multiple sources via Google Earth Engine:

Sentinel-5P satellite: Provides Satellite-driven modelling of NO₂ and PM?.? across Germany (2019 – 2024): A multi-sensor machine-learning approach

A new research study published in Environmental Pollution presents a contribution to air quality monitoring developed through collaboration between scientists and students. This paper is the result of the course “Environmental Modeling and Health” at the University of Augsburg. Under the supervision of Dr. César Alvarez, master's students played an important role in investigating how open-access satellite data and machine learning can fill critical gaps in national air pollution assessments.

Limitations of stationary monitoring networks

Although Germany has one of the densest monitoring networks in Europe, stationary sensors cannot capture every detail. Pollutant levels often vary considerably within a few hundred meters, due to traffic flow, “street canyons,” and local land use patterns. Relying only on these fixed monitoring stations creates “spatial gaps,” especially in regions with fewer sensors.

The health implications of this variability are significant: long-term exposure to nitrogen dioxide (NO?) and particulate matter (PM?.?) is associated with cardiovascular disease, stroke, and premature mortality. In 2022 alone, approximately 32,600 deaths in Germany were attributable to PM?.? and 9,400 to NO?. Even as levels decline, more than 80% of the urban population in the EU continues to be exposed to concentrations that exceed the strict guidelines set by the World Health Organization (WHO).

A multi-sensor approach driven by open data

• To address these challenges, the students developed a framework that uses only publicly available global datasets. The study integrated daily data on tropospheric NO? and carbon monoxide (CO)
• MODIS sensors: Capture vegetation density (NDVI) and aerosol distribution (AOD)
• ERA5 land reanalysis: Provides meteorological data such as temperature and surface pressure

The team evaluated seven different machine learning algorithms, including random forest and gradient boosting. To ensure that the models were physically consistent, they used “explainable AI” to assess which factors most strongly influenced the results.

Key findings: Pollutant trends 2019 – 2024

The models, created with a resolution of 10 km, showed clear trends across Germany between 2019 and 2024:

• NO? success: Nitrogen dioxide levels showed a steady decline across the country, particularly in large urban and industrial centers such as the Ruhr region. This is attributed to improvements in the vehicle fleet and pandemic-related changes in mobility. For NO?, vegetation structure (NDVI) and the satellite NO? signal were the strongest predictors.
• PM?.? complexity: Fine particulate matter proved more difficult to model due to its diverse sources, such as agriculture, wood burning, and long-range transport. Surface pressure proved to be an important factor in PM?.? accumulation.
• The WHO gap: A notable finding of the study is that although Germany complied with EU limits nationwide for the first time in 2024, 97% of measuring stations still exceed the stricter WHO guidelines for NO?.

?

One of the most significant implications of this research is its global applicability. Since the framework is based entirely on geospatial open data rather than proprietary local inventories, it can be transferred to regions with little or no monitoring infrastructure. This “open science” approach promotes transparency and supports international efforts to comply with the new EU Air Quality Directive (2024/2881).

?

Source Reference: Miller, R., Olbrich, J., Wierer, M., Chen, J., Wurm, M., & Alvarez, C. I. (2026). Satellite-driven modelling of NO? and PM?.? across Germany (2019–2024): A multi-sensor machine-learning approach. Environmental Pollution, 396, 127898. https://doi.org/10.1016/j.envpol.2026.127898.

Source: Modified from Семен Саливанчук / Fotolia.com and Miller et al., 2026