Potential of next-generation imaging spectrometers to detect and quantify methane point sources from space

We examine the potential for global detection of methane plumes from individual point sources with the new generation of spaceborne imaging spectrometers (EnMAP, PRISMA, EMIT, SBG, CHIME) scheduled for launch in 2019–2025. These instruments are designed to map the Earth’s surface at high spatial resolution (30m× 30m) and have a spectral resolution of 7–10 nm in the 2200– 2400 nm band that should also allow useful detection of atmospheric methane. We simulate scenes viewed by EnMAP (10 nm spectral resolution, 180 signal-to-noise ratio) using the EnMAP end-to-end simulation tool with superimposed methane plumes generated by large-eddy simulations. We retrieve atmospheric methane and surface reflectivity for these scenes using the IMAP-DOAS optimal estimation algorithm. We find an EnMAP precision of 3 %–7 % for atmospheric methane depending on surface type. This allows effective single-pass detection of methane point sources as small as 100 kg h−1 depending on surface brightness, surface homogeneity, and wind speed. Successful retrievals over very heterogeneous surfaces such as an urban mosaic require finer spectral resolution. We tested the EnMAP capability with actual plume observations over oil/gas fields in California from the Airborne Visible/Infrared Imaging Spectrometer – Next Generation (AVIRIS-NG) sensor (3m×3m pixel resolution, 5 nm spectral resolution, SNR 200–400), by spectrally and spatially downsampling the AVIRIS-NG data to match EnMAP instrument specifications. Results confirm that EnMAP can successfully detect point sources of ∼ 100 kg h−1 over bright surfaces. Source rates inferred with a generic integrated mass enhancement (IME) algorithm were lower for EnMAP than for AVIRIS-NG. Better agreement may be achieved with a more customized IME algorithm. Our results suggest that imaging spectrometers in space could play an important role in the future for quantifying methane emissions from point sources worldwide.

Methane is a powerful greenhouse gas, yet its sources are highly uncertain. Quantifying methane emissions 35 from different sources is critical for developing strategies to reduce atmospheric methane levels. Anthropogenic 36 emissions originate from a large number of point sources (coal mine vents, oil/gas facilities, livestock operations, 37 landfills, wastewater treatment plants) that are individually small, spatially clustered, often intermittent, and difficult to 38 quantify (Allen et al., 2013;Frankenberg et al., 2016). Here we investigate the unique potential of new-generation 39 satellite instruments designed to map the Earth's surface (imaging spectrometers) to also detect methane point sources 40 in the shortwave infrared (SWIR). 41 There has been considerable interest in using SWIR satellite observations of atmospheric methane columns by 42 solar backscatter to detect methane sources and test emission inventories (Jacob et al., 2016). These observations are 43 traditionally made by atmospheric sensors with high spectral resolution (<1 nm) to capture the fine structure of 44 methane absorption lines (Table 1). The requirement of high spectral resolution has generally implied a coarse pixel 45 resolution (>1 km) to achieve satisfactory signal-to-noise ratios (SNR), but this limits the ability to localize and 46 quantify individual point sources. Inverse analyses of observations from the SCIAMACHY instrument with 60 km 47 pixel resolution, and from the GOSAT instrument with sparse sampling at 10 km pixel resolution, have quantified 48 emissions over regional scales (Bergamaschi et al., 2009;Kort et al., 2014;Turner et al., 2015). The recently launched 49 TROPOMI instrument with global daily coverage at 7 km pixel resolution (Hu et al., 2018) will refine the regional 50 characterization but still cannot resolve point sources . Planned instruments with ~1 km pixel 51 resolution (MethaneSat, Propp et al., 2017;Geo-FTS, Xi et al., 2016) should be able to detect large point sources after 52 inversion of several days of observations (Cusworth et al., 2018;Turner et al., 2018) but would not resolve densely 53 clustered or temporally variable sources. 