30 October 2012

Sandy and the Top 20 Normalized US Hurricane Losses

UPDATE #4 13 Feb 2013: Yesterday the NHC issued its final report on Hurricane Sandy, and includes $50 billion as the total US damages from Sandy, which would place it 7th in the table below.

UPDATE #3 11/24: There is still a lot of uncertainty in Sandy damage estimates. PCS, which tabulates actual insured losses, has released a first estimate of $11 billion in insured losses. This is at the low end of the range, bu is very likely to rise. How far? We'll have to wait for that. Once better numbers are in, I'll do a new post with apples to apples numbers from Sandy to compare to our normalized tabulation.

UPDATE #2 11/1: Moody's has published an estimate of $30 billion in direct total damages due to Sandy. That would place Sandy at #10 all time in the normalization table below.

UPDATE 11/1: EQECAT has published updated estimates of Sandy's losses today, $10-20 billion insured and $30-50B total. Depending on the amount of flood damage include in the total (the NWS hurricane losses separate out flood) the new estimates, if they hold up near the high end of the range, would push Sandy into the top 10 all-time losses in the normalization table below.

Here is a table showing the top 20 hurricane losses 1900 to 2011, normalized to 2012 dollars. In other words, the figures show an estimate of what the losses would be were historical storms to occur in 2012. The numbers come from ICAT based on an extension of Pielke et al. 2008.

Great Miami Sep 18,1926 1 180,220,000,000
Galveston Sep 08,1900 2 105,570,000,000
Galveston Aug 17,1915 3 84,910,000,000
Katrina Aug 29,2005 4 84,620,000,000
Andrew Aug 24,1992 5 64,410,000,000
Storm 11 in 1944 Oct 19,1944 6 53,940,000,000
Donna Sep 10,1960 7 49,810,000,000
New England Sep 21,1938 8 46,840,000,000
Lake Okeechobee Sep 16,1928 9 44,890,000,000
Wilma Oct 24,2005 10 25,960,000,000
Hazel Oct 18,1954 11 24,260,000,000
Diane Aug 19,1955 12 24,110,000,000
Camille Aug 17,1969 13 23,040,000,000
Charley Aug 13,2004 14 20,380,000,000
Ike Sep 13,2008 15 20,370,000,000
Hugo Sep 21,1989 16 20,020,000,000
Carol Aug 31,1954 17 19,290,000,000
Agnes Jun 22,1972 18 19,010,000,000
Ivan Sep 16,2004 19 18,590,000,000
Storm 2 in 1949 Aug 26,1949 20 18,510,000,000

While it will be some time until we have apples to apples estimates from Sandy, the current estimates of $20 billion would place Sandy at #17 all time out of 242 loss-producng storms 1900 to present (in the top 10%). If the damage gets to $30 billion it would crack the top 10 and (top 5%). Right now it seems unlikely that Sandy will climb any higher on the table. (Note that inland flood damage is not included in the tabulations above.)

In historical context, Sandy sits alongside Carol, Diane and Hazel. One big difference however -- Carol, Diane and Hazel hit the US Atlantic coast within a single 13 month period in 1954-1955.  Imagine that. 


  1. Yeah, imagine what the likes of Romm and McKibbon would have been doing in 1955. I made a similar comment at Tobis's planet3.org Grim trajectories posting.

  2. I realize it might be poor form to repeat the same comment on successive posts - but then again these last two posts are on the same topic. I'd like to see what responses people might have - so I guess I'll just see if it gets through moderation.

    Although I understand the rhetorical power of these kinds of comparisons, I fail to understand their value analytically.

    Just as it would be fairly meaningless to make comparisons of the costs of storm damages w/o controlling for inflation, so it is pointless to make these comparisons w/o controlling for infrastructure changes - whether they be increases in property exposed to risks or increases in infrastructure built to protect against damages.

    These numbers are basically meaningless w/o that type of control for related variables. Even more direct comparisons such as surge levels and flood levels are not informative w/o that type of control of data.

