Appendices

Appendix D: Review of Air Quality Modeling Studies

ATSDR views environmental sampling data as critical inputs to the public health assessment process. As evidence of this, ATSDR strongly recommends the use of validated sampling data as the basis for public health decisions. In some circumstances, however, sampling data are not sufficient to characterize all site-specific exposures. For instance, few air samples were collected on Vieques between the early 1970s and 1999–the years when the Navy's military training exercises using live bombs were most extensive–and the few samples that were collected are of questionable quality. In such cases, models are arguably the best tools available to evaluate the nature and extent of contamination. ATSDR emphasizes that models are only capable of estimating exposure concentrations, based on a scientific understanding of how chemicals move in the environment. All models, however, have assumptions and uncertainties and may not accurately represent actual environmental conditions. Therefore, ATSDR carefully reviews all modeling applications to determine whether they provide meaningful estimates of environmental contamination and whether they can be used in the public health assessment process.

When evaluating the four key questions in this PHA (see Section V), ATSDR determined that the available sampling data were sufficient to address two of the key questions, without the need for modeling. On the other hand, insufficient sampling data were available to characterize air quality during live bombing exercises and to evaluate releases from the Navy's periodic use of certain materials (e.g., depleted uranium, chaff). ATSDR decided to use modeling analyses to put these two exposure scenarios into perspective.

The remainder of this appendix presents ATSDR's review of the modeling studies available for the island of Vieques. This includes modeling studies conducted by contractors to the Navy (Appendix D.1), by an engineer from Vieques (Appendix D.2), and by contractors to ATSDR (Appendix D.3). Sections V.C and V.D describe how ATSDR used these modeling analyses to reach public health conclusions.

D.1 Review of the Navy's Modeling Study of Live Bombing Activities (IT 2000, 2001)

In February 2000, contractors to the Navy completed an air dispersion modeling study of selected air emissions sources on the island of Vieques (IT 2000). The modeling study had two objectives: to determine whether certain environmental regulations apply to the Navy's operations on Vieques and to estimate ambient air concentrations of contaminants released to the air during military training exercises and open detonation of unexploded ordnance. In May 2001, the Navy released a revision to this air dispersion modeling study to correct a computational error (IT 2001). Once corrected, the estimated emission rates (and likewise the estimated ambient air concentrations) increased slightly, by less than 5% for most contaminants. Copies of both versions of this dispersion modeling report are in the Vieques site's records repositories, which are located at Biblioteca Publica on Vieques, the Vieques Conservation and Historical Trust, and the University of Puerto Rico School of Public Health.

D.1.A Overview of the Navy Contractor's Modeling Approach

The following paragraphs review three key features of the Navy contractor's modeling analysis: how emission rates were estimated, how atmospheric fate and transport was simulated, and how results were presented and interpreted. Refer to Section D.1.B for ATSDR's evaluation of the scientific rigor of the Navy contractor's modeling analysis.

• Approach to estimating emissions. The Navy contractor estimated emissions from a variety of air pollution sources on the Navy property at Vieques. The majority of emissions for most of the contaminants originated from sources on the LIA. These included emissions from air-to-ground exercises, ship-to-shore exercises, land-based exercises, and open detonation of unexploded ordnance. Other sources considered included emissions from generators and small arms firing ranges. However, for almost every contaminant, emissions from these sources accounted for a very small portion of the total emissions calculated. Accordingly, ATSDR focused its review of this dispersion modeling study on the approach used to estimate emissions from activities associated with bombing exercises.

  The Navy contractor estimated emissions for a single calendar year, 1998–a year the authors asserted was representative of prior years' activities at Vieques. Further, emissions were estimated only on an annual basis. Estimating emissions over shorter time frames, such as highest daily emissions, was not conducted. The following paragraphs describe how emissions were estimated for different categories of contaminants. ATSDR's comments on these approaches are presented in Section D.1.B.

  Emissions of particulate matter. Particulate matter emissions were estimated using a model that the Army Research Laboratory developed to predict how much dust, smoke, and debris is released to the air during realistic battlefield situations (Army Research Laboratory 2000). In general, this model characterizes the size of craters formed by explosions and then quantifies the amount of particulate matter that may be released to the air as a result. According to this model, the emission rate of particulate matter following an explosion is a function of several parameters, including the net explosive weight (NEW) of the ordnance fired and soil properties.

  When calculating emission rates using this model, the Navy contractor assumed that only those bombs that contain at least 10 pounds of high explosives will generate craters that release particulate matter to the air upon detonation. During 1998, the base year for the modeling analysis, only four types of ordnance–all air-to-ground bombs–met this 10-pound criterion. Because all land-based and ship-to-shore ordnance used in 1998 contained less than 10 pounds of NEW, this approach effectively asserted that none of the ship-to-shore or land-based activities caused emissions of particulate matter.

  To estimate particulate matter emissions, the Navy contractor first estimated the volume of craters generated by bombs. The mass of soil apparently ejected from craters was estimated by multiplying the crater volumes listed above by an assumed soil density (1.5 g/cm³). To estimate how much of the soil released is emitted as PM10, the Navy contractor multiplied the mass of soil ejected by a scaling factor documented by the Army Research Laboratory as the proportion of crater ejecta that is believed to be released as "small particles," or particles with radii less than 10 microns (Army Research Laboratory 2000). This scaling factor was 0.01007, meaning that roughly 1% of the crater ejecta is assumed to be released as PM10 emissions, and the remaining crater ejecta will be larger particles that settle to the ground in the vicinity of the impact location.

  Following the aforementioned approach, the Navy contractor estimated that total PM10 emissions from the air-to-ground bombing exercises and open detonation activities were 76 tons per year. The combined PM10 emissions from all other sources evaluated (e.g., wind-blown dust, dust from driving on dirt roads, generator exhaust) was less than 5 tons per year. Thus, by the Navy contractor's approach, emissions from the bombing exercises account for an overwhelming majority of the estimated PM10 emissions. ATSDR has many comments on the approach described in the previous paragraphs, as Appendix D.1.B describes further.

  Emissions of explosion by-products. The Navy contractor used emission factors to estimate releases of inorganic and organic explosion by-products. The emission factors were derived from a series of source tests, known as "Bangbox" studies, that measured the amounts of selected inorganic and organic chemicals released during the open detonation of various types of ordnance. The Bangbox is a flexible structure in which ordnance is detonated. Because the Bangbox is completely enclosed, pollutants released during the detonation do not escape the structure and can be measured by air sampling equipment. The Bangbox has thus allowed scientists to estimate emission factors for various types of ordnance, many of which are similar to those the Navy uses at Vieques. The emission factors used estimate the amounts of chemicals released to the air per weight of NEW detonated.

