Google DeepMind Launches WeatherNext 3 with Real-Time Satellite Integration

Google has introduced WeatherNext 3, an advanced global AI weather model that updates hourly by incorporating real-time satellite observations. [1][2] Previous AI models relied heavily on data from traditional weather simulations, which can result in a roughly six-hour lag. [3] By feeding real-time satellite observations directly into the model alongside sparse weather station observations, WeatherNext 3 addresses the six-hour lag common in traditional systems. [3] This rapid processing helps the new model predict precipitation stemming from fast-moving weather systems more effectively. [1]

The updated model visualizes surface variables like temperature and moisture at a five-kilometer spatial resolution. [1][2] Atmospheric variables, such as wind speed, are forecast at a 25-kilometer resolution. [2][3] Google claims the new system can deliver precipitation forecasts up to 50 percent more accurately a day or more in advance. [3] The company has integrated the model across its product ecosystem, including Search, Maps, Gemini, and Cloud. [2]

WeatherNext 3 shifts AI weather forecasting from relying on lagging simulation data to digesting direct, real-time satellite feeds. This transition to immediate observational data allows for faster updates that are critical for tracking volatile, localized weather events like sudden storms. By integrating this model directly into its massive consumer platforms, Google is positioning its AI research as a primary utility for everyday logistical decisions, bypassing traditional meteorological delays. What the sources don't address: whether Google plans to open-source the model weights or offer an API for third-party developers outside of Google Cloud.