Estimation of local daily PM2.5 concentration during wildfire episodes: integrating MODIS AOD with multivariate linear mixed effect (LME) models

Abstract

Seasonal peaks of air pollution from wildfires are increasing in frequency and severity in the western provinces of Canada. During these episodes, populations are exposed to adverse short-term health effects due to elevated levels of fine particulate matter, which is the primary pollutant associated with smoke. The spatial resolution of ground-based monitoring records generally is not sufficient for emergency preparedness and epidemiological studies of such events. Accurate spatial and temporal models of smoke pollution for the study of smoke exposure effects require reliable, fine-scale input data. Satellite aerosol optical depth (AOD) measures can provide a valuable alternative to the coarse spatial resolution of ground PM2.5 monitoring network measurements. Numerous statistical approaches have been used to estimate the link between AOD and PM2.5, most of which consider the relationship between AOD and PM2.5 as being fixed over space and for an entire day; however, due to time-varying meteorological variables, that relationship changes over time and space. Hence, to capture the effects of temporal and spatial variations on the AOD-PM2.5 relationship, two nested linear mixed effect (LME) models are developed herein. Daily estimation of PM2.5 concentration is derived by incorporating nested period-zone-specific random effects of the AOD-PM2.5 relationship over the province of Alberta, Canada. Model validation shows that LME improves the estimation performance of the model compared with ordinary multivariate linear regression by more than 115%. Our findings indicate that the potential of the LME model increases when additional variables are integrated with AOD measures in a multivariate framework. This single model yields an array of reliable spatial-temporal estimates of daily PM2.5 concentrations from wildfire at fine spatial resolution.