The complexity of Global Navigation Satellite System Reflectometry signals and the diversity of surface environments pose significant challenges to the retrieval of Vegetation Optical Depth (VOD). Traditional retrieval methods mainly rely on physical models and empirical formulas, which often fail to account for the nonlinear characteristics of complex surface conditions. To address this, the paper proposes a VOD retrieval method based on the eXtreme Gradient Boosting (XGBoost) algorithm, which utilizes the advantages of the high temporal resolution and high spatial coverage of Cyclone Global Navigation Satellite System (CYGNSS) data to enhance the adaptability of the retrieval model for complex surface conditions. The method uses CYGNSS feature parameters as inputs and Soil Moisture Active Passive VOD as reference outputs, optimizing model parameters through grid search and evaluating model accuracy using ten-fold cross-validation. Comparative analysis with Random Forest, Light Gradient Boosting Machine, Artificial Neural Networks, and K-Nearest Neighbors models demonstrates that the XGBoost model achieves the best retrieval accuracy, with a root mean square error of 0.002, mean absolute error of 0.030, and a correlation coefficient of 0.963. In addition, the CYGNSS VOD retrieved using XGBoost effectively reflects vegetation growth, with its distribution characteristics consistent with land cover types. The spatial correlation analysis with Enhanced Vegetation Index shows that over 80% of the areas have a correlation greater than 0.600. The proposed method demonstrates high accuracy and applicability, offering an effective approach for dynamic vegetation monitoring.