The Earth may be much more sensitive to global warming than previously thought, according to the first results from a massive distributed-computing project.
The project tested thousands of climate models and found that some produced a world that warmed by a huge 11.5°C when atmospheric carbon dioxide concentrations reached the levels expected to be seen later this century.
This extreme result is surprising because it lies far outside the 1.4°C to 4.5°C range predicted by the Intergovernmental Panel on Climate Change (IPCC) for the same CO2-level increase - a doubling of CO2 concentration from pre-industrial times. But it is possible the IPCC range was wrong because its estimate is based on just a handful of different computer models.
"We have anecdotal evidence that people tend to tune their models to be similar to other people's," says David Stainforth, from the University of Oxford, UK. "Nobody wants to have a model that's terribly different, particularly when there are only 8 or 10 in the world," he explains.
Stainforth and his colleagues set up www.climateprediction.net to see what happened when models were not tuned in this way. They start with a climate model that divides the Earth's surface into boxes hundreds of kilometres square and then change some of the 29 or so parameters that govern aspects of the atmosphere and weather.
These tweaked models are farmed out to volunteers who run them on their home computers via a screensaver. Models that accurately simulate today's climate are then dosed with carbon dioxide, to give double pre-industrial levels, and projected forward 45 years to see how the climate responds.
Since the project launched in 2003 (New Scientist, 12 September 2003) more than 95,000 people from over 150 countries have donated spare computing time. The results have now come back from 60,000 simulations and the team have analysed around 2000 of these, focussing on six parameters.
While most of the models showed the global mean temperature rising by between 3°C and 4°C, some experienced much stronger heating. "When you see large areas of the northern hemisphere at 11°C above pre-industrial levels, you think this is quite scary," says Stainforth.
Geological data shows the Earth's climate has been much warmer in the past. Temperatures were around 6°C higher during the Cretaceous period, for example, when dinosaurs roamed the Earth. But Bob Spicer, an expert in the palaeoclimate at the Open University in Milton Keynes, UK, says there is no evidence that temperatures have ever been as high as in some of the climateprediction.net simulations.
Some iterations of the models showed the climate cooling after an injection of CO2, but these were discarded after close examination because the temperature fall resulted from an unrealistic physical mechanism, says Stainforth. In these scenarios, cold water welling up in the tropics could not be carried away by ocean currents because these were missing from the models.
There are no obvious problems with the high temperature models, he says. The climateprediction.net team were left with a range of 1.9°C to 11.5°C. "The uncertainty at the upper end has exploded," says team-member Myles Allen.
Clouds, which climate scientists have already recognised as the Achilles' heel of climate prediction (New Scientist, 24 July 2004), were the main cause of the variability in the high temperature models. The two most sensitive parameters governed the humidity at which clouds form and convection in the tropics. More observations of these critical processes could now help to narrow the uncertainty in the climate models' prediction.
But the climateprediction.net team stress that they are not saying we will see double-digit temperature rises if CO2 emissions go unchecked. "We're saying we can't rule it out," says Stainforth.
The next batch of climateprediction.net models will be making predictions of the timescales over which these changes might be seen, says Stainforth. "We need a lot more help, and we encourage people to continue getting involved."
Journal reference: Nature (vol 433, p 403)