arxiv: 2604.10094 · v1 · submitted 2026-04-11 · 💻 cs.CV · cs.LG· physics.ao-ph

Global monitoring of methane point sources using deep learning on hyperspectral radiance measurements from EMIT

Vishal V. Batchu , Michelangelo Conserva , Alex Wilson , Anna M. Michalak , Varun Gulshan , Philip G. Brodrick , Andrew K. Thorpe , Christopher V. Arsdale This is my paper

Pith reviewed 2026-05-10 16:30 UTC · model grok-4.3

classification 💻 cs.CV cs.LGphysics.ao-ph

keywords methane point sourceshyperspectral radiancevision transformerplume detectionEMIT instrumentremote sensingdeep learningatmospheric monitoring

0 comments

The pith

A vision transformer trained on synthetic plumes detects 79% of known methane sources in EMIT data and twice as many as human analysts.

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

The paper introduces the MAPL-EMIT model to automatically detect, quantify, and localize methane point sources from full hyperspectral radiance measurements collected by the EMIT satellite instrument. Existing manual plume identification limits the scale of monitoring for these emissions, which contribute to near-term climate forcing. The model is an end-to-end vision transformer trained on 3.6 million synthetic plumes created by injecting physics-based signals into real global EMIT radiance scenes. On real benchmarks it recovers most hand-annotated plume complexes while surfacing additional plausible sources, and independent checks against airborne data, landfills, and controlled releases support its ability to find previously missed emitters. If the approach holds, it replaces labor-intensive workflows with rapid processing of the entire EMIT catalog for facility-scale global mapping.

Core claim

The MAPL-EMIT framework is an end-to-end vision transformer that uses the complete radiance spectrum across all pixels in an EMIT scene to jointly retrieve methane enhancements, delineate plumes, and localize sources, including cases with multiple overlapping plumes. Trained on 3.6 million physics-based synthetic plumes injected into real radiance data, the model shows high recall and precision on synthetic tests and, on 1084 real granules, captures 79% of known NASA L2B plume complexes while identifying twice as many plausible plumes as human analysts. Additional validation with coincident airborne observations, top-emitting landfills, and controlled release experiments confirms detection,

What carries the argument

The MAPL-EMIT end-to-end vision transformer that fuses spectral and spatial context from the full hyperspectral radiance spectrum to retrieve methane enhancements across every pixel.

If this is right

The model supports simultaneous quantification, delineation, and source localization even when multiple plumes overlap in one scene.
High-throughput processing becomes feasible for the full EMIT catalog, enabling global facility-scale methane mapping.
Incorporation of model outputs such as spectral fit scores and noise estimates can further reduce false-positive detections.
Validation on airborne, landfill, and controlled-release data indicates the framework finds sources missed by prior manual methods.

Where Pith is reading between the lines

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

The same synthetic-injection training strategy could be adapted to data from other hyperspectral satellite instruments to expand coverage beyond EMIT.
If processing speed scales with catalog size, the method might support repeated global surveys to track changes in emission rates over time.
Facility-scale plume maps could be cross-referenced with activity data to help attribute emissions to specific industrial sites.

Load-bearing premise

Physics-based synthetic plumes injected into real EMIT radiance data capture enough of the spectral, spatial, and atmospheric variability of actual methane emissions for the trained model to generalize reliably to new real-world scenes.

What would settle it

A new test set of EMIT granules with independent airborne plume confirmation in which MAPL-EMIT detects fewer than half of the confirmed plumes would show that the synthetic training data does not represent real conditions well enough.

read the original abstract

Anthropogenic methane (CH4) point sources drive near-term climate forcing, safety hazards, and system inefficiencies. Space-based imaging spectroscopy is emerging as a tool for identifying emissions globally, but existing approaches largely rely on manual plume identification. Here we present the Methane Analysis and Plume Localization with EMIT (MAPL-EMIT) model, an end-to-end vision transformer framework that leverages the complete radiance spectrum from the Earth Surface Mineral Dust Source Investigation (EMIT) instrument to jointly retrieve methane enhancements across all pixels within a scene. This approach brings together spectral and spatial context to significantly lower detection limits. MAPL-EMIT simultaneously supports enhancement quantification, plume delineation, and source localization, even for multiple overlapping plumes. The model was trained on 3.6 million physics-based synthetic plumes injected into global EMIT radiance data. Synthetic evaluation confirms the model's ability to identify plumes with high recall and precision and to capture weaker plumes relative to existing matched-filter approaches. On real-world benchmarks, MAPL-EMIT captures 79% of known hand-annotated NASA L2B plume complexes across a test set of 1084 EMIT granules, while capturing twice as many plausible plumes than identified by human analysts. Further validation against coincident airborne data, top-emitting landfills, and controlled release experiments confirms the model's ability to identify previously uncaptured sources. By incorporating model-generated metrics such as spectral fit scores and estimated noise levels, the framework can further limit false-positive rates. Overall, MAPL-EMIT enables high-throughput implementation on the full EMIT catalog, shifting methane monitoring from labor-intensive workflows to a rapid, scalable paradigm for global plume mapping at the facility scale.

