April 2021

Inventory of U.S. Greenhouse Gas Emissions and Sinks 1990-2019:

Updates for Natural Gas and Petroleum Systems CO2
Uncertainty Estimates

EPA updated the approach to estimate uncertainty for CH4 emissions from natural gas and petroleum systems
in the 2018 Inventory of U.S. Greenhouse Gas Emissions and Sinks (GHGI). EPA previously did not calculate the
uncertainty for C02 emissions specifically, but instead applied the CH4 uncertainty bounds to the estimated C02
emissions. This memorandum discusses an update included in the 2021 GHGI to calculate uncertainty bounds
specific to C02 emissions from National Gas and Petroleum Systems.

1 Background and 2020 (Previous) GHGI Methodology

For each annual GHGI, EPA conducts a quantitative uncertainty analysis using IPCC Approach 2 methodology
(i.e., Monte Carlo simulations technique). IPCC suggests the use of a 95% confidence interval, which is the
interval that has a 95% probability of containing the unknown "true" value. Therefore, EPA uses @RISK, a
Microsoft Excel add-in tool to estimate the 95% confidence bound around CH4 emissions from both the natural
gas and petroleum systems inventories. Due to the significant number of emissions sources in natural gas and
petroleum systems (i.e., each contains more than 100 emission sources), EPA does not calculate the
uncertainty for every emission source. Rather, EPA calculates the uncertainty for the highest-emitting sources
that cumulatively contribute at least 75% of gross emissions in natural gas and petroleum systems in the most
recent GHGI year, and then applies those results via Monte Carlo simulations to the emissions for the other
smaller sources to estimate the overall uncertainty. The 75% cumulative contribution was determined, through
the stakeholder process, to be an appropriate level of precision given the large number of emission sources
included in both the natural gas systems and petroleum systems.

In previous GHGIs, prior to 2021, EPA did not calculate uncertainty bounds specific to C02 emissions. Instead,
EPA applied the calculated CH4 bounds for natural gas and petroleum systems inventories, expressed as the
percent (%) deviation above and below, to the C02 emissions estimates.

To develop a 95% confidence interval for an emission estimate from a chosen sector (e.g., natural gas
systems), it is necessary to characterize the probability density function (PDF) of the average emission and
activity factors for each emission source contributing to that source category emission estimate. The PDF
describes the range and relative likelihood of possible values for the average emission and activity factors
corresponding to that emission source (e.g., flares in the natural gas processing segment). EPA develops
uncertainty model parameters based on published studies, Greenhouse Gas Reporting Program (GHGRP)
Subpart W data, and/or expert judgment for each of the top emission sources. If the modeling input (e.g.,
emission factor) is based on GHGRP Subpart W data, EPA employs bootstrapping to determine the shape and
other parameters of the sampling distribution of the mean value. The bootstrapping analysis enables the
determination of the PDF (e.g., normal, lognormal) as well as applicable statistical parameters (e.g., standard
deviation, maximum, minimum) needed for the Monte Carlo simulation. For modeling inputs based on
recently published studies (e.g., Zimmerle et al. 2019), EPA directly uses uncertainty information included in
the study.1 For modeling inputs based on older data sets (e.g., 1996 EPA/GRI study) or macro parameters,
which are used as inputs to several emission source estimates (e.g., total active well counts from Enverus
Drillinglnfo), EPA treats these input parameters as a uniformly distributed estimate and refers to published
estimates and expert judgment to estimate upper and lower bounds. For input values obtained from certain
data sources where uncertainty data are not available, EPA assigns uncertainty bounds based on expert

1 Gathering and boosting CH4 emissions were a top source in the 2020 GHGI uncertainty analyses. Zimmerle, Daniel et al.,
Characterization of Methane Emissions from Gathering Compressor Stations. Available at
https://mountainscholar.org/handle/10217/195489. October 2019.

Page 1 of 13


-------
April 2021

judgment based on a characterized level of confidence; for example, EPA assigns uncertainty bounds of 5% to
the U.S. Energy Information Administration (EIA) data.

