Zhang et al. extensively build upon a number of previously proposed algorithms, including concentration retrieval, plume detection, and emission rate estimation, and integrate them into a practical, open-source toolkit for greenhouse gas point-source detection and quantification. In particular, the use of the open-source Python package tobac, originally developed for cloud tracking, for emission source detection is novel and appears to perform well. Overall, this work provides a valuable reference for the greenhouse gas emission quantification community. The manuscript is well within the scope of Geoscientific Model Development and is suitable for publication after minor revisions. Specific comments are provided below.

Since this framework is intended not only for spaceborne hyperspectral imagers but also potentially for airborne instruments in the future, using unit absorption spectra (k) in units of ppm·m is recommended to improve applicability and inter-instrument comparability (see, e.g., DOI: 10.1016/j.rse.2021.112574).

In addition, unlike the commonly used per-column basis analysis, this work derives reference spectra based on clustering. While this approach is reasonable, as also noted by the authors, detector arrays often exhibit cross-track variability, especially for instruments with strong smile effects. Under such conditions, I do not expect a clustering-based approach (regardless of how surface types are classified) to outperform the per-column approach, at least over homogeneous surfaces. Have the authors conducted any comparison between the clustering-based and per-column-based methods?

Line 14: It would be better to include the chemical formula (CH₄) after methane, consistent with the notation used for CO₂.

Line 16: It may be more appropriate to replace “for identifying emission sources” with “for quantifying point source emissions” in this context.

Line 19: It may be more appropriate to replace “methane and CO₂ emission plumes from individual facilities” with “methane and carbon dioxide point sources.” Also, please avoid mixing full names and abbreviations (methane vs. CO₂); abbreviations should be used consistently after being defined (see DOI: 10.1126/sciadv.adh2391). Please check and correct this writing issue throughout the entire manuscript (e.g., Line 26, Line 37, etc.).

Line 116: Please define FWHM (full width at half maximum) at its first occurrence before using the abbreviation.

Equations (3) and (5) appear to be inconsistent with the corresponding equations in the cited reference. Please check and revise them accordingly.

Line 165: It would be better to replace “in (for example) urban areas” with “in heterogeneous areas.”

Regarding Figure 5, it is unclear what the 30° azimuth difference refers to and how it is calculated. Please clarify what angle is being compared and how the orientation of the rectangular masks is defined. In Figure 5(b), when compared with Figure 5(a), the gray rectangle on the right also appears to contain a methane plume, but in Figure 5(c) this potential plume is excluded. Could the authors clarify why this candidate was removed and which specific criterion was responsible for the exclusion? Moreover, the two orange rectangles in Figure 5(b) appear more likely to originate from the same point source. Visual discontinuities in plumes are common, especially in high-spatial-resolution but relatively low-precision (compared to TROPOMI) methane plume detection. Please justify why these are treated as two separate candidates rather than merged into a single plume.

Line 185: The description of how the mask is determined based on the angle is confusing. Please reorganize and clarify this part.

Line 235: Why not directly resample the original 25 m data to 30 m or 60 m, instead of using the current resampling strategy?

General comments:



This manuscript presents HyperGas, an open-source Python package that provides an end-to-end pipeline for greenhouse gas (CH₄ and CO₂) retrieval, plume detection, and emission rate quantification from spaceborne hyperspectral imagers (PRISMA, EnMAP, EMIT). The paper covers data preparation (L1), concentration enhancement retrieval via matched filter and lognormal matched filter (L2), semi-supervised plume detection using watershed segmentation (L3), and emission estimation through the IME and CSF methods (L4), along with a Streamlit-based GUI. The software is validated against 19 Stanford controlled methane releases and demonstrated across several sectors (oil & gas, coal mining, landfill, and power plants).

