Estimating daily PM2.5 and PM10 over Italy using an ensemble modeling approach
Year: 2019
Type: conference-abstract
Abstract: OPS 47: Increasing spatiotemporal resolution in assessment of exposure to outdoor air pollutants, Room 412, Floor 4, August 27, 2019, 4:30 PM - 5:30 PM Background: Exposure to particulate matter (PM) pollution has been associated with various adverse health effects. PM concentrations are measured worldwide sparsely by monitors; therefore, models that allow continuous spatiotemporal estimations are needed for better exposure assessment. The primary aim of this research is to estimate PM2.5 and PM10 concentrations over Italy for 2013-2015 years using an ensemble model. Methods: In the first stage multi-angle implementation of atmospheric correction (MAIAC) aerosol optical depth (AOD) product is imputed for all locations using a novel spatiotemporal land-use random-forest (RF) model. In the second stage PM estimations are produced using four different modeling approaches: linear mixed effects (LME), RF, extreme gradient boosting (XgBoost) and an Italian chemical transport model, the flexible air quality regional model (FARM). The imputed MAIAC AOD alongside additional spatial and temporal predictors are used as inputs in the three first models. In the third stage a general additive model (GAM) ensemble model is used to fuse the estimations from the four models by providing different weights for each model and allowing them to vary over space and time. Results: Cross-validated performance was robustly evaluated separately for each individual model showing variability in the root mean squared error (RMSE) and R2 of the five different models that were used. For example, in 2015, the GAM ensemble model showed the lowest RMSE (6.3) and highest R2 (0.81) for PM2.5 in comparison to those of the four separate models (7.8, 0.71 for LME; 6.9, 0.77 for RF; 6.58, 0.79 for XgBoost; and 10.42, 0.48 for FARM). Conclusions: This study shows that the ensemble model is a strong method for estimating PM concentrations, combining PM estimations from different models and taking the advantages of each individual model. The continuous PM estimations produced by this model can be applied in further epidemiological studies in Italy.
Source: Environmental Epidemiology
Authors: Shtein A, Kloog I, Schwartz J, Silibello C, Michelozzi P, Gariazzo C, Viegi G, Forastiere F, Stafoggia M
Institutions: Ben-Gurion University of the Negev, Harvard University, ASL Roma, Istituto Nazionale per l'Assicurazione Contro gli Infortuni sul Lavoro, National Academies of Sciences, Engineering, and Medicine, Istituto di Biomedicina e di Immunologia Molecolare Alberto Monroy, National Research Council, National Research Council, Karolinska Institutet
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,200
OpenAlex ID: W2981413552