Decadal prediction skill for Eurasian surface air temperature in CMIP6 models

The Eurasian surface air temperature (SAT) is experiencing decadal variations against the background of global warming. The prediction skill for the seasonal mean SAT in CMIP6 Decadal Climate Prediction Project (DCPP) models is investigated in this study. The large decadal variations of winter and autumn Eurasian SAT are barely predicted by the CMIP6 models. IPSL-CM6A-LR and the multimodel ensemble have skill in predicting the variations of spring Eurasian SAT, with significant anomaly correlation coefficients, but not for the amplitude, with negative mean-square skill scores. Significant skill is apparent for the summer SAT over Mongolia and North China, with the CMIP6 models showing their best skill for the summer Eurasian SAT. Compared to external forcing, model skills for Eurasian SAT may derive more from the initialization. It should be noted that there are model system errors in the form of false strong relationships of SAT between winter and other seasons when in fact the variations of other seasons' SATs are independent of the winter SAT in observations.