Filtering CMIP6 models in the Euro-Mediterranean based on a circulation patterns approach

The performance of a set of 26 CMIP6 global climate models (GCMs) in the Euro-Mediterranean region is analyzed based on a classification of atmospheric circulation patterns (CPs). Their spatial and temporal variability representation, including the associated surface conditions in ERA5 during 1950–2014, allows a ranking of the best-performing GCMs. GCMs manage to reproduce the annual cycle of the CPs frequency, with a dominant summer CP enhancing warm and dry conditions. However, the correct timing of this pattern and the transitional CPs often need to be more accurate. The analysis of the surface patterns related to the different CPs presents overall good model performance, higher for temperatures than for precipitation, particularly in the transition seasons, for which the GCMs spread in their skill score increases. By blending both the spatial and temporal features of the CPs, the EC-Earth3-CC, IPSL-CM6A-LR, EC-Earth3-Veg-LR, MIROC6, and GFDL-ESM4 arise as the best-performing GCMs. This ranking is used to construct multiple model ensembles of climate projections, also taking into account model dependence and spread. Results from this assessment show that future projections of extreme climate indices (2070–2 100)—including the expected increases in the frequency of warm days and dry spells—can be “performance-constrained” and their uncertainty can be more reliably assessed by selecting specific subsets of GCMs, generating tailored climate information at a regional scale. In particular, the warming and drying signals are clearer in the best-performing GCMs, with more robust results in summer than in winter.