Airborne estimates of greenhouse gas emissions are becoming more
prevalent with the advent of rapid commercial development of trace gas
instrumentation featuring increased measurement accuracy, precision, and
frequency, and the swelling interest in the verification of current emission
inventories. Multiple airborne studies have indicated that emission
inventories may underestimate some hydrocarbon emission sources in US oil-
and gas-producing basins. Consequently, a proper assessment of the accuracy
of these airborne methods is crucial to interpreting the meaning of such
discrepancies. We present a new method of sampling surface sources of any
trace gas for which fast and precise measurements can be made and apply it to
methane, ethane, and carbon dioxide on spatial scales of
Accurate national inventories of greenhouse gas emissions (primarily carbon
dioxide – CO
In principle, the aircraft top-down measurements can be conducted at all the
atmospheric scales to better understand and identify the emissions at
comparable scales. For long-lived greenhouse gases, which readily disperse
throughout the atmosphere, the global scale is very instructive. The seminal
experiment began with Keeling's acclaimed CO
Aircraft in situ measurements are particularly useful for top-down
methods at the sub-mesoscale because they can be used to measure the air
both upwind and downwind of a source region. However, deployments tend to be
costly and thus sporadic. As far as we know, the aircraft methods used so
far can be categorized into three types. First, there is the eddy covariance
technique that is carried out at low altitudes wherein the vertical fluxes
of gases carried by the turbulent wind are measured by tracking rapid
fluctuations of both concentrations and vertical wind (Hiller et al.,
2014; Ritter et al., 1994; Yuan et al., 2015). This method is generally
thought to be the most direct, but it is limited to small footprint regions
which must be repeatedly sampled for sufficient statistical confidence,
requires a sophisticated vertical wind measurement, and can be subject to
errors due to flux divergence between the surface and the lowest flight
altitude and acceleration sensitivity of the gas sensor. The second and by
far the most common approach is what chemists usually refer to as “mass
balance” and what is known in the turbulence community as a “scalar
budget” technique. Many different sets of assumptions and sampling
strategies are employed, but the overall goal is to sample the main
dispersion routes of the surface emissions as they make their way into the
overlying atmosphere after first accumulating near the surface. The scales
that can be addressed by this method are from a few kilometers (Alfieri
et al., 2010; Hacker et al., 2016; Hiller et al., 2014; Tratt et al.,
2014) to tens of kilometers (Caulton et al., 2014; Karion et al., 2013;
Wratt et al., 2001) to even potentially hundreds of kilometers (Beswick
et al., 1998; Chang et al., 2014), and this approach has been the focus of
recent measurements in natural gas production basins. These basins present a
source apportionment challenge in that emissions from multiple sources
(agriculture, oil and gas wells, geologic seepage, etc.) commingle as the
air mass travels across the basin. The third method of source quantification
is to reference measurements of the unknown trace gas to a reference trace
gas with a metered release (tracer) or otherwise-known emission rate and
assume that the tracer and the scalar of interest have the same diffusion
characteristics. Typically this tracer release technique is applied to small
scales of tens to hundreds of meters (Czepiel et al., 1996; Lamb et al.,
1995; Roscioli et al., 2015), but the principle has been attempted at the
basin (Peischl et al., 2013) and continental (Miller et al., 2012) scales using a reference trace gas
with a suitable known emission rate such as CO
The airborne mass balance flight strategies can be grouped into three basic patterns: a single height transect around a source assuming a vertically uniformly mixed boundary layer (Karion et al., 2013), single height upwind/downwind (Wratt et al., 2001) or sometimes just downwind flight legs (Conley et al., 2016; Hacker et al., 2016; Ryerson et al., 1998), multiple flight legs at different altitudes (Alfieri et al., 2010; Gordon et al., 2015; Kalthoff et al., 2002), or just a “screen” on the downwind face of the box (Karion et al., 2015; Lavoie et al., 2015; Mays et al., 2009).
Here we describe a new airborne method borne out of a necessity to identify and quantify source emissions to within 20 % accuracy in a large heterogeneous field of potential sources. The novel technique applies an aircraft flight pattern that circumscribes a virtual cylinder around an emission source and, using only observed horizontal wind and trace gas concentrations, applies Gauss's theorem to estimate the flux divergence through that cylinder. By integrating the outward horizontal fluxes at each point along the circular flight path, the flux contributions from enclosed sources can be accounted for. Making an accurate estimate, however, requires the selection of an appropriate circling radius based on the micrometeorological conditions inferred in flight from measurements onboard the aircraft. The pattern must be far enough downstream for the plume to mix sufficiently in the vertical, yet not so far that the trace gas plume enhancements do not stand out sufficiently from the background concentration.
