Data fusion of monitor data and chemical transport model simulations

Abstract

Associations between air quality and acute health effects vary across pollutants and across spatial and temporal metrics of concentration. Studies that investigate these associations require data that are both spatially and temporally complete across many pollutants. The objective of this study was to create accurate and complete pollutant concentration fields by combining the benefits of observed data and a chemical transport model, CMAQ, while reducing the effects of their incomplete spatial and temporal coverage and limited accuracy, respectively.Using a previously developed approach, these spatially and temporally resolved pollution fields were created over the domain of Georgia for 12 pollutants (8-hr maximum O3, 1-hr maximum NO2, NOx, CO and SO2, and 24-hr average PM10, PM2.5 and five PM2.5 subspecies) and four years (2009 - 2012). It was found that the results from this data fusion agree very well with observations as well as results from previous studies. Through a cross-validation analysis, it was found that the fusion is also able to estimate concentrations far from monitor locations with reasonable accuracy. SO2 is predicted most poorly due to difficulties in capturing plumes from coal combustion. For the other 11 pollutants considered, R2 values ranged from 35.8% to 83.8% from the cross-validation analysis. Because of their ability to capture spatial and temporal variations, concentration fields produced here are well suited for use in epidemiological studies. Two one-step methods were also investigated. When implemented for NO2 and PM2.5 in 2010, these alternatives were not able to predict concentrations as well as the original method,but are computationally much more efficient. It was found that developing and using models of annual mean concentration fields can account for some of the mismatch between point measurements and 12-km gridded CMAQ simulations and thus improves predictions. For larger scale applications, such as over the entire U.S., it is recommended that a one-step method incorporating annual mean models be implemented to provide results for use in health studies.