Exploring Innovative Methods for Air Dust Pollution Monitoring using Satellite Data and Products: a Comparative Study of Chalidiki (Greece) and Ihalainen (Finland)

Surface mining can have a significant impact on various aspects of the environment, including human health. One of the most harmful effects on the human body has the particulate pollution that can be released into the air during mining operations. Particles smaller than 10 micrometres, known as PM10, can be inhaled and linked to respiratory and cardiovascular diseases. Monitoring this type of pollution is challenging because of its spatial distribution. The atmosphere is a three-dimensional space, and it is impossible to cover an entire area with in-situ monitoring stations. However, remote sensing satellite data provides an effective alternative for monitoring airborne dust. Although satellite-based data have limitations, they provide stable spatial and temporal coverage that is valuable for long-term monitoring and offers advantages for monitoring large, inaccessible areas.

This study was carried out at two sites, on the Chalkidiki peninsula in Greece and in the Finnish inland near the city of Lappeenranta, between years 2015-2024 and 2013-2024, respectively. To achieve the objectives of this work, local in-situ measurements of PM10 pollution provided by the Finnish Meteorological Institute and the Hellas-Gold environmental platform and two remote sensing data products were used: the CAMS European air quality re/analysis dataset provided by the Copernicus Atmosphere Monitoring Service (CAMS) and the Aerosol Optical Depth (AOD) data at 0.55 micrometres from NASA's MODIS Terra and Aqua satellites at 10, 3 and 1 km resolution.

All data were aggregated to daily values for further evaluation, from which CAMS data showed strong correlations with in-situ data, making it a valuable resource for monitoring, with correlation coefficients (rs) around 0.8, depending on the site. In contrast, MODIS data presented a different scenario, with correlations peaking at 0.5 and in some cases becoming negative. This incompatibility may be due to long periods of missing data because of cloud cover. CAMS data were also aggregated into monthly and annual values to observe trends and seasonality. These showed a decreasing trend in concentrations, suggesting an improvement in air quality over the years at both locations. Machine learning techniques were also used, including Prophet and BSTS for trend and seasonality prediction, and XGBoost and Random Forest for short-term prediction. Prophet and XGBoost were found to be more accurate with this type of data.

Future steps in this research will include further data modelling and prediction methods, with a focus on integrating other machine learning methods, and exploring additional remote sensing datasets suitable for developing an optimal workflow for this use case.

Acknowledgments:

The presented analysis was conducted under the support of the EC grant MultiMiner. The MultiMiner project is funded by the European Union’s Horizon Europe research and innovation actions programme under Grant Agreement No. 10109137474.