DeepMind releases SatWeave and PARDIG weather datasets for research and development

This item follows Google DeepMind’s official blog post of the same date. SatWeave combines observations from geostationary satellites into a global imagery dataset; the post names sources including GOES-East, GOES-West, Meteosat and Himawari. The same article describes the satellite count as both five and six, so this item does not adopt either figure. The post says SatWeave updates hourly and that a historical archive spanning more than a decade is being released.

PARDIG stands for Precipitation AI Reanalysis: Densely Inferred GPM. The post says it uses a method similar to WeatherNext 3, with sparse NASA GPM spaceborne radar observations, SatWeave imagery and ECMWF atmospheric fields, to produce a global precipitation map at about 15-minute and roughly 10-kilometre resolution, updated with an average latency of about five hours, plus historical archives from 1999 to 2026. Google says that in its evaluations against ground radar networks in the US, Europe and Japan and more than 36,000 weather stations worldwide, PARDIG outperformed several global baselines including NASA’s IMERG-Final. That is the source’s own reported evaluation, not a test by this site.

The post says the live and historical datasets are openly available on Google Cloud Storage, and that PARDIG’s historical catalogs are also available in Google Earth Engine. This is not an update to a consumer weather app. Actual download or use still depends on a user’s own cloud account, quotas and terms.