Exploring Satellite Visible Band Observations for Quantitative Precipitation Estimation

Accurate high‐resolution precipitation information is essential for hydrologic modeling, extreme‐event monitoring, and climate assessment, yet, conventional satellite precipitation products are often constrained by the relatively coarse spatial resolution of infrared and microwave observations. We present a quantitative precipitation estimation (QPE) framework that exploits 2‐km visible‐band observations from the Himawari‐8 geostationary satellite within a U‐Net convolutional neural network architecture. The model uses six consecutive daytime multispectral visible images to encode cloud‐motion and cloud‐evolution information for hourly precipitation retrieval. Evaluation against the China multi‐source merged precipitation analysis shows that multispectral and temporally continuous visible‐band information improves precipitation‐area delineation and false‐alarm control compared with single‐time or single‐band visible configurations. The VIS‐based retrieval shows performance comparable to IR QPE in selected light‐to‐moderate rain‐rate ranges, and the VIS+IR QPE configuration provides a more balanced performance by combining VIS‐derived cloud texture and evolution with selected IR thermal and water‐vapor information. Case studies further demonstrate the value of temporal continuity, showing that consecutive visible imagery helps capture fine‐scale cloud structures and associated precipitation patterns during daytime events. These results highlight the potential of high‐spatial‐resolution visible imagery as a complementary daytime source for fine‐scale precipitation monitoring, whereas infrared and microwave observations remain essential for more physically complete precipitation retrieval across the diurnal cycle.