Operational machine learning for remote spectroscopic detection of CH 4 point sources Journal: Remote Sensing of Environment
Vít Růžička, Gonzalo Mateo-García, Itziar Irakulis-Loitxate, Juan Emmanuel Johnson, Manuel Montesino San Martin, Anna Allen, Luis Guanter, David R. Thompson · 2025
Read Operational machine learning for remote spectroscopic detection of CH 4 point sources Journal: Remote Sensing of Environment by Vít Růžička, Gonzalo Mateo-García, Itziar Irakulis-Loitxate, Juan Emmanuel Johnson, Manuel Montesino San Martin, Anna Allen, Luis Guanter, David R. Thompson (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-07719.txt.
Introduction
Vít Růžička and colleagues run the first working system that spots methane leaks from space automatically. Methane cuts offer cheap, fast climate gains, but matched-filter satellite retrievals flag many false sources that humans must check by hand. Imaging spectrometers on several satellites already scan the planet.
A machine learning detector now operates inside the United Nations Methane Alert and Response System. It screens spectrometer data for point sources with regular global coverage. Automation turns occasional expert hunts into routine monitoring that can scale to future satellite fleets. Faster alerts help operators fix leaks while the climate benefit is largest.
Catalog record
| CREATOR | Vít Růžička, Gonzalo Mateo-García, Itziar Irakulis-Loitxate, Juan Emmanuel Johnson, Manuel Montesino San Martin, Anna Allen, Luis Guanter, David R. Thompson |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-07719
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Operational machine learning for remote spectroscopic detection of CH 4 point sources Journal: Remote Sensing of Environment Vít Růžička, Gonzalo Mateo-García, Itziar Irakulis-Loitxate, Juan Emmanuel Johnson, Manuel Montesino San Martin, Anna Allen, Luis Guanter, David R. Thompson Abstract Mitigating anthropogenic methane sources is one the most cost-effective levers to slow down global warming. While satellite-based imaging spectrometers, such as EMIT, PRISMA, and EnMAP, can detect these point sources, current methane retrieval methods based on matched filters still produce a high number of false detections requiring laborious manual verification. This paper describes the operational deployment of a machine learning system for detecting methane emissions within the Methane Alert and Response System (MARS) of the United Nations Environment Programme’s International Methane Emissions Observatory. We created the largest and most diverse global dataset of annotated methane plumes from three imaging spectrometer missions and quantitatively compared different deep learning model configurations. Focusing on the requirements for operational deployment, we extended prior evaluation methodologies from small tiled datasets to full granule evaluation. This revealed that deep learning models still produce a large number of false detections, a problem we address with model ensembling, which reduced false detections by over 74%. Deployed in the MARS pipeline, our system processes scenes and proposes plumes to analysts, accelerating the detection and analysis process. During seven months of operational deployment, it facilitated the verification of 1,351 distinct methane leaks, resulting in 479 stakeholder notifications. We further demonstrate the model’s utility in verifying mitigation success through case studies in Libya, Argentina, Oman, and Azerbaijan. Our work represents a critical step towards a global AI-assisted methane leak detection system, which is required to process the dramatically higher data volumes expected from new and current imaging spectrometers. 1 Introduction The reduction of anthropogenic methane emissions represents a critical component of global climate mitigation strategies. As the second largest contributor to global warming after CO 2 ( 23 ) , methane’s short atmospheric lifetime makes its reduction a high-priority lever for achieving the climate targets outlined in the Intergovernmental Panel on Climate Change (IPCC) ( 25 ) . A key initiative of the United Nations Environmental Programme (UNEP) International Methane Emissions Observatory (IMEO) is the Methane Alert and Response System (MARS), which detects large point sources of methane emissions using satellite data. MARS utilizes a diverse array of imagery from various public space missions to identify plumes, notify stakeholders, and monitor sources globally over time. This initiative leverages both multispectral and hyperspectral satellite data. Although multispectral sensors offer higher data frequency and longer historical archives, they have limitations over heterogeneous landscapes such as dark, vegetated, or urban areas ( 45 ; 12 ) . In contrast, hyperspectral sensors provide the enhanced sensitivity required to detect smaller emissions, even across the most complex surfaces ( 19 ; 16 ; 31 ) . The current landscape