A Camera on the ISS Is Now Policing Methane: Google and NASA Flag 23,000 Leaks
A space camera originally built to look at desert minerals has been fitted with deep learning and turned into Earth's most uncompromising methane inspector.
Three key facts
Google and NASA's Jet Propulsion Laboratory published the MAPL-EMIT model in PNAS this September, using hyperspectral data from the EMIT instrument on the International Space Station to locate methane plumes globally. EMIT's day job is mapping mineral composition in arid regions; methane detection is a side gig — its hyperspectral capabilities happen to capture methane's chemical fingerprint.
The scoreboard: roughly 50% more methane leaks detected than human expert analysis, over 23,000 previously unrecorded plumes locked on, and 24 of the world's 25 largest-emitting landfills flagged. The model was trained on 3.6 million physics-simulated methane plumes, letting it pick methane signals out of complex surface backgrounds and estimate emission sources.
Google has made the global plume database, the trained model, and the inference tools all public — researchers and policymakers can pick them up and use them directly. In effect, the evidence chain for enforcement has been turned into an open-source project.
Wang Dou's Take
The rot at the heart of the methane business is invisibility: it's the second-biggest greenhouse gas, far more heat-trapping than CO2, yet its sources are scattered and concealed, so regulation basically runs on corporate self-reporting. What self-reporting is worth, this number tells you — the moment AI got involved, it found 50% more leaks. Which is to say the old "expert analysis plus corporate reports" system was missing half the picture.
The 24-out-of-25 hit rate cuts deeper: nearly every one of the world's biggest-emitting landfills has been named and shamed. The hardest part for enforcers used to be gathering evidence; now evidence is couriered from space straight into a public database. Anyone planning to deny it should first learn how to lie to a satellite. With EU methane rules tightening, this open database is a ready-made ledger handed to regulators.
It's also the most dignified answer yet to the "AI burns too much power" crowd. Large models guzzle electricity while turning invisible emissions into a public ledger. Whether AI is climate's sinner or its accountant doesn't depend on how much power it eats — it depends on what the compute is aimed at. This time, it's aimed at the debts humanity tried to skip out on.
Source: Google Research Blog · PNAS · ESG Dive