Per the Intergovernmental Panel on Climate Change (IPCC) Guidance, an uncertainty analysis should be seen as
a means to help prioritize national efforts to reduce the uncertainty of inventories in the future, and guide
decisions on methodological choice.2 Uncertainty estimates in the GHGI capture quantifiable uncertainties in
the input activity and emission factors data, but do not account for the potential of additional sources of
uncertainty such as modeling uncertainties, data representativeness, measurement errors, and misreporting
or misclassification.

2 CO2 Uncertainty Analysis

EPA updated the uncertainty methodology for the 2021 GHGI and applied the Monte Carlo simulation
technique to calculate the 95% confidence interval for C02 emissions in natural gas and petroleum systems.
For this initial C02 uncertainty analysis, EPA examined year 2018 emissions from the 2020 (previous) GHGI and
did not update the analysis to use year 2019 emissions from the 2021 GHGI. The C02 uncertainty bounds
(expressed as a percent) calculated for year 2018 in the 2020 GHGI were applied to year 2019 emissions in the
2021 GHGI, as shown in Table 8.

As a first step, EPA reviewed the 2020 (previous) GHGI C02 emissions for year 2018 to assess the highest-
emitting sources and identify those that cumulatively contribute at least 75% of emissions. Table 1 and Table 2
show the top 15 sources of 2018 emissions for natural gas and petroleum systems, respectively.

Table 1. Top 15 Sources of C02 Emissions for Natural Gas Systems in 2020 (Previous) GHGI

Industry
Segment

Emission Source

2018 CO2
Emissions
(mt)

% of Total

CO2
Emissions

% of Total CO2
Emissions,
Cumulative

Source in
top 75%?

Processing

Acid Gas Removal (AGR) Vents

17,451,105

49.9%

49.9%

Yes

Processing

Flares

6,981,114

20.0%

69.9%

Yes

Production

G&B Stations - Flare Stacks

4,205,760

12.0%

81.9%

Yes

Production

Miscellaneous Onshore
Production Flaring

1,380,268

3.9%

85.8%



Production

G&B Stations - Tanks

1,294,821

3.7%

89.5%



Production

Condensate Tanks

844,923

2.4%

92.0%



Production

G&B Stations - Dehydrators

801,603

2.3%

94.2%



Production

G&B Stations - AGR

643,969

1.8%

96.1%



Exploration

HF Completions

391,897

1.1%

97.2%



LNG Export

LNG Export Terminals

273,956

0.8%

98.0%



Production

Pneumatic Controllers

111,831

0.3%

98.3%



Production

HF Workovers

106,196

0.3%

98.6%



Transmission +
Storage

Flaring (Storage)

80,016

0.2%

98.8%



Transmission +
Storage

Flaring (Transmission)

75,251

0.2%

99.1%



Production

G&B Stations - other

70,463

0.2%

99.3%



TOTAL

34,971,601



2 2006 IPCC Guidelines for National Greenhouse Gas Inventories; Chapter 3 - Uncertainties, https://www.ipcc-
nggip.iges.or.jp/public/2006gl/pdf/l_Volumel/Vl_3_Ch3_Uncertainties.pdf

Page 2 of 13


-------
April 2021

Table 2. Top 15 Sources of C02 Emissions for Petroleum Systems in 2020 (Previous) GHGI

Industry
Segment

Emission Source

2018 C02
Emissions
(mt)

% of Total

CO2
Emissions

% of Total CO2
Emissions,
Cumulative

Source in
top 75%?

Production

Associated Gas Flaring

18,980,470

51.6%

51.6%

Yes

Production

Oil Tanks

6,369,067

17.3%

68.9%

Yes

Production

Miscellaneous Production
Flaring

4,226,320

11.5%

80.3%

Yes

Refinery

Flaring

3,648,222

9.9%

90.2%



Exploration

HF Well Completions

2,729,682

7.4%

97.7%



Production

Offshore Facilities (GoM
Federal)

411,412

1.1%

98.8%



Production

Offshore Facilities (AK)

122,362

0.3%

99.1%



Production

HF Workovers

92,895

0.3%

99.4%



Production

Pneumatic Controllers

81,375

0.2%

99.6%



Refinery

Process Vents

53,693

0.1%

99.7%



Refinery

Asphalt blowing

32,559

0.1%

99.8%



Exploration

Non-completion Well
Testing

31,698

0.1%

99.9%



Production

Offshore Facilities (Pacific)

8,688

<0.05%

99.9%



Production

Chemical Injection Pumps

7,834

<0.05%

100.0%a



Production

Associated Gas Venting

5,484

<0.05%

100.0%a



TOTAL

36,814,372



a. Cumulative emissions are less than 100%, but value is rounded to show to one decimal point.

