Industries / Energy, water and waste / Pipelines and midstream
Finding leaks
Leak detection is how operators, the public and instruments notice that oil or gas is escaping, from smell to software that balances flow in and out, to satellites.
5
Milestones achieved, from 1937 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1937
A smell added to gas
On 18 March 1937 gas exploded under the school in New London, Texas, and about 298 of the roughly 540 students and teachers inside died. To save about $300 a month, the school had tapped a residue gas line belonging to a gasoline company; the gas had no smell, leaked under the building and was lit by a switch. The disaster led to a Texas law requiring that a strong-smelling chemical be added to gas used commercially and industrially, so people can smell leaks.
Lessonpeople cannot report a leak they cannot sense.
Credit
the Texas Legislature
1989
A pressure drop ignored
On 4 June 1989 a faulty pipeline carrying natural gas liquids leaked about a kilometer from a railway, about 50 km east of Ufa in the Soviet Union. Engineers had noticed a pressure drop and restored the pressure without checking for leaks. The gas cloud ignited as two passenger trains passed, killing 575 people, according to The Moscow Times.
Lessona falling pressure reading is a leak until shown otherwise.
Credit
the pipeline's operators, not named in the source; nine officials were charged
2019
Leak detection required by rule
A US rule of 1 October 2019 requires hazardous liquid pipelines carrying liquid only (not liquid mixed with gas), apart from offshore and rural gathering lines, to have an effective leak detection system, judged against the line's length, product, leak history and how fast leaks can be found. Any computational pipeline monitoring system, software that compares flows and pressures to spot a leak, must follow the American Petroleum Institute's practice RP 1130.
Credit
PHMSA
2022
Satellites see pipelines venting
Thomas Lauvaux of France's Laboratory of Climate and Environmental Sciences and colleagues used the European TROPOMI satellite instrument to count more than 1,800 very large methane releases from oil and gas sites in 2019 and 2020, published in Science. These accounted for 8 to 12% of the sector's methane emissions in that period, and most were deliberate venting from pipelines or equipment before repairs, as NPR reported.
Lessonmany of the largest releases are planned, so they can be avoided.
Credit
Thomas Lauvaux and colleagues
2025
Alerts sent, mostly unanswered
Since 2022 the UN Environment Programme's Methane Alert and Response System has sent more than 3,500 alerts about large methane releases in 33 countries, based on satellite data and AI-supported analysis. Governments and companies answered 12% in the year to October 2025, up from 1% the year before, and the programme has recorded 25 cases of action in ten countries.
Lessonfinding a leak does nothing until someone answers.
Credit
UNEP International Methane Emissions Observatory
Watch for
Aerial patrol with AI
Forsite says its PatrolView platform combines aircraft patrols with AI detection of digging near pipelines, hyperspectral cameras (which see many narrow bands of light) for liquid leaks, and methane plume detection, trained on more than 30 million images. It gives no detection rates.
Achieved whenresults from blind tests against controlled releases, run by an outside party, are published.
Claimed by
Forsite, July 2026
Open
01
A software leak test bench
A public set of simulated pipeline runs, with and without leaks of known size, place and time, plus scoring code. It must include shutdowns, restarts and column separation, which happens when a line is not full and the liquid column breaks, giving readings that can look like a leak, or hide one.
The briefCredit · unclaimed02
Tell blowdowns from leaks
For methane plumes seen over US gas transmission compressor stations, a public table that marks each plume as planned venting (a blowdown, when a section is emptied for work) that matches a posted maintenance notice, or as an unexplained release.
The briefCredit · unclaimed
Open 01 · the brief
A software leak test bench
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of simulated pipeline runs, with and without leaks of known size, place and time, plus scoring code. It must include shutdowns, restarts and column separation, which happens when a line is not full and the liquid column breaks, giving readings that can look like a leak, or hide one.
Why it is still open
Water networks have public leak benchmarks such as LeakDB, which comes with scoring code. This session found no equivalent public set for oil or fuel pipelines. At Marshall in 2010, controllers read rupture alarms as column separation for 17 hours.
Built with what exists
TSNet, an open-source Python package for simulating pressure surges, adapted to oil properties; Python; a coding agent to write the scenarios.
First step
Model one line with a pump, a valve and a rise in elevation, and reproduce column separation on shutdown.
Done looks like
At least 1,000 labelled runs, 100 of them shutdown and restart cases with column separation and no leak; a simple flow-balance method scored on time to detect, smallest leak found and false alarms; a second group reruns the scores to within 5%.
Open 02 · the brief
Tell blowdowns from leaks
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
For methane plumes seen over US gas transmission compressor stations, a public table that marks each plume as planned venting (a blowdown, when a section is emptied for work) that matches a posted maintenance notice, or as an unexplained release.
Why it is still open
Carbon Mapper, a nonprofit that publishes plume data, says short bursts can come from planned maintenance and pipeline blowdowns. UNEP says nearly 90% of its satellite alerts go unanswered. No public study matching plumes to posted notices was found.
Built with what exists
Carbon Mapper's public plume data; HIFLD station locations; pipeline companies' public notice boards; a web agent to collect notices (Note: "Deleting 80% to start over").
First step
Take 20 plumes over one company's stations and look for notices posted within two days of each.
Done looks like
For every public plume over an interstate compressor station in one year, a label of matched or unmatched within 48 hours, with the notice linked; a second person checks 50 matches.