Operational Tropical Cyclone Forecasting with AI

A new artificial intelligence system has been deployed for operational tropical cyclone forecasting, offering improved accuracy and lead times compared to existing models. Developed by researchers including scientists from Google DeepMind, the National Oceanic and Atmospheric Administration (NOAA), and several academic institutions, the system called WeatherNext Cyclones (WN-C) provides advanced ensemble forecasts for tracking, intensity, and size of tropical cyclones globally. According to the study published in Nature, the model's performance demonstrates a measurable improvement in predicting cyclone behavior, potentially enhancing early warning systems and disaster preparedness efforts. The research team trained WN-C using a combination of global analysis data and a comprehensive database of historical tropical cyclones spanning decades. This training enabled the model to generate large ensembles of potential weather scenarios, extending up to 15 days ahead. When tested against real-world cyclones from 2023 to 2025, the model showed an average lead time advantage of one day or more over current operational forecasting models. The improvements in accuracy were described as being comparable to the advancements made in operational meteorology over the past decade. Unlike traditional high-resolution models, WN-C operates effectively with input data that is significantly coarser in detail. Researchers noted that this suggests high-resolution data might not always be essential for accurate intensity forecasting. Moreover, they found that the coarser atmospheric data used by WN-C contained more relevant signals related to cyclone intensity than previously assumed. This discovery could have implications for how weather models are designed and optimized moving forward. Incorporating predictions from WN-C into a weighted-average consensus ensemble further enhances its predictive capabilities. The model’s design allows for scalable ensembles of up to 1,000 members, which is far greater than the typical 50-member ensembles used in conventional forecasting. Larger ensembles enable better representation of rare and extreme weather events, improving the reliability of long-range forecasts. The development of WN-C involves collaboration between multiple organizations, including Google DeepMind based in London, the University of Waterloo in Canada, Google Research in the United States, NOAA's National Hurricane Center in Florida, and the Cooperative Institute for Research in the Atmosphere at Colorado State University. Additional contributions came from the UK Met Office in Exeter. Key contributors include Ferran Alet, Tom R. Andersson, Ilan Price, Stratis Markou, Andrew El-Kadi, and Dominic Masters from Google DeepMind, along with Wallace Hogsett, David Zelinsky, John Cangialosi, and Jonathan Martinez from NOAA, and James Franklin, Mark DeMaria, Kate Musgrave, Caroline L. Bain, and Helen Titley from other institutions. The findings highlight the growing role of artificial intelligence in advancing weather prediction technologies. While the study acknowledges the need for further validation and integration into operational forecasting practices, it underscores the potential of AI-driven models to complement and enhance human expertise in meteorological forecasting. The ability of WN-C to provide detailed ensemble guidance to human forecasters marks a significant shift toward more reliable and timely weather predictions, which could ultimately help reduce the risks associated with tropical cyclones. Researchers emphasized that while the model shows promise, its implementation requires careful evaluation within existing operational frameworks. They noted that ongoing testing and refinement would be necessary before widespread adoption. Additionally, the study pointed out that the success of WN-C depends on the quality and availability of input data, which must be continuously updated and maintained. As the model moves closer to operational deployment, meteorologists and climate scientists are expected to conduct extensive evaluations to assess its effectiveness under varying conditions. These assessments will involve comparing WN-C's predictions against actual storm developments and refining the model based on feedback from field observations. The ultimate goal is to integrate this technology into national and international weather services to improve public safety and disaster response strategies.