Explainable Machine Learning Reveals the Unknown Sources of Atmospheric HONO during COVID-19

Abstract

Nitrous acid (HONO) is a key precursor of the hydroxyl radical (•OH), playing an important role in atmospheric oxidation capacity. However, unknown sources of HONO (P_unknown) are frequently reported and the potential sources are controversial. Here, we explored P_unknown during COVID-19 in different seasons and epidemic control phases in Shanghai by eXtreme Gradient Boosting (XGBoost) and Shapley Additive Explanations (SHAP) for the first time. They demonstrated that the decrease of anthropogenic activity would inhibit secondary formation of HONO, as epidemic control policies turned strict. The explainable machine learning revealed that nitrogen dioxide (NO₂) had significant impacts on the P_unknown during spring 2020 (P1), where P_unknown could be fully explained by including light-induced heterogeneous conversion of NO₂ on ground, building, and aerosol surfaces. With the untightening of epidemic control in spring 2021 (P3), the HONO budget came to balance after further addition of the photolysis of particulate nitrate (NO₃^–) and soil HONO emission. As for P2 (summer), P_unknown decreased by 54% with all new sources added. These results provide new insights into HONO chemistry in response to reduced anthropogenic emissions, improving the predictions of atmospheric oxidation capacity.