Google DeepMind releases SatWeave satellite mosaic and PARDIG precipitation datasets openly on Google Cloud Storage

Google DeepMind has opened two datasets that feed its WeatherNext 3 weather model. SatWeave is an hourly global mosaic of geostationary satellite imagery across 17 channels, with an archive spanning more than a decade. PARDIG is an AI-generated global precipitation dataset at 15-minute and roughly 10 km resolution, with about five hours of average latency and archives from 1999 to 2026. Both are on Google Cloud Storage with live feeds, and PARDIG's historical catalogues are also in Google Earth Engine.

Researchers and developers working on flood, drought, landslide, solar power or wildfire monitoring get a free, regularly updated global rainfall and satellite source, which matters most where ground radar is sparse. The five-hour delay suits monitoring and analysis better than real-time warnings. The practical effect is mostly for specialists, and the datasets are tied to Google's cloud and Earth Engine ecosystem, so adoption will depend on licence terms and how easy access is in practice.

The claim that PARDIG beats NASA's IMERG-Final by more than 30% on hourly detection of light-to-moderate rain is Google's own result, and we have no independent validation. "Better at extreme precipitation" is also unquantified in the summary, and a dataset made by an AI model can carry model biases that benchmarks hide. This is a data release, not a new frontier model, so any framing as a major AI breakthrough would overstate it.

1. 01Independent evaluations of PARDIG against IMERG-Final and ground gauges, especially for extreme rainfall and in data-poor regions, over the next few months.
2. 02Evidence of real use: published papers, hydrology or disaster-agency pipelines, or startups building on SatWeave and PARDIG, plus any reported download or Earth Engine usage.
3. 03Licence terms, access costs (egress or compute charges), and whether the live feeds keep their stated latency and update schedule reliably, along with any updates beyond the 2026 archive.