Comparison of stack measurement data from R&D facilities to regulatory criteria: A case study from PNNL

Abstract

Chemical emissions from research and development (R&D) activities are difficult to estimate because of the large number of chemicals used and the potential for continual changes in processes. In this case study, stack measurements taken from R&D facilities at Pacific Northwest National Laboratory (PNNL) were examined, including extreme worst-case emissions estimates and alternate analyses using a Monte Carlo method that takes into account the full distribution of sampling results. The objective of this study was to develop techniques to estimate emissions from stack measurement data that take into account a high degree of variability in the actual emissions. The results from these analyses were then compared to emissions estimated from chemical inventories. Results showed that downwind ambient air concentrations calculated from the stack measurement data were below acceptable source impact levels (ASILs) for almost all compounds, even under extreme worst-case analyses. However, for compounds with averaging periods of a year, the unrealistic but simplifying extreme worst-case analysis often resulted in calculated emissions that were above the lower level regulatory criteria used to determine modeling requirements or to define trivial releases. Compounds with 24-hr averaging periods were nearly all several orders of magnitude below all, including the trivial release, criteria. The alternate analysis supplied a more realistic basis of comparison and an ability to explore effects under different operational modes.

Implications:

Air emissions from research operations are difficult to estimate because of the changing nature of research processes and the small quantity and wide variety of chemicals used. Stack measurements can be used to verify compliance with applicable regulatory criteria. This study shows that while extreme worst-case assumptions can be used for a relatively simple initial comparison, methods that take into account the full range of measurement data are needed to provide a more realistic estimate of emissions for comparison to regulatory criteria, particularly those criteria that define trivial levels of environmental concern.

Introduction

Title V of the Clean Air Act Amendments (42 U.S. Code 7661; U.S. Code, 1990) established in 1990 creates an operating permit program for federal, state, and local authorities to regulate air pollution emissions from large sources. Research and development (R&D) facilities are not currently regulated as a unique category, and no guidance has been developed to estimate emissions from R&D activities. However, these facilities may become major sources as stand-alone major sources (emissions exceed hazardous air pollutant [HAP] major source thresholds), or they may need to be incorporated into a Title V permit if they are collocated with a manufacturing site that is considered to be a major source. The facilities also can be excluded based on either actual emission threshold rates or individual size and production rate criteria. Emissions of hazardous air pollutants from R&D facilities are difficult to estimate because these facilities typically use a large number of chemicals in small quantities, they engage in numerous and diverse activities, and chemicals and activities change over time. The diversity and variability of R&D processes result in emissions that are unique and different from large manufacturing plants, which typically have larger quantities and a smaller number of chemicals, in addition to stable processes that do not change over time.

In addition to federal regulations, some states have regulations that apply to air chemical emissions. Applicability to R&D facilities varies depending on the state and/or interpretation of the enforcing authority. In Washington State, these regulations include General Regulations for Air Pollution Sources (Washington Administrative Code [WAC] 173-400), Operating Permit Regulation (WAC 173-401), and Controls for New Sources of Toxic Air Pollutants (WAC 173-460), all of which contain requirements for controlling the emissions of a specific list of toxic air pollutants (TAPs), as well as addressing the criteria pollutants of particulate matter, carbon monoxide, nitrogen dioxide, sulfur dioxide, lead, and ozone in the air.

The list of 400 TAPs includes carcinogens and noncarcinogens and three associated criteria for determining new source review requirements: acceptable source impact level (ASIL), small quantity emission rate (SQER), and de minimis emission threshold. ASILs are screening concentrations of the pollutants in the ambient air; SQERs are emission rates below which dispersion modeling is not required to demonstrate compliance with ASILs; and de minimis emission thresholds are emission rates not subject to new source review. For carcinogenic compounds, the ASIL is an annual average concentration (μg/m³) based on an increased cancer risk of 1 in 1 million, while for noncarcinogenic compounds the 24-hr ASIL is based on a threshold concentration for toxic effects.

The Washington State Title V regulation allows emissions to be evaluated against a list of “thresholds for hazardous air pollutants” to determine whether an emission unit or activity within a permit can be considered insignificant. The list of threshold values is from WAC 173-401, while the list of TAPs and their associated ASIL, SQER, and de minimis values are from WAC 173-460. Many but not all of the same compounds are on both lists.

Pacific Northwest National Laboratory (PNNL) sampled stack concentrations from R&D laboratory buildings during 1998–2008 to obtain data to evaluate emission estimation methods and to identify and quantify air toxics emitted for comparison with compliance limits established by Washington State. Just as many different and changing processes in an R&D facility make it difficult to estimate emissions, the variability of measured emissions also makes it difficult to compare with compliance limits for specific averaging periods. This study applies three methods using measured stack concentration data from R&D facilities at PNNL to estimate emissions for comparison with regulatory criteria. The three methods considered are (1) an extreme maximum emissions estimate, (2) a Monte Carlo simulation method that considers the full distribution of stack sampling results, and (3) an inventory-based method. The data set of measured stack concentrations from R&D facilities is unique and this is currently the only known study of its kind.

Methods

Stack sampling

PNNL sampled air chemical emissions from the stacks of four buildings during 1998–2008: a Life Sciences Laboratory (331 Building), a Chemical Sciences Laboratory (329 Building), a Radiochemical Processing Laboratory (325 Building), and an Environmental Molecular Sciences Laboratory (EMSL). A picture of the 331 Building and closeup of the final exhaust stack are provided in Figure 1 to show the general geometry for one of the buildings. The four buildings and exhaust stacks had a range of sizes, geometries, and distances to the nearest public receptor. For example, the 325 Building has a tall vertical stack with an elevation more than twice that of the building compared to the mostly horizontal 331 Building stack.

