the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 22 May 2026
GESat GEN1: In-Orbit Performance Review of a Compact SWIR Fizeau Interferometer for Facility-scale Methane Detection and Quantification
Abstract. The GESat GEN1 mission is a 16U microsatellite developed and operated by Absolut Sensing to demonstrate high-resolution methane monitoring using a compact SWIR Fizeau interferometer. GEN1 provides a 50m ground sampling distance to monitor facility-scale methane plumes. It is rated for an average detection rate of 500 kg.h−1, with the capacity to resolve leaks as small as 100 kg.h−1 in pristine environmental conditions. The system is designed to achieve a column retrieval precision of 120 ppbv, defined as the mean standard deviation of the retrieved column-averaged dry-air mole fraction of methane over operational conditions. This corresponds to a random error of approximately 6.5% relative to a background
concentration of 1850 ppbv.
The paper first presents the GEN1 mission and its interferometric payload, designed to enable selective methane sensing from a highly compact platform. It then describes the retrieval architecture, based on a physics-guided parametric algorithm that occupies an intermediate complexity between matched-filter methods, as used for missions such as Tanager-1 and EMIT (Carbon Mapper, 2024), and full physics schemes using high accuracy radiative transfer code such as libRadtran or 4A/OP (Mayer and Kylling, 2005; Scott and Chédin, 1981; Chéruy et al., 1995). The proposed model generates top-of-atmosphere radiances in the SWIR [1550, 1700nm] region with no systematic bias relative to reference full-physics model and a random error below 0.2 ppbv in methane column retrieval. Moreover, on identical computing hardware, the physics-guided parametric approach achieves a processing speed approximately 20×103 times faster than the full-physics model. This strategy therefore provides a balanced trade-off between robustness, accuracy, and computational cost, enabling large-scale processing compatible with constellation-level operations while preserving radiometric fidelity required for quantitative methane retrievals. Supported by this instrument and retrieval design, representative in-orbit results are reported, including methane plume detections over industrial facilities, together with an assessment of current limitations and planned evolutions of the processing chain. Finally, the paper concludes by evaluating the mission’s performance against its design specifications.With more than 160 acquisitions processed, the system achieves an average precision of 111 ppbv, successfully meeting the mission’s primary target, with an average detection sensitivity of 450 kg.h−1. Notably, several measurements achieved precision better than 80 ppbv, which enables an ultimate methane plume detection limit corresponding to emission rates below 100 kg.h−1 under ideal meteorological conditions. This on-orbit demonstration confirms the viability of the Fizeau interferometric approach for high-resolution monitoring and paves the way for future constellation deployment.
- Preprint
(13248 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 08 Oct 2026)
- RC1: 'Comment on egusphere-2026-1269', Anonymous Referee #1, 21 Aug 2026 reply
-
RC2: 'Comment on egusphere-2026-1269', Anonymous Referee #2, 21 Sep 2026
reply
The authors present the GESat instrument designed for methane remote sensing and provide an analysis of their retrieval and validation program, intended to show that the instrument and retrieval system produce precision within their requirements. It is certainly an interesting result and exciting to see increase satellite remote sensing capabilities of methane. Therefore, this manuscript fits firmly into the scope of AMT. I do have several reservations of the analysis as presented and I outline these concerns below, which I ask the authors to address.
1. General flow - needs some additional editing as much of the manuscript reads as more stream-of-consciousness than academic writings. For example, many paragraphs are 1-2 sentences that read as notes or considerations instead of core analyses, methods, or conclusions. The manuscript would read and flow much better if the authors condensed to what was necessary and started each paragraph with a topic sentence. It's also not clear how the arguments from Section 3 actually apply to the retrieval approach described in Section 4.
2. Equation 1 is an overly optimistic estimation of plume detection performance. A probabilistic estimate of detection limit can be better estimate via assessment of global distributions of detected sources. E.g., Jervis et al. (2026)[https://doi.org/10.1021/acs.est.5c17089]. This study finds a several factors difference between the MDL estimated via Equation 1 in your paper and a probabilistic detection limit (e.g., 50% POD).3. Line 69. I don't think referring to 0.5 m/s wind speed as "pristine" is correct. Yes, it is advantageous for Equation 11, but that is not an optimal observing condition for actually quantifying plumes. Also, this highlights another general problem with Equation 1 - that Q_min goes to zero as U goes to zero, which is clearly not physical.
