Unveiling the non-linear impact mechanism of multi-factor synergy on habitat quality based on stacking ensemble learning

Rapid urbanization imposes complex nonlinear constraints on ecological sustainability, while conventional linear models often fail to capture critical turning points in interactions among variables. To address this limitation, we developed an analytical framework integrating the InVEST model, stacking ensemble machine learning (EML), SHapley Additive exPlanations (SHAP), and partial least squares structural equation modeling (PLS-SEM) to examine the spatiotemporal drivers of habitat quality (HQ). Using Jiangxi Province, China, as a case study from 2005 to 2023, we quantified the trade-offs between human activities and ecological integrity. The results revealed a distinct core–periphery degradation pattern, with significant declines in HQ concentrated primarily in urban agglomerations. At the county scale and for the dataset analyzed, the stacking ensemble model achieved strong predictive performance, with a test-set R 2 of 0.995, and outperformed the individual base learners in capturing complex multidimensional interactions. SHAP dependence analysis identified model-conditional ecological response turning points of approximately 0.18 for the construction land proportion and 0.51 for the landscape division index. Beyond these values, the modeled negative responses of HQ became more pronounced. Topographic conditions also exerted an important constraint, with a slope turning point of 9.59° marking a shift in the modeled marginal effect from negative to positive. PLS-SEM further showed that the negative statistical association between anthropogenic pressure and HQ dynamics strengthened over the study period. By 2023, this association exceeded that of natural factors, making anthropogenic pressure the most strongly associated component in the specified path model. These findings provide model-based quantitative references for delineating urban growth boundaries and evidence-based guidance for differentiated spatial management, thereby supporting a balance between urban densification and the conservation of critical ecosystem services.