Figure 1. The 331 Building Life Sciences Laboratory (top) and closeup of final exhaust stack (bottom).

Samples were extracted over a 100-min time period during normal workday hours (i.e., 8:00 a.m.–5:00 p.m.), except for the initial year of sampling, during which 300- and 50-min samples were obtained. Usually only one sample was taken in a given day, although occasionally both a morning sample and an afternoon sample were taken. A limited number of samples were acquired during nights and weekends to represent off-shift times when R&D activities are not likely to occur. Thus, sampling results are identified as on-shift and off-shift. Table 1 summarizes the number of sampling events representing on-shift and off-shift activities for each building and year sampled. Actual emissions were sampled less than 1% of the time.

Table 1. Number of sampling events

	325		329		331		EMSL
Year(s) sampled	Off-shift	On-shift	Off-shift	On-shift	Off-shift	On-shift	Off-shift	On-shift
1998	1	3	3	9		8
1999–2000		24		18		22
2001							3 (Stack 3)	17 (Stack 3)
							1 (Stack 4)	18 (Stack 4)
							2 (Stack 5)	13 (Stack 5)
2002	1	12	1	14	1	16
2003						23
2005				27
2006					3	33
2007	1	35
2008					1	34
Total	3	74	4	68	5	136	6	48

Although samples were mostly obtained from exhaust stacks, early sampling campaigns included locations such as lobbies and corridors to evaluate non-research-related contributions to emissions. Data from these alternate locations were compared with off-shift samples to evaluate whether off-shift concentrations were similar to nonresearch emissions or whether they contained additional emissions from research being conducted in off-normal hours.

The sampling method used throughout all sampling campaigns was the collection of air samples onto triple-sorbent traps that were subsequently analyzed for volatile organic compounds (VOCs) using gas chromatography/mass spectrometry (GC/MS) analysis. The sampler drew air simultaneously through two tubes in parallel flow paths with flow rates independently adjusted for each path as based on the U.S. Environmental Protection Agency (EPA) Compendium Method TO17 (EPA, 1999a). This method provides guidelines for collection of ambient air toxic organic compounds on a variety of different sorbent media. All samples were analyzed by thermal desorption and GC/MS using procedures consistent with the mass spectrometry and quality assurance/quality control guidelines set forth in EPA Compendium Method TO-15 (EPA, 1999b).

Depending on the year, in total 46–49 target compounds (Table 2) were analyzed by GC/MS. The compounds were selected primarily based on a standard containing a mixture of 39 compounds specified in EPA Compendium Method TO-14 (EPA, 1989), plus a short list of supplementary analytes in a second standard. Depending on the building, the number of stack samples taken varied from 54 to 141 with a total of 344 samples. Thirty-two of the compounds were listed as TAPs in WAC 173-460-150 with associated ASIL ambient air concentrations, SQERs, and de minimis emission rates. The averaging periods for the sampled compounds were either 1 yr (19 compounds) or 24 hr (13 compounds), representing carcinogenic and noncarcinogenic effects, respectively. ASILs are established as described in WAC 173-460-110 using a risk-based calculation and a criterion of an increased cancer risk of 1 in 1 million or threshold-based determinations as described in the regulation.

Table 2. Target compounds and associated regulatory criteria

CAS number	Compound name	State TAPs ^a			Title V ^b
		ASIL	SQER	De minimis	Threshold
		(μg/m^3)	(lb/avg period)	(lb/avg period)	(ton/yr)
Target compounds listed as TAPs in WAC 173-460 with year averaging period
79-34-5	1,1,2,2-Tetrachloroethane	0.0172	3.3	0.165	0.15
79-00-5	1,1,2-Trichloroethane	0.0625	12	0.6	0.5
75-34-3	1,1-Dichloroethane	0.625	120	6	0.5
106-93-4	1,2-Dibromoethane	0.0141	2.71	0.135	0.05
107-06-2	1,2-Dichloroethane	0.0385	7.39	0.369	0.4
106-99-0	1,3-Butadiene	0.00588	1.13	0.0564	0.035
106-46-7	1,4-Dichlorobenzene	0.0909	17.4	0.872	0.5
107-05-1	3-Chloropropene	0.167	32	1.6	0.5
75-05-8	Acetonitrile	60	1.15E+04	576	0.5
71-43-2	Benzene	0.0345	6.62	0.331	0.5
56-23-5	Carbon tetrachloride	0.0238	4.57	0.228	0.5
67-66-3	Chloroform	0.0435	8.35	0.417	0.45
100-41-4	Ethylbenzene	0.4	76.8	3.84	0.5
87-68-3	Hexachloro-1,3-butadiene	0.0455	8.73	0.437	0.5
75-09-2	Methylene chloride	1	192	9.59	0.5
78-87-5	Propane, 1,2-dichloro-	0.1	19.2	0.959	0.5
127-18-4	Tetrachloroethylene	0.169	32.4	1.62	0.5
79-01-6	Trichloroethene	0.5	95.9	4.8	0.01
75-01-4	Vinyl chloride	0.0128	2.46	0.123	0.1
Target compounds listed as TAPs in WAC 173-460 with 24-hr averaging period
71-55-6	1,1,1-Trichloroethane	1000	131	6.57	0.5
75-35-4	1,1-Dichloroethene	200	26.3	1.31	0.2
98-82-8	1-Methylethylbenzene	400	52.6	2.63	0.5
78-93-3	2-Butanone	5000	657	32.9	0.5
74-83-9	Bromomethane	5	0.657	0.0629
108-90-7	Chlorobenzene	1000	131	6.57	0.5
75-00-3	Chloroethane	3.00E+04	3940	197	0.5
74-87-3	Chloromethane	90	11.8	0.591	0.5
67-56-1	Methanol	4000	526	26.3	0.5
95-47-6	o-Xylene	221	29	1.45
106-42-3	p/m-Xylene	221	29	1.45
100-42-5	Styrene	900	118	5.91	0.5
108-88-3	Toluene	5000	657	32.9	0.5
Target compounds previously listed as TAPs in WAC 173-460 with 24-hr averaging period
76-13-1	1,1,2-trichloro-1,2,2-trifluoroethane	2.7E+04	3547.8	177.39
120-82-1	1,2,4-Trichlorobenzene	120	15.77	0.7884	0.5
76-14-2	1,2-dichloro-1,1,2,2-tetrafluoroethane	2.3E+04	3022.2	151.11
95-50-1	1,2-Dichlorobenzene	1000	131.4	6.57
67-64-1	Acetone	5900	775.26	38.763
75-71-8	Dichlorodifluoromethane	1.6E+04	2102.4	105.12
64-17-5	Ethanol	6300	827.82	41.391
109-66-0	Pentane	6000	788.4	39.42
75-69-4	Trichlorofluoromethane	1.9E+04	2496.6	124.83
Target compounds not currently or previously listed as TAPs in WAC 173-460
CAS number	Compound name	CAS number		Compound name
541-73-1	1,3-Dichlorobenzene	108-67-8		1,3,5-Trimethylbenzene
10061-01-5	cis-1,3-Dichloropropene	611-14-3		1-Ethyl-2-methylbenzene
10061-02-6	trans-1,3-Dichloropropene	622-96-8		1-Ethyl-4-methylbenzene
95-63-6	1,2,4-Trimethylbenzene	156-59-2		cis-1,2-Dichloroethene
Notes: ^aWAC 173-460-150; ASIL applies to ambient air concentration and the other criteria apply to stack emissions; significant figures are the same as those given in the regulation;
^bWAC 174-401-531.

Most of these compounds were also listed in the table of Title V Thresholds for hazardous air pollutants in WAC 173-401-531. The current list of TAPs has been in place since June 2009, at which time WAC 173-460 was revised to reflect current toxicology data, with some compounds previously listed as TAPs removed (Washington State Department of Ecology, 2008). For purposes of this evaluation, delisted chemicals were assigned their previous ASIL as surrogate values for comparison and provided evaluation criteria for an additional nine compounds.

Extreme maximum emissions estimate method

A worst-case scenario is assumed to compare emissions based on measured stack concentrations with regulatory criteria. This worst case uses the single highest stack concentration measured at each building to calculate maximum 24-hr or annual emissions for each target compound. These maximum emissions and maximum AERMOD dispersion factors (poorest dispersion) are used to calculate ambient concentrations to compare with ASILs. Equations and assumptions used are described in this section. This approach was applied to all target compounds with 24-hr and year averaging periods for all four buildings.

ASILs are ambient air concentrations at locations where there is no restriction or control of public access. Thus, a dispersion factor is needed to calculate ambient air concentrations from the measured stack concentration data as shown in eq 1:

(1)

where C_ambient is the receptor ambient air concentration in μg/m³, C_stack the stack concentration in mg/m³, V_stack the stack volumetric flow rate in m³/s, and DF the dispersion factor, the ambient air concentration at a specific location per unit of stack emission rate, calculated with units of μg/m³ per g/s. The location and time of the maximum will vary with meteorology and emission parameters. Dispersion factors used were the maximum single annual and 24-hr averages in 5 years at any location.

Alternately, a stack concentration corresponding to the ASIL can be calculated by rearranging eq 1 as shown in eq 2:

(2)

where ASILis in μg/m³; ASILs based on 24-hr and year averaging periods were evaluated.C_ASIL is the stack concentration in mg/m³ resulting in the air concentration equaling the ASIL at the specific location where the DF applies.

Dispersion factors were calculated for both annual and 24-hr average concentrations using the AERMOD atmospheric dispersion modeling system. Developed by a collaborative team of the American Meteorological Society and EPA, AERMOD has been adopted as the EPA's preferred regulatory model for both simple and complex terrain and is promulgated as the preferred regulatory model (EPA, 2012).

Uncertainties in air dispersion models including AERMOD have been the subject of other studies (e.g., Auld et al., 2003; Irwin and Hanna 2005: Hanna et al., 2007; Poosarala et al., 2010). The studies indicate that uncertainties are up to a factor of two when considering meteorological input and dispersion parameters. Uncertainties in emissions input have been reported to contribute more to the total uncertainty than do the uncertainties of the of the meteorological input and dispersion parameters (Hanna et al., 2007). The focus of this analysis is on emissions and not on the uncertainties is the dispersion factor calculated using AERMOD.

AERMOD dated 11103 (the latest version available at the time the analysis was performed) was used in this application to calculate dispersion factors for each of the sampled buildings with inputs that included stack height, and exit gas temperature, as well as velocity, inner stack diameter, and building and emission point location/dimensions with respect to the surrounding area. All of these values were relatively constant over time for each emission point. Distances from the stacks to the nearest facility boundary ranged from 190–400 m and AERMOD identified the location outside of these boundaries that resulted in the highest dispersion factor (poorest dispersion). Five years of meteorological data were used as input to AERMOD to provide the range of dispersion conditions with which to evaluate impacts. Meteorological data for 2004–2008 were chosen because those years corresponded with the most recent sampling campaigns. Although there is year-to-year variability in the hourly data sets, a 5-yr period captures most of this variability. Wind roses for other time periods outside of this study were examined and found to be comparable to those used in this study.

