AI model improves tropical cyclone forecasts by a day or more

Researchers have introduced WeatherNext Cyclones (WN-C), an AI-driven operational weather model that produces ensemble forecasts for tropical cyclone track, intensity, and wind radii worldwide. The model was trained on global analysis data and a historical database of tropical cyclones, generating up to 1,000 ensemble members that project global weather and cyclone scenarios up to 15 days ahead.

Evaluated on cyclones from 2023 to 2025, WN-C's predictions provided an average lead-time advantage of a day or more over leading operational models for track, intensity, and wind radii. The accuracy improvement is described as comparable to the progress achieved over the previous decade of conventional operational development.

The model achieves these results using input data orders of magnitude coarser than that used by regional forecasting models. The authors say this suggests high resolution is not a strict prerequisite for state-of-the-art intensity forecasting and that coarser atmospheric data contains more intensity signal than previously recognized.

When WN-C predictions were included in a weighted-average consensus ensemble, the ensemble's overall skill improved substantially. The scalability of the AI approach allows for much larger ensembles than the typical 50-member ensembles used in conventional operations, which the researchers say better captures rare events.

The study frames the work as a step-change toward more reliable and timely forecasts and warnings that can help protect lives and reduce the impacts of tropical cyclones. The model is designed to provide operational ensemble guidance to human forecasters.

Limitations noted in the abstract include the evaluation period covering 2023–2025 and the reliance on existing global analysis and historical cyclone databases for training. The abstract does not detail computational costs, real-time deployment status, or performance across all cyclone basins individually.