Predicting daily PM2.5 concentration using satellite data at 1-kilometer resolution

This study integrates satellite and ground-based measurements to estimate fine particulate matter (PM2.5) concentrations in Thailand at a high spatial resolution (1 km). PM2.5 data from 2011 to 2020 were collected by Thailand's Pollution Control Department (PCD) and Bangkok’s Air Quality and Noise Management Division, while satellite data were obtained from NASA’s Earth Observing System Data and Information System (EOSDIS) using the Moderate Resolution Imaging Spectroradiometer (MODIS). Key satellite-derived predictors identified for PM2.5 estimation include Aerosol Optical Depth (AOD), Land Surface Temperature (LST), Normalized Difference Vegetation Index (NDVI), and Elevation (EV). Incorporating temporal variables, such as Week of the Year (WOY) and Year, further improved model performance by capturing seasonal variations in PM2.5 levels.To address the limited coverage of ground-based monitoring stations, this study develops predictive models using multiple linear regression and three machine learning algorithms. Among these, the random forest model demonstrated the highest accuracy. Model performance was evaluated using the coefficient of determination (R²) and root mean square error (RMSE), yielding strong results for training (R² = 0.95, RMSE = 5.58 µg/m³), validation (R² = 0.78, RMSE = 11.18 µg/m³), and testing datasets (R² = 0.71, RMSE = 8.79 µg/m³).Regional analysis revealed that the model performed best in northern Thailand (R² = 0.82), followed by the central (R² = 0.69), northeastern (R² = 0.69), and southern regions (R² = 0.45). The stronger performance in the north may be attributed to persistently high PM2.5 levels, whereas other regions exhibit slight declines. These findings highlight the potential of satellite data to improve PM2.5 monitoring, support air quality management, and inform policy decisions to mitigate air pollution in Thailand.