Curry et al. (2006.)

I’ve replicated the hurricane part of their Figure and included 2006 data as they should have – it’s amazes me that authors don’t make the effort to have up-to-date data. Here are some comments.

First, here is Figure 1 of Holland and Webster. They provide no statistical evidence that the trend is “statistically significant”, much less whether is a “strong statistically significant trend”. Is it possible that the apparent trend is simply chance? You’d think that the authors would attempt to show some statistical evidence, but I haven’t been able to locate any in the article. It may be “statistically significant”, but no evidence is provided.

I’ve been able to pretty much replicate the above graph from my collation of hurricane data through to 2006 as shown below – together with 2006 values inexcusably omitted by Holland and Webster.


Figure 2: My emulation of Holland and Webster Figure 1.

Now if this trend were “statistically significant” as claimed by the authors, then one would expect the “trend” to be discernable in the subset from 1986-2006 in the total of all basins. The figure below shows my calculations through to the end of 1986.


Figure 3: Total over all basins. Smoothed using gaussian 9-point smooth with padded mean.


  1. David Smith
    Posted Dec 29, 2006 at 4:56 PM | Permalink

    Steve, one point regarding 2006: the NHC studied their records and decided that a swirl was actually a tropical storm (news release here ). This unnamed weak tropical storm existed for all of 1080 minutes (18 hours).

    So, I think 2006 officially had 10 storms, not 9.

    I wonder how often such swirls were reanalyzed in 1950, or 1900 or 1860, using a reanalysis of QUIKSCAT satellite data from those early (1950 and 1900 and 1860) satellites. It’s an apples-to-oranges comparison.

  2. Posted Dec 29, 2006 at 5:00 PM | Permalink


    HW only discuss the Atlantic. So to be fair your 1986-2006 subset should look only at the Atlantic. Phil Klotzboch (GRL, 2006) has shown that there are no global trends since 1986, except the Atlantic.

    HW claim that there are distinct climate regimes, 1905-1930, 1931-1994, and 1995-2006 with abrupt phase changes between them. I think that it is about certain that the Atlantic is more active (by any metric) 1995-2006 as compared to 1970-1994.

    I’m not sure if you agree, but “trends” doesn’t seem to be the right approach anyway (as Jean S. has stated on another thread).

    Interestingly HW is an indication of how the hurricane/GW debate has come full circle. In fall 2005 it was often claimed that one cannot only look at the Atlantic for a GW signal as it represents about 11% of all storms. Now attention is back focused on the Atlantic.

    Also, does your Fig. 2 include 2006 in the smoothed curve?

  3. peter Webster
    Posted Dec 29, 2006 at 5:18 PM | Permalink


    One would like to include up-to-date data but the 2006 season had hardly started when the paper was submitted. One has to draw a line somewhere. Note that WHCC (2005) does not include 2005 data even though it was published in 2005. No one complained that we didn’t include this very active year. Strange, eh?

    W/r to Pielke ‘s comment: Come on Roger! We concentrated on the Atlantic becasuse the focus of the study was the Atlantic. Our aim was to work out what (if anything) had changed in the Atlantic. BTW, I still say that you can’t look at global issues by focussing on one basin. You seem to infer you can, or perhaps you can ignore global effects becasue the trend is larger in one basin. Perhaps you can but it defies me.

    Peter W

  4. Mark H
    Posted Dec 29, 2006 at 5:20 PM | Permalink

    Hmmmm, how about drawing a trend line of number of weather satelites in orbit vs. cyclones. Eyeballing the graph
    suggests a strong rising trend – so that must mean its statistically significant, no?

  5. Posted Dec 29, 2006 at 5:34 PM | Permalink


    Why the “come on”? I was asking Steve to recognize your focus on the Atlantic.

    As far as the global/NATL focus, here are two statements from you:

    (1) “One of the reasons that we had decided to do the [Webster et al. 2005] study is that we were concerned that previous studies had looked only at the North Atlantic Ocean and had made statements pertaining to global warming. This “looking locally and inferring globally” concerned us because only 12 percent of hurricanes each year occur in the North Atlantic.”

    15 Sept 2005

    (2) “Given the strong relationship between east Atlantic SST anomalies and tropical cyclone variability presented here and corroborated by several independent studies, we are lead to the confident conclusion that the recent upsurge in tropical cyclone frequency is due in part to greenhouse warming and this is most likely the dominant effect.”

    Holland-Webster, in press.

    Now there is noting inconsistent in these statements (unless the statement in (2) refers to global trends), but there is obviously the shift in focus that I mentioned. A science evolves so will perspectives, often in interesting ways …

    Now how about those actual statistical results Steve asked for?

  6. Posted Dec 29, 2006 at 6:12 PM | Permalink


    As long as you are reading and commenting, and since we both want to focus on science, I’d welcome your reactions to this substantive comment from the other thread by Willis E. (I’m sure other readers would as well!):


    Thanks much!

  7. Ken Fritsch
    Posted Dec 29, 2006 at 7:05 PM | Permalink

    Re: #3

    ..I still say that you can’t look at global issues by focussing on one basin. You seem to infer you can, or perhaps you can ignore global effects becasue the trend is larger in one basin. Perhaps you can but it defies me.

    Now I am totally confused — almost like there was an identity change that I missed.

    What is your take on the Kossin presentation linked below and his statements excerpted from it?

    Click to access Kossin102006.pdf


    The trends in the Atlantic and East Pacific intensity records are well supported by our new record. Upward in the Atlantic. Downward in the East Pacific. No upward trends were found in the West Pacific, Northern and Southern Indian, and South Pacific oceans. The vast majority (85%) of global hurricane activity takes place in these basins. Similar warming trends are found everywhere in the tropics. Why is the Atlantic behaving so differently? If the data are not good enough to accurately measure long-term hurricane behavior, then our path to understanding how hurricanes will change in a warming world must be through better physical understanding. This is our present research challenge.

  8. Posted Dec 29, 2006 at 7:47 PM | Permalink

    Relevant to this discussion:

    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.

    See also:

    Clearly both Solow/Moore and Holland/Webster cannot both be correct in their assessment of hurricane trends. I’ll be citing the Solow/Moore papers in my short discussion paper. According to Google scholar both Solow/Morre papers have been little cited.

  9. David Smith
    Posted Dec 29, 2006 at 8:42 PM | Permalink

    Re ## Hello, Dr. Webster, I’m glad you’re checking things out here as CA benefits when you audit what’s written. It’s appreciated, though that may be hard to imagine at times.

    Let me clarify my post #1. It had two goals:

    (1.) The first goal was to immediately, and quite publicly, point out to Steve McIntyre that his chart contains an error with regards to the number of 2006 storms. There is a mistake on the chart: the current 2006 storm number is 10 while the Unisys database used by Steve shows but 9. I want Steve, me and everyone to “get it right” to the extent possible, even on the minor stuff.

    When these happen, Steve acknowledges the comment, even when it is inconsequential, and makes a correction if he deems it to be material. His choice. In any case, my pointing out an error in a graph by Steve remains quite public.

    (For the record, I’ve spotted errors in hurricane papers which I’ve noted to the paper co-authors by private e-mail, and which they’ve acknowledged. I’ve made no mention of those here on CA. I can provide examples if desired. They were of no particular consequence but were the type of “cleanup” which, if I was the author, I’d like to know and avoid duplicating.)

    My point is, if I see what I think is a Steve McIntyre mistake, I point it out publicly. Roger did the same in his post #2. That is part of the spirit of this website. With others I sometimes tread more lightly.

    (2.) The second goal was to try to illustrate some of the issues with comparing modern storm count with storm count from pre-satellite and even pre-modern Weather Bureau days. Analysts in 2006 look at Quikscat and a few other pieces of data and decide that a weak rotation far at sea had tropical characteristics for a day. That would not have happened in 1950 or 1900 or 1860, yet it becomes part of the record and part of the comparison.

    It’s frustrating to me to see 1886 data, for example, compared to 2005 data.

    Anyway, those were my goals. Thanks again for posting.

  10. TAC
    Posted Dec 29, 2006 at 9:37 PM | Permalink

    #8 Roger: FWIW, the time series of counts of landfalling U.S. hurricanes (1851-2005) got some attention on CA a while back (here). It is an easy dataset to analyze — the data are consistent with a simple Poisson process, white noise error structure, no trends, no LTP.

    However, Judith Curry expressed some doubts and concerns:

    Note, the U.S. landfalling data is very confusing, since HURDAT may include multiple landfalls. Also, the U.S. landfalling data shows little correlation with total NATL stats, can’t really be used to infer anything about AGW or causes of basin or global hurricane/TS stats (this is discussed in BAMS article)

  11. Steve McIntyre
    Posted Dec 29, 2006 at 11:28 PM | Permalink

    If one compares Atlantic track locations pre-1950 and post-1950, the median longitude of an Atlantic track post-1950 is further east than the median longitude post-1950. Here’s a diagram of the median longitude of a reported track from 1851 to 2006.  This easting of reported storm tracks is probably statistically significant.  Is it a product of climate variability or of improved ocean coverage to the east? I’m not familiar enough with the data to venture an opinion, but I presume that there have been relevant technological changes in this period that could have contributed to improved coverage to the east.

