A Neural Network Method for Precipitation Estimation from Elektro-L No. 4/MSU-GS Spectroradiometer Measurements

The paper presents an algorithm for the precipitation intensity estimation from the data of the MSU-GS spectroradiometer on board the Elektro-L No. 4 satellite. The algorithm is based on two neural networks with transformer and convolutional architectures and aimed at detection of precipitation-forming clouds and quantification of precipitation, respectively. The Global Precipitation Measurements (GPM) Integrated Multi-SatellitE Retrievals (IMERG) product is used as a reference source of information about precipitation intensity for training and testing the neural network models. The presented algorithm takes into account spectral, textural, and microphysical properties of clouds and allows obtaining precipitation maps for both daytime and nighttime conditions. The test results have shown an RMSE value of 1.27 mm/hour and F1 score of about 0.59 for the warm and cold seasons.