3DTCR: Vortex-following generative reconstruction improves tropical cyclone intensity prediction from global AI weather forecasts
Current weather models struggle to accurately predict tropical cyclone (TC) intensity because they lack fine-scale 3D structural details. This research introduces 3DTCR, a generative AI framework that reconstructs high-resolution, three-dimensional TC structures from coarser global forecasts. Trained on physical simulations, it significantly reduces errors in wind speed prediction compared to existing systems like FuXi and ECMWF.