Blue–green water scarcity in irrigation districts is influenced by both hydrological processes and farmer management, yet most studies treat agricultural decision-making as exogenous and static. We develop a spatially explicit, bidirectionally coupled framework integrating Soil and Water Assessment Tool (SWAT) with an agent-based model (ABM) of boundedly rational farmers, embedding crop choice and irrigation scheduling within basin-scale hydrology and crop growth. Applied to the Yaohekou Irrigation District in the Han River Basin, China, the model quantifies how behavioral heterogeneity and management portfolios affect blue–green water use, irrigation supply adequacy, ecological-flow pressure, and equitable water access in the water–ecology–food–society (WEFS) nexus. The district shows persistent supply–demand gaps and strong sensitivity to behavior. Profit-driven regimes concentrate cropping and synchronize irrigation during critical stages, increasing dry-year peak blue-water withdrawals, shrinking safety margins, and amplifying drought-time deficits and inequity. Conservative social-learning regimes maintain crop diversity and stagger demand, buffering drought impacts. Policy experiments show that supply augmentation alone is partly absorbed by demand catch-up (diminishing returns); uniform water-price increases raise efficiency but reduce equity via heterogeneous responses; combining efficiency improvements with moderate supply support lowers water use per unit output and dampens sensitivity during wet-to-dry transitions. Overall, sustainable management should target the chain of demand synchronization, peak extraction, and constraint triggering, supported by technology diffusion and protective mechanisms to build socially inclusive resilience.