4. Labels and legend on Figure 2 are hard to read.
5. Line 169. Matched filtering seems not as applicable to this study, so why bring up? Did you do any tests with matched filters? If so, what were there results? If not, I don't know if this paragraph adds much beyond what is already known.
6. Section 3.3.1. Did you do any comparative tests with polynomials? Just curious if your intuition is verified in practice through application. Also, not clear how you apply your PCA decomposition. You perform a PCA analysis on the entire ASTER/ECOSTRESS library, then try and match that to the spectrum in question? Or something else? The r_k coefficients are retrieved?
7. Related to comment #5, maybe you can make a table with all the parameters/coefficients you retrieve as part of your retrieval for easy reference? Maybe in section 3.4?
8. Figures 4 and 5 could be collapsed into a single figure.
9. Other trace gas retrieval algorithms that employ optimal estimation perform iteration as they problem is non-linear (i.e., the K matrix needs to be recomputed for iterative updates to a prior). Here, it appears you are assuming that your problem is linear and/or that a single iteration starting from your prior is sufficient. Do you have evidence that this assumption works?
10. Equation 16. You only have a 2 parameter state vector? Equation 13 suggests there are additional parameters retrieved. Again, going back to my comment #6 - please make clear what it retrieved.
11. Section 4.3 What were the conditions of these 163 global scenes? Could you summarize the SZA, albedo, etc?
12. Figure 14. It does look fairly clear that there is a systematic high bias in the retrieval, which you do discuss. However, if you were to remove that mean bias compared to TROPOMI or TCCON, and then compared to TCCON/TROPOMI, what would the standard deviation of the difference between the bias-corrected GESAT vs TROPOMI/TCCON look like? That is actually closer to retrieval precision (or is a metric more commonly adopted for trace gas sounder retrieval) than quantifying background standard deviation.
Citation: https://doi.org/10.5194/egusphere-2026-1269-RC2 -
RC3: 'Comment on egusphere-2026-1269', Anonymous Referee #3, 22 Sep 2026
reply
General comments:
This manuscript presents the instrument concept, retrieval methodology, and initial in-orbit performance of GESat GEN1, a compact SWIR Fizeau interferometer deployed on a 16U satellite with a 50 m ground sampling distance. The combination of interferometric measurements and a fast parametric retrieval is relevant to facility-scale methane monitoring. The reconstructed interferograms, repeated plume observations, and preliminary comparisons with independent column measurements provide useful evidence of the mission’s capabilities. Reporting the observed positive column bias is also valuable for assessing current limitations.
However, several central performance claims require stronger support. Emission detection thresholds are inferred from column precision, while emission-rate quantification is not demonstrated. The reported precision statistic is inconsistent between the text and Fig. 10, and the relationship between the validated forward model and the operational retrieval remains unclear. Additional information is needed on retrieval sensitivity, calibration errors, and external validation. I therefore recommend major revision.
Main concerns:
1. The abstract and conclusions present detection thresholds of approximately 450 kg h-1 under nominal conditions and 100 kg h⁻¹ under favorable conditions. These values follow from Eq. (1) and the reported column precision. Such calculations are useful estimates under specified assumptions, as discussed by Worden et al. (2025), but they do not establish an empirical probability of plume detection. In particular, the detectability factor q = 2 does not characterize the false-positive and false-negative rates of the complete processing chain.
Please describe the operational plume detection procedure and clearly distinguish estimated thresholds from demonstrated detection performance. If empirical threshold claims are retained, support them with suitable reference observations, preferably controlled releases. Realistic plume injections into measured backgrounds could provide complementary evidence, provided the simulation assumptions and limits of that validation are stated. The 0.5 m s-1 wind assumption should be identified as a specific favorable case for enhancement detection, with discussion of the associated transport uncertainty and its implications for emission-rate estimation.
