MAPL-EMIT methane model from Google and NASA JPL uncovers 23,000 hidden plumes
Google Research and NASA's Jet Propulsion Laboratory have introduced MAPL-EMIT, short for Methane Analysis and Plume Localization with EMIT, an AI system that detects, quantifies, and pinpoints methane plumes worldwide using observations from NASA's EMIT imaging spectrometer aboard the International Space Station. The MAPL-EMIT methane model, detailed this week in the Proceedings of the National Academy of Sciences, tracks a greenhouse gas whose warming effect over a 100-year horizon is roughly 30 times that of carbon dioxide. Benchmarked against NASA's gold-standard L2B plume dataset, the system recovered 84% of expert-annotated plumes and flagged about 50% more that human reviewers had missed.
Extended across the satellite archive, that performance rewrites what is known about global emissions. Working through roughly 1,100 EMIT observation granules, the model identified more than 23,000 additional plumes worldwide, and its hit list includes 24 of the 25 highest-emitting landfills on the planet. Plume hunting previously depended on analysts inspecting imagery by eye, a process far too slow to cover the full EMIT catalog. Lifting that ceiling is what turns a research result into an operational capability, and the detections land at facility scale, so individual sources can be named and addressed instead of entire regions being flagged.
Inside the MAPL-EMIT methane model
MAPL-EMIT is an end-to-end vision transformer that ingests the complete radiance spectrum EMIT records at a ground resolution near 60 meters per pixel and estimates methane enhancement across every pixel in a scene at once. To train it, the Google and NASA JPL teams generated 3.6 million physics-simulated methane plumes and injected them into real EMIT scans. That augmentation strategy teaches the network to pull weak methane signals out of noisy, complex terrain instead of memorizing idealized plume shapes, and it is the reason the model generalizes beyond anything a human annotator would have labeled.
Detection is fundamentally a problem of seeing the invisible. Methane and carbon dioxide are transparent to the human eye, but each leaves a distinct spectral fingerprint, and EMIT's hyperspectral readings were designed to capture exactly those traces. MAPL-EMIT contributes the ability to interpret that signal across a full scene without waiting on a reviewer, which is what makes global coverage feasible on data already in orbit.
The central figures from the study and its validation:
| Metric | MAPL-EMIT result |
|---|---|
| Training data | 3.6 million physics-simulated plumes on real EMIT scenes |
| Ground resolution | ~60 meters per pixel |
| Expert plume recall | 84% of NASA L2B gold-standard annotations |
| Extra detections | ~50% more plumes across ~1,100 EMIT granules |
| New plumes found | 23,000+ worldwide |
| Top landfill emitters | 24 of the world's 25 largest |
The landfill blind spot
Waste sites matter because self-reported inventories in the sector have proven unreliable, and satellites repeatedly catch what the books miss. In a NASA analysis of 70 high-emitting U.S. landfills, median emissions measured from space ran 77% higher than the levels those facilities reported to the EPA. Among the 38 sites in that group that recover methane gas, emissions averaged roughly 200% above reported values. Detection that leans on self-reporting leaves the biggest sources invisible, so MAPL-EMIT's identification of 24 of the 25 largest landfill emitters effectively hands operators a prioritized list of where verification and repair will pay off.
Landfills are one slice of a much larger methane ledger. Oil, natural gas, and coal operations release an estimated 97 million metric tons of methane a year, atmospheric concentrations of the gas have more than doubled over the past two centuries, and methane drives roughly a quarter of human-caused warming. Because a ton of methane traps about 30 times more heat than a ton of carbon dioxide over the next century, every plume that stays undetected carries an outsized climate cost, which is why the search bottleneck has frustrated scientists as much as the emissions themselves.
Throughput versus analyst judgment
The design deliberately favors recall, and the trade-off is worth stating plainly. Recovering 84% of expert-annotated plumes while surfacing 50% more candidates means the MAPL-EMIT methane model output reads as a triage list: sources are found at scale, and the strongest candidates are confirmed and quantified with follow-up attention before anyone acts. The extra detections are labeled plausible plumes precisely because an automated detector will produce some false positives. What changes is the economics of the workflow: analysts who once worked plume by plume can direct field checks at the highest-probability sites, and third parties can audit any region against the same public standard.
Google has distributed the system as open tooling. The global plume database is published on Google Earth Engine with a companion visualization app, the trained models are downloadable from Kaggle, and inference code is available on GitHub. That packaging puts a functioning detection pipeline into the hands of researchers, policymakers, and facility operators without requiring any of them to build one internally, and it lets all three groups work from a single shared plume record.
The release also positions MAPL-EMIT within a crowded methane-monitoring field. Dedicated instruments such as Carbon Mapper's Tanager-1 satellite have been launched to close the same gap, and NASA's planned Surface Biology and Geology mission carries a spectrometer designed to extend EMIT's capabilities. MAPL-EMIT attacks the problem from the software side, extracting extra value from instruments already in orbit and from their historical archives, so detection power improves without waiting on the next launch. Since EMIT's spectrometer can also read carbon dioxide, the same end-to-end approach has an obvious next target in data already being collected.
What this means for operators
For companies in the energy and waste sectors, the immediate consequence is that the cost of knowing what a facility emits has dropped toward zero, on data that predates any new satellite. A public, AI-flagged plume record lets anyone compare a site's reported numbers against an independent measurement, and the 77% and 200% gaps found at U.S. landfills illustrate the exposure for firms whose inventories understate actual releases. Screening owned assets against the Earth Engine database before a regulator or an investor does is the natural first move, and operators can also run the open MAPL-EMIT methane model code over EMIT scenes covering their own facilities for a private check.
For technologists, the transferable lesson sits in the training recipe. Injecting millions of physics-simulated plumes into real spectral scenes produced a detector that outperforms human review on its own benchmark, and that same augmentation approach can be reused for other imaging spectrometers as they reach orbit. For investors and insurers, the public record functions as an independent check on reported sustainability figures in sectors where books and satellite measurements diverge.
Why this matters
Measurement has been the choke point in methane policy: a gas driving roughly a quarter of warming cannot be regulated or repaired while most of its sources go undetected. With more than 23,000 additional plumes identified and the world's largest landfill emitters named, Google and NASA JPL have converted satellite observations into an open, auditable record that operators, investors, and regulators can all act on.
Sources
AI maps global methane emissions from space
Related Articles
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.