Detection of Methane Emission from a Local Source Using GOSAT Target Observations

Emissions of atmospheric methane (CH4), which greatly contributes to radiative forcing, have larger uncertainties than those for carbon dioxide (CO2). The Thermal And Near-infrared Sensor for carbon Observation Fourier-Transform Spectrometer (TANSO-FTS) onboard the Greenhouse gases Observing SATellite (GOSAT) launched in 2009 has demonstrated global grid observations of the total column density of CO2 and CH4 from space, and thus reduced uncertainty in the global flux estimation. In this paper, we present a case study on local CH4 emission detection from a single-point source using an available series of GOSAT data. By modifying the grid observation pattern, the pointing mechanism of TANSO-FTS targets a natural gas leak point at Aliso Canyon in Southern California, where the clear-sky frequency is high. To enhance local emission estimates, we retrieved CO2 and CH4 partial column-averaged dry-air mole fractions of the lower troposphere (XCO2 (LT) and XCH4 (LT)) by simultaneous use of both sunlight reflected from Earth’s surface and thermal emissions from the atmosphere. The time-series data of Aliso Canyon showed a large enhancement that decreased with time after its initial blowout, compared with reference point data and filtered with wind direction simulated by the Weather Research and Forecasting (WRF) model.


Introduction
Among trace gases in the atmosphere, methane (CH 4 ) contributes significantly to radiative forcing, second only to carbon dioxide (CO 2 ), although its emissions estimates have larger uncertainties than those of CO 2 . Major anthropogenic source sectors are oil/gas production, coal mining, livestock, landfills, wastewater, and rice cultivation. A better understanding of the distribution and amount of emissions contributes to effective reductions in CH 4 emissions. In situ measurements taken on the ground or by aircraft, using such instruments as cavity-ringdown spectrometers, have high accuracy and precision, but the number of observation points is limited. Airborne remote sensing has demonstrated CH 4 point-source emissions and their plumes using imaging capabilities [1][2][3]. With its high spatial resolution, these measurements can detect the fine horizontal structure of plumes from individual emission source, but these are temporary. Long-term measurements are needed for more systematic estimates of CH 4 emissions. Satellite observations can provide global CH 4 data in a long time-series (over a decade) to investigate anthropogenic CH 4 emissions. Compared to the wide-spread urban CO 2 emissions, such as those from transportation, most anthropogenic CH 4 source areas are smaller than a single pixel in satellite footprints. Such areas can be treated as point sources for existing  [17,18]. To capture all CH 4 emissions, the satellite footprint must cover the point source area. In this study, we modified the pointing pattern locally and retrieved the columnar densities of lower troposphere (LT) CH 4 and CO 2 by simultaneous use of reflected shortwave-infrared (SWIR) solar light and thermal infrared (TIR) emissions. This paper describes how a local CH 4 emission could be detected using existing GOSAT functionalities and wind simulation. We also discuss the possible causes of uncertainty in the local flux estimates and suggest possibilities that might reduce such uncertainties using imaging capabilities on the local flux estimate through satellite data.

GOSAT Observations
GOSAT carries the Thermal And Near-infrared Sensor for carbon Observation Fourier-Transform Spectrometer (TANSO-FTS), which simultaneously observes both reflected SWIR solar light and TIR emissions with a single FTS mechanism. Table 2 lists the principal specifications of the GOSAT orbit and the TANSO-FTS spectral band. As described in Table 3, it measures narrow spectral bands at 0.76 µm for oxygen (O 2 ), 1.6 µm for CO 2 and CH 4 , and 2.0 µm for CO 2 , with a spectral sampling interval of 0.2 cm −1 . It also has a wide spectral band to measure thermal emissions from Earth's surface and atmosphere, including CO 2 and CH 4 . The reflected SWIR solar light passes through Earth's troposphere, reflects over the surface, and then returns to space. The O 2 A band at 0.76 µm provides the surface pressure (P surf ). Two linear polarization bands provide information on light-path modification and make accurate remote sensing possible, even under aerosol and thin-cloud contaminated conditions. To maximize the optical throughput of FTS, GOSAT has a circular instantaneous field of view (IFOV) of 15.8 mrad to collect enough photons to improve the signal-to-noise ratio (SNR). The footprint is 10.5 km in diameter for nadir observations.

