For years, methane satellites have produced images that look like evidence: bright plumes over oil fields, pipelines and industrial sites. The harder question has been whether those images can support a defensible number.
A new Nature study offers the clearest answer yet. Under controlled conditions, purpose-built methane instruments detected most large releases and estimated their size with useful accuracy. The same experiment also showed why a methane plume is not a verdict. Instruments performed differently, clouds degraded results and automated methods generated most of the false alarms.
The satellites are becoming credible measuring instruments. They are not omniscient witnesses.
A test with the answer hidden from the analysts
The year-long experiment took place in Arizona, where researchers released measured quantities of methane during satellite overpasses. The study included 746 events: 481 controlled releases ranging from 9.5 to 1,183.7 kilograms per hour, along with 265 blank controls in which no methane was released.
Twelve government, commercial and nonprofit satellite systems participated. Fifteen analysis teams from nine countries produced more than 2,250 estimates. The teams were not given complete information about when or where releases occurred, creating a single-blind test rather than a retrospective comparison in which analysts already knew the answer.
That design is the study’s central strength. Researchers could compare three separate questions: whether a system detected a release, whether it produced a false alarm and how closely its estimate matched the metered flow on the ground. Those questions are often blurred when satellite observations are compared with industry inventories that may themselves contain assumptions and omissions.
Some instruments performed very well
The results were encouraging, but not uniform.
Purpose-built methane instruments detected between 92 and 100 percent of the controlled releases. Their false-positive rates ranged from zero to 11 percent, and their average absolute percentage error in quantifying emissions was approximately 44 percent. Reported regression slopes ranged from 0.83 to 1.00, indicating useful agreement between estimated and actual release rates under the conditions tested.
Repurposed instruments were more variable. Their detection rates ranged from 62 to 88 percent, with approximately 46 percent mean absolute percentage error. Broad land-imaging scanners performed poorly at the release rates tested, although their global coverage and historical archives may make them valuable for screening and long-term analysis.
“Methane satellite” is therefore not a single technology. Instruments differ in spatial resolution, spectral design, revisit rate, coverage and intended use. A system designed to identify a concentrated point source is solving a different problem from one designed to survey large areas repeatedly.
Detection is not the same as quantification
The most important distinction for readers is between seeing methane and measuring how much methane is being emitted.
A plume’s visible shape depends on more than the source itself. Wind speed and direction, atmospheric turbulence, cloud cover, surface reflectance, instrument resolution, plume geometry, retrieval algorithms and the duration of the release all affect the signal.
A satellite observation may therefore provide strong evidence that methane was present while providing weaker evidence about the exact emission rate. The National Institute of Standards and Technology’s guidance on methane-plume quantification treats detection and estimation as related but separate methodological tasks. It emphasizes metadata, assumptions, validation and quality information rather than treating a plume image as self-explanatory.
This distinction becomes consequential when a measurement enters an enforcement case, a corporate disclosure, a carbon-market claim or a public dispute over industrial activity. “Detected,” “quantified,” “attributed” and “verified” are different claims. They should not collapse into one another merely because they appear in the same report.
The atmosphere and the ground still matter
The Arizona site was useful precisely because it offered a controlled and relatively favorable testing environment. The researchers could schedule releases around satellite overpasses, measure the flow on the ground and compare results across repeated observations.
But a desert test is not a guarantee of performance over every landscape where methane matters. Oil and gas fields, landfills, livestock operations, coal mines, pipelines and urban distribution networks present different surfaces, source patterns and atmospheric conditions. Forests, wetlands, snow-covered ground and highly cloudy regions may pose additional difficulties.
The study found that ground-measured cloud cover above 20 percent degraded performance across systems. That is not a minor technical footnote. A monitoring system that works well only when the atmosphere and surface cooperate will produce an uneven record of emissions, with some regions observed more successfully than others.
Nor does a successful point-source measurement automatically translate to diffuse or intermittent emissions. A landfill, livestock operation or large production field may not behave like a single metered release from a stack. The measurement strategy must match the physical source.
Artificial intelligence adds capability—and another error pathway
One of the study’s more consequential findings concerned automated analysis. Artificial-intelligence-based methods accounted for 79 percent of the false positives in the experiment.
That does not mean machine learning is useless, or that human analysts are automatically more reliable. Automated systems can scan large areas and identify candidate plumes at a scale that manual review cannot match. A false-positive rate may be acceptable for a screening system if the results are clearly labeled, independently reviewed and followed by confirmation.
The problem arises when a preliminary flag is treated as a confirmed emission. Users need to know how often a system is wrong, what data it used, how its confidence was calculated and whether an analyst can inspect the underlying observation. Operators also need to preserve the original data and the decision path that turned an image into a reported number.
That is a measurement-governance problem as much as a software problem. A system can be useful while imperfect, but only if its imperfections remain visible.
What the study does—and does not—establish
The experiment establishes that large, concentrated methane sources can be detected from space. It shows that purpose-built instruments can perform well under defined conditions and that quantification is possible without being perfectly precise. It also shows that false positives vary by instrument and analytical method, clouds matter and independent blind testing can reveal weaknesses that promotional demonstrations may not.
It does not establish that satellites can measure all methane emissions, that every detected plume represents a sustained leak or that an observation alone identifies the responsible company or piece of equipment. It does not establish equal accuracy for diffuse, intermittent or low-rate sources, nor does it show that satellite data by themselves are sufficient for regulatory penalties or a complete national emissions inventory.
Responsibility still requires interpretation. Someone must determine whether repeated observations refer to the same source, how weather affects confidence, how a plume is assigned to a facility and what corroborating evidence is required before action is taken.
A new measurement system, not a final answer
Methane accounted for approximately 12 percent of U.S. greenhouse-gas emissions from human activities in 2022, according to the Environmental Protection Agency. Major sources include oil and gas systems, livestock and landfills. Because methane traps substantially more heat than carbon dioxide over a 100-year period while remaining in the atmosphere for a shorter time, reducing large releases is important to near-term climate policy.
Those stakes make reliable observation valuable. They also make overstatement dangerous. A false positive can wrongly accuse an operator. An undetected release can create false confidence. An imprecise estimate can acquire an appearance of precision once it is copied into a regulatory filing, financial report or news story.
The most defensible conclusion is therefore neither that satellites have solved methane monitoring nor that their limitations make the technology untrustworthy. The evidence shows that some systems can detect and estimate large point-source releases with useful accuracy when the source location is approximately known and conditions are favorable.
The next advance will be less dramatic than another striking plume image. It will be agreement among independent satellite, aircraft, ground and engineering measurements across different terrains, weather conditions and source types. That is how a promising observation becomes a trusted measurement.
Satellite methane monitoring has passed an important test. The harder work is determining where its numbers remain strong, where they need qualification and how decision-makers should act when the atmosphere cannot provide a perfectly clear answer.
