On November 13, Google DeepMind released the two datasets behind its WeatherNext 3 model. SatWeave stitches imagery from geostationary satellites such as GOES, Meteosat and Himawari into one global mosaic across 17 visible, water vapor and infrared channels, updated hourly, with an archive spanning more than a decade. PARDIG is an AI-generated global precipitation dataset trained on NASA GPM spaceborne radar, with 15-minute time steps, roughly 10 km resolution and about five hours of average latency. DeepMind says that, evaluated against radar networks in the US, Europe and Japan and more than 36,000 weather stations, it improves hourly detection of light-to-moderate rain by more than 30% over NASA IMERG-Final, with archives from 1999 to 2026. Live feeds and archives for both are on Google Cloud Storage, and PARDIG historical catalogs are also in Earth Engine.
Key Takeaways
- ✓SatWeave fuses geostationary satellites including GOES-East/West, Meteosat 0, Meteosat Indian Ocean and Himawari across 17 channels, updated hourly, with an archive of more than a decade (the post lists five satellites in the text and six in a caption)
- ✓PARDIG stands for Precipitation AI Reanalysis: Densely Inferred GPM; it takes SatWeave imagery and ECMWF atmospheric fields as input and is trained on radar from NASA's GPM Core Observatory
- ✓PARDIG runs at 15 minutes and about 10 km with roughly five hours of average latency; DeepMind reports more than 30% better hourly detection skill (CSI) than IMERG-Final on light-to-moderate rain and strong results on extreme rain
- ✓Live feeds and historical archives of both datasets are open on Google Cloud Storage, and PARDIG historical catalogs are also in Google Earth Engine
- ✓Hourly SatWeave input lets WeatherNext 3 issue a new global forecast every hour instead of waiting for analysis data refreshed every six hours
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and pain points
A forecast is only as good as its view of the atmosphere right now. Satellites and ground stations observe constantly, but each satellite uses different projections, formats, sensors and scan schedules (10 to 30 minutes), so building one coherent global picture is hard, and ground radar and rain gauges miss many regions. DeepMind is now opening the two datasets that feed WeatherNext 3.
Architecture highlights
SatWeave stitches geostationary satellites such as GOES, Meteosat and Himawari into a global mosaic that keeps 17 visible, water vapor and infrared channels and updates hourly; most existing global composites offer four or fewer channels or update every few hours. PARDIG uses an AI architecture similar to WeatherNext 3, takes SatWeave imagery and ECMWF atmospheric fields as input, and is trained directly on sparse spaceborne radar from NASA's GPM Core Observatory to output 15-minute, roughly 10 km global precipitation maps.
Benchmarks and data
DeepMind evaluated PARDIG against radar networks in the US, Europe and Japan and more than 36,000 weather stations (2016 to 2023). It beat global baselines including NASA IMERG-Final at every rain rate tested, improved hourly detection skill (CSI) on light-to-moderate rain by more than 30%, and was especially strong on extreme rain. These are vendor-reported results.
Getting started
Live feeds and historical archives for both datasets (SatWeave over a decade, PARDIG 1999 to 2026) are open on Google Cloud Storage, and PARDIG historical catalogs are available in Earth Engine. Use cases include flood warnings, landslide and drought models, wildfire smoke tracking and solar power forecasting, especially where ground radar is missing.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.