Targeting Point Sources
TANSO-FTS has a two-axis agile pointing system, which allows cross-track (CT) and along-track (AT) motions of ±35 • and ±20 • , respectively, and compensates for the satellite image-motion during interferogram acquisition. It was originally designed for grid scan observations and viewing onboard calibration sources. It takes 4 s to acquire a single interferogram and 0.6 s for pointing and turnaround of the FTS mechanism. For the first six years of GOSAT operations, TANSO-FTS operated mainly in three-point CT scan mode to monitor global CO 2 and CH 4 with a three-day revisit cycle, using the primary part of the redundant pointing systems with a grid of about 300 km. After the pointing mechanism was switched from primary to secondary on 26 January 2015, the pointing showed more accurate geolocation and stable settling than the degraded primary system. We decided to make more frequent target observations, by uploading AT and CT pointing angles and observation timing as commands from the ground every day [12]. The pointing bias and its fluctuations have generally been smaller than 0.1 • , corresponding to a 1 km surface footprint bias or fluctuations. When a sufficiently strong point source is located within the 10.5 km footprint in nadir, GOSAT can capture total CH 4 emissions that spread horizontally and vertically. Over the three-day revisit cycle, data of 56,000 locations are acquired, most of which are grid observations and onboard calibrations. About 1000 locations are allocated to target observations, such as the Total Carbon Column Observing Network (TCCON) validation sites, desert areas for radiometric calibration, megacities, or large emission sources [12,19].

Derivation of CH 4 Partial Column-averaged Density in the Lower Troposphere
Estimation of the CH 4 local emission requires the columnar density of the LT CH 4 in order to enhance near-surface density and reduce the inflow in the upper atmosphere. Existing GOSAT Level 2 products have provided the column-averaged dry-air mole fractions of CO 2 and CH 4 (XCO 2 and XCH 4 , respectively), but only from the three SWIR-band data [20][21][22][23][24][25]. The standard deviation of XCO 2 and XCH 4 are 2.1 ppm or 0.5% and 13 ppb or 0.7%, respectively, compared with the TCCON data. Their biases are −1.48 ppm and −5.9 ppb for XCO 2 and XCH 4 , respectively. Columnar CH 4 density is the average value of the LT, upper troposphere (UT), and the stratosphere. Therefore, it is influenced by inflows of CH 4 in the UT and the stratosphere from a much wider area. In estimating CH 4 emissions from a local point source, it is highly desirable to extract columnar density only of the LT CH 4 . TIR spectra emitted from CO 2 and CH 4 contain vertical profile information, due to the temperature gradient. Conventional algorithms for the TIR band have attempted to acquire CO 2 and CH 4 densities for ten vertical layers in the troposphere, resulting in unstable retrievals due to the uncertainties of radiometric calibrations [26]. In addition, TIR data alone do not contain enough information of near-surface CO 2 and CH 4 , and thus are inadequate for local emission analysis.
We reduced the number of vertical layers for a more robust retrieval [27,28]. We can retrieve the difference between the partial column-averaged dry-air mole fractions of the two individual layers of LT and UT (XCO 2 (LT), XCO 2 (UT), XCH 4 (LT), and XCH 4 (UT)) by combining TIR and SWIR spectra data simultaneously, thereby constraining the accurate total column density of XCO 2 and XCH 4 .
We retrieved five vertical layers: two in the troposphere and three in the stratosphere. We used the entire spectral range acquired with the FTS listed in Table 3 including absorption bands of O 2 A at 0.76 µm, CO 2 and CH 4 at 1.6 µm, and CO 2 at 2.0 µm, and thermal emission from atmospheric CO 2 and CH 4 . The simultaneous use of two liner polarizations can provide accurate light-path modification under thick aerosol conditions. The retrieval method is the maximum a posteriori solution found by minimizing a cost function. For reflected solar light, we used the vector equation of radiative transfer for the diffuse components of the Stokes parameters. We define vertical layers of LT and UT not by temperature, but by the retrieved P surf from individual O 2 A band data as retrieved vertical temperature has a larger uncertainty. The pressure-height ranges of the LT and UT were taken as 0.6-1 P surf and 0.2-0.6 P surf , respectively. Both LT and UT have the same air masses. The exact ranges of each vertical layer were determined by individually retrieved P surf . In the case of ocean data, the vertical range of LT becomes 0-4 km. We allocate three layers in the stratosphere to stabilize the retrieval. Most pieces of information on the stratosphere come from a priori data. The typical degrees of freedom for the signal (DFS) are roughly 1.8 for XCH 4 (the total column) and 0.8 for XCH 4 (LT). Therefore, most of the lower troposphere data are acquired from GOSAT observations. As the LT includes the entire boundary layer, analysis using XCO 2 (LT) and XCH 4 (LT) can double the signal of local emissions and remove the effects of CO 2 and CH 4 variability in the UT, which typically extends over a much wider area. We validated our XCO 2 (LT) and XCH 4 (LT) retrieval algorithm by coincident spiral flights with an airborne spectrometer during the annual calibration and validation campaigns at Railroad Valley in Nevada, USA [29][30][31][32]. The airplane carried the PICARRO cavity-ringdown spectrometer and measured vertical profiles of in-situ CO 2 and CH 4 density from the surface to 8 km. For this analysis, we used the Level 1B radiance spectra of version 210 to produce XCO 2 (LT), XCO 2 (UT), XCH 4 (LT), and XCH 4 (UT) [33].

