Abstract
Estimates of exposure to PM(2.5) are often derived from geographic characteristics based on land-use regression or from a limited number of fixed ground monitors. Remote sensing advances have integrated these approaches with satellite-based measures of aerosol optical depth (AOD), which is spatially and temporally resolved, allowing greater coverage for PM(2.5) estimations. Israel is situated in a complex geo-climatic region with contrasting geographic and weather patterns, including both dark and bright surfaces within a relatively small area. Our goal was to examine the use of MODIS-based MAIAC data in Israel, and to explore the reliability of predicted PM(2.5) and PM(10) at a high spatiotemporal resolution. We applied a three stage process, including a daily calibration method based on a mixed effects model, to predict ground PM(2.5) and PM(10) over Israel. We later constructed daily predictions across Israel for 2003-2013 using spatial and temporal smoothing, to estimate AOD when satellite data were missing. Good model performance was achieved, with out-of-sample cross validation R(2) values of 0.79 and 0.72 for PM(10) and PM(2.5), respectively. Model predictions had little bias, with cross-validated slopes (predicted vs. observed) of 0.99 for both the PM(2.5) and PM(10) models. To our knowledge, this is the first study that utilizes high resolution 1km MAIAC AOD retrievals for PM prediction while accounting for geo-climate complexities, such as experienced in Israel. This novel model allowed the reconstruction of long- and short-term spatially resolved exposure to PM(2.5) and PM(10) in Israel, which could be used in the future for epidemiological studies.