« Disasters »

There is not "plenty of time remaining for additional storms." November hurricanes are extremely rare, and there are no hurricanes on the horizon in the three weeks left in October. Hurricane predicters have been completely off the mark this year, including William Gray's group at CSU--so badly off the mark that none of their future predictions should be taken seriously. They clearly don't know what they're talking about.

Roger,

Doesn't this data point suggest that in fact the level of activity of a hurricane season has a huge impact on damages incurred? It would follow then that on average more active storm seasons are indeed a cause for concern. Perhaps you can clear that up for me, I will defer to your expertise on this subject.

Coby-

Thanks. The overall level of Atlantic hurricane activity has a very weak relationship with damages. There is no relationship, e.g., between SST and damages. A weak relationship of PDI or ACE with damage. And a good relationship of landfall PDI with damage.

The reason for the nature of these relationships that there is not a strong relationship Atlantic basin PDI and US landfall PDI. All of this is discussed in these short exchanges from Nature:

http://sciencepolicy.colorado.edu/admin/publication_files/resource-1890-2005.48.pdf

Thanks!

David-

Seasonal forecasts have shown skill. See:

http://sciencepolicy.colorado.edu/prometheus/archives/scientific_assessments/000860another_problem_wi.html

Our dataset of 218 landfalling storms has 38 making landfall in October and November, with 6 in November. Likely, no. But not impossible either.

Thanks.

Given the strong negative influence of el nino on Atlantic hurricane production, wouldn't it be fairer to compare this year with other e lnino years, instead of the whole record?

Since 1950, I get the following 18 years with positive enso from the CPC: 51,57,63,65,68,69,71,76,77,82,86,87,91,94,97,02,04,06 (forecast).

squinting at the storm tracks from unisys, I see the following average continental US landfalls for these 18 years:
TS: 1.83 +/- 1.6
H: 0.72 +/- .8
MH: 0.39 +/1 .8

The 2006 record is:
TS: 3
H: 0
MH: 0

So in terns of landfalls, the year is within 1 standard deviation of the el nino average for all three storm sizes.

You wanna more sophisticated analysis? Ask someone who isn't a geochemist.

So, Lab Lemming, you are saying we can use the level of hurricane activity beforehand as a predictor of El Niños?

No, I don't see how that follows.

In fact, one of the problems with enso forcasting is that the skill is very low prior to June, when the hurricane season starts and the Australian winter wheat crop is sewn.

An improvement in forcasting skill would help both hurricane forcasters and people adversely impacted by enso, but that has little to do with my previous post, which can be sumarized thusly:

El Nino is probably responcible for the small season, and we know that el inio has blown hurricane forcasts before because it cannot be predicted before the start of the season. so this season falls into the catagory of "blown due to known, currently unpredictable weather events" category.

This paper is by now a bit dated, but perhaps relevant to this discussion:

Pielke, Jr., R.A., and C.W. Landsea, 1999: La Niña, El Niño, and Atlantic Hurricane Damages in the United States. Bulletin of the American Meteorological Society, 80, 10, 2027-2033.
http://sciencepolicy.colorado.edu/admin/publication_files/resource-85-1999.14.pdf

A small suggestion: Focus first on noneconomic criteria, such as storm numbers, intensity, etc., and rank the 2006 season thereon. I don't know squat about hurricane forcasting, but I know enough about economics to say that normalizing the damage valuation function over a long time series is really, really complicated and certain to be really, really controversial.

From your 1998 paper describing the normalization procedure you used, your formula is storm loss|1995 = f{year of occurrence, inflation, wealth, population}, where f is presumed to be a simple multiplicative function. Population data are easily obtained, and although nitpicky economists might prefer a GDP deflator to a GNP deflator, both capture inflation relatively well for these purposes. The proxy measure used for wealth is BEA's “fixed reproducible tangible wealth” statistic, which seems reasonable, though I'm not sure how best to distinguish insured from uninsured assets. (Insurance matters because its existence may be capitalized into market-based asset values.) Proportionality in the relationship is certainly plausible, but it's not obvious that it's the only functional form the damage function could take. Perhaps I missed it, but I don't see any expirical testing of this damage function to ascertain its goodness of fit to the data.

All this is a roundabout way of inquiring about whether the wealth function was adjusted for the fraction of national wealth held by residents in hurricane-vulnerable counties. I read the text to say that you used nationwide per capita wealth. It is almost certain that that the faction of national owned by coastal redidents is much higher today than in years past (you make that point in your paper), and plausible that the fraction of national wealth owned by residents of hurricane-vulnerable counties has risen greater than the fraction owned by residents of coastal counties in general (a point I did not find, but could have missed).