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.

Desk Editor's Note private letter to a colleague

MAPL-EMIT delivers better plume detection on real EMIT data than prior methods but its success depends on the fidelity of the synthetic training set.

read the letter

The main takeaway is that this vision transformer model trained on millions of synthetic methane plumes in real EMIT radiance data achieves 79% recall on known plumes across more than a thousand granules and identifies twice as many plausible ones as human analysts. It also passes checks with airborne measurements and known sources. What the paper does well is move beyond separate matched-filter retrievals to a single model that handles the full spectrum and spatial context for enhancement, delineation, and localization together. The use of actual global EMIT radiance as background for the synthetics is a practical choice that likely helps with realism. The multiple validation sets give the results more grounding than a synthetic-only study would have. The soft spots are around the synthetic data itself. If the physics-based injection misses some real atmospheric or surface interactions, the model could be picking up artifacts in the extra detections. The abstract does not detail the exact criteria for calling something a plausible plume or provide error bars on the metrics, so those numbers are harder to interpret without the full methods. The reliance on hand-annotated L2B products for the main benchmark is reasonable but introduces some dependence on prior processing. This paper is aimed at researchers and agencies working on satellite-based greenhouse gas monitoring who want to scale up detection. A reader interested in applying ML to hyperspectral remote sensing will find the approach and benchmarks useful. It deserves a serious referee because the real-world performance claims are specific and testable, even though the synthetic assumption needs close examination. I recommend sending it to peer review with attention to the validation details and any ablations on the training data.

Referee Report

2 major / 2 minor

Summary. The paper presents MAPL-EMIT, a vision transformer model for joint methane enhancement retrieval, plume delineation, and source localization from EMIT hyperspectral radiance data. Trained on 3.6 million physics-based synthetic plumes injected into global real EMIT scenes, it reports 79% recall on 1084 hand-annotated NASA L2B plume complexes, twice as many plausible plumes as human analysts, and supporting validations from coincident airborne data, top-emitting landfills, and controlled-release experiments. The approach aims to enable scalable, automated global monitoring at facility scale.

Significance. If the central claims hold, the work offers a meaningful advance toward automated, high-throughput methane point-source monitoring by demonstrating that a single end-to-end model trained on large-scale synthetics can match or exceed manual L2B annotation while identifying additional sources confirmed by independent airborne and release data. The explicit use of real radiance backgrounds for injection and the multi-validation strategy are strengths that support practical deployment on the full EMIT catalog.

major comments (2)

[§3] §3 (Synthetic Training Data): The central generalization claim—that 3.6 million physics-based plume injections into real EMIT radiance suffice to capture spectral, spatial, and atmospheric variability—remains load-bearing for both the 79% recall and the 'twice as many plausible plumes' result. No quantitative assessment of domain shift (e.g., residual statistics between synthetic and real plume spectra after injection, or ablation on aerosol/terrain variability) is provided, leaving open the possibility that extra detections reflect synthetic artifacts rather than true sources.
[Results] Results (real-world benchmarks paragraph): The claim of capturing 'twice as many plausible plumes' than human analysts is central to the performance narrative, yet the manuscript provides no explicit definition or decision criteria for 'plausible' (e.g., exact thresholds on spectral fit scores, estimated noise levels, or spatial coherence metrics). Without this, the factor-of-two improvement cannot be independently verified and risks circularity with the model's own outputs.

minor comments (2)