  3. Thanks for posting that comment, Roger...

    So let me add another point - Perhaps the most relevant variables not controlled for:

    The differential impact of dramatically better prediction of severe weather events, increased availability of resources, and improved preparatory strategies and methodologies.

    This is all that much more important, since I just read about an analysis that argues that Paul Ryan's budget proposals would lead to a 50% reduction in the accuracy of NOAA's ability to predict blizzards and hurricanes. Please consider not only the impact of that budget not only in terms of #'s of people suffering the impact of severe weather - but also the cost ineffectiveness of cutting services that would save money in the long run.

  4. Shouldn’t there be a “correction” or normalization factor that takes into consideration that Sandy hit an area that probably has greater asset values at risk than (most) other areas hit by hurricanes? I suspect if this were considered, the relative “fury” of Sandy – as measured by losses attributed to it -- would be diminished.

  5. @ #4 (indur)

    Maybe it's not appropriate to try to correct for that aspect?

    For one, the area hit by Sandy was also disproportionately populated for more than the last 100 years.

    Secondly, while I wouldn't argue there's any evidence of this at the moment, for the sake of argument, take the possibility that something about current and future climate change leads storms to land more frequently land in this highly populated area. Couldn't it be more useful to have an index reflect that increasingly national cost, even if that increase wasn't caused by storms being stronger?

  6. Joshua - The values give a sense of what an equivalent level of damage would be today. This does give a sense of how better forecasting and infrastructure have lessened damage over time. Roger could have been a bit more clear - it's not an equivalent storm this is measuring, but what an equivalent amount of damage caused by a historical storm would look like (the way it was described confused me initially as well).

  7. Roger by the time every insurance looter takes advantage of this storm, NJ alone will be 10 billion.

    Doorbell rings:

    Homeowner: Yes?

    Insurance looter: Would you like a free inspection of your roof? I'm sure if I "inspect" your roof I will find damage that will get you assistance in getting a new roof (even if that damage was post-Sandy damage caused by the roof inspector).

    Homeowner: Sounds great!

    Roger do you (is it possible to) factor in growing dishonesty and increases in fraudulent claims that are paid in your formula for storm damage normalization. Don't you think that people were more honest during your father's generation and would not as likely have filed the fraudulent claims that are filed with modern day disasters?

  8. @ Indur, @ kmye:

    Don't confuse the normalized historic losses with the figures you see linked to Sandy and estimates of her loss. the $20B from KA, or $15B from AIR, are insured losses that do account for higher priced homes, property, etc. in the Northeast.

    @ Papa Zu:

    I don't think the level of fraud for any catastrophe is too great that it isnt accounted for in the uncertainty of the loss estimates. The values you see in the media are usually the high end of a range. And that range has uncertainty around it as well.

    Then again, it is Jersey...

  9. (#7) Papa -

    ==]] Don't you think that people were more honest during your father's generation and would not as likely have filed the fraudulent claims that are filed with modern day disasters? [[==

    Do you have any evidence to support your supposition about a trend of increasing dishonesty?

  10. (#6) Christopher

    I'm still not sure that the variables across the entities being compared are well-enough controlled to make any comparison meaningful.

    To assess the degree to which forecasting and infrastructure have reduced the damage for a given level of storm, we'd need to quantify the improvements in forecasting and infrastructure, and control for changes in the value of property exposed to risk, changes in sea level, amount of subduction, etc.

    Again - it seems to me that these types of unspecified comparisons serve no scientific function, only a rhetorical/emotional appeal function. I consider it to be essentially a form of politicizing science.

  11. -10-Joshua

    Thanks for your comments ... I'd encourage you to read the actual literature on normalization methods and how they are evaluated, e.g., by looking for known climate signals (like ENSO) in adjusted economic data and also comparing trends in normalized data with trends in related physical quantities. This area of research is pretty robust by now. You might also compare the results of normalization methods with the dynamical models used by cat modelers and see how things stack up.

    Informed critiques always welcomed.


  12. So Roger - you're saying that in addition to the methodologies you just described, these estimates also rest on quantification of the effects of the advantages of better forecasting and differences in methodologies for responding to potential dangers? For example, FEMA - which does do work on mitigation - was established only in 1979. How do they adjust for that when comparing to storms that hit in the 1800s?

