April 2018 Inventory of U.S. Greenhouse Gas Emissions and Sinks 1990-2016: Updates to Natural Gas and Petroleum Systems Uncertainty Estimates This memo documents the uncertainty analysis used in EPA's 2018 Inventory of U.S. Greenhouse Gas Emissions and Sinks (GHGI) as well as historical analyses and considerations. In previous versions of this memo released in June and October 2017, and during stakeholder webinars and workshops held in April, June, and August 2017, EPA presented preliminary considerations and sought stakeholder feedback on updating the uncertainty estimates for natural gas and petroleum systems in the 2018 GHGI. Before the 2018 GHGI update described in this memo, the most recent uncertainty analysis for the natural gas and petroleum systems emissions estimates in the GHGI was conducted for the 1990-2009 GHGI that was released in 2011. The analysis was based on a detailed assessment of the activity data and emission factor data available at that time. Since the analysis was last conducted, several of the methods and data sources used in the GHGI have changed, and industry practices and equipment have evolved. In addition, new studies and other data sources such as the EPA Greenhouse Gas Reporting Program (GHGRP) have provided more information on emissions and the underlying conditions that lead to emissions. For the 2018 GHGI, EPA implemented updates for the natural gas and petroleum systems uncertainty analyses to reflect new information, current GHGI methodologies, and stakeholder feedback on the uncertainty quantification methodology. Section 1 of this memo documents the uncertainty modeling approach and results used in EPA's 2018 GHGI; the remainder of the sections contain historical information that was presented in the October 2017 memo, for reference (note that the October 2017 analyses reflect estimates and methodologies used in the 2017 GHGI; therefore, resulting estimates changed for the final 2018 GHGI). 1. 2018 GHGI Uncertainty Analysis In recent years, EPA has made significant revisions to the Inventory methodology to use updated activity and emissions data. To update its characterization of uncertainty, EPA has conducted a quantitative uncertainty analysis using the IPCC Approach 2 methodology (Monte Carlo Simulation technique). EPA used Microsoft Excel's @RISK add-in tool to estimate the 95 percent confidence bound around CH4 emissions from natural gas systems and petroleum systems for the 2018 GHGI, then applied the calculated bounds to both CH4 and C02 emissions estimates. 1.1. Natural Gas Systems Uncertainty For the natural gas systems analysis, EPA focused on the 16 highest-emitting sources for the year 2016, which together emitted 78 percent of methane from natural gas systems in 2016, and extrapolated the estimated uncertainty for the remaining sources. The @RISK add-in provides for the specification of probability density functions (PDFs) for key variables within a computational structure that mirrors the calculation of the inventory estimate. The IPCC guidance notes that in using this method, "some uncertainties that are not addressed by statistical means may exist, including those arising from omissions or double counting, or other conceptual errors, or from incomplete understanding of the processes that may lead to inaccuracies in estimates developed from models." The uncertainty bounds reported below only reflect those uncertainties that EPA has been able to quantify and do not incorporate considerations such as modeling uncertainty, data representativeness, measurement errors, misreporting or misclassification. The understanding of the uncertainty of emission estimates for this category evolves and improves as the underlying methodologies and datasets improve. The results presented below provide the 95 percent confidence bound within which actual emissions from this source category are likely to fall for the year 2016, using the IPCC methodology. The results of the Approach 2 uncertainty analysis are summarized in Table 1. Natural gas systems CH4 emissions in 2016 were estimated to be between 138.0 and 191.8 MMT C02 Eq. at a 95 percent confidence level. Natural gas systems non-energy C02 Page 1 of 27 ------- April 2018 emissions in 2016 were estimated to be between 21.5 and 29.9 MMT C02 Eq. at a 95 percent confidence level. Uncertainty bounds for other years of the time series have not been calculated, but uncertainty is expected to vary over the time series. For example, years where many emission sources are calculated with interpolated data would likely have higher uncertainty than years with predominantly year-specific data. Table 1. 2018 GHGI Quantitative Uncertainty Estimates for CH4 and Non-energy C02 Emissions from Natural Gas Systems (MMT C02 Eq. and Percent) Source Gas 2016 Emissions Estimate (MMT CO2 Eq.) Uncertainty Range Relative to Emissions Estimate3 (MMT CO2 Eq.) (%) Lower Boundb Upper Boundb Lower Boundb Upper Boundb Natural Gas Systems ch4 163.5 138.0 191.8 -16% +17% Natural Gas Systems0 co2 25.5 21.5 29.9 -16% +17% a - Range of emission estimates estimated by applying the 95 percent confidence intervals obtained from the Monte Carlo Simulation analysis conducted for the year 2016 CH4 emissions. b - All reported values are rounded after calculation. As a result, lower and upper bounds may not be duplicable from other rounded values in the 2018 GHGI report. c - An uncertainty analysis for the non-energy C02 emissions was not performed. The relative uncertainty estimated (expressed as a percent) from the CH4 uncertainty analysis was applied to the point estimate of non-energy C02 emissions. 1.2. Petroleum Systems Uncertainty For the petroleum systems analysis, EPA focused on the five highest methane-emitting sources for the year 2016, which together emitted 78 percent of methane from petroleum systems in 2016, and extrapolated the estimated uncertainty for the remaining sources. The @RISK add-in provides for the specification of probability density functions (PDFs) for key variables within a computational structure that mirrors the calculation of the inventory estimate. The IPCC guidance notes that in using this method, "some uncertainties that are not addressed by statistical means may exist, including those arising from omissions or double counting, or other conceptual errors, or from incomplete understanding of the processes that may lead to inaccuracies in estimates developed from models." The uncertainty bounds reported below only reflect those uncertainties that EPA has been able to quantify and do not incorporate considerations such as modeling uncertainty, data representativeness, measurement errors, misreporting or misclassification. The understanding of the uncertainty of emission estimates for this category evolves and improves as the underlying methodologies and datasets improve. The results presented below provide the 95 percent confidence bound within which actual emissions from this source category are likely to fall for the year 2016, using the IPCC methodology. The results of the Approach 2 uncertainty analysis are summarized in Table 2. Petroleum systems CH4 emissions in 2016 were estimated to be between 27.1 and 51.9 MMT C02 Eq., while C02 emissions were estimated to be between 16.0 and 30.6 MMT C02 Eq. at a 95 percent confidence level. Uncertainty bounds for other years of the time series have not been calculated, but uncertainty is expected to vary over the time series. For example, years where many emission sources are calculated with interpolated data would likely have higher uncertainty than years with predominantly year-specific data. Table 2. 2018 GHGI Quantitative Uncertainty Estimates for CH4 and Non-energy C02 Emissions from Petroleum Systems (MMT C02 Eq. and Percent) Source Gas 2016 Emissions Estimate (MMT CO2 Eq.) Uncertainty Range Relative to Emissions Estimate3 (MMT CO2 Eq.) (%) Lower Boundb Upper Boundb Lower Boundb Upper Boundb Petroleum Systems ch4 38.6 27.1 51.9 -30% +34% Petroleum Systems0 CO2 22.8 16.0 30.6 -30% +34% a - Range of emission estimates estimated by applying the 95 percent confidence intervals obtained from the Monte Carlo Simulation analysis conducted for the year 2016 CH4 emissions. b - All reported values are rounded after calculation. As a result, lower and upper bounds may not be duplicable from other rounded values in the 2018 GHGI report. Page 2 of 27 ------- April 2018 c - An uncertainty analysis for the non-energy C02 emissions was not performed. The relative uncertainty estimated (expressed as a percent) from the CH4 uncertainty analysis was applied to the point estimate of non-energy C02 emissions. 