Google upgrades WeatherNext AI model with satellite data for hourly forecasts

Google announced the release of WeatherNext version 3, an AI-driven weather forecasting model that now ingests satellite data to shorten the lag between observed conditions and forecast output. By moving from six-hour to hourly forecast cycles, the model can provide more timely predictions. The update also raises spatial resolution and expands the underlying machine-learning architecture, prompting efficiency tweaks to keep computing demands in check.

Additionally, a separate ML model trained on satellite precipitation estimates has been added, offering multiple rainfall forecasts. These enhancements aim to bring AI-based forecasts closer to the performance of traditional physics-based models while using less computing power.