54 Figure 2 shows a simulated red-blue-green (RGB) EeteS image over Berlin. We consider four scenes within 117 this domain to add WRF-LES methane plumes and perform subsequent retrievals. The scenes -Grass, Dark (water), 118 Bright, Urban -have mean SWIR surface reflectances of 0.09, 0.02, 0.30, 0.13, respectively. The urban scene is highly 119 heterogeneous. The WRF-LES simulation is conducted with 30 ´ 30 m 2 resolution (the EnMAP pixel resolution), 100 120 W m -2 sensible heat flux (moderately unstable meteorological conditions), and a mean wind speed of 3.5 m s -1 . We 121 generate an ensemble of 15 instantaneous plumes by sampling the WRF-LES simulation at five time slices and for 122 three source rates of 100, 500, and 900 kg h -1 . This range is typical of large (but not unusually large) point sources 123 (Jacob et al., 2016). 124 We compute the optical depth of the methane plume τ(λ) at wavelength λ by multiplying HITRAN absorption 125 cross sections (sH; Kochanov et al., 2016) by the methane mixing ratio enhancement (ΔVMR) and density of dry air 126 (VCD) in the 72-layered atmosphere of the MERRA-2 meteorological reanalysis (Gelaro et al., 2017): 127 τ(λ) = ' Δ , , s /,, (λ). 23 ,45 (1) 128

129
The plume transmission T(λ) is the negative exponential of τ(λ) weighted by the geometric airmass factor A (AMF) for 130 the backscattered solar radiation: 131 132 (λ) = exp{− τ(λ) }. (2)  133 134 Each pixel's EeteS radiance spectrum is multiplied by this additional plume transmission. We do not add noise or 135 aerosol effects to the plume transmission spectra because the EeteS scene already accounts for those in the computation 136 of back-scattered radiances, and the plume transmission is just a multiplicative factor on these back-scattered radiances. 137 Figure 3 shows an example WRF-LES plume (500 kg h -1 source rate) superimposed over the Grass and Urban scenes. 138 EnMAP has a specific spectral resolution and SNR. We examined the sensitivity of the retrieval to these 139 parameters by generating synthetic spectra for different spectral resolutions and SNRs, thus extending our analysis to 140 other new-generation imaging spectrometers (Table 1). For this purpose, we interpolated EeteS surface radiance spectra 141 to the desired spectral resolution assuming no instrument noise. We then multiplied these radiance spectra by the 142 standard atmosphere plus WRF-LES plume transmission spectra and added uncorrelated instrument noise as per the 143 specified SNR. 144 https://doi.org /10.5194/amt-2019-202 Preprint.  To test our EnMAP retrievals on actual data, we also downsampled AVIRIS-NG images taken from aircraft 145 over California (CARB, 2017) to match EnMAP spatial resolution, and further convolved these spectra with a 10 nm 146 Gaussian filter to match EnMAP spectral resolution and wavelength positions. AVIRIS-NG flew at 3-4 km above the 147 ground, so we simulated additional extinction at higher altitudes based on the U.S standard atmosphere (Kneizys et al., 148 1996). We compared the retrieved methane from AVIRIS-NG and the synthetic EnMAP to determine the ability of 149 EnMAP to detect and quantify the methane point sources identified by AVIRIS-NG. 150 151

Methane retrieval 152
We retrieved methane from the synthetic imaging spectrometer spectra by adapting the Iterative Maximum A 153 Posteriori -Differential Optical Absorption Spectroscopy (IMAP-DOAS) algorithm developed for AVIRIS 154 (Frankenberg et al., 2005b;Thorpe et al., 2017;Ayasse et al., 2018). DOAS retrievals isolate higher frequency features 155 resulting from gas absorption from lower frequency features that include surface reflectance as well as Rayleigh and 156 Mie scattering (Bovensmann et al., 2011). A polynomial term accounts for the low frequency features (Thorpe et al., 157 2017 In addition to methane (CH4), the retrieval must account for variable absorption by water vapor (H2O) and 161 nitrous oxide (N2O) over the 2210-2400 nm spectral region. We parameterize low frequency spectroscopic features as a 162 sum of Legendre polynomials of order k = [0, K] with coefficients ak. The state vector (x) optimized through the 163 retrieval is therefore composed of the following elements: 164 where s is a scaling factor applied to the column mixing ratio of each gas from the U.S standard atmosphere (Kneizys et 166 al., 1996). We do not include aerosols in the retrieval as they play little role at the relevant spatial and spectral 167 resolution (Ayasse et al., 2018). Methane point sources generally do not co-emit aerosols. 168 169

