Methane Mapping with Future Satellite Imaging Spectrometers

This study evaluates a new generation of satellite imaging spectrometers to measure point source methane emissions from anthropogenic sources. We used the Airborne Visible and Infrared Imaging Spectrometer Next Generation(AVIRIS-NG) images with known methane plumes to create two simulated satellite products. One simulation had a 30 m spatial resolution with ~200 Signal-to-Noise Ratio (SNR) in the Shortwave Infrared (SWIR) and the other had a 60 m spatial resolution with ~400 SNR in the SWIR; both products had a 7.5 nm spectral spacing. We applied a linear matched filter with a sparsity prior and an albedo correction to detect and quantify the methane emission in the original AVIRIS-NG images and in both satellite simulations. We also calculated an emission flux for all images. We found that all methane plumes were detectable in all satellite simulations. The flux calculations for the simulated satellite images correlated well with the calculated flux for the original AVIRIS-NG images. We also found that coarsening spatial resolution had the largest impact on the sensitivity of the results. These results suggest that methane detection and quantification of point sources will be possible with the next generation of satellite imaging spectrometers.


Introduction
Methane is a prominent greenhouse gas and is responsible for about 20% of all atmospheric radiative forcing [1,2]. As our planet continues to warm due to radiative forcing caused by greenhouse gases, understanding, monitoring, and reducing these emissions will become increasingly important. However, there remains a high level of uncertainty regarding the magnitude of methane sources and sinks globally [2,3]. Anthropogenic sources, which make up 50-65% of all methane emissions, mostly come from the energy, waste, and agriculture sectors [2,4]. These are generally point sources, or emissions emanating from a single easily identifiable source like a gas storage tank or a landfill [5,6]. Although these sectors can also contain diffuse sources, or emissions coming from a larger area, a few point sources can dominate the emissions budget. Identifying emission sources and quantifying their rates can improve regional greenhouse gas budgets and mitigation strategies [6,7].

Methods
In order to test the capability of future sensors for methane mapping, we resampled AVIRIS-NG data to create a simulated satellite product ( Figure 1) and then tested methane retrievals using this new product. We based our simulated satellite data on the sensor design for the upcoming EMIT mission [29]. The EMIT sensor design favors radiometric sensitivity over spatial resolution, leading to a 60 m spatial sampling that is slightly coarser than the 30 m spatial sampling slated for many orbital VSWIR sensors. Therefore, we simulated two sensor designs. The first was characteristic of a hypothetical future NASA mission after EMIT as well as those already operating or planned for launch by other space agencies. It has a 30 m spatial resolution with lower SNR. The second was a direct analogue of the EMIT instrument, with a 60 m spatial resolution and a higher SNR. Both simulations Remote Sens. 2019, 11, 3054 3 of 19 are important for understanding the methane mapping potential for NASA's future generation of sensors. The following section describes the original AVIRIS-NG data used, the process to resample it to a simulated satellite product, and the methane retrieval algorithm.

Aircraft to Top-of-Atmosphere
To simulate the additional atmosphere between the original aircraft height and the top of the atmosphere (TOA), we used MODerate resolution atmospheric TRANsmission 6 (MODTRAN6) to simulate one-way transmittance between the two heights [35]. Each transmittance simulation was generated with the aircraft height, latitude, longitude, day, and time of day of the original AVIRIS-NG flight. In addition, we assumed a background carbon dioxide, methane, and water vapor profile. Pixel radiance values from each image were multiplied by each simulated transmittance spectrum. For more details on the MODTRAN6 inputs and an example of a radiance spectrum adjusted to TOA, please see Appendix A.

Imagery
AVIRIS-NG is an airborne imaging spectrometer that measures radiance in the visible through shortwave-infrared (SWIR). It measures in the spectral range from about 380 to 2500 nm with a 5 nm spectral sampling [30][31][32]. AVIRIS-NG has a 34 • field of view and a one milliradian instantaneous field-of-view that results in spatial resolutions ranging between approximately 1 and 8 meters, depending on the height of the aircraft. The AVIRIS-NG data used in this study were collected in 2016 and 2017 throughout California as part of the California Methane Survey and an image from a 2018 controlled release experiment [33]. We selected images of plumes from a landfill, a dairy, a dairy digester, a wastewater treatment facility, and oil/gas infrastructure. In this study, eight examples of plumes (Table 1), one or more from each source, were resampled to satellite specifications using the methods described below. To simulate the additional atmosphere between the original aircraft height and the top of the atmosphere (TOA), we used MODerate resolution atmospheric TRANsmission 6 (MODTRAN6) to simulate one-way transmittance between the two heights [34]. Each transmittance simulation was generated with the aircraft height, latitude, longitude, day, and time of day of the original AVIRIS-NG flight. In addition, we assumed a background carbon dioxide, methane, and water vapor profile. Pixel radiance values from each image were multiplied by each simulated transmittance spectrum. For more details on the MODTRAN6 inputs and an example of a radiance spectrum adjusted to TOA, please see Appendix A.

Spectral Resampling
The exact band centers and full width half maximum (FWHM) for EMIT are still to be determined. However, EMIT will most likely have a 7.4 nm spectral sampling with an 8.2 to 8.8 nm FWHM. We resampled spectra from AVIRIS-NG radiance images using a Gaussian convolution with band centers and FWHM based on these specifications. In total, the resampled spectra had 288 bands between 376.4 and 2500 nm.

