Resource recovery reshapes adaptation–sustainability trade-offs in urban drainage renewal: From high-throughput exploration to generative AI–enabled decision support

Climate-driven pluvial flooding is accelerating urban drainage renewal, yet each increment of protection locks cities into long-lived material, energy, and maintenance liabilities. To secure robust flood mitigation without unsustainable life-cycle burdens, we examine whether resource-oriented low impact development (Rec-LID), manufactured from recycled construction waste and agricultural residues, can reconfigure the adaptation–sustainability trade-off at scale. We link facility-level life cycle assessment with catchment-scale hydrologic–sustainability simulation and decision-landscape mapping. Facility results show Rec-LIDs can lower life-cycle burdens relative to traditional LIDs, but advantages are conditional, eroding with intensive processing, transport, and variable material performance. At the system level, we generate extensive intervention portfolios and organize them in a global adaptation–cost landscape, from which the Global Minimum Adaptation Cost Trajectory is extracted to reveal stage-specific priorities and tipping behavior as marginal returns diminish and costs escalate. Incorporating Rec-LID shifts feasible solution sets and can reorder preferred strategies across economic–environmental weightings within the evaluated decision framework. Finally, we operationalize the landscape evidence through an optimization-grounded natural-language decision interface, in which large language models are grounded in the study's optimization results to retrieve, compose, and validate recommendations under local constraints. Across representative planning scenarios, the system delivered full feasibility with prediction errors for key indicators below 5 %. Together, these results position resource recovery as a conditional but transformative lever for Sponge City renewal and provide a scalable route to convert trade-off evidence into robust, constraint-consistent decisions.