Atmos. Meas. Tech., 14, 7999–8017, 2021
https://doi.org/10.5194/amt-14-7999-2021
© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.
Remote sensing of methane plumes: instrument tradeoff analysis for
detecting and quantifying local sources at global scale
Siraput Jongaramrungruang
1
, Georgios Matheou
2
, Andrew K. Thorpe
3
, Zhao-Cheng Zeng
1
, and
Christian Frankenberg
1,2
1
Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA 91125, USA
2
Department of Mechanical Engineering, University of Connecticut, Storrs, CT 06269, USA
3
NASA Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA
Correspondence:
Siraput Jongaramrungruang (siraput@caltech.edu) and Christian Frankenberg (cfranken@caltech.edu)
Received: 12 July 2021 – Discussion started: 20 July 2021
Revised: 18 October 2021 – Accepted: 20 October 2021 – Published: 22 December 2021
Abstract.
Methane (CH
4
) is the second most important an-
thropogenic greenhouse gas with a significant impact on ra-
diative forcing, tropospheric air quality, and stratospheric
water vapor. Remote sensing observations enable the detec-
tion and quantification of local methane emissions across
large geographical areas, which is a critical step for under-
standing local flux distributions and subsequently prioritizing
mitigation strategies. Obtaining methane column concentra-
tion measurements with low noise and minimal surface inter-
ference has direct consequences for accurately determining
the location and emission rates of methane sources. The qual-
ity of retrieved column enhancements depends on the choices
of the instrument and retrieval parameters. Here, we studied
the changes in precision error and bias as a result of dif-
ferent spectral resolutions, instrument optical performance,
and detector exposure times by using a realistic instrument
noise model. In addition, we formally analyzed the impact
of spectrally complex surface albedo features on retrievals
using the iterative maximum a posteriori differential opti-
cal absorption spectroscopy (IMAP-DOAS) algorithm. We
built an end-to-end modeling framework that can simulate
observed radiances from reflected solar irradiance through a
simulated CH
4
plume over several natural and artificial sur-
faces. Our analysis shows that complex surface features can
alias into retrieved methane abundances, explaining the ex-
istence of retrieval biases in current airborne methane ob-
servations. The impact can be mitigated with higher spec-
tral resolution and a larger polynomial degree to approxi-
mate surface albedo variations. Using a spectral resolution
of 1.5 nm, an exposure time of 20 ms, and a polynomial de-
gree of 25, a retrieval precision error below 0.007 mole m
−
2
or 1.0 % of total atmospheric CH
4
column can be achieved
for high albedo cases, while minimizing the bias due to sur-
face interference such that the noise is uncorrelated among
various surfaces. At coarser spectral resolutions, it becomes
increasingly harder to separate complex surface albedo fea-
tures from atmospheric absorption features. Our modeling
framework provides the basis for assessing tradeoffs for fu-
ture remote sensing instruments and algorithmic designs. For
instance, we find that improving the spectral resolution be-
yond 0.2 nm would actually decrease the retrieval precision,
as detector readout noise will play an increasing role. Our
work contributes towards building an enhanced monitoring
system that can measure CH
4
concentration fields to deter-
mine methane sources accurately and efficiently at scale.
1 Introduction
Anthropogenic greenhouse gas emissions have been ris-
ing continuously, affecting the global climate and the en-
vironment (Stocker et al., 2013). Among the most impor-
tant anthropogenic emissions are carbon dioxide (CO
2
) and
methane (CH
4
). Due to a much shorter lifetime of CH
4
(
≈
9 years) compared to CO
2
(
≈
500 years), CH
4
has gained
attention as a target for mitigation efforts to achieve short-
and medium-term reductions in global warming (Montzka
et al., 2011; Prather et al., 2012; Shindell et al., 2012). In gen-
eral, anthropogenic methane emissions are also much more
uncertain than those of carbon dioxide, which can often be
Published by Copernicus Publications on behalf of the European Geosciences Union.
8000
S. Jongaramrungruang et al.: Remote sensing of methane plumes: instrument tradeoff analysis
characterized to within approximately 10 % just from bud-
get assumptions (Gurney et al., 2019). For instance, just the
question of whether or not the leak rate in the natural gas
extraction system is 1 % or 2 % is equivalent to a 100 % un-
certainty in methane emissions from these leaks. At the same
time, leak rate outliers (Frankenberg et al., 2016; Duren et al.,
2019; Cusworth et al., 2021) are often local in nature and eas-
ily fixable, representing a win–win scenario if faulty equip-
ment or practices can be readily detected and then efficiently
mitigated. As CH
4
reduction plays a significant role in cli-
mate mitigation efforts, one key step in emission reduction
is determining where these emissions are coming from. This
is underpinned in the 2018 NASA Decadal Survey, which
names the identification and understanding of CH
4
emissions
as one of the top priorities in the efforts to improve future cli-
mate projections and help lead the way in emission reduction
(National Academies of Sciences and Medicine, 2018).
