Abstract:A Hybrid Statistical Downscaling Prediction method (HSDP) was developed to predict summer precipitation in Qinghai Province, which comprehensively used the highly predictable circulation information predicted in real time by climate forecast system 2.0 (CFSv2) and the climate factors with high correlation with summer precipitation in Qinghai Province, so as to realize the climate prediction of Qinghai summer precipitation by combining dynamic and statistical methods. According to the correlation coefficient between the interannual increment of global climate factors and summer precipitation in Qinghai Province, and the evaluation of CFSv2 prediction products on the ability of actual simulation, this paper selects the interannual increment of climate variables in the following key areas as the prediction factors: (1) CFSv2 model predicts the 500 hPa height field in summer, including the Baikal Lake Ridge, Oral Ridge and Xinjiang Ridge regions; (2) CFSv2 model predicts the zonal wind field at 200 hPa in the west of the Tibetan Plateau; (3) the sea surface temperature field in autumn and winter of the previous year in the tropical Pacific region; (4) based on the observation data of the sea level pressure field in Siberia in autumn and winter of the previous year, the statistical downscaling prediction of summer precipitation in Qinghai Province was carried out. The statistical downscaling method is used to model the precipitation from 1983 to 2011, and the spatial distribution and temporal variation of the precipitation from 1983 to 2018 in Qinghai Province were hindcasted. Results show that the statistical downscaling model can significantly improve the prediction ability of CFSv2 for summer precipitation in Qinghai Province, and can well reproduce the characteristics of less precipitation in the Northwest Plateau and more precipitation in the Southeast of Qinghai Province. The temporal variation of summer precipitation in Qinghai Province from 2012 to 2018 predicted by the model also has a high correlation coefficient (0.76), which is good for the years with significantly less precipitation (such as 2015) and those with significantly more precipitation (such as 2012, 2018). The cross-checking results (the correlation coefficient is 0.46, which is slightly lower than the correlation coefficient of 0.48) show that the model has high stability and reliability.