DeepMind's open weather data give regional climate projects an alternative to ground radars
Listen to this article
Read by Anchor
When sudden floods hit the region’s valleys or solar radiation levels fluctuate, meteorology teams and infrastructure clash with the lack of dense ground-based radar networks that Western countries rely on. Google DeepMind opened a technical pathway to address this gap by providing two open Earth-science datasets, SatWeave and PARDIG, via Google Cloud Storage and Earth Engine, to give developers and researchers data that feed weather, hydrology and climate forecasting applications.
The SatWeave system combines images captured by five satellites covering different parts of the globe, including GOES-East, GOES-West, MetOp-Sat-0, MetOp-Sat Indian Ocean and the Himawari satellite, to create a global mosaic that blends observations from six geostationary satellites across 17 channels of visible light, water vapor and infrared.The decisive operational difference lies in the update speed.While conventional models require a six-hour wait to refresh analysis data, SatWeave data are refreshed every hour, allowing the WeatherNext 3 model to generate new global forecasts on a continuous hourly cadence, supported by a historical archive of more than a decade that the platform makes available alongside the live stream.
Along the rainfall track, the PARDIG suite arrives as an AI-driven reanalysis of precipitation built on intensive inference from NASA’s Global Precipitation Measurement (GPM) Core Observatory. This model was trained directly on the observatory’s scattered space-borne radar data, using SatWeave images and atmospheric fields from the European Centre for Medium-Range Weather Forecasts (ECMWF) as primary inputs. The system not only launched a live data stream but also provided a historical archive from 1999 to 2026, whose accuracy was evaluated against three continental radar networks in the United States, Europe and Japan, and more than 36 000 weather stations worldwide.
The scientific effort was developed by joint teams from Google DeepMind and Google Research, in collaboration with Google Cloud and Google Geo, relying on open-data contributions from leading agencies such as NASA, ECMWF, the U.S. National Oceanic and Atmospheric Administration (NOAA) and space and meteorological agencies in Japan, Korea and the Netherlands.In our analysis we examined the announced field applications of the two suites.They appear to target real-world challenges ranging from issuing earlier and more accurate flood warnings in areas lacking ground radar networks to tracking wildfire smoke and optimizing solar-energy generation efficiency.
What this means for the region
Most countries in the region, outside major airport zones, do not have ground-based Doppler radar networks covering their vast deserts and coastlines as are available in Europe or Japan. Providing a 27-year historical precipitation archive together with direct coverage from the MetOp-Sat Indian Ocean satellite gives data engineers in Gulf municipalities and the water ministries of Egypt and the Levant a reliable digital foundation for building early-warning systems for flash floods without needing billion-dollar investments in new ground stations.
For operators of large solar farms in Saudi Arabia, the United Arab Emirates and Egypt, breaking the six-hour update loop and moving to hourly updates across 17 spectral channels reduces the cost of purchasing commercial weather products, and enables fine-tuning of instantaneous generation forecasts to avoid wasted storage on electricity transmission networks.
The operational burden now shifts from the difficulty of obtaining raw data to inference-engineering skills; the data are available on Google Cloud, and technical talent is required to master the integration of these hourly streams with local AI models to protect critical infrastructure and manage water resources in real time.