Machine Learning‐Based Turbulence Intensity Estimation Near Convective Clouds in East Asia Using GK‐2A Satellite Observations

Abstract The detection of atmospheric turbulence, particularly near deep convection, are essential for flight safety and efficiency. Geostationary satellites provide continuous coverage where radar is unavailable, but their lack of vertical information limits detection to upper levels. This study addresses these limitations by leveraging multi‐channel satellite data with machine learning. A U‐Net model is developed to estimate warm‐season turbulence intensity at four vertical layers using data from GEO‐KOMPSAT‐2A. The model is trained on turbulence diagnostics derived from the outputs of the Local Data Assimilation and Prediction System. Evaluation against in situ turbulence measurements shows good performance at the lower three layers, with slightly reduced accuracy at the tropopause and above. Case studies further demonstrate reliable detection of moderate‐or‐greater turbulence in or near convective regions. The results highlight the potential of geostationary satellites as real‐time observational tools for turbulence monitoring in East Asia, offering an important complement to existing ground‐ and model‐based systems.