WeatherNext: AI for Cyclone Forecasting

Google DeepMind has developed WeatherNext, an AI model that significantly improves the forecasting of cyclones, hurricanes, and typhoons. This model offers a full day's advantage in terms of accuracy compared to previous methods, equivalent to a decade of meteorological progress.

WeatherNext overcomes the challenge of simultaneously modeling the trajectory (influenced by global currents) and intensity (dependent on fine local processes) of cyclones, using a unified approach. It relies on Functional Generative Networks (FGNs) to produce thousands of scenarios, allowing for the assessment of the probability of extreme risks with surprisingly low data resolution (28x28km).

The model has already proven its effectiveness in real-world situations, notably during Hurricane Melissa in 2025. Google DeepMind has decided to open-source WeatherNext and WeatherNext Cyclones to accelerate research and enable meteorological agencies and researchers worldwide to adopt and develop it, thereby helping to strengthen community resilience to natural disasters.