Satellite-driven modelling of NO₂ and PM₂.₅ across Germany (2019-2024): A multi-sensor machine-learning approach

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Remote Sensing

Background:

• Mapping air pollution spatially is challenging due to localized variations and uneven monitoring networks.
• Germany faces challenges in national air quality assessments because pollutant levels vary significantly over short distances.
• Existing monitoring stations are often unevenly distributed, limiting comprehensive spatial analysis.

Purpose of the Study:

• To evaluate the utility of openly available satellite observations and reanalysis data for annual air quality modeling in Germany.
• To assess the performance of seven machine learning algorithms in modeling nitrogen dioxide (NO2) and fine particulate matter (PM2.5) concentrations.
• To provide a transparent and reproducible framework for national-scale air quality assessment using open global datasets.

Main Methods:

• Combined satellite data (Sentinel-5P for NO2/CO, MODIS NDVI/MAIAC AOD) with ERA5-Land meteorological variables and EuroAirnet observations.
• Evaluated seven machine learning algorithms, including Random Forest and Gradient Boosting, for NO2 and PM2.5 modeling.
• Assessed model performance using random and spatial cross-validation, employing SHAP values for predictor interpretation.

Main Results:

• Random Forest achieved the highest accuracy for NO2 modeling (R² = 0.68), with tropospheric NO2 and NDVI as key predictors.
• Gradient Boosting performed best for PM2.5 modeling (R² = 0.50), with surface pressure, NDVI, and co-emitted gases being influential.
• Annual aggregated MAIAC AOD showed limited independent information for PM2.5 modeling; road density improved local NO2 estimates.

Conclusions:

• Openly available satellite and reanalysis data can support annual air quality modeling for NO2 and PM2.5 in Germany.
• The study successfully reproduced national air pollution patterns, showing a decline in NO2 and regional variability in PM2.5.
• The developed framework offers a reproducible method for national air quality assessment, highlighting potential for integrating more Earth observation data.