I would prefer to use event-specific data: damage and SST values for each storm. It’s a mistake to count individual damages within storms as separate events. They aren’t independent. But averaging over a year’s time dulls a true model’s ability to detect a real effect.

1950 MJ SSTs are 26.11

This is calculated from:

http://www.cpc.noaa.gov/data/indices/sstoi.atl.indices

May is 25.85
June is 26.37

Average of the two is 26.11

Apologies for the confusion! Data is now correct.

Could you please assist me with this:

How do you derive these SSTs (e.g. 26.11°C for 2005) from the dataset at http://www.cpc.noaa.gov/data/indices/sstoi.atl.indices ?

Thanks

You write, “New Orleans comes to mind as an indirect effect due to failure of the 350 year long efforts to build engineering-solution-level levee systems.”

I think New Orleans is the direct result of failure to have developed and deployed a $2 billion (or less) water-filled-tube storm surge protection system from Pascagoula, MS, to south of Point a la Heche, LA, running along the Chandeleur Islands.

I’m pretty sure such a system could be deployed in less than 48 hours, at a cost of less than $2 billion…and it would have completely blocked the storm surge that caused New Orleans to flood.

http://markbahner.typepad.com/random_thoughts/2006/08/some_responses_.html

But I agree with your general point…the damages were not so much “direct” as “indirect” (resulting from failure of protection measures).

Mark

What physical mechanisms/processes can there be to cause hurricane damage to be related to SST. The damage caused is primarily a function of the type, quality, and quatity of the structures and things in the path of the hurricane. This is shown by the fact that some hurricanes cause no damage.

It seems to me that a proper measure of the potential, or driving force, for damage is the power of the storm. Direct correlation of damage as a function of SST does not seem to be correct in either the engineering or the scientific sense. A proper measure of the potential would not allow the cases in which no damage is done under positive values of the measure of the potential. I also would expect it to be a monotonic function of the potential. In adition to also being a proper function in a mathematical sense.

Additionally, how are allowances made for indirect effects of the hurricane in contrast to direct effects? New Orleans comes to mind as an indirect effect due to failure of the 350 year long efforts to build engineering-solution-level levee systems.

I probably shouldn’t do this (read and comment during a lunch hour, on something that I’m not familiar with), but…

You write, “But I do find your response difficult to understand. I have presented data in the form of a simple linear regression and rank correlation — exactly as you have described in your paper — and arrived at very different results. It would be useful to understand why, no?”

I think one of Jim Elsner’s points is that one could (or even should?) treat each event as separate prior to doing the linear regression. You’re adding all the damage in one year together before doing your regression, right? If so, would you get a different answer from not adding all the damage in one year together…i.e., using a separate data point for each storm?

You appear to use detrended SST data in your study. Setting aside everything else for the moment, presumably the SST trend is the most direct manifestation of anthropogenic climate change (a poitn that I think you made in GRL). If your damage study is based on data that has removed this trend, then why is it relevant to a discussion of the role of anthropogenic factors in US hurricane damage?

(Note – I understand and appreicate your point about “random sums” — this is a different issue.)

Thanks!

Many thanks for your continued engagement. But I do find your response difficult to understand. I have presented data in the form of a simple linear regression and rank correlation — exactly as you have described in your paper — and arrived at very different results. It would be useful to understand why, no?

You do have some data issues worth mentioning:

1. You use Collins and Lowe wchich include data generated via a catastrophe model (TopCat). this raises the possibility of some circular reasoning in your analysis. In addition, a focus on insured losses includes many more “moving parts” (such as coverage, premium levels, etc.), and thus raises the probability of spurious correlations.

Question: Did you look at your results using our normalized data for total economic losses?

2. You append 2001-2005 insured loss values without any apparent adjustment to 2000 values. this “splicing” of very different datasets will be problematic unless the data is normalized (back to 2000 values).

Question: What are your results 1950-2000?

3. If you throw out storms of less than hurricane strength and also throw out storms with <$100 million in damages, then according to our dataset, you are throwing out 150 out of 219 events from 1950-2005. This is an enormous amount of relevant data. But I will examine the relationships later today based on this subset of the data as you suggest.

Question: Are your results sensitive to including either TS or <$100M events?

4. What is the physical basis for MJ SSTs to influence intra-seasonal damage?

5. How about this — if you would simply share your data, as I have, I’d be happy to do these comparisons. Unfortunately, your paper does not provide the data or access to it. Explaining the differences in our analyses should be quite easy once I have your dataset. We are talking about approximately 80 events 1950-2005 plus the SST data that you used. You can send it to me by email if you’d prefer not to post.

Thanks!

I reiterate. An important limitation of your analysis is the failure to disaggregate the loss data. A large annual loss can be the result of many small losses or one big loss, and the number of loss events should be independent of the magnitude of the event. Thus neither of your arguments are compelling. This logic is apriori and should be addressed before any discussions of the data.

You state in your WMO Report “Different methods exist for calculating a disaster’s impacts and result in correspondingly different estimates for the same event. Consequently extreme caution should be taken when integrating analyses or conclusions across datasets.” Yet this is exactly what you do with this new post. This has little to do with the original point from your previous point concerning which argument is most compelling.

Obviously we are using different datasets. As one example we do not include tropical storms. You have a $7bn loss for 2001 (I assume mostly from TS Allison); we have $0. Our data and sources are clearly explained in our paper.

Best,
Jim

Their are some questions to raise regarding the more sophisticated elements of what they’ve done, but let’s take it one step at a time;-)

Thanks!