Research on prediction of near-road PM_2.5 concentration by integrating traffic flow and meteorological factors

Near-road particulate matter poses significant risks to public health and the ecological environment, and its levels are affected by the meteorological and traffic factors significantly. However, the contributions of these factors to particulate matter concentrations and the interactions among these factors were not well studied. In this study, the causal relationships among traffic flow (TF), near-road PM_2.5 levels, and meteorological factors were elucidated based on the long-term real-time data on near-road PM_2.5 concentrations alongside concurrent meteorological and traffic data. A predictive modeling framework was developed to predict near-road PM_2.5 concentrations using traffic and meteorological data as input. The results indicate that the correlation between TF and near-road PM_2.5 concentrations is significant (P < 0.05). Furthermore, robust causal relationships were identified between TF and meteorological parameters such as temperature and atmospheric pressure. It is suggested that TF could indirectly influence the level of near-road PM_2.5 by altering meteorological factors. By comparing the prediction performance among Long Short-Term Memory (LSTM), Backpropagation (BP) and Extreme Learning Machine (ELM) models for near-road PM_2.5 concentrations, combined with Shapley Additive exPlanations (SHAP) for feature importance analysis, it revealed that the inclusion of TF data markedly improves model accuracy in near-road PM_2.5 concentrations prediction.