  The Navy contractor reviewed many Bangbox emission factors to select appropriate factors to apply to the military training activities at Vieques. Emission factors were identified for several types of contaminants, including criteria pollutants, metals, volatile organic compounds, and semi-volatile organic compounds. The Navy contractor selected two different sets of Bangbox emission factors to estimate air releases from the following sources:

  • Emission factors for air-to-ground exercises. For the year being evaluated (1998), range utilization statistics indicated that usage of MK82, MK83, and MK84 bombs accounted for more than 99% of the explosives used during all air-to-ground exercises. The explosive charges in these bombs are composed of either 2,4,6-trinitrotoluene (TNT) with aluminum powder or a mixture of TNT, RDX, and aluminum powder. When estimating air emissions from air-to-ground exercises, the Navy contractor considered only those Bangbox studies that tested these specific types of explosives. For every contaminant, the Navy contractor selected the highest emission factor from the relevant Bangbox results.
  • Emission factors for ship-to-shore exercises and land-based exercises. In any given year, the Navy uses ordnance with varying compositions for its ship-to-ground and land-based exercises. Rather than attempting to model each composition individually, the Navy contractor instead used an approach designed to provide an upper-bound estimate of actual emissions. For every chemical considered, the Navy contractor selected the highest emission factor from all of the Bangbox tests identified. These tests included emissions from various propellants and high explosives.

  Once the Navy contractor selected emission factors for the two types of activities, air emissions were estimated by multiplying the chemical-specific emission factors by the corresponding weight of explosives (expressed as NEW) used.

  Emissions of metals. To estimate the amounts of metals released during military training exercises, the Navy contractor used emission factors published in an open detonation burn plan conducted for another Navy installation (Radian 1996). ATSDR notes that these emission factors combine together the amounts of metals detected during Bangbox experiments and the entire mass of four metals–aluminum, copper, manganese, and zinc–commonly found in bomb casings. Therefore, the emission factors assume that the entire metallic content of the casings is vaporized during an explosion.

  As with the emission factors for the by-products of explosions, two different sets of emission factors for metals were selected. First, emission factors for air-to-ground exercises were selected from the Bangbox studies involving TNT, RDX, and aluminum powder explosives. Second, emission factors for all other exercises were determined from the highest factors reported for all Bangbox studies identified.

• Approach to modeling atmospheric transport. The Navy contractor used the INPUFF dispersion model to estimate ambient air concentrations of the pollutants released during the military training exercises. This model was originally designed to simulate atmospheric transport for instantaneous or "puff-like" releases, like the emissions that occur from individual bombing events. The modeling scale for INPUFF ranges from downwind distances of several meters to tens of kilometers (EPA 1986)–a scale that is therefore sufficient for modeling transport from the LIA to the residential areas of Vieques. The INPUFF model does not explicitly account for terrain or chemical reactions that might take place in the atmosphere (EPA 1986).

  The INPUFF simulations calculated ambient air concentrations of contaminants at downwind locations from a limited set of input parameters. To run INPUFF, the Navy contractor used meteorological conditions observed at the US Naval Air Station Roosevelt Roads, located on the eastern shore of the main island of Puerto Rico, less than 10 miles from the western shore of Vieques. The Navy contractor also selected values for the locations and size of dust clouds generated by explosions during the military training exercises. Section D.1.B presents ATSDR's comments on the selected input values.

• Presentation and interpretation of results. The dispersion modeling report summarizes estimated ambient air concentrations in two sections of the report. First, two figures show how concentrations of two contaminants (manganese and RDX) vary with location in the residential areas of Vieques. Second, a table documents the estimated annual average concentrations of all pollutants considered in the modeling analysis, except for particulate matter. ATSDR does not present the Navy contractor's specific findings in this PHA, because ATSDR's conclusions regarding live bombing exercises are based entirely on the agency's own dispersion modeling outputs (see Appendix D.3). Nonetheless, ATSDR notes that the Navy contractors analyses predict that the highest ambient air concentrations of pollutants occur in the northeast portion of the residential area of Vieques, which is also the residential area located closest to the LIA. The Navy contractor's modeling study did not estimate ambient air concentrations over shorter averaging periods (e.g., maximum 24-hour air concentrations).

D.1.B ATSDR's Review of the Navy Contractor's Modeling Analysis

Because limited environmental sampling data are available to characterize how the Navy's live bombing activities affected air quality at Vieques, ATSDR thoroughly evaluated all modeling studies of these activities to determine if the modeling results can be used to reach scientifically defensible public health conclusions. ATSDR's specific comments on the Navy contractors' dispersion modeling analysis follow, organized by topic.

• Scope of modeling study. As Section D.1.A indicates, the Navy contractor's modeling study examined only annual average air quality impacts. When evaluating environmental contamination, however, ATSDR examines the public health implications of both long-term and short-term exposures. For the air exposure pathway, this typically involves characterizing both annual average and highest 24-hour average ambient air concentrations. Therefore, the Navy contractor's study is not sufficient for evaluating acute exposure scenarios. As Appendix D.3 indicates, ATSDR designed its modeling study to estimate both annual average and maximum 24-hour average ambient air concentrations.

  By design, the Navy contractor's modeling study was based entirely on range utilization statistics for 1998, and the amount of ordnance used in 1998 was assumed to be representative of amounts used in previous years (IT 2000). This is a critical assumption, because the model predictions for 1998 will not be representative of other years if the range utilization statistics for 1998 were unusually high or low. To evaluate whether 1998 is an adequate base year for the modeling application, ATSDR thoroughly evaluated range utilization statistics available for the years 1983 to 1998–the time frame for which the most complete statistics are available. An overview of ATSDR's review of those statistics follows:

  • In fiscal year 1998, the Navy used the bombing range on 197 days (see Figure 4). Between fiscal years 1983 and 1998, the Navy used the bombing range, on average, 187 days per year.
  • In fiscal year 1998, the Navy used 458 tons of high explosives on the bombing range (see Figure 5). Between fiscal years 1983 and 1998, the average annual usage of high explosives was 353 tons per year.

  Based on these observations and on the fact that emissions from the bombing exercises depend directly on the number of exercises and amounts of high explosives used, ATSDR finds that 1998 indeed appears to be an adequate base year for the modeling analysis. Because range usage in 1998 was greater than the long-term average, using 1998 as a base years might lead to a slight overestimate of emission rates.

• Comments on approach used to estimate emissions of particulate matter. As Section D.1.A explains, the Navy contractor used a model developed by the Army Research Laboratory to estimate particulate matter emissions from bombing exercises at Vieques. The model–the Combined Obscuration Model for Battlefield Induced Contaminants (COMBIC)–was developed to predict how much smoke and dust certain battlefield activities may emit. Accurate predictions of these emissions are necessary because high levels of smoke and dust can interfere with critical electro-optical systems that the military needs to operate in battlefield environments.

  For several reasons, ATSDR believes COMBIC is an adequate basis for estimating emissions of particulate matter from bombing exercises. First, ATSDR doubts that COMBIC grossly underestimates emissions of dusts and particles, because significant underestimates may cause modelers to reach incorrect conclusions that have potentially serious consequences to military personnel in battlefield environments (e.g., predicting that critical electro-optical equipment will function, when they might not). Further, according to the model documentation, several key input parameters have been established empirically from field studies involving high explosive ordnance, much like the ordnance the Navy uses at Vieques. Moreover, the model predictions have proven consistent with observations documented in other publications.