    Similarly - how do they quantify the assessment of "related physical quantities" to evaluate the impact of new infrastructure built to protect against damages?

    And should I conclude that by putting up these posts, you are implying that you trust those assessment methodologies to be valid - perhaps because you have "audited" them in some detail?

    ===]] Informed critiques always welcomed. [[[===

    Responses to skepticism are always welcomed.

  13. -12-Joshua

    Thanks ... a few replies:

    1. There is no evidence that improvements in forecasting or building practices/oversight over the past century have led to a bias in the results. See the literature for a discussion of how this conclusion has been reached.

    2. There is a big debate over whether FEMA (and NFIP specifically) has led to a decrease or an increase in disaster costs. This debate is unresolved. A PhD student of mine wrote a dissertation on this subject, and the answer is yes -- FEMA has done both.

    3. New infrastructure does not always lead to reduced damages (Google the "levee effect"). For instance, a survey of damage after Andrew (1992) found that the homes built in the 1930s and 1940s performed better in the storm than those build in 70s-90s. So such a bias can work both ways -- however we see no signal of such a bias in the normalization results for hurricanes or tornadoes (you can make a case for such a bias in floods and earthquakes however).

    4. You can be sure that the work we have published in the peer reviewed literature and often discussed here has our full confidence. That is not to say that everything will stand the test of time -- that is not how science works. However, our hurricane work has been replicated so often using multiple methods over 15+ years that I am pretty confident that it is robust. Our complete data is online, so feel free to check it out yourself.

    Thanks again.

  14. How is population growth handled in these analyses? The population in 1900 would have been far less than today and thus much less "stuff" would have been around that could be destroyed by a cyclone, which would in turn result in less damage than we see today - even if the cyclone itself was as strong.

  15. Joshua said:
    "Again - it seems to me that these types of unspecified comparisons serve no scientific function, only a rhetorical/emotional appeal function. I consider it to be essentially a form of politicizing science."

    Granted, these comparisons aren't perfect (how could they be?), but they are still a lot more informative than simply comparing the raw numbers or even the raw numbers with an inflation adjustment. If you have ideas for a better normalization method, have at it...

  16. (15) Mike -

    Yes. Expecting perfection is unrealistic. I guess my point is that these types of comparisons should be accompanied with caveats, qualifications, discussion of uncertainties, etc.

  17. Measured by number and duration of power outages Sandy was less damaging than Irene. However, Irene thinned out the weak points - trees and poles - so that number may be skewed. My driveway remained clear and my chainsaw was much less busy this year.

    That there was anything left for Hazel and Diane to destroy after Carol had already passed through underscores just how powerful they must have been.

  18. @Pasteur01:

    Most of the damage that Diane did was in flooding because it made landfall only about 5 days after another hurricane, Hurricane Connie (not on the list), and thus a lot of land was already saturated with water.

    Diane was not especially strong, but the additional water was simply too much after Connie.

  19. Just watched Roger's excellent intro of Chris Landsea and learned so much from Chris on the process of normalizing the data. I'm feel much more confident in this normalization methodology after watching the presentation on how it evolved.

  20. Weather.com has their own, very different list of the most expensive storms "in history". Surprise, surprise, they all happened since Andrew. They also give a price tag of 108 billion for Katrina.


  21. 20 - Those are the actual costs at the time; they don't even correct for inflation!

  22. Inflation isn't the only problem with the numbers that suggest that recent storms are more serious. Insurance coverage has also been expanding (covering loss of use, for example). Both biases has been changing over time.
    Since I agree that climate change is a serious threat, I like the results that the biases create, but I don't like the way they were reached. The later numbers need to be adjusted and your adjustments seem to be the best there are.

  23. There is an AP story yesterday published in the Albany Times-Union that describes Katrina as the costliest hurricane in history and Sandy as #2. The National Hurricane Center is given as the source, but no link. That is not in accord with your table of last year.

  24. -23-John Tepper Martin

    Thanks .. I'd guess that ranking is non-normalized data, see Tables 3a and 3b here: