Tropical cyclones can transform from distant swirls over warm ocean water into life-threatening disasters within hours. Their winds can tear apart buildings, storm surges can inundate coastal communities, and torrential rainfall can trigger floods far from the point where a storm makes landfall. Forecasting these systems remains exceptionally difficult because small errors in a storm’s position, structure, or intensity can produce dramatically different outcomes days later. A new artificial intelligence system, however, is offering forecasters a potentially powerful new advantage: more accurate predictions of where tropical cyclones will move, how strong they will become, and how far their damaging winds may extend.
Called WeatherNext Cyclones, or WN-C, the system is an operational AI weather model designed to produce ensemble forecasts for tropical cyclones across the globe. Rather than generating a single prediction, WN-C creates many plausible versions of the future, each representing a different possible evolution of the atmosphere. These scenarios can extend as far as 15 days, allowing meteorologists to assess not only the most likely track but also the range of outcomes that could emerge if the storm unexpectedly accelerates, weakens, intensifies, or changes direction.
The model was trained using two major sources of information. The first is global atmospheric analysis data, which provides a detailed reconstruction of past weather conditions, including pressure, temperature, humidity, winds, and other variables. The second is a global historical database of tropical cyclones, containing records of storm tracks, intensity changes, and wind-field characteristics. By learning from both the broad atmospheric environment and the observed behavior of past storms, WN-C aims to connect large-scale weather patterns with the complex processes that govern cyclone development.
That connection is one of the central scientific challenges in tropical cyclone prediction. A storm’s future depends on its interaction with the surrounding atmosphere and ocean, including vertical wind shear, moisture, sea-surface temperatures, pressure patterns, and the structure of the storm itself. These factors can change rapidly and interact in ways that are difficult for conventional forecasting systems to represent. WN-C approaches the problem by using machine learning to identify patterns in historical data that may reveal how a cyclone is likely to evolve under comparable environmental conditions.
According to the study, WN-C was evaluated on tropical cyclones occurring between 2023 and 2025. Its forecasts for track, intensity, and wind radii provided an average lead-time advantage of at least one day compared with leading operational models. In practical terms, that means forecasters could receive a reliable indication of a storm’s likely behavior earlier than before. Even a few additional hours can make a critical difference for emergency planning, evacuation decisions, the positioning of response teams, and public warnings.
The reported improvement is particularly notable because the system uses atmospheric inputs that are orders of magnitude coarser than those employed by many regional forecasting models. Regional models often rely on very high-resolution simulations to represent the inner structure of a cyclone, including its eyewall, rainbands, and tightly concentrated wind maximum. The performance of WN-C suggests that detailed local-scale information may not be the only pathway to accurate intensity prediction. Coarser global atmospheric data may contain more useful information about cyclone strength and development than scientists previously assumed.
This finding challenges a long-standing expectation in numerical weather prediction: that increasingly fine spatial resolution is always necessary for better forecasts. High-resolution models remain essential for many applications, especially for predicting local rainfall, storm surge, and detailed impacts near land. But the results from WN-C indicate that artificial intelligence can extract predictive signals from relatively broad atmospheric patterns, perhaps identifying relationships that are difficult to encode directly in traditional physical simulations. The model does not eliminate the need for physics-based forecasting; instead, it adds a different form of evidence to the forecasting process.
WN-C also appears to improve forecasts when combined with existing systems. The researchers found that including its predictions in a weighted-average consensus ensemble substantially increased forecast skill. Consensus forecasting works by combining multiple models, giving greater influence to systems that perform better for a particular type of prediction. Because different models can make different errors, their combination often produces a more reliable estimate than any single forecast. WN-C’s contribution may therefore be valuable even when it is used alongside established operational guidance rather than as a replacement for it.
The scalability of the AI system creates another important advantage. WN-C can generate ensembles containing up to 1,000 members, compared with the approximately 50 members commonly used in conventional ensemble forecasting. A larger ensemble can sample a wider range of possible futures and improve the chances of detecting rare but dangerous outcomes, such as rapid intensification, an unusual turn toward a densely populated coastline, or a storm that becomes significantly larger than expected. These low-probability scenarios are difficult to capture consistently, yet they are often the cases with the greatest consequences.
The researchers emphasize that WN-C is intended to provide guidance to human forecasters, not to make warnings independently. Meteorologists must still interpret the model’s predictions, compare them with satellite observations and other forecasting systems, and account for local vulnerabilities and rapidly changing conditions. Nevertheless, the combination of longer lead times, improved track and intensity estimates, and larger ensembles could mark a significant shift in operational tropical cyclone forecasting. As climate change continues to influence ocean temperatures, rainfall extremes, and the risks faced by coastal populations, tools capable of identifying dangerous possibilities earlier may become increasingly important. By turning vast historical weather records into thousands of potential storm futures, WN-C offers a glimpse of how artificial intelligence could reshape the race to understand—and prepare for—the next major cyclone.
Subject of Research: Artificial intelligence for operational tropical cyclone forecasting, including track, intensity, wind radii, and ensemble prediction.
Article Title: Operational Tropical Cyclone Forecasting with AI
Article References: Alet, F., Andersson, T.R., Price, I. et al. Operational Tropical Cyclone Forecasting with AI. Nature (2026). https://doi.org/10.1038/s41586-026-10953-2
Image Credits: AI Generated
DOI: 10.1038/s41586-026-10953-2
Keywords: WeatherNext Cyclones, WN-C, artificial intelligence, tropical cyclones, hurricane forecasting, ensemble prediction, storm track, cyclone intensity, wind radii, operational meteorology, extreme weather, climate risk
