• MAPL-EMIT detects about 50% more plausible methane plumes than human analysis across roughly 1,100 satellite observations.
• The model was trained on 3.6 million simulated plumes and identified more than 23,000 additional methane plumes globally.
• Open data and models could give regulators, companies and researchers new tools to identify emissions and target mitigation.

Google and NASA’s Jet Propulsion Laboratory have developed an artificial intelligence model capable of mapping methane emissions from space at global scale.

The model, known as Methane Analysis and Plume Localization with MAPL-EMIT, uses data collected by NASA’s Earth Surface Mineral Dust Source Investigation instrument. EMIT is installed on the International Space Station and captures hyperspectral observations of the Earth’s surface.

Research detailing the system was published in the Proceedings of the National Academy of Sciences. It could substantially expand the ability of governments, companies and researchers to locate methane sources that previously escaped detection.

Methane has around 30 times the warming potential of carbon dioxide over a 100-year period. It has also contributed roughly one-quarter of human-induced warming since the start of the industrial era. Its relatively short atmospheric lifetime makes methane reduction an important route for slowing near-term warming.

AI expands methane detection

Finding methane from space remains technically difficult. Conventional approaches often require analysts to manually inspect satellite observations or use methods that struggle to distinguish weaker emissions from background noise.

MAPL-EMIT instead combines spectral information with the spatial context surrounding each pixel. This allows the model to identify the shape of a methane plume and trace it towards its likely source. It can also separate overlapping plumes in dense industrial areas.

Google and NASA JPL trained the model using 3.6 million physics-based synthetic methane plumes inserted into real EMIT observations. The simulations represented different emission rates, atmospheric conditions and geographic environments.

When tested against real-world observations, MAPL-EMIT captured 84% of expert-annotated plumes. Across approximately 1,100 EMIT granules, it also identified around 50% more plausible plumes than human analysts.

The wider deployment identified more than 23,000 additional methane plumes globally. It also detected emissions at 24 of the world’s 25 highest-emitting landfill sites.

That increased sensitivity comes with a challenge. Highly sensitive detection systems can produce false positives, particularly over complex terrain. Google said the outputs therefore include confidence measures and spectral-fit information. Users can filter results depending on the level of certainty required.

RELATED ARTICLE: Google Backs Texas Electric Freight Route

From detection to emissions accountability

The implications extend beyond climate science. Faster identification of methane sources could support regulatory enforcement, corporate emissions management and independent verification of reported reductions.

For oil and gas operators, waste companies and other large emitters, improved satellite monitoring could make persistent or previously undetected releases more visible. Investors could also gain another source of data when assessing operational emissions and transition risks.

MAPL-EMIT operates at 60-metre resolution and provides estimates of methane enhancement, plume boundaries and probable source locations. However, coverage depends on where and when EMIT collected observations. Cloud cover, shadows and dark surfaces can also reduce model performance.

Google opens methane data and models

Google has released the global plume database through Earth Engine, together with an interactive visualization application. The trained model and synthetic plume datasets are available through Kaggle, while inference tools have been released through GitHub.

Opening the system to researchers, policymakers and operators could widen access to facility-level methane intelligence. That becomes increasingly important as satellite monitoring capacity expands.

NASA is preparing a new generation of imaging spectrometers that Google says could increase observation coverage by 30 to 50 times. Processing that volume manually would be difficult, making automated detection systems increasingly important.

For governments and companies facing greater pressure to quantify methane reductions, that combination of satellite coverage and AI could reshape emissions oversight. The challenge now moves beyond simply finding methane. It is whether operators, regulators and capital providers can turn better detection into faster reductions.