Uncertainties in climate change projections covered by the ISIMIP and CORDEX model subsets from CMIP5

Abstract. Two international projects, ISIMIP (Inter-sectoral Impact Model Inter-comparison Project) and CORDEX (Coordinated Regional Climate Downscaling Experiment), have been established to assess the impacts of global climate change and improve our understanding of regional climate, respectively. Model selection from the GCMs (general circulation models) within CMIP5 (fifth phase of the Coupled Model Inter-comparison Project) was conducted by the different approaches for each project: one is a globally consistent model subset used in ISIMIP and another is a region-specific model subset for each region of interest used in CORDEX. We evaluated the ability to reproduce the regional climatological state by comparing the subsets with the full set of CMIP5 multimodel ensemble. We also investigated how well the subsets captured the uncertainty in the climate change projected by the full set, to provide increased credibility for the scientific outcomes from each project. The spreads of the biases and Taylor’s skill scores from the ISIMIP and CORDEX subsets are smaller than that from the full set for the regional means of surface air temperature and precipitation. However, the spreads in ISIMIP and CORDEX extend beyond the spreads from high performed models from full set, despite using a smaller number of models. It was shown that better subsets exist that would have smaller biases and/or higher scores than the current subset. The ISIMIP subset captures the uncertainty range of the regional mean of temperature change projections by the full set better than the CORDEX subsets in 10 of 14 terrestrial regions worldwide. Compared with the randomly selected 10,000 subset samples, CORDEX uses a subset with relatively low coverage of the uncertainty range for the temperature change in some regions, and ISIMIP uses a subset with relatively high coverage in all regions. On the other hand, for the regional mean of precipitation change projections, although the coverage from the CORDEX subsets is lower among the 10,000 subset samples in half of the regions, the CORDEX subsets indicate a tendency for better coverage of the uncertainty range than the ISIMIP subsets. In the regions where CORDEX used nine models or more, good coverage (> 50 %) is evident for the projections of both temperature and precipitation. The globally consistent model subset used in ISIMIP could have difficulty in capturing uncertainties in the regional precipitation change projections, whereas it widely covers uncertainties in the temperature change projections. The region-specific model subset, like CORDEX, can cover the relative-widely uncertainties in both temperature and precipitation changes, but it depends on the number of models used.