Top Highlights
- Our Functional Generative Networks (FGNs) swiftly generate 15-day weather forecasts with uncertainty estimates, scaling from 50 to 1,000 ensemble predictions to capture rare storm scenarios.
- WeatherNext models deliver high-accuracy cyclone predictions at surprisingly coarse resolutions (28x28km and 111x111km), challenging traditional assumptions about resolution needs.
- We are open-sourcing our models and code, enabling global researchers and agencies to improve weather prediction, including specialized and local forecasts.
- The breakthrough has extended cyclone prediction lead time by over a day, equating to a decade’s worth of progress, and can be explored via Google’s Weather Lab integrated with Google Earth AI.
Breakthrough in Cyclone Forecasting with AI
Recently, Google DeepMind made a big step forward in predicting cyclones. Using a new model called WeatherNext, they can forecast storms more accurately and faster than ever before. The system uses Functional Generative Networks to generate many different weather predictions quickly. Now, a 15-day forecast takes less than a minute to produce. This speed helps meteorologists assess risks more promptly and plan better responses.
How the Technology Works and Its Impact
DeepMind’s model creates thousands of predictions, capturing the uncertainty of weather patterns. Last year, it made 50 predictions at once. This year, the team increased that to 1,000. This larger set helps detect rare but dangerous events, like Hurricane Melissa in 2025. Surprisingly, their system works well even with data at much lower resolutions than traditional models. For example, WeatherNext only needs data at 28x28km resolution, compared to more detailed models. Researchers are still studying why this simpler approach yields such accurate results.
Open Source and Future Potential
DeepMind is sharing their code and models openly with the global community. They want scientists, weather agencies, and nonprofits to use and improve these tools. The models include WeatherNext Cyclones and WeatherNext 2, as well as a smaller version called WeatherNext 2-mini, which runs on a single TPU. Additionally, users can explore forecasts on Weather Lab, a new interface that visualizes storm predictions alongside global weather data. This collaboration aims to give communities more time to prepare and protect lives from storms and climate change.
Discover More Technology Insights
Learn how the Internet of Things (IoT) is transforming everyday life.
Stay inspired by the vast knowledge available on Wikipedia.
AITechV1
