New Changnon paper on winter storm losses

August 20th, 2007

Posted by: admin

Keeping in line with similar research being done here on hurricanes (Roger and colleagues) and earthquakes (me), Stanley Changnon has a new paper out on winter storm losses. The abstract:

Winter storms are a major weather problem in the USA and their losses have been rapidly increasing. A total of 202 catastrophic winter storms, each causing more than $1 million in damages, occurred during 1949–2003, and their losses totaled $35.2 billion (2003 dollars). Catastrophic winter storms occurred in most parts of the contiguous USA, but were concentrated in the eastern half of the nation where 88% of all storm losses occurred. … The time distribution of the nation’s 202 storms during 1949–2003 had a sizable downward trend, whereas the nation’s storm losses had a major upward trend for the 55-year period. This increase over time in losses, given the decrease in storm incidences, was a result of significant temporal increases in storm sizes and storm intensities. Increases in storm intensities were small in the northern sections of the nation, but doubled across the southern two-thirds of the nation, reflecting a climatic shift in conditions producing intense winter storms.

The interesting zeroth- or first-order conclusion is that when using damage trends as a proxy for climatic trends, no climatic trends can be seen in hurricanes while a strong one can be seen in winter storms. From the latest Pielke et al. hurricane paper:

…it should be clear from the normalized estimates that while 2004 and 2005 were exceptional from the standpoint of the number of very damaging storms, there is no long-term trend of increasing damage over the time period covered by this analysis.

Whereas from the Changnon paper on winter storms:

Significant temporal increases in storm losses, storm sizes, and storm intensity have occurred in the United States. The national increase over time in losses, given the decrease in storm incidences, was a result of the increases over time in storm sizes and intensities. The marked temporal increases in storm sizes and storm intensities were greatest across the southern two-thirds of the nation.

6 Responses to “New Changnon paper on winter storm losses”

  1. Roger Pielke, Jr. Says:

    Hi Kevin-

    Changnon’s paper has a few problems with it, unfortunately.

    First, the paper that first reported this database (Changnon and Changnon 1998) indicated that the insurance industry changed its definition of “catastrophe” from $1 million to $5 million in 1983. This creates a significant discontinuity in the dataset which will have the effect of artificially increasing the losses per event after 1983.

    Changnon and Changnon (1998) try to avoid this by looking only at storms causing >$35 million in their adjusted loss database, but this also is in error. The threshold should have been more like $175 million, which corresponds to their largest adjustment times $5 million.

    So the differences reported in Table 2, which hinge on 1983, surprisingly enough, may be in some (or all) part due to the discontinuity.

    So I would exercise great caution using this paper to make claims about climate behavior. probably best to rely on actual studies of extratropical cyclones.

    And, as usual, such studies are far more valuable when the orginal data are made available and methods described precisely, rather than leaving the reader to guess and surmise!

  2. 2
  3. Chip Knappenberger Says:

    Roger (re#1),

    What you suggest hardly seems to make sense. Nowhere (that I can find anyway) does Changnon indicate in his new paper that there was a different definition of ‘catastrophe’ from 1983 onwards. Clearly a change in definition was noted in Changnon and Changnon (1998) and deemed important. Perhaps the data were recompiled so that they were consistent from 1949 through 2003?

    For instance, in the new paper, Table 1 lists the most and least damaging storms. One of the least damaging storms is listed as occurring on 1/21/83 and thus should, according to your proposition, reflect a >$5 million storm (if the definition change applied to January 1983 events). The total damages (adjusted) are listed as $8.3 million. This would mean the adjustment from 1983 to 2003 was at most 1.66 — which seems, to me, improbably low. On the other hand, if the storm only caused $1 million in damages in 1983 dollars, then the adjustment would be 8.3 which, seems at least more reasonable.

    So perhaps the ‘catstrophe’ data used in Changnon and Changnon (1998) were resampled so that the definition was consistent over time.

    Do you *know* otherwise, or are you surmising?



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

    Chip- No it does not make sense. Nor does their paper explain sufficiently to understand what they have done. There is clearly a bias if there are no storms between $1M and $6.2M. In their 1998 paper they explained that the smallest adjustment from the early 1950s was (from memory) like a factor of 25, meaning that the smallest value from that period would only be an adjusted $25M. On the other hand, if as is hinted obliquely at the top of p. 136 they first removed all storms of less than $5M prior to normalizing then this would introduce a bias. The total number of storms reported in Fig. 4 far exceeds the 202 that they report, yet they apparently use this larger total for calculating damage per storm. I don’t see a logical explanation for any of this based on the 1998 paper that they reference, or the current work.

