Comparison of the Industrial Source Complex and AERMOD dispersion models: case study for human health risk assessment.

Terrain elevation data representative for each site were obtained from various U.S. Geological Survey (USGS) data sources. Because 7.5-min terrain data at a scale of 1:24,000 were not readily available for both sites, 1[degrees] USGS digital elevation models (DEMs) at a scale of 1:250,000 were used instead. To determine surface roughness, a circle with a radius of 3 km was drawn around the center of each site and the circular area was divided into 12 sectors of 30[degrees] each, starting with sector 1, which was centered on 0[degrees](i.e., due north). The land use within each sector was classified as either water, urban, deciduous forest, coniferous forest, or grassland, consistent with EPA guidance. (18) The areal extent of each land use classification, in square meters and as a percentage of the total sector area, was determined using the Geographical Information System (GIS) software package, ArcView (ESRI).

In practice, emissions emanating from stacks, vent boxes, and tank vents are modeled as point sources. Fugitive emissions from tank farms are modeled as area sources in which the emission rate is divided by the source area to obtain an area-weighted emission rate. Fugitive emissions from process pads and buildings are modeled as volume sources and assigned dimensions on the basis of the building size in accordance with EPA guidance. (13) Following this practice, a representative stack, process building, and tank farm were chosen on each site for this study. Under the MACT rules, it is common practice for high production volume facilities to vent many of their process and fugitive emissions to a common pollution control device such as a scrubber or a thermal oxidizer unit, thus minimizing the number of emission points to be modeled. Table 1 summarizes the source parameters used in the modeling analyses.

The structures near the stacks are typical of structures found at industrial facilities (e.g., pipe racks, sheds, process pads, and process buildings) and are generally less than 6 m in height. The flow of air past structures can result in wakes and cavities forming on the downwind side of the structure, which contributes aerodynamic downwash. (13) As a rule of thumb, if the height of the stack is greater than 2.5 times the height of the nearby structures, the effects of aerodynamic downwash will be avoided. (14) In this study, the heights of both stacks were greater than 2.5 times the height of the nearby structures.

The commercial software packages, BREEZE ISC GIS Pro and BREEZE AERMOD GIS Pro (Version 4.0., Trinity Consultants Inc., 2002) were used for all modeling runs. A steady-state, unit emission rate of 1 g/sec x [m.sup.2] was used for all point and volume sources in all modeling runs. A steady-state, unit emission rate of 1 g/sec x [m.sup.2] was used for all area sources. Emissions were assumed to occur 8760 hr/yr with no downtime. Use of the unit emission rate allows the air modeling output (the ambient air concentration) to be expressed on a unit emission rate basis (i.e., [micro]g/[m.sup.3] per g/sec). The unit emission rate is not chemical specific and its use precludes having to run the model for each individual chemical emitted. To calculate the ambient air concentration of a particular chemical (in [micro]g/[m.sup.3]), the air modeling output (in [micro]g/[m.sup.3] per g/sec) is simply multiplied by the chemical emission rate (in g/sec).

A receptor grid for off-site receptors was set up using a Cartesian grid with a 100-m grid spacing out to a distance of 3 km from the approximate center of each site. Ground level was chosen as the height of all receptors. A fenceline was drawn around each site and the on-site receptors removed from the analysis. The distance from the stack to the fenceline was 150 m for Site 1 and 60 m for Site 2. The models were run in concentration mode for all sources using the 1-hr and total period averaging options, rural dispersion coefficients (ISC only), and the regulatory default options.

For Site 1, all runs were made with a meteorological dataset that contained 4 consecutive years of data and had been approved by the state agency for use in air dispersion modeling at this site. The ISC dataset originally contained a consecutive 5 yr of meteorological data. However, when the dataset was being compiled for AERMOD, from the same stations and for the same 5 yr, we were not able to locate upper air data for the first year. Therefore, for consistency, the ISC and AERMOD data were processed from the same stations for the same 4 yr. The predominant wind direction for Site 2 was from the southwest.