Detectivity and Site Selection
For a detectivity study, the CH 4 flux emitted from a point source using GOSAT target observation is modeled by Equation (1).
where F CH 4 , a p , V, and ∆XCH 4 (LT) denote CH 4 local flux, the partial air mass of the LT within the GOSAT footprint, wind speed in the boundary layer, and enhancement by emissions, respectively; L s is the average cross-wind length, which is the average distance between the source location and edge of the GOSAT footprint; and the suffixes s and b denote source and background. We assume that CH 4 emitted from a point source remains in the LT from the overpass time until the plume reaches the edge of the footprint. Considering the uncertainty of 13 ppb in XCH 4 (LT) s and XCH 4 (LT) b and the footprint area of 87 km 2 in Equation (1), the local flux must be larger than the uncertainty level of 5.0 tons of CH 4 per hour in the case of V = 2 m/s to be detected by GOSAT. In our case study for demonstrating the observation capability of GOSAT for monitoring large emission trends, we selected Southern California, USA, where clear skies are frequent. The largest emission in the EDGAR grid database in Southern California is lower than 10.0 tons of CH 4 per hour, which is close to the level of uncertainty. For a case study, we need a significantly higher emission, thus we selected the natural gas leak at Aliso Canyon (Figure 1a), a blowout event. According to the company's measurements, a gas leak over three months amounted to the emission of at least 97 Gg of CH 4 , which gives an average emission rate of 45 tons of CH 4 per hour [34]. GOSAT can view Aliso Canyon from east-orbit path 36 on the first day of the three-day revisit and from west-orbit path 37 on the following day. Considering that TANSO-FTS needs 4.6 s to acquire a single datum and has a ground speed of 6.9 km/s, we can allocate four target observations for each of paths 36 and 37. After the blowout from natural gas storage was reported on 23 October 2015 [35], GOSAT started observing the target twice every three days.  Figure A2 shows a reduction in the positive bias after normalization. Figure  As illustrated in Figure 1b, the leak point was located at the edge of Aliso Canyon-N and outside of Aliso Canyon-S. The CH4 plume should not have entered the GOSAT footprint from any direction between 90° (east) and 315°. We rejected the data not meeting this criterion, based on Weather Research and Forecasting (WRF) model calculations [37]. We used simulated wind direction at the location of the leak point (34.31°N, 118.56°W), height of 906.9 hPa, and time of 20:50 UTC. As a result, large fluctuations in data were removed. The selected data showed a large enhancement in November (just after the blowout), followed by a clear decrease. The outliers above the uncertainty level of 0.056 ppb/ppm of △XCH4(LT)/XCO2(LT) were mostly removed, as shown in Figure 4. The flat topography of Aliso Canyon-S allowed for more accurate monitoring capability in detecting the blowout and confirming blocking of the leak. Figure 4 presents some data that have negative values larger than the uncertainty level, probably due to CH4 inflow, which is not correlated with CO2 enhancement. Simi Valley is not an ideal upwind reference.   Thompson et al. [1] indicated that the wind direction over the leak point changed frequently. For a north wind, the downwind plume entered the GOSAT footprint of Aliso Canyon-S. Figure 2a,b shows observed XCH 4 and XCH 4 (LT) and their DFS in retrieval. The LT data showed enhancement soon after the blowout event in 2015, which is much clearer than total column data. DFS values close to 1 demonstrate information content in the observed XCH 4 (LT). Figure 3 shows the retrieved XCH 4 (LT) and XCH 4 (UT) at Aliso Canyon-S. The UT data did not indicate effect by the blowout event. Year-long data for both XCH 4 (LT) and XCH 4 (UT) showed decreases in summer, whereas, in the other seasons, the data largely fluctuated. Figure 3 shows XCH 4 (UT) data that are higher than those of XCH 4 (LT). The possible cause is an inflow from other CH 4 emission sources. There are several CH 4 emission sources such as oil fields and dairy farms in California. We selected pairs of successfully retrieved data at Simi Valley as a reference to XCH 4 (LT) at Aliso Canyon-S under clear-sky conditions. The detailed processes of background subtraction, normalization, and filtering are described in the Appendix A. Figure A1 shows the XCH 4 (LT) enhancement from Simi Valley at:

