Abstract
Surface precipitation phase transition is conducive to devastating snowstorms and avalanches yet remains a global challenge due to the scarcity of surface observations. Here, we present the Real-time Precipitation Phase-Intensity Collaborative Retrieval Network (RePPIC-Net), a hybrid AI framework that quantifies surface precipitation phase from satellite observations. By integrating real-time 3D atmospheric physics fields from the AI-driven FuXi model with operational geostationary satellite observations through a hierarchical architecture, our system enables real-time monitoring of surface precipitation phase, as opposed to at least 4-hour latency of current operational systems. Validated against ground stations in China, RePPIC-Net achieves a Critical Success Index for Phase and Detection of 0.1574 (snowfall) and 0.3147 (rainfall) for 0.1-5 mm/h precipitation, outperforming 4-hour latency operational products' respective scores of 0.1001 and 0.3064. The real-time precipitation phase discrimination capability of RePPIC-Net allows the development of a satellite-based surface precipitation phase nowcasting system, meeting the need for 1-3 hour global surface precipitation phase transition warnings. RePPIC-Net provides a replicable blueprint for AI-powered real-time weather monitoring, filling a gap in wintertime weather disaster warnings.