Flaring and acid gas removal (AGR) emissions are the primary source of C02 emissions in natural gas and
petroleum systems and most of the top individual emission sources include either a flare or an AGR unit. Based
on year 2018 emissions, each sector has one emission source that accounts for approximately 50% of total C02
emissions: processing plant AGR units for natural gas systems and associated gas flaring for petroleum
systems. Each sector also needs only three emission sources to achieve the 75% emissions threshold for the
uncertainty analysis. In general, the largest C02 emission sources are different than the largest CH4 emission
sources.

It should be noted that each of the flaring and AGR emission sources that cumulatively contribute at least 75%
of emissions to natural gas and petroleum systems rely on emission factors and activity factors calculated from
Subpart W data. In each of these instances, EPA used a bootstrapping analysis to characterize the PDF (e.g.,
normal, lognormal) and statistical parameters (e.g., standard deviation) for the Monte Carlo simulation.
Bootstrapping analyses are further discussed in the following section. The uncertainty results from the sources
that cumulatively contribute at least 75% of emissions were used to estimate the uncertainty for the other
smaller emission sources and the overall uncertainty via Monte Carlo simulation (as discussed in Section 1).

2.1 Bootstrapping Results

EPA performed the bootstrapping analyses for each of the Subpart W emission factors (EFs) and activity factors
(AFs). Results are provided below. Table 3 and Table 4 provide the GHGI mean value for the year 2018, the PDF
and relevant inputs for the Monte Carlo simulation as determined by the Microsoft Excel @ RISK add-in tool,
and the simulated 95% interval for the GHGI mean emission source EFs and AFs for natural gas (Table 3) and
petroleum (Table 4) systems. The 95% interval is shown as the percent above and below the GHGI mean and is
for contextual purposes only.

Page 3 of 13


-------
April 2021

The PDF for each EF and AF was chosen using a best fit analysis performed in @RISK. This approach is slightly
different than the current approach for natural gas and petroleum system CH4 emissions, as well as the overall
US GHGI uncertainty analysis, which both limit the possible PDF shapes to the most common types (e.g.,
normal, lognormal, etc.). The IPCC Guidance3 notes that there can be large differences between different
distribution functions at the extremes, where there are few or no data to constrain distribution type. This
highlights the importance of identifying the PDF of best fit during this step of the uncertainty analysis. As the
GHGRP data evaluated here are considered to be robust and large datasets, the EPA did not limit the PDF
shapes fit by @RISK. EPA sought stakeholder feedback on this approach, but none was received. Table 5 shows
an example of each PDF assigned by @RISK using a best fit function, a pictorial representation of that assigned
shape, and a histogram with 1,000 datapoints as a result of the bootstrapping.

3 2006 IPCC Guidelines provide 'Good Practice Guidance' for selecting PDFs (Section 3.2.2.4). "In many cases, several functions will fit
the data satisfactorily within a given probability limit. These different functions can have radically different distributions at the
extremes where there are few or no data to constrain them, and the choice of one function over another can systematically change the
outcome of an uncertainty analysis. Cullen and Frey (1999) reiterate the advice of previous authors in these cases that it must be
knowledge of the underlying physical processes that governs the choice of a probability function. What the tests provide, in the light of
this physical knowledge, is guidance on whether this function does or does not satisfactorily fit the data" (pg 24).

