Near-Real Time Weather Forecasting for Urban Air Mobility Using Deep Learning Emulators

Urban Air Mobility (UAM) operates at altitudes of 1,000–2,000 feet within urban environments, where dynamic airflows require precise and real-time meteorological forecasts. Traditional large-scale forecasting systems often fail to capture intra-urban phenomena such as turbulence, while high-resolution Computational Fluid Dynamics (CFD) simulations are prohibitively computationally expensive for real-time applications. To address 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 (LDAPS). The proposed model integrates Residual Dense Blocks (RDBs) with a wind direction classification system, significantly enhancing both predictive accuracy and computational efficiency. RDBs enable the effective learning of complex data patterns, while the wind direction classification system accurately predicts real-time wind direction changes, crucial for safe route planning and flight management in UAM operations. Experimental results demonstrate that the emulator reduces Mean Square Error (RMSE) compared to traditional forecasting models and achieves high accuracy in wind direction classification. Additionally, the emulator exhibits markedly lower computational costs and faster processing times than conventional CFD simulations, confirming its suitability for real-time applications. This deep learning-based emulator facilitates high-resolution urban-scale weather forecasting essential for the safe and efficient integration of UAM into urban airspaces. Furthermore, the approach holds significant potential for broader applications in urban meteorology and real-time computational tasks, establishing a new standard for meteorological forecasting tools in complex urban environments

Key Words: Urban Air Mobility (UAM), Urban meteorology, Deep Learning, Emulator, Computational Fluid Dynamics (CFD).