By Rasmus Benestad & Michael Mann
Just as Typhoon Nargis has reminded us of the destructive power of tropical cyclones (with its horrible death toll in Burma–around 100,000 according to the UN), a new paper by Knutson et al in the latest issue of the journal Nature Geosciences purports to project a reduction in Atlantic hurricane activity (principally the ‘frequency’ but also integrated measures of powerfulness).
The close timing of the Knutson et al and Typhoon Nargis is of course coincidental. But the study has been accorded the unprecedented privilege (that is, for a climate change article published during the past 7 years) of a NOAA press conference. What’s the difference this time? Well, for one thing, the title of the paper: “Simulated reduction in Atlantic hurricane frequency under twenty-first-century warming conditions” (emphasis added).
The idea that climate change might actually decrease the frequency of tropical cyclones (TCs) is not an entirely new idea. Indeed, similar conclusions have been reached in earlier work using climate model projections (e.g. Yoshimura et al., 2006, J. Meteorol. Soc. Japan; Bengtsson et al., 2006 J. Clim.; Chauvin et al., 2006 Clim. Dyn.). So what are the key developments in this latest work?
Before we tackle that question, it is helpful provide a bit more background on the problem. First, it needs to be recognized that GCMs are too coarse to provide a realistic description of ‘small-scale’ (mesoscale) features such as TCs. The atmospheric components of climate models were never really designed for the study of TCs, but the fact that they can produce features with TC-like character when run at sufficiently high resolutions, gives us increased confidence in the possibility that climate models can be used to analyze climate change impacts on TCs. In order to get a more realistic description of the TCs in coarsely resolved climate models, one needs howeveer to ‘downscale‘ the model results.
Knutson et al project future changes in Atlantic TC behavior by using a regional climate model (RCM) which produces tropical cyclones (though ones that are too weak–see discussion below) to ‘downscale’ climate change impacts. This is accomplished by driving the RCM with boundary conditions provided from the various 21st century model projections described in the IPCC 4th Assessment report (IPCC AR4).
Contrasting two recent studies
In certain respects, this new paper is closely related to a paper published last month by Emanuel et al in the Bulletin of the American Meteorology Society (‘BAMS’ to those in the know) which received some press of its own (some of it quite distorted). Emanuel et al . also use a downscaling approach applied to more-or-less the very same climate model simulations. And both studies project a decrease in the frequency of Atlantic tropical cyclones (though see caveats below). But here is where the similarities end.
Emanuel et al use a very different downscaling approach. The use a ‘seeding’ method to randomly generate small vortices analogous to ‘short wave’ tropical disturbances in the real world (the tracks they take are defined in terms of the background atmospheric circulation of the model combined with the so-called ‘self advection’ of the TC itself). They define the probability of development of these vorticies into TCs through a ‘genesis’ model that conditions the favorability of development on various characteristics of the background climate state, and they use a theoretical model to predict TC intensities.
The differences between the conclusions of the two studies are significant. Using essentially the same IPCC model projections, the two studies come to very different conclusions with regard to key projected quantities, such as the seasonally-integrated powerfulness of TCs or ‘power dissipation index’ (PDI). While the Emanuel et al study predicts a clear increase in PDI, the Knutson et al study does not. So which is right?
Well, Knutson et al fully acknowledge that their RCM still has too low resolution to produce realistic TCs (the model resolution is about 20 km, while theoretical estimates indicate that a resolution of about 1 km is likely required to simulate the inner core of intense TCs). TCs are very likely being artificially prevented from intensifying in a warmer climate in the Knutson et al study because of this. By contrast, Emanuel et al‘s approach does not suffer from such resolution limitations.
To their credit, Knutson et al openly acknowledge this weakness in their treatment of TC intensity and PDI. What about their conclusions about a projected decrease in Atlantic TC frequency, which are, after all, the central point of the paper? Here we have reservations as well (and if we were the betting kind might even put forward a wager with regard to future trends). In part, these reservations are a result of the very same issues. The limitations of the RCM for example, as the authors note, also lead to incorrect seasonal and geographic distributions of TC genesis.
In the real world, small-scale phenomena such as convection, clouds, gravity waves, and various sorts of eddies may influence mesoscale systems such as TCs, and these unresolved small-scale phenomena are represented through often somewhat simplistic statistical ‘parameterizations’. This limitation is of course common to essentially all atmospheric models, and in and of itself is no reason to dismiss the conclusions of the study.
But TCs do also play a role in terms of the larger scales, as they facilitate transport and a redistribution of heat, moisture and momentum (‘upscaling’). This action is simulated more explicitly in RCMs and more implicitly in GCMs by their parameterization schemes, but it is still not really known if these two levels of modelling provide a physically consistent picture of the scale interactions. The RCM solutions, however, are constrained by the results generated by the GCMs and thus depend on how well the parameterization schemes capture this upscaling effect. Yoshimura et al., 2006 have shown that the solutions may be sensitive to the choice parameterization schemes: they found an increase in TC number over the Indian Ocean if the model used the Kuo cumulus parameterization but a decrease if the Arkawa-Schubert cumulus parameterization scheme was used.
Also significant, perhaps, is the seasonality issue touched on above. The Knutson et al study involves an assumption of a fixed August-October TC season. Yet one important impact of large-scale climate factors which influence Atlantic TC frequency such as ENSO and the NAO (see e.g. one of our own papers on this topic) is their influence on activity during the latter part of the Atlantic hurricane season (one might argue for example that the primary reason for the low TC count during the 2006 season was the early ‘shut-down’ of the season due to increasingly strong El Nino-related wind shear in the autumn as the El Nino set in).