Page 4 of 13


-------
April 2021

Table 3. Overview of Natural Gas Systems Year 2018 C02 Uncertainty Inputs for @RISK Modeling

Emissions Calculation Input

Year 2018 GHGI
Mean Value

PDF

Relevant Inputs

2.5%
Percentile

97.5%
Percentile

EF - Processing - AGR Vents (Metric tons
C02/plant/year)

24,771

Beta General

Shape Parameter 1 = 6.4
Shape Parameter 2 = 20
Min = 13,766
Max =58,076

18,565
(-25%)

32,572
(32%)

EF - Processing - Flares (Metric tons
C02/plant/year)a

10,466

Lognorm

Standard Deviation = 1,538
Shift = 1,179

7,752
(-26%)

13,831
(32%)

EF - Production - G&B Stations - Flare Stacks
(Metric tons C02/flare)

920

Gamma

Shape = 14
Scale = 67
Shift = -8.6

531
(-45%)

1,527
(59%)

AF - Production - G&B Stations - Flare Stacks
(flare count)

4,254

Gamma

Shape = 6.5
Scale = 341
Shift = 1,992

2,839
(-33%)

6,215
(47%)

Page 5 of 13


-------
April 2021

Table 4. Overview of Petroleum Systems Year 2018 C02 Uncertainty Inputs for @RISK Modeling

Emissions Calculation Input

Year 2018 GHGI
Mean Value

PDF

Relevant Inputs

2.5%
Percentile

97.5%
Percentile

Production - Associated Gas Flaring

Basin
220

AF - Percent of Production with Assoc.
Gas Flaring or Venting

3.9%

Lognorm

Mean = 0.047
Standard Deviation = 0.022
Shift = -0.0050

1.3%
(-69%)

9.5%
(128%)

AF - Percent of Production with Assoc.
Gas that is Flared

97.6%

Pert

Min = 0.86

Most Likely Value for Shape = 1.0
Max = 1

93.1%
(-5%)

99.9%
(2%)

EF - C02 (standard cubic feet/billion
barrels)

633

Invgauss

Mean = 653
Shape = 3,863
Shift = 60

340
(-52%)

1,423
(99%)

Basin
360

AF - Percent of Production with Assoc.
Gas Flaring or Venting

0.09%

Pearson5

Shape: 36
Scale: 0.065
Shift: -0.00094

0.03%
(-60%)

0.15%
(79%)

AF - Percent of Production with Assoc.
Gas that is Flared

86.5%

Kumaraswamy

Shape Parameter 1 = 1.8
Shape Parameter 2 = 0.33
Min = 0.18
Max = 1.0

55.3%
(-36%)

100%
(17%)

EF - C02 (standard cubic feet/billion
barrels)

5,987

Gamma

Shape = 7.7
Scale = 1,016
Shift = -1,798

1,492
(-75%)

11,899
(97%)

Basin
395

AF - Percent of Production with Assoc.
Gas Flaring or Venting

58.8%

Gamma

Shape = 45
Scale = 0.016
Shift = -0.12

41.3%
(-32%)

83.2%
(38%)

AF - Percent of Production with Assoc.
Gas that is Flared

100%

Kumaraswamy

Shape Parameter 1 = 1.0
Shape Parameter 2 = 0.20
Min = 1.0
Max = 1.0

100%
(-0.02%)

100%
(0.01%)

EF - C02 (standard cubic feet/billion
barrels)

683

Beta General

Shape Parameter 1 = 4.6
Shape Parameter 2 = 16
Min = 331
Max = 1,960

453
(-34%)

1,007
(46%)

Basin
430

AF - Percent of Production with Assoc.
Gas Flaring or Venting

37.8%

Weibull

Shape = 2.0
Scale = 0.36
Shift = 0.065

13.1%
(-66%)

76.8
(99%)

AF - Percent of Production with Assoc.
Gas that is Flared

99.0%

Minimum Extreme
Value

Location = 0.99
Shape = 0.0065

96.1%
(-3%)

99.9%
(1%)

EF - C02 (standard cubic feet/billion
barrels)

293

Invgauss

Mean = 327
Shape = 1,185
Shift = 20

130
(-62%)

769
(121%)

Page 6 of 13


-------
April 2021

Emissions Calculation Input

Year 2018 GHGI
Mean Value

PDF

Relevant Inputs

2.5%
Percentile

97.5%
Percentile

Other
Basins

AF - Percent of Production with Assoc.
Gas Flaring or Venting

4.2%

Gamma

Shape = 4.3
Scale = 0.0089
Shift = 0.0053

1.7%
(-61%)

8.8%
(102%)

AF - Percent of Production with Assoc.
Gas that is Flared

92.5%

Pert

Min = 0.52

Most Likely Value for Shape = 1.0
Max = 1.0

74.5%
(-19%)