  12. bender
    Posted Dec 29, 2006 at 11:50 PM | Permalink

    Interesting and relevant graphic in #11. Suggests a potential trend in reporting bias.

  13. Steve McIntyre
    Posted Dec 30, 2006 at 12:02 AM | Permalink

    The easting effect is even stronger with the 25% quantile.

  14. bender
    Posted Dec 30, 2006 at 1:26 AM | Permalink

    Why haven’t we seen a plot like this from the hurricane climatologists? I don’t want to assume this is yet another example a failure to self-criticize, but what else is one to conclude? Maybe Dr Curry or Dr Webster can answer? Or Dr Pielke Jr?

    [I realize this comment is not focused on technical issues, but it does harken back to the technical issues that got us started on hurricane climatology – the lack of confidence intervals on regression trend lines in Dr Curry’s BAMS article, and then Dr Emanuel’s (2005) binning & pinning effects (exaggerating the recent trend in hurricane PDI). I’m trying to connect some dots here. Dr Pielke used the phrase “pathology” to describe this seeming aversion to robust statistics. Maybe he’s right. This appears to be a general feature of what one might term “alarmist science”.]

  15. Posted Dec 30, 2006 at 2:13 AM | Permalink


    Can you produce the statistics of the correlation between “number of tropical cyclones” and “Eastern Atlantic SST anomaly”?

  16. David Smith
    Posted Dec 30, 2006 at 7:50 AM | Permalink

    The period 1945-1950 included the developement of storm reconaissance flights and the start of around-the-clock tropical coverage ( = staff) at the US Weather Bureau.

    Crude satellite coverage began in the 1960s and got better over time, allowing better detection and characterization of weaker systems.

    The tropical Atlantic between Africa and the Lesser Antilles was not an area of heavy ship traffic in years past and even today it is not a busy area.

    The farther back in time one goes, the harder it is to tell if a track was based on an observation or whether it was just someone’s guess.

    Here are several more “storms” from the modern era (the record 2005 season)which I believe would likely have not have been detected/reported in earlier years. These are tracks, with the yellow color being when the “storm” had 35-40 knot winds and the green being non-storm strength:

    Bret, 2000

    Jose, 2005

    Lee, 2005


  17. Willis Eschenbach
    Posted Dec 30, 2006 at 8:02 AM | Permalink


    Well, I decided to sub-sample the Holland-Webster data. Using their 9-year trailing average data so a direct comparison is possible, I took a look at the trends over sequential 50-year subsets of the data. Here are the results:

    There was one surprise to me. I was not surprised that none of the R^2 values were significant, because of the high autocorrelation of the averaged data. Nor was I surprised that the R^2 value was high for some of the subsets.

    I was surprised, however, by the fact that for about half of the data, the trends were negative. I hadn’t expected that. For the fifty years periods trailing 1978 – 2000, when the ocean warmed up, there were fewer hurricanes … go figure. Judith or Peter Webster, any comments about that?

    My best to everyone,


  18. Posted Dec 30, 2006 at 9:11 AM | Permalink

    Hi all, a few replies:

    #10, TAC, thanks for the link, I will have a look. Note that the US landfall and storm counts that I use in my short paper are not directly generated by me from the Best Track dataset, but by NOAA (see the paper for the links). The question I’d pose to hurricane climatologists is why do we observe the remarkable stability in landfall statistics over 150+ years? The point is not to say something about global warming (as Curry suggests in your excerpt) but to say something about landfalls!

    #11, Steve- Interesting! This is aspect of the data that ought to be explained before coming to conclusions based on other attributes of the dataset.

    #14, Bender- Andrew Solow has this explanation in his 2003 paper “Statistics and Atmospheric Science”:

    “As in other fields of application, there has been an ongoing tension between the level of statistical sophistication favored by statisticians working in this area and the practical requirements of atmospheric scientists for whom statistics is not the main object of interest. A particularly gratifying development of the past few years has been a general relaxation of this tension.”


    Well, my own personal experiences with some in the hurricane community and some of what I see in Steve’s broader experiences over the past few years suggest that this tension is not as relaxed as we might wish. Things are very different in my experience in the hydrology community. The very public and political nature of climate science raises the stakes even more. This is probably the main reason why Science magazine recently highlighted climate science as one area where submissions would get a higher degree of scrutiny. IMO a site like Climate Audit can serve a very valuable role of putting some sunshine on some of these studies.

    I would encourage more scientists to put pre-publication, pre-submission papers online for such commentary. The professional downside is missing out on the big splash of an embargoed Science/Nature paper, and I suppose that there are the trolls to deal with and the possibility of being shown publicly to have made a mistake. Of course the upside is better science. If I had a dollar for every time over the past few years that I’ve heard climate scientists say that they couldn’t share their work because of their hopes for publication in Science/Nature . . .

    #17 Willis- Great graph. This is exactly the sort of comprehensive analysis that is needed in more of climate science in my opinion. This will help to place data analyses into broader context and allow readers to identify statistical cherrypicking.

    What your graph shows is something well recognized in the community, namely the period 1970-1994 was different. Things changed in 1995. There are now a range of different explanations for this and clearly there is no strong consensus on any of them.

  19. Posted Dec 30, 2006 at 9:13 AM | Permalink

    #11, #17

    Steve and Willis- I’d like to add these figures to my files for possible use in presentations. Would that be OK? I will of course credit directly to you and Climate Audit, unless you tell me differently. Let me know … Thanks!

  20. J. Curry
    Posted Dec 30, 2006 at 10:15 AM | Permalink

    Re the graph in #17. Willis, in your graph it look like you used hurricanes rather than total TC’s? before getting too excited about this graph, we once again have the problem of doing statistics without considering the physics. The 50 yr sampling period cuts partway through a big natural internal oscillation. Peak of AMO ca 1950, decreasing TC activity. The AMO started turning around in the late 80’s. The very substantial activity since 1995 reflects the ascending mode of the next AMO cycle (which is projected to peak ca 2020) plus AGW (Chris landsea and others argue that this is just the positive phase of the AMO).

    I have argued that the NATL signal from AGW seems to be associated with a change in frequency distribution of tropical cyclones: more tropical storms, and more cat45. little change in the number of hurricanes, and an apparent decrease in cat3. Apart from data quality issues associated with TC intensity, the increase in cat45 with decrease in cat 3 could simply mean that once conditions are sufficiently favorable for cat3 (wind shear and all that), then there is a greater probably of increasing to cat45 owing to increasing SST. the larger number of tropical storms seems to be associated with increased season length (more storms in non peak months, that don’t intensify much)

    Again, the statistical analyses are interesting, but i’ve seen nothing here that changes my interpretation on the NATL time series (as described in previous two paragraphs).

  21. Francois Ouellette
    Posted Dec 30, 2006 at 10:31 AM | Permalink

    #11 Steve,

    Standard deviation of the median distribution post-1950 also seems larger than pre-1950. This could be consistent with a “wider coverage” hypothesis. Do the actual numbers confirm the eyeballing? Also, it would be interesting to plot the actual standard deviations of the annual data along with the median, and see if there is a trend.

  22. bender
    Posted Dec 30, 2006 at 10:38 AM | Permalink

    Re #17

    I was surprised, however, by the fact that for about half of the data, the trends were negative. I hadn’t expected that.

    I noticed that myself last night. Look back at the time series and you can spot the 50y windows where the series are negatively correlated. I don’t think it’s an interpretible result. It is the kind of thing that happens when a correlation is dominated by a very few regime changes (of which there are two in this system).

  23. Posted Dec 30, 2006 at 10:53 AM | Permalink

    #20 Hi Judy-

    This exchange might be a good example of the “tension” that Solow describes between statistics and atmospheric science. Willis’ graph is significant not for what it shows about a 50-year trend, or even anything to do with the climate system, but because it shows how easy it is to comprehensively present how the “trend” changes over time. If you don’t like 50 years and hurricanes, use 9 years and TCs, or whatever. The point as I take it is that being comprehensive in presentation of “trends” is far more robust than being selective. More specifically, the entire notion of “trends” might be misplaced in this context.

    On the physical basis for observations, as you imply, the data is itself underdetermined — meaning than any of a number of physically-based processes might be used to explain the data. Since everyone seems to agree, at a general level, what coming years, or decades will see, it is unlikely that the arguments about the physical processes will be resolved anytime soon based on predictions, unless someone really sticks their neck out and is proven correct.

  24. J. Curry
    Posted Dec 30, 2006 at 11:55 AM | Permalink

    Here is some text from from an in press book chapter (Curry and Webster), that discusses the range of what NATL TC activity would look like ca 2025 in an average sense (near the peak of the current AMO cycle). I have presented this in numerous public presentations as well. So I have stuck my neck out, and considered the known uncertainties. Here is the rationale of my “forecast”, which encompases my understanding of the NATL TCs:

    Projections of Future Hurricane Activity

    While both groups of scientists (those that support the natural variability explanation and those that support the global warming contribution) agree that hurricane activity in the North Atlantic will remain elevated for some years, the implications for future projections of hurricane activity are quite different. Based upon the hypothesis of natural variability being the cause of the high hurricane activity in the North Atlantic since 1995, there have been several predictions of a forthcoming downturn in hurricane activity: Goldenberg et al. (2001) imply a downturn in 10-40 years; and Gray (2006) anticipates a downturn in 3-8 years associated with a global cooling.

