Abstract:The highly dense observations are used in the regional analysis, so the regional data assimilation may produce more accurate small scale analysis. However, the large scale aspect of a global analysis is superior to that of a regional analysis for assimilating more observations of the satellite observations and is not affected by lateral boundaries. To improve the forecast qualities of regional model, a blending method to merge the T639 global analysis with the regional analysis from the GRAPES Meso analysis system is implemented using discrete cosine transform (DCT) filter. The experiments show that the simulated kine tic energy spectrum of GRAPES analysis is a little bigger than T639 analysis in the small meso scale. Meanwhile, the simulated kinetic energy spectrum of blending analysis is the closest to the atmospheric kinetic energy spectrum in the small meso scale. The results indicate the qualities of geopotential height, temperature and wind of blending analysis are obviously improved, the forecast of wind within 6 h is improved, and the ETS verification of accumulated precipitation of blending analysis is higher than the regional analysis.