FY-4B Precipitation Estimation Based on Transformer

Precipitation is a key link in the earth's energy cycle and matter exchange, and is also one of the core driving elements of terrestrial ecosystems. However, there is a problem of sparse meteorological stations in areas with complex terrain, and it may even be impossible to deploy observation stations, resulting in missing ground observation precipitation data or serious time lag. In address these issues, this paper proposes a transformer based method for precipitation estimation from Fengyun-4 (FY-4B) satellite, which estimates precipitation rate in near real time based on FY-4B satellite infrared remote sensing and latitude and longitude auxiliary information. Comparisons with FY4B-QPE and CFSV2 show that the proposed method reduces the RMSE (root mean square error) of FY4B-QPE and CFSV2 by 50.64% (FY4B-QPE) and 30.03% (CFSV2), respectively, and meanwhile increases the Pearson correlation coefficient of the above products by 160.89% (FY4B-QPE) and 44.38 (CFSV2) respectively. This method is expected to supplement the completeness and timeliness of precipitation observation data in complex terrain areas.