More Details About the "Overall Confidence" Rankings

The uncertainty in the individual pieces of the national-scale assessment (emissions, ambient concentration, and exposure estimates) consist of both quantitative (e.g., model-to-monitor ratios) and qualitative (e.g., quality of data, review of emission inventories) judgements. Below is a discussion on criteria EPA used to estimate the confidence for each of these pieces of the modeling system.

Emission Inventories

The National Toxics Inventory (NTI) is the underlying basis for the 1996 emission estimates in the national-scale assessment. EPA's two primary considerations in ranking the emissions estimates are: Consistency: The 1996 NTI is a composite of emissions estimates generated by state and local regulatory agencies, industry, and EPA. Because the estimates originated from a variety of sources, as well as for differing purposes, they will in turn vary in quality, included pollutants, method of reporting compound classes, level of detail, and geographic coverage.

Variability in Quality and Accuracy of Emission Estimation Methods:

The accuracy of emission estimation techniques vary with pollutants and source categories. In some cases, an estimate may be based on a few or only one emission measurement at a similar source. The techniques used, and quality of the estimates will vary between source categories and between area, major, and mobile source sectors. Generally, the more review and scrutiny given to emissions data by States and other agencies, the more certainty and accuracy there is in that data. The uncertainties outlined in items 1, 2, were considered when ranking each toxic air pollutant for soundness of emissions information. A "higher" certainty rank indicates that the emissions information for that particular toxic air pollutant is sound with regard to its consistency (item 1 above) and in its review for accuracy (item 2 above). Air toxics that fall into this "higher" level of certainty category include: Acrylonitrile, Benzene, Carbon Tetrachloride, Chloroform, Coke Oven Emissions, Ethylene Dichloride, Ethylene Oxide, Hydrazine, Methylene Chloride, Propylene Dichloride, Perchloroethylene, Trichloroethylene, and Vinyl Chloride.

A "medium" certainty rank was assigned to those pollutants that were weak with respect to either items 1 or 2 above. These air toxics include: Acetaldehyde, Acrolein, Arsenic Compounds, Beryllium Compounds, Cadmium Compounds, Chromium Compounds, Diesel PM, Ethylene Dibromide, Formaldehyde, Lead Compounds, Manganese Compounds, Mercury Compounds (the level of review of mercury emissions data is greater than those for other metallic compounds), Nickel Compounds, Polycyclic Organic Matter (POM), 7-PAH, Quinoline, 1,3-butadiene, 1,3-Dichloropropene, and 1,1,2,2-Tetrachloroethane. A "lower" certainty rank was assigned to those pollutants for which both items 1 and 2 listed above were weak points in assimilating its emission inventory. These air toxics include: Hexachlorobenzene (HCB) and PCBs. Atmospheric Dispersion Modeling EPA used the ASPEN model to estimate ambient air toxics concentrations in the national-scale assessment.

To enable the simulation of a large number of air toxics nationwide, a number of simplified approaches must be incorporated into the model, giving rise to the following uncertainties and limitations in predicting concentrations of individual air toxics: Atmospheric Chemistry: The model uses a simplified approach to simulate atmospheric chemistry such as reactive decay and secondary formation. To permit timely computation for the full modeling domain, a simplified algorithm is used to predict secondary formations (production of an air toxics compound due to chemical reactions in the atmosphere). Thus, the more a compound is prone to secondary atmospheric transformation, the more uncertain the modeling estimated for that compound. Background/Long-range Transport Considerations: The dispersion modeling approach simulates only pollutant transport within 50 kilometers of any individual source. To compensate for this limitation, a uniform background concentration is added for some pollutants for which information is available. This simplified approach will impact our confidence of the model predictions for some pollutants more than others (for example, if the percentage of background contribution to the total is high, then the value of the background concentration will be more important). For those air toxics with high levels of background concentrations, the models ability to predict ambient concentrations will be heavily impacted by the accuracy of background estimates.