2. Overview of Uncertainty Analysis in the GHGI (from October 2017 memo) In conformance with the United Nations Framework Convention on Climate Change (UNFCCC) reporting requirements, EPA follows the Intergovernmental Panel on Climate Change (IPCC) Guidelines for National Greenhouse Gas Inventories (IPCC Guidelines, IPCC (2006)) to develop uncertainty estimates for sources included in the national GHGI. The IPCC Guidelines note the essential role of uncertainty estimates for guiding improvements to national inventories: "An uncertainty analysis should be seen, first and foremost, as a means to help prioritize national efforts to reduce the uncertainty of inventories in the future, and guide decisions on methodological choice. For this reason, the methods used to attribute uncertainty values must be practical, scientifically defensible, robust enough to be applicable to a range of categories of emissions by source and removals by sinks, methods and national circumstances, and presented in ways comprehensible to inventory users." The uncertainty analysis is performed by developing confidence intervals, which give the range within which the "true" value of an uncertain quantity is thought to lie for a specified level of confidence. The IPCC Guidelines suggest the use of a 95% confidence interval, which is the interval that has a 95% probability of containing the unknown "true" value. To develop a 95% confidence interval for an emission estimate from a chosen source category (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., reciprocating compressors in the natural gas transmission segment). Ideally, the PDF would be derived from source-specific information. However, in the absence of such data, it is also possible to use information developed through elicitation of expert judgment (IPCC 2006). Once the applicable PDFs are characterized, a Monte Carlo analysis can be conducted to characterize the composite uncertainty for each emission source (e.g., national reciprocating compressor emissions in the natural gas transmission segment) as well as the overall source category (e.g., national natural gas system emissions). Although default uncertainty values are provided by IPCC and propagation of error is a valid approach, the Monte Carlo approach is more rigorous and recommended for sources that use more sophisticated estimation methodologies, where PDFs may be non-normal, and if uncertainties are large (IPCC 2006). As described in the IPCC guidelines, Monte Carlo analysis involves selecting random values for emission factors and activity data from the respective PDFs and calculating the resulting emission estimate. This procedure is repeated numerous times and the results of each simulation are used to characterize the PDF for the overall emission estimate for the source category (IPCC 2006). Figure 1 depicts the steps involved in conducting a Monte Carlo analysis. From the figure, only Steps 1 and 2 require user input (e.g., specification of PDFs for emission and activity factors); Steps 3 through 5 are conducted through use of a software package such as @RISK.To develop uncertainty bounds around total estimated emissions at the national level, the source-specific emissions that correspond to the lower and upper confidence bounds can be summed, as the total national level estimate is the simple sum of source- specific emissions. The approach described estimates the national emissions associated with a source category as the product of the average emission factor and average activity factor for that source category. While recent studies show that the emissions from certain source categories may be highly skewed with potentially fat right-hand tails due to the existence of "super-emitters," per the Central Limit Theorem (CLT), the impact of such is expected to be minimal on the average emission factors and average activity factors used in calculating national estimated emissions (see section 4 below. Page 3 of 27 ------- April 2018 Figure 1: Illustration of Monte Carlo Method (Adapted from IPCC 2006) This example assumes three emission sources each where the emission is calculated as Activity Data ¦ Emission Factor Steps 1 and 2 - Specify uncertainties, width, and PDFs for all input data Page 4 of 27 ------- April 2018 3. Background on Uncertainty for Natural Gas and Petroleum Systems (from October 2017 memo) EPA conducted the last complete uncertainty analyses for natural gas and petroleum systems for the 1990-2009 GHGI that was released in 2011. For that analysis, EPA obtained many of the emission factors and associated uncertainty parameters (e.g., PDF and standard deviation) from the 1996 EPA-Radian study of the natural gas industry and the 1999 EPA-Radian study of the petroleum industry. EPA adopted the same source category-level uncertainty intervals for natural gas and petroleum systems emission estimates subsequent to the 1990-2009 GHGI. 3.1. Basis of the 2011 GHGI Natural Gas Systems Uncertainty Analysis The 2011 GHGI uncertainty analysis for natural gas systems included a detailed analysis for the twelve top- emitting sources in 2009 (ranked according to the 2011 GHGI estimates), in which all elements of each emission source estimate were defined in the uncertainty analysis. For the remaining sources, EPA employed a simpler methodology as described in further detail below. For natural gas systems, calculations are commonly more complex than simply multiplying an emission factor by an activity factor. For example, the activity data calculation for production site upset emissions from pressure relief valves (PRVs) involves three distinct elements: count of PRVs associated with all gas wells as originally estimated in the 1996 EPA-Radian study and updated by EPA in 2007; NEMS region-specific fraction of all gas wells for a given year; and the ratio of total gas wells in a given year compared to that in year 1992. Table 3 provides the twelve top-emitting natural gas sources along with their year 1992 emissions used in the 2011 uncertainty analysis. As can be observed from the table, EPA examined individual emission sources at the NEMS region level for the production segment (due to the calculation methodology varying by region for many production sources), and at the national level for other segments. Although the top twelve sources were identified based on the year 2009 emissions estimate, EPA conducted the actual uncertainty analysis on estimates for the year 1992, because it was the base year (i.e., year of key input data collection) of the emissions and activity data estimates for many emission sources. To define the uncertainty model parameters (steps 1 and 2 in Figure 1) of every element of the activity and emission factors for the top twelve sources, EPA combined judgments of an industry expert and a statistical expert along with data published in the 1996 EPA-Radian study. For all top twelve sources as well as the remaining sources (that were analyzed using a simplified methodology), EPA assumed a lognormal PDF as default. Then using the Monte Carlo simulation method in @RISK (steps 3 through 5 in Figure 1), EPA calculated the upper and lower estimates representing the 95% confidence interval for each of the top twelve sources listed in Table 3. These top twelve sources contributed nearly 49% of the total 1992 methane emissions from natural gas systems. For the hundreds of non-top-twelve sources collectively representing approximately half of natural gas systems emissions, EPA evaluated uncertainty using a simplified method which involved assigning uncertainty values to each source activity and emission factor but not to the activity drivers associated with that source. This simplified method does not completely capture the uncertainty associated with all the sources but does ensure that the uncertainty of the sources that are not among the top twelve is represented; assuming activity drivers have associated uncertainty, this approach would lead to underestimating overall uncertainty. Also, using the Monte Carlo simulation method in @RISK, EPA calculated the upper and lower estimates representing the 95% confidence interval for the non-top twelve sources collectively. To develop the uncertainty bounds for 1992, EPA compiled the upper and lower modeled estimates for the top twelve and non-top twelve sources and then translated these figures to +/- percentages of the GHGI estimate. EPA calculated the 95% confidence interval for natural gas systems emissions for 1992 at -19% and +30% of the GHGI-reported value. EPA then assumed that the 95% confidence interval for each of the other years was equivalent to these +/- percentage values. Page 5 of 27 ------- April 2018 Table 3. Top 12 Emission Sources for Natural Gas Systems in Previous (2011) Uncertainty Analysis for GHG Inventory Published in 2011 (2011 GHGI) 2011 GHGI CH4 Emissions, year 1992 Source (MMT CChe) Liquids Unloading (production segment, North East region) 34.8 Reciprocating Compressor Fugitives (transmission segment) 18.6 Liquids Unloading (production segment, Gulf Coast region) 17.5 Reciprocating Compressor Fugitives (processing segment) 8.1 Liquids Unloading (production segment, Mid Central region) 7.9 Shallow Water Offshore Platforms (production segment) 7.4 Wet Seal Centrifugal Compressors (transmission segment) 6.2 Pneumatic Controllers (production segment, Mid Central region) 5.6 Liquids Unloading (production segment, Rocky Mountain region) 3.4 Pneumatic Controllers (production segment, Rocky Mountain region) 2.1 Unconventional Gas Well Workovers (production segment, Rocky Mountain region) 0.0 Unconventional Gas Well Workovers (production segment, South West region) 0.0 Other Emission Sources 116.8 Total Potential Emissions from Natural Gas Systems (before Gas STAR reductions) 228.4 3.2. Basis of the 2011 GHGI Petroleum Systems Uncertainty