Identifying optimal green plot ratio thresholds for mitigating PM2.5 in street canyons: Integrating LiDAR, CFD, and explainable machine learning.

Vegetation significantly influences PM2.5 dispersion in urban street canyons, but the interplay between aerodynamic and deposition mechanisms remains poorly quantified at this scale. This study comprehensively utilized LiDAR, YOLOv8, scenario analysis, and Computational Fluid Dynamics (CFD) to derive realistic 3D geometry, to identify precise traffic flow for estimating PM2.5 emissions, and to simulate PM2.5 distribution, respectively, in a real street canyon of Xi'an. The model was validated against field measurements, achieving strong agreement (R² = 0.91, RMSE = 1.87 µg/m³). Results show that dense or obstructive vegetation mainly increases near-surface concentrations by creating low-wind zones that hinder ventilation. At the same time, deposition helps remove particles, with low shrubs proving more effective at 0.3 m than trees at higher levels (4.5 m and 7.5 m). Among the tested configurations, scenarios with low hedges along both sides of the street canyon, with or without central trees, performed best at reducing pedestrian-level PM2.5, as they maintained good airflow while enhancing near-surface deposition, with Averaged Relative Difference in Concentration (ARDC) values of -2.49% and -0.34%. Further analysis using Shapley Additive Explanations (SHAP) identified Green Plot Ratio (GPR) as the most influential metric and revealed an actionable, model-derived threshold range: GPR values between 0.54 and 0.56 are associated with effective concentration reductions. These findings demonstrate that net PM2.5 mitigation in street canyons requires balancing aerodynamic suppression against deposition enhancement through careful vegetation configuration, providing a quantitative basis for mechanism-informed urban greening design.