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
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

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
Downloaded via PEKING UNIV on March 23, 2024 at 12:07:38 (UTC).

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 carbon-
rich 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
emissions. Finally, fugitive emissions are generally unintentional
emissions, due either to human error or a process or
component failure, making any attempt at accurate quantifica-
tion 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: December 19, 2014
Revised: March 14, 2015
Accepted: March 16, 2015
Published: March 25, 2015

© 2015 American Chemical Society 4742 DOI: 10.1021/acs.est.5b00099

Environ. Sci. Technol. 2015, 49, 4742−4748
Environmental Science & Technology
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 time-
consuming, 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 atmos-
pheric 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
Article
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 real-
time 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 Spec-
trometer (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 real-
time 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
DOI: 10.1021/acs.est.5b00099
Environ. Sci. Technol. 2015, 49, 4742−4748
Environmental Science & Technology
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
Article
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 environ-
ments 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 experi-
ments 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 experi-
ments 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
DOI: 10.1021/acs.est.5b00099
Environ. Sci. Technol. 2015, 49, 4742−4748
Environmental Science & Technology
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 open-
source 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
4745
Article
quantify the minimum leak rate, using the trapezoidal
integration method described in the SI).
■ 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
DOI: 10.1021/acs.est.5b00099
Environ. Sci. Technol. 2015, 49, 4742−4748
Environmental Science & Technology
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 dot-
dashed 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 under-
represented 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
Article
• 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 log-
normal 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
DOI: 10.1021/acs.est.5b00099
Environ. Sci. Technol. 2015, 49, 4742−4748
Environmental Science & Technology
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.
4747
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.
■ 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
DOI: 10.1021/acs.est.5b00099
Environ. Sci. Technol. 2015, 49, 4742−4748
Environmental Science & Technology Article

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