Article pubs.acs.org/est
Measuring Emissions from Oil and Natural Gas Well Pads Using the Mobile Flux Plane Technique Chris W. Rella,* Tracy R. Tsai, Connor G. Botkin, Eric R. Crosson, and David Steele Picarro, Inc., 3105 Patrick Henry Drive, Santa Clara, California 95054, United States S Supporting Information *
ABSTRACT: We present a study of methane emissions from oil and gas producing well pad facilities in the Barnett Shale region of Texas, measured using an innovative ground-based mobile flux plane (MFP) measurement system, as part of the Barnett Coordinated Campaign.1 Using only public roads, we measured the emissions from nearly 200 well pads over 2 weeks in October 2013. The population of measured well pads is split into well pads with detectable emissions (N = 115) and those with emissions below the detection limit of the MFP instrument (N = 67). For those well pads with nonzero emissions, the distribution was highly skewed, with a geometric mean of 0.63 kg/h, a geometric standard deviation of 4.2, and an arithmetic mean of 1.72 kg/ h. Including the population of nonemitting well pads, we find that the arithmetic mean of the well pads sampled in this study is 1.1 kg/h. This distribution implies that 50% of the emissions is due to the 6.6% highest emitting well pads, and 80% of the emissions is from the 22% highest emitting well pads.
■
INTRODUCTION Methane, the primary constituent of natural gas, is a potent greenhouse gas with a global warming potential of 28−86 times that of an equivalent mass of carbon dioxide (Myhre et al.2). When fugitive emissions are kept under control, methane represents a clean burning, high energy content fuel that can reduce carbon dioxide emissions relative to other more carbonrich fuels. However, when methane emissions are a significant fraction of natural gas production, the climate benefit of methane is reduced or eliminated (Howarth et al.;3 Alvarez et al.4). Natural gas production, transportation, and distribution encompass an extensive nationwide infrastructure involving hundreds of thousands of wells and millions of miles of pipeline, as well as processing and storage facilities. Estimating the fugitive methane emissions from such a complex and heterogeneous system is highly challenging. Recognizing this challenge and the importance of reducing the uncertainty of fugitive emissions estimates, several recent studies (Allen et al.,5 Eastern Research Group6) have focused on quantifying fugitive emissions at various stages of the process, from well-drilling to well pad production, through processing, storage, transport, and distribution. Among the different source categories, fugitive emissions from production well pads are poorly understood. The roughly half million natural gas wells in the United States (Energy Information Administration7) are distributed geographically in dozens of distinct basins, each with unique geophysics and resulting natural gas component mixtures that require different production and processing equipment. Furthermore, the regulatory burden for emissions reduction is different from state to state, leading to different practices and resulting © 2015 American Chemical Society
emissions. Finally, fugitive emissions are generally unintentional emissions, due either to human error or a process or component failure, making any attempt at accurate quantification via an emissions factor calculation difficult. Recent studies have made direct emissions measurements of well pads to in an effort to provide quantitative estimates of well pad emissions nationwide. In Allen et al.,4 measurements of well pad emissions were made using two methods. 150 production well pads were studied using the primary quantification method, called direct source measurement, in which leaks are individually identified using an infrared camera, and the methane emissions of all detected leaks were quantified using a Hi-Flow Sampler. Because not all leaks can be found with an infrared camera, the direct source measurement method tends to provide a lower limit on total emissions. The secondary method of dual tracer release was performed on a small subset (20) of the well pads. In the tracer release method (Shorter et al.,8 Czepiel et al.9), tracer gases (in this instance, nitrous oxide, and acetylene) are released from compressed gas cylinders at a controlled rate in the vicinity of potential sources on the well pad. The advantage of this methodology is that total well pad emissions are measured, including small leaks below the detection limit of infrared cameras and physically inaccessible sources. Even with this limited sample set, the emissions from the well pads show a great deal of variability. Out of 19 well pads where a clear plume was identified, the maximum emissions Received: Revised: Accepted: Published: 4742
December March 14, March 16, March 25,
19, 2014 2015 2015 2015 DOI: 10.1021/acs.est.5b00099 Environ. Sci. Technol. 2015, 49, 4742−4748
Article
Environmental Science & Technology
