Comparison and Evaluation of Spatiotemporal Air Quality Exposure Fields Developed using Ten Methods

摘要

A variety of measures are currently being used to assign air pollutant concentrations to individuals in air quality exposure-related studies. The outcomes of those assessments reflect the methods used and may differ between methods. Comprehensive comparison of multiple methods to a common data set is needed to inform users of potential issues that may impact their use in health studies. Here, we compare pollutant concentration fields generated using ten methods and evaluate them regarding their use in exposure related studies. The application is the Atlanta (GA, USA) metropolitan region and the pollutants are CO, NO2, SO2, 03, PM2.5 and PM2.5 constituents elemental carbon, organic carbon and sulfate. The selected methods involve the use of central monitor data, multi-site averaging, three spatial interpolation methods (inverse distance weighting, tessellation and kriging), land use regression with satellite AOD information, chemical transport modeling (CAAAQ-derived fields) combined with data assimilation/fusion methods, and fine scale dispersion modeling. The central monitor and site averaging methods are spatially, temporally and chemically incomplete, and performance varies across pollutants, depending on the number and locations of monitors. Interpolation methods are able to generate spatially resolved fields, although the spatial patterns of primary pollutants are poorly captured due to sparse monitoring networks. The results are similar between IDW and kriging, but could also vary depending on other interpolation approach applied or interpolation parameters used. The CAAAQ model provides complete fields and reasonable spatial concentration patterns, but CAAAQ model performance shows substantial biases and other errors that vary between pollutants and over time. Incorporating data fusion approaches improves model performance. Satellite AOD-derived fields perform similarly to using CAAAQ fields with data fusion. Results of this study contribute to our understanding of the strengths and weakness of different methods regarding their application in exposure related studies.