The manuscript also promises emission-rate quantification but provides no flux-inversion method or source-specific estimates. To support this claim, quantify representative plumes and document background estimation, plume delineation, wind inputs, the transport formulation, and propagated uncertainties. Independent comparisons would strengthen these results.
2. Figure 10 on p. 18 labels 111 ppbv as the median scene standard deviation, whereas the text describes this value as an average and the abstract defines the 120 ppbv requirement in terms of the mean standard deviation. Given the skewed distribution, the distinction may materially affect the assessment of mission performance. Please report the arithmetic mean, median, quartiles, and fraction of acquisitions meeting the requirement, and evaluate compliance using the originally stated metric.
Background spatial variability is a useful performance measure, but it may include random noise, spatially structured retrieval errors, and genuine atmospheric variability. Describe scene selection, numerical reflectance thresholds, cloud and plume screening, and retained pixel fractions. Explain the origin and use of the assumption that 99% of retained pixels represent background. Quality filtering is reasonable, but the reported performance should be explicitly tied to the resulting valid-pixel population.
Assess sensitivity to the filtering choices and summarize performance across scene brightness and surface conditions. A representative comparison of background methane and radiance maps, together with a spatial-correlation diagnostic, would help determine whether coherent surface-related artifacts contribute to the reported standard deviation.
3. Sections 3.3 and 3.4 need to be connected more explicitly. Section 3.3 introduces PCA reflectance coefficients and several scattering and aerosol parameters, whereas Sect. 3.4 retains only methane and water-vapor profile scalings after normalization. Identify which terms are fixed, fitted beforehand, retrieved, or omitted in operational processing. Clarify whether the libRadtran comparison evaluates precisely this operational configuration. Normalization can remove a common amplitude factor, but residual surface spectral structure and scattering-induced changes in photon path length may still affect methane retrievals.
The interpretation of Eqs. (10) and (11) also requires clarification. In-scattering contributes radiance through the source term of the radiative transfer equation; an exponential extinction law alone does not establish a multiplicative representation of this contribution. Equation (11), which scales molecular absorption optical depth, likewise needs a physical justification or an explicit description as an empirical correction.
The single validation scenario in Sect. 3.3.6 is insufficient to establish a generally applicable methane modeling error below 0.2 ppbv. State whether any coefficients were fitted to the reference spectrum used for evaluation. Test the operational model on independent conditions representative of the observed geometry, surface reflectance, atmospheric state, and methane enhancements. Report methane retrieval errors after instrument-response integration and normalization, and explain how the spectral radiance residuals propagate into the quoted column error. Clarify whether 0.1 nm denotes internal calculation spacing or output sampling and demonstrate sufficient spectral convergence.
4. The optimal-estimation formulation lacks several inputs needed for reproducibility. Specify the methane and water-vapor prior profiles, their covariance, the measurement-error covariance, auxiliary meteorological data, parameter bounds, and convergence criteria. Define the normalization operator and apply it consistently to the observations, forward calculations, and Jacobians. Account for error correlations introduced by normalization and interferogram reconstruction.
Posterior uncertainty depends on the measurement errors, Jacobians, and prior covariance. Please show the contribution of the prior to the reported 60 ppbv theoretical precision, using averaging kernels, degrees of freedom for signal, or equivalent sensitivity diagnostics. Assess the response to methane enhancements concentrated near the surface, since scaling an entire atmospheric profile does not automatically establish sensitivity to a boundary-layer plume.
The approximately 5% water-vapor cross-sensitivity mentioned in Sect. 2.3 should be distinguished from the posterior CH4–H2O correlation of 0.05. These quantities are not interchangeable. Evaluate the methane retrieval error caused by realistic water-vapor profile perturbations and document the assumptions behind that calculation. Include the dependence on prior strength and atmospheric humidity.
5. The performance of this instrument depends strongly on spectral calibration and registration of successive frames. Sections 2.4 and 4.1 should provide the number of samples per interferogram, exposure time, frame rate, acquisition duration, scene footprint, and achieved registration accuracy. Explain how the spectral-response map is established and monitored in orbit, including the residual uncertainty after correction.