For Site 2, all runs were made with a meteorological dataset that contained 3 consecutive years of meteorological data and had been approved by the state agency for air modeling at this site. The meteorological data for ISC and AERMOD were processed from hourly surface and 12-hr upper air observations recorded at the same National Weather Service stations and for the same 3 yr. The predominant wind direction (i.e., direction from which the wind is blowing) for Site 1 was from the northwest.

For ISC modeling, the meteorological preprocessor, PCRAMMET, was used to create the ISC ready files. The AERMOD files were created using the AERMET preprocessor. In both models, the appropriate land use data were entered directly into the preprocessors. The appropriate land use data were determined from an assessment of the land usage in a 3-km radius around each site. In the case of AERMOD, seasonal variation and land use data were used in the preprocessor to yield different values of albedo, bowen ratio, and surface roughness for the four standard seasons.

All modeling output was collected in plot files that contained geographical coordinates (i.e., 'X' and 'Y' coordinates) for each receptor as well as the modeled ground-level air concentration for the appropriate averaging period. The modeled air concentrations were expressed as [micro]g/[m.sup.3]. However, because the modeled air concentrations were based on a unit emission rate of 1 g/sec they were expressed as [micro]g/[m.sup.3] per g/sec. The modeled air concentrations were multiplied by the source-specific emission rate to generate the predicted air concentrations. The modeled 1-hr and total period average air concentrations were imported into ArcGIS for data analysis and interpretation to assess the impact at the maximally exposed individual (MEI) as well as the spatial distribution of air concentrations and resultant human health risk.

Human Health Risk Assessment

The risk assessment evaluated the potential harm to the modeled receptors due to inhalation of the modeled maximum total period average (i.e., the exposure concentration). Risk assessors refer to the potential harm from exposure to carcinogens as risk and the potential harm from exposure to noncarcinogens as hazard. For noncancer effects, the exposure concentrations are compared with toxicity reference concentrations (RfCs). RfCs are an estimate (with uncertainty spanning perhaps an order of magnitude) of a continuous inhalation exposure to a chemical that is likely to be without an appreciable risk of deleterious effects to the human population (including sensitive subgroups) during a lifetime. For inhalation exposures, noncancer hazards are estimated by dividing the modeled exposure concentration (EC) by the RfC to yield a hazard quotient (HQ) for an individual chemical. (22) The HQ is calculated using eq 1.

HQ = EC/RfC (1)

A HQ of 1 or less for the inhalation pathway indicates that exposure to that chemical is not likely to result in any adverse health effects.

For carcinogenic effects, the lifetime incremental cancer risk (LICR) evaluates the degree to which a receptor may have an increased likelihood of developing cancer over a lifetime due to a lifetime of exposure to a chemical. (22) For carcinogenic effects, the exposure concentrations are compared with the inhalation unit risk (IUR) for a chemical. The LICR is calculated using eq 2.

LICR = EC x IUR (2)

For the great majority of chemicals, the LICR provides an upper-bound prediction of the risk of contracting cancer over a lifetime as a result of a lifetime of exposure (via inhalation) to the chemical at the modeled exposure concentration. LICRs are expressed as a unitless probability and are represented in scientific notation as a negative exponent of 10. For example, the probability of developing cancer of one chance in 10,000 is written as 1 x [10.sup.-4]. In reality, the actual risk may be lower than the predicted risk. (22) EPA cites an acceptable risk range of 1 x [10.sup.-4] to 1 x [10.sup.-6] for potential cancer risk. (23) Table 2 lists the RfCs and IUR values used in this case study. It should be noted that the above equations produce a quite simplistic and conservative estimate of hazard or risk. In reality, a distribution of hazard or risk would more accurately reflect the natural variability observed in humans.

RESULTS

Modeling Results

The study involved four separate model runs to predict the total period average and the 1-hr average air concentrations at all receptors. Because only point sources are affected by building downwash, the comparisons between the standard models and their enhanced versions incorporating the PRIME algorithm were only evaluated for point sources. Table 3 presents the maximum total period average and the maximum 1-hr average air concentrations predicted by the various models and scenarios described above.