Skip to content
Neural Daily

ResearchNews

DeepMind opens the datasets behind WeatherNext 3

🇭🇺 Olvasd el magyarul →

Google DeepMind released SatWeave and PARDIG, the two datasets behind its WeatherNext 3 weather model. One is an hourly global satellite mosaic, the other an AI-based precipitation dataset, both with archives and live feeds.

· 2 min read

Rétegzett időjárási adattérképek a Föld körül, a Google DeepMind bejelentő képe
Image: Google DeepMind · Press kit · Source

What DeepMind released

The WeatherNext team opened two datasets, both key to WeatherNext 3, released in September (Google DeepMind). Live feeds and historical archives are openly available on Google Cloud Storage, and PARDIG's archive is also in Google Earth Engine.

SatWeave: one view of Earth

  • It merges geostationary imagery from GOES-East, GOES-West, Meteosat 0, Meteosat Indian Ocean and Himawari into a global mosaic.
  • It keeps 17 visible, water vapor and infrared channels, where most global composites offer four or fewer.
  • It updates hourly, letting WeatherNext 3 issue new global forecasts every hour instead of relying on analysis data refreshed every six hours.
  • The historical archive spans more than a decade.

PARDIG: precipitation without ground radar

PARDIG uses an architecture similar to WeatherNext 3, trained on sparse spaceborne radar observations from NASA's GPM Core Observatory, with SatWeave imagery and ECMWF atmospheric fields as inputs.

  • Resolution: 15 minutes in time and about 10 km in space, with average latency of about five hours.
  • Evaluated against radar networks in the US, Europe and Japan and over 36,000 weather stations.
  • DeepMind says it beats NASA's IMERG-Final at all rain rates and improves hourly detection skill by more than 30% for light to moderate precipitation.
  • Archive from 1999 to 2026.

Why it matters

Regions without dense radar have struggled to get good precipitation data for flood, drought and landslide models. PARDIG fills that gap, and SatWeave could support solar forecasting and wildfire monitoring. In AI weather, open models are now being followed by open input data (NVIDIA StormScope EU: short-range weather AI for Europe).

    By Mikola Máté Patrik

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles