Estimating Daily PM_2.5 and PM₁₀ over Italy Using an Ensemble Model

Area of Science:

• Environmental Science
• Epidemiology
• Remote Sensing

Background:

• Accurate spatiotemporal estimates of particulate matter (PM) are crucial for understanding exposure in epidemiological studies.
• Existing methods often face challenges with data gaps and spatial resolution.

Purpose of the Study:

• To develop and validate an improved method for estimating daily PM2.5 and PM10 concentrations over Italy.
• To enhance the accuracy of PM exposure assessment for epidemiological research.

Main Methods:

• Utilized satellite-derived aerosol optical depth (AOD) data, imputing missing values with a spatiotemporal random forest model.
• Employed an ensemble modeling approach, combining linear mixed effects, random forest, extreme gradient boosting, and a chemical transport model.
• Fused individual model estimations using a geographically weighted generalized additive model (GAM) ensemble for spatiotemporal weighting.

Main Results:

• The ensemble GAM model significantly improved PM estimation accuracy compared to individual models, reducing cross-validated root mean squared error by 1-42%.
• Successfully generated spatiotemporally resolved daily estimates for PM2.5 and PM10 concentrations over Italy for 2013-2015.

Conclusions:

• The developed ensemble modeling approach provides reliable and high-resolution PM exposure data.
• These enhanced PM estimates are suitable for application in future epidemiological studies in Italy.