2026-11-10 · ← News
Google opens the data that teaches WeatherNext 3 to see rain every 15 minutes
Google has released SatWeave and PARDIG, two data foundations behind WeatherNext 3. PARDIG maps global precipitation at roughly 10 km every 15 minutes, while SatWeave combines satellite imagery into 17 channels and refreshes hourly.
The image could not be loaded.
Google DeepMind has opened historical archives and live feeds for two datasets used by WeatherNext 3. SatWeave unifies geostationary satellite imagery, while PARDIG uses AI to infer global precipitation from sparse spaceborne radar observations.
SatWeave rebuilds the global sky hourly and PARDIG maps rain every 15 minutes
SatWeave converts imagery from different satellites, projections and scan schedules into a global mosaic with 17 visible, infrared and water vapor channels. Its live feed refreshes hourly and the historical archive spans more than 10 years. That allows WeatherNext 3 to produce a new global forecast every hour instead of waiting for analysis data refreshed every 6 hours.
PARDIG, short for Precipitation AI Reanalysis: Densely Inferred GPM, learns from radar measurements collected by NASA’s GPM satellite. It uses SatWeave and ECMWF atmospheric fields as inputs. The output has roughly 10 km resolution, a 15 minute interval and average latency of about 5 hours. Its history runs from 1999 through 2026.
Open data can accelerate flood, energy and agricultural applications
The value extends beyond Google’s own model. Research teams can use the same inputs for flood, landslide and drought models as well as solar power and wildfire applications. Live data and archives are available in Google Cloud Storage, with historical PARDIG data also offered through Google Earth Engine.
This matters most in regions without dense ground radar networks. A local team gets a global, regularly updated base that it can combine with local measurements instead of first spending years building the underlying data infrastructure.
A global map still depends on unevenly distributed checkpoints
Google evaluated PARDIG against radar networks in the US, Europe and Japan and against more than 36,000 weather stations worldwide. The company says it improves detection skill for light to moderate hourly precipitation by more than 30% over NASA’s IMERG Final. That is a strong result, but the regional radar comparison does not cover the whole planet.
PARDIG also has average latency of about 5 hours. It can support hydrology, climate analysis and training, but it cannot replace local radar for warning people about a cloudburst in the next few minutes. Independent teams in poorly observed regions therefore need to test Google’s expectation of reliable accuracy everywhere on Earth.
Local models beyond the best measured regions will reveal the value
The next useful signal will be reproducible tests in Africa, Latin America, oceanic regions and mountainous terrain. Teams should also watch live feed reliability, licensing, transfer costs and the gap between the historical archive and data available in operations.
If SatWeave and PARDIG produce better local forecasts where radar is scarce, opening the data will matter more than WeatherNext 3 itself. If gains hold only in well observed regions, developers have received an impressive global layer with local blind spots.
Lilith's verdict
Google has placed a rain map in front of forecasters who often draw with too few points. Local teams must now learn whether it truly sees the storm between those points or merely paints a smooth sky over the gaps.