    If our hypothesis is correct that greenhouse warming is causing an increase in hurricane intensity globally and also an increased number of storms in the North Atlantic, what does this imply for future hurricane activity as sea surface temperatures continue to rise and the oceanic warm pool continues to expand, especially in the spring and fall? The following analysis addresses specifically the projection of North Atlantic tropical cyclone activity. We consider a range of projections from two different approaches: climate model sensitivity to increasing greenhouse gases, and simplistic projections based upon the historical data record. A projection is made for average conditions in the year 2025 (such that high frequency fluctuations from short term oscillations such as El Nino are ignored), corresponding to an increase in tropical SST of 1oF that is attributable to greenhouse warming.

    Webster et al.’s observations scale to a 6% increase in maximum wind speeds for a 1oF SST increase. By contrast, high resolution climate model simulations (Knutson and Tuleya 2004 and Oouchi et al. 2006) have found a 2% increase in intensity when scaled for a 1oF SST increase, which is a factor of 3 times smaller than that determined from the observations. Oouchi et al. also found that the number of North Atlantic tropical cyclones increased by 30% for a 2.5oC increase in SST, which scales to an increase of 1 tropical cyclone per 1oF increase in SST. By contrast, based upon the historical data record in the North Atlantic, an increase of 1oF in tropical SST implies an additional 5 tropical storms per season (Figure 8), which is a factor of 5 greater than the number inferred from climate model simulations.

    Projections of future hurricane variability must include both natural variability and greenhouse warming. Estimates of the magnitude of the impact of the Atlantic Multidecadal Oscillation (AMO) on the total number of tropical cyclones per year range from 0 (no effect) to 4-6 (the AMO explains the entire magnitude of the trough to peak variability in Fig. 1). Assuming that the AMO continues in the current cycle with approximately a 70-year periodicity, the peak of the next cycle is expected in 2020 (70 years after the previous 1950 peak), so 2025 is very near the peak of the AMO cycle. Proponents of the natural variability explanation refer to active and quiet phases rather than actual cyclic behavior; their analysis states that we are currently in an active phase that will last 10-40 years, and there is no implicit assumption that the level of activity in 2025 will be higher than the activity of the past decade.

    Based upon these assumptions, consider the following simple statistical model. The average number for the past decade of total North Atlantic tropical cyclones is 14.4. We assume that the effects of greenhouse warming and the AMO are separable and additive. Table 2 compares the simple statistical projections including both greenhouse warming (AGW) and AMO, AGW only, and AMO only. The combination of AGW+AMO would result in the greatest elevation in the number of named storms, and an unprecedented level of tropical cyclone activity. The range of the different assumptions ranges from an increase of 0 to 6.5 named storms per year. In terms of the intensity of the storms, Fig 1 suggests that the distribution of the storm intensity is changing with warming, whereby the increase is in the number of tropical storms and in the number of category 4+5 storms (NCAT45), rather than in the weaker hurricanes. Consideration of U.S. landfalling hurricanes (Fig 4) suggests a continued increase over the next two decades in the ascending mode of the AMO, which when combined with AGW may result in an unprecedented number of U.S. landfalling hurricanes. Once the AMO begins its descending mode ca. 2020, continued warming makes it doubtful that we will ever again see the low levels of hurricane activity of the 1980’s and we can expect a leveling off rather than significant decrease in activity until the next ascending phase of the AMO.

    Table 2. Projections for the average total number of North Atlantic tropical cyclones (named storms) for 2025. AGW refers to anthropogenic greenhouse warming; AMO refers to Atlantic Multidecadal Oscillation.
    (note couldn’t figure out how to get the table into the post, here is the essence)

    AMO only: + 0 to 1.5
    AGW only +1 to 5 (e.g. Emanuel)
    AMO + AGW: +1 to 6.5 (Curry/Webster)

    starting from 14.4 TC average yields projections for 2025 ranging from 14.4 to 20.9 TCs on average

    The projections in Table 2 show a broad range and any projections of future hurricane activity must be viewed as highly uncertain at this time, and it may take a decade or two before it is clear which of these models has the better predictive capability. Nevertheless, this range of projections provide some broad constraints on the conceivable elevation of North Atlantic tropical cyclone activity in coming decades.

  25. Posted Dec 30, 2006 at 1:36 PM | Permalink


    Thanks! This is indeed a bold set of predictions. I do wonder about the basis (statistical or physical) that you have for predicting the possibility of “an unprecedented number of U.S. landfalling hurricanes.” Unprecedented would be 8 or more in a single season.

    Of the 10 seasons historically with 15 or more total named storms in the record, they have averaged 3 landfalls per year, with a minimum of 0 and maximum of 6 (twice). Over the past 30 years (or so) the 10-year average proportion of storms making landfall has been between 10-20%. If this holds in the future then for your predicted range of 14.4-20.9 storms per year we should expect between 1-4 landfalls per year (rounded). To reach “unprecedented levels” would require a landfall rate of 40-60%, which has heppened within an individual season only 2 times in the past 75+ years. Over a ten-year period unprecedented landfall amounts levels would require 25 landfalls, or 2.5/year, or at observed landfall rates, for an average storm frequency at the very high end of your range.

    Can you explain a bit further the basis for expecting increasing landfalls?


  26. Steve McIntyre
    Posted Dec 30, 2006 at 1:45 PM | Permalink

    Judith, I personally think that it’s quite plausible that the number of Atlantic hurricanes have been increasing in the past century and that this increase is related to increased SST (as to the relative levels of medieval or Holocene Optimum hurricanes, that’s a different issue.)

    Having said that, this article is a lot like John Hodgman’s skit – appropriately about a hurricane meterologist (minoring in crisp fall days) – asserting something  in forceful tones as though that made it true.
    If you look at red noise series taken from persistent data of the type that TAC likes, you can easily find “regimes” and “regime change” – but these are only metaphors. HW spend lots of time denying the relevance of various oscillations but exactly how does a “regime” differ from an “oscillation”? There are attempts in econometrics to analyse whether there has been a “regime change”. Without vouching for any of these methodologies, someone who ventures into this turf should at least familiarize themselves with the issues.

  27. Steve McIntyre
    Posted Dec 30, 2006 at 2:06 PM | Permalink

    #19. Roger, of course. HEre’s a similar graphic for the median latitude of Atlantic storm tracks. There seem to multidecadal variations in the median latitude, though these “regimes” are different than regimes TC1,TC2 and TC3 -whatever they are.  BTW visually this seems to be a very spiky and antipersistent series – why would that be?

  28. J. Curry
    Posted Dec 30, 2006 at 3:19 PM | Permalink

    Two comments:

    Roger, the number 8 in my “forecast” refers to TCs, not hurricanes

    Steve, on the other thread I said that I was not defending the regime change thing, that is Holland’s idea
    Here is what I said:
    “Re Pat Michaels, i don’t believe that I have ever been publicly critical of him. In fact i think his paper has some useful points (I referenced it in my BAMS article). In particular he talks about SST thresholding for hurricane intensity. Rather than Holland’s regime change, i prefer either the SST thresholding and/or a combination of AMO+global forcing (including AGW) as contributing to the explanation for the lower frequency (non el nino) variations in the TC time series”

  29. Posted Dec 30, 2006 at 3:42 PM | Permalink


    Thanks, but your draft text refers explicitly to landfalling hurricanes:”Consideration of U.S. landfalling hurricanes (Fig 4) suggests a continued increase over the next two decades in the ascending mode of the AMO, which when combined with AGW may result in an unprecedented number of U.S. landfalling hurricanes.”

    Hence my focus in comment #25 is on landfalling hurricanes, and 8 is the correct number for an unprecedented level — this does not refer to all TCs.

    Can you explain a bit further the basis for expecting unprecedented hurricane landfalls?


  30. J. Curry
    Posted Dec 30, 2006 at 3:54 PM | Permalink

    My “forecast” is for average TC counts. 8 as an average number of landfalling TC counts would be unprecedented in the record since 1851, although we have seen 8 TCs in individual years, most recently in 2004, 2005, and also 2002 and 1985. What I am saying is that these extreme years could become the average ca 2025. These years with 8 landfalling TCs were associated respectively with 6, 6, 4, 6 landfalling hurricanes. An average of 5.5 landfalling hurricanes would be unprecedented.

  31. Posted Dec 30, 2006 at 4:09 PM | Permalink


    I must have missed something as I don’t see anything in your excerpted text in #24 on landfalling TCs. Nonetheless, my question stands regarding landfalling hurricanes.

    FYI 2002 had only 1 landfalling hurricane.

  32. Tim Ball
    Posted Dec 30, 2006 at 4:20 PM | Permalink

    Re #20
    The classification from 1 to 5 is based on wind speed. I understand today for ocean based hurricanes this is primarily achieved by a formula converting wind speeds recorded by aircraft flying through the storm. As I recall Katrina was reported as 5 before it made landfall based on one such measure, but when it came ashore was barely a 3. How was wind speed determined in offshore hurricanes prior to aircraft flights? When did the flights begin? Was the determination of wind speed standard throughout the period in question?