of hyperspectral remote sensing missions includes a diverse and growing array of data sources. These include airborne hyperspectral sensors, such as NASA’s AVIRIS series and EDF’s MethaneAIR; instruments hosted on the International Space Station (ISS), such as NASA’s EMIT; and a host of satellite missions, including DLR’s EnMAP, ASI’s PRISMA, EDF’s MethaneSAT, the Chinese Gaofen (GF-5) and Ziyuan (ZY-1) series, and the recently launched Tanager-1 mission from Carbon Mapper and Planet. Furthermore, this collection of sensors is expected to expand significantly, with several next-generation methane-sensitive satellites under development, such as NASA’s SBG (estimated to generate 10x more data than EMIT) and Carbon-I, ESA’s CHIME, or DLR’s CO2Image. The increasing volume of data from current and forthcoming hyperspectral missions requires the automation of methane emission detection. Existing detection workflows, which are based on labour intensive manual review, are inefficient for systematically monitoring large areas and are only suited to monitoring and studying previously identified emission sources. Consequently, a critical need exists for automated tools that can support analysts in monitoring not only known sites but also in discovering new emission events over vast geographical regions. Furthermore, automation can systematize the otherwise arduous and subjective task of delineating the geographic extent of methane plumes, which is a critical step for quantifying emission rates that currently depends on inconsistent human interpretation. This paper describes a fully operational methane plume detection system integrated into MARS at UNEP’s IMEO. By automating the initial screening of hyperspectral satellite data, the models developed in this study facilitate the systematic monitoring of vast areas and the discovery of previously unknown emitters. This capability substantially reduces the manual workload for analysts, allowing a small team to efficiently process a global data stream. In the operational workflow, the system flags candidate plumes for subsequent manual verification, which ensures the correctness and identifies relevant stakeholders for notification ( 43 ) . The integration of this model into the MARS pipeline enhances operational efficiency by enabling timely detection of emission events for rapid stakeholder engagement and by scaling monitoring capabilities to support a more comprehensive and systematic analysis of methane dynamics and regulatory accountability. Operationalizing a robust and scalable machine learning system for this task required significant advances beyond prior research. Our key contributions are: (i) the curation and public release of the largest and most diverse global dataset of manually annotated methane plumes, compiled from three distinct hyperspectral missions; (ii) a rigorous evaluation of various model configurations and a comprehensive performance analysis on full satellite granules, culminating in an ensemble approach that reduces false alerts by over 74% compared to the prior state-of-the-art ( 35 ) ; (iii) the demonstration of successful zero-shot generalisation across sensors, with models trained on EMIT data evaluated on PRISMA and EnMAP, whose performance is further enhanced with fine-tuning; (iv) the successful integration of our models into the UNEP IMEO’s MARS operational pipeline, where it processed almost 20,000 scenes, resulting in 1,351 verified methane leaks, and 479 stakeholder notifications over the first five months of operation of the model; and (v) a demonstration of the model’s utility not only for initial detection but also for the verifying successful mitigation, substantiated by case-studies in Libya, Argentina, Oman, and Azerbaijan. The remainder of the paper is organized as follows: Section 2 reviews the literature on methane detection and prior machine learning approaches. Section 3 describes the datasets from the EMIT, PRISMA, and EnMAP missions. Section 4 details our machine learning methodology and Section 5 outlines our experimental setup for both single-sensor training and cross-sensor adaptation. Section 6 presents our findings on model performance, false positive reduction, operational deployment results, and mitigation case-studies. Finally, Section 7 offers concluding remarks and discusses future work. 2 Background literature Here, we briefly review prior work using machine learning models for methane detection. For a more in-depth review, we refer the reader to surveys such as ( 42 ) or the background chapters of ( 36 ) . Early scientific papers of ( 10 ; 41 ; 7 ; 18 ; 19 ) systematically explore large selected geographical regions and demonstrate that methane leaks can be tracked with both multispectral (MSI) ( 45 ; 37 ; 19 ) and hyperspectral (HSI) ( 7 ; 18 ; 16 ; 31 ) instruments. These articles measure methane enhancement using band ratios for MSI data ( 45 ) and matched filter approaches for HSI data