Coupling multi-level agent system with ant colony optimization for land use allocation in Wuhan, China

To address spatial disputes from human-land conflicts, multi-agent system (MAS) has emerged for simulating stakeholder interactions and developing systemic solutions in land use optimization. However, existing MAS often fail to capture spatiotemporal land-use dynamics, while integrating systematic domain knowledge into agent-based decision-making remains challenging, constraining improvements in accuracy and interpretability. This study addresses these gaps by constructing a spatiotemporal data-driven domain knowledge using machine learning to enhance agent perception. Furthermore, we construct MLAS-ACO, a model coupling a multi-level agent system (MLAS) with ant colony algorithm, to refine agent mapping and interactive decisions through bio-inspired strategies. The Wuhan case study demonstrates: (1) CatBoost-based suitability evaluation achieves high accuracy (F1 score > 0.88), and integration with spatial agglomeration features provides a reliable basis for optimization. (2) MLAS-ACO achieves 6.84% higher overall objective, improves landscape pattern, and yields a layout consistent with development strategies. (3) Compared to MLAS, MLAS-ACO improves convergence speed by 18.7% and the overall objective by 0.35%. (4) Scenario results indicates coordinated scenario enables multi-objective synergy, underscoring construction land agglomerating in highly suitable zones. MLAS-ACO contributes to enhancing interpretability and decision adaptability of spatial optimization, offering insights for refined spatial simulation and supporting planning under policy scenarios.