The features of this study are:

• A unique partnership: Study design, data, and findings were all reviewed by the study team, Environmental Defense Fund, participating companies, and an independent Scientific Advisory Panel
• Direct access: Participating companies provided access to production sites and equipment, and assisted in the design of safe sampling protocols, making possible measurements of methane emissions, directly at the source
• First measurements: For several source categories, these data are the first reported direct, on-site measurements of methane emissions
• Multiple measurement methods: Downwind measurements at over 20% of the well completion sites and 13% of the production sites were used to confirm that potential sources were accounted for

The study team focused on measuring emissions from (i) well completion operations; (ii) pneumatic devices and equipment leaks of wells in normal operation, and (iii) liquids unloading and workovers of wells in production. The highest priority was on collecting data on well completion operations, since limited emission measurements were available for these operations prior to this study, and because engineering estimates of the magnitudes of these emissions varied widely.

Because the study team's highest priority was on collecting data for well completion operations, the team visited regions where large numbers of new wells were being brought into production. There were relatively few well workover and unloading events occurring in some of these regions. Only a few workovers were sampled, and for unloading, the study team focused on unloadings that were conducted on a known schedule. Some of these unloadings were one-time events for the wells.

Source types that were not measured included combustion devices such as the engine exhaust from compressors, and the exhausts from combustors and flares. In a few cases, downwind measurements allowed estimates to be made of methane associated with combustors and flares. These results are described in the study reports and indicated that combustors and flares were generally very efficient in combusting methane (>98% combustion efficiency).

A variety of measurements were made on the several different types of tanks employed at gas production sites. Emissions were measured, using temporary stacks, on open-top tanks that receive flowback fluids. Emissions from produced water and flowback tanks (which are typically enclosed) were also collected using temporary stacks. Permanent tanks at a production site (once the well starts producing gas for sale) may include oil or condensate and produced water tanks. These tanks may have flashing emissions in addition to working and breathing losses. Flashing emissions occur when higher pressure liquids from the separator are transmitted into these permanent tanks, which operate at atmospheric pressure. Emission points from such tanks include the weighted hatches on the top of the tank and also elevated or manifold vents that can only be accessed using lifts. Emissions for these types of tanks were not analyzed because access to the multiple potential emission sources on tanks would have required a lift at each site, severely limiting the number of sites that could have been visited. Emissions for such tanks were estimated using generally accepted models and are included in some of the data analyses.

Methane emissions were measured directly at 190 natural gas production sites in the Gulf Coast, Midcontinent, Rocky Mountain and Appalachian production regions of the United States. The sites included 150 production sites with 489 wells. In addition to the 150 production sites, 27 well completion flowbacks, 9 well unloadings, and 4 well workovers were sampled; the sites were operated by 9 different companies. The types of sources that were targeted for measurement account for two thirds of methane emissions from all onshore and offshore natural gas production, as estimated in EPA's national greenhouse gas emission inventory. Of the nine companies that provided sites for sampling, at least three companies provided sites in each of the regions.

While the data presented in this work represents one of the most extensive datasets available on methane emissions from current natural gas production activities, the sites sampled still represent a small fraction of the total number of sites nationwide. Representative sampling was believed to be achieved by:

• Selecting a large number of participant companies
• Selecting a range of geographic areas to sample
• Setting minimum number of sampling targets in each area

The nine companies that participated in this study included mid-size and large companies. While there are thousands of oil and gas companies in the U.S., the participants do represent a sizable sample of overall U.S. production and well count. Participants account for almost 12% of all US gas wells, account for 16% of gross gas production, and almost half of the new well completions. Representativeness cannot be assured. The companies volunteered, and were not randomly selected.

Randomization in the selection of sites was achieved in a variety of ways, depending on the type of source. For completions, the study team provided time windows when the measurement team would be available in certain regions and host companies identified completions that would begin as soon as possible after the study team arrived. In most cases this scheduling completely determined which sites would be sampled. To illustrate this, consider that the total number of well completions, nationwide in 2011, for all the participating companies combined, averaged roughly 10 per day. That meant that in any given production region, on any particular day, just one or two new completions, for all of the companies combined, was likely to be starting.

The time commitment associated with sampling completions was extensive. Completions lasted up to two weeks; sampling equipment set up and tear down by the study team required a day before and a day after the completion. Unloading, workover and production site sampling was much shorter in duration, typically a few hours to a half day. Consequently, sites selected for unloading, workover and production site sampling were selected based on proximity to completion sampling. Typically, a list of candidate sites was provided by the host company. If the list was too long to be entirely sampled in the allotted time, the study team selected sites based on an ability to sample as many sites as possible in the time available.

