Predicting fine particulate concentrations near a busy intersection in Sydney using artificial neural networks

Abstract

Scientific evidence of the connection between vehicle emissions and public health outcomes continues to grow. Key to this connection is the accurate monitoring and prediction of pollution concentrations within transport microenvironments at fine levels of spatial and temporal disaggregation. This paper explores the potential for using Artificial Neural Networks for such a purpose, focusing on temporallydisaggregate prediction of PM2.5 concentrations for a busy intersection in Sydney. The main findings are that with knowledge of ambient PM2.5 concentrations, traffic volumes and weather conditions, the approach is able to explain over 90 percent of the variation in PM2.5 concentrations at the intersection, and over 70 percent of the variation when applied to an independent data set collected at the same location.