pith. sign in

arxiv: 2511.21777 · v3 · submitted 2025-11-26 · 💻 cs.LG

Artificial intelligence for methane detection: from continuous monitoring to verified mitigation

Gonzalo Mateo-Garcia , Anna Allen , Itziar Irakulis-Loitxate , Manuel Montesino-San Martin , Marc Watine , Cynthia Randles , Tharwat Mokalled , Alma Raunak
show 8 more authors
Carol Casta\~neda-Martinez Juan E. Jonhson Javier Gorro\~no James Requeima Claudio Cifarelli Luis Guanter Richard E. Turner Manfredi Caltagirone
This is my paper

Pith reviewed 2026-05-17 05:14 UTC · model grok-4.3

classification 💻 cs.LG
keywords methane detectionsatellite imagerymachine learningemission monitoringremote sensinggreenhouse gasesmitigation verification
0
0 comments X

The pith

A machine learning model detects methane emissions from satellite images and has enabled verified mitigation at six persistent emitter sites.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents MARS-S2L, a model that identifies methane plumes in publicly available multispectral satellite data. Trained on a large curated set of over 80,000 images, it delivers detections every two days with 78 percent success and low false positives on new locations. When used in practice, the system sent over 2,700 alerts to operators in 25 countries. This process has led to confirmed and lasting reductions in emissions from major sources, such as a long-running super-emitter in Algeria.

Core claim

MARS-S2L is a machine learning model trained on over 80,000 images that detects methane emissions in satellite imagery, achieving 78% identification of plumes with an 8% false positive rate at 697 unseen sites. Its operational use has issued 2,776 notifications resulting in verified permanent mitigation of six persistent emitters including a super-emitter in Algeria releasing 27,000 tonnes annually and a new discovery in Libya.

What carries the argument

MARS-S2L, a machine learning model for detecting methane emissions in multispectral satellite imagery that enables facility-level attribution and frequent monitoring.

If this is right

  • High-resolution methane detections become available every two days for continuous monitoring.
  • Facility-level attribution supports direct notifications to asset owners.
  • Verified mitigations demonstrate a pathway from detection to actual emission reductions.
  • Scalable application across 25 countries shows potential for global coverage.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Such detection systems could be adapted for monitoring other atmospheric pollutants with distinct spectral signatures.
  • Combining these alerts with ground-based verification might strengthen causal links to mitigation actions.
  • Broader adoption could influence international agreements on methane reduction by providing independent data.

Load-bearing premise

That the verified mitigations are caused by the model's notifications rather than unrelated factors and that the performance metrics generalize to all operational data.

What would settle it

Observation of continued high emissions at a notified site without mitigation, or a substantial increase in false positives when applied to new satellite imagery streams.

read the original abstract

Methane is a potent greenhouse gas, responsible for roughly 30% of warming since pre-industrial times. A small number of large point sources account for a disproportionate share of emissions, creating an opportunity for substantial reductions by targeting relatively few sites. Detection and attribution of large emissions at scale for notification to asset owners remains challenging. Here, we introduce MARS-S2L, a machine learning model that detects methane emissions in publicly available multispectral satellite imagery. Trained on a manually curated dataset of over 80,000 images, the model provides high-resolution detections every two days, enabling facility-level attribution and identifying 78% of plumes with an 8% false positive rate at 697 previously unseen sites. Deployed operationally, MARS-S2L has issued 2,776 notifications to stakeholders in 25 countries, enabling verified, permanent mitigation of six persistent emitters, including a super-emitter in Algeria that had been releasing approximately 27,000 tonnes of methane annually for at least a decade and a previously unknown emitter in Libya first identified by MARS-S2L. These results demonstrate a scalable pathway from satellite detection to quantifiable methane mitigation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces MARS-S2L, a machine learning model for detecting methane emissions in publicly available multispectral satellite imagery. Trained on a manually curated dataset of over 80,000 images, the model achieves 78% plume identification with an 8% false positive rate on 697 previously unseen sites. Operationally deployed, it has issued 2,776 notifications to stakeholders in 25 countries, which the authors state enabled verified permanent mitigation of six persistent emitters, including a super-emitter in Algeria releasing ~27,000 tonnes of methane annually for at least a decade and a previously unknown emitter in Libya.

Significance. If the mitigation verification and causal attribution claims are substantiated, the work provides a concrete demonstration of scaling AI-based satellite monitoring to actionable methane reductions. The operational volume (2,776 notifications), geographic reach (25 countries), and specific high-impact examples (Algeria super-emitter, Libya discovery) represent a strength in moving beyond detection metrics to reported real-world outcomes. The every-two-days revisit capability and use of public data further support potential for broad deployment in climate mitigation efforts.

major comments (2)
  1. [Abstract and operational results] Abstract and operational results section: The central claim that the 2,776 notifications 'enabled verified, permanent mitigation of six persistent emitters' is load-bearing for the paper's headline contribution, yet no protocol is provided for independent verification (e.g., follow-up satellite quantification, regulatory records, or ground reports), no timeline linking specific notification dates to mitigation dates, and no comparison to baseline emission trends or selection effects in which sites received follow-up. This leaves the causal attribution between model detections and the reported mitigations under-specified.
  2. [Results / Evaluation] Performance evaluation on 697 unseen sites: The 78%/8% figures are reported only on a curated held-out set; the manuscript does not analyze or bound potential distribution shift to the full operational image stream (different cloud cover, sensor conditions, or facility types), which directly affects whether the reported precision supports the scale of the 2,776 notifications and the six mitigations.
minor comments (2)
  1. [Methods / Evaluation] Clarify the exact definition of 'plume identification' and 'false positive' in the context of the 697-site test set (e.g., spatial overlap threshold or emission rate threshold).
  2. [Methods] The training dataset curation process (manual labeling of >80,000 images) would benefit from a brief description of inter-annotator agreement or quality control steps.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive review and for recognizing the potential significance of the operational deployment. We address each major comment below with clarifications and commit to revisions that improve transparency without overstating the results.

read point-by-point responses

  1. Referee: [Abstract and operational results] Abstract and operational results section: The central claim that the 2,776 notifications 'enabled verified, permanent mitigation of six persistent emitters' is load-bearing for the paper's headline contribution, yet no protocol is provided for independent verification (e.g., follow-up satellite quantification, regulatory records, or ground reports), no timeline linking specific notification dates to mitigation dates, and no comparison to baseline emission trends or selection effects in which sites received follow-up. This leaves the causal attribution between model detections and the reported mitigations under-specified.

    Authors: We agree that the manuscript would benefit from greater detail on verification. In the revised version we will add a subsection describing the verification protocol, which relies primarily on follow-up multispectral satellite observations confirming the absence of plumes after notification, supplemented where available by operator reports and regulatory records. We will also include timelines for each of the six emitters that link notification dates to the dates of confirming follow-up observations. A discussion of selection effects (sites chosen for follow-up due to repeated high-magnitude detections) and the practical difficulties of establishing pre-notification baselines in an observational setting will be added. These changes will make the supporting evidence more explicit while acknowledging the inherent limits on causal attribution. revision: yes

  2. Referee: [Results / Evaluation] Performance evaluation on 697 unseen sites: The 78%/8% figures are reported only on a curated held-out set; the manuscript does not analyze or bound potential distribution shift to the full operational image stream (different cloud cover, sensor conditions, or facility types), which directly affects whether the reported precision supports the scale of the 2,776 notifications and the six mitigations.

    Authors: The 697-site held-out set was assembled to capture diversity in geography, facility type, and imaging conditions representative of operations. We nevertheless accept that an explicit treatment of distribution shift is warranted. The revised evaluation section will add a discussion of robustness to cloud cover, sensor conditions, and facility types, including any available stratified performance metrics and qualitative bounds derived from the training and test distributions. This will better contextualize how the reported metrics relate to the operational image stream. revision: yes

Circularity Check

0 steps flagged

No significant circularity in model training, evaluation, or operational claims

full rationale

The paper describes an ML model trained on a manually curated dataset of over 80,000 images and evaluated on 697 previously unseen sites, yielding independent performance metrics (78% detection rate at 8% false positive rate). Operational results such as 2,776 notifications and six verified mitigations are presented as downstream outcomes of deployment rather than quantities derived from or equivalent to fitted parameters within the reported results. No equations, self-definitional steps, fitted-input predictions, or load-bearing self-citations appear in the abstract or described content; the derivation chain from data curation to detection to mitigation reporting remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that a supervised model trained on a human-curated image set will generalize to operational satellite streams and that stakeholder notifications will produce independently verifiable emission reductions.

axioms (2)
  • domain assumption A manually curated dataset of 80,000 images is representative of real-world methane plumes across diverse geographies and facilities.
    Invoked to justify training and evaluation on previously unseen sites.
  • domain assumption Stakeholder notifications based on model detections lead to measurable and permanent mitigation actions.
    Required to connect the 2,776 notifications to the six verified mitigations.

pith-pipeline@v0.9.0 · 5582 in / 1531 out tokens · 80843 ms · 2026-05-17T05:14:54.302543+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages · 1 internal anchor

  1. [1]

    hysteresis type phenomenon

    Examples of MARS-S2L predictions compared to expert-annotated ground-truth are shown in (b). MARS-S2L successfully detects emissions from this source until its cessation. site in 2024, with examples of detections in Figure 5 (b). In Libya, MARS-S2L was run on the locations of over 300 oil and gas facilities. This lead to the identification of new emitters...

    work page doi:10.13039/501100011033 2024

  2. [2]

    T. Stocker,Climate change 2013: the physical science basis: Working Group I contribution to the Fifth assessment report of the Intergovernmental Panel on Climate Change(Cambridge university press) (2014)

    work page 2013

  3. [3]

    J. F. Dean,et al., Methane feedbacks to the global climate system in a warmer world.Reviews of Geophysics56(1), 207–250 (2018)

    work page 2018

  4. [4]

    Lauvaux,et al., Global assessment of oil and gas methane ultra-emitters.Science375(6580), 557–561 (2022)

    T. Lauvaux,et al., Global assessment of oil and gas methane ultra-emitters.Science375(6580), 557–561 (2022)

    work page 2022

  5. [5]

    Guanter,et al., Mapping methane point emissions with the PRISMA spaceborne imaging spectrom- eter.Remote Sensing of Environment265, 112671 (2021)

    L. Guanter,et al., Mapping methane point emissions with the PRISMA spaceborne imaging spectrom- eter.Remote Sensing of Environment265, 112671 (2021)

    work page 2021

  6. [6]

    S ´anchez-Garc´ıa, J

    E. S ´anchez-Garc´ıa, J. Gorro˜no, I. Irakulis-Loitxate, D. J. Varon, L. Guanter, Mapping methane plumes at very high spatial resolution with the WorldView-3 satellite.Atmospheric Measurement Techniques Discussions2021, 1–26 (2021)

    work page 2021

  7. [7]

    Irakulis-Loitxate, L

    I. Irakulis-Loitxate, L. Guanter, J. D. Maasakkers, S. Pandey, I. Aben, Satellite-based O&G emitter detection and analysis in Algeria, inEGU General Assembly Conference Abstracts(2022), pp. EGU22– 9451

    work page 2022

  8. [8]

    J. Li, B. Chen, Global revisit interval analysis of Landsat-8-9 and Sentinel-2A-2B data for terrestrial monitoring.Sensors20(22), 6631 (2020)

    work page 2020

  9. [9]

    T. Ehret,et al., Global Tracking and Quantification of Oil and Gas Methane Emissions from Recurrent Sentinel-2 Imagery.Environmental Science & Technology56(14), 10517–10529 (2022), publisher: American Chemical Society, doi:10.1021/acs.est.1c08575,https://doi.org/10.1021/acs.est. 1c08575

    work page doi:10.1021/acs.est.1c08575 2022

  10. [10]

    Gorro ˜no, D

    J. Gorro ˜no, D. J. Varon, I. Irakulis-Loitxate, L. Guanter, Understanding the potential of Sentinel-2 for monitoring methane point emissions.Atmospheric Measurement Techniques16(1), 89–107 (2023)

    work page 2023

  11. [11]

    Vaughan,et al., CH4Net: a deep learning model for monitoring methane super-emitters with Sentinel-2 imagery.EGUsphere2023, 1–17 (2023)

    A. Vaughan,et al., CH4Net: a deep learning model for monitoring methane super-emitters with Sentinel-2 imagery.EGUsphere2023, 1–17 (2023)

    work page 2023

  12. [12]

    Zortea, J

    M. Zortea, J. L. De Sousa Almeida, L. Klein, A. C. Nogueira Junior, Detection of methane plumes using Sentinel-2 satellite images and deep neural networks trained on synthetically created label data, in2023 IEEE International Conference on Big Data (BigData)(2023), pp. 3830–3839, doi: 10.1109/BigData59044.2023.10386482,https://ieeexplore.ieee.org/document...

    work page doi:10.1109/bigdata59044.2023.10386482 2023

  13. [13]