In this study we first present the general analytical method used to derive
emission estimates using airborne measurements. Next, we investigate the
structure of a generalized dispersing plume using large eddy simulation (LES)
to better understand the optimal sampling strategies for quantifying
near-surface gas sources. Because the wind fields of turbulent flows cannot
be predicted in detail, we do not attempt to compare specific features of
our observations with specific LES results, but rather we use the numerical
experiments to guide the development of the observational methodology. For
example, by investigating the LES flux divergence profiles in the layer
below the lowest flight altitude, we are able to estimate the contribution
of this unmeasured component to the overall source strength. We then
evaluate the accuracy of the approach using coordinated planned release
experiments and by applying the method to CO
The airborne detection system is flown on a fixed wing single-engine Mooney
aircraft, extensively modified for research as described in Conley et al. (2014).
Ambient air is collected through
In order to study the plume behavior of surface emissions as it relates to
sampling in the stacked circles, we use the LES module of WRF V3.6.1. WRF-LES explicitly
resolves the largest turbulent eddies by filtering the Navier–Stokes scalar
conservation equations at some scale in the inertial subrange, and allowing
the smaller motions beyond the cut-off to be modeled using a sub-grid (also
called a sub-filter) scale turbulence parameterization that is based on
properties of the larger-scale, resolved flow. Because the aircraft data is
typically sampled at 1 Hz and the true airspeed is around 70 m s
Domain and micrometeorological parameters for the three WRF-LES
experiments in this study.
Map of the airplane flight pattern sampling a methane plume emanating
from an underground storage facility. Wind direction is indicated by the white
arrow and the methane mixing ratio is given by the color bar to the right. This
flight was conducted on 28 June 2016 and took place between 12:46 and 13:52 LT
at altitudes ranging from 91 to 560 m with a loop diameter of approximately
3 km. The measured methane emission rate was 763
The standard WRF-LES module is not set up to allow for effluent release, so we
implemented a modified version of the WRF source code (S.-H. Chen, personal
communication, 2015) that includes a
surface effluent release with a specified position and release rate. Three
different convective simulations were run with varying resultant mean wind
speeds in the boundary layer, and each was allowed 4–5 h spin-up
dynamically before the effluent was released at a rate of between 2.9 and 3.5 kg h
We use an integrated form of the scalar budget equation for a passive,
conservative scalar in a turbulent fluid to estimate the emission of a gas
of interest within a cylindrical volume
The surface integral can be broken into three elements: a cylinder extending
from the ground up to a level above significant modification by the
emission, the ground surface circumscribed by a low-level (virtual) circular
flight path (
Combining Eqs. (4) and (5) leads to the result that is the basis for this
measurement technique where a series of horizontal loops at different
altitudes are flown around a source region:
In order to estimate the relative error in the horizontal divergence term
that we are eliminating, we perform a scale analysis of the relative size of
the two terms in Eq. (2), using some typical values of the CBL parameters
(convective velocity scale
We calculated a comparable estimate of
Graphical representation of the relative magnitude (%) of the
contribution of the horizontal wind divergence to the horizontal advective terms
in Eq. (3), as a function of wind speed and source magnitude for methane, using
a typical global background of 1.9 ppm and divergence of 10
Figure 3 shows the crosswind-integrated concentration profile for the plume
release in the UCD50B WRF-LES run as function of
Relative cross wind integrated concentrations of an effluent plume released at the surface in the UCD50B simulation. The data are averaged over 15 min of simulation time and normalized by the maximum concentration.
Horizontal turbulent fluxes are generally ignored in boundary layer budget
studies due to the fact that while they are often sizeable in magnitude they
do not change significantly over horizontal length scales under
consideration (the horizontal homogeneity assumption). In the vicinity of a
point source, however, this is not likely. The method outlined here
estimates source emissions using a measured horizontal flux that
incorporates wind and scalar measurements at 1 Hz sample rate, resolving
scales of
This conclusion is supported by several previous studies. For example, in a
wind-tunnel study of flux–gradient relationships, Raupach and Legg (1984)
reported that the mean stream-wise horizontal heat flux calculated by
multiplying the mean wind by the mean temperature overestimates the total
heat flux by approximately 10%, which suggests that the turbulent
component of the horizontal heat flux is negative; that is, the turbulent
flux is upwind, directed counter to the mean flow. Other researchers have
reported an even larger disparity. Field experiments by Leuning et al. (1985)
indicate that the horizontal turbulent flux of a
trace gas is
Average cospectrum of the outward-directed component of the observed
wind and the methane concentration from 70 laps around a point source near San
Antonio, Texas. The peak at 10
Further evidence of this is shown in the average cospectrum of the outward
wind and concentration fluctuation observed in the flight loops in Fig. 4.