  Though ATSDR believes COMBIC is a reliable model for estimating particulate emissions from bombing exercises, assumptions made when applying the model may lead to biases in emissions estimates. ATSDR identified the following potential shortcomings in the Navy contractor's application of COMBIC at Vieques:

  • As Appendix D.3 describes further, COMBIC can be used to estimate emissions of different size fractions of particulate matter. The Navy contractor appropriately focused on identifying releases of "small particles," which COMBIC defines as being "less than 10 microns in radius," or presumably particles less than 20 microns in aerodynamic diameter (Army Research Laboratory 2000). COMBIC includes algorithms for quantifying emissions of these small particles in three distinct parts of dust clouds formed in explosions: the fireball, the stem of the fireball, and the "skirt" (i.e., particles released near ground level at the base of the stem). The Navy contractor calculated small particle emissions for the fireball and the stem of the dust cloud, but not for the skirt. According to the COMBIC documentation, the amount of small particles in the skirt is 1.875 times greater than the combined mass of small particles in the fireball and stem of the dust cloud. Therefore, by not considering particles in the skirt of explosions, the Navy contractor's emission rates for PM10 are underestimated by nearly a factor of two.
  • When estimating emissions, the Navy contractor assumed that particulate matter releases from ordnance containing less than 10 pounds of explosives are negligible. According to the 1998 range utilization statistics, the weight of ordnance used that contained less than 10 pounds of explosives was 441 tons–all of which the Navy contractor assumed generates no particulate matter emissions. ATSDR notes that the COMBIC model indicates that particulate matter emissions for high explosives scale with the net explosive weight (in TNT equivalents) raised to the 1.111 power (Army Research Laboratory 1999); the model does not imply that any lower bound threshold determines whether particulate matter emissions occur. Simply stated, ordnance containing less than 10 pounds of explosives will generate particulate matter emissions, though clearly in less quantities than ordnance containing hundreds of pounds of high explosives. In ATSDR's modeling analysis (see Appendix D.3), all ordnance containing high explosives was considered when estimating particulate matter emissions.
  • When computing mass emission rates from crater volumes, the Navy contractor assumed a default soil density of 1.5 kg/m³. ATSDR recently reviewed the results of numerous soil sampling studies at Vieques (ATSDR 2001b) and identified two reports that document soil density measurements from Vieques. One sampling effort documents the soil density in three samples collected at the LIA, with an average density of 1.76 kg/m³ (CH2MHILL 2000). Another report indicates that the average density in the top foot of soils on Vieques ranges from 1.25 kg/m³ to 1.5 kg/m³ (Lugo-López, Bonnet, García 1953). Based on these observations, ATSDR believes the Navy contractor's assumed default density of 1.5 kg/m³ is a reasonable value in the absence of more extensive site-specific data.
• Comments on approach used to estimate emissions of explosion by-products. The Navy contractor used emission factors derived from Bangbox studies to estimate emissions of chemical by-products of bombing activities. These emission factors have been widely used to assess environmental impacts from open burning and open detonation activities. For instance, the Open Burn/Open Detonation Model (OBODM), available from EPA's clearinghouse of dispersion models on the agency's technology transfer network, also estimates air emissions from the Bangbox emission factors. ATSDR acknowledges that the representativeness of static detonation tests to live bombing exercises has not been established. However, source testing (or emissions measurements) during live bombing exercises is an extremely complicated endeavor, given the potential safety hazards associated with placing field surveying equipment in the proximity of bombing targets. In the absence of such source testing results, ATSDR believes the Bangbox emission factors are reasonable indicators of chemical releases from explosions.

  ATSDR further believes the Navy contractor's approach used to select emission factors from the available Bangbox studies was appropriate. For instance, to characterize emissions from air-to-ground exercises, the Navy contractor first identified the subset of Bangbox studies that tested explosives with similar compositions to those used at Vieques, and then selected the highest emission factor for every chemical from the various tests. As a result, the emission factors used are the highest measured releases of chemical by-products from the available Bangbox studies. Moreover, when applying the emission factors to the net explosive weight of explosives in ordnance, the Navy contractor included quantities of aluminum dust in the explosive charge toward the net explosive weight. This approach likely leads to overestimates of organic by-products of explosions, because the aluminum dust is not an explosive chemical that releases energy (and forms organic by-products) during explosions.

  Though the Navy contractor's approach includes assumptions that appear to overstate emissions of explosion by-products, it remains unclear exactly how representative the Bangbox studies are to live bombing exercises. Appendix D.3 discusses this issue further. Ultimately, ATSDR used the same set of emission factors, with one exception, to estimate releases of chemical by-products of explosions. As the exception, the emission factor for 2-nitrophenylamine was apparently transcribed incorrectly in the Navy contractor's modeling analyses. This error caused the Navy contractor to underestimate this chemical's emissions by a small margin (7%).

• Comments on approach used to estimate emissions of metals. The Navy contractor used two sources of information to estimate emissions of metals from bombing exercises. First, emission factors from the Bangbox studies were considered. These emission factors represent the amount of metals detected within the Bangbox following explosion of various types of ordnance. The metals detected in the Bangbox tests presumably originated from casings or impurities in the explosives themselves. Second, the Navy contractor considered compositions data from casings and assumed that the entire metallic portion of the casings vaporizes upon explosions. The casings composition data, however, only account for quantities of aluminum, copper, manganese, and zinc. Combined, these metals comprise roughly 3% of the total casing material. ATSDR's specific comments on the approach used to estimate emissions of metals follows:
  • The Navy contractor's approach does not account for the fact that particulate emissions from craters formed during bombing exercises will include metals that were originally in the soils. Omitting this potential source causes the Navy contractor's modeling analysis to underestimate emissions and ambient air concentrations of metals. ATSDR's previous public health evaluations for Vieques have shown that soils throughout the LIA contain metals (ATSDR 2001b), which can become airborne when craters are formed. As Appendix D.3 indicates, ATSDR's modeling analyses accounts for emissions of metals in crater ejecta.
  • The Navy contractor's emission factor for aluminum assumes that 0.0435 pounds of aluminum are released for every pound of high explosives that is used. This emission factor accounts for aluminum that might be in the casings but does not account for the fact that many types of ordnance used at Vieques contain aluminum dust in the explosive charge. As a result, the Navy contractor may have considerably underestimated emissions of aluminum. ATSDR's modeling analysis considers the entire weight of aluminum in bombs used at Vieques, including amounts in the casing and in the explosive charge.
  • The Navy contractor used emission factors that account for roughly 3% of the metals within bomb casings. Moreover, these emission factors for casings considered only potential releases of aluminum, copper, manganese, and zinc. ATSDR has identified more detailed composition data on bomb casings which identify additional metals that might be released, though in relatively low quantities. Appendix D.3 lists these other metals and their estimated emissions.
• Comments on the approach used to model atmospheric transport. The Navy contractor used the INPUFF dispersion model to predict the fate and transport of chemicals released from the LIA. ATSDR thoroughly reviewed the modeling approach and findings and presents selected comments on this analysis here:
  • Model selection. Since it was designed to model dispersion from sources of instantaneous releases, like an explosion's dust cloud, INPUFF appears to be an adequate model selection for this application. INPUFF does not explicitly account for complex terrain in its simulations. However, because the estimated release heights for all air-to-ground bombing exercises (217 to 324 meters) were higher than the highest local terrain feature (Cerro Matias, 137 meters), use of a simple terrain dispersion model is justified for this type of source.
  • Meteorological data. The Navy contractor processed meteorological data collected at US Naval Air Station Roosevelt Roads for use in the INPUFF modeling analysis. As Appendix D.3 explains further, ATSDR believes this data set is the most representative available information for conducting dispersion modeling at Vieques.
  • Other model inputs. Several other model inputs were specified in the Navy contractor's simulations, including the dimensions and height of the explosion clouds and the locations and elevations of the different receptors. The values selected for the cloud dimensions appear to be consistent with those published in various reports on high explosives, as Appendix D.3 describes. Ambient air concentrations were estimated at receptor locations in the residential areas of Vieques on a very fine grid with 10 meter by 10 meter spacing. This resolution is more than adequate to characterize exposures, especially considering that the source being modeled is several miles from the receptor grid.
• Comments on the presentation and interpretation of results. The Navy contractor estimated annual average ambient air concentrations of all pollutants considered but the summary report does not interpret the significance of the estimates nor does it present estimates of air quality impacts over shorter averaging periods. ATSDR designed its modeling analysis (see Appendix D.3) to provide perspective on the public health implications of exposure, including both acute and chronic exposure scenarios.