and u(z) = component of wind speed normal to the surface A as a function of height above the ground. The biggest challenge in implementing a surface flux integral of a plume is in retrieving the concentration measurements on the surface that intersects the downwind plume. Thoma et al.11 demonstrated quantification of emissions from landfills using open path measurements to obtain the concentration map. Closed path instruments are much simpler to deploy, but only provide measurements at a single point in space at a given time. Making measurements on a surface with closed path instrumentation has meant either using multiple instruments (which can be an expensive and ultimately impractical solution), or multiplexing multiple spatial points into a single instrument using a multiport valve manifold (which has the disadvantage that the measurements are not performed simultaneously). We have circumvented this limitation in the MFP method by using a novel gas storage manifold to sequentially analyze several inlet ports. The MFP method consists of (a) deploying a mast with 4−6 gas inlet ports on a vehicle, (b) connecting each of these inlet points to long tubes that can store approximately 50 s of gas, (c) driving the vehicle through the downwind plume, thus tracing out a measurement surface defined by the mast and the path of the vehicle, (d) measuring the gas stored sequentially with a single high-accuracy methane analyzer, and (e) combining the concentration data with measured position of the vehicle to retrieve the concentration map on the surface. This concentration map, together with realtime measurement of the wind through the surface, allows a direct calculation of the methane flux through the surface. The MFP measurement has distinct advantages over other techniques. First, knowledge of the location of the source is not required for the calculation of the emission rate. In addition, only a local measurement of the wind through the surface is required; there is no need to employ an atmospheric transport model. Obstructions and terrain effects that distort the plume do not affect the flux plane measurement provided that the flux plane captures the full extent of the plume. Unlike inverse methods, multiple upwind sources are automatically aggregated into a single leak rate without affecting the accuracy. Finally, the concentration map provides clear evidence of whether the plume is fully captured. A schematic of the complete MFP system is shown in Figure 1. The methane analyzer is modified version of a commercially available near-infrared laser-based Cavity Ring-Down Spectrometer (CRDS) (Model G2301, Picarro, Santa Clara, CA, USA, described in Crosson12). The analyzer is capable of measuring ambient methane concentrations in real-time (2 Hz acquisition rate) with subparts-per-billion accuracy and precision. In addition to measuring real-time methane concentrations with the CRDS instrument, wind speed and direction were measured with a sonic anemometer (Model 102779, Climatronics, Bohemia, NY, U.S.A.). The apparent wind speed as measured from atop the car (1 m above the vehicle roof) is corrected in real-time for the motion of the vehicle to reconstruct the true wind speed and direction with respect to the ground. A submeter-resolution GPS unit (R100, Hemisphere, Scottsdale, AZ, U.S.A.) is used to provide realtime location and car speed data. Air is sampled using a gas inlet tube mounted in front of the car and located 1.7 m above the ground. In addition to this monitoring gas inlet, there are either four (for the validation experiments) or six (for the Barnett shale study) additional gas
observed was 4.16 SCFM and the minimum was 0.08 SCFM, with a range of a factor of 50 between the two. The distribution is clearly non-Gaussian, with relatively few emitters responsible for the bulk of the emissions: the top 4 emitters (20%) account for 55% of the emissions. As with any skewed population, accurate assessment of the aggregated emissions requires sufficient sampling of the entire population, especially the relatively rare tail of strongly emitting sources. As the distribution becomes more skewed, larger sample sets (on the order of hundreds) involving more well pads are required. Quantifying emissions from well pads presents a significant technical challenge. Direct measurement using a Hi-Flow sampler and infrared requires physical access, is timeconsuming, and not all sources are be identified or measured. Tracer release also requires site access and use of compressed gas bottles, which are both significant impediments to rapid measurement surveys of well pads. Inverse methods form another general category of quantification techniques. In an inverse method, an atmospheric transport model is used in conjunction with downstream concentration measurements to calculate upwind emissions. Thoma et al.10 employed such an inverse method from a mobile platform to quantify emissions from well pads in Texas, Colorado, and Wyoming. The technique requires knowledge of the source distance and height above the ground, as well as an accurate assessment of the atmospheric turbulence and advective wind flow. Furthermore, terrain and the physical arrangement of tanks and other physical obstructions like trees and fences can modify the wind field and thus bias the emissions measurement. In this paper, we present a new methane emission quantification technique called the mobile flux plane (MFP) method capable of rapid, efficient quantification of total well pad emissions. In this technique, a vehicle equipped with a GPS, an anemometer, and a methane analyzer is driven downwind of well pads. A complete measurement of the total emissions from the well pad can be made in 6 min and does not require site access. Below, we first describe the MFP method. We then summarize controlled release experiments that validate the accuracy and precision of the technique. We then present a field demonstration of the technique, in which the emissions from about 200 well pads in the Barnett Shale, Texas were measured during a two-week campaign1 using the MFP vehicle. We conclude the paper with a discussion of the results and the implications for well pad emissions in this basin.
