Potential for use of model assessment metrics to reduce uncertainty in multi-model regional predictions for the 10-50 year horizon
In this workstream the question focussed on was - can the large uncertainty in climate change projections for some regions of Africa (from across all models used in IPCC's 5th assessment report) be reduced by disregarding models that have a poor simulation of current climate in these regions? . A large number of metrics were developed and used to assess the performance of more than twenty models for East and West Africa, focussing on seasonal-average temperature and rainfall. It was found that for Africa the method does not provide an unequivocal basis for discriminating among models. The models that do less well at simulating contemporary climate are not generally outliers in the prediction scenarios and so disregarding them does not reduce the uncertainty expressed by the large spread in model predictions.
Although disappointing in terms of potential for practical application, the result is important since it tells us that the underlying assumption in the method may be flawed: i.e. simple metrics designed to characterise current climate processes may not be sufficiently sensitive to the processes important in driving regional climate change - and that we may well do better with more targeted metrics. Our result also sends a caution regarding methodologies for using model outputs in adaptation studies: evaluation using simple metrics of contemporary climate have sometimes been used to deliberately "screen out" models likely to have outlying (possibly unrealistic) future projections - our result suggests this may be a mistaken approach.
Research must now focus on developing means of identifying and assessing the processes leading to the differences in model regional responses to CO2 increases. This will allow the development of process-based reliability metrics that could enable effective ranking of models for particular regions in future IPCC assessments.
Last updated: 12 May 2016