2024 Volume 80 Issue 16 Article ID: 23-16065
In developing countries where meteorological observation networks are not sufficiently deployed, the provision of accurate satellite rainfall using satellite products is expected not only for disaster prevention, but also to expand the range of application of technology and research requiring observed rainfall. The aim of this study is to propose a new satellite rainfall estimation algorithm HiDREDv2 by using a fully convolutional neural network, on the geostationary meteorological satellite Himawari. In this study, a model suitable for meteorological phenomena is constructed by combining the features of each of the existing models. The accuracy of this model for 6-hour rainfall accumulation was 13.26 for RMSE and 0.69 for FSS. The accuracy of the model was significantly improved over GSMaP for the 4-day accumulated rainfall during the heavy rainfall period due to the heavy rainfall in July 2008.
