Particle number emissions on mountainous roads: machine learning insights from on-road testing

Mountainous roads pose unique challenges for controlling vehicular fine particulate number (PN) emissions, a critical pollutant impacting air quality and public health. This study integrates on-road testing with interpretable machine learning to analyze PN emission characteristics of light-duty gasoline and diesel vehicles in China's Qinling Mountains, assessing terrain, driving, and environmental influences. On-road testing results indicate that steep gradients reduce vehicle speeds by 10-50 % and elevate vehicle-specific power (VSP), increasing PN emissions by up to 132 % during uphill driving. High altitudes (>1.6 km) exacerbate PN emissions due to reduced air density, with diesel vehicles showing greater sensitivity. Ensemble learning models (R² > 0. 91) and SHAP analysis uncover nonlinear terrain-driver interactions, identifying a 25-70 km/h speed range for minimizing PN, a synergistic altitude-gradient effect that elevated emissions by 1.2-3.6 times. Terrain-integrated prediction models and speed optimization strategies are proposed to mitigate PN emissions, providing a scientific basis for managing traffic emissions in mountainous regions.