Analysis The 2011 GHGI uncertainty analysis for petroleum systems included a detailed analysis for the seven top-emitting sources in 2009 (ranked according to the 2011 GHGI estimates), in which all elements of each emission source estimate were defined in the uncertainty analysis. As with natural gas systems, calculations of emission estimates for petroleum systems sources are more complex than simply multiplying an emission factor by an activity factor. They usually involve additional data elements for which PDFs need to be estimated for uncertainty analysis purposes. Table 4 provides the seven top-emitting petroleum sources along with their year 1995 emissions used in the uncertainty analysis. Although the top seven sources were identified based on the year 2009 emissions estimate, EPA conducted the actual uncertainty analysis using estimates for the year 1995, because it was the base year (i.e., year of key input data collection) of the emissions and activity data estimates for many emission sources. In the 2011 GHGI, the above seven sources contributed nearly 94% of the total 1995 methane emissions from petroleum systems. To define the uncertainty model parameters (steps 1 and 2 in Figure 1) of every element of the activity and emission factors for the top seven sources, EPA combined judgments of an industry expert and a statistical expert along with data published in the 1999 EPA-Radian study. For all top seven sources, EPA assumed a lognormal PDF as default (except for oil tanks, for which EPA assumed a combination of normal and triangular distributions to represent inputs). Then, using the Monte Carlo simulation method in @ RISK (steps 3 through 5 in Figure 1), EPA calculated the upper and lower estimates representing the 95% confidence interval for each of the top seven sources. Table 4. Top Seven Emission Sources for Petroleum Systems in Previous (2011) Uncertainty Analysis for GHG Inventory Published in 2011 (2011 GHGI) Source 2011 GHGI CH4 Emissions, year 1995 (MMT CChe) Shallow Water Offshore Platforms (production segment) 16.1 High-Bleed Pneumatic Controllers (production segment) 9.0 Oil Tanks (production segment) 5.6 Page 6 of 27 ------- April 2018 2011 GHGI CH4 Emissions, year 1995 Source (MMT CChe) Low-bleed Pneumatic Controllers (production segment) 2.6 Gas Engines (production segment) 2.0 Chemical Injection Pumps (production segment) 1.3 Deep Water Offshore Platforms (production segment) 0.4 Other Emission Sources 2.6 Total Emissions from Petroleum Systems 39.7 For petroleum systems, the 2011 analysis assumed that uncertainty for these top seven emissions sources is an indication of uncertainty for the remaining emissions sources, and therefore extended the uncertainty of aggregate emissions estimates for the top seven emissions sources to the remaining sources. With that assumption, the overall uncertainty combining the top seven sources and remaining sources was re-estimated using the @ RISK model. To develop the uncertainty bounds for 1995, the upper and lower modeled estimates for the source category were translated to +/- percentages of the GHGI estimate. EPA calculated that for 1995, the 95% confidence interval for petroleum systems emissions is -24% and +149% of the GHGI-reported value. These +/- percentage values were assumed to represent the 95% confidence interval for all other years of the time series. 4. Updated Uncertainty Analyses for Natural Gas and Petroleum Systems in the 2018 GHGI (from October 2017 memo) In recent years, EPA has made significant revisions to the GHGI methodology to use updated activity and emissions data in calculating estimates for recent years of the time series. For the 2016 and 2017 GHGIs, EPA used multiple recently published studies as well as GHGRP Subpart W data to revise the emission factors and activity data for many natural gas systems emission sources and petroleum systems production segment emission sources. To update its characterization of uncertainty, EPA has conducted a draft quantitative uncertainty analysis similar to that conducted for the 2011 GHGI using the IPCC-recommended Approach 2 methodology (Monte Carlo Simulation technique). 4.1. Approach to Estimating Source-specific Uncertainty For its updated analysis, as in the 2011 GHGI analysis, EPA first identified a select number of "top" emission sources for each source category. Page 7 of 27 ------- April 2018 Table 5 and Table 6 show the top emission sources that cover at least 75% of gross emissions in natural gas and petroleum systems for year 2015, respectively, based on the 2017 GHGI. The top 14 natural gas systems sources cover approximately 77% of total source category emissions for the year 2015; the top 5 petroleum systems sources cover 79% of total source category emissions for the year 2015. EPA seeks stakeholder feedback on how many top emission sources to include in the detailed uncertainty analysis for each source category (see next section). Page 8 of 27 ------- April 2018 Table 5. Top 14 Natural Gas Systems CH4 Emission Sources in the 2017 GHGI Year 2015 Gross Emissions % of Source Emission Source (segment) (MMT C02 Eq.) Category Emissions G&B stations (production) 49.2 27% Pneumatic controllers (production) 25.5 14% Station total fugitives (transmission) 14.3 8% Engine combustion (transmission) 6.3 3% Engine combustion (production) 6.3 3% Engine combustion (processing) 5.8 3% Liquids unloading (production) 5.2 3% G&B episodic events (production) 4.9 3% Pipeline venting (transmission and storage) 4.6 3% G&B pipeline leaks (production) 4.0 2% Station venting (transmission) 3.8 2% Shallow water offshore platforms (production) 3.1 2% Chemical injection pump venting (production) 3.0 2% Separator fugitives (production) 2.9 2% Subtotal, Top Sources 139.1 77% Natural Gas Systems Total 181.1 100% Table 6. Top 5 Petroleum Systems CH4 Emission Sources in the 2017 GHGI Year 2015 Gross Emissions % of Source Emission Source (segment) (MMT CO2 Eq.) Category Emissions Pneumatic controllers (production) 18.6 48% Shallow water offshore platforms (production) 4.2 11% Associated gas venting and flaring (production) 3.7 9% Engine combustion (production) 2.3 6% Oil tanks (production) 2.0 5% Subtotal, Top Sources 30.8 79% Petroleum Systems Total 39.0 100% Next, EPA developed uncertainty model parameters based on published studies, GHGRP Subpart W data, expert consultation, and/or the 2011 uncertainty analysis for each of the top emission sources. Appendix A documents the uncertainty parameters values for the top sources in natural gas and petroleum systems used in conducting the Monte Carlo analysis, including: • Basis of the GHGI value, • Basis of the uncertainty parameter values, • PDF, • Point estimate (i.e., estimate in GHGI which is modeled as the mean or most likely value), and • Uncertainty range (e.g., standard deviation or minimum and maximum). If the modeling input (e.g., emission factor) was based on GHGRP subpart W data, EPA's general approach was to employ bootstrapping to determine the shape and other parameters of the sampling distribution of the mean value. The bootstrapping analysis enabled the determination of the PDF (e.g., normal, lognormal, triangular, etc.) as well as applicable statistical parameters (e.g., standard deviation, maximum, minimum, etc.) needed for the Monte Carlo simulation. Most model inputs from GHGRP were determined to have a normal PDF as expected due to the Central Limit Theorem.1 For modeling inputs based on recently published studies (i.e., Marchese et al. and 1 GHGRP subpart W data sets contain information (e.g., methane emissions, number of pneumatic controllers, etc.) submitted by hundreds of facilities which generally include the majority of activity in each industry segment (e.g., natural gas production). Hence, the bootstrap samples drawn from these GHGRP subpart W data sets were sufficiently large for the purposes of the CLT. Page 9 of 27 ------- April 2018 Zimmerle et al.), uncertainty information available in the study were directly used for the EPA's analyses.2 For modeling inputs based on older data sets (e.g., EPA/GRI study) or "macro parameters/' that are used as inputs to several emission source estimates (e.g., total active well count), EPA generally treated the input as a point estimate and referred to published estimates, the previous uncertainty analysis, and expert judgment to estimate upper and lower bounds. For input values obtained from certain data sources—for example, EIA or Drillinglnfo— EPA assigned default uncertainty bounds as documented in the Appendix A tables; EPA specifically seeks feedback on these default bounds in the next section. 