The big problem with much of the discussions about trends in hurricane activity is that the databases that everyone is working from are known to have significant inhomogeneities due to changes in observing practice and technology over the years. So it’s not surprising that a new re-analysis (Kossin et al, published yesterday) has been generating significant interest and controversy among the hurricane research community (see e.g. Prometheus or Chris Mooney). However, rather than this study being taken for what it is – a preliminary and useful attempt to make homogeneous a part of the data (1983 to 2005) – it is unfortunately being treated as if it was the definitive last word. We’ve often made the point that single papers are not generally the breakthroughs that are sometimes implied in press releases or commentary sites and this case is a good example of that.
Kossin et al develop an algorithm based on North Atlantic data that can be theoretically used with the coarsest data available from the earlier parts of the record and in more remote regions. While the technique works well in the North Atlantic (picking up almost all of the storms seen in the standard data), it doesn’t work as well in other basins – possibly because the characteristics of tropical cyclones are not universal, or because the coarse early remote sensing data are still not sufficient. The poorer performance in the other basins is surely a reason to anticipate that further work will be necessary to refine these estimates, and should serve as a caution to those wanting definitive conclusions.
How does this fit in with some of the previous work? Well, it confirms the large trend in the North Atlantic (seen in Emanuel, 2005), but doesn’t show significant trends in the other basins (from 1983). This isn’t directly comparable with Webster et al (2005) though, since their trends start in the 1970s, and the shortness of the new reanalysis (only 23 years) emphasizes interannual and decadal variability associated with e.g. El Nino. The Kossin et al study is therefore unlikely to shed much light on the potential global warming/hurricane intensity link.
In summary, read the papers and the comments but don’t believe the hype.
We’ll start off the discussion with a few comments we have already received on the provocative study:
Based upon the new results of Kossin et al. (GRL, 2007), it looks like the IPCC SPM just barely covered itself in its proclamations on observed hurricanes:
There is observational evidence for an increase of intense tropical cyclone activity in the North Atlantic since about 1970, correlated with increases of tropical sea surface temperatures. There are also suggestions of increased intense tropical cyclone activity in some other regions where concerns over data quality are greater. Multi-decadal variability and the quality of the tropical cyclone records prior to routine satellite observations in about 1970 complicate the detection of long-term trends in tropical cyclone activity. There is no clear trend in the annual numbers of tropical cyclones.
From the results presented in Kossin et al. the “suggestions” of increases in intense tropical cyclone activity in regions other than the Atlantic basin are not really so well supported, at least for the last 23 years.
(disclosure: I have, to some degree, been funded by the fossil fuel industry since 1992)
1) The methodology is trained on the Atlantic. It has no parameters to allow for different structures or size of storms, and there is no good reason why it should work well on storms in other basins. Given the different land-sea configurations and the different role of ENSO in the different basins, and the fact that disturbances in other basins do not form from easterly waves from off of Africa, there is every reason to expect that storms in other basins have different characteristics. For instance, there is greater activity in the Pacific Northwest, and the tropopause is higher in the western Pacific, and this affects brightness temperatures at tops of clouds. If the size of storms differs then the fixed form of EOFs will not be able to capture that form. The analysis must be able to account for differences among basins in order to have confidence in variability or trends. It would be easy enough to test whether the storms in other basins had different characteristics by also performing an EOF analysis for each region. This basic test was not done. It should be.
2) The results are suggestive of these problems. In the SIO where the method gives 0, 1 or 2 storms vs up to 6 in the best track data, there is a serious bias. Similar large biases exist in the SPAC (up to 2 vs 5 in best track). Obviously the threshold is effectively different and it is a comparison of apples and oranges.
3) In addition, this version of the paper deals with PDI. The earlier version of the paper dealt with intensity of storms and that was abandoned because the results were not very good. In particular, the presumption is that the older results were the problem because operational methods have improved. But the Kossin et al results showed bigger and greatest discrepancies with those from best track in recent years: there is no convergence over time. This is harder to see with PDI, because the biggest storms are emphasized, but the question of why is there not good agreement in recent years is not answered.
3. Dr. Judith Curry:
The most vexing thing about the tropical cyclone data sets is the uncertainty that analyst subjectivity contributes to this. The Dvorak scheme for determining tropical cyclone intensity is notoriously subjective, see the recent BAMS article on this. The importance of what Jim Kossin has done is to take this subjectivity out of the analysis.
Kossin’s method matches well the historical data in the North Atlantic (NATL) and East Pacific (EPAC). The method was trained using North Atlantic data, and the East Pacific regime in terms of dynamical and thermodynamical conditions is very close to the North Atlantic conditions. However, Kossin’s method diverges substantially from the established data sets in the Western Pacific, South Pacific, and Indian Oceans. Does this mean that the established data sets or in error, or that Kossin’s method (trained in the Atlantic) does not translate well to the other ocean regions?
Owing to problems with dealing with historical satellite data, Kossin’s study was extended only back to 1983 (the period for which the satellite data are well calibrated), and it is almost certain that this data set cannot be extended back prior to 1977. By itself, this data set is too short to say anything about a trend in intensity. But it can in principle be used to assess uncertainties in the established data sets.
My own analysis of the discrepancies has focused on the Western Pacific (WPAC) data, where 40% of the global tropical cyclones form. During the period 1983-1987, Kossin’s data overlaps with aircraft reconnaissance data; I would expect the WPAC TC data during this period to be of comparable quality to that in the NATL and WPAC, but large discrepancies are seen. I would also expect the agreement to be better in more recent years with the advent of more sophisticated satellite systems, but the discrepancies are largest during the most recent period. Its would not be surprising for this method trained in the NATL not to work well in the other regions. The method does not allow for different structures or sizes of storms in different basins. The NATL cyclones form primarily from easterly waves, while those in other regions do not. The role of ENSO is different in the different basins. The tropopause height is higher in the WPAC. etc.
Many people in the tropical cyclone community have questioned the Emanuel and Webster et al. papers owing to uncertainties in the data sets. Other than anecdotal analyses, little has been done to quantitatively assess the uncertainties. Kossin’s paper is arguably the first important word on this subject, but it certainly won’t be the last word. To establish the credibility of Kossin’s data set outside the NATL, considerably more analysis is needed to understand discrepancies in individual basins and the nature of the discrepancies on a storm-by-storm basis.