KARACHI: Google DeepMind and Google Research have introduced WeatherNext 3, an advanced artificial intelligence-powered global weather forecasting model designed to provide localized, real-time climate intelligence at an unprecedented resolution.
According to Google, the new system directly uses live geostationary satellite mosaics and ground station data, reducing dependence on slow and compute-heavy numerical approximations traditionally used in weather forecasting.
WeatherNext 3 produces fresh forecasts every hour and offers spatial resolution of up to 5 kilometres, which Google says is roughly five times sharper than its predecessor. The system can forecast surface temperature and moisture at 5-kilometre resolution and surface winds at 10-kilometre resolution.
Google said the model will provide crucial lead time as storms, weather fronts and extreme weather conditions develop rapidly. It is also expected to improve forecasts in areas where weather services have historically been limited or less reliable.
The company claimed that WeatherNext 3 reduces error metrics by up to 50 percent on key precipitation benchmarks compared with traditional numerical weather prediction baselines. It said users planning a day or more ahead would see up to 50 percent more accurate precipitation forecasts, with the biggest improvements in regions where forecasts have previously been less dependable.
The new model also provides specialised parameters for the energy sector, including 100-metre turbine-height wind speeds, multi-layer cloud cover and solar irradiance. Google said these outputs could help grid operators and clean-energy producers, including wind and solar power projects, make better use of weather information.
WeatherNext 3 has been rolled out from today across weather experiences in Google Search, the Gemini app, Google Maps, Google Maps Platform Weather API and Google Earth Engine.
