Google DeepMind releases PARDIG and SatWeave, the datasets behind its WeatherNext 3 model
SatWeave is an hourly global mosaic of geostationary satellite imagery across 17 channels; PARDIG is an AI-built global precipitation record at 15-minute, roughly 10 km resolution, with archives back to 1999 and live feeds.
Why it matters
Most global satellite composites offer four or fewer channels or update every few hours, which is why DeepMind says SatWeave lets WeatherNext 3 start new global forecasts hourly instead of every six hours. PARDIG is trained on NASA's GPM spaceborne radar, and DeepMind reports it beats NASA's research-grade IMERG-Final, improving hourly detection of light-to-moderate rain by more than 30%. Releasing both openly lets outside researchers test those claims and build flood, hydrology and solar applications.
Key facts
- Released: 13 November 2026
- SatWeave: hourly global mosaic, 17 channels
- PARDIG: 15-minute, roughly 10 km global precipitation
- PARDIG archive: 1999 to 2026
- PARDIG beats IMERG-Final by more than 30% on light-to-moderate rain detection
- Evaluated against 36,000+ weather stations