The topic is timely and relevant. As the authors stated, the lack of standardized, open-source tools for satellite-based greenhouse gas analysis has confined this capability to a few specialized groups. HyperGas addresses a genuine community need and, if developed further, could substantially lower the barrier to entry for point-source monitoring research. The paper is generally well-written, clearly structured, and provides useful case studies. However, the manuscript has several substantive issues that should be addressed before publication. These primarily concern the limited methodological novelty relative to existing tools and literature, the scope and representativeness of the validation, the incomplete discussion of known algorithmic limitations, and certain technical choices that warrant additional justification.

Main concerns:



1. My primary concern relates to the limited algorithmic novelty and insufficient differentiation from existing tools. The core retrieval algorithm implemented in HyperGas is the standard linear matched filter (Thompson et al., 2015), supplemented by the lognormal variant (Schaum, 2021; Pei et al., 2023). The emission quantification relies on the well-established IME and CSF frameworks (Varon et al., 2018). While the authors mention existing tools such as mag1c and emit-ghg, the differentiation remains superficial. In particular, the ddeq library (Kuhlmann et al., 2024) already provides a modular, sensor-agnostic framework for point-source quantification from remote sensing images, including CSF implementation. The authors should provide a more rigorous and systematic comparison with ddeq and other existing packages (e.g., the Carbon Mapper pipeline described in Duren et al., 2025), clearly articulating what HyperGas contributes beyond integration of known methods into a single workflow. Does HyperGas offer measurable improvements in retrieval accuracy, computational efficiency, or detection sensitivity compared to these tools?

1. The controlled-release validation relies on only 19 experiments, all with plume lengths under 1 km and all conducted in limited geographic and meteorological conditions. This is a relatively small sample, and the absence of EMIT from the controlled-release validation is a notable gap, particularly since EMIT has a fundamentally different spatial resolution that could affect detection and quantification performance. Furthermore, the CSF method is not validated against controlled releases at all, the authors note that plume lengths were too short, but this leaves the CSF implementation essentially unvalidated. The CO₂ quantification is demonstrated only qualitatively (three power plant cases) with no independent validation data. The authors acknowledge this but it weakens the paper's claims about CO₂ capabilities. If possible, expand the validation dataset or at least compare their CO₂ estimates against reported emission data from the demonstrated power plants.

1. The wind calibration that underpins both the IME and CSF quantification methods is derived from only five 3-hour WRF-LES simulations. These simulations represent a limited range of atmospheric stability conditions and boundary-layer regimes. The study does not discuss how the calibration might perform under strongly stable or convective conditions, or how sensitive the results are to the turbulence parameterization in WRF-LES. Given that wind speed uncertainty typically dominates the error budget in point-source quantification (Jacob et al., 2022), the robustness of these calibrations to a wider range of meteorological scenarios should be addressed. Additionally, the 50 m resampling used for EMIT does not exactly match EMIT's ~60 m pixel size, and this discrepancy should be justified.

1. The authors state that HyperGas defaults to the standard matched filter and switches to the lognormal variant “for large methane emissions (e.g., > 10 t h^-1)” (Pei et al., 2023). However, the nonlinearity issue in matched-filter retrievals is fundamentally a function of the column enhancement magnitude (ΔXCH₄), not directly of the emission rate. A 10 t h^-1 source under strong winds may produce lower per-pixel enhancements than a 2 t h^-1 source under calm conditions. The switching criterion should be based on retrieved ΔXCH₄ values rather than on emission rates, which are not known a priori. This design choice needs either better justification or revision.

-------------------



Specific & minor points:

Section 2.1.1 (L1 radiance data): The statement “No pre-treatment is applied to the L1 data” should be qualified. PRISMA L1 data is known to exhibit spectral smile effects and across-track radiometric non-uniformity (Guanter et al., 2021). If no destriping or spectral recalibration is applied, the authors should discuss the expected impact on retrieval quality.

Section 2.2.1 (Matched filter): The manuscript uses the SPy (Spectral Python) library for the matched filter implementation. Given that MAG1C (Foote et al., 2020) introduces sparsity priors and per-pixel albedo corrections that demonstrably improve retrievals, the authors should discuss why they chose the simpler SPy implementation over MAG1C-style enhancements.

