Spatiotemporal estimation of hourly PM2.5 using AOD derived from geostationary satellite Fengyun-4A and machine learning models for Greater Bangkok

This study used four individual machine learning (ML) models (random forest, adaptive boosting, gradient boosting, and extreme gradient boosting), and a stacked ensemble model (SEM) for PM2.5 estimation over Greater Bangkok (GBK) during the dry season for 2018 –2022. Aerosol optical depth (AOD) from Fengyun-4A satellite was used as the main predictor variable. The other predictor variables include meteorological variables, fire hotspots, vegetation index, terrain elevation, and population density. Surface PM2.5 from 17 air quality monitoring stations was used for model development and evaluation. Satellite AOD aligns reasonably well with AOD from two AERONET stations in the study area in terms of correlation coefficient (r), mean bias (MB), mean error (ME), and root mean square error (RMSE). Among the individual models, adaptive boosting performed the best withr = 0.75, MB = 0.55 µg m−3, ME  = 9.1 µg m−3, and RMSE  = 12.9 µg m−3. As for SEM which comprises all individual models, it outperformed every individual model, withr = 0.84, zero MB, ME = 7.2 µg m−3, and RMSE  = 10.4 µg m−3. In two additional cases of haze hours and clean hours, SEM is best overall while adaptive boosting is superior to the other individual ML models. The case of haze hours has lower model predictability, suggesting elevated PM2.5 is difficult to predict. SEM was thus chosen to map PM2.5 as well as exposure intensity over GBK. Good agreement betwee...
Source: Air Quality, Atmosphere and Health - Category: Environmental Health Source Type: research