99.9%
(9%)

EF - C02 (standard cubic feet/billion
barrels)

450

Weibull

Shape = 2.0
Scale = 446
Shift = 108

185
(-63%)

956
(90%)

Production - Large Oil Tanks with Flares

AF - Percent of Tank Throughput That Goes
Through Large Oil Tanks with Flares

64.7%

Normal

Mean = 0.65

Standard Deviation = 0.050

54%
(-16%)

75%
(15%)

EF - C02 (standard cubic feet/billion barrels)

87.4

Gamma

Shape = 24
Scale = 3.0
Shift = 16

62
(-30%)

119
(35%)

Miscellaneous Production Flaring

Basin
220

EF - C02 (Metric tons/billion barrels)

0.0011

Pearson5

Shape = 33
Scale = 0.070
Shift = -0.0010

0.0005
(-54%)

0.002
(76%)

Basin
395

EF - C02 (Metric tons/billion barrels)

0.0035

Pert

Min = 0.000027

Most Likely Value for Shape = 0.000027
Max = 0.020

0.0001
(-96%)

0.0101
(194%)

Basin
430

EF - C02 (Metric tons/billion barrels)

0.0009

Gamma

Shape = 19
Scale = 0.000078
Shift = -0.00051

0.0004
(-61%)

0.0017
(76%)

Other
Basins

EF - C02 (Metric tons/billion barrels)

0.0007

Invgauss

Mean = 0.0010
Shape = 0.023
Shift = -0.00033

0.0003
(-50%)

0.0012
(70%)

Page 7 of 13


-------
April 2021

Table 5. PDF Supplemental Information

PFD Types

PDF Pictorial Representation

Example, Emission Calculation Input PDF

Beta General

EF — Processing — AGR Vents (Metric tons C02/plant/year)



t2> /y o?

M .& .$

Gamma

EF - Production - G&B Stations - Flare Stacks (Metric tons C02/flare)

Invgauss

Basin 220; EF - C02 (standard cubic feet/billion barrels)

^ ^ ^ ^ ^ ^ ^
#	** #' ¥>• #' 4*" #'	,<#*¦ i# irf1'	mss*

^ e.'6" #¦"

Page 8 of 13


-------
April 2021

PFD Types

PDF Pictorial Representation

Example, Emission Calculation Input PDF

Kumaraswamy

Basin 360; AF - Percent of Production with Assoc. Gas that is Flared

r v °r

oV	o\o" ePx dp' dp"1	A"'' oV" A0'' A®'

# i i i i $ 
-------
April 2021

PFD Types

PDF Pictorial Representation

Example, Emission Calculation Input PDF

Normal

Production - Large Oil Tanks with Flares;

AF - Percent of Tank Throughput That Goes Through Large Oil Tanks with Flares

(0.52,0.53) (0.55,0.57) (0.59,0.61] (0.62,0.64] (0.66,0.68) (0.70,0.71] (0.73,0.75] (0.77,0.79] [0.80,0.82)
[0.50,0.52] (0.53,0.55) (057,0.59] (0.61,0.62] (0.64,0.66] {0.68,0.70) (0.71,0.73] (0.75,0.77] (0.79,0.80)

Pearson5

Basin 360; AF - Percent of Production with Assoc. Gas Flaring or Venting

y ry v

Pert

Basin 220; AF - Percent of Production with Assoc. Gas that is Flared

&



/ / / / / *

Page 10 of 13


-------
April 2021

PFD Types

PDF Pictorial Representation

Example, Emission Calculation Input PDF

Weibull







160



Basin 430; AF- Percent of Production with Assoc. Gas that is Flared











120
100
80
60
40











































Ihh











J*?











/

/•

/¦ / / 
-------
April 2021

2.2 Monte Carlo Results

Tables 6 and 7 summarize the calculated source category level uncertainty estimates for petroleum and
natural gas systems based on year 2018 C02 emissions from the 2020 (previous) GHGI. Included as the last row
in each table is the methane uncertainty results from last year's GHGI for comparison. These Monte Carlo
results based on year 2018 C02 emissions in the 2020 GHGI were applied to year 2019 emissions in the 2021
GHGI; see Table 8. In future GHGIs the uncertainty estimates will be quantified for the most recent year of
data.

