Orphaned Wells Marquee

How to find and fix natural gas and oil wells that are abandoned and leaking

AI can help detect orphan wells across the country.

March 31, 2025

Download a print-friendly version of this article.

Oil wells and natural gas wells puncture the earth to extract energy resources from deep underground. In other words, they’re designed to make the ground leak. But if an oil well or natural gas well isn’t decommissioned properly, it may continue to leak even after it stops producing. 

“There are abandoned wells all over the country,” says Hari Viswanathan, an environmental scientist at Los Alamos. “They’re contaminating groundwater and releasing gases into the atmosphere.” Abandoned wells also affect property values and interfere with the utilization of surface and subsurface resources.

Oil and natural gas wells are plugged when no longer useful. But many wells—like the estimated hundreds of thousands drilled before the 1950s and “orphaned” by companies that no longer exist—can’t be plugged because they can’t be found (because their records have been lost or were never created). These abandoned wells are buried all over the country, and even the most basic information, such as where they are located, is often missing. Finding and plugging these orphan wells falls largely to individual states, which rely mainly on landowner complaints for prioritization.

Illustration of a drone collecting information from known oil and natural gas wells. The data appears to feed into a computer.
Undocumented oil and natural gas wells can be located using “well signatures” that Los Alamos scientists are developing with the help of machine learning (ML). Fixed-wing drones carrying multiple sensors—magnetometers to detect metal well casings, gas sensors to sniff out emissions, and LiDAR systems to reveal surface disruption—gather data about the chemical and physical attributes of confirmed well sites (grey markers). These data then go into an ML model that outputs a well signature, or set of attributes associated with wells, that can help identify hidden wells in other areas (red markers).

Now, a multi-laboratory research consortium known as CATALOG (Consortium Advancing Technology for Assessment of Lost Oil & Gas Wells), led by Viswanathan, is making it easier to find and fix orphan wells. Although locating and remediating these wells ultimately still falls to individual states, the consortium is using sensing techniques and artificial intelligence to make the process much more efficient and to prioritize the most problematic wells. CATALOG will not only lay out best practices, but it will create tools to help states maximize payoff from the time and money they spend.

One part of the project involves testing various detection methods on confirmed well sites to determine the best way to locate the most wells under a given set of conditions. The team uses multiple sensors: fixed-wing drones equipped with magnetometers to detect metal well casings, gas sensors to sniff out methane leaks, and LiDAR systems to reveal surface disruption. Machine learning is then used to merge and analyze the data from these sensors, which allows for the wells’ locations to be pinpointed with much greater accuracy than when using only a single sensor. It’s also important to test these methods on wells in different regions and under different conditions—no single detection method can be trusted to locate every well in any given area.

“If you want to find wells in a region like New Mexico or Oklahoma, you’d need to rely on completely different methods from what you’d use in a place like downtown Los Angeles,” explains Viswanathan. To develop region- or terrain-specific case studies, the chemical and physical signatures of confirmed well sites are used to build an AI model that can extract a well signature from noisy data—information that can later be used to search for hidden wells in other areas with similar conditions.

An illustration of a magnifying glass elucidating data from an old-fashioned file drawer.
Partially documented oil and natural gas wells can be added to databases with the help of an AI workflow that Los Alamos scientists are creating. Older wells are often recorded in historical documents that were never transferred to digital databases. Optical character recognition can quickly digitize historical documents, converting an image into digital lines of searchable text. Then, a large language model can extract important identifying information from this digital text and enter the well into a modern database, making quick work of an otherwise untenable task. 

There’s also a second category of orphan wells to address: those with partial documentation, but not enough to assess their environmental impact or determine appropriate remediation strategies. Many of these wells were recorded in historical documents that never made it into official databases. Sifting through these millions of documents would be overwhelming and highly inefficient for a human, but two tools, optical character recognition (OCR) and AI-driven large language models (LLMs), are perfect for the job. OCR is a digitization process that turns scanned images of historical documents into digital lines of searchable text. An LLM, an AI program trained to recognize and generate text, can then extract important information—such as a well’s depth, location, or age—from this digitized text. This information often factors into a well’s characterization when its environmental impact is assessed, and it’s an important step in the project’s efforts to improve oil and gas well databases. Because many of these documents come from wells that are missing from the databases, a good number of previously unknown wells can be located in this way.

It takes a lot of money and resources to plug even a single well, making triage a vital step. To address the biggest problems first, wells’ emissions are calculated from parts-per-million measurements in the air downwind of each well site. Then, the wells can be categorized based on these measurements, and the worst offenders are given the highest priority for remediation.

The project’s long-term impact doesn’t end when the last 1950s-era well is found and characterized; the CATALOG best practices will continue to be valuable moving forward. “As we transition toward sustainable fuels, more infrastructure will need to be decommissioned correctly,” Viswanathan says. “We don’t want to leave behind more leaky wells.”

People also ask

  • How does an oil well work? An oil well is a hole drilled into the earth’s crust to reach reservoirs of oil deep underground. The hole is reinforced with a casing, and a pump is sent down to harvest the oil and drive it to the surface, where it is collected and refined. The iconic slowly bobbing pump jacks that dot the land in oil-rich areas—also known as nodding donkeys, thirsty birds, or horseheads—are electricity-driven counterweights that help pull the oil out of the well.
  • What is natural gas? Natural gas is a naturally occurring mixture of gases, primarily methane, ethane, and propane. Its uses include heating, cooling, and cooking, as well as electricity generation and fuel for transportation. 

Share