data recovery

I see nothing in his conclusions that looks unreasonable. He plans to do a similar exercise using Kossin’s global data, which would be good.

There’s so much bad science around in climatology, mainly due to political considerations, it’s a pleasure to see some genuine scientific methodology being applied to the controversial issue of storms.

There is a paucity of accurate data in this area and much of the speculation is based on poor quality information. Anyone who can shed some light on this difficult area deserves a big thanks. I will continue to watch how this progresses with interest!!!

The NHC is making an increasing number of storm wind maps available to the public. They can be found here (select a year, say 2007, and then view the maps. Some years are populated and some are not).

I picked 14 storms from the populated years, gridded the maps and then calculated (by hand) the size of their regions of storm-force winds. Not fun but it works.

An X-Y plot of the results is here . The x-axis is hours after formation while the y-axis is the area experiencing storm-force winds. Each blue dot represents a storm at a point in time. Note that this is a sampling for illustration purposes and needs more data points (a useful project for a student). The plot shows (green line) the median area used in VK07 for tropical storms, as well as the median of the data points.

Note that these young storms, while expanding, fall well below the assumption in VK07. That creates two problems:

1. The short-duration storms, on average, simply lack time to grow to the size assumed in VK07.
2. All storms likely have their size over-estimated by VK07 for their first 12 or so hours. While this may seem like a small matter, recall that this initial time may represent 5 to 10% of the life of a storm which, with a large population, makes a difference.

(Aside: a plot which converts area into inferred radius is here . It may be that, in the VK07 technique, radius is more important than area – hard for me to tell. Note that VK07 uses a 43-mile radius cutoff (nothing smaller than that) which misses a number of the data points.)

(Another aside is that I believe they assume that any storm which nears land will be detected. While reasonably true for normal storms there’s evidence that this does not hold for short-duration storms.)

Anyway, VK07 is a good paper which I think has a young-storm problem to be sorted out. If I’m correct then the VK07 technique probably under-estimated the number of “missed” storms.

I mentioned my concern in #671 that KM04 (and thus VK07) used data based on the NHC forecast advisories (sometimes called marine advisories, with an example here ).

My concern is that these real-time advisories are intended as warnings and thus tend to err on the side of conservatism, meaning that they tend to overstate storm wind radii. Also, they are based on the best available (but limited) information at the moment the advisory is issued and involve some guesswork.

The NHC now issues wind maps for storms, based on after-action reconsideration and analysis of all the data. These are about as close to “true” radii values as we have.

So, to see if my concerns are warranted, I checked twelve storms from 1998-2000, a period of available wind maps. These are storms that never made it to hurricane strength. I compared these values against the radii given in the advisories/KM04/VK07.

An X-Y plot of “true” radii versus forecast-advisory radii is given here . If there is no bias then the values should evenly scatter around the equalization line.

In this case, however, the points almost all reside in the lower half. That is the region where advisory radii are greater than the true radii. This indicates a tendency for the advisories to overstate storm wind radii, which causes a problem for VK07.

Admittedly the sample size is small. I plan to expand it to cover all available storms but I imagine that the pattern will hold.

Before I bury some recent work into the muck of my hard drives I’ll offer three plots regarding the recent Vecchi Knutson 07 (“VK07″) paper.

VK07 is a good paper. But, I’ve expressed a concern about VK07′s use of subtropical storms and a nagging concern about the sizes of tropical storms used by VK07, especially newly-formed tropical storms.

As best as I can determine, based on my reading of VK07, the distribution of tropical storm radii observations is represented by this plot . Note that none are shown with a mean radius below about 40 miles. Also, a significant number of storm observations are shown with radii greater than 100 miles.

Here is a plot of the VK07 model with the actual observational data (red) overlaid. And here is the same plot with a couple of notes (green) added.

One concern is VK07′s lack of very small storm observations (left side of graph) which are especially important in newly-formed storms. Also, VK07 is heavy on large-radii observations, which does not agree with the observational data. I think this all leads to VK07 “under-detecting” tropical storms in the pre-satellite era with their methodology.

Beyond this, I suspect that the raw data (which is based on marine advisories, which are mainly intended as a tool to warn ships) over-states the actual extent of storm winds, especially with poorly-defined and asymmetric new storms. The windfield maps now available from the NHC are probably a better source for tropical storm parameters.

The net effect of all of this is that VK07, despite its good efforts, may still have underestimated small, weak, short-duration storms in the early decades of its study.

At some point I’ll put the NHC windfield maps to use and also write down my notes, maybe on the CA auditblog, before re-burying this vampire