Wind Speed Retrieval for CDWL via Multi-Range Gate with Sup-Adap Frequency Estimation and CNN-LSBoost Model

Multi-range-gate wind retrieval with coherent Doppler wind LiDAR (CDWL) remains challenging under a low signal-to-noise ratio (SNR) and precipitation: low SNR, spectral-edge peaks and system gain drift destabilize the radial frequency estimation, and multi-beam synthesis further amplifies the errors, directly affecting the reliability of vertical wind profiles. This study develops a multi-range-gate wind retrieval framework that couples physical constraints with data-driven regression. After spectral accumulation and quality control (QC), an adaptive super-noise centroid (sup-adap) estimator extracts radial velocities, and a CNN-LSBoost model maps the multi-beam radial velocities of adjacent range gates to the horizontal wind. The framework was evaluated with prescribed winds in simulations and validated against independent meteorological-tower observations from a field experiment under moderate rain, with training and test hours strictly separated. Against the tower, CNN-LSBoost achieved R² = 0.983, RMSE = 0.42 m/s and P95 = 0.53 m/s, outperforming KNN+GA, RF+GWO, CNN and Transformer baselines. The results indicate that the proposed framework provides a validated route to reliable wind profiling under low-SNR and precipitation conditions.