AERMOD is a straight-line plume model that utilizes a single surface/upper-air station in its calculations. A composite surface data set was generated that used wind data from the 300 Area (Hanford station #11) and atmospheric state variables (i.e., temperature, pressure, humidity) from the main Hanford measurement station (Hanford station #21). The 300 Area winds are most representative of all four buildings, three of which are located in the 300 Area and the fourth located nearby in the PNNL campus, whereas the atmospheric state variables are fairly consistent across the entire Hanford site. Twice-daily upper-air data are for the Spokane airport for the same time period.

For the worst-case scenario, emissions were modeled in AERMOD as a continuous (24 hr/day × 7 days/wk × 365 days/yr) source at a constant rate. The resulting single highest 24-hr and single highest annual average dispersion factors were used in eqs 1 and 2. An alternate operating scenario was modeled to investigate the changes in the effective dispersion factors for emissions that are not constant. The source input for this alternate operating scenario was a constant source during normal daytime hours, 8:00 a.m.–5:00 p.m. Monday through Friday (on-shift), and a reduced source at all other times (off-shift). The off-shift source was set to 10% of the daytime source. Although R&D activities are not likely to be constant or continuous even during on-shift hours, the alternative operating scenario quantifies some of the conservatism in operating mode assumptions.

Ambient air concentrations were calculated for the two scenarios using the single highest 24-hr and annual average dispersion factors from the 5-yr data set for each building coupled with the single highest 100-min concentration measured in the stack. No attempt was made to match the sampling time and date to a specific dispersion factor calculated for that time period. While unrealistic, this approach provides a basis for comparing scenarios and a screening comparison with corresponding ASIL values given in Table 2. In addition, the stack concentrations corresponding to ASIL values were calculated as shown in eq 2 and compared to the full set of sampling data, again using the single highest 24-hr and annual dispersion factor based on AERMOD modeling of the 5 years of meteorological data.

In addition to ASILs, which are screening concentrations of toxic air pollutants in the ambient air, emission rates in units of total emission for a given time period have regulatory significance. SQERs are rates below which dispersion modeling is not required to demonstrate compliance with ASILs; de minimis emission rates are trivial rates that are of no regulatory concern; and Title V threshold emission rates can be used to determine whether an emission unit or activity can be considered insignificant for Title V permitting.

Measured stack concentrations can be converted to emission rates by multiplying stack concentration by volumetric flow rate, which was relatively constant for the stacks sampled. Although R&D emissions are highly variable as demonstrated by the measured stack concentration data, an analysis is presented assuming continuous emissions at a constant rate equal to the single highest measured sample concentration in ten years for each compound. The stack concentrations that correspond to SQER and de minimis values can be back-calculated similarly to eq 2 for ASILs, except that no dispersion factor is needed and different conversion factors are required because SQER and de minimis are in units of pounds per averaging period. The stack concentrations that would correspond to SQER and de minimis rates for a source with continuous emissions at a constant rate are presented for comparison to the stack sampling data.

Assumptions used in the Extreme Maximum Emissions Estimate Method had the advantage of simplifying calculations. The assumption of continuous emissions at the single maximum measured concentration carries a much higher degree of conservatism for the longer time frame of a year than for the shorter time frame of a day. For example, a 100-min sample time represents almost 7% of a day but only 0.02% of a year. The next section describes a method that can be used to estimate annual emissions from all sampling data that better accounts for the sporadic nature of R&D operations and variability of emissions.

Monte Carlo simulation method

The Monte Carlo simulation method used in this analysis is similar to that described in Ballinger et al. (2013). With a Monte Carlo simulation, a single data point is randomly selected from the distribution of stack sampling data and used as the basis to calculate an end result. The selection and calculation is performed with 1000 randomly selected data points resulting in a distribution of calculated end results. The simulation in the referenced report created distributions of annual emission estimates and release fractions for each target compound for each year and building sampled. The simulation in this application is similar except that data sets are grouped together for all years (for each building and each compound), different operating scenarios (i.e., modes of operation) are explored, the method is applied only to compounds with annual averaging periods, and results are compared to regulatory criteria. The Monte Carlo simulation was used to estimate annual stack emissions to compare with SQER and de minimis emission rates.

Using the full set of sampling data allows incorporation of the different modes of operation such as on-shift and off-shift. Results are provided in the form of a distribution from which the user can choose the distribution statistic with the desired degree of conservatism (e.g., maximum, 95th percentile, mean, or median values) to compare to regulatory values. The Monte Carlo simulation method described in this section is applied to the data set composed of target compounds with year averaging periods measured in emissions from the 331 Building. The method is used to compare calculated stack emissions to SQER and de minimis emission rates. The 331 Building was chosen because it had the greatest number of samples, most years sampled, and concentrations of interest in comparison to ASIL values.

In this study, a Monte Carlo analysis was performed on measured stack emissions data to calculate annual emissions under three modes of operation: (1) three full operating shifts, where R&D is assumed to occur around the clock and all samples (including those taken nights and weekends) are random estimates of stack concentrations at any time of the day, week, or year; (2) on-shift + off-shift, where samples taken during the normal workday (“on-shift”) are random estimates of emissions from normal working hours (˜2000 hr/yr) and samples taken during nights and weekend (“off-shift”) are representative of reduced operations occurring the rest of the year; and (3) on-shift–off-shift, where on-shift samples are attributed to normal R&D work plus non-R&D sources (e.g., furniture, office equipment, building infrastructure) and off-shift samples are attributed to non R&D sources alone. Off-shift is subtracted from on-shift in the third mode to represent emissions from R&D operations alone.

