To What Extent Does Discounting ‘Hot’ Climate Models Improve the Predictive Skill of Climate Model Ensembles?

Abstract

It depends. The Intergovernmental Panel on Climate Change's (IPCC) Assessment Report Six (AR6) took a step toward ending so-called ‘model democracy’ by discounting climate models that are too warm over the historical period (i.e., models that ‘run hot’) when making projections of global temperature change. However, the IPCC did not address whether this procedure is reliable for other quantities. Here, we explore the implications of weighting climate models according to their skill in reproducing historical global-mean surface temperature using three other climate variables of interest: global average precipitation change, regional average temperature change, and regional average precipitation change. We find that the temperature-based weighting scheme leads to an improved prediction of global average precipitation, though we show that this prediction could be overconfident. On regional scales, we find a heterogeneous pattern of error reduction in future regional precipitation. This stands in sharp contrast with the broad regional pattern of error reduction in future temperature projections, though we do find regions where error is not significantly reduced. Our results demonstrate that practitioners using weighted climate model ensembles for climate projections must take care when weighting by temperature alone, lest they produce unreliable climate projections that result from an inappropriate weighting procedure.