Sir

Stephen Schneider argues in his Commentary “What is 'dangerous' climate change?” (Nature 411, 17–19; 2001) that the Intergovernmental Panel on Climate Change should assign subjective probabilities to the “storyline” scenarios of future climate change offered in its recent special report on emissions scenarios (SRES), because probabilities will be required by policy-makers.

Schneider raises an important concern, in that a set of plausible scenarios does not, by itself, tell policy-makers what is likely to happen or provide clear guidance for action. But his proposal does not take into account that decision-makers must form a climate policy acceptable to groups with many different, yet plausible, estimates of the likelihood of alternative futures.

This condition of deep uncertainty differs from many risk-management problems, in that little solid information exists to inform subjective probabilities for the long-term economic, social and technological trends underlying different greenhouse-gas emission scenarios. It is unlikely that scientific evidence will soon resolve the assumptions about the socio-economic future made by different groups.

Under such conditions of deep uncertainty, decision-makers often rely on robustness — that is, strategies that work reasonably well compared with the alternatives across a wide range of plausible scenarios (R. J. Lempert & M. E. Schlesinger, Climatic Change 45, 387–401; 2000). They then form a policy solution that works reasonably well for all possibilities. In contrast, an optimum strategy requires probabilities for alternative scenarios, which can invite discord, as well as leading to policies that can fail in some plausible futures.

In our view, the SRES team was correct to avoid assigning probabilities and instead recommend use of ensembles of multiple scenarios to capture what is known about the long-term climate future. Researchers can now use such scenarios to assess the robustness of alternative, near-term climate policies. Such studies are likely to suggest near-term policy choices that are largely insensitive to many uncertainties. Only when such studies also reveal a few key assumptions to which near-term choices are particularly sensitive, can scientists take the perilous step of putting limits on the range of subjective probabilities for those few assumptions.