Improved surface NO Retrieval: Double-layer machine learning model construction and spatio-temporal characterization analysis in China (2018-2023).

As an important atmospheric pollutant causing serious harm to human health and the natural environment, monitoring of surface NO (SNO) level is of critical importance. However, the current SNO retrieval models neglect to consider the influence of NO vertical hierarchical structure on the process of converting the tropospheric NO columns (XNO) to the SNO levels. Meanwhile, conventional machine learning models struggle to capture complex spatiotemporal relationships between SNO and XNO, which lead to the large differences between the current model results and the site-based measurements. To enhance the accuracy of SNO level inversion, this study incorporated the NO vertical stratification characteristics and its spatial-temporal variation mechanisms over a long time series. By leveraging the advanced Light Gradient Boosting Machine (LGBM) and Extremely Randomized Forests (ERF) models, a Double-Layer Machine Learning (DLML) framework was developed to estimate SNO levels across China from 2018 to 2023. Based on the results of this study, the temporal and spatial variation patterns of SNO levels across China, including key regions, were comprehensively analyzed. The results showed that: (1) Compared with the traditional model, the DLML model proposed in this study showed better performance, in which the R of spatio-temporal cross-validation reached 0.87. This represented an improvement of about 10 % over previous models. At the same time, MAE and RMSE were reduced to about 4.24 μg/m and 5.79 μg/m respectively. (2) The retrieved SNO levels in China mainly showed a decreasing trend from the central and eastern coastal areas to the surrounding areas, and the annual average concentration had reached the level of the World Health Organization (WHO) air quality guidelines. In terms of time, the retrieved SNO levels showed a U-shaped variation, with the highest in winter, followed by autumn, spring, and summer, reaching the peak in January and December, and then reaching the valley in June-August. (3) The two abnormal events occurred in winter, indicating that the meteorological conditions in winter were the main reason for the change of SNO in the air. Among them, the factors that cause the peak values of Wuhan and Yangtze River Delta may also be due to the high level of economic development, dense population activities, and frequent industrial activities in the two regions, resulting in their own SNO emissions.