Characteristics and prediction of traffic-related PMs and CO2 at the urban neighborhood scale

Accurate portrayal of the spatial-temporal distribution of traffic-related pollutants at the neighborhood scale contributes to refined management for the urban air environment. This study conducted mobile monitoring of traffic-related particulate matters (TRPMs, including PM10, PM2.5, and BC) and CO2 on urban roads in a neighborhood-scale area of Xi'an. Random forest land-use regression (RF-LUR) models for TRPMs and CO2 were developed to quantify the contributions of relevant influencing factors. Finally, the accuracy of the LUR and RF-LUR prediction models was evaluated and compared. The results showed that the pollutants in the neighborhood-scale regions had significant spatial-temporal heterogeneity. The concentrations of TRPMs and CO2 were significantly higher in the morning peak than in the evening peak. BC and CO2 hotspots mostly appeared on arterial roads, while PM10 and PM2.5 exhibited considerably greater spatial variability and more widely dispersed hotspots. The prediction models demonstrated that the RF-LUR model outperforms the LUR model in all evaluation indicators. Specifically, the RF-LUR model displays a higher degree of explanatory power, as evidenced by R2 values of 0.784, 0.871, 0.533, and 0.551 for PM10, PM2.5, BC, and CO2, respectively. Further analysis of the contributions of various factors showed that the concentrations of PM10 and PM2.5, which reflect the background concentrations, were the main influencing factors, with explanatory powers of 57.72% and 78.27%, respectively. Additionally, meteorological factors explained TRPMs and CO2 to a greater extent than did different land use types. Our results showed that RF-LUR can be effectively applied to the prediction of neigh-borhood scale traffic pollutants.