Sampling data for each building were grouped into on-shift and off-shift. For the three shifts mode, the Monte Carlo simulation, performed using Crystal Ball (Oracle, 2012), involved randomly selecting a value from the distribution that included both on-shift and off-shift data for all sample years for a given building and multiplying that value by the stack volumetric flow and conversion factors to result in the units of lb/yr. This calculation was executed 1000 times to result in a distribution of annual emission estimates. Calculations for the on-shift + off-shift mode were similar to the three shifts mode, except that random selections were taken separately from the on-shift and off-shift groups and weighted by the hours per year in that shift. On-shift was modeled as a normal distribution with a mean of 2000 hr/yr and sigma of 200 hr/yr with off-shift occurring the remainder of the year. The on-shift–off-shift mode investigates the possibility that off-shift stack measurements include emissions from non-research related operations, a concept supported by the similarity of off-shift concentrations to samples taken in hallways and corridors. Emissions were calculated by subtracting a randomly selected off-shift value from a randomly selected on-shift value before multiplying by the volumetric stack flow rate and the conversion factors. Negative values were set to zero for the final distribution in the on-shift–off-shift mode calculations.

An alternate approach to generating an annual emission estimate from an individual sampling result would be to treat each sample as an indicator of an individual 100-min segment of the year. The Monte Carlo simulation would then randomly select 5256 values (the number of 100 min segments in a year) to compute an emissions estimate and repeat that process 1000 times. This is more realistic than the original approach, but more computationally demanding. The alternative approach was used on a limited number of compounds and compared to the original Monte Carlo simulations.

Sample measurements for many compounds were close to or below detection limits, prompting the question of whether detection limits provided sufficient resolution for a comparison with regulatory criteria. The analytical detection limit for each compound is based on the quantity of that compound collected on the sample tube. Thus, detection limits in terms of stack concentration varied with compound and with the volume of gases used to collect the sample. The Monte Carlo method was applied to the range of concentrations representing detection limits as follows: Annual emissions corresponding to detection limit concentrations were calculated by grouping all results below detection limits for each compound and building, and performing a Monte Carlo analysis to calculate a distribution assuming three full operating shifts. This is the same calculation as the three shifts mode described previously, except that only the concentrations flagged as below detection limit were included in the source distribution. Calculated stack emissions corresponding to detection limit concentrations were compared with regulatory de minimis emission rates.

Chemical inventory method

In addition to emissions calculations performed using measured stack concentrations, emissions were estimated from chemical inventory data for the years sampled. Data were obtained from a chemical inventory system that tracks chemicals currently in the buildings and also roughly estimates chemical usage based on inventory record changes. Usage quantities are calculated within the system by assuming that the contents of each container are used uniformly between the time a container is full when initially added to the inventory and the time it is removed from inventory. Inventory and usage data were obtained for the compounds with annual averaging periods and emissions calculated by applying a release fraction of 10% to the combination of usage plus one-half inventory quantities (Ballinger et al., 2012). The release fraction of 10% is consistent with that used by PNNL (Woodruff et al., 2000). Chemical usage plus half of inventory was used because PNNL routinely uses this larger quantity to estimate emissions based on the following: In a population of many containers, it can be assumed that on average, the containers are half empty, and half the capacity has been used and was subject to emission processes. The chemical inventory method was applied to the 331 Building inventory data for the years sampled for the target compounds with annual averaging periods and with sufficient inventory data. Uncertainties in this method can be attributed to potential inaccuracies in the inventory data, the assumptions underlying the usage calculation, and the value of the release fraction. Although these uncertainties have not been quantified, this method is believed to overestimate emissions.

Results

Extreme maximum emissions estimate method

Ambient air concentrations calculated as shown in eq 1 were compared to ASILs concentrations and are displayed in Table 3 in terms of percent of ASIL. Results are provided for compounds with annual averaging periods (carcinogenic) arranged in order of highest to lowest percent. Compounds with 24-hr averaging periods (noncarcinogenic) are not shown in the table because they were several orders of magnitude less than ASILs with the highest result of 0.2% (p/m-xylene, 331 Building). As seen in Table 3, most calculated ambient air concentrations are less than 1% of the ASIL even under the worst-case assumptions of the single highest concentration emitted continuously from the stack over a year. Under these assumptions, only one measurement (chloroform in the 331 Building) resulted in a maximally conservative calculated ambient air concentration above the ASIL, and seven others resulted in calculated ambient air concentrations above 1% of ASIL values for one or more emission points.

Table 3. Comparison of calculated concentrations to ASILs ^a

Compound	325 Building	329 Building	331 Building	EMSL
Chloroform	0.1%	7.0%	130.8%	1.8%
Carbon tetrachloride	0.7%	18.7%	1.8%	0.3%
Methylene chloride	0.9%	16.1%	0.4%	6.4%
Hexachloro-1,3-butadiene	1.7%	2.9%	10.4%	2.8%
Benzene	7.6%	1.4%	9.5%	1.0%
Trichloroethene	3.5%	4.8%	7.9%	5.0%
1,1,2,2-Tetrachloroethane	0.2%	0.2%	2.0%	0.2%
Ethylbenzene	0.0%	0.1%	2.0%	0.1%
1,3-Butadiene	0.2%	0.5%	0.7%	0.5%
1,4-Dichlorobenzene	0.2%	0.4%	0.5%	0.1%
Vinyl chloride	0.1%	0.3%	0.5%	0.4%
1,2-Dichloroethane	0.1%	0.2%	0.3%	0.5%
1,2-Dibromoethane	0.1%	0.2%	0.3%	0.2%
Tetrachloroethylene	0.0%	0.2%	0.0%	0.0%
Propane, 1,2-dichloro-	0.0%	0.1%	0.1%	0.1%
1,1-Dichloroethane	0.0%	0.1%	0.0%	0.0%
3-Chloropropene	0.0%	0.1%	0.1%	NM ^b
1,1,2-Trichloroethane	0.0%	0.0%	0.1%	0.0%
Acetonitrile	0.0%	0.0%	0.0%	0.0%
Notes: ^aComparison of ambient air concentrations calculated from highest measured stack concentrations to acceptable source impact levels (target compounds with year averaging periods);
^bNM = not measured: 3-chloroprene was not part of the suite of target compounds analyzed during the time period that EMSL was sampled.