D.2 Review of Rafael Cruz Pérez's Modeling Study of Live Bombing Activities (Cruz Pérez 2000)

In 2000, Dimension Magazine, a publication of the College of Engineers and Surveyors of Puerto Rico, released an article written by Rafael Cruz Pérez, PE, about environmental contamination at Vieques (Cruz Pérez 2000). ATSDR has identified additional releases of this article from earlier years, but bases its review of the article on the most recent version. The article summarizes levels of environmental contamination, both measured and modeled, in multiple media, including soil, surface water, groundwater, and air. This review focuses specifically on an air modeling analysis documented in the article of high explosives used at Vieques. Refer to Appendix C.4 and C.5 for ATSDR's review of this article's summary of ambient air sampling on Vieques.

D.2.A Overview of Rafael Cruz Pérez's Modeling Approach

The following paragraphs review three key features of Rafael Cruz Pérez's modeling analysis: how emission rates were estimated, how atmospheric fate and transport was simulated, and how results were presented and interpreted. Refer to Section D.2.B for ATSDR's evaluation of the scientific rigor of this modeling analysis.

• Approach to estimating emissions. The modeling analysis conducted by Rafael Cruz Pérez evaluated potential air quality impacts associated with bombing activities involving one type of ordnance: 105 mm high explosive mortar projectiles. According to Navy ordnance statistics, these projectiles weigh 33 pounds and contain 5.1 pounds of high explosives. When evaluating air quality impacts, the modeling analysis considered emissions of only particulate matter and did not consider emissions of other pollutants that bombing activities release.

  To estimate air emissions, Rafael Cruz Pérez reported that firing a single 105 mm high explosive mortar will displace 400 kg of soil. Of this amount, 80% (or 320 kg) was assumed to fall to the ground immediately in the vicinity of the impact location. The remaining 80 kg of particles that remain airborne were assumed to be available for downwind transport. Rafael Cruz Pérez further estimated that 94% of these remaining airborne particles will fall to the ground within several hundred feet of the impact location. With this assumption, 4.8 kg of the soil particles released are considered available for longer range transport. Information on the assumed particle sizes is not provided. The publication by Rafael Cruz Pérez cites no references for any of the aforementioned assumptions and emissions estimates.

• Approach to modeling atmospheric fate and transport. The article by Rafael Cruz Pérez indicates that estimates of ambient air concentrations at downwind locations were calculated using a dispersion equation, but the equation is not provided. According to other text in the article, the equation assumes that ambient air concentrations of particulate matter are inversely proportional to the downwind distance raised to the 1.5 power. Rafael Cruz Pérez cites a 1976 publication by the Naval Surface Weapons Center as the source of this concentration decay term. ATSDR located this citation, which was released as "preliminary draft" by the Naval Surface Weapons Center in 1978 (Young 1978). The 1978 document, in turn, cites a 1968 publication of the U.S. Atomic Energy Association as the original source of information on the assumed concentration decay being inversely proportional to downwind distance raised to the 1.5 power (Slade 1968). Refer to Section D.2.B for ATSDR's comments on this dispersion algorithm.
• Presentation and interpretation of results. Rafael Cruz Pérez presents estimates of ambient air concentrations in the article for various averaging times, depending on the distance from the LIA. First, Rafael Cruz Pérez reports estimated ambient air concentrations for the scenario of the Navy firing a single 105 mm high explosive mortar. As an example of the results, the article indicates that estimated ambient air concentrations of particulate matter at distances between 3,000 and 4,722 meters from the LIA will be 173 µg/m³, and this concentration is assumed to occur over a duration 10.5 minutes. Further, at distances between 6,000 and 18,900 meters from the LIA, which includes the residential areas of Vieques, the estimated concentration of particulate matter is 33 µg/m³, which is assumed to last for 15.9 minutes. These concentrations represent Rafael Cruz Pérez's estimates of the incremental air quality impact of firing a single mortar and do not include contributions from other sources.

  To predict actual exposure point concentrations, Rafael Cruz Pérez presented several additional data points. First, the article indicates that an exercise involving the use of several 105 mm high explosive mortars can increase ambient air concentrations of particulate matter in the residential areas of Vieques by 98.38 µg/m³, but the article does not present the equations used to estimate this concentration nor does it indicate the averaging time for this reported increase. Next, the article indicates that actual exposure point concentrations would be higher than 197 µg/m³–a level apparently calculated by adding the 98.38 µg/m³ increase in concentration to an assumed background concentration of 99 µg/m³. No averaging period is given for the estimated concentration of 197 µg/m³ or the assumed background concentration. The article concludes by asserting that the estimated concentrations are higher than EPA's primary and secondary air quality standards, which are cited as 75 µg/m³ (annual average) and 60 µg/m³ (highest 24-hour average), respectively. Section D.2.B, below, presents ATSDR's review of Rafael Cruz Pérez's modeling analysis.

D.2.B ATSDR's Review of Rafael Cruz Pérez's Modeling Analysis

As with the Navy contractor's modeling analysis, ATSDR thoroughly reviewed Rafael Cruz Pérez's publication on environmental contamination at Vieques. ATSDR's specific comments on this modeling analysis is presented below, organized by the same three topics presented in Section D.2.A:

• Comments on approach used to estimate emissions of particulate matter. ATSDR cannot critically evaluate the approach used to estimate emissions, because the article by Rafael Cruz Pérez does not provide any references for the main assumptions used in the emissions calculations. Nonetheless, several notable observations can be made from the estimated emission rates. First, ATSDR notes that the Rafael Cruz Pérez study predicts that firing of ordnance containing less than 10 pounds of high explosives can displace considerable amounts of soil. As Section D.1.A indicates, the Navy contractor's modeling analysis assumed that all such ordnance would not generate any particulate matter emissions. ATSDR's modeling analysis (see Appendix D.3), like Rafael Cruz Pérez's study, assumes that all high explosive ordnance generates particulate matter emissions.