  33. Steve McIntyre
    Posted Dec 30, 2006 at 5:44 PM | Permalink

    Here are a few more graphics. The first graphic shows the worldwide count of tropical cyclones (calculated on the same basis as HW Figure 1), showing a rather stable global total. Red- ATL; Cyan- NIO; blue = EPAC; purple – WPAC; black – SH. I’ve plotted ATL and NIO adjacent to one another since there is an interesting multidecadal antithesis between the two, which I’ve not seen mentioned anywhere. Yes, ATL storms have increased, but NIO storms seem to have decreased. Is there any significance to this – I don’t know. But the decline in NIO storms seems every bit as “significant” (or not) as the increase in ATL storms.

    Next here is the same thing for hurricanes. What do we notice – well, for a start, the graph shows a significant increase in SH hurricanes in the mid-1980s without there being a comparable increase in storms – most notably in SH. This is almost certainly just an artifact of data coverage. Is there a trend in NH hurricane counts? Doesn’t look like it to me. Also NIO results have the same problems as SH data. NIO hurricane counts don’t change commensurately with the storm information – is this a data artifact or something in the climate? Who knows. Is it possible that the sum of NIO and ATL hurricanes remains proportional to the sum of NIO plus ATL cyclones? Seems like a plausible hypothesis and not ruled out by this data.
    You can see where a biased choice can crop up – if Webster or Curry say: we don’t have adequate hurricane data for the NIO so we’ll exclude it, the reasoning is understandable but not necessarily immune from bias (or data snooping). The storm count information certainly suggests an opposite trend in the NIO. One cannot avoid feeling that, if they wanted to use data from this region (say ATL and NIO coverage were reversed), then they might have constructed some plausible estimation procedure. In the period since 1986 when Gray indicates that more consistent statistics begin, one sees the lack of trend that characterizes the count information back to 1970.

    Finally here is the same graphic for Cat 4 hurricanes (note the change of vertical scale). Just looking at it, one feels that the SH (black) Cat 4 information probably has artifacts in it. We know that the NIO coverage in the 1970s was just as bad as the SH coverage. Were there some Cat 4 hurricanes in the NIO in the 1970s. Given that storm counts were then as high as recent ATL counts, it seems possible to me. The blue here is EPAC. I get the impression of a sort of anticorrelation between EPAC Cat 4 counts and ATL Cat 4 counts. The sum of EPAC+ ATL Cat 4 storms doesn’t show a trend that I would describe as “strongly significant” merely by eyeballing it. So how can there be a “strongly significant” ATL trend?

    PS – I’ve not spent a lot of time with this data and haven’t triple-checked it. I’ll continue reviewing the data sets.

  34. guest
    Posted Dec 30, 2006 at 5:46 PM | Permalink

    Another data oddity:

    Look at the subset of reported storms which hit land in Mexico or Central America

    27 storms reported as having landfall in Mexico/Central America during 1860-1899

    49 storms reported as having landfall in Mexico/Central America during 1955-1994

    Difference of 22 storms.

    Higher overall reported activity (1955-94 vs 1860-99) would account for about 6 of those storms.

    What accounts for the other 16?

    a. Path shift
    b. More likely, underreporting of storms hitting Mexico/Central America during the 19’th century

  35. J. Curry
    Posted Dec 30, 2006 at 6:05 PM | Permalink

    TCs: A quck glance at the 2002 TC tracks, i see the following
    Hurricane Lili
    Hurricane Kyle
    Hurricane Isadore
    TS Hannah
    Hurricane Gustav
    TS Fay
    TS Eduard
    TS Bertha
    TS Arthur

    This is 9. So I am not sure exactly which 8 were included in my landfall dataset (I don’t have all this documentation with me). But 2 GT students, James Belanger and Mark Jelinek, spent all last summer checking the TC data to put this data set together. In fact, their work resulted in many 10s of pages of proposed corections to the HURDAT data set. Their proposed corrections up to 1914 have been accepted by the HURDAT committee (their names clearly appear in the HURDAT documentation), I presume the HURDAT committee is considering the rest of the proposed corrections

    So if anyone wants my landfall dataset (the version i have with me (just has yearly counts of TC and hurricanes), i can email the file to steve and he can publish the dataset in the journal of climateaudit.

    Table 2 in its entirety could not be copied into the text (given my inability to deal with doing anything fancy like that on the site.) the forecast included a range of 5-8 TCs as an average for ca 2025. I’ve explained how this translates into unprecedented landfalling hurricane activity (using the analogue method). I’m not going to split hairs further here over the wording.

    If anyone has suggestions on how to improve a forecast llike this, I am all ears. My other forecast is that by 2025 I will be retired but probably still alive, remains to be seen whether anyone remembers this forecast or even cares at that point. That is the problem with climate, it varies slowly and it takes a long time to verify any predictions.

  36. Posted Dec 30, 2006 at 6:24 PM | Permalink

    Judy- Yes, please do send your dataset to Steve for posting, it’ll be a lot easier to discuss, as your numbers presented here don’t jibe. 2002 had 1 hurricane landfall, not 4: http://www.nhc.noaa.gov/tracks/2002atl.gif


  37. J. Curry
    Posted Dec 30, 2006 at 6:36 PM | Permalink

    Re NATL Central America landfalls, the counts should have the same reliability as the U.S. landfalls back to 1944 (prior to that, who knows). from 2001-2005, there were 31 Central American landfalls. Prior to 2001 (back to 1950 anyways) the previous periods with high counts were
    1951-1955: 23
    1955-1970: 24

    The signal for increasing landfalls seems higher in central america than for the U.S. The reason for this is that there is an increasing trend for a greater percentage of the NATL storms to enter the Gulf (Holland discusses this). whether this is AMO or AGW or both I don’t know (but i suspect it is both).

    Note: my central american landfall dataset was prepared by GT students Brandon Foskey and Paula Agudelo.

  38. Willis Eschenbach
    Posted Dec 30, 2006 at 7:26 PM | Permalink

    Well, re my graph, I can see I’m not too clear in my writing. My thanks to everyone for their comments. A few notes:

    1) Roger P., you say:

    What your graph shows is something well recognized in the community, namely the period 1970-1994 was different. Things changed in 1995. There are now a range of different explanations for this and clearly there is no strong consensus on any of them.

    Respectfully, that’s not what the graph says. What the graph shows are the 50-year trailing regression statistics. Thus, for the 50-year period from 1930-1980, when the temperature warmed by a degree, on average there were two fewer hurricanes. This has very little to do with 1970-1994 period, and everything to do with the 1930-1980 period.

    2) Judith C., you say:

    Re the graph in #17. Willis, in your graph it look like you used hurricanes rather than total TC’s? before getting too excited about this graph, we once again have the problem of doing statistics without considering the physics. The 50 yr sampling period cuts partway through a big natural internal oscillation. Peak of AMO ca 1950, decreasing TC activity. The AMO started turning around in the late 80’s. The very substantial activity since 1995 reflects the ascending mode of the next AMO cycle (which is projected to peak ca 2020) plus AGW (Chris landsea and others argue that this is just the positive phase of the AMO).

    Again respectfully, it seems that you are doing physics without considering the statistics, or perhaps, as I said, I have not made myself clear. I have updated the graph labels to avoid misinterpretation. The “Slope” in the graph (which I had called “Trend” is not the trend in the number of hurricanes. It is the slope of the regression line, and measures the change in the number of cyclones per °C change in East Atlantic SST. Regardless of the state of the AMO, if the hypothesis “warmer sst means more hurricanes” is true, we should see that relationship at all times. As we go from say a warm phase of the AMO to a cold phase, the number of hurricanes should decrease, and vice versa. After all, the AMO is nothing more than a long-period cycle in the SST. Why should that long-period SST cycle have the opposite effect on hurricanes as a shorter term SST cycle?

    What my graph shows is that there are periods of time when this hypothesized increase with SST is not happening, in fact, the reverse is happening. Is this related to the AMO? Possible, I suppose, but it would have to be related to some secondary aspect of the AMO other than SST. Don’t know … lemme look at a longer time series … back in a couple of hours …

    … OK, thanks for waiting. Here’s a longer version of the same graph, without the “p” values, and including a 50-year trailing average of the AMO.

    Is there a relationship? Still don’t know …


  39. John S
    Posted Dec 30, 2006 at 7:26 PM | Permalink

    Some references for those interested in regime changes…

    Bai, J. and P. Perron (1998), “Estimating and Testing Linear Models with Multiple Structural Changes,”Econometrica, 66, 47-78.

    Bai, J, and P. Perron (2003), “Computation and Analysis of Multiple Structural Change Models,” Journal of Applied Econometrics, 18, 1-22.

    These represent an advance over earlier approaches because the number of breaks does not need to be specified a priori and the break points are, likewise, determined by the data rather than having to be imposed (like good old Chow tests of old). The tests are pretty simple (sup F) but it’s the critical values that are the tricky bit.

    However, as alluded to above, it is virtually impossible to distinguish between integrated (or near integrated series) and stationary series with regime breaks in practical applications – the tests have virtually no power. To make that distinction you need to make an argument from first principles of the systems rather than any statistical tests.

