Based on the winter temperature and NCEP reanalysis data and the realtime prediction data of the CFS model in Southwest China, the SVD diagnostic analysis is used to select the atmospheric circulation and early sea temperature and OLR factor fields in the key areas affecting the winter temperature in Southwest China, and to establish a combined statistical drop scale prediction model combining prediction and observation fields. The results of this statistical downscale forecast model for 1982-2017 show that the spatial correlation coefficient with the observation field is significantly higher than the original results of the CFS model, with the multi-year mean increasing from -0.06 to 0.38, up to 0.85. At the same time, this scale prediction model can better restore the spatial distribution of winter temperature in Southwest China.