US EPA

1255488 

Journal Article 

The moving-window Bayesian maximum entropy framework: estimation of PM(2.5) yearly average concentration across the contiguous United States 

Akita, Y; Chen, JC; Serre, ML 

2012 

1 

Journal of Exposure Science & Environmental Epidemiology
ISSN: 1559-0631
EISSN: 1559-064X 

22 

5 

496-501 

English 

22739679 

10.1038/jes.2012.57 

WOS:000307934000010 

Geostatistical methods are widely used in estimating long-term exposures for epidemiological studies on air pollution, despite their limited capabilities to handle spatial non-stationarity over large geographic domains and the uncertainty associated with missing monitoring data. We developed a moving-window (MW) Bayesian maximum entropy (BME) method and applied this framework to estimate fine particulate matter (PM(2.5)) yearly average concentrations over the contiguous US. The MW approach accounts for the spatial non-stationarity, while the BME method rigorously processes the uncertainty associated with data missingness in the air-monitoring system. In the cross-validation analyses conducted on a set of randomly selected complete PM(2.5) data in 2003 and on simulated data with different degrees of missing data, we demonstrate that the MW approach alone leads to at least 17.8% reduction in mean square error (MSE) in estimating the yearly PM(2.5). Moreover, the MWBME method further reduces the MSE by 8.4-43.7%, with the proportion of incomplete data increased from 18.3% to 82.0%. The MWBME approach leads to significant reductions in estimation error and thus is recommended for epidemiological studies investigating the effect of long-term exposure to PM(2.5) across large geographical domains with expected spatial non-stationarity.Journal of Exposure Science and Environmental Epidemiology advance online publication, 27 June 2012; doi:10.1038/jes.2012.57. 

long-term exposure; geostatistics; moving window; Bayesian maximum entropy; PM2.5