High resolution automatic methane emissions detection in satellite data using AI

Reducing methane emissions has taken center stage in slowing global warming, but their systematic detection has remained elusive so far. After years of research, a team of scientists that spun out of the Los Alamos National Laboratory has managed to build the first global and automated high resolution methane emissions detection system, published this week in Nature Communications.

LOS ALAMOS, N.M. , May 15, 2024 Responsible for about a third of global warming to date, methane is much more effective than CO2 at trapping heat in the atmosphere. Curbing methane emissions is therefore widely considered to be among the fastest ways to slow down global warming – explaining a surge in corporate commitments and new regulations, including the recent introduction of fines on methane emissions by the US Environmental Protection Agency.

However, no solution exists today to detect methane at scale. Energy companies and government agencies rely on imperfect information to introduce remedial actions. Detections from local sensors or sensors mounted on cars, planes or drones have limited coverage. Current detections from satellites are not entirely automated and come with a trade-off between poor detection capabilities or poor coverage.

"This technology is the first able to automatically detect emissions with good spatial resolution, anywhere on Earth, every few days."

In a study that just came out in Nature Communications, the research team of Geolabe, out of Los Alamos, New Mexico, has developed the first method to automatically detect methane emissions at high spatial and temporal resolution and global scale. The team trained an AI algorithm able to parse through large amounts of data produced by the powerful Sentinel-2 satellite constellation, and autonomously identify methane signatures. As shown in the peer reviewed study, the approach can detect more than 85% of methane emitted from oil and gas basins such as the Permian Basin and is precise enough to identify the particular source of individual leaks and emissions.

"This technology is the first able to automatically detect emissions with good spatial resolution, anywhere on Earth, every few days", says Bertrand Rouet-Leduc, the main author of the research. "Automation is paramount when analyzing large areas, and we also were able to dramatically improve detection thresholds".

The AI algorithm can precisely identify emission sources that before could only be detected by airplane or by pointing a dedicated satellite. "This hasn't really been done before, especially at that level of detection thresholds and accuracy", says Claudia Hulbert, also author on the study.

"Our technology and platform will hopefully provide a reliable and inexpensive way to get at the fine spatial distribution of methane emissions and their historical trends."

The paper "Automatic Detection of Methane Emissions in Multi-Spectral Satellite Imagery Using a Vision Transformer" is appearing on May 14th, 2024 in Nature Communications:


Geolabe, LLC was founded in 2020 by scientists from the Los Alamos National Laboratory. The team, who has won a NASA award for their research, is developing AIs to extract actionable insights from complex satellite data. www.geolabe.com