A Machine Learning Approach to Estimate Surface-Level NO₂ Concentration using High Resolution Remote Sensing Observations

Current measurement systems for Nitrogen Dioxide $\left(\mathrm{NO}_{2}\right)$, a key contributor to poor air quality, remain inadequate. In situ networks provide precise surface-level readings but lack high-resolution spatial data. In contrast, satellite-based remote sensing provides wider coverage but doesn’t provide surface-level measurements. Previous studies have shown that machine learning models built using remote sensing data can predict surface-level concentrations with some degree of accuracy. Here, we present machine learning models trained with NASA’s TEMPO instrument which provides high-resolution data. Our work demonstrates that the best model trained with TEMPO data provided accurate hourly surface-level NO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> concentrations and significantly outperformed previous models.