Google DeepMind has published the foundational datasets behind its latest weather model, giving researchers free access to the observations that make its forecasts possible.

The first, SatWeave, is a global mosaic of geostationary satellite imagery. It blends observations from six geostationary satellites — GOES-East and GOES-West, Meteosat 0 and Meteosat Indian Ocean, and Himawari — which use different map projections, file formats, sensors and scan schedules. Where most existing global composites drop incompatible aspects or update only every few hours, SatWeave knits them into a single global map across 17 visible, water-vapour and infrared channels, refreshed hourly. DeepMind is releasing both the hourly live feed and a historical archive spanning more than a decade.

The second dataset, PARDIG (Precipitation AI Reanalysis: Densely Inferred GPM), is an AI-based global precipitation map. It is trained directly on sparse spaceborne radar observations from NASA's Global Precipitation Measurement core observatory, using SatWeave imagery and ECMWF atmospheric fields as inputs. The result is a global precipitation field at 15-minute temporal resolution and roughly 10-kilometre spatial resolution, updated with an average latency of about five hours.

According to DeepMind, PARDIG outperforms leading global baselines across the rain rates analysed, including NASA's IMERG-Final — a widely used research-quality product — and improves hourly precipitation detection skill by more than 30% on light-to-moderate rainfall. It proved especially effective at capturing extreme precipitation, which existing global products often miss. Because PARDIG learns directly from satellites rather than depending on ground measurements, DeepMind expects it to hold up reasonably well even in regions without dense radar or rain-gauge coverage.

Both datasets feed WeatherNext 3, the company's most advanced global weather AI, and their live and historical versions are now openly available on Google Cloud Storage, with PARDIG's historical catalogues also in Google Earth Engine.

DeepMind frames the release as a bid to widen the ecosystem: it says it wants the community to explore applications ranging from floods, landslides and drought — especially in regions lacking dense ground radar — to solar-power forecasting and wildfire monitoring.