Alternative operating scenarios were run with AERMOD to quantify changes in the effective dispersion factor if emissions are reduced or eliminated during off-shift periods. Both annual and 24-hr dispersion factors were reduced by almost half for the 331 Building if emissions during nights and weekends are 10% of emissions during normal on-shift working hours and were reduced by 45% if emissions during off-shifts are reduced to zero. Results from the other buildings were similar to 331, with reductions ranging from 16 to 65%.

The stack concentration corresponding to the regulatory criteria under worst-case assumptions were calculated for each compound and are plotted as lines in Figure 2 along with distribution plots for each compound from the full set of sampling data. This figure is based on data from the 331 Building, which was chosen because it had the greatest number of samples, had the most years sampled, and had concentrations of interest in comparison to ASIL values as shown in Table 3. The stack concentrations corresponding to ASILs were calculated using eq 2 and a worst-case dispersion factor obtained from AERMOD. As mentioned previously, the uncertainty in the dispersion factor from meteorological input and dispersion parameters may be a factor of 2. The AERMOD results from the alternate operating scenario also indicate a factor of two reduction in the dispersion factors for emissions that occur primarily during normal working hours but are reduced or eliminated during off-shift hours. The stack concentrations corresponding to regulatory criteria in Figure 2 are considered conservative and more realistic stack concentrations representing infrequent emissions would be higher. Common units of mg/m³ are used to be consistent with sampling results. As shown in the figure, almost all sample distributions were orders of magnitude below the stack concentrations corresponding to the compounds’ ASIL. Chloroform is the exception, with a single maximum stack measurement above the calculated emission that would correspond to the ASIL under assumptions of continuous emissions at the highest measured stack concentration.

Figure 2. Calculated stack concentrations corresponding to ASILs compared to stack sampling distributions (331 Building): (a) compounds with year averaging period; (b) compounds with 24-hr averaging period.

Figure 2 also displays calculated stack concentrations that correspond to the SQER and de minimis emission rates for comparison to sample distribution plots for the full set of sampling data. Several compounds with year averaging periods had sample results above SQER-related stack concentrations, but almost all were above de minimis-related criteria.

For compounds with 24-hr averaging periods, all measured stack concentrations were below ASIL-, SQER-, and de minimis-related values, except for a single p/m-xylene value that was above de minimis. Results for nine target compounds that were previously listed as TAPs (see Table 2) were similar to those shown in Figure 2b; all measured stack concentrations were below regulatory criteria formerly assigned to those compounds.

Monte Carlo simulation method

The simplifying extreme worst-case assumptions are unrealistic for the comparison of compounds that have year averaging periods with levels of regulatory significance. Consequently, Monte Carlo simulations were used to reduce the level of conservatism in the emission estimate for comparison with the SQER and de minimis emission rates by obtaining estimates of annual emissions based on the full set of measured stack data. These simulations are used to evaluate the three alternate modes of operation. Figure 3 shows the results for the 331 Building when the Monte Carlo simulation is run using the three full operating shifts mode (Mode 1). In this process, a randomly selected stack concentration measurement is treated as a constant, continuous source and is used to calculate one estimate of annual emissions. The process is repeated 1000 times, creating a distribution curve. The maximum from the curve is the same as the extreme maximum case, but less extreme estimates such as the 95th percentile and median values are also generated. Results are normalized by each compound's SQER emission rate to show the differences compared to the SQER criteria.

Figure 3. Comparison of 95th percentile annual emission estimates (331 Building).

Many of the compounds that are above the SQER using the extreme maximum concentration are below the SQER with the three shifts mode 95th percentile value, and only hexachloro-3-butadiene is above the SQER using the median of the distribution. This agrees with Figure 2a, which shows that all hexachloro-3-butadiene sample results were above the SQER-related concentration for that compound. The reduction is generally greater for the compounds that are a higher percentage of the SQER. Reductions due to shifting from assumptions of extreme maximum to 95th percentile ranged from over two orders of magnitude (trichloroethene) to 60% (hexachloro-3-butadiene).

A more rigorous alternate approach was used in which 5256 random samples were selected to represent each 100-min segment of the year and then used to calculate an annual emission. The process was repeated 1000 times to generate a distribution of emission estimates for three selected compounds: chloroform, hexachloro-1,3-butadiene, and benzene. The alternate calculations were performed for both the three shifts and the on-shift + off-shift modes. As would be expected, the more rigorous approach produced almost the same mean, but with a much narrower distribution than the original Monte Carlo simulations. For the compounds studied, the range of the distribution using the alternate more rigorous calculations was similar to the 95% confidence interval of the mean using the original process. In almost all cases, the maximum from the alternate calculations was less than the upper bound on the mean (using a 95% confidence interval) from the original Monte Carlo simulations. The range of the mean appears to be a reasonable substitute for the more rigorous alternate calculations and the upper bound on the mean (using a 95% confidence interval) represents a realistic yet still conservative estimate of annual emissions for a given operational mode.

Changing the assumptions about the modes of operations also made a significant difference in estimating emissions for many of the compounds. Figure 4 shows the changes in estimated emissions for the four compounds with emissions closest to their SQER based on the three full operating shifts model. Results using the upper mean of the distributions produced from the Monte Carlo simulations under the different modes are displayed. Chloroform is above the SQER under all operational mode assumptions, and hexachloro-1,3-butadiene is above the SQER under the three shifts and the on-shift + off-shift modes but not the on-shift–off-shift mode. All other compounds are below their SQER criteria under all operational modes.

