Huayu-Global: precipitation estimates every 30 minutes across the globe, with the quality of products published months later

By combining three types of on-board sensors on five geostationary satellites, this deep learning system aims to provide reliable rain data in time for flood warning.

Satellite precipitation products suffer from a persistent dilemma: real-time estimates come in quickly but are less accurate, while the most reliable products are not released until months after observation, too late for flood warning or disaster management. A Chinese team presents Huayu-Global, a deep learning system designed to reduce this gap.

Multiple constellations, one setting

Huayu-Global merges the streams of three types of sensors on board five geostationary satellites into a common estimation framework, and produces consistent precipitation fields every thirty minutes covering almost the entire globe. Evaluated on more than one million ground station readings, from two periods outside the training years, it detects rainfall events better and estimates cumulative events better than the real-time products compared, and matches the level of delayed products in the main indicators.

The authors see it as a fast and reliable rainfall data layer for flood forecasting, water management and disaster response. Its multi-constellation design would also allow future geostationary missions to be easily integrated into hydrological and weather forecasting systems.