Remote sensing instruments using absorption spec-
troscopy have been available as one effective solution for
measuring atmospheric CH
4
concentration over large geo-
graphical areas. Space-based CH
4
retrieval techniques from
satellite observations such as the SCanning Imaging Absorp-
tion SpectroMeter for Atmospheric CHartographY (SCIA-
MACHY; Frankenberg et al., 2005a, 2011) and the Green-
house gases Observing SATellite (GOSAT; Parker et al.,
2011, 2015; Turner et al., 2015) were dedicated missions
with CH
4
as a key target. They used CH
4
absorption fea-
tures in the 1.6 or 2.3 μm bands to retrieve column CH
4
con-
centration across the globe. The TROPOspheric Monitoring
Instrument (TROPOMI), with a spatial resolution of a few
kilometers, has also been shown to be capable of identifying
regions of high emissions (de Gouw et al., 2020; Hu et al.,
2018). These satellites, which have been designed by the at-
mospheric community, have particular sets of goals and in-
strument specifications that are mostly targeted towards ob-
taining observations of the regional-scale methane distribu-
tions with high accuracy and precision. Most of these satel-
lites were designed to measure gradients of methane concen-
tration across hundreds to thousands of kilometers of scale as
this enables surface flux inversions at the global scale. Typ-
ically, all of these instruments have one feature in common
– they have very high spectral resolution (0.05–0.25 nm) to
distinguish individual methane absorption lines from spec-
trally smooth surface albedo variations. However, due to their
coarse spatial resolutions, the measurements are not yet at a
level where local sources can be identified, attributed to a
specific source type (e.g., compressor station or well pad),
and mitigated directly.
One potential solution to fill this scale gap is to use an
airborne instrument that has a much higher spatial resolu-
tion, such as the Methane Airborne MAPper (MAMAP; Ger-
ilowski et al., 2011) or the next-generation Airborne Vis-
ible InfraRed Imaging Spectrometer (AVIRIS-NG; Thorpe
et al., 2017). The latter is based on the insight that methane
column enhancements at high spatial resolution (a few me-
ters) can be so high that the retrieval of the absorbing fea-
ture can be done, even with a moderate spectral resolution
(5–10 nm). If the methane column is expressed similar to the
Dobson unit, i.e., as the thickness of a layer of pure gas which
would be formed by the total column amount at standard
conditions, then the layer thickness at current background
methane conditions would only be about 1.6 cm. Thus, a pure
methane layer of only 1.6 mm would enhance the total col-
umn by 10 %, which is certainly realistic for measurements
of methane point sources at fine spatial resolution.
Bradley et al. (2011) and Thorpe et al. (2014) were among
the first to show that moderate resolution instruments can de-
tect methane plumes, even when the strong 2.3 μm methane
band is convolved with the AVIRIS (resolution of 10 nm) or
AVIRIS-NG (5 nm) instrument line shape functions. While
individual lines are hard to resolve, the strong methane band
in this range causes enough fine structure in terms of bulk
absorptions by a multitude of methane lines within the in-
strument resolution. Previous studies by Frankenberg et al.
(2016), Duren et al. (2019), and Cusworth et al. (2021) have
utilized AVIRIS-NG to conduct field campaigns in Cali-
fornia and the Four Corners region, where they could cre-
ate a map of methane enhancements in the area and detect
several hundreds of individual methane sources, which fol-
lowed a heavy tail flux distribution. The concept of this air-
borne spectrometer provides a promising opportunity for lo-
cal source detection and quantification. However, the instru-
ment was not originally designed for methane detection, and
it does not meet the same precision and accuracy require-
ments as those satellites from the atmospheric community for
methane retrieval at a global scale, which require precision
better than 1 %, which is equivalent to enhancements of about
19 ppb (parts per billion) in XCH
4
(or 4
×
10
17
molec
..
cm
−
2
,
0
.
007 mole m
−
2
or 152 ppm m). One significant drawback of
coarse spectral resolution is the occurrence of retrieval arti-
facts that often correlate with specific surface features (see
Fig. 1). This can confound the detection and quantification
of methane point sources in the analysis and obviate the
robust detection of subtle gradients at larger spatial scales
(Jongaramrungruang et al., 2021). Even though most strong
plumes can be observed, the uncertainties in the overall de-
tection and quantification at the regional level can present
persistent problems and often involve human judgment to
isolate plumes from artifacts. In fact, during the Califor-
nia survey, the Four Corners study, and the Permian survey
(Duren et al., 2019; Frankenberg et al., 2016; Cusworth et al.,
2021), human analysts were involved in a manual process to
look through each flight line to classify true emission sources
from false positives. Similarly, in previous space-based stud-
ies to locate and approximate a large emission source, such
as a blowout event, prior information about the location of
the source is usually already known, making it much eas-
ier to find a true methane source from space-based measure-
ments over the area after the fact. There are ongoing efforts to
develop automated plume detection for existing instruments.
