English: This image shows a frequency distribution of climate sensitivity, based on model simulations. Based on the cited Lindsey (2010) public-domain source: To understand how uncertainty about the underlying physics of the climate system affects climate predictions, scientists have a common test: they have the model predict what the average surface temperature would be if atmospheric carbon dioxide concentrations were to double pre-industrial levels (the climate sensitivity).

They run the simulation thousands of times, each time changing the starting assumptions of one or more processes. When they put all the predictions from these thousands of simulations onto a single graph, what they get is a picture of the most likely outcomes and the least likely outcomes.

The pattern that emerges from these types of tests is interesting. Few of the simulations result in less than 2 °C of warming—near the low end of estimates by the Intergovernmental Panel on Climate Change (IPCC). Some simulations result in significantly more than the 4 °C, which is at the high end of the IPCC estimates.

This pattern (statisticians call it a “right-skewed distribution”) suggests that if carbon dioxide concentrations double, the probability of very large increases in temperature is greater than the probability of very small increases.

Our ability to predict the future climate is far from certain, but this type of research suggests that the question of whether global warming will turn out to be less severe than scientists think may be less likely than whether it may be far worse.