    S. Zhao, Y. Zhang, S. Zhao, X. Wang, D. J. Varon, A Data-Efficient Deep Transfer Learning Framework for Methane Super-Emitter Detection in Oil and Gas Fields Using Sentinel-2 Satellite.EGUsphere2024, 1–34 (2024). 13

    work page 2024

  14. [14]

    P. Joyce,et al., Using a deep neural network to detect methane point sources and quantify emissions from PRISMA hyperspectral satellite images.Atmospheric Measurement Techniques16(10), 2627–2640 (2023), publisher: Copernicus GmbH, doi:10.5194/amt-16-2627-2023,https://amt.copernicus. org/articles/16/2627/2023/

    work page doi:10.5194/amt-16-2627-2023 2023

  15. [15]

    Radman, M

    A. Radman, M. Mahdianpari, D. J. Varon, F. Mohammadimanesh, S2MetNet: A novel dataset and deep learning benchmark for methane point source quantification using Sentinel-2 satellite imagery. Remote Sensing of Environment295, 113708 (2023), doi:10.1016/j.rse.2023.113708,https://www. sciencedirect.com/science/article/pii/S0034425723002596

    work page doi:10.1016/j.rse.2023.113708 2023

  16. [16]

    Rouet-Leduc, C

    B. Rouet-Leduc, C. Hulbert, Automatic detection of methane emissions in multispectral satellite im- agery using a vision transformer.Nature Communications15(1), 3801 (2024)

    work page 2024

  17. [17]

    M. Omara,et al., Developing a spatially explicit global oil and gas infrastructure database for character- izing methane emission sources at high resolution.Earth System Science Data15(8), 3761–3790 (2023), doi:10.5194/essd-15-3761-2023,https://essd.copernicus.org/articles/15/3761/2023/

    work page doi:10.5194/essd-15-3761-2023 2023

  18. [18]

    C. Aybar,et al., CloudSEN12, a global dataset for semantic understanding of cloud and cloud shadow in Sentinel-2.Scientific Data9(1), 782 (2022), doi:10.1038/s41597-022-01878-2,https://www. nature.com/articles/s41597-022-01878-2

    work page doi:10.1038/s41597-022-01878-2 2022

  19. [19]

    Mu ˜noz-Sabater,et al., ERA5-Land: A state-of-the-art global reanalysis dataset for land applications

    J. Mu ˜noz-Sabater,et al., ERA5-Land: A state-of-the-art global reanalysis dataset for land applications. Earth system science data13(9), 4349–4383 (2021)

    work page 2021

  20. [20]

    Lucchesi,File specification for GEOS-5 FP (Forward processing), Tech

    R. Lucchesi,File specification for GEOS-5 FP (Forward processing), Tech. rep. (2013)

    work page 2013

  21. [21]

    Irakulis-Loitxate, L

    I. Irakulis-Loitxate, L. Guanter, J. D. Maasakkers, D. Zavala-Araiza, I. Aben, Satellites Detect Abatable Super-Emissions in One of the World’s Largest Methane Hotspot Regions.Environmental Science & Technology56(4), 2143–2152 (2022)

    work page 2022

  22. [22]

    D. J. Varon,et al., High-frequency monitoring of anomalous methane point sources with multispectral Sentinel-2 satellite observations.Atmospheric Measurement Techniques14(4), 2771–2785 (2021)

    work page 2021

  23. [23]

    Vaughan,et al., CH4Net: a deep learning model for monitoring methane super-emitters with Sentinel-2 imagery.Atmospheric Measurement Techniques17(9), 2583–2593 (2024)

    A. Vaughan,et al., CH4Net: a deep learning model for monitoring methane super-emitters with Sentinel-2 imagery.Atmospheric Measurement Techniques17(9), 2583–2593 (2024)

    work page 2024

  24. [24]

    Aybar,et al., CloudSEN12+: The largest dataset of expert-labeled pixels for cloud and cloud shadow detection in Sentinel-2.Data in Briefp

    C. Aybar,et al., CloudSEN12+: The largest dataset of expert-labeled pixels for cloud and cloud shadow detection in Sentinel-2.Data in Briefp. 110852 (2024), doi:10.1016/j.dib.2024.110852,https:// www.sciencedirect.com/science/article/pii/S2352340924008163

    work page doi:10.1016/j.dib.2024.110852 2024

  25. [25]

    E. D. Sherwin,et al., Single-blind validation of space-based point-source detection and quantification of onshore methane emissions.Scientific Reports13(1), 3836 (2023), number: 1 Publisher: Na- ture Publishing Group, doi:10.1038/s41598-023-30761-2,https://www.nature.com/articles/ s41598-023-30761-2

    work page doi:10.1038/s41598-023-30761-2 2023

  26. [26]

