Search

Fine particulate matter (PM2.5) level in Bangkok and vicinity areas, Thailand, has increased for a half decade and exceeds the national air quality standard in dry season. However, there are a limited number of air quality monitoring stations measuring PM2.5 concentration. A land use regression (LUR) model can be applied to an estimate of PM2.5 concentrations at nonmonitored locations. This is the first study that develops the LUR model with the random forests (RF) algorithm incorporating WRF/CMAQ for predicting daily PM2.5 concentration for the year 2017 in Bangkok and vicinity areas. CMAQ-simulated PM2.5 concentrations, WRF-simulated meteorological parameters, population density, and land use variables were used as predictors of the LUR model. The predictor variables were assessed by variable importance measurements. A cross-validation (CV) technique was performed to evaluate the prediction performance in spatial and temporal aspects and obtained the coefficient of determination (R²) and root mean square error (RMSE) values. The daily PM2.5 concentrations were estimated by the developed model at 1x1 km resolution in the study area. The predictive performance of the developed land use random forests (LURF) model was compared with that of the conventional LUR model built with multiple linear regression to reveal the advantages of the RF algorithm. Accordingly, the LURF model obtains a spatial CV-R² of 0.63 and a temporal CV-R2 of 0.70, which are higher than that of the conventional LUR model (a spatial and a temporal CV-R² of 0.46). The findings demonstrate that the LURF model incorporating WRF/CMAQ is advantageous to accurately estimate ambient PM2.5 level in Thailand. [ABSTRACT FROM AUTHOR]