[Abstract] Abstract: The statement that the model 'significantly lower[s] detection limits' relative to matched-filter approaches lacks any quantitative comparison (e.g., minimum detectable enhancement or precision-recall curves at fixed false-positive rate).
[Abstract] Abstract and Results: No error bars, confidence intervals, or statistical tests accompany the 79% recall figure or the factor-of-two plume count, making it difficult to judge robustness across the 1084-granule test set.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the constructive review and for recognizing the potential of MAPL-EMIT for scalable methane monitoring. We address each major comment below with specific revisions to the manuscript.

read point-by-point responses

Referee: [§3] §3 (Synthetic Training Data): The central generalization claim—that 3.6 million physics-based plume injections into real EMIT radiance suffice to capture spectral, spatial, and atmospheric variability—remains load-bearing for both the 79% recall and the 'twice as many plausible plumes' result. No quantitative assessment of domain shift (e.g., residual statistics between synthetic and real plume spectra after injection, or ablation on aerosol/terrain variability) is provided, leaving open the possibility that extra detections reflect synthetic artifacts rather than true sources.

Authors: We agree that a quantitative domain-shift analysis would strengthen the generalization argument. The training design injects physics-based plumes into real global EMIT radiance scenes precisely to retain authentic spectral, spatial, and atmospheric backgrounds; synthetic hold-out tests and independent real-world validations (airborne overflights, landfill inventories, and controlled releases) already indicate that additional detections are not artifacts. Nevertheless, we will add an appendix containing (i) residual statistics comparing synthetic versus observed plume spectra after injection and (ii) ablation experiments that vary aerosol optical depth and terrain complexity. These additions will be included in the revised manuscript. revision: yes
Referee: [Results] Results (real-world benchmarks paragraph): The claim of capturing 'twice as many plausible plumes' than human analysts is central to the performance narrative, yet the manuscript provides no explicit definition or decision criteria for 'plausible' (e.g., exact thresholds on spectral fit scores, estimated noise levels, or spatial coherence metrics). Without this, the factor-of-two improvement cannot be independently verified and risks circularity with the model's own outputs.

Authors: We accept that an explicit, reproducible definition of 'plausible' is required. In the manuscript, a plume is labeled plausible when it satisfies three model-derived criteria: (1) spectral fit score above a stated threshold, (2) estimated per-pixel noise below a stated threshold, and (3) spatial coherence consistent with plume morphology. These same criteria are used to filter false positives. To eliminate any appearance of circularity, we will revise the real-world benchmarks paragraph to state the exact numerical thresholds and the spatial-coherence metric, and we will add a short methods subsection that documents how these thresholds were chosen from the validation sets. The factor-of-two comparison will then be reported with respect to these fixed, transparent criteria. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's core pipeline trains MAPL-EMIT on 3.6 million physics-based synthetic plumes injected into real EMIT radiance, then measures recall (79%) against external hand-annotated NASA L2B complexes on a held-out test set of 1084 granules plus independent checks from airborne data, landfills, and controlled releases. These evaluation targets are not constructed from the model's fitted outputs, self-generated labels, or prior self-citations; the 'plausible plumes' count is cross-validated externally rather than defined by the model itself. No load-bearing step reduces by construction to its inputs, and the derivation remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the fidelity of physics-based synthetic plume injection into real radiance data and the assumption that model performance on held-out real benchmarks generalizes to operational use. All neural network parameters are learned from the synthetic distribution.

free parameters (1)

Vision transformer weights and training hyperparameters
All model parameters are optimized on the 3.6 million synthetic examples; exact values and selection process are not stated in the abstract.

axioms (1)

domain assumption Physics-based synthetic plumes accurately reproduce the spectral and spatial signatures of real methane emissions in EMIT observations.
This underpins both training and the claim that the model identifies previously uncaptured real sources.

pith-pipeline@v0.9.0 · 5642 in / 1528 out tokens · 73978 ms · 2026-05-10T16:30:47.489009+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Fully Automatic Trace Gas Plume Detection
cs.LG 2026-05 unverdicted novelty 6.0

An automated ML-plus-physics pipeline detects trace gas plumes in EMIT spectrometer data, flagging major events in real time and recovering at least 25% of plumes missed by prior human review.

