Google DeepMind’s WeatherNext Cyclones model uses Functional Generative Networks to efficiently generate large probabilistic ensembles for tropical cyclone track and intensity forecasting, trained end-to-end on roughly 20 terabytes of global atmospheric data alongside nearly 5,000 historical storms from the IBTrACS database. Despite operating at a comparatively coarse 28-by-28-kilometer resolution, about 100 times lower than traditional regional intensity models, the system produces state-of-the-art forecasts and can generate a 1,000-member, 15-day ensemble in under a minute on a single TPU. DeepMind reports the model delivers more than a full day of additional lead time in track, intensity, and wind-structure forecasts compared with existing operational systems.
