Vegetation restoration enhances drought resilience through threshold-dependent carbon–water regulation in a semi-arid sandy ecosystem

Drought increasingly constrains ecosystem functioning in semi-arid regions, yet the nonlinear thresholds through which vegetation restoration regulates carbon–water processes and ecosystem resilience remain poorly understood. In this study, a drought–carbon–water–resilience framework was developed to assess how vegetation restoration affects ecosystem drought resilience in the Mu Us Sandy Land, a representative semi-arid sandy ecosystem in northern China. MODIS products, including MOD17A2H/MOD17A3H for gross and net primary productivity (GPP and NPP), MOD16A2 for evapotranspiration (ET), MOD13A2 for the normalized difference vegetation index (NDVI), MOD11A2 for land surface temperature (LST), and MOD44B for vegetation cover, were used for the period 2001–2020 and harmonized to a 1000-m spatial resolution and an annual temporal scale to quantify water use efficiency (WUE), carbon use efficiency (CUE), the temperature–vegetation drought index (TVDI), ecosystem resistance, recovery, resilience, and a restoration intensity index (RII). The results showed that drought intensity increased slightly over time, while WUE and CUE exhibited contrasting spatial patterns and different sensitivities to drought. At the pixel scale, temporal correlations between annual TVDI and WUE/CUE were positive over 83.74% and 69.78% of the study area, respectively, indicating that carbon–water use efficiency often showed compensatory increases under mild-to-moderate drought conditions. In contrast, the regional drought-gradient analysis revealed threshold-dependent responses of WUE and CUE to increasing TVDI, with critical thresholds at approximately TVDI = 0.67 and TVDI = 0.66, respectively; beyond these thresholds, both water and carbon use efficiency declined rapidly. Strong restoration areas showed higher RSwue and RScue than weak/no restoration and degradation areas, and the proportion of high-resilience areas increased with restoration intensity, particularly for CUE-based resilience. Hierarchical regression further showed that RII provided additional explanatory power for ecosystem resilience after controlling for drought intensity, although its effects differed between water-use and carbon-use processes. These findings indicate that vegetation restoration enhances drought resilience through threshold-dependent carbon–water regulation, but its effectiveness depends on vegetation structure and water availability. Threshold-based monitoring and structure-optimized restoration should therefore be prioritized in semi-arid sandy ecosystem management.