Evaluation of Domestic Emission Reduction Effects on Decreases in Concentrations of PM2.5 and Its Compositions Using a CMAQ-ML Model

초록

Although PM2.5 concentrations in South Korea have decreased over time, they still exceed the World Health Organization guideline, indicating the need for fast and reliable tools to evaluate emission reduction policies. In this study, a CMAQ-ML framework was developed that includes an observation-based concentration prediction model and a Relative Response Factor (RRF) model, enabling rapid assessment of emission reduction effects. The model was constructed using the XGBoost algorithm with hyperparameter optimization. The model was trained using one-year data (year of 2022) and tested to another one-year data (year of 2023). Performance evaluation showed that the concentration prediction model reproduced the observed PM2.5 and its component concentrations more accurately than the CMAQ model alone, while the RRF prediction model successfully reproduced the CMAQ-based RRF response characteristics. In terms of computational efficiency, the CMAQ model required approximately 190 hours for the one-year simulation, whereas the CMAQ-ML model produced similar results within seconds. Application of the emission reduction scenario based on the 2(nd) comprehensive plan for fine dust management showed that the annual mean PM2.5 concentration decreased by approximately 0.6 mu g/m(3), which was mainly driven by decreases in NO3- and NH4+. In addition, the emission reduction scenario reduced the frequency of high PM2.5 concentration days. The CMAQ-ML framework enables rapid and accurate evaluation of domestic emission reduction policies and can serve as an effective decision-support tool for air quality management.

키워드