Figure 4. Comparison of annual emission estimates under different operational modes (331 Building; using upper bound of mean from Monte Carlo distribution).

For over half of the compounds with annual averaging periods, the difference between three shifts and on-shift + off-shift mode is less than 20%, but between on-shift + off-shift and on-shift–off-shift is 90% or greater. These are compounds where the on-shift and off-shift concentrations are similar, with many less than the detection limit so that subtracting off-shift from on-shift gives a result close to zero. For example, a high percentage of sampling results for hexachloro-1,3-butadiene and 1,3-butadiene were below analytical detection limits and resulted in zero or near-zero values for the on-shift–off-shift calculation. In addition, the inventory for many compounds was found to be low or zero, granting credibility to the on-shift–off-shift mode, which assumes off-shift sample results represent underlying background contributions from sources other than R&D operations.

A similar evaluation was made comparing calculated emissions to de minimis rates, which are established at 5% of SQER values. All of the compounds except for 1,1-dichloroethane and acetonitrile are above de minimis rates under extreme maximum assumptions. The Monte Carlo simulation results allow for more realistic emission estimates to compare with de minimis rates. The upper bound on the mean from distributions generated for each operating mode were compared to de minimis criteria and Table 4 indicates those compounds that were above de minimis. The list in Table 4 identifies the compounds that remain above the de minimis criteria as levels of conservatism are reduced from extreme maximum to more realistic estimates of emissions. Only four compounds have estimated emissions above de minimis criteria under on-shift–off-shift mode assumptions. However, this may be an artifact of detection limits that are not sensitive enough to detect concentration levels for the comparison.

Table 4. Compounds with calculated emissions above de minimis criteria (331 Building)

		Upper 95% confidence interval of mean
Compound	Extreme maximum	Mode 1: three shift	Mode 2: on-shift + off-shift	Mode 3: on-shift–off-shift	Mean of distribution using < DL sampling results
Hexachloro-1,3-butadiene	X	X	X	X	X
Chloroform	X	X	X	X
1,1,2,2-Tetrachloroethane	X	X	X	X	X
Benzene	X	X	X	X	X
1,3-Butadiene	X	X	X		X
1,4-Dichlorobenzene	X	X	X		X
Carbon tetrachloride	X	X	X		X
Vinyl chloride	X	X	X		X
1,2-Dibromoethane	X	X	X		X
1,2-Dichloroethane	X	X	X		X
Ethylbenzene	X	X
3-Chloropropene	X	X	X		X
Methylene chloride	X
1,2-Dichloropropane	X	X	X		X
Trichloroethene	X	X
1,1,2-Trichloroethane	X
Tetrachloroethylene	X
1,1-Dichloroethane
Acetonitrile

Some of the compounds listed in Table 4 had a high percentage of sampling results that were under the detection limit. These results are assigned detection limit concentrations. A Monte Carlo analysis was applied to the sample results that were under detection limit concentrations to obtain a distribution that represents the lowest quantity of emissions that could be detected. Three shifts operating assumptions were used for the simulation. The mean of this distribution for each compound was compared to de minimis criteria. As shown in Table 4, the outcome of this detection limit analysis was that more than half of the compounds with annual averaging periods were above de minimis criteria using detection limit concentrations. For these compounds, sampling did not provide sufficient resolution to compare annual emissions with de minimis criteria. Similarly, the detection level does not provide sufficient resolution for the on-shift–off-shift mode calculations for those compounds.

Title V threshold emission rates are given in tons/yr and are generally much higher than SQER rates, although two compounds, trichloroethene and acetonitrile, have threshold rates that are less than SQER rates. Comparing estimated emissions based on the highest measured stack concentrations from the 331 Building resulted in two compounds with extreme maximums above threshold rates: chloroform and trichloroethene. Calculated emissions using upper bound on the mean of the distribution under three shifts operational assumptions reduces the estimates so that both of these compounds are below threshold criteria.

Chemical inventory method

Chemical inventory data were used as an alternate method to estimate annual emissions. This included chemical inventory at the beginning of the year corresponding to the years sampled for each building and usage data for those years. The inventory data indicated no inventory for some compounds for specific buildings or years. For example, only 11 of the 19 compounds with year averaging periods had inventory data for the 331 Building during the 7 years sampled, and three of those 11 had no inventory for some of those years. Figure 5 shows the highest estimated annual emissions from the inventory calculations for the six compounds with the highest percent of SQER. Inventory-based emission estimates are compared to upper bound of the mean values from the Monte Carlo-generated distributions under the different operational modes. In all cases, only chloroform was shown to have calculated emission estimates above SQER criteria. The inventory-based calculated emission value for chloroform was above the SQER for 2 of the 7 years sampled, but the average was just under the SQER. In general, the inventory-based emission estimates are most similar to the upper bound of the mean from the on-shift + off-shift calculated distributions.

Figure 5. Comparison of emission estimates based on inventory vs. measurements (331 Building; using upper bound of mean for modes).

Discussion

Comparing stack emissions based on stack concentration measurements to regulatory criteria for emissions requires assumptions about the temporal variability of the emissions. Even though stack measurements show a wide range of concentrations, using an extreme worst-case approach that maximizes emissions estimates simplifies the analysis and provides an initial screening evaluation. This approach revealed that almost all of the target compounds measured in the four R&D buildings sampled over a 10-yr time frame were below ASIL concentration, even using the single highest 24-hr and annual dispersion factors based on 5 years of meteorological data and continuous release plus the single highest concentration measured for each compound. Only one compound, chloroform, was above its ASIL using these extreme maximum assumptions, and seven others were greater than 1% of their ASIL.