  Next, ATSDR notes that Rafael Cruz Pérez's study and the Navy contractor's study are quite similar in terms of estimating the proportion of displaced soil that is available for longer range downwind transport. Specifically, Rafael Cruz Pérez assumed that 1.2% of the soil displaced by an explosion will travel in the plume for long distances, though information on the particle sizes is not provided. The Navy contractor, on the other hand, assumed that 1.007% of the soil displaced will be emitted as PM10 and will remain airborne for long distances. Section D.3.B presents ATSDR's approach to estimating emissions of airborne particles, as well as more detailed information on the particle sizes.

  Finally, to get a sense for how the two modeling studies compare, ATSDR used the Navy's total annual usage of high explosives to extrapolate Rafael Cruz Pérez's emissions estimates to an annual value. To do this calculation, ATSDR noted that Rafael Cruz Pérez's study predicts that 4.8 kg of particulate matter (available for long-range transport) are generated for every 5.1 pounds of high explosives used; further, the Navy's usage of high explosives in 1998 was 771,734 pounds (IT 2000). Assuming, to a first approximation, that particulate matter emissions vary linearly with the amount of high explosives in the ordnance, Rafael Cruz Pérez's emissions estimates imply that the annual releases of particulate matter may be as high as 800 tons per year. This emission rate is 10 times higher than the particulate matter emission rate used in the Navy contractor's modeling analysis. Therefore, the approaches used by the Navy contractor and Rafael Cruz Pérez lead to considerably different emissions estimates. Section D.3.B presents ATSDR's best estimate of particulate matter emissions from military training exercises at Vieques. ATSDR's estimate is higher than the Navy's, and lower than Rafael Cruz Pérez's.

• Comments on the approach used to model atmospheric fate and transport. ATSDR cannot critically evaluate the approach Rafael Cruz Pérez used to model atmospheric fate and transport, because the article does not provide sufficient information (e.g., equations) for ATSDR to reproduce the estimated ambient air concentrations. ATSDR can comment on the general approach, however, which assumed that ambient air concentrations of contaminants decrease by the downwind distance raised to the 1.5 power. This assumed rate of concentration decay is a reasonable first approximation for estimating ambient air concentrations for continuous plumes, but releases from high explosive mortars generate instantaneous plumes. Instantaneous plumes have concentrations that decay more rapidly with downwind distance by virtue of dispersion along the downwind direction, which need not be accounted for with continuous plumes. An expert reviewer of this modeling analysis suspected that concentrations from an instantaneous plume would probably decay with downwind distance raised to an exponent between 2 and 2.5 (Hanna 2001). In other words, concentrations within an instantaneous plume would likely decay much faster than predicted in Rafael Cruz Pérez's article. Consequently, the approach used to estimate dispersion overstates actual concentrations.

  The appropriate value of the concentration decay term notwithstanding, ATSDR emphasizes that Rafael Cruz Pérez's approach to estimating ambient air concentrations likely provides only a very rough approximation of actual air quality. Many years of research have established that atmospheric dispersion is not only a function of downwind distance, but is also a function of atmospheric stability. Further, the approach used by Rafael Cruz Pérez does not account for varying wind speeds, wind directions, mixing heights, and other meteorological phenomena that affect how contaminants move through the atmosphere. As Section D.3.C describes, ATSDR used a model that accounts for how site-specific meteorological conditions at Vieques affect atmospheric fate and transport.

• Comments on the presentation and interpretation of results. The article by Rafael Cruz Pérez presents various ambient air concentrations as results of its modeling analysis. The final analyses in the article presents an estimated concentration (98.38 µg/m³), which is apparently an estimated 24-hour average concentration resulting from the firing of numerous 105 mm high explosive mortars. The article does not describe how this result was calculated and how many mortars were assumed to be fired to generate this level of contamination. To evaluate the significance of this estimate, Rafael Cruz Pérez estimated an exposure point concentration in the residential areas of Vieques by adding the estimated ambient air concentration resulting from the mortar fire (98.38 µg/m³) to an assumed background concentration (99 µg/m³). The article then compares the resulting exposure concentration (197 µg/m³) to EPA's former primary and secondary standards for TSP.

  ATSDR has several comments on the article's interpretation of the estimated ambient air concentrations. First, ATSDR notes that not enough information is provided to evaluate how Rafael Cruz Pérez estimated the increase in particulate matter concentrations resulting from the mortar firing (i.e., 98.38 µg/m³), which is apparently based on a 24-hour averaging period. Nonetheless, the interpretations of this estimated concentration appear to be flawed. Specifically, ATSDR notes that the assumed background concentration used in the article is the highest ambient air concentration of TSP (99 µg/m³) measured in two different studies (see Appendix C.4 and C.5). The background concentration selected is more than twice as high as the average TSP levels that PREQB recently measured at Vieques using rigorous sampling methods. Therefore, the estimated background concentration appears to be more representative of a maximum concentration than of an average concentration.

  More importantly, ATSDR does not believe comparing an estimated 24-hour average concentration to an annual average health-based standard is appropriate. A more appropriate interpretation would compare the estimated 24-hour average concentration (197 µg/m³) to the former 24-hour average health-based standard for TSP (260 µg/m³), assuming the estimated concentrations are indeed total suspended particulates. ATSDR's modeling analysis, documented throughout Appendix D.3, estimates both annual average and maximum 24-hour average concentrations of particulate matter, and compares these estimates to the appropriate health-based standards.

  Finally, to assess whether the predicted air quality impacts are reasonable, ATSDR extrapolated the predicted concentrations to a larger-scale bombing event: firing a single 2,000-pound bomb containing 1,000 pounds of high explosives. To extrapolate to this scenario, ATSDR notes that Rafael Cruz Pérez analysis predicts that firing a single round of ordnance containing 5.1 pounds of high explosives would cause short-term average air concentrations in the residential areas of Vieques to increase by at least 33 µg/m³. To a first approximation, firing a single bomb that contained 200 times as much high explosives would cause approximately a 200-fold higher air quality impact, or a short-term concentration of 6,600 µg/m³. Although extensive sampling data are not available to determine whether or not such predicted concentrations are reasonable, these increases in concentrations, if correct, would likely be associated with significantly impaired visibility throughout the residential areas of Vieques. ATSDR has heard no accounts of such air quality impacts and has not witnessed such effects on visibility during open detonation events involving much more ordnance than a single bomb. These observations, combined with the previous comments, suggest that Rafael Cruz Pérez's modeling analysis may overstate air quality impacts from military training exercises on Vieques.