Reference graph

Works this paper leans on

63 extracted references · 63 canonical work pages · cited by 1 Pith paper · 1 internal anchor

[1]

N., Kostis, H.-N., & Thorpe, A

Armstrong, Z. N., Kostis, H.-N., & Thorpe, A. (2024, May 21). NASA Scientific Visualization Studio. NASA Scientific Visualization Studio. https://svs.gsfc.nasa.gov/5272

work page 2024
[2]

J., Razak, A., Yawanarajah, S., Lunt, M., Tse, E

Ashfold, M., Varkkey, H., Wong, Y. J., Razak, A., Yawanarajah, S., Lunt, M., Tse, E. O. Y., & Oi, E. Y. T. (2023). Global Methane Pledge: A review of data, policy and transparency in reducing methane emissions in Malaysia. In Global Methane Pledge: A Review of Data, Policy and Transparency in Reducing Methane Emissions in Malaysia (pp. vii – viii). ISEAS–...

work page 2023
[3]

(2025, December 5)

AVIRIS-3 L2B Greenhouse Gas Enhancements, Facility Instrument Collection. (2025, December 5). NASA Earthdata; Earthdata. https://www.earthdata.nasa.gov/data/catalog/ornl-cloud-av3-l2b-ghg-2358-1

work page 2025
[4]

K., Cusworth, D

Ayasse, A. K., Cusworth, D. H., Howell, K., O’Neill, K., Conrad, B. M., Johnson, M. R., Heckler, J., Asner, G. P., & Duren, R. (2024). Probability of detection and multi-sensor persistence of methane emissions from coincident airborne and satellite observations. Environmental Science & Technology, 58(49), 21536–21544

work page 2024
[5]

K., Thorpe, A

Ayasse, A. K., Thorpe, A. K., Cusworth, D. H., Kort, E. A., Negron, A. G., Heckler, J., Asner, G., & Duren, R. M. (2022). Methane remote sensing and emission quantification of offshore shallow water oil and gas platforms in the Gulf of Mexico. Environmental Research Letters, 17(8), 084039

work page 2022
[6]

Berk, A., Conforti, P., Kennett, R., Perkins, T., Hawes, F., & van den Bosch, J. (2014). MODTRAN® 6: A major upgrade of the MODTRAN® radiative transfer code. 2014 6th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS), 1–4

work page 2014
[7]

G., Thorpe, A

Brodrick, P. G., Thorpe, A. K., Thompson, D. R., & Green, R. O. (2025). EMIT L2B Greenhouse Gas Algorithm Theoretical Basis Document (ATBD). Jet Propulsion Laboratory, California Institute of Technology

work page 2025
[8]

M., Brodrick, P

Chlus, A. M., Brodrick, P. G., Thorpe, A. K., Chapman, J. W., Jensen, D. J., Coleman, R. W., Fahlen, J., Olson-Duvall, W., Thompson, D. R., & Green, R. O. (2025). AVIRIS-3 L2B greenhouse gas enhancements, facility instrument collection. ORNL Distributed Active Archive Center. https://doi.org/10.3334/ORNLDAAC/2358

work page doi:10.3334/ornldaac/2358 2025
[9]

A., Shephard, M

Clough, S. A., Shephard, M. W., Mlawer, E. J., Delamere, J. S., Iacono, M. J., Cady-Pereira, K., Boukabara, S., & Brown, P. D. (2005). Atmospheric radiative transfer modeling: a summary of the AER codes. Journal of Quantitative Spectroscopy and Radiative Transfer, 91(2), 233–244

work page 2005
[10]

W., Thorpe, A

Coleman, R. W., Thorpe, A. K., Brodrick, P. G., Chlus, A., Fahlen, J. E., Bue, B., Chapman, J., Eckert, R. F., Jensen, D. J., Eastwood, M. L., Thompson, D. R., Elder, C. D., Beltran, S. G., Amezquita, D. R., Hopkins, F., & Green, R. O. (2026). Identifying and quantifying greenhouse gas emissions with the AVIRIS-3 airborne imaging spectrometer. Remote Sens...

work page 2026
[11]

H., Jacob, D

Cusworth, D. H., Jacob, D. J., Varon, D. J., Chan Miller, C., Liu, X., Chance, K., Thorpe, A. K., Duren, R. M., Miller, C. E., Thompson, D. R., Frankenberg, C., Guanter, L., & Randles, C. A. (2019). Potential of next-generation imaging spectrometers to detect and quantify methane point sources from space. Atmospheric Measurement Techniques, 12(10), 5655–5668

work page 2019
[12]

