AAAR 33rd Annual Conference
October 20 - October 24, 2014
Rosen Shingle Creek
Orlando, Florida, USA
Abstract View
Estimating Spatiotemporal Variations of PM$_(2.5) over the Pittsburgh Metropolitan Area Using Aerosol Optical Depth
Tao Xue, RICHARD BILONICK, Daniel Connell, Evelyn Talbott, Judith Rager, LuAnn Brink, University of Pittsburgh
Abstract Number: 31 Working Group: Health Related Aerosols
Abstract Epidemiological studies have associated PM$_(2.5) with adverse health outcomes. Exposure assessment plays a critical role in estimating health risk of PM$_(2.5) but is limited by sparsely-distributed monitoring stations. Previous studies have applied satellite remote sensing of aerosol optical depth (AOD) to predict spatiotemporal variations of PM$_(2.5) mass concentration measurements but rarely considered the time-varying PM$_(2.5)-AOD association and potentially complex spatiotemporal correlation. This study is a part of the Pittsburgh Aerosol Research and Inhalation Epidemiology Study (PARIES) and aims to predict PM$_(2.5) over the Pittsburgh metropolitan area from 2001-2008.
We collected PM$_(2.5) mass concentration measurements from 102 monitors at 47 sites and MODIS AOD (Level 2) measurements from the Terra satellite. Measurements were aggregated at the daily level resulting in 61,346 PM$_(2.5) measurements (square-root transformed) and 309,919 AOD observations. We designed a two-stage approach. First, AOD was smoothed using space-time kriging with a product-sum covariance structure to reduce spatial misalignment between PM$_(2.5) and AOD. Next, smoothed AOD was associated with PM$_(2.5) through a time-varying coefficient mixed effects model, producing best linear unbiased predictors of PM$_(2.5) at given spatiotemporal coordinates. We used cross validation (CV) to evaluate our models.
PM$_(2.5) had a mean of 15.2 mu-g/m$^3 with a standard deviation of 9.6 mu-g/m$^3 and was higher in autumn (19.8 mu-g/m$^3) and summer (15.2 mu-g/m$^3), lower in spring (13.1 mu-g/m$^3) and winter (12.8 mu-g/m$^3). Daily PM$_(2.5) was 1.7% more highly correlated with smoothed AOD (R$^2=0.4466) than raw AOD (R$^2=0.4392). PM$_(2.5)-AOD association was found to vary seasonally and was higher in summer/autumn than in winter/spring. Ten-fold CV confirmed that space-time kriging with the varying-coefficient mixed effects model performed well with CV R$^2 of 0.9882 and 0.9157 (CV RMSE 0.0313 and 0.4435), respectively, for smoothing AOD and predicting PM$_(2.5). We will present details of the statistical methodology and examples of the resulting daily exposure maps.