Mapping the daily nitrous acid (HONO) concentrations across China during 2006-2017 through ensemble machine-learning...

Understanding nitrous acid (HONO) distribution is critical for atmospheric chemistry, as HONO contributes to hydroxyl radical formation. However, China lacks detailed spatial data on HONO concentrations. Prior research has shown that HONO influences air quality and photochemical processes, but its sources and spatial variability remain unclear. Existing studies focus on localized measurements, leaving large regions unmonitored. This gap motivated the need for a nationwide HONO model. No prior work had resolved how to estimate HONO across China using machine learning. The absence of long-term, high-resolution HONO data limits progress in air pollution modeling. This study addresses the lack of spatially resolved HONO data by proposing a new modeling approach. The goal is to enhance understanding of HONO's role in atmospheric chemistry and pollution patterns.

This study aimed to estimate daily HONO concentrations across China from 2006 to 2017 using an ensemble machine learning model. The specific problem addressed is the lack of spatially resolved HONO data, which hinders accurate atmospheric chemistry modeling. The motivation stems from the need to understand HONO's role in air quality and its variability over time. The model integrates random forest, gradient boosting, and back propagation neural network algorithms. The study also sought to identify key factors influencing HONO concentrations. The ensemble approach was chosen to improve prediction accuracy and robustness. The model's performance was validated using cross-validation techniques. This work provides a novel framework for HONO estimation at a national scale.

The study employed an ensemble machine learning model combining random forest, gradient boosting, and back propagation neural network algorithms. Input variables included NO3− concentration, urban area, and NO2 column density. The model was trained on ground-level HONO observations collected from 2006 to 2017. Spatial resolution was set at 0.25° to capture regional variability. Cross-validation techniques were used to assess model performance. Sample-based, site-based, and by-year validation methods were applied. Variable importance analysis identified key predictors of HONO concentrations. The model's output included daily HONO maps at a national scale.

The ensemble model achieved an R² of 0.7 with an RMSE of 0.36 ppbv for sampled-based validation. Site-based validation showed an R² of 0.67 and RMSE of 0.36 ppbv. By-year validation yielded an R² of 0.62 and RMSE of 0.40 ppbv. HONO hotspots were identified in the BTH, PRD, YRD, and Sichuan Basin regions. These hotspots correlated with tropospheric NO2 columns and surface NO3− levels. HONO concentrations remained stable from 2006 to 2013 but declined after 2013. This decline was attributed to the Air Pollution Prevention and Control Plan. NO3− concentration, urban area, and NO2 column density were top predictors of HONO levels.

The ensemble model successfully estimated HONO concentrations across China from 2006 to 2017. The model's performance was validated using multiple cross-validation methods. HONO hotspots aligned with known pollution regions and NO2 and NO3− distributions. The study found that HONO levels stabilized until 2013 and then decreased. This decline was linked to air pollution control policies. NO3− concentration and urban area were key variables in predicting HONO. Agricultural land and forest had minor effects on HONO concentrations. The model provides essential data for atmospheric chemistry research in China.