Novel spatiotemporal nonlinear regression approach for unveiling the impact of urban spatial morphology on carbon emissions

Abstract

Understanding the impacts of urban spatial morphology on carbon emissions is crucial for promoting sustainable urban development. However, traditional models have limitations when analyzing complex spatiotemporal heterogeneity and nonlinear relationships. Therefore, we proposed an Integrated Spatiotemporal Nonlinear Regression (ISTNR) model to explore the complex relationship between urban spatial morphology and carbon emissions. This model combines Geographically and Temporally Weighted Regression (GTWR) to capture spatial and temporal dependencies, the Random Forest (RF) model to address nonlinear relationships, and the game theory-based Shapley Additive Explanations (SHAP) tool to enhance the interpretability of the results. The data encompassed urban morphology and carbon emissions across specific regions and periods, and the robustness and adaptability of the model were validated in various urban morphology environments. The ISTNR model demonstrated significant superiority over traditional regression models, achieving an R² of 0.924, a substantially lower MSE (18.06×10⁶), and higher predictive accuracy and stability in complex urban environments. Additionally, bootstrap uncertainty analysis indicated that the model's prediction intervals were relatively narrow, suggesting low prediction uncertainty and high stability. The SHAP analysis quantified the specific contributions of various urban morphological features to carbon emissions, further revealing their mechanisms impacting emission predictions. This study presents an effective quantitative tool for urban planning and carbon emissions control, offering practical support for future urban policymaking.