Section 2.2.2 (Lognormal matched filter): The claim that “the lognormal matched filter only supports positive radiances” is somewhat misleading. All physical radiances are positive; the issue is that noise can produce negative values in certain bands. This should be clarified.

Section 2.3 (Plume detection): The use of the full NIR window (1300–2500 nm) for plume detection following Roger et al. (2024) is a sensible choice. However, the combined use of Chambolle TV denoising with J-Invariance calibration and watershed segmentation from tobac introduces multiple tunable parameters. The sensitivity of the final emission estimate to these parameter choices is not systematically assessed. A sensitivity analysis, at least for the most critical parameters, would strengthen confidence in the method's robustness.

Section 2.4.2 (Wind calibrations): The three background scenes (Xinjiang, Anna Creek, Madrid) are all over land. No water-adjacent or coastal scenes are included. How representative are the calibrations for plumes occurring near coastlines or over partially water-covered scenes?

Section 3.2 (Carbon dioxide emissions): The discussion of buoyant plume effects on CO₂ wind calibration is important but incomplete. The authors note that “actual CO₂ emissions are released as hot, buoyant plumes at heights greater than 10 m” but power plant stacks are typically 100–200 m tall, and the effective plume rise can be several hundred meters. This substantially changes the relevant wind speed. Using U₁₀ for elevated sources is a known and significant limitation that deserves more thorough treatment.

Line 361: "HypeGas v1.0" appears to be a typographical error, should be "HyperGas v1.0."

General Writing: The manuscript is clearly written overall, but there are minor instances of inconsistent notation (e.g., ΔX vs. ΔXCH₄) and a few grammatical issues (e.g., Line 269: "We have performed", should be lowercase "we").

References:

Duren R, Cusworth D, Ayasse A, et al. The Carbon Mapper emissions monitoring system[J]. Atmospheric Measurement Techniques, 2025, 18(22): 6933-6958.

Foote M D, Dennison P E, Thorpe A K, et al. Fast and accurate retrieval of methane concentration from imaging spectrometer data using sparsity prior[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(9): 6480-6492.

Guanter L, Irakulis-Loitxate I, Gorroño J, et al. Mapping methane point emissions with the PRISMA spaceborne imaging spectrometer[J]. Remote Sensing of Environment, 2021, 265: 112671.

Jacob D J, Varon D J, Cusworth D H, et al. Quantifying methane emissions from the global scale down to point sources using satellite observations of atmospheric methane[J]. Atmospheric Chemistry and Physics, 2022, 22(14): 9617-9646.

Kuhlmann G, Koene E, Meier S, et al. The ddeq Python library for point source quantification from remote sensing images (version 1.0)[J]. Geoscientific Model Development, 2024, 17(12): 4773-4789.

Pei Z, Han G, Mao H, et al. Improving quantification of methane point source emissions from imaging spectroscopy[J]. Remote Sensing of Environment, 2023, 295: 113652.

Roger J, Guanter L, Gorroño J, et al. Exploiting the entire near-infrared spectral range to improve the detection of methane plumes with high-resolution imaging spectrometers[J]. Atmospheric Measurement Techniques, 2024, 17(4): 1333-1346.

Schaum A. A uniformly most powerful detector of gas plumes against a cluttered background[J]. Remote Sensing of Environment, 2021, 260: 112443.

Thompson D R, Leifer I, Bovensmann H, et al. Real-time remote detection and measurement for airborne imaging spectroscopy: a case study with methane[J]. Atmospheric Measurement Techniques, 2015, 8(10): 4383-4397.

Varon D J, Jacob D J, McKeever J, et al. Quantifying methane point sources from fine-scale satellite observations of atmospheric methane plumes[J]. Atmospheric Measurement Techniques, 2018, 11(10): 5673-5686.