Relative congener scaling of Polychlorinated dibenzo-p-dioxins and dibenzofurans to estimate building fire contributions in air, surface wipes, and dust samples

Abstract

The United States Environmental Protection Agency collected ambient air samples in lower Manhattan for about 9 months following the September 11, 2001 World Trade Center (WTC) attacks. Measurements were made of a host of airborne contaminants including volatile organic compounds, polycyclic aromatic hydrocarbons, asbestos, lead, and other contaminants of concern. The present study focuses on the broad class of polychlorinated dibenzo-p-dioxins (CDDs) and dibenzofurans (CDFs) with specific emphasis on the 17 CDD/CDF congeners that exhibit mammalian toxicity. This work is a statistical study comparing the internal patterns of CDD/CDFs using data from an unambiguous fire event (WTC) and other data sets to help identify their sources. A subset of 29 samples all taken between September 16 and October 31, 2001 were treated as a basis set known to be heavily impacted by the WTC building fire source. A second basis set was created using data from Los Angeles and Oakland, CA as published by the California Air Resources Board (CARB) and treated as the archetypical background pattern for CDD/CDFs. The CARB data had a congener profile appearing similar to background air samples from different locations in America and around the world and in different matrices, such as background soils. Such disparate data would normally be interpreted with a qualitative pattern recognition based on congener bar graphs or other forms of factor or cluster analysis that group similar samples together graphically. The procedure developed here employs aspects of those statistical methods to develop a single continuous output variable per sample. Specifically, a form of variance structure-based cluster analysis is used to group congeners within samples to reduce collinearity in the basis sets, new variables are created based on these groups, and multivariate regression is applied to the reduced variable set to determine a predictive equation. This equation predicts a value for an output variable, OPT: the predicted value of OPT is near zero (0.00) for a background congener profile and near one (1.00) forthe profile characterized by the WTC air profile. Although this empirical method is calibrated with relatively small sets of airborne samples, it is shown to be generalizable to other WTC, fire source, and background air samples as well as other sample matrices including soils, window films and other dust wipes, and bulk dusts. However, given the limited data set examined, the method does not allow further discrimination between the WTC data and the other fire sources. This type of analysis is demonstrated to be useful for complex trace-level data sets with limited data and some below-detection entries.