Skip to main content

Google DeepMind details SatWeave and PARDIG weather datasets

13 NOVEMBER 2026·2 MIN READ·4 SOURCES

Google DeepMind has described two datasets intended to support weather, hydrology and climate applications: SatWeave, a global satellite mosaic, and PARDIG, an AI-based precipitation dataset.

Google DeepMind details SatWeave and PARDIG weather datasets

Key takeaways · 3

  • 01

    Google says SatWeave lets WeatherNext 3 initialize new global forecasts hourly, compared with six-hour updates for conventional numerical weather prediction analysis cycles.

  • 02

    PARDIG has an average latency of about five hours and is used to train WeatherNext 3 precipitation forecasts.

  • 03

    Researchers can access SatWeave on Google Cloud Storage under CC BY 4.0 without submitting an access-request form if they have a Google Cloud account.

Two datasets for weather applications

Google DeepMind says SatWeave and PARDIG are intended to help researchers and developers build weather, hydrology and climate applications.[1] The company describes both datasets as key to advances in WeatherNext 3.[1] They serve different roles: SatWeave brings together satellite observations, while PARDIG is an AI-based global precipitation dataset used to train WeatherNext 3’s precipitation forecasts.[2][1] The announcement therefore links a broad set of atmospheric observations with a precipitation-specific training resource, rather than presenting the two datasets as interchangeable.[2][1]

SatWeave combines varied satellite inputs

Google says SatWeave combines observations from GOES-East, GOES-West, Meteosat Prime, Meteosat Indian Ocean and Himawari into a global satellite mosaic.[2] It covers 17 spectral bands, including visible, near-infrared, shortwave-infrared, water-vapor and thermal-infrared bands.[2] Bringing the inputs together involves differences in projections, calibration and file formats, and the source satellites have scan intervals ranging from 10 to 30 minutes.[2] Google says SatWeave is publicly available on Google Cloud Storage under CC BY 4.0, and Google Cloud account holders can query it without an access-request form.[2]

Hourly forecast initialization

Google says SatWeave lets WeatherNext 3 initialize new global forecasts hourly; conventional numerical weather prediction analysis cycles update every six hours.[2] Google DeepMind says it has used SatWeave in WeatherNext 3 medium-range forecasts and is exploring applications in solar power and wildfire monitoring.[1] The stated hourly cadence describes forecast initialization, not a guarantee that every downstream application or forecast product updates hourly.[2] For teams assessing the dataset, that distinction matters: the evidence describes how Google uses the mosaic, but does not establish results for solar-power or wildfire-monitoring systems.[1]

PARDIG’s training data and coverage

Google says PARDIG is trained on precipitation observations from NASA/JAXA’s GPM Core Observatory and uses satellite observations and ECMWF atmospheric fields as inputs.[3] WeatherNext 3 uses PARDIG as a precipitation training target, and Google reports an average latency of about five hours.[3] Google says it evaluated PARDIG against radar networks in the United States, Europe and Japan, as well as more than 36,000 weather stations worldwide.[3] The company says learning from spaceborne sensors allows consistent coverage over oceans and regions without dense radar or gauge networks.[3]

These datasets offer professionals distinct resources to assess: a multi-satellite view that Google uses for hourly forecast initialization, and a precipitation dataset used to train WeatherNext 3. Teams considering applications in areas such as solar power or wildfire monitoring should distinguish Google’s stated exploration of those uses from demonstrated results, and account for PARDIG’s reported latency.

Why it matters
For teams

AI training for your team

kju.ai is built for teams: daily sessions, team analytics, and rollout support for your organization.

How this developed

  1. 13 November 2026

    Google DeepMind details SatWeave and PARDIG weather datasets

Sources

AI fluency, one session a day, built for your work.