Vegetation stability in global arid-semiarid regions is critical for planetary ecological balance and climate change mitigation. Northwest China (NWC), a typical arid-semiarid region and core of global mid-latitude drylands, is pivotal to global dryland ecosystem research-its vegetation resilience dynamics shape regional ecological security and inform vegetation stability enhancement in analogous regions worldwide. However, spatiotemporal patterns of NWC's vegetation resilience, regionally heterogeneous driving mechanisms, and quantifiable restoration thresholds (especially nonlinear multifactor synergies) remain unresolved. Compared with existing dryland resilience studies that focus on qualitative driver descriptions or single-scale analysis, this study innovatively integrates the XGBoost-SHAP framework with AC1 (first-order lag autocorrelation coefficient) to quantitatively disentangle multifactor coupling mechanisms, identify subregional heterogeneous rules, and derive actionable thresholds, forming a "mechanism-quantification-solution" framework. We analyzed vegetation resilience in NWC during 2000-2022 using the XGBoost-SHAP framework, with resilience quantified using the AC1, where higher AC1 values indicate lower resilience. Results showed: (1) resilience exhibited a significant "higher resilience in the east and lower in the west" pattern, with high-resilience areas concentrated in the Qinling-Daba Mixed Forest Ecosystem (QDMF) and Northwest China Forest-Steppe Transitional Ecosystem (NWFT), and low-resilience belts in the Northwest China Desert-Oasis Ecosystem (NWDO) and Northwest China Steppe-Loess Ecosystem (NWSL); (2) regional resilience overall enhanced (P < 0.01) with a 2017 turning point: NWDO reversed to rapid improvement, NWSL's improvement slowed, NWFT continuously weakened, and QDMF shifted from improvement to degradation; (3) driving mechanisms featured "climate dominance with regional heterogeneity" and refined regulation by multifactor synergistic interactions: mean annual precipitation (MAP) and mean annual temperature (MAT) served as dual dominants at the regional scale, with optimal resilience achieved when MAP > 800 mm and MAT < 0 degrees C (quantitative restoration criterion); subregionally, NWDO was temperature-dominated, NWSL by hydrothermal coupling, NWFT by precipitation, and QDMF by hydrothermal-radiation-nutrient coupling, while synergistic effects further modulated resilience dynamics. This study provides actionable technical parameters (critical thresholds, sub-regional priorities) for differentiated ecological restoration, offering a scalable model for global analogous arid-semiarid regions and advancing UN SDG 15.3.