Because the integral of the cospectrum yields the total flux (scalar and
wind covariance), this function is useful for examining the contributions to
the overall flux from each of the scales of motion (represented by aircraft
speed divided by frequency). The results shown in Fig. 4 are from a
CH
Determining the optimal sampling distance from the targeted point source is
a balance of several factors. First, not surprisingly, the largest plume
signal occurs closest to the source (Fig. 3).
Second, a high degree of confidence in the results is contingent upon
sampling the majority of the plume at and above the lowest flight altitude,
which only occurs downwind after a sufficient time has elapsed to loft the
initially near-surface plume. And third, an attempt is made to sample the
plume before it reaches the top of the boundary layer so that the vertical
turbulent entrainment flux does not become appreciable violating the
assumption of negligible flux through the top of the volume
Dimensionless flux divergence profiles generated from averaging over
3 different WRF-LES runs using 30 time steps for each one. The horizontal flux
per unit altitude (
To gain further insight into the second feature of the dispersing plume,
Fig. 5 shows the average horizontal flux divergence profiles derived from
the three WRF-LES runs. Here we discuss a dimensionless
Based on the simulation results presented in Fig. 5, we see the gradient
below the lowest flight safe altitude typically becomes very small for
Rate of convergence toward the final leak rate estimation as a function
of the number of loops for LES CASE UCD508. By 15 laps, the emissions estimate
(blue line) has stabilized to 2.5 kg h
The atmospheric boundary layer is a turbulent medium, meaning that two passes across a plume at the same altitude and distance downwind will likely make very different measurements of the trace gases of interest. A natural question arises as to how many passes are required to develop a statistically sound estimate of the emission rate. We investigate the number of passes required to obtain a statistically robust estimate using the WRF-LES results and a controlled release experiment. By calculating the horizontal flux divergences with a virtual airplane flying through the simulated tracer field, and then randomly sampling the flux divergences from each of the legs and plotting the resultant estimated emission rate as a function of the number of samples used, we obtain the results presented in Fig. 6. The gray region around the red line mean represents the standard deviation of estimates based on a random set of loops. Figure 7 shows results from an analysis of actual flight data from the controlled ethane release test near Denver, Colorado, on 19 November 2014. It is evident from both the simulation data and the field data that a statistically stable estimate seems to be achieved somewhere between 20 and 25 loops around the source.
Averaged LES estimates for the Aerodyne case. This leak shows a slightly
higher number of laps before convergence (
Measurements of the relevant scalars (e.g., CH
Our method assumes a stationary emission source. The leg-to-leg variability is primarily driven by the stochastic nature of turbulence (e.g., we may sample the plume on one lap and miss it on another). By aggregating the laps into vertical bins, we can use the standard deviation of the horizontal fluxes within each bin as an estimate of the uncertainty within that bin. Then the total uncertainty in the estimate of the flux divergence is simply estimated by adding up the individual bin uncertainties in quadrature. The first term on the rhs of Eq. (6) is the time rate of change of the scalar mass within the cylindrical flight volume. This storage term is estimated by performing a least squares fit of the methane density with time and altitude. The resulting uncertainty in the time rate of change is then combined (summed in quadrature) with the uncertainty from the altitude bins to achieve a total uncertainty in the measurement.
We use measurements from three sets of flights to characterize the accuracy
of this estimation method. We flew 2 days measuring an controlled ethane
release provided by Aerodyne Research, Inc., 4 days measuring a controlled natural gas release provided by the Pacific Gas & Electric Company (PG & E),
and six power plant flights where our estimates are compared with
reported hourly power plant CO
Two experiments with known/controlled ethane releases were performed in
collaboration with the Aerodyne Mobile Laboratory team. Pure ethane was
released and measured with a flowmeter by the Aerodyne ground crew. The
Colorado site (November 2014) was in a remote area approximately 170 km
NE of Denver. This site was chosen because of the flat terrain and lack of
other nearby ethane sources that could pollute the controlled release plume.
