Machine learning-assisted diagnosis of abrupt air pollution emissions: insights from firework emissions in China

The identification and quantification of abrupt atmospheric pollutant emissions are crucial for the implementation of effective emergency control measures. Firework emissions during the Chinese New Year represent a widespread and distinct scenario for studying such abrupt emissions. In this study, we applied a machine learning-based approach to quantify the contribution of firework emissions to key air pollutants, using surface observations from across China during 2017–2023. The results reveal a sharp increase in PM2.5 concentrations from New Year’s Eve into the early hours of the following day, observed across multiple cities. Notably, smaller cities (e.g., Huangshan and Gannan) experienced markedly higher peak PM2.5 concentrations compared to larger cities (e.g., Beijing and Shenzhen). The relative contribution (PCSF) of firework emissions to PM2.5 was generally higher in parts of the Southwest, Northeast, and Northwest regions, with the PCSF value from 2017 to 2019 ranging from 150 to 600 μg/m³, while during the pandemic period (2020-2022), it decreased to 100-300 μg/m³. These results suggest that strict environmental control policies and population mobility patterns have a significant impact on reducing PM2.5 concentrations. Our study highlights the potential of machine learning techniques in diagnosing and assessing the occurrence of abrupt air pollution events.