Optimal estimation 170
To retrieve the state vector from the Eetes TOA radiances, we use a forward model similar to previous IMAP-171 DOAS algorithms (Thorpe et al., 2017, Ayasse et al., 2018, with a modification to the polynomial term for surface Here F h is the high-resolution backscattered TOA radiance at wavelength l, I0(λ) is the incident TOA solar intensity, 175 τn,l is the default optical depth from the US standard atmosphere for trace gas element n = [1,3] of the state vector at 176 vertical level l = [1,72], sn is the scaling factor to that default optical depth optimized in the retrieval, Pk(λ) is the k th 177 Legendre polynomial, and the ak are coefficients optimized in the retrieval. The optical depth τn,l is computed in the 178 same fashion as Equation 1, using information from the MERRA-2 reanalysis and HITRAN absorption cross sections. 179 For satellite retrievals, the AMF is a scalar describing the optical path through the atmosphere. In Section 4.3, we apply 180 the IMAP-DOAS algorithm to airborne AVIRIS-NG scenes and use a vector-valued AMF that depends on the height of 181 the aircraft. 182 Previous IMAP-DOAS algorithms used a simple polynomial approximation for the surface reflectance, but 183 here we use Legendre polynomials to exploit their orthogonality. We find that K = 4 provides sufficient spectral 184 resolution whereas previous applications using simple polynomials required K = 6 (Ayasse et al., 2018). 185 We compute the TOA backscattered radiances K ( , λ) over the 2210-2410 nm spectral range at 0.02 nm 186 resolution, and assemble these in a vector F h (x) representing the high-resolution spectrum as simulated by the forward 187 model for a given x. We convolve this spectrum with the instrument FWHM and then sample at the known wavelength 188 positions. For example, for EnMAP, we convolve ( ) with a 10 nm FWHM and sample the resulting spectra at 189 EnMAP's 10 nm intervals to get the low-resolution F(x). We also explored performing separate convolutions on the 190 high resolution transmission and polynomial terms in Equation 3, and then multiplying them together to get F(x). We 191 found little difference in the results between methods. 192 Observed backscattered TOA radiances (y) can be represented as 193 where the observational error is the sum of instrument and forward model errors. As is commonly done for satellite 195 retrievals, we assume that the forward model error is small compared to the instrument error characterized by the SNR. 196 The forward model is non-linear so that the solution must be obtained iteratively. A Jacobian matrix is calculated for 197 each iteration of the state vector 198 199 and we employ a Gauss-Newton iteration to solve iteratively for the optimal state vector (Rodgers, 2000): 200 Here SO = [εε T ] is the observation error covariance matrix defined by the instrument SNR, xA is the prior estimate of the 202 state vector, and SA is the prior error covariance matrix. We set a weak prior error variance for methane, m nop 3 = 5, to 203 accommodate large plume enhancements. The prior XCH4 estimate is 1800 ppb. The iterative analytical solution to the 204 inverse problem as described by equation (6) also provides the posterior error covariance matrix ( q ) as part of the 205 solution: 206 (7)  207 208 q gives information on the error correlation between retrieved methane and surface reflectivity, which is a major 209 concern for methane retrievals (Butz et al., 2012). 210 the Grass and Urban scenes. Near the emission source, the 500 kg h -1 plume is clearly defined in the Grass scene. It is 215 also detectable in the Urban scene but obscured by surface retrieval artifacts. The 900 kg h -1 plume is better captured 216 over both surfaces, though major retrieval artifacts remain in the Urban scene. 217 Varon et al. (2018a) previously estimated the theoretical ability of a satellite instrument to quantify source 218 rates from point sources as a function of instrument precision, assuming a uniform surface reflectance. They concluded 219 that an instrument with 1-5% precision on XCH4 would be able to quantify point sources with an error of 70-170 kg h -1 . 