A spatiotemporal land-use-regression model to assess individual level long-term exposure to ambient fine particulate matters

We aimed to establish a spatiotemporal land-use-regression (ST-LUR) model assessing individual level long-term exposure to fine particulate matters (PM(2.5)) among 6627 adults enrolled in Guangdong province, China from 2015 to 2016. We collected weekly average PM(2.5) concentration (from the air quality monitoring stations) and visibility, population density, road density and types of land use of each air quality monitoring station and participant’s residential address from April 2013 to December 2016. A ST-LUR model was established using these spatiotemporal data, and was employed to estimate the weekly average PM(2.5) concentration of each individual residential address. Data analysis was carried out by R software (version 3.5.1) and the SpatioTemporal package was used. The results showed that the ST-LUR model applying the land use data extracted using a buffer radius of 1300 m had the best modelling fitness. The results of 10-fold cross validation showed that the R(2) was 88.86% and the RMSE (Root mean square error) was 5.65 μg/m(3). The two-year average of PM(2.5) prior to the date of investigation were calculated for each participant. This study provided a novel method to precisely assess individual level long-term exposure to ambient PM(2.5), which may extend our understanding on the health impacts of air pollution. • Variables input in the spatiotemporal land-use-regression (ST-LUR) model include visibility, population density, road density, and types of land use. • The land use data should be extracted using a buffer radius of 1300 m. • The R(2) of the ST-LUR model was 88.86% and the RMSE was 5.65 μg/m(3), indicating the good performance of the model.