Overview
- Google announced WeatherNext 3 on Thursday and has begun integrating its outputs into Search, Maps, Gemini, the Google Maps Platform, and Google Cloud.
- The model pulls low‑latency geostationary satellite mosaics and sparse station observations to generate a fresh global forecast every hour, cutting the six‑hour data lag of traditional systems.
- WeatherNext 3 reports higher spatial detail for many surface variables — up to about 5 km for temperature and moisture — and Google says precipitation forecasts one day or more ahead are up to 50% more accurate.
- The system produces 64‑member probabilistic ensembles with select cycles to 15 days and has been documented in an arXiv paper and ranked top on Brightband’s Operational WeatherBench in independent live evaluations.
- Practical questions remain about commercial access, validation by national agencies, and how organizations will adopt ensemble outputs, but the model could give farmers, emergency services, and renewable‑energy operators faster, more local forecasts where ground observations are sparse.