220 Here we characterize the EnMAP instrument precision as the relative root-mean squared-error (RRMSE) between the 221 true and retrieved column methane concentrations for individual 30 ´ 30 m 2 pixels in the scenes of Figure 2 including 222 the WRF-LES plumes. Figure 4 summarizes the results for the four scenes of Figure 2. We find precisions of 8.2 ± 223 0.7% for Grass, 13 ± 0.7% for Urban, and 3.7 ± 0.5% for Bright scenes. The standard deviations refer to the RRMSEs 224 computed for the 15 different realizations of the WRF-LES plumes and for the 3 source rates of 100, 500, and 900 kg h -225 reflectance over the scene including dark pixels. As illustrated in Figure 3, the 8% precision over the relatively uniform 229 grass surface should enable EnMAP to successfully quantify 500 kg h -1 point sources in a single pass. 230 Beyond the precision for the methane retrieval, an additional limitation for retrieving point sources is the error 231 correlation with variable surface reflectance. This is illustrated in Figure 3 with the retrieved XCH4 enhancements over 232 Grass and Urban scenes relative to the background. In the case of the Grass scene with a 500 kg h -1 source, the 8% 233 precision limits the ability to observe the downwind plume but there is a clear enhancement over background at the 234 source location. With a 900 kg h -1 source the downwind plume becomes well-defined against the background. In the 235 case of the Urban scene, the detection of the 500 kg h -1 plume is far more problematic because of large positive artifacts 236 over dark (water) pixels. The 900 kg h -1 plume is still difficult to distinguish from the artifacts and would require prior 237 knowledge of source location to be identified and quantified. The error correlation between methane and surface 238 reflectance in the retrieval can be reduced by increasing the spectral resolution of the instrument as discussed in Section 239 4.2. 240 241

Sensitivity to instrument spectral resolution and SNR 242
Here we examine the potential of future instruments with improved spectral resolution and SNR relative to 243 EnMAP (Table 1) to achieve improved retrievals of point sources. Figure 5 shows the change in retrieval precision as 244 we vary the spectral resolution from 10 to 1 nm and the SNR from 100 to 500. The precision estimates are calculated 245 using two methods. First, we estimate the precision by evaluating the RRMSEs averaged over the Grass, Urban, and 246 Bright scenes of Figure 2, for 3 source rates and 15 instantaneous plume realizations, following the procedure of 247 Section 4.1. Since SNR varies on a per-pixel basis, the plotted SNRs for this method represent the mean scene SNR. 248 Specifications of the instruments in Table 1 are identified on the plot. Precision improves as spectral resolution and 249 SNR increase, as expected. The dependencies are not linear, and the contours are concave, meaning that precision is 250 more effectively improved by increasing spectral resolution by a certain factor than by increasing SNR by the same 251 factor. Increasing the spectral resolution improves precision through multiple independent factors: by increasing the 252 number of independent measurements across the methane interval; by increasing the effective squared depth of the 253 sharpest methane absorptions, for improved spectral contrast relative to the continuum; and by better resolution of the 254 unique methane absorption shape, which improves discrimination against potential surface confusers. 255 Second, we estimate theoretical precision in Figure 5  precision method, doing multiple along-track samples improves the theoretical precision from 5% to 1%. Varon et al. 262 (2018a) found that an instrument with 5% precision could constrain most anthropogenic point sources above 170 kg h -1 . 263 Using both the RRMSE and theoretical precision methods of Figure 5, we find that a spaceborne AVIRIS-NG 264 instrument (spectral resolution 5 nm, SNR 200-400) would have a precision of 5.5 -1.0%, meaning that such an 265 instrument could constrain a majority of anthropogenic methane point sources. 