Google has released WeatherNext 3, an artificial intelligence system designed to improve predictions for rain and snowfall. The new model claims unprecedented resolution by generating global forecasts five times sharper than its predecessor. It achieves this by training on real-time weather observations rather than relying solely on standard datasets. Samier Merchant, a research engineer at Google Research, explained that the system now incorporates data sources most global models typically ignore. This update allows the software to process live satellite imagery and ground-level measurements directly. The company states this approach reduces errors in forecasting precipitation patterns across different regions.
The practical value lies in the ability to resolve smaller weather features that previous models smoothed over. Higher resolution means localised storms and microclimates become visible in the forecast data. This precision helps meteorologists and planners anticipate specific conditions without relying on broad regional averages. The shift from static training data to dynamic observation inputs represents a significant technical adjustment in how weather is modelled.
- Five times sharper global resolution than WeatherNext 2
- Direct training on real-time satellite and ground observations
- Improved accuracy for predicting rain and snowfall
