Spatiotemporal modeling of long-term PM_2.5 concentrations and population exposure in Greece, using machine learning...

Area of Science:

• Environmental Science
• Atmospheric Science
• Public Health

Background:

• Lack of high-resolution, long-term PM2.5 data in Greece and the Eastern Mediterranean hinders accurate health exposure assessments.
• Spatial modeling is crucial for understanding air pollution variability and its impact on populations.

Purpose of the Study:

• To develop a high-resolution (1 km2) gridded dataset of PM2.5 concentrations in Greece from 2015-2022.
• To assess PM2.5 spatial variability and population exposure nationwide.
• To evaluate various modeling approaches for PM2.5 prediction.

Main Methods:

• Integration of in situ observations, meteorology, emissions, and satellite Aerosol Optical Depth (AOD) data.
• Evaluation of seven statistical, machine learning, and hybrid models, with a focus on Random Forest (RF).
• Development of a national-level PM2.5 dataset using the best-performing RF model.

Main Results:

• The RF model achieved high accuracy (R2 = 0.73, MAE = 2.2 μg m-3).
• Winter months exhibited the highest PM2.5 levels (avg. 16.8 μg m-3) due to biomass burning and dispersion limitations.
• The entire Greek population exceeded the WHO PM2.5 guideline (5 μg m-3), with significant exceedances in Central and Northern regions.

Conclusions:

• The developed high-resolution PM2.5 dataset is a reliable tool for air quality management and health studies in Greece.
• Targeted interventions are urgently needed, especially in regions like Ioannina, to mitigate severe air pollution impacts.
• The study underscores the pervasive issue of PM2.5 exposure across Greece, necessitating continued monitoring and policy action.