Model-to-Monitor Evaluations:

In an effort to apply a "reality check" on the ASPEN estimates, modeled air quality estimates were compared with available, but geographically limited ambient air monitoring data. EPA selected a representative subset of seven air toxics (benzene, perchloroethylene, formaldehyde, acetaldehyde, cadmium, chromium, and lead) mainly because these air toxics have the largest number of monitoring sites. For the purpose of evaluating the certainty of the ASPEN modeling system, EPA used the median value of the model-to-monitor ratio (as computed at the geographic location of the monitoring sites) as an indicator of how well the modeling system predicts ambient concentrations of the seven air toxics evaluated. Based on the median value of the model-to-monitor ratios, it is seen that: 1) benzene carries a relatively higher level of certainty; 2) Perchloroethylene and the two aldehydes carry a mid-level of certainty, and 3) the metals (Cadmium, Chromium, and Lead) carry a very low level of certainty. As with the emissions ranking procedure, EPA evaluated each toxic air pollutant to determine how sound it can be modeled (via ASPEN) under the limitations and evaluation techniques outlined in items 1-3 above. Toxic air pollutants that are ranked "higher" for level of modeling certainty are those that are not highly reactive, not dominated by background levels, and those for which the model-to-monitor evaluation was either not conducted or showed a high level of agreement.

These air toxics include: Acrylonitrile, Benzene, Carbon Tetrachloride (mostly from background, but background levels are well characterized), Coke Oven Emissions, Diesel PM, Ethylene Oxide, Hexachlorobenzene, Methylene Chloride, Polycyclic Organic Matter (POM), 7-PAH, Propylene Dichloride, Quinoline, Trichloroethylene, Vinyl Chloride, 1,3-Dichloropropene, and 1,1,2,2-Tetrachloroethane. Air toxics that rank "medium" for level of modeling certainty are those that are: either highly reactive or formed secondarily in the atmosphere, dominated by uncertain background concentrations, or indicative of being poorly (at the mid-level) modeled by the model-to-monitor evaluations. The air toxics that fall into the medium category because they are either highly reactive or formed secondarily in the atmosphere include: Acetaldehyde (secondary formation), Acrolein (secondary formation), 1,3-Butadiene (highly reactive), Formaldehyde (secondary formation), and Hydrazine (highly reactive). The air toxics that fall into the "medium" category because of dominance by uncertain background concentrations include: Chloroform, Ethylene Dibromide, Ethylene Dichloride, and PCBs. Perchloroethylene falls into the medium category because of its model-to-monitor evaluation.

Finally, all metal compounds are ranked "lower" for modeling certainty due to the very low level of correlation seen in the model-to-monitor evaluation of the metals evaluated. Thus, the air toxics that rank "lower" for modeling certainty include all metallic compounds: Arsenic Compounds, Beryllium Compounds, Cadmium Compounds, Chromium Compounds, Lead Compounds, Manganese Compounds, Mercury Compounds and Nickel Compounds. Combining Emissions Ranking with ASPEN Modeling Ranking EPA assigned a combined ranking for the emissions and ASPEN modeling ranking which was driven by the lower of the two rankings.

The following matrix describes how EPA combined the two rankings. The entries in the cells in grey are the combined ranking. For example, if the emissions certainty rank is "higher" and the ASPEN certainty rank is "medium," than the combined rank is a "medium"

Applying the criteria in the above matrix to each of the air toxics evaluated leads to the combined ranking shown in Table B. below.

Table B: Combined Judgements for Emissions and ASPEN Modeled Ambient Concentrations

Exposure Modeling Using HAPEM4

Once EPA estimates ambient concentrations using ASPEN, the next step is to input these values into an exposure model (HAPEM4) to get exposure estimates for each of the pollutants. This is the step to try and account for the movement of people from one location to another, outside to inside, etc.

Many of the factors (e.g, pollutant penetration from outdoors to indoors) considered in the exposure calculations vary from pollutant to pollutant. For some pollutants, these factors are better understood than others. The specific contributors to certainty EPA considered include the following factors:

1. Availability of Exposure Data for Comparison. Does actual exposure data exist on the toxic air pollutant being evaluated? Is surrogate personal exposure data available (a pollutant with similar physical properties)? For some of the toxic air pollutants, there is little or no data available.
   
   

2. Uncertainty due to microenvironment factors. For some air toxics the relationship between indoor and outdoor air concentrations is better understood. Do studies exist for that toxic air pollutant? Can a surrogate pollutant be used? What is the quality of the study? Is the study clear on what portion of the pollutant is from indoor versus outdoor origin?

Using these criteria and others, Table C. summarizes the certainty rank for each of the pollutants evaluated:

Table C: Certainty Ranking for Exposure Modeling Results