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.