Measured acceleration · Climate · Physics
WeatherNext Cyclones gains more than a day of forecast lead time
A global operational ensemble model for tropical-cyclone track, intensity, and wind-structure forecasting, evaluated on storms from 2023 through 2025.
Summary
A peer-reviewed Nature paper introduces WeatherNext Cyclones, an AI ensemble model for global tropical-cyclone forecasting. On storms from 2023 through 2025, its track, intensity, and wind-radii predictions averaged more than a day of lead-time advantage over leading operational models. The model was developed with operational forecasting organizations, used alongside National Hurricane Center workflows, and released with code and weights.
AI role
Learned global weather and cyclone dynamics from atmospheric analyses and historical storm records to generate probabilistic forecasts up to 15 days ahead.
Narrative role
This extends the weather-model timeline from faster inference to an operationally relevant gain in warning-quality lead time, with peer-reviewed evaluation, expert-forecaster collaboration, and an open release for downstream research and deployment.
Caveat
The headline results combine retrospective evaluation with early operational use in one high-stakes domain. Multi-season independent evidence is still needed to measure reliability and real-world warning outcomes, and the model does not replace official human forecasts.