Climate Audit

Earlier we noted that the number of hurricane-days in 1933 actually exceeded the number of hurricane-days in 2005. However, the 2005 PDI was significantly higher than 1933 PDI, which indicates fairly trivially that hurricane speeds in 2005 were estimated to be higher than 1933. So here is a histogram of 1933 compared to 2005 wind-speed counts in 5 knot increments – first a 1933 histogram, then a 2005 histogram and then the same data side-by-side. Obviously the 2005 distribution is skewed towards higher wind speeds. When you cube the wind speeds in a PDI calculation, this difference is exacerbated.

The shape of the 2005 histogram is actually a little odd – it’s got a bimodal look that surprises me. Does anyone know what is the explanation for this?  There definitely seems to be a bias to measurement in even 10s that is more pronounced in the 2005 measurements and which seems surprising given that these are supposedly scientific measurements. The peak in the 110 knot tranche is also very remarkable in the 2005.

1933 was a notable year in other respects. As I recall, it is still the record year for U.S. temperature. It was the bottom of the Great Depression, with a dust bowl in the Prairies.

Here’s a figure showing Judith Curry’s PDI as compared to calculations from my collation of the ATL track data previously archived, also comparing PDI – which is the integral of wind speed cubed against other measures: a count of hurricane-days, an integral of wind speed and an integral of wind speed squared (which I think is what they call “ACE”). I’ve plotted the data back to 1851 ti see what it looks like prior to truncations by Curry, HW, Eamnuel or others.

All of this is plotted here for the Atlantic and thus is biased through progressively improved measurements of eastern storms through the century. I haven’t done any smoothing since the data is simple enough that people should be allowed to look at the raw data – rather than take the risk that impressions are introduced through smoothing artifacts such as Emanuel’s end-point pinning.

As an impression, it looks to me like the transformations by which wind speeds are summed, then squared, then cubed, results in a difference between late 20th century levels and (say) late 19th century levels that isn’t present in the simple count. This indicates that the count of hurricanes is similar but that the average speed and/or average length of hurricane has increased in the later period. Obviously the measurement of wind speeds in the late 19th century was not done in the same way as in the late 20th century. Without parsing through the data on every storm, I don’t see how one can draw any conclusions based on this data as to whether the differences result from technical changes or from climatological changes.
Continue reading →

The following is from Judith Curry:

Whether changes in the characteristics of tropical storms observed in the last few decades [1,2] are the result of only natural variability, due to climate change, or a combination of both factors is the subject of intense debate [3]. Central to the debate is the quality of the tropical cyclone data [3,4]. Continue reading →

Perhaps the key issue in Emanuel 2005 (as observed in Landsea’s 2005 comment) was the adjustment to the ATL data set performed by Emanuel (citing an old publication, Landsea 1993). We’ve snickered a bit recently at Martin Juckes using older data rather than newer data – only in climate science would this seem possible. And we see another such incident in Holland and Webster – this time with wind speed methodology – where they use a method of Landsea 1993 rather than the recommendations of Landsea 2005.
Holland and Webster stated:

We use the “best-track” tropical cyclone data base from the National Hurricane Center (Jarvinan et al. 1984). The only changes to the data set data have been to include the intensity corrections recommended by Landsea (1993).

Sounds innocent enough. But underneath this is the raging debate between Emanuel and Landsea about adjustment of wind speeds in the Hurdat database. I reviewed Landsea 1993 and Landsea 2005 here which, upon re-reading, stands up as a pretty reasonable representation of the debate. As noted in my earlier post, Landsea 2005 stated:

It is now understood to be physically reasonable that the intensity of hurricanes in the 1970s through to the early 1990s was underestimated, rather than the 1940s and 1960s being overestimated. To examine changes in intensity over time, it is therefore better to use the original hurricane database than to apply a general adjustment to the data in an attempt to make it homogeneous.

Despite this explicit statement from Landsea, Holland and Webster apply the method of Landsea 1993 without even citing Landsea 2005. Whether Landea 2005 is right or wrong, this adjustment is a topic that any competent data analyst needs to deal with before analyzing trends; however, Holland and Webster simply ignore the issue.

I noticed this matter when I experimented with a plot of the average length of a ATL cyclone and the average length of an ATL hurricane (which I examined in the course of examining the supposed constancy of proportion of hurricane-days to cyclone-days.) The average length of an ATL hurricane declines in the late 1960s, coinciding with the issue raised in Landsea 1993. Holland Webster presumably reduce wind speeds prior to 1970, but don’t provide any details on the adjustment other than what I’ve quoted. Landsea 1993 is specific to the aircraft reconnaissance period – but one would need to check whether H and W adjusted wind speeds for earlier periods as well.
The early years with high hurricane-days mentioned previously – 1886, 1887, 1933 – were not done with aircraft reconnaissance and would not, on the face of it, be subject to Landsea 1993 adjustments in any event. But H and W would need to be checked somehow to see if they adjusted data prior to the aircraft reconnaissance period as well. The results that I’ve presented so far have been based on the methodology recommended in Landsea 2005 – using the original Hurdat data – and will therefore not reconcile precisely to HW results (but are obviously a valid sensitivity study that the original data analysts should have done as well). I urge people interested in the topic to read or re-read my earlier post on Landsea 1993 versus Landsea 2005, as this affects H and W as well.