27 August 2026

 

Methane has always presented polluters with a convenient defence: you cannot see it.

The greenhouse gas can escape from a faulty valve, pipeline, coal mine or landfill for days without attracting attention. Unless an operator detects the problem, reports it accurately and arranges a repair, the scale of the release may never be properly reflected in official emissions inventories.

That defence is becoming harder to sustain.

Earth observation satellites can now identify some of the world’s largest methane plumes from orbit. A recently completed European Space Agency project called AI4CH4 has gone further, combining observations from multiple satellite missions with artificial intelligence to automate methane plume detection and estimate emission rates.

It is a significant technical advance. Methane is responsible for nearly 30% of global warming since pre-industrial times, according to ESA, but remains in the atmosphere for much less time than carbon dioxide. Reducing large methane leaks can therefore produce relatively rapid climate benefits.

 

How do satellites detect methane leaks?

 

Methane detection does not work like conventional satellite photography. The gas is invisible to the human eye, but it absorbs sunlight at particular wavelengths. Satellite instruments can identify these spectral fingerprints in the light reflected from Earth’s surface and passing through a methane plume.

Some specialist satellites use hyperspectral imaging, dividing reflected light into hundreds of narrow wavelength bands. This makes it possible to distinguish methane from other atmospheric gases and estimate how much is present. Multispectral satellites such as Sentinel-2 collect fewer, broader bands but can still reveal sufficiently large plumes under favourable conditions.

The emerging satellite methane monitoring network combines instruments with different strengths.

The Copernicus Sentinel-5P satellite carries the TROPOMI instrument, which provides near-global daily coverage and is well suited to finding large emission hotspots. Its methane measurements have a spatial resolution measured in kilometres, however, making it difficult to connect a plume directly to an individual piece of infrastructure.

Higher-resolution observations from Sentinel-2, commercial satellites such as those operated by GHGSat and hyperspectral systems such as Carbon Mapper’s Tanager-1 can then provide a more detailed view. This tiered observation model uses broad scans to find suspicious areas and targeted or higher-resolution instruments to narrow the source down to a facility or smaller area.

The comparison is not exact. Sentinel-2 systematically images land surfaces but was not originally designed as a methane-monitoring mission. Satellites such as GHGSat and Tanager are purpose-built or optimised for detecting concentrated greenhouse-gas sources.

Ground sensors, aircraft and drones therefore remain important. They can detect smaller leaks, provide local meteorological information and verify what is happening at the site. Satellites are becoming the first layer of an integrated methane monitoring system rather than a complete replacement for measurements closer to the ground.

 

How AI4CH4 automates methane plume detection

 

Turning spectral measurements into a reliable emission estimate is difficult. Clouds, surface reflectance, terrain and atmospheric conditions can all affect the signal. Wind is particularly important because it determines the shape, direction and concentration of a plume.

Traditional approaches generally combine a measured methane enhancement with atmospheric transport models and wind data. Errors in local wind speed can translate into substantial uncertainty about how fast methane is escaping.

AI4CH4, led by Canadian research organisation C-CORE, was designed to reduce these bottlenecks. Its automated processing chain brings satellite-data ingestion, plume detection, emission quantification and visualisation into a single workflow.

The project explored convolutional neural networks, vision transformers and automated image-segmentation techniques. It also incorporated foundation-model technology, including Meta’s Segment Anything Model, alongside vision-language methods to delineate methane plumes from complex satellite scenes.

Training such systems creates another problem: there are not enough accurately labelled leaks for which both the satellite image and true emission rate are known. The researchers therefore constructed benchmark datasets combining real satellite observations with simulated methane-plume scenarios.

These synthetic examples allow AI models to encounter far more plume shapes and environmental conditions than a purely observational dataset could provide. They also help compensate for the shortage of verified industrial leak data available for training.

ESA says the models were validated using independent datasets and real-world case studies across several regions. The project also produced five peer-reviewed papers, three of which had been published by June 2026.

However, this does not mean wind has become irrelevant or that every methane release can now be quantified autonomously. ESA describes reduced dependence on ancillary data such as wind information as one of the project’s aims. Detection limits still depend on the instrument, cloud cover, surface characteristics, weather and the size and duration of the release.

