Prediction of pedestrian exposure to traffic particulate matters (PMs) at urban signalized intersection

Abstract

To evaluate the prediction performance of three models (CAL3QHC – California Line Source Model 3 with Queuing and Hot Spot calculations, BPNN - Back Propagation Neural Network, and WNN - Wavelet Neural Network) on pedestrian exposure to traffic-related particulate matter (PM), the models were applied for evaluating PM2.5 concentrations, and an investigation was performed using mobile measurement devices, at one of the busiest signalized intersections in Xi’an, China. The results revealed that the concentrations of fine particles (PM2.5) at the intersection were highest in the southeast and northeast corners, and that the spatial distribution of PM is related to wind and the layout of buildings around. Additionally, the results indicated that the concentration of PMs during the weekends was lower than that for weekdays, and the PMs concentration in the off-peak period was slightly higher than in the morning peak period. Further detailed analysis showed that the WNN model could make better predictions for varied meteorology and traffic conditions, compared to the other two models. The prediction errors of PM2.5 for the CAL3QHC, BPNN, and WNN are 8.25 %, 5.55 %, and 4.61 %, respectively. The results suggest the WNN model is a useful and fairly accurate tool for predicting PM at signalized intersections.