Table 6. Summary of Petroleum Systems Year 2018 C02 Uncertainty Results

Emission Source

Mean Year
2018 Emissions
(MT C02)

2.5% Lower Bound of Mean
Year 2018 Emissions
(MT C02)

97.5% Upper Bour
Year 2018 Em
(MT C02

d of Mean
ssions

Value

%

Value

%

Associated Gas Flaring

220 Gulf Coast

686,281

162,148

-76%

1,859,022

171%

360 Anadarko

37,482

6,334

-83%

100,827

169%

395 Williston

10,131,704

5,636,295

-44%

16,568,499

64%

430 Permian

7,248,710

1,548,479

-79%

20,402,926

181%

Other

876,292

204,927

-77%

2,237,256

155%

Production - Large Oil Tanks with Flares

6,369,067

4,315,997

-32%

8,963,286

41%

Miscellaneous Production
Flaring

220 Gulf Coast

686,842

305,011

-56%

1,223,911

78%

395 Williston

1,653,170

62,280

-96%

5,152,768

212%

430 Permian

1,182,863

455,995

-61%

2,086,585

76%

Other

703,446

338,856

-52%

1,199,926

71%

Total for Sources Modeled a

29,575,857

20,514,329

-31%

43,877,159

48%

Total for Sources Not Modeled

7,238,515

4,643,803

-36%

10,697,527

48%

Source Category Total

36,814,372

26,890,336

-27%

51,923,681

41%

a. Those sources that cumulatively contribute at least 75% of emissions.

Table 7. Summary of Natural Gas Systems Year 2018 C02 Uncertainty Results

Emission Source

Mean Year
2018 Emissions
(MT C02)

2.5% Lower Bound of Mean
Year 2018 Emissions
(MT C02)

97.5% Upper Bour
Year 2018 Em
(MT C02

d of Mean
ssions

Value

%

Value

%

Acid Gas Removal Vents

16,522,287

12,304,773

-26%

21,834,132

32%

Flares

6,981,114

5,218,862

-25%

9,211,674

32%

Gathering & Boosting - Flare Stacks

4,205,760

1,997,940

-52%

7,567,846

80%

Total for Sources Modeled a

27,709,161

22,338,840

-19%

34,076,332

23%

Total for Sources Not Modeled

7,262,440

5,675,255

-22%

8,865,628

22%

Source Category Total

34,971,601

29,295,317

-16%

41,463,998

19%

a. Those sources that cumulatively contribute at least 75% of emissions.

Table 8. Summary of 2021 GHGI C02 Uncertainty Results

Sector

Mean Year
2019 Emissions
(MMT C02)

2.5% Lower Bound of Mean
Year 2019 Emissions
(MMT C02)

97.5% Upper Bound of Mean
Year 2019 Emissions
(MMT C02)

Value

%

Value

%

Petroleum Systems

47.3

34.5

-27%

66.6

+41%

Natural Gas Systems

37.2

31.3

-16%

44.3

+19%

3 Requests for Stakeholder Feedback

EPA sought stakeholder feedback in the November 2020 memo and in the public review draft of the GHGI, but
did not receive any stakeholder comments.

Page 12 of 13


-------
August 2020

The questions below were not updated for this memorandum and are copied from the November 2020 memo.
Questions to Stakeholders

EPA seeks stakeholder feedback on the approach under consideration and the questions below.

1.	EPA seeks general feedback on the approach of calculating uncertainty bounds for C02 emissions
separately from CH4 emissions.

2.	EPA seeks feedback on applying the CH4 emissions uncertainty methodology to C02 emissions (e.g.,
calculate the uncertainty for the highest-emitting sources that cumulatively account for at least 75% of
total C02 emissions and use Monte Carlo simulations to calculate the uncertainty for the other smaller
sources and the overall uncertainty).

3.	EPA seeks feedback on whether the PDFs incorporated into the uncertainty analysis should be limited
(e.g., normal, lognormal, uniform, triangular, and beta) or if other distributions should be considered
(e.g., Weibull, Kumaraswamy, Pearson5).

Page 13 of 13


-------