Assessment of PM2.5 concentrations over bright surfaces using MODIS satellite observations

Exposure to particles with an aerodynamic diameter smaller than 2.5 μm (PM2.5) adversely impacts human health. In many geographical regions where ground PM2.5 monitoring is spatially sparse and unsuitable for environmental health inference, satellite remote sensing can potentially be used for estimating human exposure to PM2.5. However, retrieval of the aerosol optical depth (AOD) using the Dark Target (DT) algorithm is uncertain in many regions worldwide (e.g. western USA, the Middle East and central Asia) due to low signal-to-noise ratio as a result of high surface reflectivity in the spectral bands used by the algorithm. In this study we use the Deep Blue (DB) algorithm as well as a combined DB-DT algorithm for AOD retrievals. The AOD products are used to predict ground PM2.5 using mixed effects models and the daily calibration approach. Models for the two study areas (Israel and San Joaquin Valley, Central California) were developed independently and then compared to each other. Using the AODDB within a mixed effects model considerably improved PM2.5 prediction in high reflectance regions, revealing in both study areas enhanced model performance (in terms of both R2 and the root mean square prediction error), significant increase in the spatiotemporal availability of the AOD product, and improved PM2.5 prediction relative to using AODDT retrievals.