[Urban Ozone Driving Factors Based on Explainable Machine Learning].

Abstract

Sixteen sites in the coastal city of Qingdao, including eight national control sites, seven provincial control sites, and one background site, were selected. By coupling the extreme gradient boosting （XGBoost） model with the interpretability SHapley Additive exPlanations （SHAP） module, the impact of meteorological elements and atmospheric pollutant emissions on ozone （O₃） pollution was investigated. The results indicated that from 2019 to 2023, meteorological factors contributed 67.7% to O₃ formation, whereas emissions from atmospheric pollutants accounted for 32.3%. Surface solar radiation significantly affected O₃ formation from 10：00 to 17：00. A positive correlation existed between temperature and O₃ concentration, peaking at 14：00. A relative humidity below 70% was conducive to O₃ formation and a relative humidity above 70% had a 94% probability of negatively contributing to O₃ production. Particularly between 12：00 and 16：00, relative humidity significantly and positively contributed to O₃ formation. When the boundary layer height was below 500 meters, it positively affected O₃ concentration, whereas above this height, its impact weakened. In the morning and late afternoon, boundary layer height promoted the formation of O₃ concentration. Easterly （E） to southwesterly （SW） winds had a positive effect on O₃ concentrations in Qingdao. NO₂ showed a negative response in the morning （06：00-11：00） and a positive response in the afternoon （12：00-15：00）. PM_2.5 had a nonlinear positive correlation with O₃, positively affecting O₃ concentration from 07：00 to 14：00 PM and suppressing it from 15：00 to 18：00. Significant differences existed in the dominant factors of O₃ concentration across different areas and seasons. In the western Laoshan District, Yangkou, the western site of the Shinan District, and eastern and western parts of the West Coast New Area, surface solar radiation had a noticeably lower impact on O₃ than in other locations. The effect of NO₂ was most significant in the eastern parts of the West Coast New Area, western Laoshan District, and Yangkou. PM_2.5 affected O₃ formation more in these sites than in others of Qingdao. In spring, the impact of NO₂ was more significant in the western Laoshan and Shinan Districts. In summer, surface solar radiation was more influential in Yangkou, the eastern and western parts of the West Coast New Area, the western Laoshan District, and the Shibei District, whereas, relative humidity was the key factor in other locations. In autumn, temperature and surface solar radiation were the main factors affecting O₃. In winter, the contribution of NO₂ was higher than that in different seasons, with anthropogenic emissions playing a more important role in O₃ formation. The analysis of days exceeding O₃ standards showed that surface solar radiation and NO₂ were the main drivers of exceeding O₃ concentrations. For all sites, the total SHAP values of PM_2.5 and PM₁₀ on days exceeding O₃ standards ranged between 6.1 μg·m^-3 and 12.4 μg·m^-3.