J., Orth-Lashley, B., Girard, M., France, J., Fisher, R

Dowd, E., Manning, A. J., Orth-Lashley, B., Girard, M., France, J., Fisher, R. E., Lowry, D., Lanoisellé, M., Pitt, J. R., Stanley, K. M., O’Doherty, S., Young, D., Thistlethwaite, G., Chipperfield, M. P., Gloor, E., & Wilson, C. (2024). First validation of high-resolution satellite-derived methane emissions from an active gas leak in the UK. Atmospheric ...

work page 2024
[13]

Cusworth, A

Duren, R., D. Cusworth, A. Ayasse, K. Howell, A. Diamond, T. Scarpelli, J. Kim, et al. ‘The Carbon Mapper Emissions Monitoring System’. Atmospheric Measurement Techniques 18, no. 22 (2025): 6933–58. https://doi.org/10.5194/amt-18-6933-2025

work page doi:10.5194/amt-18-6933-2025 2025
[14]

M., Thorpe, A

Duren, R. M., Thorpe, A. K., Foster, K. T., Rafiq, T., Hopkins, F. M., Yadav, V., Bue, B. D., Thompson, D. R., Conley, S., Colombi, N. K., Frankenberg, C., McCubbin, I. B., Eastwood, M. L., Falk, M., Herner, J. D., Croes, B. E., Green, R. O., & Miller, C. E. (2019). California’s methane super-emitters. Nature, 575(7781), 180–184

work page 2019
[15]

Ehret, T., De Truchis, A., Mazzolini, M., Morel, J.-M., d’Aspremont, A., Lauvaux, T., Duren, R., Cusworth, D., & Facciolo, G. (2022). Global Tracking and Quantification of Oil and Gas Methane Emissions from Recurrent Sentinel-2 Imagery. Environmental Science & Technology. https://doi.org/10.1021/acs.est.1c08575

work page doi:10.1021/acs.est.1c08575 2022
[16]

Eldering, A. (2026). Methane Plume Detection and Quantification Intercomparison Analysis Based on EMIT Measurements. In Data Foundation. https://datafoundation.org/news/reports/812/812-Methane-Plume-Detection-and-Quantification-Int ercomparison-Analysis-Based-on-EMIT-Measurements-

work page 2026
[17]

(2023, August 30)

EMIT GHG ATBD V1. (2023, August 30). https://lpdaac.usgs.gov/documents/1696/EMIT_GHG_ATBD_V1.pdf

work page 2023
[18]

Ester, M., Kriegel, H.-P., Sander, J., & Xu, X. (1996). A density-based algorithm for discovering clusters in large spatial databases with noise. Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD-96), 226–231

work page 1996
[19]

Forster, P., Storelvmo, T., Armour, K., Collins, W., Dufresne, J.-L., Howden, D., Knutti, R., Mauritsen, T., & et al. (2021). Short-lived Climate Forcers. In V. Masson-Delmotte, P. Zhai, A. Pirani, S. L. Connors, C. Péan, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M. I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J. B. R. Matthews, T. K. Maycock, T. Waterfiel...

work page 2021
[20]

L., Marshall, B

Gordley, L. L., Marshall, B. T., & Allen Chu, D. (1994). Linepak: Algorithms for modeling spectral transmittance and radiance. Journal of Quantitative Spectroscopy & Radiative Transfer, 52(5), 563–580

work page 1994
[21]

E., Rothman, L

Gordon, I. E., Rothman, L. S., Hargreaves, R. J., Hashemi, R., Karlovets, E. V., Skinner, F. M., Conway, E. K., Hill, C., Kochanov, R. V., Tan, Y., Wcisło, P., Finenko, A. A., Nelson, K., Bernath, P. F., Birk, M., Boudon, V., Campargue, A., Chance, K. V., Coustenis, A., … Yurchenko, S. N. (2022). The HITRAN2020 molecular spectroscopic database. Journal of...

work page 2022
[22]

J., Irakulis-Loitxate, I., & Guanter, L

Gorroño, J., Varon, D. J., Irakulis-Loitxate, I., & Guanter, L. (2023). Understanding the potential of Sentinel-2 for monitoring methane point emissions. Atmospheric Measurement Techniques, 16(1), 89–107

work page 2023
[23]

Green, R. (2022). EMIT L1B at-sensor calibrated radiance and geolocation data 60 m V001 [Dataset]. NASA Land Processes Distributed Active Archive Center. https://doi.org/10.5067/EMIT/EMITL1BRAD.001

work page doi:10.5067/emit/emitl1brad.001 2022
[24]

