Can climate projection uncertainty be constrained over Africa using metrics of contemporary performance?

Projections of climate change over Africa are highly uncertain, with wide disparity amongst models in their magnitude of local rainfall and temperature change, and in some regions even disparity in the sign of rainfall change. This has significant implications for decision-makers within the context of a vulnerable population and few resources for adaptation. One approach towards addressing this uncertainty is to rank models according to their historical climate performance and disregard those with least skill. This approach is systematically evaluated by defining 23 metrics of model skill and focussing on two vulnerable regions of Africa, the Sahel and the Greater Horn of Africa. Some discrimination in the performance of 39 CMIP5 models is achieved, although divergence amongst metrics in their ranking of climate models implies some uncertainty in using these metrics to robustly judge the models' relative performance. Importantly, when the more capable models are selected by an overall performance measure, projection uncertainty is not reduced because these models are typically spread across the full range of projections (except perhaps for Central to East Sahel rainfall). This suggests that the method's underlying assumption is false, this assumption being that the modelled processes that most strongly drive errors and uncertainty in projected change are a subset of the processes whose errors are observed by standard metrics of historical climate. Further research must now develop an expert judgement approach that will discriminate models using an in-depth understanding of the mechanisms that drive the errors and uncertainty in projected changes over Africa.