Anyway, assuming your model is generally correct, and your proxy measure for wealth is national per capita wealth unadjusted for the rising secular share owned by residents of coastal (and especially, hurricane-vulnerable) counties, then it's possible that your conclusion was an incomplete refutation of the hypothesis that climate change is responsible for rising hurricane damages. In other words, the apparent null effect observed in Figure 4 might hide a downward slope.

Richard-

Thanks for these comments.

In the 1998 study we do adjust the wealth by the per capita change in population at the coastal county level. This will likely get some, but perhaps not all of the wealth effects. In our new study (coming soon) we also include an approach explicitly based on housing units rather than population, which we feel should address this issue somewhat more robustly.

But our ENSO study should provide some confidence about the ability to detect climate signals in the normalized record:

Pielke, Jr., R.A., and C.W. Landsea, 1999: La Niña, El Niño, and Atlantic Hurricane Damages in the United States. Bulletin of the American Meteorological Society, 80, 10, 2027-2033.
http://sciencepolicy.colorado.edu/admin/publication_files/resource-85-1999.14.pdf

Given than there are no trends in hurricane landfall statistics over the past century (a point on which everyone seems to agree), there is not reason to expect a trend in losses due to trends in landfall. So the damage data won't be a good place to find a climate change signal in any case. For more on this subject please see:

http://sciencepolicy.colorado.edu/prometheus/archives/climate_change/000960climate_change_and_d.html

Thanks!

I appreciate the response and new citations. I'm still a bit confused about the definition of the wealth factor, which in the 1998 paper is:

"Wy = wealth factor, determined by the ratio of the inflation-adjusted 1995 fixed reproducible tangible wealth expressed as per capita to that of year y"

Unlike the population factor, which is specified as county-level, I don't see any spatial attributes in the wealth definition -- just a constant nationwide value for each year.

A county-level population variable might capture some county-level wealth effects, but if it did so then the model would overstate the contribution of population changes and understate the contribution of wealth changes, leaving both coefficients biased. There's no reason to believe these biases cancel out, meaning that there is still likely to be a wealth effect residing in the error term. Were you unable to create a county-level wealth factor?

I read the ENSO paper to say that climate (but not climate change) can be shown to be responsible for significant differences in hurricane damages. I'm less sure what to make of this conclusion for policy purposes. You recommend that Congress "set aside supplementary funds for disaster-related costs during years that are identified as particularly active," but Congress doesn't work that way. It is institutionally incapable of setting aside funds for any purpose, and its budget rules exempt emergency supplementals so that routine events now are routinely treated as "emergencies." (Before the Budget Impoundment and Control Act of 1974, presidents were able to set aside funds for emergencies, and they are highly motivated to do so. George H.W. Bush might have been re-elected in 1992 if he had a pile of cash to spend after Andrew. George W. Bush had to go back to Congress for a pork-laden supplemental after Katrina. If Katrina had struck in 2004 instead, he would have been electoral toast.)

Richard-

You have accurately characterized our wealth calculation. We convert national wealth data (this is all that is available) into a per capita basis and then use coastal county population (which is available at the county level). Note that if we are missing some amount of excess wealth accumulation in coastal counties which exceeds national per capita averages over time, that this will tend to suppress the normalized values of storms the further into the past you look. So this would in fact tend to bias the dataset in favor of seeing an increasing trend over time (i.e., past storms in fact would have higher normalized results with faster wealth accumulation in coastal counties than national averages). Given the possible direction of this bias, this provides even stronger evidence that there is no such long-term trend.

But I emphasize that there is no reason to expect such a long-term trend as landfall metrics also show no long-term trends. The fact that the dmaage data reflects the data on the physical characteristics of storms at landfall should also give some confidence in the approach.

The importance of the ENSO study is that it hows that the data is of sufficient quality to be able to identify a climate signal in the dataset. A climate change signal of the same magnitude as the difference between warm and cold ENSO events should in principle be identifiable.

Finally, our suggestion for a different way to think about emergency spending was made in recognition that the current approach does not allow such foresight. There are of course understandable political reasons for the current approach. But at one time FEMA did not use weatehr forecasts to anticipate looming disasters. Now it does. Policies can change.

Thanks!