D.3 ATSDR's Modeling Study of Navy Exercises Using Live Bombs (ERT 2001)

Much of ATSDR's efforts evaluating this site have focused on air quality between the 1970s and 1999–the years when the Navy conducted military training exercises on Vieques using live bombs. Though three parties conducted air sampling projects during this time frame, all of which did not find ambient air concentrations of pollutants at levels above EPA's air quality standards, the quality of the sampling data are not known because original documentation on the sampling projects is limited or not available. As a result, ATSDR used modeling studies to evaluate potential exposures to contaminants released from live bombing activities.

Before estimating emissions and modeling fate and transport, ATSDR first obtained and thoroughly reviewed the two air quality modeling studies that were readily available for Vieques. In so doing, ATSDR not only could build upon the strengths of the work already completed but also could identify and improve upon potential shortcomings noted in Appendix D.1 and D.2. Key features of ATSDR's dispersion modeling analysis are reviewed in the following sections.

D.3.A Goal of ATSDR's Modeling Study

ATSDR designed its modeling study to generate reasonable estimates of how air-to-ground, ship-to-shore, and land-based military activities at Vieques affect air quality in the residential areas of the island. Because this PHA is evaluating potential inhalation exposures, the emphasis in ATSDR's modeling was to make reasonable estimates of ambient air concentrations; characterizing deposition of air particles was not considered in this study, since ATSDR's other PHAs have already addressed (or will soon address) levels of contamination present in other environmental media, including drinking water supplies, soils, and biota. Recognizing that military training exercises at Vieques are not continuous and vary in intensity from one exercise to the next, ATSDR estimated both annual average and maximum 24-hour average exposure point concentrations. These concentrations were then used to evaluate chronic exposure scenarios and acute exposure scenarios, respectively.

The rest of this appendix describes the approaches ATSDR used to estimate emissions from the various military training exercises (Section D.3.B) and to model the atmospheric fate and transport of these emissions (Section D.3.C). Section D.3.D then presents key findings from the modeling analyses. ATSDR's public health interpretations of the modeling results are documented in Sections V.C and V.D.

D.3.B Emissions Estimates

This section describes how ATSDR estimated emission rates from the Navy's military training exercises, including both maximum 24-hour emissions and annual average emissions. Consistent with the goal of the modeling study, ATSDR estimated the combined emissions from the use of high explosives ordnance during air-to-ground, ship-to-shore, and land-based exercises. ATSDR notes that the Navy periodically collects unexploded ordnance from the LIA and destroys the explosive charges in open detonation events. ATSDR's approach to estimating emissions assumed that all ordnance fired on the LIA explodes upon impact. With this approach, performing separate calculations for open detonation events is unnecessary, because ATSDR has already accounted for the potential explosion by-products in its calculations for the bombing exercises.

ATSDR estimated emissions using the range utilization statistics for 1998–the same base year that the Navy contractor used in its modeling analysis (see Appendix D.1.A). ATSDR selected this base year for several reasons, but primarily because 1998 has the most detailed range utilization statistics of all years of data that ATSDR has reviewed. Further, the Navy's use of the range in 1998 is representative of that of previous years. More specifically, ATSDR found that the number of days the Navy used the range in 1998 and the amount of high explosives that were fired on the range in 1998 exceed the long-term average for these parameters over a 16-year period (see Appendix D.1.A). Finally, by using the same base year as the Navy contractor, ATSDR can compare emissions estimates between the studies on the same basis. The remainder of this section describes how ATSDR estimated emissions for different classes of pollutants released during military training exercises:

• Emissions estimates for particulate matter. ATSDR is unaware of any studies that have directly measured the amount of particulate matter that an explosion during a military training exercise releases to the air. As Appendix D.1 noted, measuring emissions from explosions is inherently difficult, because measurement devices cannot easily be placed in close proximity to the site of an explosion. Nonetheless, researchers have long observed explosions and have been able to estimate the amounts of particles ejected by evaluating crater sizes, deposition, and other relevant phenomena. ATSDR's particulate emission estimates were made using the COMBIC model, which Appendix D.1 describes. ATSDR emphasizes that this model has been developed to perform realistic simulations of battlefield scenarios, for which accurate predictions are needed to determine whether critical equipment can function in combat situations. Though the intended application of the model provides confidence that the estimated emission rates will be reasonable, it does not guarantee that the predictions will match actual conditions.

  ATSDR's use of the COMBIC model differs from the evaluation performed by the Navy contractor (see Appendix D.1) in three important regards. First, ATSDR considered emissions to the "skirt" of the explosion cloud, which the Navy contractor did not–a factor that results in approximately a 2-fold difference in the emission rates, all other inputs considered equal. Second, ATSDR assumed that all ordnance used during military training exercises, and not just those with more than 10 pounds of high explosives, generate particulate emissions. Third, although bombs at Vieques are fuzed to detonate on impact, ATSDR assumed that the bombs penetrate the surface, which leads to higher emissions estimates than a surface detonation. Table D-1 lists several key inputs used to, and assumptions made when, estimating emissions of particulate matter. Based on these inputs, including a detailed distribution of high explosive ordnance types (as documented in IT 2000), ATSDR estimated that the military training exercises release 277 tons of particulate matter into the air per year. A much greater amount of soil is displaced during the explosions and falls back to the ground in the immediate vicinity of the craters. ATSDR assumed that the 277 tons of emissions are in the form of PM10, even though the COMBIC model documentation indicates that these particles range in sizes from 0-20 microns. More detailed information on particle size distributions is not available.

  To estimate maximum daily particulate emissions, ATSDR reviewed nearly 6 years of daily range utilization statistics to characterize the most intense bombing activity over a 24-hour time frame. Only 6 years were considered because only annual range utilization statistics are available for other years. The daily bombing activity selected to calculate the highest 24-hour average emission rate occurred in October 1995, when 94.5 tons of ordnance containing 39 tons of high explosives were used on a single day. Based on this level of activity and the assumption that the distribution of ordnance types used was the same as the annual average, ATSDR estimated the daily worst-case emission rate to be 28 tons of PM10 released on this one day identified as being representative of the most intense military training exercises. The emission rate for this one day should not be viewed as being representative of typical conditions. In fact, the range utilization statistics indicate that less than 10 tons of ordnance were fired per day of military training exercises scheduled.

  ATSDR acknowledges that these estimated emission rates have inherent uncertainties, and the actual emission rates may be higher or lower than the levels calculated. The estimated emission rates used in these analyses are believed to be based on the best information currently available. Though predicting the amount of emissions from a single explosion is extremely difficult, due to the variability in blast behavior and soil properties from one event to the next, the COMBIC model is designed to given reasonable predictions for a series of events, such as those that occur over a year or a day of intense activity.

• Emissions estimates for explosion by-products. ATSDR estimated emission rates for chemical by-products of explosions using BangBox emission factors, which Appendix D.1 describes. ATSDR's approach is nearly identical to that used by the Navy contractor in its modeling analysis. The BangBox emission factors are also documented in OBODM (Bjorklund et al. 1998), which is the only atmospheric dispersion model on EPA's Support Center for Regulatory Air Models designed specifically to characterize emissions of explosion by-products. These emission factors are believed to be the best information currently available, and arguably the most widely used basis, for estimating air emissions from detonations of high explosives.

