Study area: The study area is the inland river watershed of northwest China, the upper Manas River basin. Study focus: Baseflow ensures water supply and ecological stability, especially in arid regions. In arid regions, hydrological processes differ significantly between dry and wet seasons; natural hydroclimatic factors jointly regulate the dynamic variation of watershed baseflow. To investigate the causes of baseflow changes in the Manas River basin in northwestern China, we used a machine learning model to provide a detailed perspective. We adopted multi-source remote sensing data and quantitatively evaluated the impacts of hydroclimatic variables and vegetation conditions on baseflow during dry and wet seasons through Categorical Boosting (CatBoost) coupled with SHapley Additive exPlanations (SHAP) analysis. New hydrological insights for the region: Results show that Normalized Difference Vegetation Index (NDVI) exhibits the strongest importance. Meanwhile, surface air temperature (Temp) plays an important role. In the dry season, Evapotranspiration (ET) and Vapor Pressure Deficit (VPD) rank higher, and snowmelt serves as an important source of baseflow recharge. NDVI exerts a strong influence through its interactions with most hydroclimatic variables, and this effect is particularly significant in the wet season. In the wet season, NDVI and Precipitation (Prec) show two-way positive interaction with clear thresholds (0.21 and 2.83 mm respectively), and the synergistic effect between vegetation cover and precipitation amplifies their impact on baseflow. During the dry season, ET (0.41 mm) and VPD (0.60 kPa) may be the key feature thresholds for suitable vegetation growth in the upper Manas River basin. The study's results provide important support for exploring baseflow generation mechanisms, optimizing water resource management, and preventing water resource disasters in arid regions.