    E. D. Sherwin,et al., Single-blind test of nine methane-sensing satellite systems from three continents. Atmospheric Measurement Techniques17(2), 765–782 (2024), publisher: Copernicus GmbH, doi: 10.5194/amt-17-765-2024,https://amt.copernicus.org/articles/17/765/2024/. 14

    work page doi:10.5194/amt-17-765-2024 2024

  27. [27]

    Bonazzi,et al., An Eye on Methane 2024 (2024),https://www.unep.org/resources/ eye-methane-2024, international Methane Emissions Observatory (IMEO) report

    G. Bonazzi,et al., An Eye on Methane 2024 (2024),https://www.unep.org/resources/ eye-methane-2024, international Methane Emissions Observatory (IMEO) report

    work page 2024

  28. [28]

    Abichou,et al., An Eye on Methane 2024 (2025),https://www.unep.org/resources/ eye-methane-2025, international Methane Emissions Observatory (IMEO) report

    T. Abichou,et al., An Eye on Methane 2024 (2025),https://www.unep.org/resources/ eye-methane-2025, international Methane Emissions Observatory (IMEO) report

    work page 2024

  29. [29]

    United Nations Environment Programme, Eye on Methane: Invisible but not Un- seen (2024),https://wedocs.unep.org/bitstream/handle/20.500.11822/46541/eye_on_ methane_2024_invisible_but_not_unseen.pdf?sequence=3, accessed: 2024-11-19

    work page 2024

  30. [30]

    UN Environment Programme, Catalyzing Methane Action - Harnessing UNEP’s IMEO Data for Methane Emissions Reduction in the MENA Region, Session at the Iraqi Pavilion, 29th Conference of the Parties (COP29) (2024), baku, Azerbaijan, November 11-24, 2024

    work page 2024

  31. [31]

    S. Pandey,et al., Daily detection and quantification of methane leaks using Sentinel-3: a tiered satellite observation approach with Sentinel-2 and Sentinel-5p.Remote Sensing of Environment296, 113716 (2023)

    work page 2023

  32. [32]

    T. A. de Jong,et al., Daily global methane super-emitter detection and source identification with sub-daily tracking.Geophysical Research Letters52(8), e2024GL111824 (2025)

    work page 2025

  33. [33]

    Hong,et al., SpectralGPT: Spectral remote sensing foundation model.IEEE Transactions on Pattern Analysis and Machine Intelligence(2024)

    D. Hong,et al., SpectralGPT: Spectral remote sensing foundation model.IEEE Transactions on Pattern Analysis and Machine Intelligence(2024)

    work page 2024

  34. [34]

    K. Cha, J. Seo, T. Lee, A billion-scale foundation model for remote sensing images.arXiv preprint arXiv:2304.05215(2023)

    work page arXiv 2023

  35. [35]

    Irakulis-Loitxate, J

    I. Irakulis-Loitxate, J. Gorro ˜no, D. Zavala-Araiza, L. Guanter, Satellites Detect a Methane Ultra- emission Event from an Offshore Platform in the Gulf of Mexico.Environmental Science & Technology Letters(2022), publisher: American Chemical Society, doi:10.1021/acs.estlett.2c00225,https:// doi.org/10.1021/acs.estlett.2c00225

    work page doi:10.1021/acs.estlett.2c00225 2022

  36. [36]

    CV AT.ai Corporation, Computer Vision Annotation Tool (CV AT) (2023),https://github.com/ opencv/cvat

    work page 2023

  37. [37]

    B. J. Schuit,et al., Automated detection and monitoring of methane super-emitters using satellite data. Atmospheric Chemistry and Physics Discussions2023, 1–47 (2023)

    work page 2023

  38. [38]

    N. Balasus,et al., A blended TROPOMI+ GOSAT satellite data product for atmospheric methane using machine learning to correct retrieval biases.Atmospheric Measurement Techniques16(16), 3787–3807 (2023)

    work page 2023

  39. [39]

    P. Joyce,et al., Using a deep neural network to detect methane point sources and quantify emissions from PRISMA hyperspectral satellite images.Atmospheric Measurement Techniques16(10), 2627– 2640 (2023)

    work page 2023

  40. [40]

    Joyce,et al., Using a deep neural network to detect methane point sources and quantify emissions from PRISMA hyperspectral satellite images.EGUsphere2022, 1–22 (2022)