W., Jensen, D., Bender, H., Vinckier, Q., Xiang, C., Olson-Duvall, W., Lundeen, S., & Thompson, D

Green, R., Thorpe, A., Brodrick, P., Chadwick, D., Lopez, A., Elder, C., Villanueva-Weeks, C., Fahlen, J., Coleman, R. W., Jensen, D., Bender, H., Vinckier, Q., Xiang, C., Olson-Duvall, W., Lundeen, S., & Thompson, D. (2023). EMIT L2B estimated methane plume complexes 60 m V001 [Dataset]. NASA Land Processes Distributed Active Archive Center. https://doi....

work page doi:10.5067/emit/emitl2bch4plm.001 2023
[25]

H., Varon, D

Guanter, L., Irakulis-Loitxate, I., Gorroño, J., Sánchez-García, E., Cusworth, D. H., Varon, D. J., Cogliati, S., & Colombo, R. (2021). Mapping methane point emissions with the PRISMA spaceborne imaging spectrometer. Remote Sensing of Environment, 265(112671), 112671

work page 2021
[26]

Harris, F. J. (1978). On the use of windows for harmonic analysis with the discrete Fourier transform. Proceedings of the IEEE, 66(1), 51–83. https://doi.org/10.1109/PROC.1978.10837

work page doi:10.1109/proc.1978.10837 1978
[27]

Huber, P. J. (1964). Robust estimation of a location parameter. The Annals of Mathematical Statistics, 35(1), 73–101

work page 1964
[28]

Hurley, P. (1994). PARTPUFFA Lagrangian Particle-Puff Approach for Plume Dispersion Modeling Applications. Journal of Applied Meteorology and Climatology, 33(2), 285–294

work page 1994
[29]

International Energy Agency. (2025). Global Methane Tracker 2025. IEA. https://www.iea.org/reports/global-methane-tracker-2025

work page 2025
[30]

J., Varon, D

Jacob, D. J., Varon, D. J., Cusworth, D. H., Dennison, P. E., Frankenberg, C., Gautam, R., Guanter, L., Kelley, J., McKeever, J., Ott, L. E., Poulter, B., Qu, Z., Thorpe, A. K., Worden, J. R., & Duren, R. M. (2022). Quantifying methane emissions from the global scale down to point sources using satellite observations of atmospheric methane. Atmospheric Ch...

work page 2022
[31]

Ledoit, O., & Wolf, M. (2004). A well-conditioned estimator for large-dimensional covariance matrices. Journal of Multivariate Analysis, 88(2), 365–411

work page 2004
[32]

Liu, Z., Hu, H., Lin, Y., Yao, Z., Xie, Z., Wei, Y., Ning, J., Cao, Y., Zhang, Z., Dong, L., Wei, F., & Guo, B. (2022). Swin transformer V2: Scaling up capacity and resolution. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 11999–12009

work page 2022
[33]

Loshchilov, I., & Hutter, F. (2017). Decoupled Weight Decay Regularization. http://arxiv.org/abs/1711.05101

work page internal anchor Pith review Pith/arXiv arXiv 2017
[34]

Makridakis, S. (1993). Accuracy measures: theoretical and practical concerns. International Journal of Forecasting, 9(4), 527–529

work page 1993
[35]

Manolakis, D., & Shaw, G. (2002). Detection algorithms for hyperspectral imaging applications. IEEE Signal Processing Magazine, 19(1), 29–43

work page 2002
[36]

Manolakis, D., Lockwood, R., Cooley, T., & Jacobson, J. (2014). Is the matched filter good enough for hyperspectral target detection? IEEE Transactions on Geoscience and Remote Sensing, 52(7), 4216–4229. https://ieeexplore.ieee.org/document/6678280

work page arXiv 2014
[37]

Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Jia, Y., Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, … Xiaoqiang Zheng. (2015). TensorFlow: Large-Scale Machine Learning on Hete...

work page 2015
[38]

Perlin, K. (2001). Simplex Noise. https://userpages.cs.umbc.edu/olano/s2002c36/ch02.pdf

work page 2001
[39]

Planck, M. (n.d.). 9. On the distribution law of energy in the normal spectrum. Retrieved January 15, 2026, from https://web.pdx.edu/~pmoeck/pdf/planck-paper.pdf

work page 2026
[40]