266 A benefit of increasing spectral resolution is to improve decoupling of surface and methane spectroscopic 267 features. We saw in Figure 3 that this was a major source of error over inhomogeneous surfaces such as the Urban 268 scene. It is manifested in the retrieval by an error correlation between state vector elements sCH4 (scaling factor for 269 methane column mixing ratios) and ak (coefficients for the surface reflectivity described by Legendre polynomials). 270 This error correlation is described by the posterior error covariance matrix q obtained as part of the retrieval (Equation 271   6). For example, the error correlation decreases significantly between EnMAP (r = -0.33) and AMPS (r = -0.19). This 272 driven by the increase in spectral resolution from 10 nm to 1 nm. A separate test shows that simply increasing the SNR 273 to 300 (as for SBG) does not improve the error correlation. 274 An important implication of decoupling XCH4 from the surface reflectance in the retrieval is to improve the 275 capability for plume pattern recognition, which is necessary to convert observed plume methane enhancements into 276 source rates (Varon et al., 2018a). Figure 6 illustrates this for the Grass and Urban scenes of Figure 3 including the 277 plume from the 500 kg h -1 point source. Following Varon et al. (2018a), we define the plume for the retrieved scenes 278 with a plume mask that applies median and Gaussian filters to pixels above the 80 th percentile of XCH4 within the scene. 279 Retrievals are performed with the specifications of the EnMAP instrument (10 nm spectral resolution, SNR 180), SBG 280 (10 nm, 300), and AMPS (1 nm, 400). 281 For the Grass scene we find that all three instruments can discern the plume pattern near the emission source 282 and separate it from surface features. SBG and AMPS capture larger plume domains because of their higher precisions 283 ( Figure 5), but a source rate can still be estimated successfully with EnMAP by taking into account the dependence of 284 the retrieved plume extent on instrument precision (Varon et al., 2018a). For the Urban scene, EnMAP plume detection 285 is swamped by surface artifacts. Simply increasing the SNR as in the SBG instrument does not improve the situation.
Increasing the spectral resolution to 1 nm as in the AMPS instrument enables detection of the plume though 287 quantification would still be prone to surface artifacts. 288 289

Evaluation with AVIRIS-NG observations 290
To test the EnMAP retrieval capability with actual observations, we downsampled AVIRIS-NG spectra taken 291 over California methane emitting facilities (CARB, 2017). We chose three scenes observed by AVIRIS-NG on 292 different days over oil and gas facilities. Figure 7 shows the RGB images, the AVIRIS-NG plume retrievals performed 293 by applying the method of Section 3 with a variable AMF, and the downsampled EnMAP retrievals. Plume masks 294 have been applied in the same way as for Figure 6. At the altitudes used for the California survey, AVIRIS-NG has 3´3 295 m 2 pixel resolution and hence features much sharper methane enhancements than EnMAP (note the different scales for 296 the middle and right panels). 297 We see from Figure 7 that EnMAP is able to detect the same plumes as AVIRIS-NG (two plumes in the 298 bottom panels). This is facilitated by the brightness of the surfaces. The surface reflectivities retrieved simultaneously 299 with the methane enhancements in our IMAP-DOAS algorithm are 0.39-0.49, brighter than the Bright EeteS scene in 300 Section 4.1. 301 The plume observations can be related to the corresponding source rates by computing the integrated mass 302 enhancements (IME) within the plume mask (Frankenberg et al., 2016;Varon et al., 2018a). The IME is calculated as: 303 where ΔΩ , is the plume mass enhancement in pixel i relative to background (kg m -2 ), Λ , is the corresponding area of 305 the pixel, and the summation is over the N pixels within the plume mask. The point source rate Q is then inferred from 306 the IME as (Varon et al., 2018a) ,45 , is a characteristic plume size and ~•• is an effective wind speed describing the rate of turbulent 309 dissipation of the plume (L/Ueff is the lifetime of the plume against turbulent dissipation to below the detection limit). 