Satellites pass a year-long blind test of methane leak detection

Methane is the stealth bomber of greenhouse gases. It traps heat far more effectively than carbon dioxide over its first two decades in the atmosphere, and much of it escapes silently from oil and gas infrastructure, landfills, and agricultural operations before anyone on the ground notices. For the past several years, a growing fleet of satellites has promised to catch these leaks from orbit, flagging plumes of the invisible gas as they drift across the landscape. But a pressing question has lingered behind the marketing materials and the dazzling plume imagery: how well do these orbital sentinels actually perform when nobody knows the answer in advance?

A team led by researchers at Stanford University, the University of Michigan, the University of California-Berkeley, and the University of Iowa has now delivered the most comprehensive answer to date. In a year-long, single-blind experiment conducted in Arizona, the researchers released precisely metered quantities of methane into the open air and challenged twelve satellites, operated by governments, commercial companies, and non-profit organizations, to detect and quantify the emissions. Fifteen independent analysis teams from nine countries submitted more than 2,250 estimates without ever knowing the true release rates. The results, published in Nature, paint a picture of a technology that is genuinely powerful but far from uniform in quality.

The scale of the test is unprecedented. In total, the experiment comprised 746 events: 481 controlled methane releases ranging from 9.5 to 1,183.7 kilograms per hour, interspersed with 265 blank controls in which no gas was released at all. The blanks are the crucial acid test of any detection system, because they reveal how often an instrument or algorithm cries wolf. Every release and every blank was timed to coincide with overpasses of the participating satellites, ensuring that the orbital instruments had a genuine opportunity to observe each event. According to the authors, this is the largest controlled test of methane-sensing satellites conducted so far, and one of the first evaluations of three newer instruments: EMIT, a NASA imaging spectrometer originally designed to study mineral dust; Tanager-1, operated by the non-profit Carbon Mapper; and GHOSt, a hyperspectral observation satellite.

The headline finding is that purpose-built instruments excel. Satellites designed specifically to hunt methane, such as those in the GHGSat constellation, achieved true positive detection rates of 92 to 100 percent while keeping false positives between 0 and 11 percent. Their quantification accuracy was equally impressive: when the teams’ estimated emission rates were compared against the metered values, the regression slopes fell between 0.83 and 1.00, meaning the estimates tracked the true releases almost linearly, with a mean absolute percentage error of 44 percent. For a technology measuring a gas plume from hundreds of kilometers above the Earth, that level of agreement is a remarkable achievement, and it validates the use of these systems for holding individual facilities accountable for large leaks.

Instruments that were repurposed for methane sensing, including hyperspectral mappers built for other scientific missions, performed more variably. Their detection rates ranged from 62 to 88 percent at the tested release rates, with a mean absolute percentage error of 46 percent. Many of these systems nonetheless demonstrated significant skill, which matters because repurposed instruments vastly expand the world’s collective methane-monitoring capacity without requiring new spacecraft. The weakest performers were broad land-imaging scanners, satellites such as Sentinel-2 and Landsat 8 that were designed for general surface observation and only retrospectively adapted for methane plume detection. At the release rates tested in Arizona, these instruments detected the plumes poorly, although the authors note that their consistent global coverage and deep historical archives continue to motivate work to improve their algorithms.

One of the most striking results concerns artificial intelligence. Machine-learning methods are increasingly being applied to satellite imagery to automate plume detection and quantification, and several participating teams employed them. The study found genuine promise in these approaches, but also a sobering caution: 79 percent of the false positives recorded in the experiment emerged from AI-based methods. In other words, when a satellite team reported a methane plume that turned out to be nothing, it was disproportionately likely to be an algorithm rather than a human analyst making the call. For a field in which satellite alerts can trigger regulatory scrutiny or reputational damage for facility operators, that statistic is a warning that automated detection must be validated as rigorously as the instruments themselves.

Clouds emerged as another fundamental constraint. When ground-measured cloud cover exceeded 20 percent, performance degraded across all the tested systems. The physics is unforgiving: these instruments detect methane by measuring absorption of sunlight reflected off the Earth’s surface, and clouds interrupt that light path entirely. The authors highlight this as a particular concern for tropical regions, where persistent cloudiness means that some of the world’s most methane-intensive landscapes remain poorly observed. This blind spot matters for global methane accounting, because estimates built from satellite surveys may systematically underrepresent emissions from precisely the wet, cloudy regions where wetlands, landfills, and oil and gas operations overlap.

The experiment’s single-blind design deserves attention on its own terms. In previous validation efforts, some of the same researchers had run smaller single-blind tests of nine satellite systems, but controlled evaluation of this scope, duration, and geographic consistency had never been attempted. By keeping the release schedule, rates, and even the existence of blank events hidden from the analysis teams, the experimenters eliminated the optimistic biases that can creep in when operators know they are being tested. The metered flow rates, meteorological data, and cloud imagery from all experimental phases have been released in a public repository, along with the code needed to reproduce the paper’s figures, giving the broader community a durable benchmark against which future instruments and algorithms can be measured.

The timing of this work is significant. More than 150 countries have pledged under the Global Methane Pledge to cut methane emissions 30 percent by 2030 relative to 2020 levels, and regulators in the European Union and the United States are moving toward rules that treat satellite detections as evidence of leaking infrastructure. The UN Environment Programme’s Methane Alert and Response System already notifies governments and companies of plumes spotted from orbit, and California’s air resources board is building a satellite-based methane monitoring program of its own. Those systems will only carry the credibility their users demand if their accuracy is known, and this experiment provides exactly that foundation.

The bottom line is quietly optimistic but clear-eyed. When the approximate location of a source is known, satellite-based measurements can achieve accurate detection and quantification of methane point sources, according to the study’s authors, and the best purpose-built instruments now detect large releases nearly every time they look. Yet the technology is not a magic eye in the sky: repurposed and general-purpose scanners miss substantial fractions of emissions, AI methods over-trigger, and clouds still hide a meaningful slice of the planet. For journalists, investors, and regulators alike, the lesson of the year-long blind test is that the satellite methane revolution is real, but every detection deserves the same scrutiny the Stanford team brought to the Arizona desert.

Subject of Research: Blind controlled testing of satellite-based methane detection and quantification systems

Article Title: Evaluating space-based methane detection through a one-year blind test

Article References: Reuland, F. E., Adams, T., Sherwin, E. D., El Abbadi, S. H., Kort, E. A., & Brandt, A. R. (2026). Evaluating space-based methane detection through a one-year blind test. Nature, 658(8136), 673-681. https://doi.org/10.1038/s41586-026-11115-0

Image Credits: AI Generated

DOI: 10.1038/s41586-026-11115-0

Keywords: methane, satellite remote sensing, greenhouse gases, controlled release experiment, EMIT, Tanager-1, GHOSt, GHGSat, artificial intelligence, cloud cover, climate mitigation, single-blind validation