4.2. Approach to Estimating Aggregate Uncertainty for Total CH4 Emissions Across All Sources In response to stakeholder feedback, EPA considered multiple approaches to: 1) estimating uncertainty for non- top sources; and 2) combining uncertainty for top and non-top sources to determine uncertainty for total CH4 emissions across all sources. Estimating Uncertainty for Non-top Sources • Simple Summation Approach: The uncertainty bounds for total CH4 emissions from non-top sources can be set equal to the approximated uncertainty bounds for total CH4 emissions from all top sources combined, which are computed using a simple summation approach. This approach was presented in the previous version of this memo. • Propagation of Error Approach: Similar to the simple summation approach, the uncertainty bounds for total CH4 emissions from non-top sources can be set equal to the approximated uncertainty bounds for total CH4 emissions from all top sources combined, which are computed using the propagation of error formulas (IPCC Approach l3). Propagation of error formulas are most suitable when errors for source- specific estimates being combined are uncorrelated and random. However, because certain source- specific parameters are used in estimating emissions from multiple top sources in both natural gas and petroleum systems, the error terms are expected to be correlated. • Monte Carlo Approach: The uncertainty bounds for total CH4 emissions from non-top sources can be estimated using a multi-step process that relies on Monte Carlo simulation as described below. Due to the existence of unmodeled sources, this approach is based on IPCC Approach 2, and similar to what EPA used in the 2011 GHGI petroleum systems uncertainty assessment. o Step 1 - The uncertainty bounds for total CH4 emissions from all top sources combined is determined using Monte Carlo simulation, o Step 2 - The probability density function (PDF) (i.e., normal, lognormal, etc.) along with its parameters (e.g., mean, standard deviation) of the total CH4 emissions from all top sources combined is derived by evaluating the obtained distribution in Step 1, o Step 3 - The PDF for total CH4 emissions from non-top sources can be approximated by the PDF obtained in Step 2 such that the adjusted mean corresponds to the estimated total CH4 emissions from non-top sources and the standard deviation corresponds to the product of the ratio of the estimated total CH4 emissions from non-top sources to-the estimated total CH4 emissions from all top sources combined. o Step 4 - The uncertainty bounds around total CH4 emissions from non-top sources can be estimated using Monte Carlo simulation. 2 In most cases, the needed uncertainty information (e.g., standard deviation, PDF type) had to be statistically imputed using information provided in the study. For example, when a study only reported 90% confidence bounds and the shape of the PDF, the standard deviation of the distribution needed for the Monte Carlo analysis had to be back-calculated. 3 See discussion beginning in Volume 1, Chapter 3, Section 3.2.3 of 2006 IPCC Guidelines for National Greenhouse Gas Inventories: http://www.ipcc-neeip.iees.or.ip/public/ep/enelish/6 Uncertaintv.pdf. Page 10 of 27 ------- April 2018 Combining Uncertainty for Top and Non-top Sources to Determine Uncertainty for Total CH4 Emissions Across All Sources • Simple Summation Approach: The uncertainty bounds for total CH4 emissions from top and non-top sources are combined using a simple summation approach. This results in the same 95% uncertainty bounds for the top, non-top, and overall (i.e., sum of total CH4 emissions from top and non-top sources) CH4 emissions. Results from this approach were presented in a previous version of this memo. • Propagation of Error Approach: The uncertainty bounds for total CH4 emissions from top and non-top sources can be combined by applying propagation of error formulas (IPCC Approach 1). Even though this approach is easily implementable, it is predicated on the assumption that errors for total CH4 emissions from top and non-top sources are uncorrelated and random. • Monte Carlo Approach: The uncertainty bounds for total CH4 emissions from top and non-top sources can be combined by using Monte Carlo simulations (IPCC Approach 2) where the PDF for total CH4 emissions from non-top sources is estimated as noted in the Monte Carlo Approach described above. 4.3. Results Table 7 and Table 8 below summarize calculated source category level uncertainty estimates for natural gas and petroleum systems, respectively, based on year 2015 emissions from the 2017 GHGI. Note that the reported uncertainty intervals only reflect those uncertainties that EPA has been able to quantify and do not incorporate considerations such as modeling uncertainty, data representativeness, measurement errors, misreporting or misclassification. According to the IPCC guidelines, if inventory source category uncertainties are correlated, the use of a stochastic simulation (the Monte Carlo method) is preferable (i.e., IPCC Approach 2) for combining uncertainties provided that the PDFs and correlation structure of the source-specific uncertainties are quantified. Thus, EPA is considering using the IPCC Approach 2 for both estimating uncertainty for non-top sources and combining uncertainty for top and non-top sources for the draft 2018 GHGI. Page 11 of 27 ------- April 2018 Table 7. Summary of Natural Gas Systems Year 2015 CH4 Uncertainty Draft Update Results3 Emission Source Mean Year 2015 Emissions (MT CO? Eq.) 2.5% Lower Bound of Mean Year 2015 Emissions (MT CO? Eq.) 97.5% Upper Bound of Mean Year 2015 Emissions (MT CO? Eq.) Value % Value % G&B Stations (Production) 49,192,568 44,648,570 -9% 53,729,999 9% Pneumatic Controllers (Production) High-bleed Pneumatic Controllers 2,368,036 1,120,764 -53% 3,993,889 69% Intermittent-bleed Pneumatic Controllers 22,380,215 12,772,278 -43% 33,786,265 51% Low-bleed Pneumatic Controllers 757,911 122,599 -84% 1,585,524 109% Station Total Fugitives (Transmission)15 Station, Incl. Compressor Components 2,934,282 1,943,931 -34% 4,265,232 45% Reciprocating Compressors 8,484,047 5,290,921 -38% 13,237,091 56% Centrifugal Compressor (Wet Seals) 1,424,742 867,790 -39% 2,301,693 62% Centrifugal Compressor (Dry Seals) 1,467,867 873,408 -40% 2,370,718 62% Engine Combustion (Production) 6,323,058 456,914 -93% 22,434,220 255% Engine Combustion (Transmission) 6,299,036 2,088,945 -67% 8,384,561 33% Engine Combustion (Processing) 5,806,032 1,959,080 -66% 7,438,841 28% G&B Episodic Events (Production) 4,879,055 189,869 -96% 26,068,511 434% Pipeline Venting (Transmission and Storage) 4,590,999 1,210,468 -74% 6,310,292 37% G&B Pipeline Leaks (Production) 4,038,975 1,537,376 -62% 6,857,736 70% Station Venting (Transmission) 3,849,139 735,035 -81% 7,436,119 93% Chemical Injection Pump Venting (Production) 3,034,943 1,999,797 -34% 4,106,240 35% Liquids Unloading With Plunger Lift (Production) 3,016,831 2,522,890 -16% 3,516,669 17% Liquids Unloading Without Plunger Lift (Production) 2,211,607 1,689,719 -24% 2,749,717 24% Shallow Water Offshore Platforms (Production) 3,086,499 453,544 -85% 5,728,239 86% Separator Fugitives (Production) 2,924,891 1,758,786 -40% 4,162,242 42% Total for Sources Modeled in Uncertainty Assessment Simple Summation 139,070,729 84,242,680 -39% 220,463,797 +59% Propagation of Error Approach 139,070,729 123,166,394 -11% 169,558,504 +22% Monte Carlo Approach 139,070,729 115,487,232 -17% 167,501,749 +20% Total for Sources Not Modeled in Uncertainty Assessment Simple Summation 23,354,601 14,147,148 -39% 37,023,204 +59% Propagation of Error Approach 23,354,601 20,683,734 -11% 28,474,512 +22% Monte Carlo Approach 23,354,601 19,141,032 -18% 28,226,024 +21% Source Category Total Simple Summation 162,425,331 98,389,827 -39% 257,487,001 +59% Propagation of Error Approach 162,425,331 146,298,291 -10% 193,340,018 +19% Monte Carlo Approach 162,425,331 138,228,918 -15% 190,790,170 +17% a - Referto footnote'd' of Table A1 for explanation of how a key modeling parameter was updated since the previously presented analysis, impacting calculated bounds shown in this table for transmission station fugitive sources. Page 12 of 27 ------- April 2018 Table 8. Summary of Petroleum Systems Year 2015 CH4 Uncertainty Draft Update Results Emission Source Mean Year 2015 Emissions (MT C02 Eq.) 2.5% Lower Bound of Mean Year 2015 Emissions (MT C02 Eq.) 97.5% Upper Bound of Mean Year 2015 Emissions (MT C02 Eq.) Value % Value % Pneumatic Controllers (Production) High-bleed Pneumatic Controllers 2,126,086 638,737 -70% 4,181,207 97% Intermittent-bleed Pneumatic Controllers 15,887,354 7,755,094 -51% 26,307,172 66% Low-bleed Pneumatic Controllers 619,806 83,942 -86% 1,442,226 133% Shallow Water Oil Platforms (Production) Shallow Water Oil Platforms 4,207,887 1,053,746 -75% 11,552,844 175% Associated Gas Flaring & Venting (Production) Associated Gas Flaring 2,642,647 1,209,503 -54% 4,459,852 69% Associated Gas Venting 1,062,962 75,820 -93% 2,626,185 147% Oil Tanks (Production) Large Oil Tanks with Flares 202,495 84,691 -58% 340,088 68% Large Oil Tanks with VRU 99,012 12,738 -87% 209,784 112% Large Oil Tanks without Controls 1,443,504 593,070 -59% 2,511,470 74% Small Oil Tanks with Flares 1,726 280 -84% 4,830 180% Small Oil Tanks without Controls 115,514 4,828 -96% 386,500 235% Large Oil Tank Separators with Malfunctioning Dump Valves 149,605 21,233 -86% 533,238 256% Gas Engine Combustion (Production) Gas Engine Combustion 2,254,932 36,467 -98% 7,215,509 220% Total for Sources Modeled in Uncertainty Assessment Simple Summation 30,813,532 11,570,150 -62% 61,770,904 +100% Propagation of Error Approach 30,813,532 21,469,826 -30% 44,926,168 +46% Monte Carlo Approach 30,813,532 19,504,086 -37% 44,868,886 +46% Total for Sources Not Modeled in Uncertainty Assessment Simple Summation 9,062,042 3,402,699 -62% 18,166,386 +100% Propagation of Error Approach 9,062,042 6,314,124 -30% 13,212,468 +46% Monte Carlo Approach 9,062,042 5,551,116 -39% 13,716,381 +51% Source Category Total Simple Summation 39,875,574 14,972,849 -62% 79,937,290 +100% Propagation of Error Approach 39,875,574 30,136,175 -24% 54,585,861 +37% Monte Carlo Approach 39,875,574 27,292,147 -32% 54,063,731 +36% Page 13 of 27 ------- April 2018 5. Requests for Stakeholder Feedback (from October 2017 memo) The EPA sought feedback on the questions below in the version of this memo released in October 2017. The EPA continues to welcome additional stakeholder feedback on these questions for potential updates to future GHGIs. 