The role of ENSO
ENSO itself, and how it’s influences are represented in the analysis, is potentially an even more fundamental issue. It is well known (and openly acknowledged in both the Emanuel et al and Knutson et al studies) that tropical Atlantic TC frequency is heavily influenced by ENSO variability. This is primarily through its influence on vertical wind shear in the Caribbean and tropical Atlantic, which in turn determines how favorable of an environment incipient TCs encounter as they form and intensify. We have discussed this here in detail before.
Given that ENSO is the dominant source of variability on interannual timescales, it is likely that future changes in ENSO (more specifically, the mean state of the climate and whether it is more “El Nino” or “La Nina” like, i.e. is there a strengthened or weakened ‘Walker Circulation’) could have a profound influence on Atlantic TC frequency. Although the IPCC models project overall a more El Nino like mean state with a weakened Walker circulation, there is far from a consensus among the models. Several credible state-of-the-art coupled models project precisely the opposite. And all of the models used in the IPCC assessment suffer to a varying extent from certain fundamental biases (the inability to produce a realistic ‘ITCZ’ over a large part of the equatorial Pacific ocean–the so called ‘split ITCZ problem’).
The CMIP3 model projections are essentially evenly split as to whether they project an increase or decrease in the magnitude of individual El Nino and La Nina events. Yet the frequency of large El Ninos and large La Ninas means everything in terms of the likelihood of very active Atlantic tropical storm seasons. If all of this sounds familiar to you, its because we made essentially the same point about a year ago in response to a paper that was more or less making the same argument as Knutson et al, though not quite as fleshed out.
The fact that the RCM-based downscaling approach can reproduce the observed changes when fed modern reanalysis data is used by Knutson et al as a ‘validation’ of the modeling approach (in a very rough sense of the word–there is in fact a non-trivial 40% discrepancy in the modeled and observed trends in TC frequency). But this does not indicate that the downscaled GCM projections will provide a realistic description of future TCs in combination with a multi-model GCM ensemble mean. It only tells us that the RCM can potentially provide a realistic description of TC behavior provided the correct input.
Indeed, other purely statistical approaches using large-scale climate predictors of Atlantic TC activity, and which seem to imply different relationships between projected climate change and future Atlantic TC activity (more on this in the future!), also pass similar validation tests with flying colors. So validation against the modern record alone (be it with a dynamical or statistical model) cannot demonstrate the reliability of the future projections. It can only indicate the self-consistency of the analysis.
Scientifically, where do we go from here? How do we achieve greater clarity on these issues? Obviously, there is need for significant increases in resolution of the RCMs, as past studies indicate a significant sensitivity of results to model resolution. The arguably required, aforementioned 1 km resolution may not be practically achievable in the near term, but the community must strive to move in that direction, particularly if projections of future changes in TC strength, intensity, and power dissipation are to be useful.
Better yet would be to run the coupled ocean-atmosphere models themselves at very high resolution (e.g. 10 km or even finer). This could in principle eliminate many of the thorny issues discussed above, including the potential artifacts of using embedded models with one-way only coupling. But this may be wishful thinking, at least for the foreseeable future.
Of most fundamental significance to assessing the reliability of these current projections, in our view, is the “junk in/junk out” factor. The detailed projections made using either the RCM approach of Knutson et al or the ‘random seeding’ approach of Emanuel et al, can only be as good as the large-scale scenarios used to drive them. And since key aspects of those large-scale scenarios as far as Atlantic TC activity is concerned (i.e. what really happens to the ENSO mean state and amplitude of variability) are currently not confidently known, neither can we be confident using the model projections to say what will happen to Atlantic TC activity in the future.
In this respect, we have to consider the entirety of currently available evidence that can inform our assessment of climate change impacts on Atlantic TCs. We know, for example, from the work of Santer et al. that the warming trend in the tropical Atlantic cannot be explained without anthropogenic impacts on the climate. Knutson et al. do not contest this. Furthermore, they do not dispute that the late 20th century increase in Atlantic TC frequency is tied to large-scale SST trends (though they argue that the influence may be non-local rather than local). So we know that (i) the warming is likely in large part anthropogenic, and (ii) that the recent increases in TC frequency are related to that warming. It hardly seems a leap of faith to put two-and-two together and conclude that there is likely a relationship between anthropogenic warming and increased Atlantic TC activity.
What Knutson et al are asking us to do in essence is to put all that aside (because, they argue–in short–that its not the warming but the pattern of warming that matters here) and instead take on faith the perhaps not-much-more-than 50/50 proposition that the mean changes in ENSO state and variability projected by the IPCC multimodel ensemble (which are a key determinant in the projected future Atlantic TC activity) should be trusted.
Given these considerations, we would argue that coastal homeowners, insurers, the re-insurance industry, and every other potential stakeholder in this debate would be wise not to take false comfort from the notion (which the headlines resulting from this paper will inevitably feed) that climate change poses no future Atlantic hurricane threat. In fairness to Knutson et al, they do explicitly point out that their projected decrease in frequency is mostly coming from the weak end of the TC intensity spectrum. In principle, therefore, we imagine that they might perhaps even agree with this message themselves. Indeed, we invite them to comment here!
Update: The authors of the paper have put out an FAQ about their research.
Correction (5/21/08): It has been brought to our attention by NOAA representatives that the NOAA press conference for the Knutson et al paper was not an unprecedented event in recent years. In fact, similar press conferences were held for two other papers (both also questioning the premise that climate change is likely to lead to an increase in tropical cyclone activity) by Wang and by Vecchi and Soden