Analyzing spatiotemporal truck emission pattern using explainable machine learning: A case study in Xi’an, China

Abstract

This study investigates the spatial distribution pattern of CO₂ emissions from Heavy duty trucks (HDTs) in Xi’an City across different periods and their influencing factors. Five study periods were initially proposed considering the temporal distribution of emissions and the city’s traffic restriction policy. The study area was divided into 2,839 spatial grids. Then, CO₂ emissions from HDTs were calculated at the spatial grid scale, and variables related to road density, accessibility to freight hubs, POI density, and demographic indicators were also integrated into the spatial grids. Finally, five XGBoost models were constructed using the spatial data from different periods, and the spatial and temporal heterogeneity of each variable’s impact on CO₂ emissions was thoroughly analyzed using the SHAP (SHapley Additive exPlanations) explainer. The results demonstrate that the predictive ability of the XGBoost model surpasses that of the OLS model and the GWR model, providing better insight into the spatiotemporal heterogeneity.