Prediction of NO3 Radical Concentrations from 2015 to 2020 in Beijing and Shanghai Based on Air Quality Models and Machine Learning Methods

Nitrate radicals (NO₃) in the nocturnal boundary layer are key oxidants that influence nighttime atmospheric oxidation capacity and the nitrogen cycle. However, their low concentrations, short lifetimes, and complex nighttime chemistry pose challenges for large-scale spatiotemporal observations. In this study, we couple a machine learning model with the CMAQ model and historical observation data to predict NO₃ over the long term. This approach combines an exact physical-chemical framework with observational data support to better capture the spatial and temporal characteristics of NO₃. Our results show that: (1) CMAQ accurately simulates NO₂ and O₃ concentrations and also performs well for N₂O₅, indicating that the nighttime NO₃ reaction framework in CMAQ is correct. (2) Combined with the CMAQ result, the stacking model improves the R of NO₃ predictions by an average of 0.17 compared with single models, and its SHAP results align with current atmospheric chemistry. (3) After predicting NO₃ levels and comparing summer and winter conditions in Shanghai and Beijing, our results reveal a notable decrease in NO₃ in Shanghai during summer, likely due to declining nighttime O₃. However, reduced heterogeneous hydrolysis during Shanghai’s winter nights may lead to a slight rise in NO₃ concentrations. Additionally, NO₃ in Beijing do not show a strong decrease and even increases slightly.