Co-Occurring Extremes of PM<sub>2.5</sub> and Ozone in Warm Seasons of the Yangtze River Delta of China: Insights from Explainable Machine Learning
Yan Lyu, Danni Wu, Fuliang Han, Huiying Zhang, Fangfang Lv, Azhen Kang, Yijia Hu, Xinyu Pang
Abstract
Recently, summertime PM 2.5 and ozone extremes were reported to frequently co-occur in southern China. In this study, we further demonstrate that their co-occurring extremes can spread into warm seasons in the Yangtze River Delta (YRD) region of China. The annual co-occurrence frequency ranged from 26% to 50% in the YRD from 2015 to 2022, with higher frequencies observed in coastal cities. Notably, the co-occurrence frequency was higher during the COVID-19 pandemic, implying that such co-occurrence may be more spatially widespread with continuous PM 2.5 reduction in China. Taking the pandemic period as an example, we leveraged a machine learning algorithm (i.e., Random Forest) coupled with SHapley Additive ExPlanation (SHAP) to identify higher relative importance of solar radiation-related variables (e.g., surface net solar radiation) during co-occurrence periods compared to non-co-occurrence periods in the YRD. Additionally, incorporating volatile organic compounds (VOCs) measurements, we further showed the higher relative importance of VOCs to the extremes of ozone and PM 2.5 through a case study at Shaoxing (a typical city in the YRD). Overall, the findings highlight the increasing potentials for co-occurring extremes with ongoing PM 2.5 reductions in the YRD and suggest that reducing VOCs (e.g., halocarbons) may help mitigate these extremes in the future.