    They don’t provide enough detail to discern what they did, which really limits the usefulness of the study.

  6. 4
  7. Chip Knappenberger Says:


    According to p.136:

    “When a storm caused losses in two or more districts, a count of an occurrence was made
    in each district; hence, the sum of the values on Fig. 4 exceeds the storm total of 202.”

    The storm loss averages are then calculated for the storms impacting each district (regardlss of whether or not they were also counted in another district). I don’t see a big problem there.

    It is curious about the $5 million comment in reference to Kocin and Uccellini (2004) paper. Not sure what the context of that comment is.

    I guess I don’t really understand these types of analyses anyway…if a $1 million storm in 1950 is adjusted to be a $35 million storm today, and yet $1 million storms today are also included, then isn’t there a bias towards smaller storms over time? To avoid this, shouldn’t the lower limit as to which storms to include in the analysis be something like $35 million in today’s dollars? Now I see this is exactly the point you were making in your first comment…which makes the contents of Changnon’s Table 1 (for the lowest storms) quite difficult to comprehend, for at least the 5 least damaging storms all caused less damage than $35 million.

    Even if they reclassified the post-1982 events to include actual losses of 1 million (as opposed to $5 million from the 1998 paper), the very least damaging storm in the entire analysis should not have caused less than $35 million in losses in 2003 dollars.

    Isn’t that right?

    But by not doing so, which seems to be the case, doesn’t this induce a bias towards smaller storms with time–opposite to the trend that is reported?

    Quite a bit confused,


  8. 5
  9. Roger Pielke, Jr. Says:

    Hi Chip-

    Further confusions: On p. 136 they write of Table 4,

    “The frequency of winter storm catastrophes in the nation’s nine climate districts (Fig. 4)
    shows a maximum of 95 storms in the Northeast, with other high values in the Central,
    Southeast, and South districts. A study of the impacts of snowstorms in the northeast during
    1950–2003 identified 66 bad storms (Kocin and Uccellini 2004), but only 32 of these
    storms were also winter storm catastrophes. The other 34 storms identified as being severe
    did not produces losses amounting to $5 million.”

    So presumably there would only be 32 storms in their database? If so, then why calculate damage per storm as in terms of the 95 storms? or if the storms are doubled counted then you can’t use these numbers for loss-per storm!

    You write, “shouldn’t the lower limit as to which storms to include in the analysis be something like $35 million in today’s dollars?” Yes!

    You ask — “But by not doing so, which seems to be the case, doesn’t this induce a bias towards smaller storms with time–opposite to the trend that is reported?”

    Actually, I think the opposite. Think of it his way, if they removed all storms of a non-normalized $1-5M prior to 1983 then this has the effect in the earliest years of removing all storms that when normalized would be $35-175M! So by removing these largish events it creates an artificial bias towards large storms.

    This is my best guess, but as I said, I can’t be sure based on the study, which seems to me a missed opportunity.

  10. 6
  11. Chip Knappenberger Says:


    The p.136 passage the you quoted, especially the Kocin numbers, seem almost seems like an aside. I would expect to see some discussion of why the Kocin numbers were so dramatically different from Stan’s, but there really isn’t any. Odd, to say the least.

    As far as the inclusion of storms go, I see it like this. If only storms causing $1 million in real (then) damages were included and if the weighting factor decreased from 35x in the 1950s to 1x in 2003, then, it seems like there are a lot more small storms that are included over time. So that by 2003, storms that only caused $1 million dollars in damages are included, while the smallest storm (in 2003 dollars) that could be included in 1950 would be a $35 million storm. Thus, only big storms are included in the 1950s while smaller and smaller storms are included as the weighting factor declines. To me, this should increase the frequency and decrease the average severity as we approach the present…all else being equal. To keep things level, storms causing 1,000,000/35 or $28,571 in damages in 1950 should be included (or conversely, the minimun storm included should be $35 million in adjusted (2003) dollars. Otherwise, there is a deficit of small storms early in the record.

    Your point, I think, is that if they changed the minimum criteria for damage to be included as a ‘catastrophe’ from $1 million to $5 million in 1983, that this change would have caused continuity problems. But, I don’t think it is established that they actual did this (I know you think they did, and perhaps you are correct, but I still am not sure based upon my argument in my first comment).

    Nevertheless, it certainly appears that sufficient detail is lacking, since neither of us can seem to figure out what actually is going on. And that, in and of itself, seems to pose a problem. :^)