A., & Brandt, A

Reuland, F., Adams, T., Galvin, K., Kort, E. A., & Brandt, A. R. (2026). Large-Scale Controlled Methane Releases for Satellite-Based Detection and Emission Quantification of Point-Sources [Dataset]. Stanford Digital Repository. https://doi.org/10.25740/qh001qt3946

work page doi:10.25740/qh001qt3946 2026
[41]

Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, 234–241

work page 2015
[42]

Rouet-Leduc, B., & Hulbert, C. (2024). Automatic detection of methane emissions in multispectral satellite imagery using a vision transformer. Nature Communications, 15(1), 3801

work page 2024
[43]

Růžička, V., Mateo-Garcia, G., Gómez-Chova, L. et al. Semantic segmentation of methane plumes with hyperspectral machine learning models. Sci Rep 13, 19999 (2023). https://doi.org/10.1038/s41598-023-44918-6

work page doi:10.1038/s41598-023-44918-6 2023
[44]

J., Maasakkers, J

Schuit, B. J., Maasakkers, J. D., Bijl, P., Mahapatra, G., van den Berg, A.-W., Pandey, S., Lorente, A., Borsdorff, T., Houweling, S., Varon, D. J., McKeever, J., Jervis, D., Girard, M., Irakulis-Loitxate, I., Gorroño, J., Guanter, L., Cusworth, D. H., & Aben, I. (2023). Automated detection and monitoring of methane super-emitters using satellite data. At...

work page 2023
[45]

D., El Abbadi, S

Sherwin, E. D., El Abbadi, S. H., Burdeau, P. M., Zhang, Z., Chen, Z., Rutherford, J. S., Chen, Y., & Brandt, A. R. (2024). Single-blind test of nine methane-sensing satellite systems from three continents. Atmospheric Measurement Techniques, 17(2), 765–782

work page 2024
[46]

Stewart, R., Andriluka, M., & Ng, A. Y. (2016, June). End-to-end people detection in crowded scenes. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA. https://doi.org/10.1109/cvpr.2016.255

work page doi:10.1109/cvpr.2016.255 2016
[47]

Swinehart, D. F. (1962). The Beer-Lambert Law. https://doi.org/10.1021/ed039p333

work page doi:10.1021/ed039p333 1962
[48]

Team, J. (2025). JEO: Model training and inference for geospatial remote sensing and Earth Observation in JAX. https://github.com/google-deepmind/jeo

work page 2025
[49]

R., Green, R

Thompson, D. R., Green, R. O., Bradley, C., Brodrick, P. G., Mahowald, N., Dor, E. B., Bennett, M., Bernas, M., Carmon, N., Chadwick, K. D., Clark, R. N., Coleman, R. W., Cox, E., Diaz, E., Eastwood, M. L., Eckert, R., Ehlmann, B. L., Ginoux, P., Ageitos, M. G., … Zandbergen, S. (2024). On-orbit calibration and performance of the EMIT imaging spectrometer...

work page 2024
[50]

R., Thorpe, A

Thompson, D. R., Thorpe, A. K., Frankenberg, C., Green, R. O., Duren, R., Guanter, L., Hollstein, A., Middleton, E., Ong, L., & Roberts, D. A. (2015). Real-time remote detection and measurement for airborne imaging spectroscopy: A case study with methane. Atmospheric Measurement Techniques, 8(10), 4383–4397. https://doi.org/10.5194/amt-8-4383-2015

work page doi:10.5194/amt-8-4383-2015 2015
[51]

K., Frankenberg, C., & Roberts, D

Thorpe, A. K., Frankenberg, C., & Roberts, D. A. (2014). Retrieval techniques for airborne imaging of methane concentrations using high spatial and moderate spectral resolution: application to AVIRIS. Atmospheric Measurement Techniques, 7(2), 491–506

work page 2014
[52]

K., Green, R

Thorpe, A. K., Green, R. O., Thompson, D. R., Brodrick, P. G., Chapman, J. W., Elder, C. D., Irakulis-Loitxate, I., Cusworth, D. H., Ayasse, A. K., Duren, R. M., Frankenberg, C., Guanter, L., Worden, J. R., Dennison, P. E., Roberts, D. A., Dana Chadwick, K., Eastwood, M. L., Fahlen, J. E., & Miller, C. E. (2023). Attribution of individual methane and carb...