1. The following elements of EPA's general approach to uncertainty analysis: a. Performing a detailed uncertainty analysis for "top" sources that cover a specified percent (e.g., 75%) of gross emissions in natural gas (Table 5) and petroleum systems (Table 6) for year 2015, and extending the uncertainty of aggregate emissions estimates for the top emissions sources to the remaining sources (as illustrated in Table 7 and Table 8). b. Using a Monte Carlo Approach to calculate source category total uncertainty (as illustrated in Table 7 and Table 8). c. Calculating uncertainty for a select year, then assuming the same relative uncertainty as the 95% confidence interval for all other years of the time series. 2. The availability of additional information and data from statistical and industry experts that are relevant to characterizing the uncertainty parameters for the sources listed in Page 14 of 27 ------- April 2018 3. Table 5 and Table 6 and detailed in Appendix A. 4. How to compare estimated uncertainty ranges from different studies and measurement/calculation approaches, and important caveats and considerations. Appendix B of this memorandum compares the GHGI uncertainty ranges to uncertainty characterizations presented in several recently published studies. Which other studies have information on uncertainty that could be compared with the GHGI uncertainty ranges? 5. The uncertainty ranges in the GHGI reflect the uncertainty in the available emissions and activity data. Any systematic errors that may arise because of imperfections in conceptualization, models, measurement techniques, or other systems for recording or making inferences from data are not reflected in the uncertainty analysis because we lack information on them. Such errors, if they exist, could bias the results leading to either under- or over-estimates. The EPA requests additional information to characterize systematic errors in the GHGI, how and where these could be described in the GHGI, and how they could be incorporated into the uncertainty analysis. 6. Additional steps that could be taken to improve characterization of the PDFs. Most model inputs for this uncertainty assessment were determined from our analysis to have a normal PDF, as expected due to the Central Limit Theorem (CLT). As discussed earlier in this memo, the CLT states that the means of random samples drawn from a population with any type of distribution will be normally or near-normally distributed if the sample size is large enough (Mendenhall, Wackerly, & Scheaffer, 1990). The EPA seeks feedback on general approaches to consider for data sources for which sample sizes are comparatively small; the sampling methodology could be biased (e.g., may not result in a nationally representative sample); or only certain statistical parameters (e.g., mean and standard deviation) rather than the full underlying dataset were available in the source material. 7. While they do not provide overall uncertainty estimates for CH4 emissions from natural gas systems, Brandt et al. (2016) argue that uncertainty ranges in the previous GHG Inventories might be too narrow for some source categories due to existence of extreme distributions in natural gas data sets.4 EPA seeks feedback on approaches that would improve characterization of extreme distributions for the GHGI uncertainty analysis. 8. As shown in tables Al and A2, EPA has assigned default uncertainty bounds for point estimates obtained from certain data sources. EPA seeks feedback on these values: a. National estimates of gas production, gas consumption, and oil production (EIA): +/-1% b. National estimate of transmission pipeline miles (PHMSA): +/-1% c. National well count estimates (developed by EPA from Drillinglnfo data): +/- 5% 9. How improved uncertainty results can be used to target improvements for the GHGI. 6. References Allen, et al. (2013). David T. Allen, Vincent M. Torres, James Thomas, David W. Sullivan, Matthew Harrison, Al Hendler, Scott C. Herndon, Charles E. Kolb, Matthew P. Fraser, A. Daniel Hill, Brian K. Lamb, Jennifer Miskimins, Robert F. Sawyer, John H. 4 More specifically, they assert that "...(1) heavy-tailed distributions are a pervasive characteristic of natural gas leak size distributions; (2) natural gas leaks are more heavy-tailed than other natural and social phenomena, (3) the largest 5% of leaks are (by median expectation) responsible for over 50% of the leaked methane from a given source category; (4) the recent use of log-normal distributions to model the distribution of leaks within a source category is not supported and systematically underestimates the importance of large emitters; (5) heavier-than-log-normal distributions lead to larger uncertainty than currently included in official estimates; (6) robustly characterizing heavy-tailed distributions will require sample sizes much larger than currently used in most studies; (7) aggregating results across studies to improve accuracy and robustness is statistically challenging." Page 15 of 27 ------- April 2018 Seinfeld, et al. 2013. "Measurements of methane emissions at natural gas production sites in the United States." Proceedings of the National. Academy of Sciences, Vol. 110 17768-17773. 2013. Allen, et al. (2014a). David T. Allen, Adam P. Pacsi, David W. Sullivan, Daniel Zavala-Araiza, Matthew Harrison, Kindal Keen, Matthew P. Fraser, A. Daniel Hill, Robert F. Sawyer, and John H. Seinfeld. "Methane Emissions from Process Equipment at Natural Gas Production Sites in the United States: Pneumatic Controllers." Environmental Science & Technology, Vol. 49 633- 640. 2014. Allen, et al. (2014b). David T. Allen, David W. Sullivan, Daniel Zavala-Araiza, Adam P. Pacsi, Matthew Harrison, Kindal Keen, Matthew P. Fraser, A. Daniel Hill, Brian K. Lamb, Robert F. Sawyer, and John H. Seinfeld. "Methane Emissions from Process Equipment at Natural Gas Production Sites in the United States: Liquid Unloadings." Environmental Science & Technology, Vol. 49 641-648. 2014. Brandt, et al. (2016). Adam R. Brandt, Garvin A. Heath, and Daniel Cooley. "Methane Leaks from Natural Gas Systems Follow Extreme Distributions." Environmental Science & Technology, Vol. 5012512-12520. 2016. IPCC (2006). Montreal: Intergovernmental Panel on Climate Change, National Greenhouse Gas Inventories Programme. Guidelines for National Greenhouse Gas Inventories, Volume 1 General Guidance and Reporting. 2006. Lamb, et al. (2015). Brian K. Lamb, Steven L. Edburg, Thomas W. Ferrera, Touche Howard, Matthew R. Harrison, Charles E. Kolb, Amy Townsend-Small, Wesley Dyck, Antonio Possolo, and James R. Whetstone. "Direct Measurements Show Decreasing Methane Emissions from Natural Gas Local Distribution Systems in the United States." Environmental Science & Technology, Vol. 49 5161-5169. 2015. Littlefield, et al. (2017). James A. Littlefield, Joe Marriott, Greg A. Schivley, and Timothy J. Skone. "Synthesis of recent ground- level methane emission measurements from the U.S. natural gas supply chain." Journal of Cleaner Production, Vol. 148 118- 126. 2017. Lyon, et al. (2015). David R. Lyon, Daniel Zavala-Araiza, Ramon A. Alvarez, Robert Harriss, Virginia Palacios, Xin Lan, and Robert Talbot. "Constructing a Spatially Resoved Mehane Emission Inventory for the Barnett Shale Region." Environmental Science & Technology, Vol. 49 8147-8157. 2015. Marchese, et al. (2015). Anthony J. Marchese, Timothy L. Vaughn, Daniel J. Zimmerle, David M. Martinez, Laurie L. Williams, Allen L. Robinson, and Austin L. Mitchell. "Methane Emissions from United States Natural Gas Gathering and Processing." Environmental Science & Technology, Vol. 49 10718-10727. Mendenhall, Wackerly, & Scheaffer, 1990. William Mendenhall, Dennis D. Wackerly, and Richard L. Scheaffer. Mathematical Statistics With Applications. Pws Pub Co (1990). Zimmerle, et al. (2015). Daniel J. Zimmerle, Laurie L. Williams, Timothy L. Vaughn, Casey Quinn, R. Subramanian, Gerald P. Duggan, and Bryan Willson. "Methane Emissions from the Natural Gas Transmission and Storage System in the United States." Environmental Science and Technology, Vol. 49 9374-9383. 2015. Zavala-Araiza, et al. (2015). Daniel Zavala-Araiza, David R. Lyon, Ramon A. Alvarez, Kenneth J. Davis, Robert Harriss, Scott C. Herndon, Anna Karion, Eric Adam Kort, Brian K. Lamb, Xin Lan, Anthony J. Marchese, Stephen W. Pacala, Allen L. Robinson, Paul B. Shepson, Colm Sweeney, Robert Talbot, Amy Townsend-Small, Tara I. Yacovitch, Daniel J. Zimmerle, and Steven P. Hamburg. "Reconciling Divergent Estimates of Oil and Gas Methane Emissions." Proceedings of the National Academy of Sciences in the United States of America, Vol. 112 15597-15602. 