OFICIAL Google DeepMind Blog

AI model achieves breakthrough in forecasting cyclones — Google DeepMind

What happened
Based on Google DeepMind Blog · Aug 06, 2026

Google DeepMind and collaborators open-sourced WeatherNext AI models, delivering cyclone forecasts with an extra day of lead time, equivalent to a decade of meteorological progress.

🚩 Report a correction ↗ Open source
Key points
·
WeatherNext enables accurate cyclone forecasts that can give an extra day of warning.
·
Predicting how dangerous cyclones develop is a longstanding challenge where every hour counts.
·
Tropical cyclones — also known as hurricanes or typhoons — are among the most destructive weather phenomena on Earth, responsible for more than 700,000 deaths and $1.4 trillion in economic losses globally over the past 50 years.
·
For forecasters, issuing timely, accurate warnings is a constant race against time.
Key numbers
·
The model achieves state-of-the-art accuracy while using lower-resolution inputs (28x28km), surprising scientists and challenging traditional high-resolution approaches.

Google DeepMind and Google Research, in partnership with agencies including the National Hurricane Center and UK Met Office, developed WeatherNext, an AI model that improves cyclone track, intensity, and wind structure predictions. The model provides forecasters with an additional day of accuracy, meaning three-day forecasts now match the reliability of prior two-day predictions. This advance is described as roughly equivalent to a decade of progress in meteorological forecasting techniques.

During the 2025 hurricane season, WeatherNext assisted the National Hurricane Center in forecasting Hurricane Melissa, predicting rapid intensification and landfall in Jamaica. The model generated 1,000 possible scenarios per cyclone to support decision-making, enabling timely warnings and preparation. The open-sourced models, WeatherNext 2 and WeatherNext Cyclones, aim to empower researchers and forecasters globally by improving disaster preparedness and renewable energy planning.

WeatherNext bridges a longstanding gap in cyclone modeling by combining global atmospheric data with fine-scale thermodynamic processes in a single AI system. The model achieves state-of-the-art accuracy while using lower-resolution inputs (28x28km), surprising scientists and challenging traditional high-resolution approaches. It was trained on nearly 20 terabytes of global atmospheric data and historical cyclone records, enabling efficient ensemble predictions that capture weather uncertainties.

The open-source release includes WeatherNext Cyclones, WeatherNext 2, and a compact version, WeatherNext 2-mini, designed for accessibility. Forecasts are available via Weather Lab, part of Google Earth AI, which visualizes predictions for temperature, precipitation, and wind speed. The initiative invites collaboration to enhance global weather resilience, combining AI with human expertise to improve early warnings and climate adaptation strategies.

Original source → Deals on Clipraptor.com →