Study region: Loess Plateau (LP), China Study focus: Understanding the impact of elevated CO2 (CO2 ) concentrations on water-carbon interactions is critical for informing climate mitigation strategies in semi-arid regions susceptible to hydrological stress. Although CO2 fertilization effects on terrestrial processes are well documented, existing models often oversimplify land cover heterogeneity and neglect region-specific feedbacks, limiting their applicability to complex landscapes like the LP. This study addresses these limitations by employing the Community Land Model (CLM5), which represents diverse land types, including forests, grasslands, and croplands (such as corn and wheat), and is validated against flux tower and remote sensing data. New hydrological insights for the region: This work reveals that eCO2 nearly doubles gross primary productivity by 2.78 gC/m(2)/yr, while suppressing evapotranspiration by 1.42 mm/yr and increasing baseflow by 1.09 mm/yr. Despite vegetation carbon gains (similar to 19.16 gC/m(2)/yr), soil organic carbon exhibited a continued decline at 6.01 gC/m(2)/yr, suggesting slower soil carbon recovery. Water use efficiency (WUE) improved disproportionately, with forests exhibiting a 50 % increase due to deep-rooted systems and stomatal optimization, and croplands showing only a 10 % improvement, likely constrained by irrigation practices. Spatial heterogeneity emerged as a dominant control, with the marginal gain in WUE per unit increase in CO2 (WUE/CO2 ratio) declined by 14 % over two decades, highlighting water constraints in semi-arid ecosystems. By integrating land cover-specific responses, this study advances the homogenized CO2 representations in land surface models and offers actionable insights for targeting afforestation in ecological restoration zones. These findings highlight the necessity of resolving spatial heterogeneity in Earth system models to enhance climate adaptation strategies and optimize water-carbon synergies under escalating CO2 scenarios.