Constraining Climate Model Projections of Regional Precipitation Change.

As communities prepare for the impacts of climate change, policy makers and stakeholders increasingly require locally resolved projections of future climate. Statistical downscaling uses low‐resolution outputs from climate models and historical observations to both enhance the spatial resolution and correct for systematic biases. By examining the downscaled rainfall over land, we show that although bias corrections are effective in reducing biases in the current climate, they do not reduce the intermodel spread in future rainfall projections. This failure stems from the strong dependence of future rainfall change upon the current climatological rainfall patterns. Even after bias corrections are applied, the downscaled projections of precipitation change retain this dependence upon their native climatology. However, we show that this dependence can be exploited; even very simple methods to subset models according to their ability to resolve the observed rainfall climatology can substantially reduce the intermodel spread in rainfall projections. Plain Language Summary: To prepare for future climate change, policy makers and stakeholders require reliable model projections with high spatial resolution. To meet this need, statistical methods have been developed to postprocess the model output so correct for systematic biases and enhance the spatial resolution. We focus on rainfall over land and find that the postprocessing only yields consistent values for the current climate and no reduction in the uncertainty is obtained for future projections even after the postprocessing. We show that patterns of future rainfall change exhibit a strong dependence on the current climate. This dependence can be exploited by selecting models based on their ability to reproduce the observed climate. We show that this screening of models can significantly reduce the uncertainty in future rainfall projections. The success of this simple method emphasizes the importance of model evaluation in reducing the uncertainty of multimodel climate change projections. Key Points: Statistical downscaling improves the current climatology but does not reduce the intermodel spread of future rainfall projectionsModels with more similar base climates yield more similar projections of change both with and without statistical downscalingScreening models provides a simple method for constraining future projections and is insensitive to the choice of observational data sets [ABSTRACT FROM AUTHOR]