■
EXPERIMENTAL SECTION The MFP method is a ground-based mobile methane measurement system that measures the methane flux through a vertically oriented measurement surface to infer emissions at a distance while driving downwind on roads near oil and gas producing facilities. Given an observed concentration profile, the flux of methane passing through the plane may be described as follows: q=
∬A k(C(y , z) − Co)u(z) dydz
(1)
where q = measured flux of methane passing through the plane; A = the surface over which the measurement is made; k = factor for unit conversion from a volume of methane to kg hr−1; C(y,z) = methane concentration as a function of horizontal position, y (along the axis of the vehicle’s motion) and height above the ground, z; Co = background methane concentration; 4743
DOI: 10.1021/acs.est.5b00099 Environ. Sci. Technol. 2015, 49, 4742−4748
Article
Environmental Science & Technology
speed varies vertically and is different for each of the six sampling inlets, the function u(z) is the vertical wind-speed gradient. For this work, we used a wind gradient of the form (u(z)/u0) = (z/zo)α. The coefficient α was determined to be 0.17 ± 0.05 in an experiment described in the Supporting Information (SI), u0 is the wind speed measured by the 2-D anemometer mounted on the vehicle roof, and zo is the height of this anemometer above the ground. The error the flux due to uncertainty in the vertical wind speed gradient is estimated to be no more than 10% (95% coverage). We note that this wind gradient corresponds to relatively open field conditions. Physical obstructions on or near the road would distort the wind field and lead to erroneous flux estimates. We have not quantified this error for these measurements, although we note that these measurements were made in largely rural environments with relatively few trees, buildings, parked vehicles, and other obstructions. Future work will include multiple vehicle mounted anemometers to better assess the transverse wind field. We performed a series of controlled methane release experiments under a variety of atmospheric conditions to validate the performance of the MFP system. These experiments are described in detail in the SI. Two plume integration techniques have been explored: (a) a trapezoidal approximation to quantify the portion of the plume that intercepts the MFP inlet mast, which represents a reliable lower bound on emissions, and (b) a limited Gaussian plume model to estimate emissions outside the physical dimensions of the flux plane. For the conditions under which the validation work was performed, the 4 m mast captures about 50−80% of the total emissions, with the remaining flux either slipping below the bottom inlet or above the top inlet on the mast. Using a modified Gaussian model of the plume in which the vertical width of the measured instantaneous plume is determined from a fit to the measured concentration data (i.e., it is not determined from atmospheric conditions), we are able to extrapolate this additional flux accurately, as demonstrated in the discussion in the SI. We have selected the Gaussian plume model method for estimating flux for the Barnett measurements. It is important to note that although the instantaneous plume is not necessarily fit well by this simple Gaussian model, the errors induced by this assumption have been characterized in the validation experiments and are captured in our stated precision and accuracy. From these validation experiments, we have demonstrated a measurement precision characterized by the geometric standard deviation of 1.9, an accuracy of 24%, and a detection limit of 0.034 kg/h (at a mean distance of 34 m). A geometric