Google DeepMind has announced the release of two foundational datasets, PARDIG and SatWeave, aimed at advancing global weather, hydrology, and climate research. These datasets, which underpin the capabilities of the recently launched WeatherNext 3 model, are now openly available to researchers and developers worldwide. The initiative seeks to address critical challenges in observing and predicting Earth's weather and water cycles.

SatWeave: A Unified Satellite Mosaic

SatWeave integrates data from six geostationary satellites—GOES-East, GOES-West, Meteosat 0, Meteosat Indian Ocean, and Himawari—into a cohesive global mosaic. Unlike existing composites that offer limited channels and infrequent updates, SatWeave provides 17 visible, water vapor, and infrared channels with hourly updates. This enables WeatherNext 3 to produce global predictions more frequently than traditional models, which rely on six-hourly data refreshes. Historical archives spanning over a decade and live feeds are now accessible, opening avenues for applications such as solar power forecasting and wildfire monitoring.

PARDIG: High-Resolution Global Precipitation Data

PARDIG (Precipitation AI Reanalysis: Densely Inferred GPM) delivers highly accurate global precipitation maps with a 15-minute temporal resolution and approximately 10-kilometer spatial resolution. Built on AI trained with NASA's Global Precipitation Measurement (GPM) Core Observatory data, SatWeave imagery, and ECMWF atmospheric fields, PARDIG addresses gaps in precipitation tracking, particularly in regions lacking dense ground radar networks. Evaluations against radar networks in the US, Europe, and Japan, as well as over 36,000 weather stations, show that PARDIG outperforms NASA's IMERG-Final dataset, improving detection skill for light-to-moderate precipitation by over 30%.

Empowering Global Research

Both datasets are available via Google Cloud Storage, with PARDIG's historical catalogs also hosted on Google Earth Engine. By providing live and historical data, Google DeepMind aims to empower the global research community to develop improved models for critical applications, including flood, landslide, and drought prediction. This release is part of Google's broader Earth AI portfolio, which leverages advanced AI systems to address pressing environmental challenges.