Google DeepMind releases SatWeave and PARDIG datasets

Google DeepMind's release builds on its recent WeatherNext 3 model, which relied on both datasets for its advances. SatWeave addresses a long-standing difficulty in combining satellite imagery: the five satellites it draws from use different map projections, file formats, sensors, and scan schedules ranging from 10 to 30 minutes. Most existing global composites drop incompatible aspects, offering only four or fewer channels or updating every few hours. SatWeave instead blends six geostationary satellites into a 17-channel mosaic that updates hourly, allowing WeatherNext 3 to produce new global predictions rather than waiting for analysis data refreshed every six hours.

PARDIG tackles precipitation tracking, where ground radars and rain gauge networks are not always available or well-distributed. Built on an AI architecture similar to WeatherNext 3 and trained on sparse spaceborne radar observations from NASA's GPM Core Observatory, it produces a global precipitation map at 15-minute temporal resolution and roughly 10-kilometer resolution with an average latency of about five hours. Evaluations against three continental radar networks and over 36,000 weather stations worldwide found it outperforms leading global baselines, including NASA's IMERG-Final, and proved particularly effective at capturing extreme precipitation events that existing products struggle to detect.

Because PARDIG learns directly from satellites rather than relying solely on ground measurements, DeepMind expects it to have reliable accuracy everywhere on Earth, including areas without ground data. The company is positioning the open datasets as a way to let researchers explore improved models for floods, landslides, and drought, especially in regions without dense ground radar networks.