standard deviation of 1.9 means that 67% of measurements are within a factor of 1.9 of the geometric mean μ (i.e., between μ ÷ 1.9 and 1.9 × μ), and that 95% of measurements are between a factor of 1.92 = 3.6 of μ. For sources with emissions far above the detection limit (i.e, ≫ 3.6 times the detection limit of 0.034 kg/ h), the detection efficiency is >95%. However, at or near the detection limit, the detection efficiency drops. We now consider MFP measurements performed in the Barnett Shale during a 16-day field campaign to quantify well pad emissions. During this campaign, we did not have physical access to well pads or to the private lease roads leading from public roads to the well pads, and so we were restricted to performing measurements on public roads downwind of the sources. Given the relatively low density of public roads in the majority of the oil and gas producing region, and given the practical detection distance of about 150 m in the downwind
Figure 1. Schematic of the vehicle with forward sampling pole with 6 inlet ports, a 6 port sampler with 50 s gas storage time, and a 2 Hz CH4 analyzer. The motion of the sampling pole through space as the vehicle moves creates the plane through which the flux is measured. (Note, the validation experiments were performed with a 4 inlet port version of the instrument). Geospatial position and ambient wind speed and direction are measured at 1 Hz. During survey mode, the instrument monitors the concentration at the monitoring inlet. When a plume is detected, the system switches to reanalysis mode and measures the samples stored in the six storage tubes in sequence. Reanalysis takes about 5 min.
inlets evenly spaced along the 4.1-m (12-ft) mast. Forward mounting of the inlets means that the turbulent effects of the vehicle motion on the plume are minimized. As the vehicle moves through the plume, the air is simultaneously sampled at each of the six inlets and stored into individual aluminum tubes called AirCores (Karion et al.13), one for each inlet, each with a length of 46 m and inner diameter of 0.48 cm (V = 832 cm3). Given this volume and sampling rate of 1000 sccm, each AirCore is capable of storing a gas sample for 50 s. The system is designed such that when a plume signal is detected on the instrument monitoring intake, sampling from the six inlets is suspended and each AirCore is individually and sequentially pumped into the CRDS instrument and analyzed, with a total analysis time ∼5 min at a flow of 1000 sccm. Alternatively, the resampling of the AirCores can be initiated manually, which was the primary mode of operation for this study. For each inlet port, a horizontal distance axis is applied based upon the measurement time and the vehicle velocity during the capture of the plume. The reconstructed plume is shown in Figure 2 as an interpolated heat map, in which each horizontal strip of dots correspond to an inlet. In this case, the plume is clearly well-captured by the vehicle inlet mast. These interpolated images of the measured plume are used to assess the quality of the measurement and quickly judge in real-time whether or not the plane swept out by the mast captured the plume. The flux is calculated using eq 1. Because the wind-
Figure 2. A typical plume image constructed in real-time by the analysis software after the measurement is made. This is an interpolated image. The dotted lines show the vertical locations of each of the respective gas inlets. 4744
DOI: 10.1021/acs.est.5b00099 Environ. Sci. Technol. 2015, 49, 4742−4748
Article
Environmental Science & Technology
quantify the minimum leak rate, using the trapezoidal integration method described in the SI).
