The paper presents an algorithm for precipitation estimation based on data from the Himawari-8/9 satellite. The algorithm is based on two neural networks of visual transformer and convolutional architectures for preliminary precipitation mask calculation and rain rate estimation. Data from the “Global Precipitation Measurements” (GPM) international project were used as a reference value of precipitation. These data are based on measurements from various active and passive microwave and infrared satellite instruments. The algorithm takes into account spectral, textural, and microphysical parameters of clouds. An accuracy assessment was carried out using GPM data and ground-based rain gauges. The results of a comparison between the algorithm and the ComsoRu-6 regional numerical weather prediction model are also given. It is shown that the presented algorithm most accurately estimates the amount of accumulated precipitation sums but it has a tendency to overestimate this value. On the other hand, GPM and CosmoRu-6 often underestimate precipitation. The comparison with the product of GPM showed a root-mean-squared error of about 2.19 mm/h.
Andreev et al. (Mon,) studied this question.