One exception to this pattern was for Gulf Coast sites, where the study team, based in Austin, Texas, could make day trips to production sites. For these sites, the study team randomly selected from hundreds of potential sites provided by host companies. A second exception was for unloadings. These events were difficult to schedule since they were often done, by site operators, immediately as needed. This often did not allow the study team to travel to the site and set up equipment prior to the unloading occurring. Therefore, special efforts were made to identify and sample unloadings that could be scheduled.

Downwind measurements were made at over 20% of the well completion sites and 13% of the production sites. In order to make downwind measurements, steady, moderate winds, suitable topography, and road access 200-1000 m downwind of the sites were needed. These constraints limited the sites that could be sampled using the dual tracer, downwind measurement method.

The full dataset is available and more information can be found at the web site of the Cockrell School of Engineering at the University of Texas at Austin: www.engr.utexas.edu/news/methanestudy.

Much uncertainty exists about the amount of methane emissions resulting from natural gas production and the focus in this work was on resolving that uncertainty, using direct measurements of emissions at the source of the emissions. Natural gas exploration and production operations can emit a variety of chemicals to the atmosphere. Expanding the chemicals targeted for measurements would have significantly expanded the scope and complexity of the study.

The study team made direct measurements at the sources of the emissions at 190 production sites. At over 20% of the well completion sites and 13% of the production sites, where wind conditions, topography and downwind access allowed, measurements were made downwind of the sites and total site emissions were estimated based on those measurements. So, for a significant subset of the sites, the study team had two independent measurements of total emissions at a site: the sum of direct source measurements made on site and total site emissions based on downwind measurements. These measurements were generally consistent with each other (details are available in the study reports).

The study team made measurements in the Appalachian, Gulf Coast, Mid-continent, and Rocky Mountain regions. Differences in the geological formations that are the source of natural gas and the condensates that are produced with natural gas can lead to differences in operating procedures and emissions. For example, differences in pressures may influence emissions from pneumatic devices; differences in pressures and gas compositions may lead to different equipment being used when a well is completed.

In general, regional differences in emissions were observed. For example, emissions from pneumatic devices and equipment leaks were lower in the Rocky Mountain region than in some other regions. Emissions from completion flowbacks were also lower in the Rockies. In contrast, the frequency of liquids unloading was much greater in the Rockies than in other regions. These differences may be due to differences in the geology, differences in equipment and operating procedures, differences in state regulation, or other factors. The study team has not attributed regional differences to particular causes, however, all of the study data are reported by region in the study dataset.

If emission factors from this work for completion flowbacks, equipment leaks and pneumatic pumps and controllers are assumed to be representative of national populations and are used to estimate national emissions, total annual emissions from these source categories are calculated to be 957 Gg methane (with sampling and measurement uncertainties estimated at ±200 Gg). The estimate for comparable source categories in the EPA national inventory is ~1200 Gg. Additional measurements of unloadings and workovers are needed to produce national emission estimates for these source categories. The 957 Gg in emissions for completion flowbacks, pneumatics and equipment leaks, coupled with EPA national inventory estimates for other categories, leads to an estimated 2300 Gg of methane emissions from natural gas production (0.42% of gross gas production). While the overall total of the estimates from this work and the EPA national emission inventory differ by only about 10%, the relative importance of some emission source categories differ. The measurements indicate a greater importance for pneumatic devices and equipment leaks and less for completion operations, than the most recent EPA analysis.

The EPA is aware of the measurements made in this work. The EPA notes on page 3-70 of the national inventory report released in April 2013,

"There are relevant ongoing studies that are collecting new information related to natural gas system emissions (e.g. GTI data on pipelines, University of Texas at Austin (UT Austin) and Environmental Defense Fund (EDF) data on natural gas systems). EPA looks forward to reviewing information and data from these studies as they become available for potential incorporation in the Inventory"

As new data emerge, such as the measurements reported in this study, EPA incorporates this information into its national inventories. This improved understanding, and replacement of best available estimates with measured data, can sometimes lead to significant changes in estimated emissions.

The GHGRP provides the first national, facility-level data on greenhouse gas emissions from across the oil and gas sector, however, the emissions reported through the GHGRP still often rely on estimates rather than direct measurements. This study focused on particular emission source types in the production sector and collected direct measurements on those sources rather than relying on empirical equations or default emission rates.

