Abstract: Gust forecasting over the complex terrain of the Qinghai–Tibet Plateau remains a major operational challenge, particularly during the frequent gales of the winter–spring dry season. To improve dryseason gust forecasts and reveal the underlying multiscale physical mechanisms, this study constructs four physically guided feature sets: European Centre for Medium-Range Weather Forecasts (ECMWF) forecast variables, historical persistence information, spatiotemporal attributes, and topographic dynamic factors. Using the Light Gradient Boosting Machine (LightGBM) algorithm, a set of progressive feature fusion?experiments (Exp1–Exp4) is designed to systematically quantify the marginal contribution of each feature category to forecast performance. Results show that the fullfeature Exp4 model achieves the best performance, with a mean absolute error (MAE) that is 52.5% lower than that of the raw ECMWF forecast. Feature importance analysis identifies longitude, altitude, and latitude as the most influential factors, with topographic dynamic factors (altitude, slope, aspect variability, etc.) contributing a cumulative 13.42%. SHapley Additive exPlanations (SHAP) is used to interpret the model. The SHAP-based analysis further reveals a threshold effect at 3000 m altitude. To address the systematic underestimation caused by the scarcity of stronggust samples (≈10% of the training set), a graded weighting strategy is introduced. For gust events of level?8 (17.2–20.7?m·s?1) and level?9 (≥20.8?m·s?1), the weighted Exp4 model increases the Threat score (TS) by 79.4% and 148.2%, and reduces the Miss rate by 23.1% and 10.5%, respectively, compared with the unweighted model. Relative to the raw ECMWF forecast, the TS improvements are 96% and 60%, respectively. Error analysis demonstrates that the proposed model effectively corrects the systematic overestimation bias of ECMWF over the highaltitude complex terrain and significantly enhances the capture of strong gusts in the afternoon. This study provides an accurate and interpretable technical solution for operational hazardous gust warning over the Tibetan Plateau.