Towards the Construction of Climate Change Scenarios
- Cite this article as:
- Mitchell, J.F.B., Johns, T.C., Eagles, M. et al. Climatic Change (1999) 41: 547. doi:10.1023/A:1005466909820
- 232 Downloads
Climate impacts assessments need regional scenarios of climate change for a wide range of projected emissions. General circulation models (GCMs) are the most promising approach to providing such information, but as yet there is considerable uncertainty in their regional projections and they are still too costly to run for a large number of emission scenarios. Simpler models have been used to estimate global-mean temperature changes under a range of scenarios. In this paper we investigate whether a fixed pattern from a GCM experiment scaled by global-mean temperature changes from a simple model provides an acceptable estimate of the regional climate change over a range of scenarios. Changes estimated using this approximate approach are evaluated by comparing them with results from ensembles of a coupled ocean-atmosphere model. Five specific emissions scenarios are considered. For increases in greenhouse gases only, the 'error' in annual mean temperature for the cases considered is smaller than the sampling error due to the model's internal variability. The method may break down for scenarios of stabilisation of concentrations, because the patterns change as the model approaches equilibrium. The inclusion of large local perturbations due to sulphate aerosols can lead to significant deviations of the temperature pattern from that obtained using greenhouse gases alone. Combining separate patterns for the responses to greenhouse gases and aerosols may improve the accuracy of approximation. Finally, the accuracy of the scaling approach is more difficult to assess for deriving changes in regional precipitation because many of the regional changes are not statistically significant in the climate change projections considered here. If precipitation changes are only marginally significant in other models, the apparent disagreement between different models may be as much due to sampling error as to genuine differences in model response.