« Disasters »

Hurricane Trends, Frequency, Prediction

This post is a slightly edited version of some random musings on hurricane science that I have shared with the Tropical Storms discussion list.

A few thoughts come to mind from the latest round of exchanges on the list.

1. Detection of trends

I call your attention to a recent paper by Rob Wilby:

Wilby, R. L. 2006. When and where might climate change be detectable in UK river flows? Geophys. Res. Lett., Vol. 33, No. 19, 14 October. (PDF)

. . .under widely assumed climate change scenarios, expected trends in UK summer river flows will seldom be detectable within typical planning horizons (the 2020s). Even where climate driven changes may already be underway, losses in deployable resources will have to be factored into long-term water plans long before they are statistically detectable.

Specifically, Wilby finds that "Assuming no change in variance, annual mean [river flow] must change by 38-46% by 2025 or 28-33% by 2055 to be detectable." These are huge numbers for changes in means and they assume perfect data quality. Wilby explains this result as "The long detection times for trends in UK river flow are due to the low signal-to-noise ratio of hydro-climatic time series at basin scales."

What does this mean for detection of trends in tropical cyclone intensity?

Assuming as the recent WMO report does that there is a 3-5% increase in windspeed for every degree increase in SST, and given that (a) windspeeds are, to date at least, measured with certainties that are arguably not at this level of precision (sometimes using S/S categories that are up to 5 times as large as these values), and (b) SSTs themselves are a noisy series, see e.g. this paper, this community may be in a situation where the science is fundamentally "underdetermined." I think that everyone in this community would benefit from an understanding of the notion of "underdetermination" in science and what it signifies for scientific debates. For an intro see this brief discussion.

For these reasons I have come to the conclusion that the search for detection of trends in TC activity will no doubt motivate much interesting research, but cannot result in a comprehensive community consensus anytime soon.

2. A Theory of (Maximum Potential) Frequency

If the above points may seem pessimistic (well, they are;-) the community need not throw up its hands. It seems to me that a way out of this quandary is for even greater attention to be paid to the development of a theory of maximum potential frequency to parallel MPI (maximum potential intensity) theory. I may be mistaken, but as an outsider looking in, it seems that much of the community is quite comfortable with existing MPI theory and is largely hopeless about a theoretical understanding of frequency (those working on further exploring MPI and developing a theory of frequency, please forgive, I am making a general point).

It also seems to me, perhaps naively, that intensity and frequency cannot be treated as independent, and continuing to treat TC research in this manner is an obstacle to further advances.

I would be interested in any efforts to develop a basin-by-basin theory of maximum potential frequency. For the Atlantic basin for instance, what is the maximum theoretically possible number of TCs that can develop during a single hurricane season, and why? Even starting with the classic six (?) conditions for development provides an upper constraint on the number of developing systems in a particular season. Inherent in the notion of a "season" are the seeds of an MPF theory. Consider that in the Atlantic 2005 saw 28 named storms, could there be 35? Why or Why not? 40? How about 5? Zero?

How about it?

3. The Importance of Prediction

In recent decades, Bill Gray, and others, have drawn a line in the sand, by arguing that in a particular basin the best that can be done in terms of expecting future activity is not to be found based on theory-based models but based on an understanding of statistical relationships which may not be fully understood from a theoretical standpoint. Debates about statistical vs. dynamical approaches to prediction occur across the sciences, and are no different here.

It seems to me that if this community wants to make progress on the debate over hurricanes and climate, it will necessarily have to engage in predictions of the future to a far greater degree that it does now, much as the ENSO community has done. There will be limited successes at first and successes and failures determined by luck. There will also have to be the careful management of public and policy makers expectations.

By predictions I mean - can anyone devise a methodology that can systematically beat out Gray, Saunders, Elsner, NOAA, climatology etc.? Such predictions might be seasonal, multi-year, or longer, but they should be verifiable by actual data on time scales that allow for feedback into the process of research. The experiences of the ENSO community are very instructive (and somewhat humbling) along these lines. If the long-term climate (i.e., over several years and longer) of TCs is indeed nonstationary, then over time those who base their predictions on historical statistical relationships will produce predictions whose skill should be easily exceeded by those using dynamical methods. In practice, in many fields, achieving such success has been difficult -- compare managed mutual fund performance to the naive baseline of the S&P 500, for instance!

Generic predictions about what will happen under 2XCO2 in 2100 are great, but they are unfalsifiable by experience on research timescales, feeding the problem of underdetermination. The alternative to making scientific predictions is that we perpetuate the state of underdetermination in this community and risk detaching ourselves from the fundamentals of this important aspect of the scientific method.