A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Environmental Engineering and Management

Thesis (M. Sc.) - Asian Institute of Technology, 2025

In Thailand, key environmental and public health concerns center on exposure to fine particulate matter with an aerodynamic diameter of 2.5 micrometers or less (PM2.5), along with ground-level ozone (O3). These pollutants have been widely studied in low- and middle-income countries over recent decades due to their adverse health effects. Despite this, air quality monitoring infrastructure in Thailand remains limited, with 65 out of 77 provinces having only one or two monitoring stations. This leaves substantial areas without direct air quality measurements. To address this gap, satellite-based observations have become instrumental in estimating the spatial distribution of PM2.5 and O3 concentrations. In particular, data from Aerosol Optical Depth (AOD) and satellite-derived ozone products are leveraged through machine learning techniques, providing a valuable complement to traditional ground-based monitoring in under resourced regions. This study aimed to estimate PM2.5 and O3 concentrations across Thailand for the year 2023, using a 9 {u00D7} 9 km2 grid resolution. Three models were employed: a multiple linear regression (MLR) model, random forest (RF) and extreme gradient boosting (XGBoost). These models incorporated data from the Geostationary Environment Monitoring Spectrometer (GEMS). The accuracy of AOD and O3 products from GEMS was evaluated against measurements from Thailand{u2019}s AERONET station. Additionally, meteorological inputs were derived from the Weather Research and Forecasting (WRF) model and validated using data from the Pollution Control Department (PCD). GEMS data showed high agreement with AERONET observations, with coefficients of determination (R2) of 0.845 for AOD and 0.999 for O3. Among the modeling approaches, the random forest model performed best in estimating PM2.5 concentrations, achieving an R2 of 0.967 and a root mean square error (RMSE) of 8.058. However, the model{u2019}s prediction for ground-level O3 showed a low correlation (r = -0.092), indicating limited reliability for O3 estimation in this context. Given these findings, only PM2.5 estimates were used for health impact assessments. The study focused on the short-term effects of PM2.5 exposure on stroke mortality at the provincial level, following World Health Organization (WHO) guidelines. Results indicated higher PM2.5-related stroke mortality in northern provinces such as Nan, Chiang Rai, Chiang Mai, Sukhothai, and Phayao. In contrast, southern and eastern regions experienced the lowest rates of stroke fatalities.