Abstract: The scenario data of Coupled Model Intercomparison Project Phase 6 (CMIP6) are processed by Delta method in statistical downscaling method.Based on the centralized root mean square error (RMSE), correlation coefficient (R), standard deviation (STD) and interannual variability skill score (Ts), comprehensively evaluate the simulation ability of scenario data from 10 CMIP6 models on the extreme temperature index in Shandong,and select models with better simulation results. The results indicate that:In the simulation of extreme temperature indices TXx and TNx that characterize intensity,after downscaling,more than 50% of the models simulate that the median absolute deviation of extreme temperature index in Shandong is closer to the observed value than before downscaling, in the simulation of extreme temperature indices TR, SU, FD and ID that characterize frequency,more than 60% of the models absolute deviation medians are closer to the observations than before downscaling.Using the ranking method, the simulation ability of the 10 models was compared.The CMCC-ESM2 score was 178 points, NorESM2-MM score was 191 points,and TaiESM1 score was 191 points.Therefore, these three models are preferred models. The error percentage of extreme temperature indices simulated by the ensemble mean of optimized models is lower than that of all models.After optimization,the absolute value of error percentage of TXx, TNx, TR and ID decreased from 4.2%, 2.2%, 28.67% and 14.3% to 3.3%, 1.8%, 23.58% and 9.7%, the error percentage of SU was within ± 1% before and after optimization.

Key words: CMIP6, Model evaluation, Extreme temperature index, Delta downscaling method, Shandong