  40. J. Curry
    Posted Dec 30, 2006 at 7:37 PM | Permalink

    Here are the reasons I like to use TC counts instead of hurricanes, even for landfalls (although i realize that if you are interested in damage, then the intensity implied by a hurricane is more relevant)

    The HURDAT dataset contains a huge number of inconsistencies. The best track data set provides track data with 6 hour increments, however it actually classifies landfall strength based on readings not provided within the data set that occurs within the 6 hour increments and may vary in strength from the track readings. In some instances there are substantial discrepancies, (TS vs hurricane designation, or 2 categories difference in hurricane strength). The Belanger/Jelinek comments submitted to hurdat identified tons of these kinds of discrepancies. While these discrepancies are most numerous for the early part of the record, they found 4 discrepancies from 95-05.

    Further, TCs do flukey things once they approach land, they invariably diminish in strength. however, before landfall (particularly for gulf strikes), much damage may occur in the forward right quadrant of the storm while a storm is still at hurricane strength, and it may degrade to TS strength when the eye actually crosses land.

    My landfalling data set was motivated by understanding the causes of the landfalls rather than correlating damage to the landfalls. therefore a hurricane in this dataset is counted as a hurricane even if it degraded to a TS before landfall.

    So this is the source of the discrepancy between my 4 vs your 1 2002 landfalling hurricanes. 2002 is NOT one of the years where Belanger/Jelinek found discrepancies in the landfalling data set. when the eye crossed land, only 1 2004 TC was a hurricane. (the discrepancies between the two methods of counting hurricanes is typically much smaller than in 2002, and is dominated by discrepancies between the best track 6 hourly designation and the landfall designation)

    And this is why I think total TC is a much more useful designation than hurricane when trying to sort out what is going on with the landfalls in terms of causal mechanisms.

  41. J. Curry
    Posted Dec 30, 2006 at 7:55 PM | Permalink

    Willis, I don’t know how to post plots here, but i can describe one that agudelo and hoyos made and perhaps you can reproduce it. They standardized the SST and TC timeseries back to 1851. They then created a 15-25 year filtered time series. Then they did a sliding correlation on the filtered time series. The sign of the TC SST correlation changes sign as follows:

    1850-1875: +
    1870-1910: –
    1920-1950: +
    1950-1980: –
    1980-present: +

    the correlation seems to change sign at the peak and bottom of the AMO. This seems to be what you are picking up also.

    again, my point is that the correlation is frequency dependent:
    el nino scale: negative
    15-25 years: switches sign
    greater than 25 years: positive

    the greater than 25 years stuff is AMO + AGW

    if anyone can give me instructions for posting plots (i use a mac), i will try to post (bender gave me instructions some time back, but i couldn’t get them to work).

    The AMO seems to excite some sort of subcycles, although at ~ 20 yrs there is conceivably a solar element. Who knows what is going on here, but I think what you found is the same thing that agudelo and hoyos found.

  42. Willis Eschenbach
    Posted Dec 30, 2006 at 8:09 PM | Permalink

    Judith, thanks for the information. Is the list of Atlantic TCs that you use available on the web? Also, there are a variety of definitions of the ACR (25-5°N, 55-20°W), etc. Which one do you use, and why?

    To post images here, you need to first have them on the web. Then, in your post, put a line of the form:

    lessthan img src=”http://homepage.mac.com/williseschenbach/.Pictures/holland_webster_50_yr_stats_long.jpg” style=width: 800px; height: 600px;” / greaterthan

    The section in quotes is the URL where the image is located, replace it with the exact URL of your desired image. Replace the lessthan and greaterthan with the corresponding symbols (I can’t type them, or the post will get munched).

    Since this is just text, it doesn’t matter whether you’re on Mac or PC, I use a Mac myself.

    All the best,


  43. J. Curry
    Posted Dec 30, 2006 at 8:26 PM | Permalink

    here is my U.S. landfall data set. note hurricanes denotes that the storm was hurricane strength at some point prior to landfall or at landfall

    Year Named Storms Hurricanes
    18513 2
    1852 3 3
    1853 1 1
    1854 3 3
    1855 1 1
    1856 3 2
    1857 2 2
    1858 1 1
    1859 2 1
    1860 3 3
    1861 4 3
    1862 0 0
    1863 2 0
    1864 0 0
    1865 4 2
    1866 2 2
    1867 2 2
    1868 1 0
    1869 4 4
    1870 3 3
    1871 6 4
    1872 2 1
    1873 4 2
    1874 3 1
    1875 2 1
    1876 2 2
    1877 3 2
    1878 4 3
    1879 6 4
    1880 5 4
    1881 4 2
    1882 4 3
    1883 1 1
    1884 1 1
    1885 5 3
    1886 7 7
    1887 6 4
    1888 6 4
    1889 3 2
    1890 1 0
    1891 3 2
    1892 3 0
    1893 7 6
    1894 3 2
    1895 4 1
    1896 4 4
    1897 4 2
    1898 5 3
    1899 5 3
    1900 3 1
    1901 6 3
    1902 3 2
    1903 2 2
    1904 3 2
    1905 2 0
    1906 5 4
    1907 3 0
    1908 3 2
    1909 7 5
    1910 2 2
    1911 2 2
    1912 5 2
    1913 3 3
    1914 1 0
    1915 4 3
    1916 8 6
    1917 1 1
    1918 2 1
    1919 2 1
    1920 3 3
    1921 2 2
    1922 1 0
    1923 4 1
    1924 3 2
    1925 2 1
    1926 4 4
    1927 1 0
    1928 3 3
    1929 2 2
    1930 1 1
    1931 2 0
    1932 5 2
    1933 7 5
    1934 5 3
    1935 2 2
    1936 7 3
    1937 4 0
    1938 4 2
    1939 3 1
    1940 4 3
    1941 4 2
    1942 3 2
    1943 4 3
    1944 4 3
    1945 4 3
    1946 4 2
    1947 7 3
    1948 4 3
    1949 4 3
    1950 4 4
    1951 1 1
    1952 2 1
    1953 7 3
    1954 4 3
    1955 5 3
    1956 2 1
    1957 5 1
    1958 2 2
    1959 6 4
    1960 4 2
    1961 3 2
    1962 1 1
    1963 1 1
    1964 5 4
    1965 2 1
    1966 2 2
    1967 2 2
    1968 3 2
    1969 3 2
    1970 3 1
    1971 5 3
    1972 2 1
    1973 1 0
    1974 1 1
    1975 1 1
    1976 2 1
    1977 1 1
    1978 2 0
    1979 5 3
    1980 2 1
    1981 2 1
    1982 1 0
    1983 2 1
    1984 2 1
    1985 8 6
    1986 2 2
    1987 2 1
    1988 4 1
    1989 4 3
    1990 1 0
    1991 1 1
    1992 2 1
    1993 2 1
    1994 3 1
    1995 5 3
    1996 4 2
    1997 1 1
    1998 7 4
    1999 5 4
    2000 2 1
    2001 3 1
    2002 8 4
    2003 4 2
    2004 9 6
    2005 8 6

  44. J. Curry
    Posted Dec 30, 2006 at 8:34 PM | Permalink

    Willis, for NATL SST, Agudelo and Hoyos used the region as per the Webster et al. (2005) paper. also, i can’t upload files to web (i pay people to do that kind of thing for me, but unfortunately they are off duty righ now)

  45. Willis Eschenbach
    Posted Dec 30, 2006 at 9:17 PM | Permalink

    Many thanks, Judith. Do you have the number of landfalling named storms and hurricanes for 2006?

  46. J. Curry
    Posted Dec 30, 2006 at 9:37 PM | Permalink

    Willis, for 2006 i count 3 U.S. landfalling TCs. one of these, Ernesto, was a hurricane before landfall. so consistent with the rest of the time series, it would be 3 TCs, 1 hurricane

  47. jae
    Posted Dec 30, 2006 at 10:00 PM | Permalink

    The problem I have with the “Hurricane Team” is that the effects are always attributed to AGW, and not just GW. We are coming out of the LIA, for crying out loud. I have to wonder about the motivation for this constant reference to AGW. Could it be related to funding? Could it be related to “belief?” Could it be related to being a syncophant? Their science would be a lot more credible, IMO, if they dropped the A.

  48. Posted Dec 30, 2006 at 10:18 PM | Permalink


    Thanks much for posting your landfall data. Just FYI in 38 years (of 154) your dataset differs from that published by NHC for hurricane landfalls. In every case in these 38 years your dataset has more landfalls.

    I think that the differences arise due to an overcount in your dataset (or at least a significantly different definition of hurricane landfall than commonly used) based on this statement: “note hurricanes denotes that the storm was hurricane strength at some point prior to landfall or at landfall.” Including storms at hurricane strength “prior to landfall” allows storms that weaken long before striking the coast to be included.

    Take 2002 as an example. There were only 4 storms total that were hurricane strength at some point during that season. Your dataset has each one recorded as a hurricane landfall. However, only one actually made landfall at hurricane strength. Two were at hurricane strength long before they approached the US coast as weakened systems.. The fourth became a hurricane far out to sea after departing the US coast.

    Similar overcounts appear for 2001: http://www.nhc.noaa.gov/2001.html
    and 1999: http://www.nhc.noaa.gov/1999.html
    and 1998: http://www.nhc.noaa.gov/1998.html

    In 2006 Ernesto made landfall in NC with maximum winds of 60 kts, below hurricane strength. It was a hurricane before striking Haiti. It was not a US hurricane landfall.