  When selecting emission factors, ATSDR first identified all of the BangBox studies that considered high explosives of similar composition to those the Navy used at Vieques, namely those that contain some combination of TNT, RDX, and aluminum powder. From this set of studies, the highest emission factor was selected for this modeling analysis for every chemical measured. Table D-2 lists the chemicals by-products of explosions that ATSDR considered, along with their emission factors, emission rates, and estimated annual average air concentrations. Section D.3.D comments on the uncertainties associated with the data presented in Table D-2.

• Emissions estimates for metals. ATSDR identified four different ways that metals may be emitted to the air during military training exercises: bomb casings may vaporize, trace metals in the explosive mixture may be released, larger amounts of aluminum in the high explosive charge may be released, and soils that contain metals may be ejected into the air. ATSDR notes that the Navy contractor's modeling analysis did not consider at least two of these factors contributing to air emissions. Approaches ATSDR used to represent these different factors follow:
  • Metals released from bomb casings. ATSDR reviewed data provided by the Navy on the composition of metals in the bomb casings. This data indicated that the following metals were present in some types of bomb casings at the concentrations specified: aluminum (5.6%), boron (0.0002%), chromium (0.02%), copper (2.35%), iron (93.11%), manganese (1.82%), molybdenum (0.001%), nickel (0.01%), titanium (0.01%), and zinc (0.45%). The estimated weight of the casings was calculated as the difference between the total amount of ordnance used in 1998 (1,295 tons) and the total amount of high explosives within the ordnance (386 tons). Metals emissions were calculated by multiplying the composition by the total weight of the casings. ATSDR conservatively assumed that the entire casings are vaporized in every explosion. This assumption clearly overstates emissions, because fragments of casings have remained on the ground after most military training exercises, including those that used live bombs.
  • Trace amounts of metals in the high explosive mixture. The BangBox studies reviewed for this public health assessment include emission factors for 14 metals: aluminum, antimony, barium, cadmium, calcium, chromium (trivalent and hexavalent), copper, lead, mercury, nickel, potassium, sodium, titanium, and zinc. These metals were presumably present in the casings or the high explosive mixture tested in the BangBox studies. Even if their origin was the casings, which were addressed separately in the emissions calculations, ATSDR considered the BangBox emission factors to estimate releases. The emission factors for these metals can be obtained from the OBODM model (Bjorklund et al. 1998).
  • Aluminum in the high explosive charge. According to the Navy's bomb composition statistics, the high explosive charges for the ordnance most commonly used at Vieques contained varying amounts of organic compounds (typically a mixture of TNT and RDX) and aluminum powder. The highest composition of aluminum powder in the bombs most commonly used was 21%. ATSDR assumed that this amount of aluminum powder was present in all rounds of ordnance fired and that all of the powder was emitted as PM10.
  • Metals in soils. The soils at the LIA contain naturally-occurring metals as well as metals contamination. To estimate the amount of metals released in crater ejecta, ATSDR multiplied the particulate emission rates by the average metals concentrations in the LIA soils (see Table 4).

  Table D-3 lists the annual emission rates that ATSDR calculated for metals, organized by the four different factors that contribute to these emissions. The table also lists the estimated annual average air concentrations. Section D.3.D comments on the uncertainty associated with the metals emissions estimates.

• Emissions estimates for explosives. Range utilization statistics indicate the total weight of high explosive charges in the ordnance used at the LIA (e.g., 386 tons in 1998). Further, ordnance composition data compiled by the Navy characterize the typical chemical composition of these high explosive charges. In recent years, these have been composed primarily of TNT and RDX; aluminum powder is also found in considerable quantities in these charges, but emissions of aluminum were calculated with those for the other metals. To estimate air emissions of the organic high explosives (TNT and RDX), ATSDR multiplied the weight of the high explosive charge by the maximum composition of the individual constituents.

  Approximately 93% of the high explosive material used during the base year was from three different types of air-to-ground ordnance, which contain TNT and RDX at concentrations up to 80% and 45.1%, by weight. ATSDR used these maximum levels to estimate the total quantity of these chemicals in the charges, even though both chemicals clearly cannot be present at these concentrations in the same mixture. The remaining 7% of high explosive material has widely varying compositions. Rather than calculating the quantities of each component in the charges, ATSDR instead calculated a single emission rate for "all other" high explosive chemicals.

  After calculating the amounts of chemicals present in the charges, ATSDR then estimated the proportion of the high explosives that are consumed during the detonation. Although destruction efficiencies for high explosives have not been measured for live bombing exercises, ATSDR notes that the BangBox emission factors suggest that open burning and open detonation activities are typically more than 99% efficient at destroying organic high explosive chemicals. High destruction efficiencies are assumed to apply to the military training exercises at Vieques, primarily because rapid destruction of the charge is needed for ordnance to be effective. The fact that only trace amounts of high explosive chemicals remain in the LIA soils (ATSDR 2001b) is consistent with the assumed high destruction efficiency. To calculate emissions for the dispersion modeling analysis, ATSDR assumed that 10% of the organic chemicals in high explosive charges are emitted. In other words, ATSDR assumed that the explosions have a 90% destruction efficiency for the organic chemicals in the charges.

  ATSDR's emissions estimates for high explosives, along with the estimated ambient air concentrations that result from the modeling analysis, are summarized below:

  Chemical	Annual Amount Used	Estimated Air Concentration
  TNT	31 tons/year	0.003 µg/m3
  RDX	19 tons/year	0.002 µg/m3
  All others	<2.8 tons/year	<0.0003 µg/m3

D.3.C Atmospheric Fate and Transport

ATSDR used the CALPUFF dispersion model to evaluate the atmospheric fate and transport of air emissions. This model was selected because it has been designed to assess many types of sources, including non-continuous (or "puff") sources, and can also assess deposition, which other "puff" models (like INPUFF) cannot do. The modeling was performed using CALPUFF Version 5.5, Level 010730_1. The following paragraphs describe key inputs selected for this application; a complete listing of these inputs is available in the final modeling report (Trinity Consultants 2002):

