Revealing the built environment impacts of idling emissions from electric ride-hailing vehicles through explainable machine learning

Understanding how built environment characteristics influence electric vehicles (EV) idling emissions is crucial for designing low-carbon transportation systems, yet this stationary emission source remains largely overlooked. This study fills this gap by integrating high-resolution GPS trajectory data, grid carbon intensity, and explainable machine learning to quantify the nonlinear impacts of the built environment on EV idling emissions. We developed a LightGBM-SHAP framework that achieves robust predictive performance (R2 = 0.889) and systematically identified nonlinear thresholds. Three key findings emerged. First, across the seven observation days, lower daily mean temperatures were generally associated with higher estimated daily CO2 emissions under the adopted temperature-dependent emission-accounting framework. Second, road network length, residential land, transportation facilities, and population density collectively accounted for nearly 80% of total SHAP importance, with multiple built environment features exhibiting nonlinear threshold effects. Third, feature significance exhibited spatiotemporal heterogeneity, with identical built environment configurations exerting markedly different impacts across varying time periods and spatial scales. These findings should be interpreted as predictive associations rather than causal planning standards, offering diagnostic evidence for future policy evaluation.