Abstract:

The ensemble mean of simulated temperature and precipitation of CMIP5 and CMIP3 in arid areas of Northwest China was compared, the results showed that CMIP5 was more close to the observed value. Comparing with CMIP3, the correlation coefficient of CMIP5 was improved by 0.15 for year, 0.13 in spring, 0.24 in summer and 0.02 in autumn, while decreased by 0.07 in winter. CMIP5 simulated better trend of mean temperature than CMIP3 in the arid area of Northwest China, the biases reduced by 0.03℃per decade for year, 0.1℃per decade for spring, 0.01℃per decade for summer, 0.06℃per decade for autumn and 0.14℃per decade for winter, respectively. The annual and seasonal bias of CMIP5 was 1-2℃ lower than CMIP3 in the arid area of Northwest China, but temperature for both CMIP3 and CMIP5 were 3-6℃ and 1-4℃ higher than observation in Tianshan Mountains, especially in summer, which was reached to 6℃ and 4℃, respectively. Two models showed little difference in the correlation coefficient between simulated and observed precipitation, which was lower than 0.1, but the bias was still higher. The precision of precipitation trend for CMIP5 was lower than CMIP3, the biases increased by 0.67 mm per decade for year, 0.23 mm per decade for spring, 0.51 mm per decade for summer, 0.11 mm per decade for autumn, and 0.14 mm per decade for winter, respectively. The root mean square error of precipitation of CMIP5 was decreased by 77.6 mm, 25.5 mm, 25.0 mm, 18.8 mm and 13.9 mm for annual, spring, summer, autumn and winter, respectively. In spatial, CMIP5 still simulated higher annual and seasonal precipitation, but was better than CMIP3. Conversely, CMIP3 and CMIP5 gave lower annual and seasonal precipitation in the Tianshan Mountains, which was 50 mm lower than the observation.

Key words: the northwest arid areas, CMIP3, CMIP5, average temperature, precipitation, simulation ability