Vegetation in alpine arid regions is highly sensitive to climate change and vulnerable to disturbances resulting from the complex interplay of environmental and anthropogenic factors. These interactions give rise to highly heterogeneous and dynamic vegetation responses across spatial and temporal dimensions. Understanding the spatiotemporal evolution of vegetation and its underlying driving mechanisms is essential for evaluating the effectiveness of ecological restoration programs and for informing the development of region-specific, evidence-based ecological protection and management policies. However, due to the nonlinear, synergistic, and often threshold-based nature of vegetation responses to diverse environmental variables, traditional statistical models may be inadequate in fully capturing these complexities and interactions. To overcome these limitations, this study integrates advanced machine learning and spatial analysis methods. We applied a Random Forest (RF) model, which was particularly effective in modeling nonlinear relationships and capturing complex variable interactions, in conjunction with SHapley Additive exPlanations (SHAP), an interpretation framework that enhanced model transparency by quantifying the marginal contribution of each predictor to the model’s output. Furthermore, the Geodetector method was employed to identify and quantify the spatially explicit effects and interactions among different environmental and anthropogenic driving factors. This comprehensive multi-method approach allows for both predictive accuracy and interpretability in the analysis of vegetation dynamics. Based on multi-source remote sensing data from 2000 to 2022—including the Normalized Difference Vegetation Index (NDVI), temperature, precipitation, and the Standardized Precipitation Evapotranspiration Index (SPEI)—we investigated the spatiotemporal variation of vegetation cover in the Qaidam Basin, a representative alpine arid region located on the northeastern edge of the Qinghai–Tibet Plateau. The results demonstrated that: (1) Over the 23-year period, NDVI in the Qaidam Basin exhibited a significant increasing trend, with an average annual growth rate of 0.0014 /a. Spatially, NDVI followed a southeast-to-northwest decreasing gradient, with 93.56% of the region showing a greening trend. These changes were especially prominent in the southwestern and northeastern subregions, indicating a general and sustained improvement in regional vegetation conditions; (2) Climate change was found to be the predominant driver of NDVI variation, influencing 88.15% of the area. In comparison, the influence of human activities was limited to 2.66%, mainly in urban and peri-urban zones. On average, climate factors contributed 89.18% to NDVI variation, while anthropogenic factors contributed 10.82%; (3) Among all variables, precipitation emerged as the strongest positive driver, followed by SPEI, whereas temperature and elevation had net negative impacts on NDVI performance. Additionally, the response of NDVI to these factors exhibited clear nonlinear threshold effects. NDVI responded negatively to precipitation levels below 191 mm, but showed marked increases beyond this threshold. Similarly, temperature exerted a suppressive effect once it exceeded 2.3?°C, and the positive influence of elevation gradually weakened and even turned negative above 3600?m. SPEI values exceeding 0.6 were associated with significantly enhanced NDVI, reflecting improved vegetation resilience under favorable moisture conditions. By combining interpretable machine learning techniques with spatially explicit factor detection tools, this study provides novel insights into the mechanisms governing vegetation dynamics in high-altitude arid environments. The findings offer critical scientific support for formulating adaptive and locally optimized ecological management strategies, particularly in the context of ongoing global climate change and environmental pressures.