Geographical-XGBoost and SHAP reveal ecosystem service supply-demand responses to landscape patterns in karst regions

Human activities influence landscape pattern (LP) change and, in turn, regulate the supply and demand of ecosystem services (ESs), which is particularly important in ecologically fragile karst regions. Focusing on Guanling County, we applied the G-XGBoost model integrated with the SHAP interpretation framework to quantify the spatiotemporal distribution patterns and supply-demand imbalances of landscape patterns and ESs in the region from 2000 to 2020. This approach identified key influencing factors, their contributions, and interactions, which also revealed spatial heterogeneity in the intensity of these impacts. Results show that (1) landscape fragmentation generally slowed, landscape heterogeneity decreased while aggregation increased, and construction land expansion was associated with more complex patch shapes. (2) All five ESs exhibit persistent spatial mismatches, with persistent habitat quality deficits, sustained carbon sequestration surplus, a tendency toward balance for soil conservation in some units, and urbanization-related shifts for grain production and annual water yield. (3) Landscape pattern indices contributed more than 50% to ES supply, with the largest patch index and landscape shape index playing dominant roles, whereas ES demand was mainly explained by socio-economic factors, except for soil conservation demand. (4) The G-XGBoost model yields detailed spatial evidence to inform zoning-based management. Accordingly, we propose targeted planning strategies: in supply-sensitive areas, priority should be given to conserving large natural patches and restricting infrastructure-driven fragmentation; while in high-demand zones, key interventions include controlling construction footprints, enhancing green-blue infrastructure, and exploring cross-jurisdictional ecological compensation.