Cross-sectional Analysis and Machine Learning Modeling of Ground-Truth and Satellite-Derived NO₂ Concentrations in Temperate Climate Zone

The gap between column-integrated satellite retrievals and patchy ground-level monitoring makes it difficult to accurately characterize the spatial distribution of nitrogen dioxide (NO₂), which is crucial for environmental health. In order to close this observational gap in temperate climate regions, this study suggests an approach for integrating machine learning. We created a geographical predictive model utilizing a Random Forest Regressor by combining ground-truth data from OpenAQ with Sentinel-5P TROPOMI tropospheric column densities from January 2023 to January 2025. To estimate ground-level concentrations, the model combines satellite observations with geographic coordinates and category location contexts. With a Coefficient of Determination (R²) of 0.4337 and a Mean Absolute Error (MAE) of 8.25 µg/m³, the model effectively established a spatial transfer function despite the different physical characteristics of the two datasets. As a proof-of-concept for low-latency air quality assessment, these measurements show a strong capacity to resolve spatial variability in surface NO₂ using satellite inputs. This study provides a scalable framework for improving surveillance in regions without dense sensor infrastructure by validating the effectiveness of machine learning in downscaling satellite products for localized monitoring.