基于深度学习的FY-4A降水估计方法OA

Traditional precipitation estimation methods primarily rely on ground station observations.However,due to the sparse and uneven distribution of stations,especially in complex terrain areas such as the"Belt and Road"initiative and the Tibetan Plateau,ground station ob-servation data often face issues of missing and incomplete data,which severely affects meteorological services and technical research.To ad-dress this,a deep learning-based FY-4A precipitation estimation method is proposed.This method utilizes a convolutional neural network guided by an attention mechanism to extract features from satellite cloud images,adaptively learning the complex relationship between satellite cloud images and ground precipitation to achieve precipitation estimation.Extensive experiments were conducted using the state-of-the-art op-erational reanalysis product ERA5 as the benchmark data,the results show that the root mean square error(RMSE)of this method is 0.425 mm/h,and the correlation coefficient is 0.541.Compared to CFSV2,GPM precipitation products,and Unet,Attention-Unet,and DLPE-MS deep learning precipitation estimation methods,the RMSE was reduced by 25.569％,51.484％,0.932％,4.709％,and 2.299％,respective-ly,while the correlation coefficient was increased by 34.577％,73.397％,3.442％,5.458％,and 5.664％,respectively.This method can bet-ter identify precipitation areas compared to other precipitation products and methods.The proposed method provides high-quality foundational data support for meteorological services and technical research such as weather forecasting and climate prediction warnings,and offers a new research approach for satellite-based precipitation estimation.