Abstract: Driving data, such as air temperature, precipitation, wind field, specific humidity and radiation, used in land surface models and land data assimilation system, can be derived using the reanalysis datasets from downscaling. In this study, the statistical downscaling of air temperature data was analyzed based on calculating the observation error covariance in west China using geostatistics, estimating the background error covariance using innovation vector method, and interpolating the daily air temperature data with spatial resolution 0.125°×0.125° in 2000 using optimal interpolation, and the precision of the analyzed air temperature field was also analyzed. The results show that the analyzed daily air temperature can well reflect the change of air temperature in west China and be directly used in land surface models under the gentle topographical conditions. For the QinghaiTibetan Plateau, however, quality of the analyzed air temperature was poor due to the effect of the topography and distribution of weather stations, but it was increased to some extent compared with the reanalyzed data.

Key words: air temperature, downscaling, land surface models, daily air temperature, west China