Google announced on November 13, 2026 that it is opening historical archives and continuing feeds of satellite imagery and AI-estimated rainfall. Weather researchers and teams exploring water management, agriculture and renewable energy can reuse the data to investigate new applications.
SatWeave combines satellite images into a common map of clouds and atmospheric conditions. PARDIG uses AI to estimate precipitation from satellite observations and atmospheric data. Both datasets already contribute to Google’s WeatherNext 3 forecasting model; this release opens those inputs to others, rather than launching a consumer weather app.
Google proposes research into flood, landslide and drought modelling with PARDIG, and solar-power forecasting and wildfire monitoring with SatWeave. Those are proposed applications, not demonstrated improvements to warnings or business results. The datasets remain experimental research products.
Quarter-hour estimates, hours of delay
PARDIG estimates rainfall at 15-minute intervals on a grid spaced roughly 10 kilometres apart. But its continuing feed arrives in six-hour batches, with estimates about two to eight hours old when published—roughly five hours on average. For a team assessing time-sensitive uses, the practical distinction is that detailed timestamps do not mean immediate delivery.
SatWeave’s current feed updates hourly with about an hour’s delay. Its 2016–2025 archive has half-hourly observations; data from 2026 onward is hourly.
The access route depends on the task. Google Earth Engine offers static PARDIG archives: the near-real-time-labelled collection currently covers January 2016 through October 22, 2026, not a continuously updating feed. Google Cloud Storage carries the continuing PARDIG updates, alongside its archives, and is also where SatWeave is available.
For studies needing earlier years, PARDIG’s separate long-record version covers 1999–2023. It uses different inputs, and Google cautions that changes in satellite coverage and the transition between versions can create apparent shifts unrelated to climate. Simply joining the records is not enough to establish a long-term climate trend.
Promising results, uneven coverage
In their PARDIG preprint, Google’s researchers report stronger rainfall-detection results than comparison products, using radar networks in the US, Europe and Japan and more than 36,000 surface stations. Those are the authors’ evaluations against ground measurements, not an independent replication or proof of equal accuracy everywhere.
PARDIG’s documented limitations include less reliable estimates near the poles and difficulties with mountains, snow and light drizzle. Its quarter-hour rainfall rates are capped at 75 millimetres per hour, which can clip local peaks. SatWeave also has coverage gaps: polar observations and some types of imagery in certain regions are missing. Sensor errors or distortions from combining images can also affect the data.
For a research project, these limits make checking local coverage and comparing estimates with suitable local measurements a sensible first step. Neither dataset should be the sole basis for decisions affecting life or property; Google directs users to official weather and emergency alerts.
What open access covers
Both the archives and continuing observational feeds are licensed under CC BY 4.0. That license permits reuse and adaptation, including commercially, with credit, a license link and disclosure of changes. Google also lists third-party attribution requirements.
The dataset guides say no separate data-access application is required for a Google Cloud account. Earth Engine still needs a registered Cloud project with its API enabled and the right permissions.
Open licensing does not make processing cost-free. Earth Engine charges commercial users for computing and storage; some plans also have monthly platform fees. Qualifying noncommercial research can use it without an additional platform charge. Teams evaluating the release therefore need to check both whether the observations suit their research and what their chosen analysis platform will cost.
For background on the forecasting side, see RohitAI’s earlier coverage of WeatherNext and cyclone models.