Methane is a powerful greenhouse gas: over a 100-year horizon its global warming potential is about 30 times that of carbon dioxide, and it has contributed roughly 25% of human-caused warming since the start of the industrial era. Because methane persists in the atmosphere for a relatively short time, rapidly cutting emissions provides a fast-action way to slow near-term warming. More than 125 countries have committed under the Global Methane Pledge to reduce methane emissions by 30% by 2030.
Cost-effective mitigation focuses on point sources from waste, agriculture and energy infrastructure. To track those sources at scale, scientists increasingly rely on space-based hyperspectral imaging.
EMIT on the ISS and its role in methane detection
The NASA Earth Surface Mineral Dust Source Investigation (EMIT) instrument on the International Space Station was originally designed to map mineralogy in arid regions. Researchers at the Jet Propulsion Laboratory (NASA JPL) and elsewhere have used EMIT’s hyperspectral capabilities to detect methane by recording hundreds of spectral bands per pixel, allowing the identification of chemical signatures of invisible gases. EMIT’s characteristics include an approximately 80 km field of view, 60-meter spatial resolution and 7.4 nm spectral sampling, enabling facility-scale methane measurements with a good signal-to-noise ratio.
MAPL-EMIT: a vision transformer that uses spectral and spatial context
The PNAS paper “Global monitoring of methane point sources using deep learning on hyperspectral radiance measurements from EMIT” describes Methane Analysis and Plume Localization with EMIT (MAPL-EMIT), an end-to-end deep-learning pipeline built on a Swin-S transformer vision architecture. Unlike methods that analyze hyperspectral data pixel-by-pixel, MAPL-EMIT processes full spectra together with their surrounding spatial context. That spatial awareness helps the model distinguish true wind-blown methane plumes—gas trails dispersing from a source—from ground materials that share similar spectral signatures and have caused false detections in the past.
MAPL-EMIT addresses three tasks simultaneously:
- Enhancement Quantification: estimating methane quantity per pixel within a plume.
- Plume delineation: segmenting plume shapes and boundaries, including overlapping plumes.
- Source localization: tracing dispersed gas backward to identify the emission origin.
Training on 3.6 million simulated plumes
Transformer models require very large training sets, and a globally labeled corpus of real methane plumes does not exist. To compensate, the team developed a physics-based simulation framework and generated 3.6 million synthetic methane plumes which were injected into real EMIT scenes. They used Lagrangian puff models to simulate particle movement and turbulent dispersion, recreating the chaotic behavior of real gas emissions. Training on these realistic simulations enabled MAPL-EMIT to learn to detect plumes across a wide range of emission strengths, terrains and atmospheric conditions, and to learn source estimation and the handling of overlapping plumes.
Key advantages of the synthetic training approach included:
- exposure to varying emission rates (large leaks and small intermittent releases),
- plume diversity across many terrain and atmospheric contexts for better generalization,
- the ability to train explicitly on source estimation and overlapping plume scenarios.
Real-world performance and limits
On real EMIT data the model shows promising scalability. When benchmarked against NASA’s expert-annotated L2B methane plume dataset, MAPL-EMIT recovered 84% of expert-annotated plumes and identified roughly 50% more plausible plumes across about 1,100 EMIT granules, demonstrating improved sensitivity to subtle signals over background noise. The model also captured weaker emissions than prior methods and mapped plumes at 24 of the world’s 25 highest-emitting landfills.
False positives remain a practical challenge, particularly in complex terrain. To help users manage that trade-off, MAPL-EMIT outputs are accompanied by physics-based plume confidence (spectral fit) scores derived from multiple detections over strided inference and other properties. Each detected plume is also labeled with a “lower” or “higher” confidence tag so users can filter results according to their tolerance for false positives versus missed detections.
Open resources and implications for mitigation
To support the broader community, the authors have released multiple resources: a global plume database on Google Earth Engine with an interactive Earth Engine App for visualization, the trained model and synthetic plumes on Kaggle, and an inference library on GitHub. These tools aim to help local stakeholders, researchers, policymakers and industry identify and act on methane emissions more quickly.
As NASA prepares next-generation imaging spectrometers that are expected to increase coverage by a factor of 30–50, automated and robust techniques like MAPL-EMIT will be increasingly important for turning expanded observations into actionable mitigation.
Acknowledgements
Contributors from Google and NASA JPL who supported this work include: Alex Wilson, Anna M. Michalak, Varun Gulshan, Philip G. Brodrick, Andrew K. Thorpe, Christopher V. Arsdale, Burak Ekim, Carl Elkin, Tal Geller, Omry Gillon, Nita Goyal, Mansi Kansal, Roy Nadler, John Platt, Sergei Shames, Bijoy Shetty, Aaron Sonabend, Deepika Sukhija, Shahar Timnat, Maxim Neumann, Anton Raichuk, Frances Reuland, Adam R. Brandt, David R. Thompson, Robert O. Green, Jay Radzinski, Vishal V. Batchu, and Michelangelo Conserva.
The PNAS paper contains full methodological details and additional results; the authors encourage the scientific community to use the released data and tools to support targeted methane mitigation efforts.
