Integrating Chemical\nMechanisms and Feature Engineering\nin Machine Learning Models: A Novel Approach to Analyzing HONO Budget

Nitrous acid (HONO) serves as the primary source of OH\nradicals\nin the atmosphere, exerting significant impacts on atmospheric secondary\npollution. The heterogeneous reactions of NO<sub>2</sub> on surfaces\nand photolysis of particulate nitrate or adsorbed nitric acid are\nimportant sources of atmospheric HONO, yet the corresponding kinetic\nparameters based on laboratory investigations and field observations\nexhibit considerable variations. In this study, we developed an explainable\nmachine learning model to analyze the HONO budget using two years\nof summer urban supersite observations. By integrating chemical mechanisms\nand feature engineering into our machine learning model, we assessed\nthe contributions of different sources to HONO and inferred the kinetic\nparameters for the primary HONO formation pathways, thereby partially\naddressing the limitations associated with predetermined rate coefficients.\nOur findings revealed that the primary source of daytime HONO in the\nsummer was the photolysis of nitric acid adsorbed on both aerosol\nand ground surfaces, accounting for over 40% of its unknown sources.\nThis was followed by the photoenhanced heterogeneous conversion of\nNO<sub>2</sub> and the photolysis of particulate nitrate. Additionally,\nwe derived the corresponding kinetic parameters, analyzed their influencing\nfactors, and confirmed that machine learning methods hold great potential\nfor the study of the HONO budget.