• Source parameters. All emissions were assumed to originate from the geographic center of the LIA (coordinates: 257.748 km East, 2,006.944 km North, Zone 20). This choice is considered acceptable because ordnance is likely to impact many different locations at the LIA. The emissions clouds generated during explosions were modeled as elevated volume sources. Three different sets of source parameters were used, corresponding to cloud heights predicted for air-to-ground exercises using 500-lb, 1,000-lb, and 2,000-lb bombs (IT 2000). These three bombs account for more than 90% of the high explosive ordnance used during the base year. The emissions were represented as puff releases with a diurnal emissions profile: emissions were set to zero between 11:00 PM and 7:00 AM every day. This diurnal profile reflects the times of day when the Navy used live bombs prior to 1999 (IT 2000). The center of the cloud heights varied from 285 to 424 meters, and the initial lateral dimensions from 44 to 66 meters. These values were used in the Navy contractor's dispersion modeling analysis, and are based on observations of explosion characteristics made by the former Defense Nuclear Agency. Unit emission rates were used in the model.
• Meteorological data. CALPUFF can use three dimensional meteorological fields when extensive meteorological data are available, particularly for multiple sites. Since the data needed to generate these meteorological fields are not available for Vieques, CALPUFF was run using meteorological data like that compiled for running EPA's Industrial Source Complex (ISC) models. The meteorological data were based on surface measurements taken at the U.S. Naval Station at Roosevelt Roads, Puerto Rico, upper air measurements from San Juan, Puerto Rico, and precipitation data from Fajardo, Puerto Rico. Data were processed for the years 1985, 1985, 1989, 1990, and 1991, such that ATSDR's results could be compared directly to those of the Navy contractor. Missing surface data observations were relatively few (less than 100 hours missing per year), and were filled according to EPA guidance. Missing data periods of greater than 5 hours were left as missing in the model ready files.
• Receptor locations. The model was run with a computational grid that spanned 100 km by 100 km. The receptor domain was limited to the residential areas of Vieques. In this area, ambient air concentrations were predicted for a receptor grid with 100 meter spacing. Receptors were also placed at 100 meter spacing along the boundary that separates the residential area on Vieques from Navy property. Terrain elevations were input to the model by interpolating from Digital Elevation Model data obtained from the U.S. Geological Survey.
• Run options. Gaseous and particulate emissions were modeled separately, given their different deposition properties. Particulates were modeled assuming that they were all present as PM10. CALPUFF's default deposition parameters were selected for all events. Liquid (0.00066 1/s) and frozen (0.00022 1/s) wet scavenging coefficients were selected for the particulate emissions; these were taken from the most recent User's Guide for the Industrial Source Complex models. Runtime options typical of regulatory applications were selected for all other parameters. Complex terrain was not considered in these evaluations because the estimated initial cloud heights were greater than the elevations of the local terrain features.
• Outputs. Normalized concentrations were calculated for several scenarios. For every year of meteorological data considered, and for each of the three different cloud types modeled, annual average and maximum 24-hour average normalized concentrations were calculated for particles (with deposition algorithms "on"), and annual average and maximum 24-hour average normalized concentrations were calculated for gaseous contaminants (with deposition algorithms "off"). The modeling results, and associated uncertainties, are summarized in the following section.

D.3.D Results

Modeling results were reported as normalized concentrations, based on unit emission rates (Trinity Consultants 2002). For all three initial cloud dimensions considered, the highest normalized concentrations occurred for receptors along the property line that separates the residential areas of the island from Navy property. These receptor locations are at least 1 mile upwind from the most heavily populated areas on Vieques.

At the location with highest predicted air quality impacts, the annual average normalized concentrations varied with initial cloud height and the year of meteorological data considered. Table D-4 summarizes the main model outputs for the various scenarios considered. The modeling results showed that concentrations did not change dramatically with initial cloud height, as annual average ambient air concentrations varied by less than a factor of two between the 500-pound and 2,000-pound bombing events, whose initial cloud heights differ by 160 meters.

The approach used to calculated air concentrations from the normalized concentrations depends on the averaging time and contaminant of concern. The normalized concentrations for particles (i.e., considering deposition) were used to estimate air concentrations for both metals and particulate matter, while those for vapors (i.e., not considering deposition) were used to estimate air concentrations for chemical by-products of explosions and high explosive chemicals. The highest daily emission rate was multiplied by the 24-hour maximum normalized concentrations when assessing worst case air quality impacts over the short term. This approach assumes that the most intense bombing activity occurred on the day that had the least favorable meteorological conditions–an unlikely scenario, but one that helps ensure that the modeling analysis does not underestimated 24-hour average concentrations. To calculate annual average air concentrations, the annual average emission rates were multiplied by the corresponding annual average normalized concentrations.

As acknowledged throughout this section, air dispersion modeling analyses have inherent uncertainties and limitations, and the concentrations predicted in this analysis may be higher or lower than the actual impacts that occurred on Vieques during military training exercises with live bombs. Specific comments on uncertainties associated with individual contaminants follow:

• Metals. The approach used to estimate emissions for metals is believed to be an upper bound estimate of actual emissions. That is, the amount of metals released to the air is likely not higher than the amount of metals in the casings, in the high explosive charge, and in the soils ejected into the air. Although predicting crater ejecta arguably involves the greatest uncertainty, assumptions that the entire bomb casings vaporize and that all of the aluminum powder in high explosive charges is emitted are conservative. As a result, ATSDR has confidence that the metals emissions data and estimated air concentrations are reasonable and do not understate the actual amounts that military training exercises contributed to air quality.
• Organic by-products of explosives. The BangBox emissions studies are widely used to characterize emissions from detonations involving high explosives. The extent to which results from these highly-controlled studies represent conditions during military training exercises is not known. However, ATSDR notes that the predicted ambient air concentration for every by-product considered was more than three orders of magnitude lower than health-based comparison values. Given this substantial difference between predicted concentrations and the concentrations that would require further evaluation, ATSDR again has confidence that the model predictions are a sound basis for making public health conclusions, even if the BangBox emissions studies do not perfectly replicate conditions in the field.
• Explosives. The modeling analysis assumed that every high explosive charge contains 80% TNT and 41.5% RDX, which are the highest cited concentrations from bomb composition data. These assumed compositions clearly overstate the total amount of high explosives released to the air. The percentage of organic high explosives that are destroyed in bombs used during military training exercises is not known. As a first approximation, based loosely on destruction efficiencies reported for open detonation events, ATSDR assumed that 90% of the organic high explosives are destroyed when ordnance is fired on the LIA. ATSDR acknowledges that the estimated air concentrations are highly dependent on this assumed destruction efficiency. However, even if ATSDR assumed that only 10% of the explosives were destroyed (an unrealistically low number), the estimated ambient air concentrations of TNT and RDX would still be below health-based comparison values. As a result, ATSDR has confidence that the conclusions made for high explosives are appropriate.
• Particulate matter. The uncertainty involved in estimating particulate matter emissions is arguably the greatest, and is also most difficult to interpret. The fact that the predicted increase in annual average PM10 concentrations (0.04 µg/m³) is at least two orders of magnitude lower than levels of health concern is reassuring. Moreover, the fact that the air sampling studies from the 1970s, though of questionable quality, did not report particulate concentrations greater than EPA's air quality standards also provides some level of comfort that the estimated concentrations do not grossly underestimate actual air concentrations.

Notes:
- Emission factors and emission rates listed are for air-to-ground activities only. ATSDR used different sets of emission factors for ship-to-shore and land-based activities, but these activities consistently accounted for approximately 5% of the total concentrations and are not summarized in this table.
- The ambient air concentration listed is for the location in the residential area of Vieques found to have the highest air quality impacts from the military training exercises. The concentrations reflect contributions from all three types of military training exercises.

Note:
- Section D.3.B discusses the assumptions made to estimate the emission rates for metals. Several assumptions are highly conservative (e.g., the casings from all high explosives completely vaporize upon impact) and most likely cause these emissions estimates to overstate actual emissions levels.

Note:
- The annual average normalized concentrations are averages of the annual average concentrations output for the five different years of meteorological data; the 24-hour average normalized concentrations are the highest daily-average level predicted for the five years of meteorological data.