Atmos. Meas. Tech., 14, 7999–8017, 2021
https://doi.org/10.5194/amt-14-7999-2021
S. Jongaramrungruang et al.: Remote sensing of methane plumes: instrument tradeoff analysis
8001
Figure 1.
Example of the systematic outliers from a retrieved
AVIRIS-NG scene
(b)
compared with an RGB image
(a)
.
Future real-time monitoring systems would greatly benefit
from next-generation instruments that would reduce retrieval
artifacts and provide retrievals with improved accuracy, such
that remote sensing measurements can be analyzed to locate
and quantify plumes automatically and at scale. This moti-
vates the origination of this study.
If we had the opportunity to design a new instrument that
is optimized for methane retrievals at fine spatial resolution
(sub 50 m), then what would the specifications of this instru-
ment look like? Thorpe et al. (2016) proposed a 1 nm in-
strument to mitigate the drawbacks of AVIRIS-NG. To fully
evaluate optimal performance metrics, we have to consider
the tradeoff between spectral and spatial resolutions and con-
comitant changes in detector noise characteristics. On the
one hand, the instrument needs to meet the requirements of
the atmospheric community so that it can unambiguously dif-
ferentiate methane from other confounding factors. On the
other hand, the instrument should have an adequate inte-
gration time to achieve a high spatial resolution with suffi-
cient signal-to-noise levels. Here, we investigate this trade-
off and evaluate the risks and benefits for methane retrievals
at fine resolutions with the purpose of successfully detect-
ing and quantifying local sources in mind. We built an end-
to-end modeling framework that can generate reflected so-
lar radiance through a methane plume of known concentra-
tion over realistic surfaces and perform the retrieval from the
corresponding observed radiance under a given instrument to
output the predicted methane concentration in each column.
Our model calculates the noise-equivalent spectral radiance
(NESR) as a function of incoming radiance and instrument
parameters, such as integration time, detector size, quantum
efficiency, readout noise, and spectral resolution, rather than
prescribing the signal-to-noise ratio (SNR) as an independent
variable. By varying the instrument and retrieval parameters,
we can derive the associated precision error and bias from
the retrieval. We also compare the tradeoff between the two
most frequently used fitting windows in the 1.6 and 2.3 μm
ranges.
Section 2 outlines the background on radiative transfer,
followed by data and methodology on the forward model
with realistic surface reflectances, instrument operators, and
retrieval setups. The results and discussion are provided in
Sect. 3. The final section contains the concluding remarks
and future steps.
2 Data and methodology
For the sake of simplicity, we ignore the impact of atmo-
spheric scattering, as Rayleigh scattering is negligible in the
near-infrared, and the impact of aerosols is rather small com-
pared to methane enhancements in the near-field of local
sources. While aerosols can cause small systematic biases
in the retrieved methane amount, their impact on measuring
anomalies caused by methane plumes should be rather small.
In addition, the precision error is not strongly affected by
neglecting atmospheric scattering, and experience with pre-
vious moderate resolution methane mapping has shown that
surface interferences are more crucial. In the absence of at-
mospheric scattering, and assuming a Lambertian surface,
the reflected radiance as measured by an instrument at the
top of the atmosphere in the nadir direction can be modeled
as follows:
L
λ
=
I
0
,λ
·
r
λ
·
T
λ
↑
·
T
λ
↓
·
cos
(
SZA
)
π
,
(1)
where
I
0
stands for the incoming solar irradiance spectrum,
T
λ
↓
is the atmospheric transmission along the photon light
path downwards to the surface,
r
λ
is the surface albedo,
T
λ
↑
is the transmission along the light path on the way up from
the surface to the instrument, and SZA is the solar zenith an-
gle. Figure 2 illustrates a schematic for Eq. (1). The subscript
λ
denotes the wavelength dependence of these variables. The
multiplication in Eq. (1) is element-wise for each wavelength
in the spectral range of interest.
2.1 Incoming solar irradiance
We constructed
I
0
,λ
by multiplying a continuum level spec-
trum with a high-resolution solar transmission spectrum that
includes absorption features in the Sun’s photosphere, which
are the so-called Fraunhofer lines. These absorption features
are caused by trace elements in the solar photosphere. The
continuum spectrum is obtained from Meftah et al. (2018)
with a 0.2 nm resolution. We fitted a third-order polynomial
to this measured spectra in a 1.4–2.5 μm range to obtain a
smooth continuum spectrum. A disk-integrated solar trans-
mission spectrum is obtained from a tabulated line list com-
piled by Toon (2015). We interpolated the baseline and trans-
mission spectra to a common 0.01 nm resolution grid and
https://doi.org/10.5194/amt-14-7999-2021
Atmos. Meas. Tech., 14, 7999–8017, 2021