Abstract: The climate system model is an important tool for studying the mechanism of climate change and predicting future climate change.Due to the extreme complexity of the climate system,there may be some errors and limitations in simulating the climate.Therefore,it is necessary to evaluate the simulation capability of climate models.Nowadays,many researches have been done on the assessment of climate system modeling temperature capability and the prediction of future temperature changes.However,there is no report on the relationship between climate modeling capacity to simulate the seasonal variation of air temperature and some influencing factors.Therefore,this paper analyzes the spatial difference characteristics of simulation capability of air temperature in China,using BCC-CSM1.1(m),GFDL-CM3 and HADGEM2-ES global climate models from the Coupled Model Inter-comparison Project Phase 5 (CMIP5),based on the daily mean temperature,maximum temperature and minimum temperature data obtained from 663 conventional meteorological stations in China during the time period from 1951 to 2004.The simulation capability of three models are verified by using the observational air temperature data,and its relationships are discussed with latitude,longitude,elevation,slope,aspect and terrain obscuration (slope,aspect and terrain obscuration are obtained from 500 m×500 m digital elevation model).The results showed that the BCC-CSM1.1(m) and GFDL-CM3 model can reproduce the seasonal variation of air temperature in China.The mean absolute error and root mean square error of daily mean temperature and daily minimum temperature simulated by models are smaller in Northeast China,North China,South China and East China,the simulation capability is stronger than those in the other areas.However,in Western China,the simulation capability is relatively weak characterized by large mean absolute and root mean square errors.Compared with the daily mean and minimum temperature,the mean absolute error and root mean square error of daily maximum temperature simulated by models are larger in North China and Northeast China.The capability of reproducing the seasonal variation of air temperature in China from HADGEM2-ES model is the weakest if compared with the results from the BCC-CSM1.1(m) and the GFDL-CM3 models,and its mean absolute error and root mean square error were increased from south to north with larger errors in parts of Western China,Inner Mongolia and Northeast China,and smaller errors in the southern part of South China.For the three models,BCC-CSM1.1(m) model is the best in the simulation of the air temperature in China,followed by the GFDL-CM3 and HADGEM2-ES models.For each model,the simulation of the daily mean temperature is the best,followed by the daily minimum temperature,and the daily maximum temperature is the worst.The latitude,longitude,altitude and slope have different influences on the simulation effect of climate models among different models.The BCC-CSM1.1(m) and the GFDL-CM3 models have high simulation capability in low and high latitudes,low longitude,low slope and areas of an altitude less than 2 000 meters.The HADGEM2-ES model has high simulation capability in low latitude areas and 112°E nearby areas.The aspect and terrain obscuration have no significant impact on the simulation effect of climate models.

Key words: global climate models, air temperature, spatial difference