DeepMind now provides historical archives and live feeds of a global geostationary satellite mosaic and an AI precipitation dataset, so researchers can build weather and hydrology models on the same inputs used for WeatherNext 3.
Key takeaways · AI-distilled
SatWeave merges five geostationary satellites (GOES-East and West, two Meteosats, Himawari) with different projections, sensors and 10-30 minute scan schedules into one hourly global mosaic; DeepMind says most composites offer four or fewer channels or update every few hours.
The hourly feed lets WeatherNext 3 start new global forecasts every hour, while models initialized on analysis data get a refresh every six hours. The SatWeave archive spans more than a decade.
PARDIG is trained on sparse spaceborne radar from NASA's GPM Core Observatory, with SatWeave imagery and ECMWF fields as inputs, and outputs global precipitation at 15-minute and roughly 10 km resolution with about five hours of average latency.
Checked against radar networks in the US, Europe and Japan and over 36,000 stations, DeepMind reports PARDIG beats NASA's IMERG-Final at every rain rate tested and improves hourly detection of light-to-moderate rain by more than 30%. Its archive runs from 1999 to 2026.
