A data-driven investigation on PM2.5 dispersion and the relevant professional driver exposure risks induced by compound turbulence in street canyons

Abstract

Traditional experimental methods for collecting and analyzing particulate matter (PM) concentration distribution data within street canyons face significant limitations, including the passivity of experimental conditions, coarse spatial resolution, and excessive time requirements. To overcome these issues, the data-driven approach that enables rapid flow field prediction based on finite simulation data from computational fluid dynamics has been developed. Firstly, PM_2.5 dispersion flow fields were simulated for 75 scenarios in a typical street canyon, Qingnian Road, Wuhan, China. Proper Orthogonal Decomposition (POD)  was then applied to extract mode coefficients from these data, followed by Least Square Support Vector Machine (LSSVM) used to establish mappings between mode coefficients and operating parameters, enabling rapid flow field prediction. The movement of the centroid of the PM_2.5 concentration field β was defined to assess the spatial variation of PM_2.5, while a normalized parameter γ was employed to evaluate the dispersion intensity of PM_2.5. Concurrently, an exposure risk assessment model was introduced to evaluate the traffic-related PM_2.5 exposure risks for professional drivers who are chronically exposed within their working environments. The results indicated that vehicle-induced turbulence is the primary factor affecting dispersion of PM_2.5, and there is a critical vehicle speed at which the dispersion intensity increases sharply once exceeded, which provides actionable guidance for drivers to adjust driving habits and optimize ventilation strategies to reduce long-term exposure to high PM_2.5 exposure risk.