310 Varon et al. (2018a) relate Ueff to the 10-m wind speed (U10) by fitting to WRF-LES simulations. Here we use their 311 relationship derived for the a 50 m pixel resolution, 5% precision instrument (Varon et al., 2018), and apply it as a 312 rough approximation to the AVIRIS-NG and downsampled EnMAP plumes: 313 where Ueff and U10 are in units of [m s -1 ]. We obtain U10 from the HRRR-Reanalysis at 3-km hourly resolution 315 (https://rapidrefresh.noaa.gov/). 316 Figure 7 shows the source rates inferred from the AVIRIS-NG and EnMAP retrievals for each point source. 317 The AVIRIS-NG source rates are a factor of 1.2-3.0 greater (average 1.9) than the EnMAP source rates. There could be 318 several factors behind this discrepancy including error correlation with surface reflectivity in the EnMAP retrieval that 319 would cause some loss of the plume, and use of a generic plume mask and IME algorithm for both instruments. As 320 pointed out by Varon et al. (2018a), the U10-Ueff relationship needs to be tailored to the pixel resolution and precision of 321 the particular instrument, and to the choice of plume mask. Nevertheless, the results do confirm that EnMAP should be 322 able to detect plumes and quantify source rates down to ~100 kg h -1 when the scene is sufficiently bright. 323 324

Conclusions 325
We examined the potential of next-generation spaceborne imaging spectrometers (EnMAP, PRISMA, EMIT, 326 SBG,) for observing atmospheric methane plumes from point sources and inferring the corresponding source rates. 327 These instruments have launch dates of 2019-2025 and focus on observing the Earth surface with fine pixel resolution 328 (30 ´ 30 m 2 ), but they also have observing channels at 2200-2400 nm with 7-10 nm spectral resolution that could be 329 used to retrieve methane plumes. This would achieve much finer spatial resolution than the standard satellite 330 instruments designed to measure atmospheric methane, and would provide a unique resource for global mapping of 331 individual methane point sources. 332 We focused our baseline analysis on EnMAP (spectral resolution 10 nm, SNR 180, 2020 launch date) as its 333 specifications are well documented (Guanter et al, 2015). We created synthetic spectra using the EnMAP End-to-End 334 Simulation Tool (EeteS) to simulate various surface scenes (Grass, Urban, Bright) with instrument errors and with 335 superimposed methane plumes generated by a WRF Large Eddy Simulation (LES). We then retrieved these scenes for 336 atmospheric methane together with surface reflectivities using the Iterative Maximum A Posteriori -Differential 337 Optical Absorption Spectroscopy (IMAP-DOAS) approach. The resulting precisions for methane are 8% for the Grass 338 scene, 13% for Urban, and 4% for Bright. A 500 kg h -1 methane plume (typical of very large point sources) is readily 339 detected over the relatively homogeneous Grass surface. The highly heterogeneous Urban surface is much more 340 challenging because of retrieval artifacts. 341 The limitation of EnMAP in detecting methane plumes over heterogeneous surfaces is caused by error 342 correlation between methane and surface reflectivity in the retrieval. We examined how precision and error correlation 343 https://doi.org/10.5194/amt-2019-202 Preprint. Discussion started: 29 May 2019 c Author(s) 2019. CC BY 4.0 License. could be improved by increasing spectral resolution and SNR. We find that spectral resolution reduces error 344 correlation more important than SNR. The proposed Atmospheric Methane Plume Spectrometer (AMPS), which 345 bridges the gap between imaging spectrometers and atmospheric sensors (1 nm spectral resolution, SNR 400), can 346 greatly decrease surface artifacts and detect a 500 kg h -1 plume even over the heterogeneous Urban surface. Alternative 347 surface parameterizations might also improve XCH4 and surface separation. For example, a channelwise representation 348 with reflectances tied through an empirical covariance structure (Thompson et al., 2018) has been used previously to 349 improve consistency in water vapor estimations. Alternative algorithms, such as matched filter approaches (Ong et al.,350 2019) may show different XCH4 sensitivities, and in particular may be better able to represent structured reflectances of 351 more complex surfaces. 352 We tested the EnMAP capability with actual observations by downsampling AVIRIS-NG images taken from 353 aircraft (3 ´ 3 m 2 pixels, 5 nm spectral resolution, SNR 200) over California methane emitting facilities (CARB, 2017). 