2015. Page 16 of 27 ------- April 2018 Appendix A: Uncertainty Parameter Values for the Top Sources in Natural Gas and Petroleum Systems Used in Conducting Monte Carlo Analysis (from October 2017 memo) As described above, the national emissions estimate associated with a source category is computed as the product of the average emission factor and average activity factor for that source category. Thus, the uncertainty parameters presented in Tables A1 and A2 below are associated with the distribution of these average values; not the distribution of emissions from that source (for more information, see section "Updated Uncertainty Analyses for Natural Gas and Petroleum Systems in the 2018 GHGI: Approach" above). Table Al. Overview of Natural Gas Systems Year 2015 ChU Uncertainty Inputs for @RISK Modeling (from October 2017 memo) Emissions Calculation Input Input Basis a Uncertainty Basis PDF Mean or Point Estimate or Most Likely Value Standard Deviation b Lower Bound or Minimum C Upper Bound or Maximum C Macro Parameters National active gas well count, 2015 Drillinglnfo Expert Judgment, 5% Uniform 421,893 - 400,798 442,988 Methane content of natural gas for NE region GTI (2001) & EIA Statistical analysis of Allen et al. (2013) methane content data Normal 0.865 0.008 - - Methane content of natural gas for MC region GTI (2001) & EIA Statistical analysis of Allen et al. (2013) methane content data Normal 0.824 0.030 - - Methane content of natural gas for RM region GTI (2001) & EIA Statistical analysis of Allen et al. (2013) methane content data Normal 0.774 0.006 - - Methane content of natural gas for SW region GTI (2001) & EIA Statistical analysis of Allen et al. (2013) methane content data Normal 0.805 0.013 - - Methane content of natural gas for WC region GTI (2001) & EIA Statistical analysis of Allen et al. (2013) methane content data Normal 0.919 0.014 - - Methane content of natural gas for GC region GTI (2001) & EIA Statistical analysis of Allen et al. (2013) methane content data Normal 0.888 0.019 - - Default methane content of natural gas EPA/GRI (1996) Statistical analysis of Allen et al. (2013) methane content data Normal 0.788 0.008 - - Gas Wells for NE Region (2015) Drillinglnfo Expert Judgment, 5% Uniform 153,380 - 145,711 161,049 Gas Wells for MC Region (2015) Drillinglnfo Expert Judgment, 5% Uniform 79,645 - 75,663 83,627 Gas Wells for RM Region (2015) Drillinglnfo Expert Judgment, 5% Uniform 75,689 - 71,905 79,473 Gas Wells for SW Region (2015) Drillinglnfo Expert Judgment, 5% Uniform 45,370 - 43,102 47,639 Gas Wells for WC Region (2015) Drillinglnfo Expert Judgment, 5% Uniform 2,417 - 2,296 2,538 Gas Wells for GC Region (2015) Drillinglnfo Expert Judgment, 5% Uniform 65,392 - 62,122 68,662 G&B Stations (Production) scfd/station Marchese, et al. (2015) Statistical analysis of study data Normal 53,066 2,468 Region Marketed Onshore Production (MMCF), 2015 EIA EIA publication default, 1% Uniform 7,499,108 - 7,424,117 7,574,099 National Marketed Onshore Production (BCF), 2012 EIA EIA publication default, 1% Uniform 23,531 - 23,295 23,766 Page 17 of 27 ------- April 2018 Emissions Calculation Input Input Basis a Uncertainty Basis PDF Mean or Point Estimate or Most Likely Value Standard Deviation b Lower Bound or Minimum C Upper Bound or Maximum C NE Region Marketed Onshore Production (MMCF), 2015 EIA EIA publication default, 1% Uniform 4,443,949 - 4,399,509 4,488,388 MC Region Marketed Onshore Production (MMCF), 2015 EIA EIA publication default, 1% Uniform 5,087,452 - 5,036,577 5,138,326 RM Region Marketed Onshore Production (MMCF), 2015 EIA EIA publication default, 1% Uniform 2,177,308 - 2,155,535 2,199,081 WC Region Marketed Onshore Production (MMCF), 2015 EIA EIA publication default, 1% Uniform 521,702 - 516,485 526,919 GC Region Marketed Onshore Production (MMCF), 2015 EIA EIA publication default, 1% Uniform 7,554,759 - 7,479,211 7,630,306 Pneumatic Controllers (Production) scfd/controller (low bleed) Subpart WRY2015 & EPA/GRI (1996) Statistical analysis of reported Subpart W data; PDF per expert judgment; statistical parameters for emission rate imputed using the reported 90% confidence bound in EPA/GRI study Normal 23 10 - - scfd/controller (high bleed) Subpart WRY2015 & EPA/GRI (1996) Statistical analysis of reported Subpart W data; PDF per expert judgment; statistical parameters for emission rate imputed using the reported 90% confidence bound in EPA/GRI study Normal 622 100 - - scfd/controller (intermittent bleed) Subpart WRY2015 & EPA/GRI (1996) Statistical analysis of reported Subpart W data; PDF per expert judgment; statistical parameters for emission rate imputed using the reported 90% confidence bound in EPA/GRI study Normal 218 42 - - Fraction of total controllers that are low bleed Subpart WRY2015 Statistical analysis of reported data Normal 0.24 0.05 - - Fraction of total controllers that are high bleed Subpart WRY2015 Statistical analysis of reported data Normal 0.03 0.01 - - Fraction of total controllers that are intermittent bleed Subpart WRY2015 Statistical analysis of reported data Normal 0.73 0.05 - - Total controllers per well Subpart WRY2015 Statistical analysis of reported data Normal 1.88 0.23 - - Page 18 of 27 ------- April 2018 Emissions Calculation Input Input Basis a Uncertainty Basis PDF Mean or Point Estimate or Most Likely Value Standard Deviation b Lower Bound or Minimum C Upper Bound or Maximum C Station Total Fugitives (Transmission) scfd/station Zimmerle, et al. (2015) PDF per expert judgment; statistical parameters imputed using the reported 95% confidence bound Normal 9,104 1,269 - - Station scaling factor for national count based on subpart W count Zimmerle, et al. (2015); Subpart W RY2012 Statistical analysis of study datad Lognormal 3.52 0.5085 - - Engine Combustion (Transmission) scf/HPhr EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Triangular 0.240 - 0.045 0.323 1992 MMHPhr EPA/GRI (1996) PDF per expert judgment; statistical parameters imputed using the reported 90% confidence bound Normal 40,380 4,194 - - 2015 Total National gas Consumption (tril ftA3 / yr) EIA EIA publication default, 1% Uniform 27 - 27 28 1992 Total National gas Consumption (tril ftA3 / yr) EIA EIA publication default, 1% Uniform 20 - 20 20 Engine Combustion (Production) scf/HPhr EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Triangular 0.240 - 0.045 0.323 MMHPhr (for All gas wells in 1992) EPA/GRI (1996) PDF per expert judgment; statistical parameters imputed using the reported 90% confidence bound Lognormal 27,460 32,531 Total Gas Wells (2015) (excluded NE) Drillinglnfo Expert Judgment, 5% Uniform 268,513 - 255,087 281,939 Total Gas Wells (1992) (excluded NE) Drillinglnfo Expert Judgment, 5% Uniform 140,758 - 133,720 147,796 Engine Combustion (Processing) scf/HPhr EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Triangular 0.240 - 0.045 0.323 MMHPhr/plant Subpart WRY2015 Statistical analysis of reported data Normal 75.3 4.9 - - 2015 national plant count O&GJ (2014) O&GJ publication default, 1% Uniform 667 - 660 674 Liquids Unloading (Production) Fraction of wells that vent using plungers Subpart WRY2015 Statistical analysis of reported data Normal 0.100 0.008 - - Fraction of wells that vent without using plungers Subpart WRY2015 Statistical analysis of reported data Normal 0.068 0.008 - - scfy/plunger well Subpart WRY2015 Statistical analysis of reported data Normal 148,589 966 - - scfy/non-plunger well Subpart WRY2015 Statistical analysis of reported data Normal 160,411 562 - - Page 19 of 27 ------- April 2018 Emissions Calculation Input Input Basis a Uncertainty Basis PDF Mean or Point Estimate or Most Likely Value Standard Deviation b Lower Bound or Minimum C Upper Bound or Maximum C G&B Episodic Events (Production) Total CH4 Emissions from G&B Episodic Events (Gg/yr), 2012 Marchese, et al. (2015) PDF per expert judgment; statistical parameters imputed using the reported 95% confidence bound reported for 2012 estimate Lognormal 169 330 - - Marketed Onshore Production (MMCF), 2015 EIA EIA publication default, 1% Uniform 27,284 - 27,011 27,557 National Marketed Onshore Production (BCF), 2012 EIA EIA publication default, 1% Uniform 23,531 - 23,295 23,766 Pipeline Venting (Transmission and Storage) Mscfy/mile EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Triangular 31.7 - 2.7 47.8 Transmission Pipeline Miles PHMSA (2015) PHMSA publication default, 1% Uniform 301,748 - 298,731 304,765 G&B Pipeline Leaks (Production) miles/well for NE Region EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Normal 0.400 0.071 - - miles/well for MC Region EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Normal 0.620 0.110 - - miles/well for RM Region EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Normal 1.120 0.194 - - miles/well for SW Region EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Normal 1.120 0.195 - - miles/well for WC Region EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Normal 1.120 0.195 - - miles/well for GC Region EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Normal 