Trend Analysis for the Gas Leak in Aliso Canyon
(a) Aliso Canyon-S; and (b) Aliso Canyon-N and Burbank after background subtraction. Figure A2 shows a reduction in the positive bias after normalization. Figure A3 shows the XCH 4 (LT)/XCO 2 (LT) ratio and the WRF wind direction at the leak point during GOSAT overpass for filtering. XCH 4 (LT) data for both Simi Valley and Aliso Canyon-N/-S were possibly influenced by pollution inflow from the greater Los Angeles area. The measured XCH 4 (LT) at these locations varied significantly, depending on wind conditions, resulting in large fluctuations. We assumed that: (1) urban pollution caused both CH 4 and CO 2 enhancements; (2) inflow to the reference point of Simi Valley and to Aliso Canyon had the same CH 4 to CO 2 ratios; and (3) no CO 2 was emitted at the gas leak point on the hills of Aliso Canyon. XCH 4 (LT) normalized by the retrieved XCO 2 (LT) should remove the inflow portion inside the footprint.  As illustrated in Figure 1b, the leak point was located at the edge of Aliso Canyon-N and outside of Aliso Canyon-S. The CH 4 plume should not have entered the GOSAT footprint from any direction between 90 • (east) and 315 • . We rejected the data not meeting this criterion, based on Weather Research and Forecasting (WRF) model calculations [37]. We used simulated wind direction at the location of the leak point (34.31 • N, 118.56 • W), height of 906.9 hPa, and time of 20:50 UTC. As a result, large fluctuations in data were removed. The selected data showed a large enhancement in November (just after the blowout), followed by a clear decrease. The outliers above the uncertainty level of 0.056 ppb/ppm of XCH 4 (LT)/XCO 2 (LT) were mostly removed, as shown in Figure 4. The flat topography of Aliso Canyon-S allowed for more accurate monitoring capability in detecting the blowout and confirming blocking of the leak. Figure 4 presents some data that have negative values larger than the uncertainty level, probably due to CH 4 inflow, which is not correlated with CO 2 enhancement. Simi Valley is not an ideal upwind reference.

Reducing Uncertainties in the Local Flux Estimation
There were four major causes of uncertainty in the case study for estimating the local flux from a point source using Equation (1): retrieval of XCH 4 (LT), V, L s , and background removal.
(a) Satellite remote sensing has common retrieval errors in forward calculation, in addition to random instrument noise and radiometric calibration errors. Multiple scattering by aerosols and clouds, bidirectional reflection on Earth's surface, uncertainties in line parameters, and the solar model cause both bias and random errors. It is difficult to significantly reduce the current XCH 4 retrieval uncertainty of 13 ppb. The GOSAT IFOV is too large, compared to the areas of the Aliso Canyon CH 4 gas blowout. By improving spatial resolution to the scale of the emissions area, differential XCH 4 (LT) measurements can be significantly enhanced, possibly enabling more accurate estimates of local CH 4 emission.
(b) The number of sites where wind speed and direction are monitored regularly is limited. The data were available only at the Burbank Airport, which is 22 km away from the emissions source. In addition, their vertical profiles were not measured. Thus, we instead have to use a simulation such as the WRF model with large uncertainties. We need more validation data for the wind speed simulation in the boundary layer.
(c) In Aliso Canyon, the wind direction during the GOSAT overpass varied each day. The point of leakage was decentered for Aliso Canyon-N and outside the footprint for Aliso Canyon-S. Without accurate wind direction information, the flux cannot be estimated. Targeting the center of the emissions point can cover a plume in any wind direction. The present single-pixel GOSAT with its two-axis pointing system limits the number of observation points. Multi-pixel data will help in inferring the wind direction and in selecting proper reference points upwind for each flux estimate.
(d) A proper upwind reference closer to the emissions source assumed as a background can remove CH 4 contamination from the surrounding area. Imaging capability and a smaller footprint with a flat topography would provide a more accurate background level than the current method using Simi Valley data. For example, XCH 4 acquired by TROMOMI may be useful for this purpose [38]. A more robust algorithm for topography is also needed.
TANSO-FTS was designed to demonstrate accurate remote sensing from space with marginal spectral resolution using a single-pixel detector. FTS needs mechanical scan time to acquire the interferogram and the spectral resolution becomes worse when the off-axis angle becomes larger. Conventional space borne FTSs use near-optical axis areas: 1, 2 × 2, and 3 × 3 pixels for GOSAT, IASI, and CrIS, respectively. A grating spectrometer with a two-dimensional detector can acquire multiple pixel data in the CT direction within an electrical integration time, which is much shorter than the mechanical scan time of the GOSAT FTS. We can allocate spectral information in one dimension and CT electrical scan for the other dimension. Aberration-corrected grating technology can widen the CT field of view. With satellite motion, we can acquire image data that have at least 1000 times more sampling points than the existing GOSAT. To achieve this, we propose an imaging capability with a 1 km 2 spatial resolution to monitor plume orientation and simultaneous measurement of short-lived species to estimate the wind speed and direction [39].