354 We showed that these EnMAP-like images are able to detect and quantify actual plumes of magnitude ~100 kg h -1 over 355 relatively bright surfaces. Source rates inferred from the plumes with a generic Integrated Mass Enhancement (IME) 356 method are a factor of 1.2 to 3 lower for EnMAP than for AVIRIS-NG, which could be due in part to unaccounted 357 dependence of the IME method on instrument pixel size and precision. 358 In summary, our analysis shows that future spaceborne imaging spectrometers designed to map the Earth 359 surface in the SWIR also have considerable potential for detecting methane plumes from point sources and quantifying 360 source rates. The detection capability of 100-500 kg h -1 over relatively bright or homogeneous land surfaces would 361 allow accounting for a wide range of point sources. The fine spatial resolution of these instruments should make them a 362 unique resource to contribute to tiered observing systems for greenhouse gases (Duren and Miller, 2012 b Spectral resolution is represented by the full-width at half-maximum (FWHM). 5 c For SCIAMACHY and GOSAT, SNR is for CO2 band used in the CO2-proxy method retrieval. For other instruments, SNR is at 2300 nm. d Airborne Visible/Infrared Imaging Spectrometer -Next Generation (Thorpe et al., 2017). AVIRIS-NG provides roughly a ground sampling distance (GSD) of 1 m per km altitude. The Frankenberg et al. (2016) and Duren et al. (2019) campaigns operated at 3-4 km altitude. 10 e Along-track oversampling increases SNR by √ where N = number of along-track frames. AVIRIS-NG routinely achieves N > 4 so AVIRIS-NG effective SNR at 2300 nm can be as much as 400.
h GreenHouse Gases Satellite (McKeever et al., 2017). Revisit times are for selected 12 ´ 12 km 2 scenes. The demonstration GHGSat-D instrument presently in space has additional instrument imperfections that limit its precision to 13% (McKeever et al. 2017). 5 i TROPOspheric Monitoring Instrument (Hu et al., 2018) j Airborne Methane Plume Spectrometer (Thorpe et al., 2016) k PRecursore IperSpettrale della Missione Applicativa (http://prisma-i.it) l Environmental Mapping and Analysis Program (Guanter et al., 2015) m Earth Surface Mineral Dust Source Investigation (Green et al., 2018) 10 n Surface Biology and Geology, previously called HyspIRI (Hochberg et al., 2015) https://doi.org /10.5194/amt-2019-202 Preprint. Discussion started: 29 May 2019 c Author(s) 2019. CC BY 4.0 License.   Retrieval of a methane plume over grass and urban EnMAP scenes. The plume was generated by WRF-LES with a source rate of either 500 kg h -1 or 900 kg h -1 . The left panels show the dry air column mixing ratio enhancements relative to the 1800 ppb background for a 500 kg h -1 methane plume superimposed on the RGB images of Figure 2. The middle panels 5 show the retrieval of those enhancements using the IMAP-DOAS retrieval algorithm applied to the EnMAP instrument specifications. The right panels show the retrieval of the 900 kg h -1 plume. The XCH4 enhancements in the right panels are scaled by 5/9 to be comparable with the other panels. Negative enhancements are reset to equal the background.   (Table 1) over different surfaces. The precisions are the relative root-mean squared errors (RRMSE) between the "true" methane columns in synthetic scenes and values obtained from the IMAP-DOAS retrieval applied to the EnMAP top-of-atmosphere (TOA) backscattered radiances. 5 The error bars represent the standard deviation over 15 WRF-LES plume realizations and 3 source magnitudes for the plume (100, 500, 900 kg h -1 ).
Precision as function of surface type G r a s s U r b a n B r i g h t  SBG (1) SBG (3) AMPS (1) AMPS (2) Precision of methane retrievals for imaging spectrometers SBG (1) SBG (3) AMPS (1) AMPS (2) AVIRIS-NG (1) (in space) https://doi.org/10.5194/amt-2019-202 Preprint. Discussion started: 29 May 2019 c Author(s) 2019. CC BY 4.0 License. Figure 6. Plume pattern recognition applied to a point source of 500 kg h -1 over Grass and Urban scenes as shown in Figure   3. The plume pattern is defined by applying median and Gaussian filters to pixels above the 80 th percentile of XCH4 in the scene. Areas excluded by the mask are shown in gray. The panels show retrievals from the EnMAP, SBG, and AMPS 5 instruments.