Low-level wind shear classification for airborne weather radar based on signal simulation and residual network

Low-level wind shear (LLWS) is a sudden and hazardous meteorological phenomenon that seriously threatens aircraft safety. Conventional detection methods rely on parameter calibration for identification but lack detailed classification capability. Additionally, the dynamic complexity of atmospheric wind fields and the substantial volume of raw radar data impose high acquisition risks and storage constraints onboard. To address these challenges, this study proposes a simulation-based approach for classifying airborne weather radar-identified LLWS during takeoff and landing. The method first establishes four typical LLWS field models—microburst, crosswind, low-level jet, and sea–land breeze—based on fluid dynamics principles. It then simulates the aircraft takeoff trajectory using a standard race-track pattern and generates airborne radar echoes via MATLAB integrated with a wind field retrieval algorithm to produce radial velocity images. A comprehensive dataset of wind shear images under varied conditions is constructed. Finally, a residual network integrated with transfer learning is developed, which achieves a classification accuracy of 95.50% across six categories and outperforms both Inception V1 and SqueezeNet in experimental trials, demonstrating a promising solution for LLWS classification using airborne radar.