Near Road PM2.5Concentration Estimation Using Artificial Neural Network Approach

Evidence has shown the increasing association between particulatematter and adverse health problems. In urban areas, most of households arelocated near arterials, which are exposure to PM2.5directly. Hence, it iscritical to accurately predict the near-road PM2.5concentration anddistribution for health risk analysis. This paper applies artificial neuralnetwork (ANN) to estimate the near-road PM2.5concentration. Factorsinfluencing the detected concentration are classified into four categories:traffic-related, weather-related, location-related and background-related.The estimated values are compared with concentrations detected bymonitoring campaigns in Gainesville, FL and Shanghai, China.Distinguished from previous research, this study illustrates the PM2.5dispersion and distribution within50m near road with portable PM andweather instruments. The results indicate that ANN-based model is capableof producing accurate estimation of pollutant dispersion near road. Besides,PM2.5concentration decayed about a half at30m distance from an arterialroad in Gainesville, FL. Background contributes to more than2/3of thedetected value at roadside in Shanghai, and the distance-decay pattern is notas obvious as that in Gainesville, which is different from previous studiesreported in the literature. An ANN-based model performs better afterremoving the background concentration and with higher concentrationvalue of PM2.5.