Conclusions
In this paper, we present a case study on CH 4 emissions from a natural gas leak at Aliso Canyon using target observations with a pointing mechanism. We used the time-series of the partial column-averaged dry-air mole fractions of the lower troposphere, XCH 4 (LT), retrieved by simultaneous use of reflected solar light and thermal emission. Frequent and long-term satellite observations, with appropriate background references, can detect CH 4 enhancements from local-point emission sources and even capture total emissions. XCH 4 (LT) minimizes the effects of inflow of CH 4 in the UT and the stratosphere. The CH 4 blowout event at Aliso Canyon was clearly seen in the temporal changes in XCH 4 (LT). Normalization with XCO 2 (LT) reduced the effects of CH 4 inflow to the target areas. Geometric mismatch between the emission point and the center of the GOSAT footprint needed screening using the WRF wind direction data. However, the large footprint made it difficult to enhance XCH 4 (LT) further with more flat topography and estimate the flux variation quantitatively. Imaging capability and a higher spatial resolution with a two-dimensional multi-pixel detector may reduce uncertainties in the local flux estimation from point sources.
Author Contributions: A.K. contributed to satellite operation, data analyses, and preparation of the manuscript. N.K. and F.K. are responsible for partial-column density retrieval and WRF calculation, respectively. H.S., K.S., and Y.K. selected target observation points and reviewed the data analysis. All authors have read and agreed to the published version of the manuscript.

Conflicts of Interest:
The authors declare no conflict of interest.

Appendix A. Background Subtraction, Normalization, and Filtering Processes
Appendix A. 1

. Background Subtraction Using Reference Point Data
Assuming the LT inflow is locally spatially constant, Equation (A1) calculates the enhancement from the point source using nearby reference data as follows: where XCH 4 (LT) s and XCH 4 (LT) r are XCH 4 (LT) over the source and reference points measured by GOSAT within the same orbit path, respectively. Figure

Appendix A.2. Normalization Process
As there are other CH 4 and large CO 2 emission sources in Southern California, we introduced the XCH 4 (LT)/XCO 2 (LT) ratio to reduce contamination by temporal inflow, which is not equal to the source and reference. Assuming the common virtual background for source and sink, the enhancement trend of Aliso Canyon can be monitored as presented in Equation (A2). The suffixes e, b, si, and ri denote enhancement by emissions from the source, background, and inflow at the source and reference points, respectively. Aliso Canyon is surrounded by the Transverse Range and urban areas. There are both CO 2 and CH 4 emission sources in the urban areas; consequently, the inflow to Aliso Canyon has mixed contamination. Assuming that XCH 4 (LT)/XCO 2 (LT) of inflow to the source and reference from urban areas have common values, the above equation can extract enhancement from the CH 4 point source. Figure A2 shows a reduction of 20 ppb in the positive bias in the case of background XCO 2 (LT) of 400 ppm. Considering the uncertainty of 13 ppb in XCH 4 (LT) and 2.1 ppm in XCO 2 (LT) in the source and reference points, and assuming that both are random, the uncertainty level in enhancement is 0.056 ppb/ppm ( Figure 4). This approach is effective for removing uncertainty due to the inflow and monitoring the trend, but we require wind information for quantitative discussion.

Appendix A.3. Filtering with Wind Direction
The sites where wind speed and direction are monitored regularly are limited. The data were available only at Burbank Airport, which is 22 km from the emissions source. Thus, we instead used the WRF model to simulate the actual atmospheric conditions. Figure A3 shows the XCH 4 (LT)/XCO 2 (LT) ratio and the WRF wind direction at the leak point (34.31 • N, 118.56 • W) during GOSAT overpass (20:50 UTC), at a height of 906.9 hPa. We filtered out data with a wind direction of East-South-Northwest (90-315 • ) using the WRF wind direction. Figure A3. XCH 4 (LT)/XCO 2 (LT) enhancement with time between the blowout and leak control (circles) as well as the WRF wind direction (crosses).