Abstract: Liaoning Province is one of the main grape production areas in China. Due to the increasing of temperature and the frequent occurrence of extreme temperature, it has a noticeable impact on the planting and production of grapes in Liaoning. Based on extreme temperature indices, Global Climate Models(GCM) are evaluated and the model with the best simulation effect in the study area are selected. It is helpful to improve the accuracy of future climate resource and disaster risk analysis. In this paper, the observation meteorological daily data during 1971−2010 was used, including 32 stations across Liaoning Province. 31 models in the CMIP5 climate downscaling dataset were assessed the simulation performance about the spatio-temporal variation characteristics of the extreme temperature indices in Liaoning Province (yearly and seasonally). And the locations of the meteorological stations were completely consistent with that of the model stations. After using SS/M2 indices and MR comprehensive rating, the best model was selected. Seven extreme temperature indices, including mean maximum temperature(TXm), mean minimum temperature(TNm), frost days(FD0), summer days (SU25), minimum minimum temperature(TNn), maximum maximum temperature(TXx) and the range of extreme temperatures(ETR), were adopted to investigate the change of extreme temperature. There were the following conclusions: for the indices representing the characteristics of mean temperature, the performance of all models were better and closest to the observed values; For the indices representing continuous extreme temperature events, the simulated results of all models were ordinary; And for indices representing extreme temperature, the simulation results were very different from the observation values. Since there was no consistency in the ranking of extreme temperature indices among the different time and space scales, the three models with the best simulation ability were MPI-ESM-LR, GFDL-ESM2M and MIROC-ESM-Chem introduced by MR index. The three models were the first choice in this paper. Comparing the error, the results of the preferred model ensemble averages were significantly better than that of others. It is helpful to use the selected models to predict the future change of agricultural climate resources and to analyze disaster risks across Liaoning Province. It can help seek advantages and avoid disadvantages and reduce disaster losses.

Key words: Climate downscaling dataset from CMIP5, Liaoning, Extreme temperature index, Model evaluation, Optimal selection models