资源环境科技发展态势分析平台

Quality of a downscaling depends primarily on the quality of the driving global climate model (GCM). In this study, historical atmospheric conditions simulated by 14 GCMs in CMIP5 are evaluated for downscaling applications centred over the Tibetan Plateau (TP) with ERA-Interim reanalysis as reference. Another reanalysis NCEP-DOE is also used to estimate the uncertainty associated with the reanalyses. Performances of six frequently used GCM variables, involving atmospheric circulation, air temperature and humidity, are evaluated in terms of biases, spatial correlation coefficient, mean absolute error as well as distinct seasonal features. To detect distributional biases, the two-sample Kolmogorov-Smirnov test (KS test) is applied to both the original time series and their anomalies on the monthly scale. A spatial ranking scheme is finally applied to objectively quantify overall relative merits of the GCMs over this region. We found that differences between two reanalysis datasets are negligible over this region. Regarding the GCMs' performances, the biases of the simulated variables show remarkable differences among models. Sea level pressure and 500 hPa geopotential height are well simulated by all the GCMs, whereas specific humidity at 600 hPa has a significant dry bias and temperature at 500 hPa has a sizable cold bias. The spatial pattern of the upper-tropospheric circulation is relatively poorly simulated. The KS test suggests that the climatic mean and higher order moments play about an equal role in causing the errors. According to the ranking scores, CCSM4, CNRM-CM5, MPI-ESM-LR, NorESM1-M, MIROC4h, MPI_ESM_MR and CSIRO-MK are relatively superior to other GCMs for this region.