[Objective] This study aims to quantitatively assess the contributions of individual meteorological factors to hydrological drought evolution in the Yuanjiang River Basin, a typical humid region in the middle reaches of the Yangtze River, and to reveal the climatic mechanisms. [Methods] We integrated historical hydrometeorological observations (1975-2022) with future climate projections (2026-2050) derived from five CMIP6 global climate models (GCMs) under three Shared Socioeconomic Pathways (SSP126, SSP245, and SSP370). To ensure regional accuracy, the future meteorological projections were bias-corrected using the quantile delta mapping method. Subsequently, we developed and validated a monthly streamflow prediction model across six hydrological stations utilizing the XGBoost algorithm. This model incorporated a comprehensive set of predictors, including meteorological variables (precipitation, relative humidity, radiation, and temperature) and human activity indicators (reservoir operation and land use changes). To ensure the interpretability of the machine learning model, the SHAP (SHapley Additive exPlanations) framework was employed to quantify the marginal contributions and unravel the directional influences of each driving factor on streamflow variations. Finally, the validated XGBoost model was driven by the bias-corrected future data to project long-term streamflow dynamics. Based on these projections, hydrological drought events were characterized using the standardized runoff index (SRI) at 3- and 6-month scales, with key drought features (frequency, duration, severity, and intensity) systematically extracted via the run theory. [Results] From 1975 to 2022, the XGBoost model demonstrated excellent performance, achieving average Nash-Sutcliffe efficiency (NSE) values of approximately 0.90 for calibration and 0.80 for validation. SHAP analysis revealed precipitation as the dominant driver of streamflow variability, accounting for 69.8% of the total explained variance, followed by relative humidity (11.9%). Furthermore, analysis of the daily-to-monthly aggregation methods showed that the monthly mean contributed 58.2% of the variance, whereas metrics capturing intra-monthly fluctuations and extreme distributions (std, p90, and p10) cumulatively contributed 41.8%. Specifically, standard deviation (20.8%) and the 90th percentile (14.4%) had high explanatory weights, highlighting the critical importance of intra-monthly meteorological variability. Historically, hydrological droughts exhibited distinct spatiotemporal heterogeneity, with more frequent occurrences in winter and spring and greater frequency and severity at upstream stations (e.g., Taoyi). Under future climate scenarios (2026-2050), precipitation was projected to follow a “drier dry season and wetter wet season” pattern, with reductions of approximately 25 mm from September to November and increases of up to 50 mm from April to August. Concurrently, relative humidity was projected to decline throughout the year, with pronounced decreases of 0.05-0.10 during the dry season. Consequently, hydrological droughts were expected to worsen, with projected increases in frequency (0.04-0.06), duration (0.2-0.4 months), and intensity (0.05). A significant structural shift in drought categories was also anticipated: moderate drought frequency increased markedly (0.04-0.08), while severe drought frequency generally declined, dropping by approximately 0.06 at Taoyi Station. This structural transition from “long-duration, high-intensity” to “high-frequency, moderate-intensity” droughts in the future was driven by seasonal moisture dynamics. Sharp declines in precipitation and relative humidity in June acted as the trigger for drought onset (with the combined SHAP negative contribution reaching approximately -230 m³/s), followed by continuous moisture deficits from July to October that drove the progression into moderate drought. However, steady moisture increases from November to May created a significant compensation effect, interrupting the deep accumulation of streamflow deficits and preventing the evolution into severe droughts. These findings highlighted that future drought mitigation in the Yuanjiang River Basin should prioritize managing seasonal consecutive droughts and the cumulative impacts of moderate droughts. [Conclusion] Precipitation and relative humidity are the dominant meteorological factors controlling hydrological drought in the Yuanjiang Basin. The XGBoost-SHAP framework effectively quantifies their individual contributions and reveals that the seasonal coupling between precipitation reduction and enhanced evapotranspiration due to declining relative humidity is the primary mechanism driving future drought intensification. Future hydrological drought across the basin is projected to intensify, characterized by increased frequency, prolonged duration, and a structural shift toward more frequent moderate drought events. Machine-learning interpretability approach provides a valuable and robust supplement to traditional physical models for understanding and projecting climate change impacts on hydrological drought. This integrated analytical framework offers scientific support for adaptive water resource management and drought mitigation strategies in the Yuanjiang River Basin and similar humid regions facing escalating drought risks under climate change.