direction, only a small subsample of all the oil and gas well pads in the region could be sampled using the MFP vehicle. That practical limitation impeded our ability to select a truly representative subsample from the greater population. As a result, the emissions of the subpopulation of measured well pads are biased to those located near public roads and may not be representative of the methane emissions of the greater Barnett Shale production area. During the field campaign (Oct. 15−30, 2013), we collected N = 207 flux measurements, distributed as follows in the nine county region centered on the major production areas: Cooke, Denton, Hood, Johnson, Montague, Parker, Tarrant, Wise, and Somervell. Surveys were planned to maximize the number of facilities that would be within the measurement range of the instrument from public roads. To guide our survey route, we estimated which facilities would be observable based on opensource maps of public roads, a database of well locations (Zavala-Araiza et al.14), and the forecasted range of wind directions. Each day, routes were chosen based on these predictions, but since the accuracy of the predictions was not perfect, any well pad facility that was found to be within range of the instrument along the planned survey route was measured. The distance between the measurement and source well pad was calculated using the MFP mobile system’s GPS data and the EDF-provided database of well locations. If there were multiple wells on a single well pad, then the average of the GPS coordinates in the database was used as the coordinates of the well pad. For this set of tests, only one measurement was made of each well-pad; it is important to note that we therefore do not have an independent assessment of the measurement uncertainty. To ensure that our sample contained only measurements of well pads and no other leak sources, we selected a subsample of the collected measurements according to the following selection criteria: • The mean lateral wind speed during the plume measurement was greater than 1.0 m/s (N = 200). This limit also effectively ensured that the angle between the advective plume and the normal vector of the flux plane was within the ±60 degrees employed in the validation experiments; • The measurement could be clearly attributed to a single well pad in both the well pad database as well as the satellite image, based on the on-board recorded mean wind direction (N = 177); • The estimated distance to the source well pad was 150 m or less (N = 150); • The centroid of the plume as obtained from the Gaussian fit of the peaks of the individual inlet measurements was below the top inlet at 4.2 m (N = 142); and • The width of the plume as obtained from the Gaussian fit of the peaks of the individual inlet measurements was less than 5 m (N = 115), which is approximately the height of the pole. 55% of all plume measurements satisfied these five criteria. Geographically, these measurements were distributed in the nine county region as follows: Cooke (1), Denton (36), Hood (1), Johnson (32), Montague (4), Parker (4), Tarrant (10), Wise (26), and Somervell (1). Forty-five percent of the measurements do not satisfy these data quality criteria, and therefore cannot be used for accurate quantification (although they can in principle be used to
■
RESULTS AND DISCUSSION The distribution of the N = 115 measurements as a function of distance is shown in the top panel of Figure 3. The average
Figure 3. top panel. Distribution of distances between the emission source and the measurement point for the N = 115 MFP measurements that meet all the data quality criteria (see text for more information). Bottom panel: emissions (on a log scale) vs distance.
distance between the source and the measurement location for these samples is 79 m. Note that of the well-pads nearer than 150 m (N = 150), 77% met the plume centroid and width criteria (N = 115); by comparison, 84% (101 out of 120 measurements) of the validation experiments met the same criteria, which is not surprising, given the mean distance of 34 m for the validation experiments vs 79 m for the field campaign. We observe a tendency toward fewer detections at greater distances, indicating a reduction in the number of well pad measurements that meet the data selection criteria, as expected given the increasing size of the plume with propagation distance. The bottom panel shows the individual source emissions measurements on a log scale vs distance, using the GF integration method. There is no discernible bias in the measured emissions with distance. The largest measured flux was 47.6 kg/h; the smallest nonzero flux measured was 0.027 kg/h. The emissions distribution for these N = 115 well pads is shown in Figure 4, using the Gaussian fit method described in the SI. The red curve is the modeled log-normal distributions, with geometric mean of 0.63 kg/h, geometric standard deviation of 4.1, and arithmetic mean of 1.74 kg/h. This is a highly skewed distribution, with the geometric mean differing from the arithmetic mean by a factor of about 2.5. In addition, we have included the precision of the measurement (yellow curve) as derived from the controlled release work. The controlled release work was performed primarily under Pasquill Gifford atmospheric stability classes (De Nevers15) A,B; the Barnett work was more in the B,C range; therefore, the controlled release precision should represent the worst case 4745
DOI: 10.1021/acs.est.5b00099 Environ. Sci. Technol. 2015, 49, 4742−4748
Article
Environmental Science & Technology