Changes in fringe contrast or spectral response may partly mimic the methane signal, while subpixel misregistration can introduce spectral artifacts over heterogeneous surfaces. Quantify these effects using representative observations or controlled perturbation tests, and discuss whether plume motion during acquisition is significant. Before-and-after correction statistics would provide stronger evidence than the two fitted interferograms in Fig. 8 alone.
The attribution of the difference between the theoretical 85 ppbv and observed 111 ppbv to structural residuals also requires quantitative support. Compare predicted and observed uncertainties for matched scene conditions. Any error decomposition should distinguish independent variance contributions from correlated or systematic effects; a 30% increase in standard deviation is not, by itself, an attribution to a particular error source.
6. The TCCON, COCCON, and TROPOMI comparisons are useful initial checks, but the limited number of coincidences does not yet establish a stable bias across operational conditions. Provide acquisition dates, spatial and temporal coincidence criteria, averaging footprints, product versions, quality filters, and definitions of the uncertainty bars. Account for differences in averaging kernels and prior profiles, or assess their contribution to the comparison uncertainty. For the revisit analysis, explain how genuine temporal variability in background methane affects the interpretation as instrumental stability.
The conclusion that the positive column offset does not affect plume quantification needs qualification. A spatially uniform additive offset cancels in local background subtraction, whereas multiplicative or spatially varying errors can bias the retrieved enhancement and inferred emissions. This distinction is discussed by Jervis et al. (2021). Examine whether the bias depends on signal level, surface properties, viewing geometry, or humidity, and assess the response to known enhancements before concluding that absolute-column errors are immaterial to plume quantification.
-------------------
Specific & minor points:Reconcile the 1625–1660 nm passband in Table 1 and Sect. 2.2 with the 25 nm interval stated in Sect. 3.4. Distinguish the optical passband, retrieval window, and broader 1550–1700 nm model-test interval.
The response in Eq. (3) includes conversion to DU, although Eq. (2) labels the output as electrons, and Eq. (15) appears to apply the gain again. Correct the units and gain placement, and index the response by interferogram element.
Define the radiometric convention for solar illumination and reflectance, including the Lambertian factor of 1/π. Explain how layer path lengths include solar and viewing geometry. State whether the PCA representation includes a mean spectrum.
Table 1. Express descending-node time as local solar time rather than UTC. Correct the dark-current unit, since amperes already represent charge per unit time, and specify the detector conditions associated with the stated value.
Explain the difference between 166 and 163 acquisitions, and reconcile the 30,000 DU value in Sect. 4.3 with 28,000 DU in the conclusions. Distinguish nominal, theoretical, and measured SNR values and their observing conditions.
Figure 5. The legend gives a radiance bias of −0.007%, whereas the text states zero systematic bias. Report the result consistently. Correct the reference to Fig. 4 as showing spectral residuals; the residuals are shown in Fig. 5.
Figure 12. Reconcile the figure’s median difference of 0.78% with the mean stated in the text. Define the relative-difference denominator, distinguish signed and absolute differences, and explain the reported 1σ value of 1.11%.
The reported China values imply a difference of 71 ppbv relative to TCCON (1975 − 1904), while 68 ppbv is the difference from COCCON. Reconcile the Pasadena bias of 4.6% with the 4.2% upper value in the conclusions.
Add dates, locations, scale bars, orientation, and readable concentration units to the plume examples. Figure 11 needs a color bar and clear masking information. Explain the numerical annotations in Figs. 13–14 and improve the readability of the interferogram panels.
Correct “ODP” to “OPD” and use “column-averaged dry-air mole fraction” consistently.
References:
Jervis D, McKeever J, Durak B O A, et al. The GHGSat-D imaging spectrometer[J]. Atmospheric Measurement Techniques, 2021, 14(3): 2127-2140.
Worden J, Green P, Eldering A, et al. Common practices for quantifying methane emissions from plumes detected by remote sensing[M]. US Department of Commerce, National Institute of Standards and Technology, 2025.
