Vegetation plays a critical role in modulating the water cycle via transpiration. Therefore, understanding how vegetation interacts with terrestrial water storage (TWS) across space, time, and soil depth is essential for developing sustainable water-resource strategies in arid regions. A major challenge is that vegetation-water relationships in drylands are highly heterogeneous, nonlinear, and often lagged, making it difficult to identify when greening alleviates or intensifies water stress. Nevertheless, the spatiotemporally heterogeneous and coupled relationships between vegetation and TWS in these areas remain poorly understood. Here we elucidate the coupling mechanisms and threshold behaviors connecting water resources and vegetation dynamics in arid Northwest China (NWC). We find that, during the historical period (1981-2023), regions with elevations above 2000 m exhibited higher terrestrial water storage anomalies (TWSA), coinciding with widespread vegetation greening. In contrast, lowland basins such as the Tarim and Qaidam exhibited structural aridity (near-zero total soil moisture (TSM) and minimal vegetation cover). Notably, we identify nonlinear responses of vegetation to TSM at different depths: (1) initial vegetation greening leads to rapid depletion of shallow soil moisture (SM), and (2) sustained greening beyond critical thresholds (NDVI approximate to 0.133, LAI approximate to 4.667) depends on lagged recharge of root-zone SM from precipitation and snowmelt. Under warming scenarios of 1.5-4 degrees C (2024-2099), vegetation trends are projected to reverse with a transition boundary near 37 degrees N. Specifically, under a 3 degrees C warming scenario, northern NWC shifts from greening to degradation, while southern NWC transitions from drying to recovery. This spatial reversal is closely linked to altered precipitation regimes: northern Xinjiang and the Loess Plateau face heightened aridity and vegetation loss, whereas southern Xinjiang and the fringes of the Tibetan Plateau benefit from increased moisture availability and vegetation regeneration. By explicitly quantifying depth-dependent thresholds, lagged recharge effects, and future spatial reversals in vegetation response, this study provides a process-based basis for defining the water limits of ecological restoration in arid landscapes.