How Google and NASA Built an AI Model to Spot Methane Leaks From Space

Google and NASA's Jet Propulsion Laboratory have introduced MAPL-EMIT, an AI model that detects methane emissions globally using data from NASA's EMIT instrument, an imaging spectrometer aboard the International Space Station. The research, published in the Proceedings of the National Academy of Sciences, found the model detects 50 percent more methane plumes than human experts working with the same satellite data, identifying more than 23,000 additional plumes worldwide, including 24 of the 25 largest-emitting landfills globally.

 

Why Detecting Methane From Space Requires Solving a Specific Signal Problem

 

Methane carries a warming potential roughly 30 times that of carbon dioxide over a 100-year timeframe, making it one of the highest-priority targets for near-term climate mitigation, since reducing methane emissions delivers a disproportionately large climate benefit relative to the volume of gas actually reduced. Satellite instruments like EMIT can detect methane plumes, concentrated clouds of released methane gas, from orbit, but identifying these plumes reliably against the backdrop of complex, varied terrain and atmospheric noise has historically required extensive manual review by human analysts, a process that is inherently slow and limited by the number of trained experts available to review the resulting satellite imagery.

MAPL-EMIT specifically targets that bottleneck by training an AI model on 3.6 million physics-simulated methane plumes, synthetic training examples generated from physical models of how methane plumes actually behave and appear in satellite data, rather than relying solely on a smaller set of real-world, manually verified examples. That large-scale synthetic training approach allows the model to learn the visual and spectral signatures of methane plumes across a considerably wider range of terrain types and atmospheric conditions than a training set limited to previously identified real-world plumes could provide.

 

Read more: Alfa Laval Selected to Supply Heat Transfer Tech for Europe's Largest Carbon Capture Project

 

Why Detecting Landfill Emissions Specifically Matters for Near-Term Mitigation

 

The model's identification of 24 out of 25 of the world's largest-emitting landfills is a particularly notable result, since landfills represent a methane source category where mitigation action, capturing and either flaring or utilising landfill gas, is generally well-established and readily deployable once a specific high-emitting site has been accurately identified and located. Unlike some other methane sources that might require more complex or costly intervention, landfill methane capture technology is mature and commercially available, meaning the primary barrier to mitigation in many cases is simply knowing precisely which specific landfills are generating the largest, most impactful emissions.

By identifying nearly all of the world's largest-emitting landfills through this AI-driven analysis of existing satellite data, the research provides a specific, actionable target list that landfill operators, regulators or climate finance organisations could use to prioritise mitigation investment toward the highest-impact sites first, rather than needing to conduct comprehensive site-by-site assessment across the full global population of landfills to identify which ones warrant priority attention.

 

Why Making the Data and Tools Publicly Available Extends the Research's Practical Value

 

Google has released the resulting global plume database on Earth Engine, its geospatial data analysis platform, alongside a dedicated app for visualising the data, and has made the open-source models available on Kaggle with inference tools on GitHub. That open access approach matters because it extends this research's practical utility beyond the original research team, allowing external researchers, policymakers, climate finance organisations or environmental regulators to directly access and build upon the model's methane plume detections rather than needing to independently replicate the underlying detection methodology or rely solely on the published paper's summary findings.

That open-source release pattern connects to the broader trend of AI-driven emissions verification and monitoring technology examined throughout recent coverage, including Everimpact's satellite-based emissions tracking platform and the various carbon credit verification methodologies relying on remote sensing, reflecting a broader industry direction toward making large-scale, satellite-derived emissions data more accessible and actionable for organisations working on methane mitigation specifically, rather than keeping such detection capability confined to specialised research institutions alone.

 

Explore OneStop ESG Marketplace: Monitoring and testing

 

Key Takeaways

 

MAPL-EMIT demonstrates how AI trained on large-scale synthetic data can outperform manual expert analysis in detecting methane plumes from satellite imagery, identifying over 23,000 additional plumes beyond what human review had previously catalogued. The model's identification of nearly all the world's largest-emitting landfills provides a specific, actionable target list for near-term mitigation given landfill gas capture technology's existing maturity. Public release of the database, visualisation tools and open-source model extends the research's practical value to external researchers and organisations working on methane mitigation.