ARTICLE pubs.acs.org/IECR
Temporal Variability in Flaring Emissions in the HoustonGalveston Area Radovan T. Pavlovic, David T. Allen,* and Elena C. McDonald-Buller Center for Energy and Environmental Resources, The University of Texas at Austin, 10100 Burnet Road, Building 133, M.S. R7100, Austin, Texas 78759, United States
bS Supporting Information ABSTRACT: Recent studies performed in the HoustonGalvestonBrazoria (HGB) area indicate that some industrial air pollutant emission sources exhibit high temporal variability that can lead to very rapid ozone formation, especially when emissions include highly reactive volatile organic compounds. This motivated the collection of a unique data set of air pollutant emissions, from industrial facilities, reported with an hourly time resolution. The industrial flares portion of this data set was utilized in this work to characterize and model the highly variable temporal patterns of flare emissions at petrochemical facilities. Petrochemical and chemical manufacturing flares were grouped into categories based on industrial process they service, chemical composition of the flared gas, and the temporal patterns of their emissions. Stochastic models were developed for each categorization of flare emissions and provide representative temporal profiles for flares in specific types of operations in the petrochemical and chemical manufacturing sectors.
’ INTRODUCTION One of the air quality impacts of flaring emissions is the formation of ground level ozone. Ozone in the troposphere is generated by photochemical reactions of oxides of nitrogen (NOx) and volatile organic compounds (VOCs) and is one of the most ubiquitous air pollutants in urban areas. Because of the complex, nonlinear relationships between emissions of ozone precursors and formation of ozone, the amount of ozone formed is dependent on the temporal patterns of emissions. An underlying assumption in most emissions inventories used in air quality modeling is that the emissions from flaring and many other industrial sources are nearly constant, since many industrial facilities operate continuously at a nearly constant rate of output. However, recent studies performed in the HoustonGalveston Brazoria (HGB) area indicate that some industrial emissions sources exhibit high temporal emissions variability, with emissions changing by orders of magnitude over hourly to daily time periods.14 These temporal variations in emissions can lead to very rapid ozone formation, especially when emissions contain highly reactive volatile organic compounds (HRVOCs, defined as ethene, propylene, 1, 3-butadiene, and butenes).5,6 The variability in these emissions may have a significant impact on ozone generation in the HGB region, or other regions with extensive industrial emission sources.610 Recognition of the importance of industrial emission variability motivated the collection of a unique data set in the HGB region during August and September 2006. Hourly emissions data were collected from 141 industrial locations in the HGB area (including more than 600 emissions points). The data were collected for a total of 32 days (August 15 through September 15, 2006) at one-hour intervals. Data reported included emissions from sources equipped with continuous emissions monitors for NOx and sulfur dioxide (SO2), as well as emissions from sources of HRVOCs that had continuous data related to emissions (e.g., mass flow rates to flares). Chemical speciation was based on r 2011 American Chemical Society
industry reported hourly data and speciation procedures developed by the Texas Commission on Environmental Quality.11,12 This emissions database is the largest collection of hourly industrial emissions in a single area of the United States, and is complemented by extensive ambient air quality data collected in the industrial source region during the same time period, as part of a large air quality field study (Second Texas Air Quality Field Study; TexAQS II).13 The emissions database was incorporated into a 2006 special inventory (SI) by the Texas Commission on Environmental Quality.14 This paper reports on analyses of flare emissions using this unique database. Flares constitute the majority of the 2006 SI VOC and HRVOC emissions (77% of HRVOC emissions), followed by stacks and cooling towers (18%) and fugitive emissions (5%). Flare emissions from refineries and chemical manufacturing plants were examined to identify similarities and differences in the temporal patterns of flare emissions. Stochastic models of the flare emissions were developed in order to generate temporal profiles for each of the flare subcategories. The temporal profiles obtained using this approach allow for the development of hourly emissions inventories for flares of similar categories for which hourly data may not be available.
