Evaluation of roadside air quality using deep learning models after the application of the diesel vehicle policy (Euro 6)

Abstract

AbstractEuro 6 is the latest vehicle emission standards for pollutants such as CO, NO2 and PM, that all new vehicles must comply, and it was introduced in September 2015 in South Korea. This study examined the effect of Euro 6 by comparing the measured pollutant concentrations after 2016 (Euro 6–era) to the estimated concentrations without Euro 6. The concentration without Euro 6 was estimated by first modeling the air quality using various environmental factors related to diesel vehicles, meteorological conditions, temporal information such as date and precursors in 2002–2015 (pre–Euro 6–era), and then applying the model to predict the concentration after 2016. In this study, we used both recurrent neural network (RNN) and random forest (RF) algorithms to model the air quality and showed that RNN can achieve higher R2 (0.634 ~ 0.759 depending on pollutants) than RF, making it more suitable for air quality modeling. According to our results, the measured concentrations during 2016–2019 were lower than the concentrations predicted using RNN by − 1.2%, − 3.4%, and − 4.8% for CO, NO2 and PM10. Such reduction can be attributed to the result of Euro 6.