Google DeepMind and Google Research have introduced WeatherNext 3, an AI-powered weather model designed to improve forecasting of rain, snow, temperature, wind and renewable energy conditions.
Google DeepMind and Google Research have introduced WeatherNext 3, a new artificial intelligence-powered weather model designed to deliver more detailed and frequent forecasts while improving predictions for rain, snow and other weather conditions.
The model is described as a major step forward from earlier AI-based forecasting systems, with Google saying it can generate forecasts at hourly intervals using direct observations from satellites and weather stations.
According to the information provided, an independent evaluation by Brightband found WeatherNext to be among the most accurate global weather models currently available.
Unlike older AI models that primarily relied on numerical weather data, WeatherNext 3 is trained using direct observations from the ground and satellites. The system can forecast key variables such as temperature and humidity at a resolution of 5 kilometers.
Other land-based variables are forecast at 10-kilometer resolution, while wind-speed forecasts use a 25-kilometer resolution.
Google says WeatherNext 3 is five times more granular than WeatherNext 2, which provided data every six hours on a 25-kilometer grid.
Improved rain, snow and renewable energy forecasts
Rainfall and snowfall forecasting have received particular attention in the new model. WeatherNext 3 has been trained using NASA’s Integrated Multi-satellitE Retrievals for GPM (IMERG) precipitation data as well as radar observations.
During testing, the model reportedly delivered improvements of up to 60% in medium-range forecasts and up to 30% in initial forecasts.
WeatherNext 3 has also been designed to support renewable energy projects. It can forecast wind speeds at an altitude of 100 meters, roughly the height at which wind turbines operate, while also assessing cloud cover and sunlight conditions relevant to solar power generation.
Google has made WeatherNext 3 data available to researchers, developers and businesses through BigQuery, Earth Engine and Google Cloud Storage.
The model is also being integrated into Google Search, the Gemini app and Google Maps, with Google saying it can improve longer-range forecasts by up to 50%.
Despite the advances, Google has cautioned that weather remains inherently uncertain. For emergencies and natural disasters, the company recommends relying on official warnings issued by local meteorological authorities.
About The Author
