The Library Media — Science, Technology & Health · AI News Desk
The Rain Data Behind the AI Forecast Is Now Open: DeepMind Releases the WeatherNext 3 Datasets
By Catalina Clarke
Photo: NOAA / Wikimedia Commons (Public domain)
Google DeepMind has opened two of the datasets behind its WeatherNext 3 forecasting system to anyone, free of charge — a global satellite mosaic called SatWeave and a global precipitation record called PARDIG that reaches back to 1999.
DeepMind announced the release in a blog post dated November 13, 2026, saying both datasets are available on Google Cloud Storage and Google Earth Engine for use with WeatherNext 3, its AI weather model.
SatWeave, as DeepMind describes it, combines imagery from geostationary weather satellites — including GOES-East, GOES-West, Meteosat and Himawari — into a single global mosaic spanning 17 visible, water-vapor and infrared channels, updated hourly.
PARDIG is the precipitation half of the release: a global rainfall record at 15-minute intervals and roughly 10-kilometre resolution, with about five hours of latency, and an archive running from 1999 to 2026.
Why rain data is the hard part
Satellite imagery shows clouds; it does not directly measure how much rain reaches the ground. Ground radar does, but radar networks cover only part of the world — much of Africa, South Asia, Latin America and the open oceans have little or none. A precipitation record built from satellites is one way researchers try to fill those gaps.
DeepMind says it tested PARDIG against three radar networks and more than 36,000 weather stations, and reports an improvement of more than 30 percent on light-to-moderate precipitation compared with IMERG-Final, the widely used satellite precipitation product from NASA’s Global Precipitation Measurement mission. Those performance figures are DeepMind’s own, reported from its testing; they have not been independently verified here.
What it could be used for
For communities in places without dense radar coverage, better rain data feeds directly into flood warnings, landslide risk tools, drought monitoring and water management — the everyday end of weather forecasting, where a good rainfall record matters more than a headline storm track.
The release also gives outside researchers the same inputs the model itself uses, which makes WeatherNext 3’s forecasts easier to test, compare and build on.
DeepMind did not date WeatherNext 3 itself in the announcement, and the release is of data, not of a new model. What changed on Friday is access: two working datasets that were internal to one of the world’s leading AI forecast systems are now sitting in public cloud storage, waiting for whoever wants to build with them.
Catalina Clarke
Editor-in-Chief, The Library Media
From The Library — the estate of our family.
