Reliability of the global climate models during 1961-1999 in arid and semiarid regions of China.

General circulation models (GCMs) are useful tools for investigating mechanisms of climate change and projecting future climate change scenarios, but have large uncertainties and biases. Accurate models are of significant importance for agriculture, water resources management, hydrological simulation, and species distribution. In this study, we examined the precipitation and temperature reproducibility of 34 GCMs during the period from 1961 to 1999 over arid and semiarid regions of China. The study area was divided into eight sub-regions; each represented a specific topography. The evaluation was conducted for the whole study area and the sub-regions. Spatial and temporal indices and weighting methodology were used to comprehensively illustrate the models' reproducibility. The results showed that the simulation ability during winter outperformed than that during summer (the weight was 0.192 higher for precipitation and 0.044 higher for temperature during winter than that during summer over the whole study area). Precipitation was more accurately simulated during spring than during autumn as opposed to temperature (the weight was 0.124 higher during spring than during autumn for precipitation and 0.1 higher during autumn than during spring for temperature for the whole region). For precipitation, the simulation ability in the basins was the best, followed by plateaus and mountains; the weights were 0.462, 0.308, and 0.231, respectively. For temperature, the mountains and plateaus had the best and poorest reproducibility, at weights of 0.446 and 0.198, respectively. The top models for precipitation and temperature at different spatial scales (whole study area, three topography types, eight sub-regions) were recommended. The results served as a reference for model selection in future studies regarding impacts of climate change on eco-hydrology.