FuXi-TC: A generative framework integrating deep learning and physics-based models for improved tropical cyclone forecasts

Tropical cyclones (TCs) are among the most devastating natural hazards, yet their intensity remains notoriously difficult to predict. NWP models are constrained by both computational demands and intrinsic predictability, while state-of-the-art deep learning-based weather forecasting models tend to underestimate TC intensity due to biases in reanalysis-based training data. Here, we present FuXi-TC, a diffusion-based generative forecasting framework that combines the track prediction strength of the FuXi model with the intensity representation of NWP simulations. By conditioning a diffusion model on the large-scale forecasts of the global FuXi model, FuXi-TC effectively downscales and delivers higher-accuracy forecasts of fine-grained variable fields such as wind speed and precipitation. In evaluations across the 2024 Western North Pacific, our approach matches the TC intensity forecast skill of the operational ECMWF deterministic model while delivering superior precipitation forecasts. Meanwhile this is achieved with significantly higher inference speeds and lower computational costs. Moreover, FuXi-TC demonstrates robust zero-shot generalization directly when applied to North Atlantic hurricanes without any fine-tuning. When applied to the FuXi ensemble model, this framework effectively yields well-dispersed probabilistic forecasts and refines the ensemble intensity predictions.