    You have 320 storms making landfall as hurricanes and NHC has 280. If the differences in the four years above are indicative of the other 34 years, then your dataset differs from the one I am using (produced by NHC) by 14%. This could be enough to make a difference in an analysis. At a minimum any analysis based on counting storms as being at hurricane strength at landfall but which in reality are far weaker systems is likely to be misleading to those of us who focus on landfalling storms.

    Three questions:

    Why not simply use the NHC dataset?

    Is your prediction for 2025 based on TCs that make landfall which had been at one time at hurricane strength, but not necessarily at the time of landfall?

    Do you use a similar convention for counting named storm landfalls?

    Thanks much!

  49. Posted Dec 30, 2006 at 11:16 PM | Permalink

    Re: #38, Willis-

    Thanks for the additional graph, very interesting. You write:

    Thus, for the 50-year period from 1930-1980, when the temperature warmed by a degree, on average there were two fewer hurricanes. This has very little to do with 1970-1994 period, and everything to do with the 1930-1980 period.

    For the 50-year period 1930-1980 you have 40 years of active hurricane seasons (1930-1969) and 10 years of inactive (1970-1980). So yes, you see a trend of fewer hurricanes across these periods. This is well established (Landsea et al. had a paper on this in 1994). Scientists disagree as to why this pattern has occurred (understatement;-); explanations include natural multi-decadal variability and also the effects of the U.S. Clean Air Act.

    Your second graph (in #38) clearly shows inflection points in the slope at about 1970 and 1994.

    Happy New Year!

  50. J. Curry
    Posted Dec 31, 2006 at 10:28 AM | Permalink

    Roger, the NHC landfalling data set is fraught with errors, as 10s of pages of documentation written by Jelinek and Belanger has shown. The more reliable number is number of TCs. I stated that very clearly several times in previous posts. The definition of hurricane in my landfalling data set is a storm that was once hurricane strength whose eye crossed land when it was at least at TS strength. My reasons for doing this were stated in the previous post. Because of these issues, my main focus is on landfalling TCs.

  51. J. Curry
    Posted Dec 31, 2006 at 10:55 AM | Permalink

    Roger, the “official” landfalls data set has numerous errors of inconsistency, including the inclusion of subtropical and extratropical storms, inconsistent designation of landfall relative to the best tracks data, and inconsistent

    For some examples from the early part of the record, see


    1908 / Storm 5 — Best track data for 09/01 shows a track that makes landfall with TS force winds over mainland NC and continues over the outer banks with the same TS level winds of 45kt. Recommendation is to adjust the header to indicate a landfall or adjust the best track data for lower winds and/or move the storm farther out in the Atlantic.

    Response (from Landsea, committee concurs): Agreed. This system should be listed as a tropical storm landfall for North Carolina.

    The more interesting discrepancies arise from cases where we clearly have sufficient data, but inconsistent info is included in the data. From list of comments submitted to hurdat, but no response as of yet, an example:

    2005 / Storm 15 — Currently the HURDAT header shows both XING=0 and SSS=1 and the trailer shows an impact of NC1. The recommendation is to adjust the XING=0 designation to XING=1, bringing inline the various HURDAT designations for this storm. (note from JC, i think this is ophelia, which they did not include as a landfall)

    1982 / Storm 2 — Currently, the HURDAT header listing for this storm has XING = 1 implying that this storm struck the U.S. as either a tropical storm or hurricane. However, the HURDAT trailer lists the maximum intensity of this system as a subtropical storm (SS). Therefore, the recommendation is to either change XING = 0 or change the intensity designation within the best track data to tropical cyclone prior to U.S. impact and the designation in the trailer to TS.

    1976 / Storm 1 — The current listing in the HURDAT header for this storm has XING = 1 implying that this storm struck the U.S. as either a tropical storm or hurricane. However, the HURDAT trailer list this system as a subtropical storm (SS) and the best track data designate the storm as subtropical throughout its entire lifespan. Therefore, the recommendation is to either change XING = 0 or change the designation within the best track data to non subtropical prior to U.S. impact and the maximum intensity designation in the trailer to TS.

    1972 / Storm 2 — Currently, the HURDAT trailer for this storm designates a category one impact for NY and CT. However, prior to landfall on 6/22 at 12Z and 18Z, maximum sustained winds are listed at 50 and 55 kts with a minimum pressure of 980 mb, and after landfall on 6/23 at 0Z, maximum winds are listed at 45 kts. Therefore, the recommendation is to either remove the impact designation for both NY and CT, since this storm had winds of only a tropical storm, or increase the winds prior to or at landfall to category one strength. Given the pressure-wind relationship with a surface pressure of 980 mb, it is very likely that the storm was a category one hurricane before landfall.

  52. guest
    Posted Dec 31, 2006 at 12:04 PM | Permalink

    Re #27 If there’s a way to filter out latitudes above 30N then the graph might show something interesting

    Storms travel westward below 30N and then recurve eastward above 30N. By looking at below-30N data one might see if the pre-curvatures travels of storms are trending more southerly.

  53. TAC
    Posted Dec 31, 2006 at 12:13 PM | Permalink

    Judith, it is reassuring that you’ve looked into the quality of the NHC hurricane dataset, and your comments (“NHC landfalling data set is fraught with errors”) are certainly cause for concern. However, I am curious as to why you have substantially more confidence in your TS dataset, particularly in light of (#50):

    The definition of hurricane in my landfalling data set is a storm that was once hurricane strength whose eye crossed land when it was at least at TS strength.

    How long have instruments been in place to say whether or not a TS was ever of hurricane strength? That would seem like a problematic determination prior to WWII, or even pre-satellite. Also, has there been a consistent definition of “hurricane strength” over the years?

  54. Ken Fritsch
    Posted Dec 31, 2006 at 12:48 PM | Permalink

    Re: #51

    I stand in anticipation of being corrected by the experts, but after reading the referenced link this layperson’s view sees a nibbling of the data at the margins without getting a feel for the overall data situation.

    After noting RPJ’s comments, it appears obvious to me that using Dr. Curry’s definition of a land falling event would certainly put the total number of TCs back into play. This definition, in my view, cuts the independency of observations used for total and landfall events — and the ability to use one as a check on the other.

  55. Posted Dec 31, 2006 at 1:10 PM | Permalink


    Thanks. My sense is that we’d all be better off working from the same shared, offical dataset, and working to improve it as errors are found. When multiple datasets are introduced it can lead to confusion, especially when different researchers use incompatible definitions of concepts like “landfalling hurricane.” the recent Wu et al. paper in EOS illustrates, for intensities how different datasets can lead to confusion rather than clarity.

    I can understand scientifically why you might want to focus on all landfalling cyclones. At the same time, errors at the TC/H boundary are largely irrelevant to those of us focused on damage. An increase in TS, cat 1 or even 2 storms is not of large societal significance as compared to cat 3-5, which cause about 90% of total damage.

    Can you tell me how many total storms there are in your dataset for hurricane landfalls using the NHC definition of landfall? NHC has 280. Yours using a different definition for “hurricane landfall” results in with 320. How much of the 40-storm difference is due to the different definitions versus errors that you caught in the NHC dataset?


  56. J. Curry
    Posted Dec 31, 2006 at 1:42 PM | Permalink

    Roger, a shared official data set would definitely be the best thing. I started out last summer using the official designated landfall dataset, but the inconsistencies were so bad that we undertook this other effort. We gave our comments to Landsea et al. last Aug. to be considered by the hurdat committee. They have evaluated our comments up to 1911 or so. They have not gotten back to us re the later years, i assume they are waiting for the HURDAT process to reconsider the later years. So at this point, the only data I have confidence in is our version of the total TC data set. Re the storms that were hurricanes at landfall, at this point, the discrepancies between the designated landfall data set and the best track data set are too large to make sense of a number of the hurricanes, in my opinion. So at this point, i don’t have a clear break down of the hurricane data, but when Jelinek and Belanger return to atlanta, we can put together documentation for storms that they found to be inconsistent. I would also like to add that i REALLY do not like being in the position of putting a TC climate data out there, since I do not want to lay myself open to have interjected bias in any way into the data set. But Jelinek and Belanger did a very careful job, and we will wait to see what the HURDAT committee has to say about all this. So my recommendation at this point is to stick to the landfallling TCs in the context of this exchange that is trying figure out causal mechanisms, etc. (i agree that damage studies need the hurricane designation). Trying to get the HURDAT process to speed up would be a really good thing.

  57. Posted Dec 31, 2006 at 2:13 PM | Permalink


    Thanks, makes sense to me. If it is OK with you I would like to use your dataset in a revision to my draft paper on landfall proportions, here:


    … simply for robustness sake. I can credit you and/or you students any way that you’d like. Let me know if this would be OK.

    Thanks again for sharing your dataset, and kudos to Steve for making this discussion resource available.

    Happy new year, Judy, and all the others on this site!

  58. J. Curry
    Posted Dec 31, 2006 at 2:56 PM | Permalink

    Roger, the credit for the data set goes to James Belanger and Mark Jelinek.