AI can accelerate satellite-data analysis and reduce some of the uncertainty associated with conventional processing. Its results still carry uncertainty of their own, and that becomes especially important when satellite findings move from scientific monitoring into enforcement.

 

Detecting a plume is not the same as identifying a polluter

 

A methane plume does not arrive with a company name attached.

After detection comes attribution: establishing which facility or activity produced it. In a dense oil and gas basin, several pipelines, wells or processing installations may sit within or close to the apparent source area. Wind can carry methane away from its origin, while intermittent emissions may disappear before an investigator reaches the site.

High-resolution imagery and infrastructure maps can narrow the possibilities. Repeated satellite observations can reveal whether the release is persistent. Ground teams can then inspect equipment with handheld sensors, drones or vehicle-mounted instruments.

This creates a methane accountability chain:

1. Detection: A satellite or ground sensor identifies an elevated methane concentration.
2. Attribution: Analysts connect the plume to a probable facility or source.
3. Verification: Further satellite passes or ground measurements confirm the release.
4. Notification: The operator and relevant authorities receive the evidence.
5. Repair: The operator investigates and stops or reduces the emission.
6. Follow-up measurement: New observations determine whether the intervention worked.

Break any link and the plume may remain an interesting data point rather than a climate intervention.

 

Methane alerts still do not guarantee action

 

The UN Environment Programme’s Methane Alert and Response System shows both the potential and the weakness of this chain.

Announced at COP27 in 2022, MARS entered a pilot phase in January 2023 and became fully operational in January 2024. It now uses data from more than 30 satellite instruments, combined with scientific analysis and AI models, to identify large methane releases and notify governments and companies.

By June 2026, UNEP had recorded more than 40 verified mitigation cases across ten countries. In Algeria, for example, a MARS alert led to the repair of a leak that UNEP said had persisted for decades.

Yet only 13% of MARS alerts worldwide were receiving a response. UNEP defines a response as an alert being investigated and information being sent back. It does not necessarily mean that an emission source has been repaired.

Satellite methane detection has advanced faster than institutional willingness to use it.

 

Can regulators use satellite methane evidence?

 

Europe is beginning to connect satellite detection with regulatory action.

The EU Methane Regulation, adopted in 2024, requires stronger measurement, monitoring, reporting and verification in the energy sector. It also requires the European Commission to establish a global monitoring tool using satellite data and a rapid-alert mechanism for super-emitting events inside and outside the EU.

This gives satellite evidence a formal role in regulatory oversight, particularly as an early-warning and investigative tool. It does not automatically turn every detected plume into legally conclusive proof.

Regulators still need documented methodologies, uncertainty estimates and a defensible connection between the emission and the responsible operator. Repeat observations and measurements made on the ground may be needed before sanctions or other enforcement action can follow.

The same data could become increasingly relevant to investors and insurers. Satellite observations can expose discrepancies between corporate disclosures and measured emissions, indicate poorly maintained infrastructure or show whether promised repairs produced results.

But an observation must be interpreted in context. A single image might capture an accidental leak, operational venting or a plume originating beyond the apparent facility boundary.

Its greatest value may therefore lie in changing the burden of explanation. Instead of campaigners or regulators having to establish that an operator is under-reporting, a company confronted with repeated independent observations may have to explain why its figures do not match what satellites see.

 

Turning satellite methane data into repairs

 

A Spanish landfill study illustrates how the full accountability chain can work.

ESA and its partners combined Sentinel-5P observations with GHGSat measurements, aircraft surveys, ground measurements and site information at Madrid’s Las Dehesas landfill. Operators used the findings to target maintenance on gas-collection wells and pipelines, while later observations assessed whether those interventions had worked.

It succeeded because scientists, local government and the site operator collaborated. Technology supplied the evidence; governance converted it into action.

AI4CH4 can make the first stages faster, more scalable and less dependent on painstaking manual analysis. It can help turn vast satellite archives into a stream of potential methane alerts.

What it cannot do is compel a company to answer the telephone, send an engineer or replace a leaking valve.

Methane plumes are becoming much harder to hide. Inaction, unfortunately, remains entirely visible.