Methane Satellites Face a Blind Test

Two trailers sit in a patch of desert near Casa Grande, Arizona, loaded with compressed natural gas bought from local filling stations. The gas is held at about 2,500 pounds per square inch (170 bar), Earth.com reports. It runs through calibrated meters and out of a vertical stack built to mimic a leaking compressor or an unlit flare. Overhead, satellites pass. The people reading their images do not know when the valve opened.

That setup produced the most thorough answer yet to a question that now carries regulatory weight: when a satellite says a pipeline or facility is leaking methane, can anyone believe it? Nature published "Evaluating space-based methane detection through a one-year blind test" on 14 October 2026. Stanford says the results give a benchmark for systems that regulators, environmental groups and companies increasingly use to detect leaks and identify super-emitters.

A test built to embarrass

Researchers had good reason to ask. Earth.com reports that only two small blind tests had ever checked satellite methane images against leaks of known size.

The new paper describes a year-long, single-blind test of 746 events: 481 methane releases of 9.5–1,183.7 kilograms per hour, plus 265 blank controls to catch false alarms. The team timed the releases to overpasses by 12 government, commercial and non-profit satellites, per Nature. Fifteen analysis teams from nine countries generated more than 2,250 estimates, which the paper calls "the largest controlled test so far" to the authors' knowledge. Earth.com lists Carbon Mapper, NASA JPL, the United Nations' methane observatory and Google among the teams, alongside a string of commercial startups.

Who passed

Instruments built for the job did best. Nature says "built-for-purpose" methane instruments showed a 92–100% true positive rate and a 0–11% false positive rate, with quantification error of 44% mean absolute percentage error. Earth.com reports that Carbon Mapper detected every one of the 31 Tanager-1 releases that clouds didn't spoil and called no false plumes on its five blank passes. GHGSat caught 92% of its releases and reported one false positive.

Instruments repurposed for methane sensing landed in the middle, with detection rates of 62–88% and 46% mean absolute percentage error, according to Nature. The broad land scanners, Sentinel-2 and Landsat, caught just 29% to 37% of plumes. Earth.com adds a caveat: the test never released gas at the very large rates those two are expected to see. Nature's paper credits them with consistent global coverage and historical archives.

Under ideal conditions, with few clouds and moderate wind, Stanford says instruments reliably detected plumes down to about 50-200 kilograms per hour, depending on the system. That sits close to the 100 kilograms per hour that triggers reviews under some state and national super-emitter programs. Stanford also says most estimates were up to 50% higher or lower than actual emissions, with errors roughly balancing out on average. Adam Brandt, the Stanford professor who led the work, put the practical upshot plainly: "As long as you know approximately how big it is, then you still know you need to go fix it."

El Abbadi, a University of Iowa co-author who developed the release equipment, goes further. She said "satellites are ready for applications ranging from enforcing regulation to updating methane inventories."

Where it frays

The weak spots are specific. Ground-measured cloud cover above 20% degraded performance across systems, Nature says, raising concerns in some locations. Earth.com reports that a sizable portion of all submitted answers was thrown out because of clouds, in Arizona, a site chosen for its clear skies. The team named Nigeria, Venezuela, Canada and Russia among cloudy regions that matter most for methane. Lead author Frances Reuland told Earth.com the limitation is "a practical limitation in cloudier regions, not a reason for inaction."

Then the false alarms. Teams reported 71 false plumes in all, and 79% came from commercial teams running newer artificial-intelligence methods on Sentinel-2 and Landsat data, Earth.com reports. The paper's verdict on those methods: "caution is warranted". Orbio Earth made the most true detections from the land scanners, 58, and also the most false ones, 33. Kayrros made a third as many detections and no false ones.

Quantification can wobble even for good data. Earth.com says ten teams working from one EMIT image of a plume near 1,760 pounds (800 kilograms) an hour reported rates that differed by more than a factor of four.

What the desert could not test

One caveat sits in the design itself. The teams knew where the stack was. Nature's conclusion is that satellites can achieve accurate detection and quantification when the source location is approximately known. Earth.com notes that this is how a satellite checks a known site, not how most plumes are found.

The commercial side concedes the point. In a reply to a referee on EGUsphere, GHGSat's Antoine Ramier said the results apply to target-mode operations over known customer assets, as opposed to survey-mode background monitoring of large areas. The referee had raised concerns about extending the results to regions other than the controlled-release locations. Earth.com reports that in real-world surveys GHGSat detects only some methane plumes even at release rates around 1,000 pounds (460 kilograms) an hour, several times the threshold measured in controlled tests.

Reuland told Earth.com the most valuable next step would be a similarly large test in very different terrain, "ideally somewhere greener and more humid".

Brandt says the results should give regulators, producers and the public "Our results should give regulators, policymakers, producers of methane, and the general public a lot of confidence moving forward." For operators, that cuts both ways. A satellite that flags a known compressor station is now a credible witness. A satellite sweeping a cloudy basin for leaks nobody has located is still untested, and no one has yet run that blind.