  59. Steve McIntyre
    Posted Dec 31, 2006 at 4:03 PM | Permalink

    #11. Here’s a variation of #11 in which I’ve calculated the median longitude by year for measurements above 65 knots and between 35 and 65 knots. There is perhaps a slight trend for wind GT 65 knots, but there definitely seems to be a “regime change” around 1950 for intermediate wind speeds between 35 and 65 knots with a distinct increase in measurements further to the east. Is it more likely that this regime change is climatic or to do with the introduction of aircraft reconnaissance. Valley girls have a word for this.  Is there any potential knock-on for Holland and Webster?  I don’t know right now.

  60. guest
    Posted Dec 31, 2006 at 5:30 PM | Permalink

    There is general reference to 1944 as being the start of Atlantic recon flights but anyone who uses that date should study the history of the recon program. It was bumpy, with starts and stops and changes in practices and handoffs from one group to another. It took about five years for it to become established.

  61. guest
    Posted Dec 31, 2006 at 5:41 PM | Permalink

    Steve, the Unisys database includes a pressure reading for storm observations where there was some basis (recon, satellite, ship, landfall) for estimating the pressure. If Unisys has a blank, then there was probably no observational basis for a pressure reading (which is the basis for wind estimates).

    Could you plot, for each year, the percent of 6-hour observations for which Unisys shows a pressure? I think you would see a pattern of few observations in older years, increasing with time.

    The farther back one goes, the more guesses one encounters about storm intensity.

  62. Steve McIntyre
    Posted Dec 31, 2006 at 10:21 PM | Permalink

    #61. Here’s the graph you requested – percent of Atlantic track data with pressure measurement by year. It took about 10 seconds to do the calculation using the collation of Track data that I’ve archived at http://data.climateaudit.org/data/hurricane/unisys/Track.ATL.txt and similarly for WPAC, EPAC, SH and NIO. The calculation was as follows:

    pct.pressure< -tapply(!is.na(Track.collation$ATL$press),Track.collation$ATL$year,sum)/

    All the other graphs that I’ve shown are typically 1-2 lines using the marvelous tapply function in R.

    Atlantic storm track – % of measurements by year with pressure data.

  63. TAC
    Posted Jan 1, 2007 at 7:28 AM | Permalink

    SteveM and all: Happy New Year!

    Re #59 and #62: Drawing inferences from small samples is always a bit tricky, but your figures are pretty convincing; it seems highly likely that the “regime change” is due to introduction of aircraft reconaissance, not a change in the natural system.

    This sort of problem arises often in longitudinal datasets, particularly where “improved” monitoring methods have been adopted and where no one anticipated the need to employ consistent sampling protocols (who could have imagined we’d be interested in trends?). In particular, this has been a recurrent issue in water quality data (e.g. concentrations of heavy metals in aquatic samples). There is some peer-reviewed literature on the topic, but most of what I’ve read has been in the form of internal white papers (dirty linen seldom gets washed in public).

    Incidentally, in such situations, the generally accepted practice (“default assumption”) is to assume that observed trends are due to the change of procedure. However, where the statistical characteristics (i.e. bias and standard error) of the different monitoring procedures can be quantifed (as was the case in some of the water quality datasets), the problem can be dealt with using bias correction and weighted least squares. In some cases it may be possible to draw inferences about the system under study, but there is always an asterisk.

  64. Ken Fritsch
    Posted Jan 1, 2007 at 11:56 AM | Permalink

    I plotted Dr. Curry’s landfalling named storms and landfalling hurricanes versus year of occurrence using Curry’s definition and the data presented in comment #43 of this thread.

    For named storms I found the trend had y = 0.0039x + 3.00 and r^2 =0.0089 and for hurricanes I found y = -0.0006 + 2.14 and r^2 = 0.0004. These results indicate no significant trend in either case.

    These data would essentially agree with that presented in RPJ’s draft paper on the subject and, at least to my current understanding would not change his indicated conclusions.

  65. Ken Fritsch
    Posted Jan 1, 2007 at 12:02 PM | Permalink

    Re: #64

    That would be y = -0.0006x + 2.14 and R^2 = 0.0004 for hurricanes. I also wanted to ask whether and/or what I was missing from the discussion on these numbers/trends?

  66. David Smith
    Posted Jan 1, 2007 at 3:17 PM | Permalink

    Re #59, #62

    In #59, the apparent eastward shift circa 1950 likely reflects better detection and tracking of weaker systems in the eastern Atlantic. Recon and patrol flights were established, the US Weather Bureau increased staff and began pay greater attention, including assigning names to systems starting in 1950.

    In #62, the lack of storm pressure readings prior to 1950 shows just how much guessing was done with regards to storm wind strength then. How can one use data that’s not based on actual observation?

  67. J. Curry
    Posted Jan 1, 2007 at 3:42 PM | Permalink

    I certainly agree that there is no “trend” in U.S. landfalling TCs. I stated this in my testimony. The relevant comparison is the number of storms in different peak periods. The average for the past 10 or 11 years is higher than any previous 10 or 11 years that you can find in the data record (the peak near 1880 comes close). If you plot the landfalling TC time series with a 11 yr running mean (its plotted in my testimony, fig 4), you see the following broad features:
    minima: ~1860
    maxima: ~1880
    minima: ~1920
    maxima: ~1950
    minima: ~1975
    maxima: 2005

    (note: the reference for my testimony and these plots is

    Click to access GT%20-%20Curry%20Testimony.pdf

    The broad features of this variability look like the AMO, which seems to have had a period of ~70 yrs during this period. The whole issue deconvoluting AMO from AGW from the tropical SST time series was addressed by Mann/Emanuel, but the overall signal of the landfalling TCs looks a lot more like AMO than the global temperature record.

    The most salient feature from an AGW point of view is the fact that we are currently about 15-20 years from the peak of the current AMO, and the recent period since 1995 has already slightly surpassed the TC peak at the previous AMO peaks.

    So on to my previous issue of a forecast for 2025 (and this is of most relevance for florida, which has been taking 50% of the recent landfall hits; the florida stuff is for bender), we would seem to have 3 general possibilities

    1) The TCs will pretty much continue as they have over the last decade, with an average of around 5/yr with much year to year variation.

    2) We are currently at the peak of the current active cycle (we must be since the counts have reached the previous peak values) and we should be heading soon back to the much lower values of the 1980s

    3) The AMO will peak ca. 2020 plus AGW will result in additional 0.5C warming. This scales to an average of 8 TCs/yr (which is the average of the 2004/2005 seasons)

    Assuming things will stay the same is a bad idea given the historical record and the bad assumption in the late 70’s that things would stay the same as the low levels during that period

    If it is #2, we would see significant decrease in 5-10 yrs

    For some insight into the implications of #3, if the current AMO peaks in 2020, then the analogue for the current position in the AMO cycle is the mid 30’s (and 1865). The 1930’s analogy would be that we are at the beginning of a broad maxima that will last for ~25 years, while the 1880’s analogy is that we are headed towards a sharper peak.

    #1 is a persistence forecast

    #2 assumes that the climatology is stable, and therefore we must see a downturn soon rather than a continued increase

    #3 includes some physics of the NATL storms, associated with both AGW and AMO, with some simple statistical relationships to infer the number of projected landfalls

    #3 has more robust assumptions behind it (some physics).

    10 more years of observations should clarify the difference between #2 and #3 (may be somewhat ambiguous relative to #1)

  68. Steve McIntyre
    Posted Jan 1, 2007 at 4:21 PM | Permalink

    If one looks at the graph in #62 above, the period from 1953-1966 or so is interesting because there is a mix of measurements with pressure readings and ones without pressure readings – which I presume correspond to different measurement technologies. So I made the following graphic showing the histogram of wind speeds for ATL measurements from 1953-1966 for measurements with pressure and measurements without pressure. Guess what – they are a lot different. The histogram of measurements without accompanying pressure measurements has a distribution that looks more like what I would expect from an extreme-value distribution – hurricanes are out on the tail of an energy distribution and one expects things to look like a tail distribution. The measurements with accompanying pressure readings don’t look like a tail distribution. What causes the difference? Maybe some hurricane specialist (Judith???) can explain. Is there a difference in the type of storm that had pressure measurements? Or is the difference an artifact of differing measurement methodologies?

  69. Posted Jan 1, 2007 at 4:59 PM | Permalink

    Judy- #67 – Just so I fully understand, can you provide your exact definition of “landfalling named storm” used in these figures? I understand your definition for “landfalling hurricane.” Thanks!

  70. TAC
    Posted Jan 1, 2007 at 5:31 PM | Permalink

    #68 SteveM: How about this for a possible explanation: It is a stratified sample. By inspection, it seems that for wind speeds above 100, the sample is 100% of population; Wind speeds 60-100, sample is about 50%; Wind speeds below 60, sample is about 10%.

    In a world of limited resources, this might be a nearly optimal sampling program, exactly the kind of experimental design that a good statistical consultant might have come up with. However, it would be essential that the sampling approach be documented in order to avoid future misinterpretation of the data.

  71. Steve McIntyre
    Posted Jan 1, 2007 at 5:41 PM | Permalink

    #70. That seems like a plausible explanation. I wonder if there’s documentation for it.