• This distance boundary has been scaled with wind speed such that there should be an equal probability of detection via a peak concentration threshold regardless of wind speed. • Using this detection distance boundary, the path our vehicle traversed, and the mean wind direction, we have generated a geospatial area in which we are likely to have seen a leak of 0.08 kg/h or greater. • Using the well pad database validated against satellite imagery, we have identified all well pads in the detection area, discarding well pads where the wind bearing was too oblique to the road direction to be confident that the plume intersected the road. A total of 193 well pads satisfying this criterion were found. • Of these 193 well pads, 71 had no measurable signal from the public road downwind of the plume. Therefore, the fraction of nondetected emissions is 37%. • The mean emission rate per well pad of 1.74 kg/h should be reduced to 1.74(1−0.37) = 1.10 kg/h assuming that the emissions from the 37% of pads that are not detected is zero, or to 1.74 (1−0.37) + 0.08(0.37) = 1.13 kg/h, assuming that these 37% of well pads emit 0.08 kg/h. Using bootstrap resampling (Mooney and Duval16), we have determined the following parameters for the measured lognormal distribution of well-pad emissions (1-sigma): geometric mean of 0.63 ± 0.09 kg/h, geometric standard deviation of 4.1 ± 0.4, and an arithmetic mean of 1.74 ± 0.35 kg/h, not including the well pads from which emissions were not detected (we will include these sources at a later point in the analysis, below). Given these statistics, 95% of the sources have emissions are between 0.63/4.12 = 0.037 kg/h and 0.63 × 4.12 = 10.5 kg/h. Since the validation measurements indicate that the precision of the measurements also has a log-normal distribution, the effect of the precision can be removed via quadrature subtraction. We have performed a Monte Carlo analysis to remove the validation distribution from the measured distribution, which leaves the following actual distribution of leaks: geometric mean of 0.72 ± 0.11 kg/h, a geometric standard deviation of 3.5 ± 0.4, and an arithmetic mean of 1.63 ± 0.33 kg/h, with a 95% confidence interval of 1.09 to 2.38 kg/h. Note that the arithmetic mean is decreased by a factor of 1.07 from the measured distribution, due to the slight overestimation of the controlled release values observed during the validation experiments. This distribution is highly skewed, leading to statistical properties that have important implications for quantification of well pad emissions. If we make the simplifying assumption that the distribution is perfectly log-normal with the central parameters above, then we can derive the cumulative total of the emissions as a function of the population of emitters (in order of decreasing emissions). This analysis is shown in Figure 5, where we have assumed that 37% of all sources have zero leak rate. From this graph, we may draw the following conclusions about the distribution of well pad emissions: • 10% of the total emissions is from the top 0.3% of the sources; • 20% of the total emissions is from the top 1.1% of the sources; • 50% of the total emissions is from the top 6.6% of the sources; • 80% of the total emissions is from the top 22% of the sources; and
Figure 4. Histogram of the natural logarithm of the measured emission rate in kg/h. The red curve is the modeled log-normal distribution; the precision of the measurement (yellow) as derived from the controlled release work (see SI). We also show the detection probability for this measurement set (gray dot-dashed line), scaled by a factor of 10.
precision, although we cannot rule out the possibility that the precision of the field measurements are less precise, since the measurements were made in different locales, under different atmospheric conditions, with different obstacles and terrain. The distribution is much wider (>2X) than the instrument precision, but the logarithmic scale conceals the actual difference between the two distributions. Contrast the geometric standard deviation of 1.9 for the validation experiments to the geometric standard deviation of 4.1 for the field measurements, which means that the 95% width of the distribution of the field measurements is a factor of 21 times broader than the 95% width of the validation measurements (21 = (4.1/1.9)4). In other words, the distribution of well-pad emissions is dramatically broader than the measurement precision, which limits the sensitivity to errors in the measurement of the measurement precision. We also display the detection probability for this measurement set (gray dotdashed line), scaled by a factor of 10, with a detection limit (50% chance of detection) of 0.045 kg/h, obtained using the scaling factor obtained from the controlled release experiments and described in the SI. Note that detection probability does not affect the emissions distribution above about 0.15 kg/h. Below this value, the detection probability drops off, as does the distribution of source emissions. This implies both that the detection limit of 0.045 kg/h is approximately correct, and that the low end of the emissions distribution may be underrepresented in this sample set. We have estimated the fraction of well pads that have emissions below the detection limit of the instrument, using the following procedure: • We have selected a nominal detection distance boundary of 85 m and a minimum leak rate of 0.08 kg/hour. Under the mean wind conditions encountered in this campaign, the probability of detection of this level of leak should be about 50% or greater at this distance or closer, for this leak rate, using the results from the validation experiment to determine the detection probability. 4746
DOI: 10.1021/acs.est.5b00099 Environ. Sci. Technol. 2015, 49, 4742−4748
Environmental Science & Technology
■
Article
ASSOCIATED CONTENT
S Supporting Information *
A detailed description of the validation of the mobile flux plane technique using controlled release experiments. This material is available free of charge via the Internet at http://pubs.acs.org.