Citation: https://doi.org/10.5194/egusphere-2026-1269-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 272 | 111 | 30 | 413 | 23 | 19 |
- HTML: 272
- PDF: 111
- XML: 30
- Total: 413
- BibTeX: 23
- EndNote: 19
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript by Pignol et al. presents GESsat GEN1, a new satellite system designed for high resolution detection and quantification of methane emissions. The manuscript describes the instrument, the methane retrieval set-up, and results from an initial performance assessment based on real on-orbit data.
I consider that this contribution may be relevant to the AMT readership, as it introduces a promising new mission for the growing methane-sensitive satellite ecosystem. On the other hand, I do not think that the manuscript does a good job to present this new system and its processing chain.
Major comments:
1) Retrieval: The authors present a relatively sophisticated retrieval set-up implemented for this mission. It includes a specific radiative transfer model and an optimal estimation retrieval framework, the description of which spans more than one third of the text. The authors also state that “to validate a physics-guided parametric retrieval framework that balances accuracy, robustness, and computational efficiency for large-scale processing” (L54) is an objective of this work.
However, the authors adopt a number of assumptions and simplifications on this general retrieval framework (spectrally-constant albedo, no scattering, fixed vertical profiles, ...), which ends up in a retrieval state vector consisting of only two elements (XCH4 and XH2O). Also, there is no description of the prior used to constrain the retrieval in the optimal estimation framework.
So it seems that despite the complexity of the initial retrieval framework, the actual retrieval simply relies on a two-parameter forward model applied to 25-nm fiting windows. I can imagine that the resulting XCH4 maps can still show XCH4 fields with a decent SNR, but also very inaccurate (confirmed by the relatively large systematic errors shown in Figs. 13-14). Also, a simplified forward model consisting in two-way gas transmittances derived from CH4 and H2O cross sections neglecting scattering would lead to a similar performance than the retrieval approach proposed here. Those transmittances could be easily generated from Hitran, e.g. using Libradtran.
I would recommend to substantially revise section 3 to properly present the actual retrieval scheme, with a proper discussion of the compromises adopted and their potential impact on the large retrieval offsets obtained in the comparison with TCCON data.
2) Scope: The manuscript presents a number of technical system performance and data quality tests, including inteferogram formation or Fizeau interferometer imperfections. However, I miss some more information on the mission characteristics and data quality that could be of interest to potential data users.
In particular, more examples of good and bad retrieval cases could be provided (only a couple of methane plumes are shown). I would also welcome a discussion of the potential and limitations of this new system with respect to other available satellite data sources, including GHGSat, Carbon Mapper or the several public imaging spectroscopy missions being used for plume mapping (EnMAP, EMIT). The comparison to GHGSat would be particularly relevant because of the technical similarity between the two missions.
Other minor comments:
L3. please also provide the instrument’s spectral window and a spectral resolution and the acquisition’s spatial swath
L11. I don’t think that Carbon Mapper is the best reference for the matched filter retrieval. For example, Thompson et al. https://agupubs.onlinelibrary.wiley.com/doi/full/10.1002/2016GL069079 would be a much better reference
L29. Jongaramrungruang et al. is not a proper reference of GWP numbers
L68. The 120 ppbv calculation corresponds to a single pixel, which is not representative of a real cases in which several pixels above the noise level would be needed to detect a plume
Section 2.1: please add information on swath and spectral window and resolution
L93 and 96: what are “low noise” and “sufficient SNR”?
Fig. 3. I am not sure this figure is relevant
Fig. 5. It is not clear to me how the radiance differences between Libradtran and the parametric model may relate to XCH4 retrieval errors?
Eq. 17-18: how is Sa generated?
L336-341: this could be in the Methods sections, rather than in Results.
Fig.8: a fit & residual of the forward model to those spectra would be ore informative of the data quality and retrieval performance
Fig. 10: what fraction of the XCH4 STD is due to surface clutter and what to measurement noise? Acquisitions with surface with different levels of spatial heterogeneity could be used to evaluate that, and it would be useful information to users.
Fig. 12: it is difficult to appreciate what is represented in this “internal absolute consistency” plot.