1.120 0.198 - - miles (equally divided for each region) EPA/GRI (1996) PDF and mileage variability per expert judgment; statistical analysis of study data Normal 14,367 3,668 - - CH4 emissions, Bcf EPA/GRI (1996) Buildup of the four pipeline materials using Monte Carlo sampling informed by the 90% confidence intervals reported in the GRI 1996 documentation. Normal 6.6 2.1 - - miles EPA/GRI (1996) PDF per expert judgment; statistical parameters imputed using the reported 90% confidence bound reported Normal 340,200 20,700 - - Page 20 of 27 ------- April 2018 Emissions Calculation Input Input Basis a Uncertainty Basis PDF Mean or Point Estimate or Most Likely Value Standard Deviation b Lower Bound or Minimum C Upper Bound or Maximum C Station Venting (Transmission) mscfy/station EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Normal 4,359 1,787 - - Station scaling factor for national count based on subpart W count Zimmerle, et al. (2015); Subpart W RY2012 Statistical analysis of study datad Lognormal 3.52 0.5085 - - Chemical Injection Pump Venting (Production) #pumps/well Subpart WRY2015 Statistical analysis of reported data Normal 0.189 0.033 - - scfd/pump Subpart WRY2015 & EPA/GRI (1996) Statistical analysis of reported Subpart W data; PDF per expert judgment; statistical parameters for emission rate imputed using the reported 90% confidence bound in EPA/GRI study Normal 216.4 6.7 - - Shallow Water Offshore Platforms (Production) scfd/platform EPA (2015) Statistical parameters imputed using the reported PDF and 90% confidence bound in 2000 GOADS analysis (EPA (2005)) Normal 8,899 3,873 - - Shallow Water Gas Platforms BOEM (2011) & EPA (2008) Statistical parameters imputed using the reported standard deviation in 2011 uncertainty analysis (EPA (2010a)) Normal 1,973 17 - - Separator Fugitives (Production) # separators/well Subpart WRY2015 Statistical analysis of reported data Normal 0.685 0.045 - - scfd/separator (NE & MC Region) EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Normal 0.899 0.045 - - scfd/separator (except NE & MC Region) EPA/GRI (1996) PDF per expert judgment; statistical analysis of study data Normal 122.016 24.439 - - indicates not applicable a - Refer to the Natural Gas Systems 2018 annex tables (https://www.epa.gov/ghgemissions/additional-information-oil-and-gas-estimates-1990-2015-ghg-inventory-published-april) for more detailed documentation of the estimate basis, b - Applicable for "mean" input values (normal PDF). c - Lower and upper bounds are applicable for "point estimate" input values (uniform PDF). Minimum and maximum values are applicable for "most likely" input values (triangular PDF), d - Upon further review, EPA updated the estimated probability density function (PDF) for the station scaling parameter so that the mean of the PDF better corresponded to the estimated mean CH4 emissions for the source. As a result, the bounds for the sources within this category shifted to the left. The previous station scaling parameter had a lognormal PDF with a mean of 3.52, standard deviation of 0.51, and shift parameter of 1.59. These were determined by fitting a distribution to the original data provided in the Zimmerle et al. study. The shift parameter shifts the whole distribution to the right, in this case. The revised station scaling parameter used in this updated memo still has a lognormal PDF with a mean of 3.52 and a standard deviation of 0.51. However, a value of zero was used for the shift parameter to ensure that the mean of the distribution used in the Monte Carlo analysis was equivalent to the value used in computing the CH4 emissions for the source. This also better aligned with how the other lognormal PDFs were characterized in the remaining source categories (all with zero-value shift parameters). Page 21 of 27 ------- April 2018 Table A2. Overview of Petroleum Systems Year 2015 Cm Uncertainty Inputs for @RISK Modeling (from October 2017 memo) Emissions Calculation Input Input Basis a Uncertainty Basis PDF Mean or Point Estimate or Most Likely Value Standard Deviation b Lower Bound or Minimum C Upper Bound or Maximum c Macro Parameters National active oil well count Drillinglnfo Expert Judgment, 5% Uniform 586,896 - 557,551 616,241 Pneumatic Controllers (Production) scfd/controller (high bleed) Subpart WRY2015 & EPA/GRI (1996) Statistical analysis of reported Subpart W data; PDF per expert judgment; statistical parameters for emission rate imputed using the reported 90% confidence bound in EPA/GRI study Normal 621.73 100.20 - - Fraction of total controllers that are high bleed Subpart WRY2015 Statistical analysis of reported data Normal 0.03 0.01 - - scfd/controller (intermittent bleed) Subpart WRY2015 & EPA/GRI (1996) Statistical analysis of reported Subpart W data; PDF per expert judgment; statistical parameters for emission rate imputed using the reported 90% confidence bound in EPA/GRI study Normal 218.32 41.90 - - Fraction of total controllers that are intermittent bleed Subpart WRY2015 Statistical analysis of reported data Normal 0.70 0.07 - - scfd/controller (low bleed) Subpart WRY2015 & EPA/GRI (1996) Statistical analysis of reported Subpart W data; PDF per expert judgment; statistical parameters for emission rate imputed using the reported 90% confidence bound in EPA/GRI study Normal 22.85 9.51 - - Fraction of total controllers that are low bleed Subpart WRY2015 Statistical analysis of reported data Normal 0.26 0.07 - - Total controllers per well Subpart WRY2015 Statistical analysis of reported data Normal 1.002 0.195 - - Shallow Water Offshore Platforms (Production) scfd CH4/platform EPA (2015) Statistical parameters imputed using the reported PDF and 90% confidence bound in 2000 GOADS analysis (EPA (2005)) Lognormal 16,552 11,146 - - Total number shallow water GOM platforms BOEM (2011) & EPA (2008) Statistical parameters imputed using the reported standard deviation in 2011 uncertainty analysis (EPA (2010b)) Normal 1,447 10.45 - - Page 22 of 27 ------- April 2018 Emissions Calculation Input Input Basis a Uncertainty Basis PDF Mean or Point Estimate or Most Likely Value Standard Deviation b Lower Bound or Minimum c Upper Bound or Maximum c Oil Tanks (Production) scf/bbl (large with flare) Subpart WRY2015 Statistical analysis of reported data Normal 0.35 0.10 - - throughput fraction (large with flare) Subpart WRY2015 Statistical analysis of reported data Normal 0.55 0.05 - - scf/bbl (large with VRU) Subpart WRY2015 Statistical analysis of reported data Normal 0.47 0.21 - - throughput fraction (large with VRU) Subpart W RY2015 Statistical analysis of reported data Normal 0.201 0.04 - - scf/bbl (large uncontrolled) Subpart WRY2015 Statistical analysis of reported data Normal 7.90 2.18 - - throughput fraction (large uncontrolled) Subpart WRY2015 Statistical analysis of reported data Normal 0.18 0.03 - - scf/bbl (small with flare) Subpart WRY2015 Statistical analysis of reported data Lognormal 0.088 0.050 - - throughput fraction (small with flare) Subpart WRY2015 Statistical analysis of reported data Normal 0.019 0.007 - - scf/bbl (small uncontrolled) Subpart WRY2015 Statistical analysis of reported data Lognormal 2.3 1.4 - - throughput fraction (small uncontrolled) Subpart WRY2015 Statistical analysis of reported data Normal 0.05 0.02 - - scf/bbl (malfunctioning dump valves) Subpart WRY2015 Statistical analysis of reported data Lognormal 0.15 0.15 - - throughput fraction (malfunctioning dump valves) Subpart WRY2015 Statistical analysis of reported data Normal 0.932 0.025 - - % National throughput managed by tanks Subpart WRY2015 Statistical analysis of reported data Normal 0.63 0.07 - - National oil production EIA EIA publication default, 1% Uniform 3.44E+09 - 3.41E+09 3.48E+09 Associated Gas Venting and Flaring (Production) Fraction of total wells with venting or flaring Subpart WRY2015 Statistical analysis of reported data Normal 0.12 0.02 - - Fraction of wells with associated gas that flare Subpart WRY2015 Statistical analysis of reported data Normal 0.83 0.06 - - mscfy/flaring well Subpart WRY2015 Statistical analysis of reported data Normal 95 21 - - fraction of wells with associated gas that vent Subpart WRY2015 Statistical analysis of reported data Normal 0.17 0.06 - - mscfy/venting well Subpart WRY2015 Statistical analysis of reported data Normal 193 94 - - Page 23 of 27 ------- April 2018 Emissions Calculation Input Input Basis a Uncertainty Basis PDF Mean or Point Estimate or Most Likely Value Standard Deviation b Lower Bound or Minimum C Upper Bound or Maximum c Gas Engine Combusion (Production) scf/HPhr EPA/GRI (1996) Statistical analysis of GRI data (NETL, 2016) Triangular 0.24 - 0.04 0.32 compressors EPA/Radian (1999) PDF per expert judgment; statistical parameters imputed using the reported 90% confidence bound calculated for 1993 estimate Normal 3,097 1,522 - - MMhp-hr/compressor EPA/Radian (1999) PDF per expert judgment; statistical parameters imputed using the reported 90% confidence bound calculated for 1993 estimate Lognormal 6.30 4.95 - - indicates not applicable a - Refer to the Petroleum Systems 2017 annex tables (https://www.epa.gov/ghgemissions/additional-information-oil-and-gas-estimates-1990-2015-ghg-inventory-published-april) for more detailed documentation of the estimate basis, b - Applicable for "mean" input values (normal PDF). c - Lower and upper bounds are applicable for "point estimate" input values (uniform PDF). Minimum and maximum values are applicable for "most likely" input values (triangular PDF). Page 24 of 27 ------- April 2018 Appendix A: References for Tables A1 and A2 Allen et al. (2013) Measurements of methane emissions at natural gas production sites in the United States. Proc. Natl. Acad. Sci. U.S.A. 110, 17768-17773 (2013) Drillinglnfo Dl Desktop® Raw Data PLUS Download. Drillinglnfo, Inc. April 2016. EIA (various published data sets). Energy Information Administration, U.S. Department of Energy. Washington, DC. EPA/GRI (1996) Methane Emissions from the Natural Gas Industry. Prepared by Radian. U.S. Environmental Protection Agency. April 1996. EPA/Radian (1999) Methane Emissions from the U.S. Petroleum Industry. Prepared by Radian International. U.S. Environmental Protection Agency. February 1999. GTI (2001) Gas Resource Database: Unconventional Natural Gas and Gas Composition Databases. Second Edition. GRI- 01/0136. OGJ (2014) "Worldwide Gas Processing." Oil & Gas Journal, PennWell Corporation, Tulsa, OK. 2014. Marchese et al. (2015) Methane Emissions from United States Natural Gas Gathering and Processing. Environmental Science & Technology, 49, 10718-10727. 