Unraveling the nonlinear relationships between blue-green spatial patterns and carbon sequestration through an interpretable XGBoost-SHAP model

Blue-green spaces (BGS) play a critical role in contributing to both direct carbon sequestration (CS) and indirect carbon reduction. However, the relationship between blue-green spatial patterns (BGSPs) and CS remains insufficiently explored. This study developed an analytical framework to investigate the nonlinear relationship between BGSP and CS. In this framework, morphological spatial pattern analysis (MSPA) and landscape pattern indices were used to characterize BGSP; the Carnegie-Ames-Stanford approach (CASA) model estimated net primary productivity (NPP) as a proxy for CS; and the eXtreme gradient boosting-SHapley additive exPlanations (XGBoost-SHAP) model was employed to reveal the underlying nonlinear relationships between BGSP metrics and NPP. Hebei Province, China, was selected as a representative case study. The results indicate that BGSPs are significantly associated with CS, with the strength of the association varying across different features. Core and percentage of landscape (PLAND) are the dominant predictors of CS, both showing positive associations with NPP. Large and continuous blue-green patches are essential for maintaining high CS, whereas small, fragmented, and isolated patches contribute relatively little. Boundary length and shape complexity are positively associated with CS above specific thresholds. Moderate fragmentation may be weakly beneficial to CS. Significantly higher NPP is associated with PLAND above 80, core above 60, LSI above 2, ED above 30, edge above 20, AI below 97.5, and SPLIT above 20. These findings provide a scientific basis for optimizing regional BGSP and inform spatial planning and ecological management strategies aimed at enhancing CS. Furthermore, the proposed framework offers a transferable methodological reference for investigating nonlinear relationships between BGSP and CS in other regions.