Spatial Network Structures and Nonlinear Association Characteristics of Synergistic Pollution Reduction and Carbon Mitigation in China: Evidence from the CatBoost–SHAP Model

Synergistic pollution reduction and carbon mitigation (SPRCM) is a critical pathway for advancing global climate governance and promoting green and low-carbon transitions. It also represents an important practice for strengthening ecological civilization and achieving regional sustainable development in China. Based on panel data from 31 Chinese provinces during 2010–2024, this study employs an improved entropy-weighted TOPSIS method to measure SPRCM levels and integrates a modified gravity model, social network analysis, and the CatBoost-SHAP explainable machine learning approach to examine its spatiotemporal evolution, spatial network structure, and nonlinear association characteristics. The main findings are as follows. First, China’s overall SPRCM level shows a gradual upward trend, while significant regional disparities remain. Second, the interprovincial spatial association network has gradually evolved toward a polycentric structure and exhibits clear hierarchical characteristics. Although network connectivity has continuously improved, linkages between core and peripheral provinces remain relatively weak. Third, individual network characteristics display distinct regional differentiation. Eastern coastal provinces consistently occupy core network positions, whereas northeastern and some western provinces remain relatively peripheral. Fourth, the block model analysis identifies four functional groups within the network: net beneficiaries, net spillovers, brokers, and two-way spillovers. These blocks play differentiated roles in spatial association and network interactions. Fifth, the CatBoost-SHAP analysis indicates that road density, freight volume, and human capital are the most important variables in explaining model predictions, with their combined mean absolute SHAP importance accounting for 54.59%. Moreover, these factors exhibit significant nonlinear association patterns with predicted SPRCM levels.