Dynamic Assessment and Multi-Scenario Prediction of County-Level SPCR Based on a Spatiotemporal Adaptive Ensemble Model

Against the growing prominence of global climate change and environmental governance challenges, synergistic pollution and carbon reduction (SPCR) has emerged as a core pathway for regional sustainable-development strategies. Focusing on the refined evaluation and dynamic prediction of intra-urban SPCR, this study selects 20 county-level administrative units in Chengdu, western China, and establishes an SPCR analytical framework based on a spatiotemporal adaptive ensemble model. First, the TOPSIS–gray relational analysis (GRA) approach is adopted to systematically measure the spatiotemporal evolution pattern of county-scale SPCR during 2001–2022. The overall SPCR level exhibits gradual improvement following a U-shaped trend. Complex-network theory is further applied to reveal the synergistic characteristics of its “core-periphery” structure. Second, the XGBoost–SHAP model identifies that the increased proportion of the tertiary industry and the control of total energy consumption constitute the most prominent positive and inhibiting driving factors for SPCR, respectively. On this basis, a Spatiotemporal Heterogeneity Ensemble Learning Model (STE-ML) is constructed. By dynamically integrating geographical proximity and socio-economic functional relevance, this model adaptively synthesizes the strengths of base learners, including the geographically and temporally weighted regression (GTWR) model, geographical optimal similarity (GOS) model, and Random Forest (RF), and achieves high-accuracy simulation of county-level SPCR (R2 = 0.806). Finally, four scenarios, namely the Business-as-Usual Scenario (BAUS), Structural Optimization Scenario (SOS), Technology Leadership Scenario (TLS), and Collaborative Governance Scenario (CGS), were established to simulate the evolutionary pathways of SPCR from 2023 to 2035. The results indicate that the BAUS, which continues the existing development pattern, will trigger further divergence in regional synergistic levels. By contrast, the CGS, which incorporates in-depth adjustment of industrial structure, low-carbon technological innovation, and cross-regional collaborative governance, can most effectively elevate the overall SPCR level and facilitate balanced development among county-level jurisdictions. In terms of robustness, a single-factor sensitivity analysis with ±10% variation in scenario parameters and a 95% prediction-interval test were conducted. The results show that the absolute value of the sensitivity index |S| does not exceed 0.0113, and the gap (1.44%) between the CGS and BAUS substantially exceeds the maximum effect of any single parameter. These findings verify that the forecasting conclusions are robust and that scenario discrepancies are not driven by a single hypothetical parameter. This study provides methodological support and empirical evidence for the implementation of differentiated and refined collaborative governance of pollution and carbon reduction within cities.