    P. Joyce,et al., Using a deep neural network to detect methane point sources and quantify emissions from PRISMA hyperspectral satellite images.EGUsphere2022, 1–22 (2022). 15

    work page 2022

  41. [41]

    Jongaramrungruang,et al., Towards accurate methane point-source quantification from high- resolution 2-D plume imagery.Atmospheric Measurement Techniques12(12), 6667–6681 (2019)

    S. Jongaramrungruang,et al., Towards accurate methane point-source quantification from high- resolution 2-D plume imagery.Atmospheric Measurement Techniques12(12), 6667–6681 (2019)

    work page 2019

  42. [42]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation, inMedical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18(Springer) (2015), pp. 234–241

    work page 2015

  43. [43]

    V. R˚ uˇ ziˇcka,et al., Semantic segmentation of methane plumes with hyperspectral machine learning models.Scientific Reports13(1), 19999 (2023), number: 1 Publisher: Nature Publishing Group, doi: 10.1038/s41598-023-44918-6,https://www.nature.com/articles/s41598-023-44918-6

    work page doi:10.1038/s41598-023-44918-6 2023

  44. [44]

    Berk,et al., MODTRAN6: a major upgrade of the MODTRAN radiative transfer code, inAlgorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XX, M

    A. Berk,et al., MODTRAN6: a major upgrade of the MODTRAN radiative transfer code, inAlgorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XX, M. Velez-Reyes, F. A. Kruse, Eds., International Society for Optics and Photonics (SPIE), vol. 9088 (2014), p. 90880H, doi:10.1117/12.2050433,https://doi.org/10.1117/12.2050433

    work page doi:10.1117/12.2050433 2014

  45. [45]

    Emde,et al., The libRadtran software package for radiative transfer calculations (version 2.0.1)

    C. Emde,et al., The libRadtran software package for radiative transfer calculations (version 2.0.1). Geoscientific Model Development9(5), 1647–1672 (2016), doi:10.5194/gmd-9-1647-2016,https: //gmd.copernicus.org/articles/9/1647/2016/

    work page doi:10.5194/gmd-9-1647-2016 2016

  46. [46]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  47. [47]

    C. Aybar,et al., CloudSEN12, a global dataset for semantic understanding of cloud and cloud shadow in Sentinel-2.Scientific Data9(1), 782 (2022), number: 1 Publisher: Nature Publishing Group

    work page 2022

  48. [48]

    I. P. on Climate Change (IPCC),Chapter 7: The Earth’s Energy Budget, Climate Feedbacks, and Climate Sensitivity(Cambridge University Press) (2021), doi:10.1017/9781009157896.009,https: //www.ipcc.ch/report/ar6/wg1/chapter/chapter-7/#7.6

    work page doi:10.1017/9781009157896.009 2021

  49. [49]

    U. E. P. Agency, Greenhouse Gas Emissions from a Typical Passenger Vehicle (2023),https://www. epa.gov/greenvehicles/greenhouse-gas-emissions-typical-passenger-vehicle

    work page 2023

  50. [50]

    M. W. Jones,et al., Annual methane emissions including land use,https://ourworldindata.org/ grapher/methane-emissions(2024), data source: Jones et al. (2024) – with major processing by Our World in Data

    work page 2024

  51. [51]

    IEA, Global Methane Tracker 2025 (2025),https://www.iea.org/reports/ global-methane-tracker-2025, licence: CC BY 4.0. 16 Supplementary Materials for Artificial intelligence for methane detection: from continuous monitoring to verified mitigation Anna Allen1,6∗†, Gonzalo Mateo-Garcia1∗†, Itziar Irakulis-Loitxate1,4†, Manuel Montesino-San Martin1, Marc Wati...

    work page 2025

  52. [52]

    Validation is al- ways performed on the real data with no simulation

    Model selection is performed using mean average precision on the validation split. Validation is al- ways performed on the real data with no simulation. For offshore locations we additionally fine-tune the model for an extra epoch using only real data. We train the model on a single Azure virtual machine with an A100 80Gb Nvidia GPU, 24 CPU cores and 220G...

    work page 2022

  53. [53]

    Notifications were issued to the operator in October 2024 and again in May and June 2025. Following the final notification on 3 June 2025, the operator conducted an investigation and attributed the leaks to extensive wear and tear on pipelines connecting several wells to a gas collection point. Maintenance and repair work were subsequently completed, elim...

    work page 2024