Air Quality Estimation with Sentinel-5P
Used pretrained models for air quality variable estimation, supporting faster, more robust environmental monitoring.
Training data
-
Primary source: Sentinel-5P TROPOMI Level 2 products for NO₂, CO, and O₃ (daily global coverage)
-
Geographic scope: Urban and peri-urban areas across Europe and sub-Saharan Africa
-
Auxiliary datasets:
-
Ground-based in situ measurements (e.g., Copernicus Atmospheric Monitoring Service stations, AirBase)
-
Meteorological reanalysis (ERA5) for contextual features (e.g., temperature, wind speed)
-
Emission inventories and population density maps for contextual correlation
-
Link to main concept
- Copernicus Foundation Models: Applied to multi-dimensional EO data to improve estimation of atmospheric pollutants.
- High-dimensional input processing: Combines spectral, temporal, and spatial information using foundation model embeddings.
- Low-label regimes: Ground station data is sparse; the pretrained models help in leveraging EO data for reliable estimation where labels are scarce or unavailable.
Impact
Using Sentinel-5P data, this spotlight application evaluates how pretrained foundation models can support fast and accurate air quality estimation. The approach improves the prediction of key atmospheric pollutants, with enhanced generalisation in diverse regions, especially where dense ground station networks are unavailable. The outcomes highlight the role of foundation models in delivering scalable environmental intelligence to support public health assessments, regulatory monitoring, and early warning systems.