■
AUTHOR INFORMATION
Corresponding Author
*Phone: (408) 962−3941; fax: (408) 962-3200; e-mail: rella@ picarro.com (C.W.R.). Author Contributions
The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS The authors would like to thank Eben Thoma at the U.S. Environmental Protection Agency, who performed controlled releases of methane in support of this research. We would also like to thank Pieter Tans, Colm Sweeney, Anna Karion, and Tim Newberger of NOAA for the development of the AirCore technology used in this work. Finally, we would also like to acknowledge the support of the Environmental Defense Fund, who provided partial support for this research.
Figure 5. Fraction of total emissions from a well pad population with a geometric standard deviation of 3.5, and with 37% of all sources having zero emissions, plotted against the fraction of the well pad population, in order of decreasing emissions.
• The bottom 50% of the sources contribute less than 2% of the total emissions. These measurements were made on a population of 115 emitting sources and 67 nonemitters. Given that 20% of the emissions is due to the top 1.1% of sources, this would indicate that we have only measured perhaps one or two well pads in this high emission bin. In other words, we are under-sampling the high emitting tail of the distribution; the actual distribution may deviate from the derived log-normal distribution, either higher or lower than the predicted distribution. Further work is needed to identify the population of high emitters which may further skew the distribution. We further note that because we were not able to sample well pads in a fully random fashion, there are potential biases built into this distribution that may not be reflected in the overall population of well pads in the Barnett Shale. Lease road and/or well pad access would be required to perform a more unbiased well pad survey. We have presented an analysis of methane emissions from oil and natural gas producing well pad facilities in the Barnett Shale region measured using an innovative mobile flux plane (MFP) measurement system. For those well pads with nonzero emissions, the log-normal distribution is highly skewed, with an arithmetic mean of 1.72 kg/h and with 95% of the well-pads emitting between 0.037 and 10.5 kg/h. Including the population of nonemitting well pads, we find that the arithmetic mean of the well pads sampled in this study is 1.1 kg/h. If our subsample is representative of the larger population of well pads, then this distribution would indicate that 50% of the emissions is due to the 6.6% highest emitting well pads, and 80% of the emissions is from the 22% highest emitting well pads. This result, if verified, has important implications for infrastructure integrity management of well pads, suggesting that deploying MFP systems to rapidly and accurately assess the largest emitters in a production field can lead to an ∼80% reduction in emissions by addressing the largest 20% of the sources.
