/Science/, Vol 307, Issue 5709, 497 , 28 January 2005 [DOI: 10.1126/science.307.5709.497a]
GREENHOUSE WARMING: Climate Modelers See Scorching Future as a Real Possibility
*Richard A. Kerr*
Researchers tapping the computer power of 26,000 idling personal computers are confirming that a searing heating of the globe in the coming centuries can't be ruled out. Their new twist is twofold. It could get even hotter than the previous worst case had it. And, contrary to earlier work, no modeler can yet say that such an extreme scenario is any less likely than the moderately strong warming that most climate scientists expect. That shakes up what had seemed to be an emerging consensus, although some balk at such an extreme perspective.
Last summer, a gathering of climate modelers, paleoclimatologists, and other climate researchers seemed to be converging on a middle ground (/Science/, 13 August 2004, p. 932 <http://www.sciencemag.org/cgi/content/full/305/5686/932>). Three different kinds of studies--a collection of the latest expert-designed climate models, perturbations of a single model, and studies of past climate--addressed the question of how sensitively the climate will respond to increasing greenhouse gases. All three seemed to be pointing to a moderately strong sensitivity. When greenhouse gases are doubled, it appeared, warming would be something like 2ºC to 4°C. A low, nearly harmless climate sensitivity much less than 2ºC seemed quite unlikely, and an extreme one of more than 6ºC or 7ºC was possible but far less likely than the consensus range.
Now comes modeler David Stainforth of the University of Oxford, U.K., and 15 colleagues with their pumped-up version of the perturbed-parameter approach. In a similar study reported at last summer's workshop and in /Nature/, James Murphy of the Hadley Center for Climate Prediction and Research in Exeter, U.K., and colleagues altered 29 model variables that control physical properties such as the behavior of clouds, atmospheric convection, and winds. Given their available computing resources, they had to perturb only one parameter at a time for a total of 53 simulations. In order to extrapolate their results into the high-sensitivity range, they had to assume that changes in two parameters simply add up.
In the Stainforth study, published this week in /Nature/, the modelers varied six parameters several at a time so that they could explore nonlinear interactions between parameters. The resulting 2578 simulations used in the study were run on personal computers whose owners--members of the general public--contributed their idle processing capacity to the climate/prediction/.net experiment.
In this distributed-computing study, perturbed parameters did in fact interact nonlinearly to heighten climate sensitivity. None of the simulations had a climate sensitivity less than 2ºC, in line with expectations. And most simulations did not fall far from the model's sensitivity of 3.4ºC when run with no parameters perturbed. But the inevitable long tail of results on the high-sensitivity side ran out to 11ºC, 2ºC farther than any kind of study before it.
How likely is an 11ºC sensitivity? "We can't yet give a probability for our results," says Stainforth. "Our [high-end] results are very sensitive to our prior assumptions," such as which parameters are perturbed and by how much. Previous studies suffer from the same limitations, he says.
Other climate researchers will take some convincing. "I just can't believe climate sensitivity is 10ºC," says paleoclimatologist Thomas Crowley of Duke University in Durham, North Carolina. Climate's responses to past, natural changes in greenhouse gases or equivalent climate drivers, such as volcanic eruptions, have just been too modest for that, he says. Until modelers can rein in their unruly simulations--perhaps using the tens of thousands of additional perturbed-parameter simulations now in hand at climate/prediction/.net--Crowley will stick with a moderately strong warming.
-- www.e-laser.org Laser@inventati.org