The flux profiles for both releases are shown in
Figure 8 and indicate that, in both cases, the
aircraft successfully flew above the ethane plume (measurements tend toward
zero with increasing altitude). An example of an individual lap is shown in
Fig. 9 and indicates a clear plume signal
downwind of the release. As the aircraft climbs, eventually the signal
disappears, as shown in the figure. Agreement was excellent, with the
estimated emission just 2 % over the actual controlled release rate. The
second Aerodyne controlled release in Arkansas on 3 October 2015 was
performed at a site surrounded by nearby emission sources and an elevation
change (
Controlled ethane releases.
Controlled natural gas release.
Ethane horizontal transport profiles for the Aerodyne controlled
releases near Denver, Colorado, on 19 November 2014 (
A significant upwind ethane source was observed during the Arkansas experiment. This source was evident on roughly half of the upwind passes, suggesting that techniques which rely on a limited number of upwind passes to characterize the background could have a large random error and thus erroneously estimate the upwind source strength. A similar problem would affect those techniques that employ a downwind transect, using the edges of that transect lying outside the plume to estimate the background concentration. These observations demonstrate the complication (and bias) that can arise from nearby sources. Since this method integrates all the emission sources in the area within the flight circle and a small distance upwind of the circle depending on the vertical mixing, estimates from Gauss's method may be biased high if there are sources within that area. The average error of the two ethane releases is 13 %.
In conjunction with PG & E, we performed two sets of 2-day ground-level
controlled release experiments from existing PG & E facilities, exactly 1
year apart. The first set was performed southeast of Sacramento near the
town of Rio Vista, CA at the Rio Vista “Y” station and the second set near
Bakersfield, CA. For the Rio Vista test, the release rate was not calibrated
with a flow meter but, based on the size of the orifice and the upstream
pressure, the release rate was estimated at 15.2
In comparison with the controlled C
Power plant estimates. The mid-point of the measurements (Hour UTC) is indicated in the third column (Hour). The reported emissions from the hour before to the hour after that time were averaged to derive the “Reported” emissions in column 5. Emissions are reported in units of metric tons (t) per hour.
Power plants in the US are required to report CO
Comparison of aircraft vs. reported power plant emissions.
This technique was developed out of the necessity to identify and quantify individual well pads in an extensive oil and gas production field. Consequently the frequent tracking of the upwind and downwind side of the source provides a very accurate determination of the location and magnitude of a given emission site. The main uncertainty arises from the effluent below the lowest flight altitude, but this is minimized by targeting a downwind distance determined by LES studies to provide very little change in the plume flux divergence from the lowest loop to the ground. In addition to the controlled release experiments, hundreds of sites have been measured using this technique with varying levels of success. Ideal conditions include flat terrain, ample sunlight to promote vertical mixing, consistent winds, and no nearby competing sources. Under optimal conditions we have demonstrated that measurement uncertainties are quite low, often better than 10 %. As the conditions deteriorate from the ideal to situations involving complex terrain, variable winds or nearby upwind sources, measured uncertainties can increase to be as large or larger than the emission estimates themselves. In the worst case of stably stratified conditions (winter or nighttime), for instance, the lack of vertical mixing may preclude the trace gases emitted at the surface from reaching the minimum safe flight altitude. Complex terrain provides a challenge to the method because the aircraft is unable to maintain a constant altitude above the ground. A possible future refinement of this technique to be applied in complex terrain would be to fit the measurements of both wind and mixing ratio to a uniform 3-D surface surrounding the source, where the grid passes through the terrain and then integrate the flux normal to this irregular virtual flight path. This would not assume level loop flight legs and would, in principle, account for individual loops being flown at differing altitudes and thus more closely track mass continuity near the terrain elevation.
Data are available upon request by the corresponding author.
The authors declare that they have no conflict of interest.
Funding for this study provided by the California Energy Commission (CEC) and the US Department of Energy (DOE). Funding for the Denver and Arkansas portion of this work was provided by RPSEA/NETL contract no. 12122-95/DE-AC26-07NT42677 to the Colorado School of Mines. Cost sharing was provided by Colorado Energy Research Collaboratory, the National Oceanic and Atmospheric Administration Climate Program Office, the National Science Foundation (CBET-1240584), Southwestern Energy, XTO, Chevron, Statoil, and the American Gas Association, many of whom also provided operational data and/or site access. We also thank Shuhua Chen for assistance with the WRF-LES code modifications and advice. The National Center for Atmospheric Research is sponsored by the National Science Foundation. This work was supported in part by the NOAA AC4 program under grant NA14OAR0110139 and the Bureau of Land Management, grant L15PG00058. We thank Ying Pan for her significant contribution to our understanding of the negative horizontal scalar flux. Edited by: Andre Butz Reviewed by: three anonymous referees