work page doi:10.1126/sciadv.adh2391 2023
[53]

K., Roberts, D

Thorpe, A. K., Roberts, D. A., Bradley, E. S., Funk, C. C., Dennison, P. E., & Leifer, I. (2013). High resolution mapping of methane emissions from marine and terrestrial sources using a Cluster-Tuned Matched Filter technique and imaging spectrometry. Remote Sensing of Environment, 134, 305–318. https://doi.org/10.1016/j.rse.2013.03.018

work page doi:10.1016/j.rse.2013.03.018 2013
[54]

Thorpe, A., Brodrick, P., Chadwick, D., Lopez, A., Villanueva-Weeks, C., Fahlen, J., Jensen, D., Bender, H., Vinckier, Q., Xiang, C., Olson-Duvall, W., Chlus, A., Lundeen, S., Thompson, D., & Green, R. (2026). EMIT L2B Estimated Methane Plume Complexes 60 m V002 [Data set]. NASA Land Processes Distributed Active Archive Center. https://doi.org/10.5067/EMI...

work page doi:10.5067/emit/emitl2bch4plm.002 2026
[55]

Tiemann, E., Zhou, S., Klaser, A., Heidler, K., Schneider, R., & Zhu, X. X. (2025). Machine learning for methane detection and quantification from space: A survey. IEEE Geoscience and Remote Sensing Magazine, PP(99), 2–28

work page 2025
[57]

J., Jacob, D

Varon, D. J., Jacob, D. J., McKeever, J., Jervis, D., Durak, B. O. A., Xia, Y., & Huang, Y. (2018). Quantifying methane point sources from fine-scale satellite observations of atmospheric methane plumes. Atmospheric Measurement Techniques, 11(10), 5673–5686

work page 2018
[58]

Vaughan, A., Mateo-García, G., Gómez-Chova, L., Růžička, V., Guanter, L., & Irakulis-Loitxate, I. (2024). CH4Net: a deep learning model for monitoring methane super-emitters with Sentinel-2 imagery. Atmospheric Measurement Techniques, 17(9), 2583–2593

work page 2024
[59]

R., Green, R

Xiang, C., Thompson, D. R., Green, R. O., Fahlen, J. E., Thorpe, A. K., Brodrick, P. G., Coleman, R. W., Lopez, A. M., & Elder, C. D. (2025). Identification of false methane plumes for orbital imaging spectrometers: A case study with EMIT. Remote Sensing of Environment, 328(114860), 114860

work page 2025
[60]

standard atmosphere transmittances

Zhang, X., Maasakkers, J. D., Roger, J., Guanter, L., Sharma, S., Lama, S., Tol, P., Varon, D. J., Cusworth, D. H., Howell, K., Thorpe, A. K., Brodrick, P. G., & Aben, I. (2025). Global Identification of Solid Waste Methane Super Emitters Using Hyperspectral Satellites. Environmental Science & Technology. https://doi.org/10.1021/acs.est.4c14196 Supporting...

work page doi:10.1021/acs.est.4c14196 2025
[61]

We compute the theoretical spectral radiance of the Sun, modeled as a blackbody at T=5778 K using Planck's Law

We perform a solar spectrum normalization to remove the dominant spectral shape of the solar irradiance (Planck, n.d.). We compute the theoretical spectral radiance of the Sun, modeled as a blackbody at T=5778 K using Planck's Law. This high-resolution blackbody curve is convolved with the specific Spectral Response Function (SRF) for each EMIT band to de...

work page
[62]

Slot linear projections

We apply a cross-track radiometric correction to mitigate detector-level striping and systematic illumination artifacts across the swath. We pre-compute the means for every band at every cross-track position across all the EMIT granules. For a given pixel, the radiance is adjusted by multiplying it by the ratio of the global mean radiance for that band (a...

work page
[63]

Spatial Proximity: The Euclidean distance between the predicted plume origins must not exceed a predefined threshold of 1.5 km

work page
[64]

The Pearson correlation coefficient is computed across overlapping pixels, requiring a correlation exceeding 0.97 for the candidates to be classified as identical

Enhancement Correlation: For candidate pairs meeting the proximity constraint, local enhancement patches centered on their respective origins are extracted and aligned. The Pearson correlation coefficient is computed across overlapping pixels, requiring a correlation exceeding 0.97 for the candidates to be classified as identical. Consolidated candidates ...

work page arXiv 2022