  72. David Smith
    Posted Jan 1, 2007 at 5:52 PM | Permalink

    Steve , if you’re taking requests, could you look at histograms for these three periods:

    1945-1960 (This is a timespan identified by HW as a major-hurricane period (see their Figure 4) using best-guesses plus some recon)

    1980-1994 (This is a dormant timespan, but it benefits from pretty good satellite estimates)

    1995-2006 (This is a timespan identified by HW as a major-hurricane period, with best-available satellite estimates)

    The first and third periods should be similar in distribution. The second period should be different from the other two, and skewed towards the lower velocities.

    If the first and third period distributions are not similar, then to me the first period numbers are suspect.

  73. David Smith
    Posted Jan 1, 2007 at 6:18 PM | Permalink

    Re #68

    My two cents:

    The histogram on the right is composed of “best-guesses”. The weather specialists guessed about storm strengths in smooth ways which approximated what they thought was typical storm behavior. That’s what I would do – I’d guess a storm was behaving “normally” unless I had a reason to think otherwise.

    The histogram on the left is composed of actual measurements. They tend to be taken in the later, mature stages of storms (typically in the western Atlantic) which explains it being centered at higher velocities than that of the right histogram. It also tends to be skewed towards the right side, which is probably a better reflection of reality during active periods than the “climatological behavior” assumed in the right histogram.

    Bottom line: once again, the historical guesses aren’t worth much. Using best-guesses in a search for climate signals is fraught with risk.

    Footnote: Note the Uptick at 65 knots in both histograms. That probably reflects forecasters tendency to name/keep a system a “hurricane” (65 knot minimum) more than it deserves, for the psychological effect of the word if populated areas are threatened.

  74. Judith Curry
    Posted Jan 1, 2007 at 6:35 PM | Permalink

    Steve, this is why i have been focusing on TC counts, rather than anything to do with intensity data. I think the NATL intensity data is ok back to 1977. So-so from 1970-1977. From 1944-1970, there is a hotly debated issue (which i think relates to what you are seeing) on the pressure-intensity relationship to use (Landsea vs Emanuel). Prior to 1944, the intensity data is almost certainly useless. Doing something to sort out the intensity data back to 1944 would be a HUGE help. The HURDAT committee hasn’t yet tackled this, but I suspect what they do may be controversial (both sides of this debate are not adequately represented on the HURDAT committee). We’ll see if anything useful can be salvaged from this.

    Emanuel’s analysis (with v cubed) is very sensitive to the wind speed. Holland/Webster analysis, who look at major (3+4+5) vs minor (1+2) hurricanes is less sensitive to this issue.

  75. Willis Eschenbach
    Posted Jan 1, 2007 at 8:39 PM | Permalink

    Judith, I’m curious about one of your statements regarding landfalling TCs:

    The AMO will peak ca. 2020 plus AGW will result in additional 0.5C warming. This scales to an average of 8 TCs/yr (which is the average of the 2004/2005 seasons)

    My curiosities are:

    1) The trend line of landfalling TCs w.r.t. temperature, although significant, is almost flat, 0.02 additional TCs/1°C of warming. A half degree C warming will make no difference.

    2) What is the source of your statement that the AMO will peak in 2020? The AMO seems to me to be peaking now. Here is a graph of the AMO, along with an analysis of the underlying cycles. This is robust w.r.t. the use of 1, 2, or 3 principal cycles. It shows that the AMO is peaking, not in 2020, but about 2006 …

    2) The trend line of TCs w.r.t. the AMO shows a best fit with a 4-year lag between AMO and TCs. However, the correlation, while visually compelling, is not statistically significant (p = 0.11)

    3) Even assuming that the correlation is significant, a) the current average is about 4.5 storms/year, b) the average has never been above 5 TCs per year, and c) the projection peaks at about 4.8 TCs per year.

    My best to you,


  76. Steve McIntyre
    Posted Jan 1, 2007 at 9:46 PM | Permalink

    #72. David, is this what you want:

    Generating script is:

    temp1=1945)&(Track.collation$ATL$year< =1960)
    temp2=1980)&(Track.collation$ATL$year< =1994)
    temp3=1995)&(Track.collation$ATL$year< =2006)


  77. Judith Curry
    Posted Jan 2, 2007 at 7:14 AM | Permalink

    Willis, the AMO does not reflect itself very much in tropical SST. The main definition of AMO is associated with SSTs up around greenland.

    for an easily accessible faq sheet on AMO
    http://www.wunderground.com/blog/JeffMasters/comment.html? entrynum=265&tstamp=200512 (watch for space, delete)

    note, the issue of separating the AMO from the global temperature signal is somewhat thorny (pointed out by mann/emanuel).

    When will the next peak occur (or the end of the positive phase?) the last postive phase lasted 44 years (as per Jeff Masters). Recent unpublished work from a conference (see http://www.realclimate.org/index.php?p=365) suggests that a slowdown of the meridional overturning circulation (associated with the AMO) is unlikely before 2030-2050.

    So it does not seem that we are currently at the peak of the AMO, or near the end of the positive phase.

  78. David Smith
    Posted Jan 2, 2007 at 8:55 AM | Permalink

    Re #76 Steve, thanks for the histograms.

    The middle and right histograms are about as good as we can get on Atlantic storm wind distribution. I think that, in almost all cases, there was at least a modern satellite image on which to estimate storm strength.

    (For reference, the middle histogram is of a modern inactive period while the right histogram is a modern active period.)

    They show what Steve calls a tail distribution, that sloping decline at the extreme wind speeds, which is what I, too, expect. They also show about 60% of readings are below 50 knots, which is weak tropical storm/depression strength.

    Two things about the left histogram, which is of the previous active period. One, the below-50 knots count is about 50% instead of 60%. To me, that indicates that the seasonal records may have missed weak systems and also that the early, weak, far-at-sea portions of some storm tracks may not be recorded due to lack of coverage.

    Two, instead of a tail distribution at higher windspeeds, the histogram looks more like a “Manx cat” tail distribution, a bit stubby.

    I’d be hesitant to use the 1945-1960 intensity data in a chart with more-relaiable modern data. It’s a mix of 20% measurement and 80% analysts’ best-guesses and it shows in the histograms.

  79. Ken Fritsch
    Posted Jan 2, 2007 at 9:41 AM | Permalink

    Would not the distributions for wind speeds as depicted by the-without-pressure graph in comment #68 and all 3 graphs in comment #76 be what would be expected for a measurement that cannot go below zero and would thus tend to have a log distribution?

    The graph of wind speed distributions for the with-pressure case in comment #68 looks less likely/typical to me.

  80. Steve McIntyre
    Posted Jan 2, 2007 at 10:23 AM | Permalink

    #79. Remember that you’re not measuring all wind speeds, just wind speeds in systems of interest. So they’ve already been “scouted”. Think of distributions of baseball skill in Bill James terms – I remember an insightful comment on this: skill for major league players is not bell-shaped, because anyone who even gets a cup of coffee is already really good. It’s a tail distribution. Same with storm wind speed measurements. They’ve already been scouted. No harm in this – just that it has to be recognized.

  81. Ken Fritsch
    Posted Jan 2, 2007 at 10:37 AM | Permalink

    Re: #80

    With-pressure = distribution with a major league pre-selection.

    Without-pressure = sandlot distribution.

    Actually on closer reading you had indicated this in the earlier post. Thanks for (re)calling this to my (in)attention.

  82. Steve McIntyre
    Posted Jan 2, 2007 at 12:30 PM | Permalink

    #81. Actually my point is a little different still. Without-pressure are still the equivalent of “pretty good” players and well out on the tail of the distribution.

  83. Ken Fritsch
    Posted Jan 2, 2007 at 5:20 PM | Permalink

    Re: #81

    To heck with the baseball analogies, they are only confusing me again. I have studied Bill James baseball statistics over the years and used a modification of a couple of them to show what I thought would indicate that great pitchers (my position when I played) were more valuable to a baseball team than a great everyday position player — not — they were very nearly the same.

    I think we agree that the distribution in the graph with-pressure is unexpected and needs explaining.

    It was of course a similar argument by Stephen Jay Gould about changing skill levels, over time, of all major league players improving to the point that they were approaching the limits of human capabilities and thus explaining the lack of 0.400 hitters since T. Williams last did it over 6 decades ago. Sort of like putting that zero barrier at the other end of the x axis.

  84. jae
    Posted Jan 2, 2007 at 5:47 PM | Permalink

    75: Willis, it is interesting to compare the trends you show with Figure 4 here. My eyball statisitical procedure shows a very good correlation.

  85. Willis Eschenbach
    Posted Jan 2, 2007 at 9:05 PM | Permalink

    jae, IIRC that figure was later shown to contain some errors. Don’t have the reference in front of me. There are a variety of correlations of climate with sunspot activity, however, which makes the lack of sunspot data in GHCs … well, I was going to say “surprising”, but given the GHCs, I’ll go with “typical”.


  86. jae
    Posted Jan 3, 2007 at 11:02 AM | Permalink

    Willis: please try to locate the reference. I’m really interested in this.

  87. David Smith
    Posted Mar 6, 2007 at 6:00 PM | Permalink

    A couple of presentations from Greg Holland are here and here . I think they’re pursuing an Atlantic-is-special approach.

    At first glance I don’t see anything new involving data.

%d bloggers like this: