Researchers Released Methane 481 Times. Could Satellites See It?

At a desert test site in Arizona, researchers deliberately let measured quantities of methane escape. Sometimes they left the gas off. The satellite analysts looking down did not get the answer sheet.

Over 2025, the experiment accumulated 481 methane-release events and 265 blank controls. Fifteen analysis teams tried to tell whether gas was present and, when it was, how much was flowing. The results, published in Nature on October 14, give methane monitoring a demanding reality check: several systems performed very well, but the conditions behind their success matter as much as the headline percentage.

This was a single-blind test. Analysts knew the approximate source location, but not whether methane had been released or at what rate. It tested the ability to inspect a known place, not to discover every unknown leak on Earth.

A real plume, or an empty scene?

The empty trials are what make the experiment more revealing than a collection of impressive satellite pictures. A system that always announces a leak will find every real one. It will also send people chasing gas that is not there.

The study's published source data let readers inspect both sides. Carbon Mapper's analysis of Tanager-1 detected all 31 release events retained for analysis and flagged none of five retained blank trials. GHGSat detected 47 of 51 retained releases and falsely flagged one of nine blanks.

Those are encouraging results from systems designed for methane sensing. They are not a clean league table. The instruments observed different samples, teams made different filtering decisions, and Tanager generally used a high-sensitivity observing mode. Five empty trials are also a small basis for promising that false alarms will never happen.

The missing observations change the story

Why keep saying “retained”? Because an unusable image and a successful inspection are different outcomes.

The study supplement allowed analysis teams to exclude observations, most commonly because of clouds. In the GHGSat source-data row, 42 of 107 observation opportunities were marked cloudy; another five were listed as not tasked or without imagery. Those opportunities are not included in the 47-out-of-51 detection result. Nor were all 107 opportunities gas releases: the total also includes blanks.

For someone monitoring a facility, that distinction is practical. “The sensor usually finds gas in the images it can use” is a different promise from “the sensor checks this site successfully whenever it passes.” A clouded-out visit cannot establish that a source is clean.

The location matters, too. Reviewers challenged how far results from a known desert site could travel. In the published review exchange, the authors narrowed their conclusion to approximately known sources and called for tests across other terrain, weather and offshore settings.

Seeing gas is not the same as measuring its flow

A methane signal in an image answers one question. Estimating kilograms escaping each hour asks another: how much gas is present, and how quickly is the wind carrying it away?

For the purpose-built systems, the paper reports a mean absolute percentage error of about 44% in emission-rate estimates. That describes the average relative size of errors in this test, not a guaranteed margin around each reading. That average does not tell you how far any one estimate missed.

This is also why the result should not be pasted onto Google's separate methane-plume catalog. A catalog's confidence labels, an instrument's performance at a controlled site and a particular facility's actual emissions answer different questions.

The experiment provides strong grounds for targeted monitoring, not proof that leaks were repaired or emissions fell. Those later steps belong to the follow-through after a detection. Its most useful lesson is how to read a blank map: no marked plume is not the same as a usable observation showing no gas.

Study disclosures: Funders included Saudi Aramco and Stanford's industry-supported Natural Gas Initiative, alongside the Global Methane Hub and Grantham Foundation. Co-authors Eric Kort and Adam Brandt hold roles on Carbon Mapper's Science and Measurement Committee. The study was first posted as a preprint in April 2026; the October publication is its peer-reviewed journal version, not a new round of releases.