Google releases weathernext 3 datasets: satweave satellite imagery and pardig global precipitation data

Google DeepMind has made two foundational datasets, SatWeave and PARDIG, publicly available for global researchers and developers. These resources offer both real-time data streams and historical archives, intended for modeling in meteorology, hydrology, and climate science. SatWeave is a comprehensive global image stitching product that integrates data from five satellites—GOES-East, GOES-West, Meteosat 0, Meteosat Indian Ocean, and Himawari—resolving issues related to differing projections and formats. It updates hourly and provides over a decade of historical records, enabling hourly global forecasts for WeatherNext 3, which surpasses the traditional six-hourly refresh rate. Google also encourages its use in applications like solar power forecasting and wildfire monitoring. PARDIG is an AI-based global precipitation reanalysis dataset, utilizing a spherical grid Transformer architecture similar to WeatherNext 3. It trains on sparse satellite radar observations from NASA's core stations, using SatWeave imagery and ECMWF atmospheric fields as input. With a spatial resolution of approximately 10 km and a 15-minute temporal resolution, PARDIG shows superior performance over benchmarks like NASA IMERG-Final, particularly in capturing extreme precipitation, even in areas lacking ground radar. Both datasets, including PARDIG's historical archive spanning from 1999 to 2026, are hosted on Google Cloud Storage under a CC BY 4.0 license.