Satellite products of atmospheric composition are abundant and useful but the retrieval products relevant for air quality typically provide only vertically integrated observations and generally have a coarse spatial resolution. For understanding everyday exposure to air pollution right where people live and walk, it is necessary to translate the total column observations from the satellite to surface information and to improve the spatial resolution of the information.
To accomplish this, NILU has recently developed a machine learning-based approach for better exploiting atmospheric satellite data for societal applications. This approach called S-MESH (Shetty et al., 2024, Shetty et al., 2025) uses data from satellite instruments such as TROPOMI and VIIRS as well as other geospatial datasets in an ML framework to convert satellite-retrieved column observations to surface concentrations and downscale the data from relatively coarse native resolution to a finer grid, thus enabling city-scale analysis.
In this Master’s thesis the student would apply S-MESH and contribute to further improving this tool. Potential improvements and the specific research questions of the thesis can be defined together with the student based on interest and qualifications and include for example one or more of the following:
- identifying useful new spatial predictor variables, including them in S-MESH, and evaluating their impact,
- applying S-MESH at higher spatial resolution than the 1 km x 1 km currently used and evaluating the achievable accuracy,
- testing neighborhood-aware ML approaches for potentially improved performance, or
- moving from daily predictions to hourly predictions using observations from the recently launched Sentinel-4 geostationary satellite instrument.
Recommended background/qualifications:
Competence in a high-level programming language (e.g. Python or R). Knowledge of statistics/machine learning and processing geospatial or satellite data sets.
References
- Shetty, S., Schneider, P., Stebel, K., David Hamer, P., Kylling, A., and Koren Berntsen, T.: Estimating surface NO2 concentrations over Europe using Sentinel-5P TROPOMI observations and Machine Learning, Remote Sensing of Environment, 312, 114321, https://doi.org/10.1016/j.rse.2024.114321, 2024.
- Shetty, S., Hamer, P. D., Stebel, K., Kylling, A., Hassani, A., Berntsen, T. K., and Schneider, P.: Daily high-resolution surface PM2.5 estimation over Europe by ML-based downscaling of the CAMS regional forecast, Environmental Research, 264, 120363, https://doi.org/10.1016/j.envres.2024.120363, 2025.