Resample Spatial Resolution
Each image was resampled to 30 m and 60 m spatial resolutions using Gaussian resampling. The stated spatial resolution represents the full-width half maximum of the Gaussian function, with the x-y coordinates for each AVIRIS-NG pixel center used to calculate each pixel's weighting for resampling. The original unorthorectified AVIRIS-NG images were used for resampling, but pixel coordinate information allowed the creation of orthorectified 30 m and 60 m output images.

Sensor Noise
The SNR from AVIRIS-NG is higher than the proposed SNR for the EMIT sensor and in addition, after spatial and spectrally resampling the data, some of the noise from the original AVIRIS-NG data is reduced through averaging. Therefore, adding sensor noise back into the data is critical for accurately modeling satellite data. To properly model the SNR of the satellite sensors, a noise component aimed at mimicking expected sensor noise was added to the radiance. Modeled sensor noise comes from two main components: photon noise (PN) and constant noise (CN). PN increases as the number of photons hitting a detector increases, and scales with the square root of radiance. CN is mainly associated with electronics and quantization in the signal chain. A primary contributor is "read noise", caused by the analog-to-digital conversion as the instrument digitally records the signal, and this parameter can be modeled as constant across wavelengths and pixels. To calculate PN, we first converted the spectral radiance to the total number of photons and then multiplied it by the etendue (G), which is calculated based on the area of the emitting source (in this case, the Sun) times the solid angle, band FWHM, integration time (τ), and the system transmittance factor (T): where PN is the photon noise; L λ is the radiance at each wavelength in each pixel; λ is the wavelength; and hc is Planck's constant times the speed of light. The AVIRIS-NG CN is 100 photons, therefore the same value was used for this simulation. The total noise was calculated as follows: where TN is the total noise. The total noise was then converted back to radiance units by substituting TN for PN in Equation (1) and inverting. TN converted to radiance was then multiplied by a random number from a Gaussian distribution with a mean of 0 and standard deviation of 1, then added to the original radiance for each wavelength for each pixel. PN increases with increasing radiance; for low radiance values, CN dominates the total noise and for higher radiance values, PN dominates. For the 60 m image, we multiplied PN by the square root of 2. The SNR was calculated by dividing the signal by the noise. In this case, the signal was the number of photons and the noise is the total noise. The resulting average SNR for the 30 m simulated satellite images in the SWIR (2100 nm-2500 nm) was 214 and the average SNR for the 60 m simulated satellite images in the SWIR was 484. Figure 2 shows the full spectrum SNR for the two simulations.

Matched Filter Methane Retrieval
The methane enhancement above the background for each pixel was estimated using an iterative matched filter technique tuned with a methane unit absorption spectrum [19,[36][37][38]. This iterative method included additional optimization factors that correct for spatially varying surface albedo. In addition, a ℓ1 sparsity prior was included to reflect the expectation that methane enhancement is a rare occurrence within an image. This sparse approach is similar to methods employed in compressed sensing techniques for the reduction of acquisition time or improved resolution in fields as varied as medical imaging, radio astronomy, and electron microscopy [39][40][41][42][43]. The iterative technique required for solving this sparse optimization problem additionally produces the explicit removal of any methane absorption signal from the background covariance matrix; failure to remove this signal is detrimental to matched filter techniques [37,38,44,45].
This iterative matched filter with sparsity and albedo correction was applied to the AVIRIS-NG images and simulated 30 m and 60 m images. For all scenes, the sparse matched filter was run for 30 iterations, which includes an additional 50% margin over the 20 iterations that produced initial observable convergence of the optimization energy. The number of iterations was based on observations of convergence from a test dataset. The methane unit absorption spectrum for each scene was computed from the change in radiance corresponding to a change in methane concentration in MODTRAN6 with each scene's sensor height and solar zenith angle. The albedo correction factor was estimated by the ratio of a pixel's radiance to the mean radiance in the scene. For the original AVIRIS-NG data, the matched filter was applied to groups of five adjacent columns, corresponding to adjacent detectors of the AVIRIS-NG instrument. This grouping improved the covariance estimate used by the matched filter to describe the background by suppressing artifacts from non-uniformity among detector elements [19]. For the simulated satellite images, column artifacts were averaged out by spatial resampling, and therefore the matched filter was applied to the entire image.
The resulting methane enhancement image is measured in ppm-m, where ppm represents concentration and m represents the path length over which absorption occurs. Due to this algorithm's sparsity prior, many pixels are reported to have no methane enhancement above the background. This allows for interpreting the enhancement values as a detection result (presence/absence of methane for a positive/zero enhancement value). We also consider the resulting images as a quantitative retrieval directly from the per-pixel methane concentration. For more information on the retrieval algorithm please see Foote et. al., In Review [46].