2015. PHMSA (2015) Transmission Annuals Data. Pipeline and Hazardous Materials Safety Administration, U.S. Department of Transportation, Washington, DC. BOEM (2011) Platform Information and Data. Bureau of Ocean Energy Management, U.S. Department of Interior. 2011. https://www.data.boem.gov/homepg/data center/index.asp. EPA (2005) Incorporating the Mineral Management Service Gulfwide Offshore Activities Data System (GOADS) 2000 data into the methane emissions inventories. Prepared by ICF International. U.S. Environmental Protection Agency. Exhibit 6, page 10. June 30, 2005. EPA (2008) Natural Gas Model Activity Factor Basis Change. Prepared by ICF International. U.S. Environmental Protection Agency. January 7, 2008. EPA (2010a) Natural Gas Uncertainty Model 11-30-10. Excel Spreadsheet. Prepared by ICF International. U.S. Environmental Protection Agency. Table 8. November 30, 2010. EPA (2010b) 2009 Petroleum Uncertainty Model 11-30-10. Excel Spreadsheet. Prepared by ICF International. U.S. Environmental Protection Agency. Table 8. November 30, 2010. EPA (2015) Inventory of U.S. Greenhouse Gas Emissions and Sinks 1990-2013: Update to Offshore Oil and Gas Platforms Emissions Estimate. Table 1, page 2. April 2015. Available online at: http://www.epa.gov/climatechange/ghgemissions/usinventoryreport/natural-gas-systems.html. Zimmerle, et al. (2015) Methane Emissions from the Natural Gas Transmission and Storage System in the United States. Environmental Science and Technology, 49, 9374-9383. 2015. Page 25 of 27 ------- April 2018 Appendix B: Comparison to Recently Published Studies Large amounts of data and information on natural gas and petroleum systems have recently become available, through the Greenhouse Gas Reporting Program (GHGRP) and external studies. In general, there are two major types of studies related to oil and gas GHG data: "bottom up" studies that focus on measurement or quantification of emissions from specific activities, processes and equipment (e.g., GHGRP data), and "top down" studies that focus on verification of estimates (e.g., aircraft and satellite studies). The first type of study can lead to direct improvements to or verification of GHGI estimates. The GHG Inventory estimates for oil and gas underwent extensive updates in recent years using data from these types of studies. The second type of study can provide general indications on potential over- and under-estimates. EPA reviews both types of studies for data that can inform GHGI updates. In this section, we compare the updated draft quantitative GHGI uncertainty estimates for CH4 emissions from natural gas and petroleum systems, using the ranges detailed in this memorandum and developed for the 2018 GHGI, to those reported in recently published studies that include a bottom up inventory component (see Tables B1 and B2). While both top down and bottom up studies often include assessments of uncertainty, a comparison of uncertainty information from studies that use a top down approach was not developed for this memorandum, and would require further considerations. All studies reviewed for uncertainty information used Monte Carlo simulation technique to examine uncertainty bounds for the estimates reported, which is in line with the IPCC recommended Approach 2 methodology. The uncertainty ranges in the studies listed in Tables Bl, B2, and B3 differ from those of EPA. However, it is difficult to extrapolate uncertainty ranges from these studies to apply to the GHGI estimates because the GHGI source category level uncertainty analysis is not directly comparable to source- or segment-specific uncertainty analyses in these studies. Further, the methodologies and data sources used in estimating CH4 emissions in these studies often differ significantly from the studies underlying GHGI methodologies. For example, the GRI/EPA study generally had smaller sample sizes and more rudimentary techniques for developing nationally-applicable emissions and activity factors from the collected data than the more recent bottom up studies that were used for 2015 estimates in the 2017 GHGI. Table Bl. Comparison of Quantitative Uncertainty Estimates for CH4 Emissions from Natural Gas Systems (MMT CO2 Eq. and Percent) Segment Study Year Emissions (MMT CO2 Eq.) Uncertainty Rangea MMT CO2 Eq. % Lower Bound Upper Bound Lower Bound Upper Bound All Segments, National EPA 2017 GHGI 2015 162.4 104.3 271.9 -36% 67% Production, Barnett Shale Lyon, et al. (2015)b 2013 3.6 3.37 3.87 -7% 6% Gathering Facilities, National Marchese, et al. (2015) 2012 42.4 37.76 47.09 -11% 11% Gathering, Barnett Shale Lyon, et al. (2015)b 2013 4.3 3.00 5.97 -30% 39% Processing, Barnett Shale Lyon, et al. (2015)b 2013 1.2 0.81 1.77 -33% 47% Trans. & Storage, National Zimmerle, et al. (2015) 2012 37.6 30.44 48.85 -19% 30% Trans. & Storage, National Lyon, et al. (2015)b 2013 0.4 0.28 0.55 -28% 39% Distribution, National Lamb, et al. (2015) 2013 9.8 NA 21.32 NA 117% Distribution, Barnett Shale Lyon, et al. (2015)b 2013 0.2 0.17 0.35 -18% 74% Oil and Gas, All Segments, Barnett Shale Zavala, et al. (2015)c 2013 12.9 10.5 16.0 -19% 24% All Segments, National Littlefield, et al. (2017) 2012 183.9 150.3 242.2 -24% 29% NA = Not available a The figures represent the 95 percent confidence intervals reported in each of the studies for the source. b The emission estimates reported are for the 25-county Barnett shale region, not the U.S. as a whole, and encompass natural gas and petroleum emissions. Therefore, the point estimates are not comparable to those reported in other studies and are italicized to emphasize such. c The Zavala et al. results represent both natural gas and petroleum activities. Page 26 of 27 ------- April 2018 Table B2. Comparison of Quantitative Uncertainty Estimates for ChU Emissions from Petroleum Systems (MMT CO2 Eq.and Percent) Segment Study Year Emissions (MMT CO2 Eq.) Uncertainty Rangea MMT CO2 Eq. % Lower Bound Upper Bound Lower Bound Upper Bound All Segments, National EPA 2017 GHGI 2015 39.9 14.8 80.5 -63% 102% Production, Barnett Shale b Lyon, et al. (2015) 2013 0.39 0.37 0.42 -6% 6% Oil and Gas, All Segments, Barnett Shale Zavala, et al. (2015)c 2013 12.9 10.5 16.0 -19% 24% a The figures represent the 95 percent confidence intervals reported in each of the studies for the source. b The emission estimates reported are for the 25-county Barnett shale region, not the U.S. as a whole, and encompass natural gas and petroleum emissions. Therefore, the point estimates are not comparable to those reported in other studies and are italicized to emphasize such. c The Zavala et al. results represent both natural gas and petroleum activities. Table B3. Comparison of Quantitative Uncertainty Estimates for CH4 Emissions from Specific Emission Sources from Natural Gas Systems (MMT CO2 Eq. and Percent) Segment & Emission Source Study Year Emissions (MMT CO2 Eq.) Uncertainty Rangea MMT CO2 Eq. % Lower Bound Upper Bound Lower Bound Upper Bound Production, National, Pneumatic Controllers EPA 2017 GHGI 2015 25.5 14.0 39.3 -45% 54% Production, National, Pneumatic Controllers Allen, et al. (2014a) 2012 15 9.9 26.3 -34% 75% Production, National, Chemical Injection Pump Venting EPA 2017 GHGI 2015 3.0 2.0 4.1 -34% 35% Production, National, Chemical Injection Pump Venting Allen, et al. (2013) 2011 1.7 0.9 2.5 -49% 47% Production, National, Liquids Unloading With Plunger Lifts EPA 2017 GHGI 2015 3.0 2.5 3.5 -16% 17% Production, National, Liquids Unloading Without Plunger Lifts EPA 2017 GHGI 2015 2.2 1.7 2.7 -24% 24% Production, National, Liquids Unloading With Plunger Lifts Allen, et al. (2014b) 2012 4.8 2.8 7.3 -42% 53% Production, National, Liquids Unloading Without Plunger Lifts Allen, et al. (2014b) 2012 2.0 1.3 4.0 -38% 100% a The figures represent the 95 percent confidence intervals reported in each of the studies for the source. Page 27 of 27 ------- |