’ METHODS The 2006 SI represents a period of one month (August 15September 15, 2006; 768 h; the reporting period coincided with the Second Texas Air Quality Field Study) during the ozone Special Issue: Industrial Flares Received: June 21, 2011 Accepted: September 20, 2011 Revised: September 18, 2011 Published: September 20, 2011 12653
dx.doi.org/10.1021/ie2013357 | Ind. Eng. Chem. Res. 2012, 51, 12653–12662
Industrial & Engineering Chemistry Research
ARTICLE
season in the HGB area and includes data from a much larger number of unit operations than was available previously at a 1-h time resolution. The 2006 SI reports hourly emissions of VOC species from sources located across 15 counties: Brazoria, Calhoun, Chambers, Galveston, Harris, Jefferson, Live Oak, Matagorda, Nueces, Orange, Victoria, Colorado, Fort Bend, Jim Wells. and Lavaca. HRVOC emissions are reported separately only from sources located in 11 reporting counties (Brazoria, Calhoun, Chambers, Galveston, Harris, Jefferson, Live Oak, Matagorda, Nueces, Orange, and Victoria).14 Texas counties with reported VOC and HRVOC emissions in the 2006 SI inventory are shown in Figure 1. Emissions are based on mass flow rates to flares and assumed destruction efficiencies. Emissions Characterization. Flare emissions sources in the 2006 SI were initially grouped based on industrial category and process type. For example, the flares that had refinery reporting codes were categorized into five major refinery processes types (six natural gas flares, eight process gas flares, six flares associated with fuel fired equipment, eight flares categorized as unclassified refinery processes, and two flares associated with catalytic cracking units). The flares were then divided into subcategories based on relative emissions variability (i.e., the ratio between the standard deviation and the mean of the hourly mass flow rate to the flare). For example, for refinery flares, three categories of emissions variability were selected. One subcategory included flares with a standard deviation of mass flow rate to the flare that was less than 95% of the average flow. A second subcategory included flares with a standard deviation of hourly flow rate that was between 95 and 160% of the mean, and a third subcategory had the highest relative variability (above 160%). The emissions time series data were analyzed by investigating cumulative probability distributions of emitted mass, as well as frequency distributions of emissions rates. For a specific flare “A”, the ratio between a single hourly emissions rate “En”, assumed constant for 1 h, and the total mass emitted was defined as follows: fn ¼
En MA
ð1Þ
where En is the nth hourly emissions in a series sorted in ascending order from lowest to highest, and MA is the total mass emitted by flare A during the 32-day period. The cumulative probability was defined as follows: n
Cn ¼
∑ fk k¼1
CN ¼
∑ fn ¼ 1 n¼1
N
ð2Þ
ð3Þ
where N represents the total number of operation hours in flare A. Figure 2a shows the cumulative distribution of total mass emitted for a representative refinery flare from the 2006 SI with 768 h of operation. The cumulative distribution represents the probability of an hourly emissions less than or equal to the corresponding value on the x-axis. For example, for the flare shown in Figure 2, more than 90% of the total flared mass is emitted with emissions rates less than or equal to 50 lbs/h. In contrast to the cumulative distribution of total mass emitted, the distribution of flaring time (i.e., the frequency distribution of emissions rates) is characterized by a histogram, which
Figure 1. East Texas counties with reported VOC and HRVOC emissions in the 2006 Special Inventory.
reports frequencies of emissions rates and corresponding flaring time probabilities (Figure 2b). For example, if an emissions rate with value En occurs for v(En) hours in a time series with total of N operating hours, then the corresponding probability for that event is given as vðEn Þ N The cumulative probability P is then defined as: pn ¼
n
Pn ¼
∑ pk k¼1
PN ¼
∑ pn ¼ 1 n¼1
N
ð4Þ
ð5Þ
ð6Þ
The cumulative distribution function based on flaring time shows the percentage of total time when hourly emissions were less than or equal to a value on the histogram x-axis. For flares that were grouped into the same subcategories, data aggregation was conducted to develop a representative temporal emissions pattern representing up to 510 flares of similar type. In the aggregated data sets, emissions data for each flare were normalized by the monthly average hourly rate for that flare (i.e., normalized emissions rate of 1.0 represents the monthly average hourly rate). The normalized aggregated data were then treated as a single data set. Stochastic Model Development. For each industrial flare category, a stochastic model was developed to characterize the temporal variability of the flare emissions. Methods used to develop the stochastic models are based on the work of Webster and co-workers, with slight modifications.3,4,15 The stochastic model for each category of flares was constructed by assuming that the flares operate in distinct operating modes. Each mode is characterized by a probability density function (PDF) describing the flow rates in the mode. The number of modes and the form of PDF for each mode (normal or log-normal distribution) is determined empirically by plotting the inverse normal of the hourly emissions against hourly emission rates or the logarithm of the hourly emission rates.3 The number of segments in these plots determines the number of modes as described in further detail in the Supporting Information. The form of the PDF representing the mode is chosen to be normal or log-normal. If the relation between the inverse normal and hourly 12654
dx.doi.org/10.1021/ie2013357 |Ind. Eng. Chem. Res. 2012, 51, 12653–12662
Industrial & Engineering Chemistry Research
ARTICLE
Figure 3. Stochastic modeling algorithm. Separate stochastic models were developed for each flare category.
Figure 2. (a) Cumulative distribution of total mass emitted and (b) distribution of flaring rate frequencies for a representative refinery flare from the 2006 SI.
emission rates is linear, then emission flow rates are normally distributed. However, if the relation between the inverse normal and logarithm of hourly emission rates is nonlinear, then emission flow rates have a log-normal distribution. As an example of the number and types of modes, one of the categories of refinery flares, described in the results section, had seven modes of operation, six of which had normal PDFs and one of which had a lognormal PDF. Note that the number of emissions components varied between two and seven for all flare categories based on the number of PDFs (i.e., distributions) fitted to the flare emissions rate data. For this work, the time that the flare stays in one emission mode before transiting to another emission mode is defined as the residence time (RT). In the approach of Webster and Nam, residence time was calculated by using an exponential distribution function or by calculating a transition probability (to remain in the same component) from the total number of hours spent in a particular emission component, respectively.3,4,15 However, residence times obtained in this way are based on the average time that a flare remains in the same component, without capturing the actual distribution of residence times. As a result, such a
modeling approach may not successfully generate episodes of unusually long residence times, which may reflect important behavior not otherwise addressed. Hence, in this work the residence time is sampled from the distribution of residence time values from the actual time-series for each component. Specifically, the RT bin was quasi-randomly sampled from a discrete distribution (i.e., the RT histogram of the actual time-series data, generated for each flare category), and the residence time was randomly selected within the bin. For example, if the quasirandom generator, based on the corresponding histogram probability, samples the RT bin of 510 h, the model will randomly select and uniformly apply a value between 5 and 10.16 As shown in Figure 3, to model a time series, the flare is placed into an initial mode that is randomly selected. Then, the number of hours that the flare will stay in that mode is randomly selected from the distribution of residence times, for that mode, abstracted from the SI data. For each hour that the flare remains in the initial mode, a flow rate is selected based on the PDF of the flow rate distribution for the selected mode (again, determined from the flow rate data). After the target number of hours in the mode is reached, a new mode is selected. The new mode selected is based on a transition matrix, again developed using the ambient data, and a new time in mode is selected. The process of selecting hourly emissions is then repeated.16 The stochastic model has a relatively large number of parameters. There may be as many as five to seven modes for each type of flare operation, and each mode has a functional form and two parameters for the PDF describing flow rates. The time spent in each mode is represented by a histogram that has multiple parameters and a transition matrix has the number of parameters equal to the number of modes squared. Thus, the stochastic model representing each type of flare may have roughly 100 parameters. These parameters are based on several thousand hours of flow data for each type of flare (multiple flares, each operating for 768 h, are represented in each group), so the model is complex, but not overdetermined.
’ RESULTS AND DISCUSSION The first step in analyzing the 2006 SI was to perform a comparison with other emission data sets for the same sources. 12655
dx.doi.org/10.1021/ie2013357 |Ind. Eng. Chem. Res. 2012, 51, 12653–12662
Industrial & Engineering Chemistry Research
ARTICLE
Table 1. 2006 SI Emissions in Tons Per Day, Averaged Throughout the Inventory Period, by County and by Emissions Point Type (Note: Blanks Are Due to Non-Reported HRVOC Emissions in Four of Fifteen Counties) flares county
fugitives
stacks and cooling towers
total
VOC [tpd]
HRVOC [tpd]
VOC [tpd]
HRVOC [tpd]
VOC [tpd]
HRVOC [tpd]
VOC [tpd]
HRVOC [tpd]
Brazoria
3.30
2.37
0.08
0.01
1.67
0.35
5.04
2.74
Calhoun
0.56
0.46
0.00
0.00
0.49
0.25
1.05
0.71
Chambers Colorado
1.08 0.00
0.30
0.00 0.00
0.00
0.36 3.94
0.00
1.45 3.94
0.31
Fort Bend
0.00
Galveston
3.87
0.46
0.84
0.00
1.23
0.04
5.94
HARRIS
8.71
3.44
2.10
0.51
7.03
0.75
17.84
4.70
Jefferson
2.18
0.66
3.12
0.00
1.73
0.41
7.03
1.07
Jim Wells
0.01
0.00
0.72
Lavaca
0.14
0.00
0.10
Live Oak Matagorda
0.04 0.05
0.00 0.04
0.00 0.02
0.00 0.00
0.05 0.17
0.00 0.00
0.10 0.25
0.00 0.04
Nueces
0.38
0.06
0.36
0.00
0.69
0.00
1.43
0.07
Orange
0.18
0.10
0.00
0.00
0.69
0.00
0.86
0.10
Victoria
0.11
0.03
0.01
0.00
0.10
0.00
0.22
0.03
20.61
7.94
6.52
0.53
18.99
1.80
46.12
10.26
total
0.00
0.01
Table 1 shows total VOC and HRVOC emissions by county and by source type as reported in the 2006 SI. For the sources reporting in the 2006 SI, 61% of VOC emissions are located in Harris, Jefferson, and Galveston counties, while 83% of all HRVOC emissions originate from sources in Harris, Brazoria, and Jefferson counties. In addition, 45% of all VOC and 77% of all HRVOC emissions in the SI originate from flares, followed by stacks and cooling towers (41% VOC and 18% HRVOC), and finally fugitives (14% VOC and 5% HRVOC). In a 2005 seasonal average inventory (2005 ozone season day (OSD) inventory17), used for comparison, the main source of HRVOCs is fugitive emissions (51%), followed by stacks and cooling towers (29%) and flares (20%). Because most fugitive emissions are not measured continuously, they are not fully represented in the 2006 SI. However, comparison of the 2005 OSD inventory and the 2006 SI suggests a higher ratio of flare emissions to cooling tower emissions than in the 2005 OSD inventory (i.e., 77% flares, 18% stacks and cooling towers, and 5% fugitives, compared to 20% flares, 29% stacks and cooling towers, and 51% fugitives in the 2005 OSD). It is important to stress that the 2006 SI is based on measured data (e.g., emissions are based on mass flow rates to flares and assumed destruction efficiencies) while the 2005 OSD inventory for the same sources is based on average emission factors. The 2006 SI also represents only a segment of industrial source emissions in the HGB area and just a selection of sources within specific industrial sites. However, even though the 2006 SI reports emissions from less than 4% of all point sources in the state of Texas (i.e., 623 emissions points are listed in the 2006 SI as opposed to 18 551 emissions points that are reported in a comprehensive TCEQ 2005 emissions inventory), cumulatively the 2006 SI HRVOC emissions represent almost a third of statewide industrial emissions of VOCs. The source sectors that contribute to the 2006 SI are representative relative to area-wide emissions. Figure 4 shows flare VOC emissions by industrial sector as reported in the 2006 SI. Almost 95% of all VOC and 99% of all HRVOC emissions in the HGB area (15 counties for VOC and 11 counties for HRVOC)
0.01 0.50
0.72 0.24
Figure 4. Flare VOC emissions distribution by industrial sector.
originate from three industrial sectors (Standard Industrial Classification, SIC): petroleum refineries (SIC 2911), chemical manufacturing (SIC 2869), and plastics production (SIC 2821). Although the 2006 SI is not a comprehensive database, the most recent inventory data reported in the 2005 OSD EI also suggest that these three industries represent the majority of VOC and HRVOC emissions in the region.17 Refinery Flares. Analysis of flare temporal profiles focused on those with reported nonzero activity during the 2006 SI inventory period. Table 2 shows 2006 SI flare VOC emissions aggregated to refinery process levels as defined by the Standard Classification Code (SCC). In the 2006 SI, there are fifteen SCC flare categories with reported emissions greater than zero; however, almost 90% of all refinery flare emissions originate from five SCC sectors: natural gas flares (SCC 30600903; 37%), followed by 12656
dx.doi.org/10.1021/ie2013357 |Ind. Eng. Chem. Res. 2012, 51, 12653–12662
Industrial & Engineering Chemistry Research
ARTICLE
Table 2. Refinery Flare VOC Emissions in the 2006 SI Aggregated by SCC Process over the Reporting Period petroleum refining (SIC 2911)
2006 SI emissions tons/month
SCC
SCC description
(768 h)
lb/day
30600903
natural gas flares
83.5
5216
30600904 30190099
process gas flares fuel fired equipment
42.6 32.6
2662 2040
30600999
flares not classified
31.4
1964
30600201
catalytic cracking units
13.1
817
30699998
petroleum products - not classified
7.1
443
30130115
miscellaneous
5.6
353
39990022
miscellaneous
4.2
264
30688801
fugitive emissions
3.0
190
30600401 30601801
blowdown systems hydrogen generation unit
1.3 0.8
80 50
64420033
miscellaneous
0.8
49
30601701
catalytic hydrotreating unit
0.2
11
30600508
wastewater treatment
0.1
7
40600240
marine vessels
0.0
1
226.3
14146
total
process gas flares (SCC 30600904; 19%), fuel fired equipment flares (SCC 30190099; 14%), unclassified flares (SCC 30600999; 14%), and fluid catalytic cracking flares (FCCU; SCC 30600201; 6%). Table 3 segregates each of these five major SCC refinery processes to the level of individual flares (six natural gas flares, eight process gas flares, six flares associated with fuel fired equipment, eight flares categorized as unclassified refinery processes, and two flares associated with FCCUs). In addition to industrial process classification, refinery flares can be also categorized based on the chemical composition of the flared gas as well as properties of the emissions time series. Analysis of the chemical composition of the major refinery flare categories in the 2006 SI revealed that fuel fired equipment, process, and natural gas flares (represented by SCCs 30190099, 30600904, and 30600903, respectively) have emissions of light gases (propane) and gasoline range volatiles (isobutane and n-butane). Fluid catalytic cracking unit flares typically contained propylene, ethene, propane, isobutene, and isopentane. Finally, VOCs from the “unclassified” flare category included propane, pentane, n-butane, isobutene, propylene, and unclassified VOC. Based on the chemical composition analysis, as well as the industrial process, refinery flares were grouped, for time series modeling, into three major categories: (1) fuel fired equipment and process and natural gas flares; (2) FCCU flares; and (3) unclassified flares with no distinct industrial process identification. Within each of these three major categories, subcategories were established based on relative emissions variability. Fuel-Fired Equipment, Process, and Natural Gas Flares. Fuel, process, and natural gas flares are used for burning excess gases from refinery processes. Flaring can occur when excess process gases are produced, during refinery startups and shutdowns, and due to maintenance activities. Therefore emissions time-series for these flares are expected to have both episodic features as well as components of the flow that are more routine. Based on the relative emissions variability (i.e., the ratio between the standard deviation and the mean), flares were grouped
into three subcategories. Subcategory I includes flares with lowest relative emissions variability (less than 95% of the mean) and typically operating above the average rate more than 50% of the time. Subcategory II represents flares with moderate relative variability (in this case between 95 and 160%) and typically operating above the average rate for approximately 3040% of the time. Finally, Subcategory III represents flares with highest relative variability (above 160%) and typically operating above the average emissions rate for less than 30% of the time. Table 3 shows fuel, process, and natural gas flares, sorted by relative variability (ascending) with the emissions statistics for each flare subcategory. Cumulative distribution plots of normalized flared VOC mass are shown in Figure 5. Emissions rates for each flare were normalized by the monthly average hourly rate in order to focus on emissions variability characteristics rather than total mass emitted. The cumulative distribution represents the probability of an hourly emissions rate less than or equal to the corresponding value on the x-axis (for example, for all flares shown in Figure 5, more than 80% of total flared mass is emitted at rates smaller than or equal to 15 times the average monthly rate). As the relative emissions variability increases from low to high and thus the frequency of high emissions rates increases, the cumulative distribution plots shift from the left to the right side of Figure 5; the cumulative distribution plots of normalized mass emitted for Subcategory I flares (light gray line in Figure 5) are on the far left; Subcategory II (dark gray) in the middle; and Subcategory III (dashed line) on the right. Whereas the cumulative distribution functions based on flaring time are relatively similar for all three of the main refinery flare categories, the distribution functions based on total mass emitted exhibit more significant differences (Figure 5). The underlying reason is that during most of the operating time (>90%), emissions rates for flares in the fuel-fired equipment, process, and natural gas category are below two times the average; however, as the magnitude of emissions excursions increase (i.e., from Subcategory I to III), the mass contributions of periods of high emissions also increase. Flares in Subcategory I emit only 2% of total mass through high emissions rates, defined as three times the average or greater; Subcategory II flares emit almost 20% of the mass through high emissions rates, and finally Subcategory III flares emit almost 40% of the mass through emissions rates of three times the average or greater. Consequently, despite the low frequency of occurrence of excursions they can significantly influence the distributions of total mass emitted (i.e., mass CDF). Fluid Catalytic Cracking Unit (FCCU) Flares. A fluidized bed catalytic cracking unit is an integral component of oil refining with the function of breaking (cracking) large hydrocarbons into gasoline and diesel range molecules. Only two FCCU flares were identified in the 2006 SI with reported VOC emissions. Table 4 provides descriptive statistics extracted from the 2006 SI for those two flares. The 2006 SI data suggest that oil refining FCCU flares in the HGB area operate with lower emissions variability, ranging between 25 and 55%, when compared to Subcategory I of fuel, process, and natural gas flares (i.e., fuel, process, and natural gas flares with lowest emissions variability ranging between 47 and 78%). Although there are two FCCU flares with nonzero emissions included in the 2006 SI, only one had considerable activity during the inventory period; the other flare (FCCU Flare 2) had an average emissions rate of less than 1 lb/h. Therefore, for purposes of this study, only FCCU Flare 1 was considered as a “representative” FCCU flare. Although the time series for this 12657
dx.doi.org/10.1021/ie2013357 |Ind. Eng. Chem. Res. 2012, 51, 12653–12662
Industrial & Engineering Chemistry Research
ARTICLE
Table 3. Emissions Statistics for Fuel Fired Equipment, Process and Natural Gas Flares, Segregated into 3 Sub-Groups Based on the Raw Data Distributions
a
average emissions
SD
relative variabilitya
% of the time above
flare ID category
rate [lb/h]
[lb/h]
[%]
the average
fuel, process and natural gas flare 1 fuel, process and natural gas flare 2
137 24
74 14
54% 57%
53% 43%
fuel, process and natural gas flare 3
39
27
68%
64%
fuel, process and natural gas flare 4
37
26
69%
61%
fuel, process and natural gas flare 5
0
0
76%
53%
fuel, process and natural gas flare 6
2
1
78%
47%
fuel, process and natural gas flare 7
4
4
98%
35%
fuel, process and natural gas flare 8
79
84
106%
44%
fuel, process and natural gas flare 9 fuel, process and natural gas flare 10
15 17
21 24
136% 139%
25% 30%
fuel, process and natural gas flare 11
16
25
157%
26%
fuel, process and natural gas flare 12
17
31
183%
28%
fuel, process and natural gas flare 13
3
5
196%
26%
fuel, process and natural gas flare 14
23
74
323%
19%
subcategory subcategory I: relative variability below 95%
subcategory II: relative variability between 95 and 160%
subcategory III: relative variability above 160%
Relative variability is defined as the ratio of the standard deviation and the average emissions rate, multiplied by 100 to yield a percentage.
Table 4. Emissions Statistics for FCCU Flares in the 2006 SI average emissions flare ID
Figure 5. Cumulative distributions of total mass emitted for fuel fired equipment, process, and natural gas flares. The shades of gray represent separate subcategories based on the statistical properties of the cumulative distributions. Emission rates for each flare are normalized by its monthly average hourly rate.
flare indicates a few excursions that lasted up to 40 h, overall the emissions variability of 25% with rates never exceeding three times the average rate can be characterized as moderate variability, compared to the fuel fired equipment, process, and natural gas flares. “Unclassified” Flares. Unclassified flares represent a group of refinery flares in the 2006 SI without a clear description of the associated process. However, collectively these flares show common behavior and can be grouped broadly into two categories based on their time series patterns and relative emissions variability. The 2006 SI includes eight flares listed as Unclassified (SCC: 30600999). Table 5 provides a summary of descriptive emissions statistics for all unclassified refinery flares. The first four flares in Table 5 comprise Subcategory I of the unclassified flares, while the remaining four flares comprise Subcategory II. Subcategory I of the unclassified flares represents flares with occasional periods of high emissions rates. When the relative
rate [lb/h]
Sd
relative
[lb/h] variability [%]
% of the time above the average
FCCU flare 1
33
8
25%
32%
FCCU flare 2