Improve OMI Observations on Ground-Level NO₂ Using Multiple Observations, Simulations, and Machine Learning

Nitrogen dioxide (NO₂ ) is a criteria air pollutant with adverse impacts on human health and the environment. An accurate ground-level NO₂ dataset with high spatial resolution is beneficial for NO₂ pollution management and public health studies. In this study, we leverage long-term ozone monitoring instrument (OMI) NO₂ observations by converting OMI NO₂ vertical column densities into ground-level NO₂ concentrations and improving its spatial resolution to 1×1 km by using the light gradient boosting machine (LightGBM) learning model in three metro areas in USA: Metro New York, New York (NY), Baltimore, Maryland (MD), and Houston, Texas (TX). Our improved ground-level NO2 products achieved robust performance across all regions, with a correlation coefficient of 0.897 in NY, 0.876 in Baltimore, and 0.914 in Houston. The corresponding root-mean-squared error (RMSE) was 3.278 ppb in NY, 2.385 ppb in Baltimore, and 2.084 ppb in Houston, respectively. Furthermore, the products effectively captured the temporal variations in observed NO₂ concentrations across all cities. We also found that emissions, population density, boundary layer height (BLH), and winds are consistently important contributors to NO₂ concentrations across all three cities, while each city also has its own significant factors. This attribution analysis will be helpful to state and local air pollution agencies in their air pollution management efforts.