Part of the conservatism is due to assumptions in calculating dispersion factors that convert stack emissions to ambient air concentrations. AERMOD was used to calculate dispersion under worst-case conditions and an alternate case representing reduced emissions during nights and weekends. Alternate dispersion factors are 50% of extreme worst-case dispersion factors.

Measured chloroform stack concentrations ranged over nearly four orders of magnitude. An assumption of continuous emissions over a year at the highest stack concentration measured may simplify the calculation, but is not realistic. This becomes evident by visualizing the sampling data compared to regulatory criteria under extreme maximum assumptions as presented in Figure 2 for the 331 Building which shows that compounds with 24-hr average regulatory limits were several orders of magnitude below the corresponding ASILs and SQER values and were also well below de minimis-related values except for maximum concentrations of p/m-xylene. However, several compounds with yearly average regulatory limits have some measured stack concentrations that would be higher than SQER-related concentrations, and almost all have measurements higher than de minimis-related concentrations.

Monte Carlo techniques were used to calculate more realistic emissions for compounds with annual averaging periods to compare with criteria used by regulators to determine modeling requirements or define trivial emissions. This technique allows consideration of the full range of data and the investigation of different modes of operation. Results vary depending on the compound, the statistical value selected from the resulting distribution, and operational mode (Figures 3 and 4). Applying the analysis to the 331 Building stack concentration measurements indicated that eight compounds had calculated emission estimates above SQER criteria under extreme maximum assumptions, but only three were above SQER criteria compared to the 95th percentile value from the distributions generated by Monte Carlo simulations and only two, hexachloro-1,3-butadiene and chloroform were above SQER criteria compared to the upper 95% confidence interval of the mean of the distributions.

Comparisons of extreme maximum and Monte Carlo scenarios generated emission estimates to the more restrictive de minimis criteria were made for the compounds with annual average regulatory limits. All but two of these compounds (1,1-dichloroethane and acetonitrile) had calculated emission estimates above de minimis rates under extreme maximum conditions. Results from the Monte Carlo analysis showed some additional compounds below these criteria, but many remained above, including those for which the analysis showed most samples to be below detection limits. To address whether detection limits provided sufficient resolution for this comparison, an analysis similar to the three shifts mode was performed on sample results flagged as less than the detection limit. The median of the resulting distribution was then compared to de minimis and SQER values.

Hexachloro-1,3-butadiene was the only compound that had calculated emission estimates above SQER values using detection limit results, but 10 others were above de minimis, as shown in Table 4. Eight compounds had detection limits that allowed for comparison with de minimis criteria, as indicated by no X in the right-most column of Table 4. The de minimis comparison for these compounds shows that acetonitrile and 1,1-dichloroethane are below de minimis under all scenarios, chloroform is above de minimis under all scenarios, and five other compounds (ethylbenzene; methylene chloride; trichloroethene; 1,1,2-trichloroethane; and tetrachloroethylene) may or may not be above de minimis depending on the operational mode assumed.

Chemical inventory data was used as an alternate method to calculate annual emissions for compounds with averaging period of a year and with sufficient inventory. Eight of the compounds met these conditions in the case of the 331 Building. Inventory-based estimates were similar to estimates based on stack concentration measurements using the Monte Carlo techniques in that they made similar predictions with respect to whether a chemical was above or below SQER and de minimis criteria.

Conclusion

Stack emission concentration measurements from R&D facilities were evaluated against regulatory criteria using simplifying conservative assumptions, and the bases for these assumptions were explored to quantify degrees of conservatism. Predicted downwind concentrations were below ASILs for almost all compounds, even under extreme maximum analyses. Chloroform was the only compound for which the extreme maximum assumptions resulted in a predicted ambient annual average concentration above the ASIL. Concentrations of compounds in the measured stack data ranged over several orders of magnitude, so using maximum measured stack concentrations to calculate annual emissions or ambient concentrations based on annual averaging periods is unrealistic. Using Monte Carlo techniques allows emissions to be estimated based on the full range of measured stack concentrations, and a comparison value can be selected from the resulting distribution that corresponds to the desired degree of conservatism (e.g., upper bound of the mean, 95th percentile). The methods shown also provide a means to incorporate different operational modes.

Applying Monte Carlo techniques to measured stack concentration data from one of the PNNL facilities was helpful in comparing estimated emissions to regulatory criteria used to determine modeling requirements or define trivial levels of emissions. Although extreme maximum analysis indicated that eight compounds with annual averaging periods had calculated emission estimates above SQER values, most of these were shown to be below SQER values using Monte Carlo analysis. Only one compound had calculated emission estimates above SQER values under all methods. Detection limits were problematic for many of the compounds when comparing estimated emissions to the lower de minimis criteria. Of the eight compounds with sufficient detection for the comparison, two were below de minimis even under extreme maximum conditions, one was above de minimis under all operating mode assumptions, and the other five had mixed results depending on the mode of operation assumed.

Chemical inventory methods provide an alternate method of estimating annual emissions and produced similar results with respect to determining which compounds had calculated emission estimates above SQER and de minimis values. Of the three operational modes investigated, inventory-based estimates were the most similar to the mode that assumes off-shift samples represent a reduced level of operations during nights and weekends. The uncertainties in the chemical inventory method have not been quantified, but in general the method is believed to conservatively estimate annual emissions.

Acknowledgment

The authors thank Jeremy Rishel for his assistance in the AERMOD calculations and Professor Christopher Simpson for his careful review of the paper and suggestions for improvements.