■
REFERENCES
(1) Harriss, R.; Alvarez, R. A., Using multi-scale measurements to improve methane emissions estimates from oil and gas operations in the Barnett Shale, Texas: Campaign summary. Manuscript in preparation. (2) Myhre, G., Shindell, D.; F.-M. Bréon, Collins, W.; Fuglestvedt, J.; Huang, J.; Koch, D.; Lamarque, J.-F.; Lee, D.; Mendoza, B., et al. (2013): Anthropogenic and Natural Radiative Forcing. In Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change; Stocker, T. F., Qin, D., Plattner, G.-K., Tignor, M., Allen, S. K., Boschung, J., Nauels, A., Xia, Y., Bex, V., Midgley, P. M., Eds.; Cambridge University Press: Cambridge/New York. (3) Howarth, R. W.; Santoro, R.; Ingraffea, A. Methane and the greenhouse-gas footprint of natural gas from shale formations. Clim. Change 2011, 106 (4), 679−690. (4) Alvarez, R. A.; Pacalab, S. W.; Winebrake, J. J.; Chameides, W. L.; Hamburg, S. P. Greater focus needed on methane leakage from natural gas infrastructure. Proc. Nat. Acad. Sci. 2012, v109 (17), 6435−6440 DOI: 10.1073/pnas.1202407109. (5) Allen, D. T.; Torres, V. M.; Thomas, J.; Sullivan, D. W.; Harrison, M.; Hendler, A.; Herndon, S. C.; Kolb, C. E.; Fraser, M. P.; Hill, A. D.; et al. Measurements of methane emissions at natural gas production sites in the United States. Proc. Natl. Acad. Sci. U. S. A. 2013, 110 (44), 17768−17773. (6) Eastern Research Group; Sage Environmental, City of Fort Worth Natural Gas Air Quality Study, Final Report, July 13, 2011, available at http://fortworthtexas.gov/uploadedFiles/Gas_Wells/ AirQualityStudy_final.pdf, accessed August 2014. (7) Energy Information Administration (2012): Number of Producing Gas Wells 2007−2012. Available at http://www.eia.gov/ dnav/ng/ng_prod_wells_s1_a.htm. Accessed August 7, 2014. (8) Shorter, J. H.; McManus, J. B.; Kolb, C. E.; Allwine, E. J.; Siverson, R.; Lamb, B. K.; Mosher, B. W.; Harriss, R. C.; Howard, T.; Lott, R. A. Collection of leakage statistics in the natural gas system by tracer methods. Environ. Sci. Technol. 1997, 31 (7), 2012−2019. (9) Czepiel, P. M.; Mosher, B.; Harriss, R. C.; Shorter, J. H.; McManus, J. B.; Kolb, C. E.; Lamb, B. K. Landfill methane emissions
4747
DOI: 10.1021/acs.est.5b00099 Environ. Sci. Technol. 2015, 49, 4742−4748
Article
Environmental Science & Technology measured by enclosure and atmospheric tracer methods. J. Geophys. Res.: Atmos. (1984−2012) 1996, 101 (D11), 16711−16719. (10) Thoma, E. D.; Squier, B. C.; Olson, D.; Eisele, A. P.; DeWees, J. M.; Segall, R. R.; Amin, M. S.; Modrak, M. Assessment of methane and VOC emissions from select upstream oil and gas production operations using remote measurements, interim report on recent survey studies. In Proceedings of 105th Annual Conference of the Air & Waste Management Association, Control No. 2012-A-21-AWMA, 2012, pp 298−312. (11) Thoma, E. D.; Shores, R. C.; Thompson, E. L.; Harris, D. B.; Thorneloe, S. A.; Varma, R. M.; Hashmonay, R. A.; Modrak, M. T.; Natshcke, D. F.; Gamble, H. A. Open-path tunable diode laser absorption spectroscopy for acquisition of fugitive emission flux data. J. Air Waste Manage. Assoc. 2005, 55 (5), 658−668. (12) Crosson, E. R. A cavity ring-down analyzer for measuring atmospheric levels of methane, carbon dioxide, and water vapor. Appl. Phys. B: Laser Opt. 2008, 92 (3), 403−408. (13) Karion, A.; Sweeney, C.; Tans, P.; Newberger, T. AirCore: An innovative atmospheric sampling system. J. Atmos. Oceanic Technol. 2010, 27 (11), 1839−1853. (14) Zavala-Araiza, D., Lyon, D., Alvarez, R. A., Hamburg, S. P. Towards a functional definition of methane super-emitters: Application to natural gas production sites. Manuscript in preparation. (15) De Nevers, N. Air Pollution Control Engineering; Waveland Press: Long Grove, IL, 2010. (16) Mooney, C. Z.; Duval, R. D., Eds. Bootstrapping: A Nonparametric Approach to Statistical Inference (Nos. 94−95); Sage: Thousand Oaks, CA, 1993.
4748
DOI: 10.1021/acs.est.5b00099 Environ. Sci. Technol. 2015, 49, 4742−4748