The focus of this study was to better characterize emissions from key sources. In comparing the results of this work to existing national data, the study team used the annual EPA greenhouse gas national emission inventory (see questions above), rather than the GHGRP. The EPA's national emission inventory has a much longer history than the GHGRP and emission estimates are made using a consistent and documented methodology.

The primary objective of the study was to collect measurements of emissions from natural gas production. Similar to the other national level emission study by the EPA and GRI in the early 1990s, the resulting dataset is necessarily a relatively small sample of a national population. For example, out of more than half a million natural gas wells in the United States, this study sampled roughly 500. Twenty seven well completions were sampled in this work (no measurement data for this source category were available prior to this work). For 2011, the EPA estimated that 8077 hydraulically fractured wells were completed.

In this work, measurements were used to develop either regional or national emission averages (e.g., leaks per well, emissions per pneumatic device, and emissions per completion). The choice of taking regional or national averages was based on the scientific judgment of the study team. National emissions were estimated by multiplying the average of emissions per well, device or event by the number of wells, devices or events. For example, the national emissions for well completions were estimated by multiplying the number of completions per region (from the EPA national greenhouse gas inventory) by the average emissions per region, then summing over all regions. Emissions from equipment leaks were estimated by multiplying the total number of wells in the region (from the EPA national greenhouse gas inventory) by a regional average of equipment leaks per well, then summing over all regions.

In some cases, such as well unloadings and workovers, the study team concluded that the measurement data were not sufficiently representative of national populations to justify a national estimate. The reasoning behind these decisions is described in the study reports. In all cases, the study reports contain all of the information and describe the assumptions used to develop the national estimates.

This study collected data on natural gas production systems, not on the entire natural gas supply chain or other fuel systems. Therefore, any comparisons made with other fuels necessarily rely on data and assumptions from other studies.

In addition, to compare greenhouse gas emissions of natural gas systems to greenhouse gas emissions from other fuels requires comparing the global warming potential of methane to the global warming potential of carbon dioxide. Since methane released to the atmosphere is slowly converted to carbon dioxide, comparing methane to carbon dioxide requires that the analyst choose a time frame over which to evaluate warming potential (e.g., immediate impact, impact over one to several decades, impact over a century).

With these cautions in mind, the study team notes that Alvarez et al (reported in 2012 in the Proceedings of the National Academy of Sciences; Volume 109, pages 6435-6440) estimated the methane leak rate, over the entire natural gas supply chain, that would need to be achieved for natural gas to have an immediate greenhouse gas benefit relative to other fuels. As a point of comparison, this study found a leak rate of 0.42% in natural gas and condensate production (methane emitted as a percentage of natural gas produced), which leads to a leak rate of 1.3% over the entire natural gas supply chain, using EPA estimates for the remainder of the supply chain, and 1.9% for "well to wheels" (i.e., when in use emissions from vehicles are included). Based on Alvarez et al.:

• Switching to natural gas combined cycle electricity generation from coal powered electricity generation results in immediate net climate benefits if methane leak rate in the natural gas supply chain is less than 2.9% (methane emitted as a percentage of natural gas produced); climate benefits increase over time
• Switching to natural gas vehicles from gasoline powered cars results in immediate net climate benefits at well to wheels methane leak rates less than 1.4%; climate benefits increase over time
• Switching to natural gas vehicles from gasoline powered cars results in net climate benefits in 100 years at well to wheels methane leak rates less than 3-4.5%, depending on whether the emissions are a long term conversion or a one-time event
• Switching to natural gas vehicles from diesel powered heavy duty vehicles results in immediate net climate benefits at well to wheels methane leak rates less than 0.9%; climate benefits increase over time

This study did not address health and safety implications of methane emissions and did not perform emission measurements of emissions that lead to photochemical smog.

The study team plans to make additional measurements of emissions from liquids unloading and pneumatic controllers. Sampling is being planned for types of operations and regions that extend the work reported here.

New EPA regulations require broad deployment of many of the operation practices on which measurements were performed in this study. Most notably, as observed in this study, reduced emission completions (RECs, sometimes referred to as green completions) were already being conducted at 66% of the completion flowbacks that the study team made measurements on. Data from this study, which demonstrate that 99% of methane that has the potential to be released can be captured or controlled using REC operations, indicate that requirements for RECs are expected to lead to emission reductions.

Information on the submission of manuscripts and the PNAS review process can be found on the PNAS website at:
http://www.pnas.org/site/aboutpnas/submiss.xhtml
and
http://www.pnas.org/site/misc/iforc.pdf