Tropical Cyclone Multi‐Level Wind‐Speed Structure Reconstruction From Sparse Dropsonde Data Via Adversarial Learning

Tropical cyclones (TCs) pose significant global hazards due to their intense winds, heavy rainfall, and associated storm surges. While in situ observations from aircraft are crucial for understanding TC structures, these measurements are often spatially sparse, limiting the characterization of the wind speed structure. In this study, we introduce a deep learning (DL) framework based on generative adversarial networks to reconstruct multi‐level wind‐speed (including that at the 10 m surface layer) from sparse dropsonde inputs within seconds. Our simulation experiments incorporate realistic flight parameters and account for the horizontal drift of dropsondes. We present a case study that reconstructs TC structures using a combination of observations from Hurricane Hunter missions. This approach demonstrates the potential of DL for reconstructing the multi‐level wind‐speed structure of TCs from sparse data. The current framework focuses on reconstructing wind speed, showing significant promise for surface wind estimation and operational storm surge modeling.