WeatherNext 3: Hourly global weather forecasts using live satellite data (blog.google)

Google DeepMind and Google Research have launched WeatherNext 3, an advanced global weather forecasting model that utilizes real-time satellite data to provide highly localized and precise hourly weather predictions. This model significantly enhances previous forecasting methods by generating forecasts at a 5-kilometer resolution, five times sharper than its predecessor, WeatherNext 2, which operated on a 25-kilometer grid. By directly training on live satellite observations and sparse weather station data, WeatherNext 3 addresses the challenges of rapidly changing weather patterns, offering a more accurate and meaningful prediction of local atmospheric conditions. This breakthrough is particularly impactful for regions that have historically been underserved by weather forecasting due to high computational costs. By delivering high-resolution forecasts every hour, WeatherNext 3 improves the accuracy of precipitation predictions, making it vital for sectors like agriculture, energy production, and emergency response. The model also forecasts critical variables for renewable energy, such as wind speeds at turbine height, allowing for better energy production planning. Integrated across Google products, WeatherNext 3 promises to enhance weather intelligence globally, offering users more reliable forecasts tailored to their local environments.