Estimating ground-level PM_2.5 levels in Taiwan using data from air quality monitoring stations and high coverage of microsensors

A widespread monitoring network of Airbox microsensors was implemented since 2016 to provide high-resolution spatial distributions of ground-level PM_2.5 data in Taiwan. We developed models for estimating ground-level PM_2.5 concentrations for all the 3 km × 3 km grids in Taiwan by combining the data from air quality monitoring stations and the Airbox sensors. The PM_2.5 data from the Airbox sensors (AB-PM_2.5) was used to predict daily mean PM_2.5 levels at the grids in 2017 using a semiparametric additive model. The estimated PM_2.5 level at the grids was further applied as a predictor variable in the models to predict the monthly mean concentration of PM_2.5 at all the grids in the previous year. The modeling–predicting procedures were repeated backward for the years from 2016 to 2006. The model results revealed that the model R² increased from 0.40 to 0.87 when the AB-PM_2.5 data were included as a nonlinear component in the model, indicating that AB-PM_2.5 is a significant predictor of ground-level PM_2.5 concentration. The cross-validation (CV) results demonstrated that the root of mean squared prediction errors of the estimated monthly mean PM_2.5 concentrations were smaller than 5 μg/m³ and the R² of the CV models of 0.79–0.88 during 2006–2017. We concluded that Airbox sensors can be used with monitoring data to more accurately estimate long-term exposure to PM_2.5 for cohorts of small areas in health impact assessment studies.