Predicting ozone formation in petrochemical industrialized Lanzhou city by interpretable ensemble machine learning

• Published in: Environmental pollution (1987) Vol. 318; p. 120798
• Main Authors: Wang, Li, Zhao, Yuan, Shi, Jinsen, Ma, Jianmin, Liu, Xiaoyue, Han, Dongliang, Gao, Hong, Huang, Tao
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 01.02.2023
• Subjects: Air Pollutants - analysis; Air Pollution - analysis; air quality; algorithms; atmospheric chemistry; China; Communicable Disease Control; COVID-19; COVID-19 infection; dust storms; Ensemble machine learning; Environmental Monitoring - methods; Ground-level ozone (O3); Humans; Machine Learning; meteorology; O3 formation sensitivity; observational studies; Ozone - analysis; photochemistry; pollution; Pollution episodes; Shapley algorithm; solar radiation; Volatile Organic Compounds - analysis; China; Ensemble machine learning; Pollution episodes; O3 formation sensitivity; Ground-level ozone (O3); Shapley algorithm; O formation sensitivity; Ground-level ozone (O)
• ISSN: 0269-7491; 1873-6424; 1873-6424
• DOI: 10.1016/j.envpol.2022.120798

Ground-level ozone (O3) formation depends on meteorology, precursor emissions, and atmospheric chemistry. Understanding the key drivers behind the O3 formation and developing an accurate and efficient method for timely assessing the O3–VOCs-NOx relationships applicable in different O3 pollution events are essential. Here, we developed a novel machine learning ensemble model coupled with a Shapley additive explanation algorithm to predict the O3 formation regime and derive O3 formation sensitivity curves. The algorithm was tested for O3 events during the COVID-19 lockdown, a sandstorm event, and a heavy O3 pollution episode (maximum hourly O3 concentration >200 μg/m3) from 2019 to 2021. We show that increasing O3 concentrations during the COVID-19 lockdown and the heavy O3 pollution event were mainly caused by the photochemistry subject to local air quality and meteorological conditions. Influenced by the sandstorm weather, low O3 levels were mainly attributable to weak sunlight and low precursor levels. O3 formation sensitivity curves demonstrate that O3 formation in the study area was in a VOCs-sensitive regime. The VOCs-specific O3 sensitivity curves can also help make hybrid and timely strategies for O3 abatement. The results demonstrate that machine learning driven by observational data has the potential to be a very useful tool in predicting and interpreting O3 formation. [Display omitted] •O3 levels increased during the COVID-19 and a heavy O3 pollution event.•O3 levels decreased during a sandstorm event.•Ensemble learning model was integrated to forecast O3 in different episodes.•Critical features of ensemble learning model were identified by SHAP method.•Ensemble learning model was used to establish O3-VOC-NOx sensitivity curves.