CFD-quality nowcasting for urban air mobility with a deep learning-based emulator

Abstract Urban air mobility (UAM) aircraft typically operate at altitudes of 1000–2000 feet over complex urban terrains, where airflow interactions with structures create dynamic conditions requiring precise, real-time meteorological forecasts. Traditional large-scale forecasting systems often struggle to adequately capture turbulent boundary-layer characteristics due to coarse grid resolutions, while high-resolution computational fluid dynamics (CFD) simulations remain too computationally expensive for real-time operations. To overcome these challenges, this study introduces a deep learning-based emulator that rapidly generates CFD-like results using data from the local data assimilation and prediction system operated by the Korean Meteorological Administration. By integrating residual dense blocks with a wind direction classification system, the emulator significantly enhances predictive accuracy and computational efficiency. This approach enables urban-scale high-resolution weather forecasts, which are critical for UAM operations and broader urban meteorological applications, establishing a new standard for the safe and effective integration of UAM.