Integrating Urban Air Mobility (UAM) into urban transportation systems requires operational planning that addresses safety and efficiency. In urban areas, local wind speed and turbulence intensity are strongly influenced by terrain features such as building geometry, elevation, rivers, and green spaces, directly affecting vertiport placement and flight path planning. In this study, we employ the massively parallel multi-GPU CFD solver (MPM-STD) to generate high-resolution urban wind datasets under diverse meteorological conditions and train a deep learning surrogate model (POD-Transformer) on these datasets. The trained surrogate model is applied to the Yeouido district of Seoul, combining simulation with meteorological observations to analyze multiple wind-related indicators for UAM operational safety. This study presents model validation results and example applications, and discusses the potential for incorporating local terrain and meteorological effects into future UAM operational planning.
 

Proper Orthogonal Decomposition,Transformer Network,Large Eddy Simulation,Urban Microclimate,UAM Operation Safety Assessment