Integrated Mass Enhancement
The integrated mass enhancement (IME) is the sum of all the methane enhancement above the background concentration present in each plume. To calculate the IME, we first need to identify the

Matched Filter Methane Retrieval
The methane enhancement above the background for each pixel was estimated using an iterative matched filter technique tuned with a methane unit absorption spectrum [19,[35][36][37]. This iterative method included additional optimization factors that correct for spatially varying surface albedo. In addition, a 1 sparsity prior was included to reflect the expectation that methane enhancement is a rare occurrence within an image. This sparse approach is similar to methods employed in compressed sensing techniques for the reduction of acquisition time or improved resolution in fields as varied as medical imaging, radio astronomy, and electron microscopy [38][39][40][41][42]. The iterative technique required for solving this sparse optimization problem additionally produces the explicit removal of any methane absorption signal from the background covariance matrix; failure to remove this signal is detrimental to matched filter techniques [36,37,43,44].
This iterative matched filter with sparsity and albedo correction was applied to the AVIRIS-NG images and simulated 30 m and 60 m images. For all scenes, the sparse matched filter was run for 30 iterations, which includes an additional 50% margin over the 20 iterations that produced initial observable convergence of the optimization energy. The number of iterations was based on observations of convergence from a test dataset. The methane unit absorption spectrum for each scene was computed from the change in radiance corresponding to a change in methane concentration in MODTRAN6 with each scene's sensor height and solar zenith angle. The albedo correction factor was estimated by the ratio of a pixel's radiance to the mean radiance in the scene. For the original AVIRIS-NG data, the matched filter was applied to groups of five adjacent columns, corresponding to adjacent detectors of the AVIRIS-NG instrument. This grouping improved the covariance estimate used by the matched filter to describe the background by suppressing artifacts from non-uniformity among detector elements [19]. For the simulated satellite images, column artifacts were averaged out by spatial resampling, and therefore the matched filter was applied to the entire image.
The resulting methane enhancement image is measured in ppm-m, where ppm represents concentration and m represents the path length over which absorption occurs. Due to this algorithm's sparsity prior, many pixels are reported to have no methane enhancement above the background. This allows for interpreting the enhancement values as a detection result (presence/absence of methane for a positive/zero enhancement value). We also consider the resulting images as a quantitative retrieval directly from the per-pixel methane concentration. For more information on the retrieval algorithm please see Foote et. al., In Review [45].

Integrated Mass Enhancement
The integrated mass enhancement (IME) is the sum of all the methane enhancement above the background concentration present in each plume. To calculate the IME, we first need to identify the pixels associated with the plume and filter out spurious signals. We applied a 200 ppm-m threshold to filter out any potential spurious signals. The 200 ppm-m threshold was determined by reviewing the mean and standard deviation for the matched filter outputs. A 200 ppm-m value represents a round average of one standard deviation below the average mean of the enhanced pixels. To further filter out spurious signals, we calculated the number of pixels in contiguous clusters, any clusters below a certain threshold were filtered out. For the original AVIRIS-NG images, any clusters below 350 contiguous pixels were filtered out. For the 30 m satellite simulation, clusters below four pixels were filtered out and for the 60 m satellite simulation, clusters below two pixels were filtered out. Different thresholds were used for some cases based on empirical adjustments and are documented in Appendix A. We then calculated an integrated mass enhancement (IME) or the sum of the methane present in each plume [6] as follows: where α is the methane value in ppm-m for the n pixels in the plume over the pixel area S and the constant k is used to convert to methane mass units [6].

Flux Estimate
A simple flux estimate can be calculated using the IME, the length of the plume, and the wind speed. The methods for calculating the IME are listed above. To calculate the length of the plume, we identified the two pixels in the plume with the furthest distance and calculated the Euclidean distance between the two pixels. HRRRv3 10 m wind fields in forecast mode were used to estimate wind speeds for the location and time of each AVIRIS-NG scene [46,47]. Using the 3 km grid product, an average wind speed and standard deviation was calculated using a total of 27 HRRRv3 grid cells, a 3 × 3 box centered on the source for three time-steps (plume detection time ±1 hour). The equation for the flux is as follows: where IME is the amount of methane in kg (calculated above); l is the plume length in meters; s is the wind speed meter/hour; and the resulting flux in the units of kg/hour. The uncertainties for the flux estimates result from uncertainties in the wind speeds.

Methane Plumes by Sector
One advantage of using a satellite system to detect methane plumes is the ability to observe a large diversity of methane emitting sectors. Plumes from different sectors have different shapes, sizes, and underlying land cover that can influence how well a plume is detected. Here, we look at plumes from the three largest anthropogenic methane emitting sectors to understand how well a satellite system would perform in each sector. The following sections present the matched filter methane retrieval results for the AVIRIS-NG, 30 m, and 60 m satellite simulations.

Petroleum and Natural Gas
The natural gas and petroleum industry represents about 30%-35% of all anthropogenic methane emissions globally [2,4,48]. Methane emissions from this sector can be challenging to characterize, given the significant variability in source type, plume shape and size, and intermittence. Here, we present three examples from different sources. The first is a gas storage facility that has two emission sources, the second is a well with a methane plume, and the third is a leak from a natural gas distribution line in a neighborhood [7]. All plumes were detectable from the 30 m and 60 m simulated satellite images. Figure 3 depicts the results of two plumes at the Honor Ranch gas storage facility. The source for the top plume is an emergency shutdown stack and for the bottom plume, it is a compressor unit. In the AVIRIS-NG image, these two plumes were distinct, which allows the plumes to be attributed to their exact sources. In the 30 m image, some of the detail of the plume is lost, but two distinct plumes are still visible. At 60 m, the methane enhancement is visible, but the distinction between the two plumes disappears. At the 30 m and 60 m resolution, we could easily attribute the plume to the facility, but not to the exact locations within the facility as we could with the original AVIRIS-NG image.
Remote Sens. 2019, 11, x FOR PEER REVIEW 7 of 19 Figure 3 depicts the results of two plumes at the Honor Ranch gas storage facility. The source for the top plume is an emergency shutdown stack and for the bottom plume, it is a compressor unit. In the AVIRIS-NG image, these two plumes were distinct, which allows the plumes to be attributed to their exact sources. In the 30 m image, some of the detail of the plume is lost, but two distinct plumes are still visible. At 60 m, the methane enhancement is visible, but the distinction between the two plumes disappears. At the 30 m and 60 m resolution, we could easily attribute the plume to the facility, but not to the exact locations within the facility as we could with the original AVIRIS-NG image.  Figure 4 shows the matched filter results from a methane plume from a well. Given the sheer quantity and high density of wells throughout the United States, a satellite system would be well suited to identify those that are emitting methane. This particular example is from a hydrocarbon extraction site within the Aliso Canyon gas storage facility. The details of the plume are most distinct in the original AVIRIS-NG image. In the 30 m and 60 m satellite images, the plume is still clearly detectable, but becomes increasingly less detailed. One concern in the natural gas and petroleum sector are leaks and other fugitive emissions. Not only do these emissions contribute to the overall greenhouse gas budget, but they also represent lost  Figure 4 shows the matched filter results from a methane plume from a well. Given the sheer quantity and high density of wells throughout the United States, a satellite system would be well suited to identify those that are emitting methane. This particular example is from a hydrocarbon extraction site within the Aliso Canyon gas storage facility. The details of the plume are most distinct in the original AVIRIS-NG image. In the 30 m and 60 m satellite images, the plume is still clearly detectable, but becomes increasingly less detailed.
Remote Sens. 2019, 11, x FOR PEER REVIEW 7 of 19 Figure 3 depicts the results of two plumes at the Honor Ranch gas storage facility. The source for the top plume is an emergency shutdown stack and for the bottom plume, it is a compressor unit. In the AVIRIS-NG image, these two plumes were distinct, which allows the plumes to be attributed to their exact sources. In the 30 m image, some of the detail of the plume is lost, but two distinct plumes are still visible. At 60 m, the methane enhancement is visible, but the distinction between the two plumes disappears. At the 30 m and 60 m resolution, we could easily attribute the plume to the facility, but not to the exact locations within the facility as we could with the original AVIRIS-NG image.  Figure 4 shows the matched filter results from a methane plume from a well. Given the sheer quantity and high density of wells throughout the United States, a satellite system would be well suited to identify those that are emitting methane. This particular example is from a hydrocarbon extraction site within the Aliso Canyon gas storage facility. The details of the plume are most distinct in the original AVIRIS-NG image. In the 30 m and 60 m satellite images, the plume is still clearly detectable, but becomes increasingly less detailed. One concern in the natural gas and petroleum sector are leaks and other fugitive emissions. Not only do these emissions contribute to the overall greenhouse gas budget, but they also represent lost One concern in the natural gas and petroleum sector are leaks and other fugitive emissions. Not only do these emissions contribute to the overall greenhouse gas budget, but they also represent lost revenue and in sufficiently large amounts can pose a public safety hazard. The ability to detect leaks over large areas could help improve the mitigation of greenhouse gas emissions and hazards. Figure 5 is an example of an underground natural gas distribution line leak that was detected in 2016 in the Chino Hills neighborhood, CA. This particular leak was detected using the real time methane mapping capability of AVIRIS-NG and the results were shared with the local gas company who quickly confirmed and repaired the leak [19]. The plume is clearly visible in the AVIRIS-NG image and a distinct methane enhancement is visible in the 30 m and 60 m simulated images. revenue and in sufficiently large amounts can pose a public safety hazard. The ability to detect leaks over large areas could help improve the mitigation of greenhouse gas emissions and hazards. Figure  5 is an example of an underground natural gas distribution line leak that was detected in 2016 in the Chino Hills neighborhood, CA. This particular leak was detected using the real time methane mapping capability of AVIRIS-NG and the results were shared with the local gas company who quickly confirmed and repaired the leak [19]. The plume is clearly visible in the AVIRIS-NG image and a distinct methane enhancement is visible in the 30 m and 60 m simulated images. Current and past satellite sensors, like SCIAMACY and TROPOMI, are able to detect regional methane enhancements over areas with heavy oil and gas production, like the methane anomaly over the Four Corners areas in the Western United Sates [50], but cannot identify the exact location of individual point sources. Currently, the only way to locate a specific source is through a ground or airborne campaign, which can be costly, time consuming, and have a limited geographic extent [6]. Therefore, a higher spatial resolution satellite sensor that could identify point sources would help us understand the dynamics of oil and natural gas methane emissions. Other studies have looked into the possibility of using spaceborne sensors to do fine scale methane plume detection. Varon et al. [13] performed a sensitivity study indicating that individual plumes could be mapped with a sufficiently sensitive constellation of point source monitoring satellites, similar to the planned GHGSat fleet. In addition, Cusworth et al. showed that plumes from oil/gas facilities could be detected with the upcoming EnMAP mission [51]. The future generation of imaging spectrometers will help us to improve our understanding of methane emissions from the oil and natural gas sectors.

Landfill/Wastewater Treatment
The waste sector, which includes landfills and wastewater treatment, is thought to make up about 16% of global anthropogenic methane emissions, with the majority coming from landfills [2,4,48]. Globally, landfills take on many different forms, and factors like the composition of the waste, the temperature, the age, and management system can all influence the amount of methane emitted from a landfill [52][53][54]. Of those landfills exhibiting plumes during the California Methane Survey, plumes appeared persistent across multiple years, therefore making them an excellent target for spaceborne studies [7]. Satellite monitoring of these sites can help understand emissions from this under sampled sector, assess gas capture systems, and generally provide a better understanding of landfill emissions. Figure 6 shows the matched filter results from the Newby Island landfill in San Jose, CA. In this image, a very large plume is present, originating from multiple places in the landfill. Note the scale on (C) Satellite 60 m is higher than the other figures, which is due to the large size of the plume. The plume remains visible in all simulated satellite images. Large plumes have been detected with Current and past satellite sensors, like SCIAMACY and TROPOMI, are able to detect regional methane enhancements over areas with heavy oil and gas production, like the methane anomaly over the Four Corners areas in the Western United Sates [49], but cannot identify the exact location of individual point sources. Currently, the only way to locate a specific source is through a ground or airborne campaign, which can be costly, time consuming, and have a limited geographic extent [6]. Therefore, a higher spatial resolution satellite sensor that could identify point sources would help us understand the dynamics of oil and natural gas methane emissions. Other studies have looked into the possibility of using spaceborne sensors to do fine scale methane plume detection. Varon et al. [13] performed a sensitivity study indicating that individual plumes could be mapped with a sufficiently sensitive constellation of point source monitoring satellites, similar to the planned GHGSat fleet. In addition, Cusworth et al. showed that plumes from oil/gas facilities could be detected with the upcoming EnMAP mission [50]. The future generation of imaging spectrometers will help us to improve our understanding of methane emissions from the oil and natural gas sectors.

Landfill/Wastewater Treatment
The waste sector, which includes landfills and wastewater treatment, is thought to make up about 16% of global anthropogenic methane emissions, with the majority coming from landfills [2,4,47]. Globally, landfills take on many different forms, and factors like the composition of the waste, the temperature, the age, and management system can all influence the amount of methane emitted from a landfill [51][52][53]. Of those landfills exhibiting plumes during the California Methane Survey, plumes appeared persistent across multiple years, therefore making them an excellent target for spaceborne studies [7]. Satellite monitoring of these sites can help understand emissions from this under sampled sector, assess gas capture systems, and generally provide a better understanding of landfill emissions. Figure 6 shows the matched filter results from the Newby Island landfill in San Jose, CA. In this image, a very large plume is present, originating from multiple places in the landfill. Note the scale on (C) Satellite 60 m is higher than the other figures, which is due to the large size of the plume. The plume remains visible in all simulated satellite images. Large plumes have been detected with AVIRIS-NG from other landfills in California and a large landfill in India, indicating that this type of methane source can be studied from space. AVIRIS-NG from other landfills in California and a large landfill in India, indicating that this type of methane source can be studied from space. Adjacent to the landfill is a wastewater treatment facility. Figure 7 shows a small plume that is evidently emanating from one of the tanks. The plume is clearly visible in the AVIRIS-NG image, but less so in the 30 m and 60 m images. Given this result, the detection of a methane plume from this particular source will be difficult for the satellite instruments modeled here.

Dairies
Enteric fermentation and manure management make up about 28% of global anthropogenic methane emissions [2,4,49]. This is a significant anthropogenic methane source, but the most difficult to map using this method. Enteric fermentation is an area source and does not produce methane from a single easily identifiable point. This makes mapping methane from cattle with imaging spectroscopy more difficult. However, there is an exception for some methods of manure management. Many manure lagoons from large dairies and some digesters in California produce small yet detectable plumes. Figure 8 shows the matched filter results of a plume from a typical dairy Adjacent to the landfill is a wastewater treatment facility. Figure 7 shows a small plume that is evidently emanating from one of the tanks. The plume is clearly visible in the AVIRIS-NG image, but less so in the 30 m and 60 m images. Given this result, the detection of a methane plume from this particular source will be difficult for the satellite instruments modeled here. AVIRIS-NG from other landfills in California and a large landfill in India, indicating that this type of methane source can be studied from space. Adjacent to the landfill is a wastewater treatment facility. Figure 7 shows a small plume that is evidently emanating from one of the tanks. The plume is clearly visible in the AVIRIS-NG image, but less so in the 30 m and 60 m images. Given this result, the detection of a methane plume from this particular source will be difficult for the satellite instruments modeled here.

Dairies
Enteric fermentation and manure management make up about 28% of global anthropogenic methane emissions [2,4,49]. This is a significant anthropogenic methane source, but the most difficult to map using this method. Enteric fermentation is an area source and does not produce methane from a single easily identifiable point. This makes mapping methane from cattle with imaging spectroscopy more difficult. However, there is an exception for some methods of manure management. Many manure lagoons from large dairies and some digesters in California produce small yet detectable plumes. Figure 8 shows the matched filter results of a plume from a typical dairy

Dairies
Enteric fermentation and manure management make up about 28% of global anthropogenic methane emissions [2,4,48]. This is a significant anthropogenic methane source, but the most difficult to map using this method. Enteric fermentation is an area source and does not produce methane from a single easily identifiable point. This makes mapping methane from cattle with imaging spectroscopy more difficult. However, there is an exception for some methods of manure management. Many manure lagoons from large dairies and some digesters in California produce small yet detectable plumes. Figure 8 shows the matched filter results of a plume from a typical dairy manure lagoon. It is clearly visible in the AVIRIS-NG images and remains detectable in the 30 m and 60 m images.  An anaerobic digester is an enclosed area where the breakdown of organic material (manure) is promoted and the biogas is then captured and used as an energy source. Many studies have measured the potential biogas produced from digesters in order to estimate gas production [55,56]. However, few studies have looked at the gas lost due to leaks. In addition, emissions from manure can be well constrained with isotopic measurements, but this does not allow for the distinction between an open manure lagoon and a digester [57]. Regular satellite measurements of large dairy production areas like California's San Joaquin Valley would improve understanding of the dynamics of methane emission from manure management. Figure 9 shows the matched filter results of three plumes in the upper left corner coming off a digester in California. In the 30 m and 60 m simulated images, the plume is still visible, but has blended into one larger plume instead of three smaller ones.

Controlled Release
In September 2018, a controlled release experiment was performed to test the accuracy of the AVIRIS-NG methane retrievals. Natural gas was released by a pipeline operator at a known flux while AVIRIS-NG flew over. We analyzed one scene from this experiment to verify that our methods were producing reasonable results (Figure 10). The flux during this flight was 103.71 kg/h and the retrieval from AVIRIS-NG with our methods was 117.59 ± 15.53 kg/h. The agreement between the An anaerobic digester is an enclosed area where the breakdown of organic material (manure) is promoted and the biogas is then captured and used as an energy source. Many studies have measured the potential biogas produced from digesters in order to estimate gas production [54,55]. However, few studies have looked at the gas lost due to leaks. In addition, emissions from manure can be well constrained with isotopic measurements, but this does not allow for the distinction between an open manure lagoon and a digester [56]. Regular satellite measurements of large dairy production areas like California's San Joaquin Valley would improve understanding of the dynamics of methane emission from manure management. Figure 9 shows the matched filter results of three plumes in the upper left corner coming off a digester in California. In the 30 m and 60 m simulated images, the plume is still visible, but has blended into one larger plume instead of three smaller ones.  An anaerobic digester is an enclosed area where the breakdown of organic material (manure) is promoted and the biogas is then captured and used as an energy source. Many studies have measured the potential biogas produced from digesters in order to estimate gas production [55,56]. However, few studies have looked at the gas lost due to leaks. In addition, emissions from manure can be well constrained with isotopic measurements, but this does not allow for the distinction between an open manure lagoon and a digester [57]. Regular satellite measurements of large dairy production areas like California's San Joaquin Valley would improve understanding of the dynamics of methane emission from manure management. Figure 9 shows the matched filter results of three plumes in the upper left corner coming off a digester in California. In the 30 m and 60 m simulated images, the plume is still visible, but has blended into one larger plume instead of three smaller ones.

Controlled Release
In September 2018, a controlled release experiment was performed to test the accuracy of the AVIRIS-NG methane retrievals. Natural gas was released by a pipeline operator at a known flux while AVIRIS-NG flew over. We analyzed one scene from this experiment to verify that our methods were producing reasonable results ( Figure 10). The flux during this flight was 103.71 kg/h and the retrieval from AVIRIS-NG with our methods was 117.59 ± 15.53 kg/h. The agreement between the

Controlled Release
In September 2018, a controlled release experiment was performed to test the accuracy of the AVIRIS-NG methane retrievals. Natural gas was released by a pipeline operator at a known flux while AVIRIS-NG flew over. We analyzed one scene from this experiment to verify that our methods were producing reasonable results ( Figure 10). The flux during this flight was 103.71 kg/h and the retrieval from AVIRIS-NG with our methods was 117.59 ± 15.53 kg/h. The agreement between the AVIRIS-NG flux and actual flux was good, however, the agreement for the 30 m and 60 m images was not as close ( Table 2). This was a smaller plume and the entire plume spanned one or two pixels in the 60 m satellite simulation, therefore calculating an accurate flux was more difficult. Regardless, the plumes were visible in both the 30 m and 60 m, which indicates that plumes with a flux as small as~100 kg/h will be visible with the next generation of satellite imaging spectrometers. AVIRIS-NG flux and actual flux was good, however, the agreement for the 30 m and 60 m images was not as close (Table 2). This was a smaller plume and the entire plume spanned one or two pixels in the 60 m satellite simulation, therefore calculating an accurate flux was more difficult. Regardless, the plumes were visible in both the 30 m and 60 m, which indicates that plumes with a flux as small as ~100 kg/h will be visible with the next generation of satellite imaging spectrometers.

Flux
The fluxes for the AVIRIS-NG, 30 m, and 60 m images were calculated using the methods presented above. Calculating the flux allows the scientific community to quantitatively assess the volume of methane emitted per source and sector. We tested an image from a controlled release experiment (presented above) to empirically validate our flux estimates. We calculated a flux estimation of 117.59 ± 15.53 kg/h for the AVIRIS-NG image of the controlled release experiment. The reported flux at the time of the image was 103.71 kg/h. These results are close enough that we feel confident that the AVIRIS-NG images provide a good estimation of the magnitude of the flux. We can therefore compare the results from the satellite simulations to the original AVIRIS-NG image to understand how well a satellite system would perform. In general, the flux estimates from the satellite simulations are consistent with the flux estimations from the AVIRIS-NG image. Table 2 provides the results of the flux estimates for the AVIRIS-NG, the 30 m, and 60 m images. Figure 11 shows the flux results from the 30 m and 60 m image plotted against the flux results from the AVIRIS-NG image. The relationship between the satellite simulations and the AVIRIS-NG image was strong (R 2 = 0.98), although the linear regression line did not match the 1:1 line, the error bars crossed or came close to the 1:1 line. The satellite simulations tended to overestimate the flux for both simulations. In addition, the overestimation appears to become larger as the size of the methane plume increases, however, when the error is normalized by the size of the plume, this relationship breaks down. One explanation for the overestimation is that the coarser spatial resolution inflates the spatial extent of the plume, which increases the IME and plume length. However, the IME tends to increase more than the plume length, causing the flux to also increase. In a similar study, Cusworth et al. found that the satellite simulation underestimated the AVIRIS-NG retrieval, but they used a different retrieval algorithm, a different plume mask, and did not simulate a 60 m instrument [51]. More work will be needed to be done to understand precisely how increasing spatial resolution changes the IME and therefore the flux.

Flux
The fluxes for the AVIRIS-NG, 30 m, and 60 m images were calculated using the methods presented above. Calculating the flux allows the scientific community to quantitatively assess the volume of methane emitted per source and sector. We tested an image from a controlled release experiment (presented above) to empirically validate our flux estimates. We calculated a flux estimation of 117.59 ± 15.53 kg/h for the AVIRIS-NG image of the controlled release experiment. The reported flux at the time of the image was 103.71 kg/h. These results are close enough that we feel confident that the AVIRIS-NG images provide a good estimation of the magnitude of the flux. We can therefore compare the results from the satellite simulations to the original AVIRIS-NG image to understand how well a satellite system would perform. In general, the flux estimates from the satellite simulations are consistent with the flux estimations from the AVIRIS-NG image. Table 2 provides the results of the flux estimates for the AVIRIS-NG, the 30 m, and 60 m images. Figure 11 shows the flux results from the 30 m and 60 m image plotted against the flux results from the AVIRIS-NG image. The relationship between the satellite simulations and the AVIRIS-NG image was strong (R 2 = 0.98), although the linear regression line did not match the 1:1 line, the error bars crossed or came close to the 1:1 line. The satellite simulations tended to overestimate the flux for both simulations. In addition, the overestimation appears to become larger as the size of the methane plume increases, however, when the error is normalized by the size of the plume, this relationship breaks down. One explanation for the overestimation is that the coarser spatial resolution inflates the spatial extent of the plume, which increases the IME and plume length. However, the IME tends to increase more than the plume length, causing the flux to also increase. In a similar study, Cusworth et al. found that the satellite simulation underestimated the AVIRIS-NG retrieval, but they used a different retrieval algorithm, a different plume mask, and did not simulate a 60 m instrument [50]. More work will be needed to be done to understand precisely how increasing spatial resolution changes the IME and therefore the flux.

Spatial Resolution and Signal-to Noise Ratio
In this analysis, the spatial resolution and SNR are the two variables that distinguish the 30 m and 60 m satellite simulations. Sensor design often requires compromises between different performance measures. For example, designers may sacrifice SNR for finer spatial resolution and vice versa. We looked at which of these factors had the largest impact on methane gas mapping. We used the Honor Ranch gas storage facility as an example and tested how our retrieval changed with different noise levels; no added noise, ~400 SNR, and ~200 SNR. We found that visually, the plume became less distinct from the background and the presence of spurious signals increased, which in turn slightly increased the flux estimate ( Figure 12). We also tested the image with the same SNR at the native resolution (1.6 m), 30 m, and 60 m. We found that the plume became less detailed with coarsening spatial resolution, indicating that spatial resolution, rather than SNR, had a larger influence on detecting plumes. In addition, we performed an Analysis of Variance (ANOVA) with a Tukey post hoc using the flux and standard deviation, and found that the only image that was significantly different (p ≤ 0.05) from the original AVIRIS-NG was the 60 m image. More plumes as well as a larger range in SNR and pixel size will be needed to accurately quantify how noise versus spatial resolution effects plume retrieval. However, these results indicate that changing the spatial resolution has a larger impact than changing the noise.

Spatial Resolution and Signal-to Noise Ratio
In this analysis, the spatial resolution and SNR are the two variables that distinguish the 30 m and 60 m satellite simulations. Sensor design often requires compromises between different performance measures. For example, designers may sacrifice SNR for finer spatial resolution and vice versa. We looked at which of these factors had the largest impact on methane gas mapping. We used the Honor Ranch gas storage facility as an example and tested how our retrieval changed with different noise levels; no added noise,~400 SNR, and~200 SNR. We found that visually, the plume became less distinct from the background and the presence of spurious signals increased, which in turn slightly increased the flux estimate ( Figure 12). We also tested the image with the same SNR at the native resolution (1.6 m), 30 m, and 60 m. We found that the plume became less detailed with coarsening spatial resolution, indicating that spatial resolution, rather than SNR, had a larger influence on detecting plumes. In addition, we performed an Analysis of Variance (ANOVA) with a Tukey post hoc using the flux and standard deviation, and found that the only image that was significantly different (p ≤ 0.05) from the original AVIRIS-NG was the 60 m image. More plumes as well as a larger range in SNR and pixel size will be needed to accurately quantify how noise versus spatial resolution effects plume retrieval. However, these results indicate that changing the spatial resolution has a larger impact than changing the noise.

Limitations
The satellite simulations indicate that point source methane mapping will be possible with the future generation of satellite imaging spectrometers. However, there are some limitations with the simulations presented here. While the specifications for the simulation represent the best available knowledge about the design of a future spectrometer (EMIT), there remains a possibility that specifications will change before launch. Other satellite designs might include different spectral, spatial, or SNR characteristics not modeled here, however these are unlikely to be drastically different. In addition, this study did not account for other factors associated with satellite sensors including potential orbits, cloud cover, sun angles, and surface brightness. In particular, surface brightness has proven to have a large influence on methane retrievals [51,58]. While there is inherent variability in surface brightness in the AVIRIS-NG scenes, we were not able to control this variable. Finally, while the flightlines from this study represent a range of emissions over variable surface terrain, more challenging examples containing lower fluxes were not tested.
Other limitations in the study include the methods we used to create a plume mask and the flux estimate. The plume mask is intended to isolate the plume and filter out spurious signals or false positives. Spurious signals were present in both the airborne data and the satellite simulations (Figures 3-10). For this study, we attempted to be as consistent as possible and therefore generally used consistent thresholds to isolate the plume. It should be noted, however, that more tailored thresholding could produce better results in future studies. The thresholding and segmentation can have a large impact on the results, more work is needed to determine the best practices for how to approach this. Additional work is required to assess the impact that these spurious signals may have on our IME and flux estimates. This study also used a simple method to estimate the flux of the plumes. More robust flux calculations could lead to more accurate results.

Conclusions
We tested the potential to use future Earth observing satellites such as EMIT or SBG to map and measure methane plumes from the three main anthropogenic emission sectors. These particular

Limitations
The satellite simulations indicate that point source methane mapping will be possible with the future generation of satellite imaging spectrometers. However, there are some limitations with the simulations presented here. While the specifications for the simulation represent the best available knowledge about the design of a future spectrometer (EMIT), there remains a possibility that specifications will change before launch. Other satellite designs might include different spectral, spatial, or SNR characteristics not modeled here, however these are unlikely to be drastically different. In addition, this study did not account for other factors associated with satellite sensors including potential orbits, cloud cover, sun angles, and surface brightness. In particular, surface brightness has proven to have a large influence on methane retrievals [50,57]. While there is inherent variability in surface brightness in the AVIRIS-NG scenes, we were not able to control this variable. Finally, while the flightlines from this study represent a range of emissions over variable surface terrain, more challenging examples containing lower fluxes were not tested.
Other limitations in the study include the methods we used to create a plume mask and the flux estimate. The plume mask is intended to isolate the plume and filter out spurious signals or false positives. Spurious signals were present in both the airborne data and the satellite simulations (Figures 3-10). For this study, we attempted to be as consistent as possible and therefore generally used consistent thresholds to isolate the plume. It should be noted, however, that more tailored thresholding could produce better results in future studies. The thresholding and segmentation can have a large impact on the results, more work is needed to determine the best practices for how to approach this. Additional work is required to assess the impact that these spurious signals may have on our IME and flux estimates. This study also used a simple method to estimate the flux of the plumes. More robust flux calculations could lead to more accurate results.

Conclusions
We tested the potential to use future Earth observing satellites such as EMIT or SBG to map and measure methane plumes from the three main anthropogenic emission sectors. These particular future satellites will measure in the SWIR with 7-10 nm spectral sampling, the spatial resolution will range from 30 m to 60 m pixels, and will have a SNR in the SWIR between 200 and 400. To test the potential of using these instruments for high resolution methane mapping, we simulated two proposed satellite designs from the AVIRIS-NG data. The first was a 30 m pixel resolution with a~200 SNR in the SWIR, the second was a 60 m pixel resolution with~400 SNR in the SWIR. We then ran a matched filter methane retrieval algorithm on the simulated satellite images and compared it to the original AVIRIS-NG.
We found that for almost all plumes, the methane enhancement remained visible at the coarser spatial resolutions. This indicated that most point source emissions from the largest anthropogenic sources should be detectable, at minimum, in a 30 m resolution image. We found that when we quantified the flux of the plume in the simulated images, it correlated well with the flux calculated from the AVIRIS-NG results. In addition, for the controlled release experiment, the flux calculated from the AVIRIS-NG data was similar to the known flux. Within the limits of our study, we also found that the spatial resolution had a larger impact on the results than the SNR. We conclude that a satellite system will be able to map and quantify the biggest of the point source emitters and should be able to detect plumes as small as~100 kg/h, although accurate quantification of flux may be more difficult. These plumes may only be a small fraction of the total number of plumes, but likely represent up to 40% of emitted methane.
While the future generation of satellite imaging spectrometers were not designed for high resolution gas mapping, this work indicates that they will be able to map methane plumes and quantify emission rates. However, imaging spectrometer concepts have been developed that would utilize finer spectral resolution to improve gas sensitivities [50,58]. Eventually, these sensors will become integrated into larger greenhouse gas monitoring schemes and will help scientists better understand global methane emissions.

Conflicts of Interest:
The authors declare no conflicts of interest.  Figure A1. Example a radiance spectrum for the original AVIRIS-NG (red) and example of the same spectrum adjusted for TOA (blue). Figure A2. Example of a radiance spectrum at 5 nm sampling (blue) and the same example at 7.5 nm sampling (red). Note the decreased detail in the gas absorption bands.  Figure A1. Example a radiance spectrum for the original AVIRIS-NG (red) and example of the same spectrum adjusted for TOA (blue). Figure A1. Example a radiance spectrum for the original AVIRIS-NG (red) and example of the same spectrum adjusted for TOA (blue). Figure A2. Example of a radiance spectrum at 5 nm sampling (blue) and the same example at 7.5 nm sampling (red). Note the decreased detail in the gas absorption bands.  Figure A2. Example of a radiance spectrum at 5 nm sampling (blue) and the same example at 7.5 nm sampling (red). Note the decreased detail in the gas absorption bands.