Scattered across the United States are hundreds of thousands of oil and gas wells that exist in no official record — drilled as far back as the 1850s, never properly sealed, and quietly leaking methane, benzene, and other hazardous substances into soil, water, and air. Estimates suggest there may be between 310,000 and 800,000 of these undocumented orphaned wells nationwide.
Finding them has long been the central challenge. But researchers may have unlocked an unlikely key: paper topographic maps drawn more than a century ago — and an AI trained to read them.
A hidden hazard hiding in plain sight
Undocumented orphaned wells — or UOWs — share a few defining characteristics: no formal records, no known or financially solvent operator, and no guarantee they were ever properly sealed. Many were simply left open or filled with improvised plugs before drilling regulations existed. That combination creates a slow-moving environmental problem, easy to overlook precisely because it stays invisible.
The hazards are real. Improperly plugged wells can allow oil and chemicals to leach into nearby water sources, while above ground they release toxic substances — benzene, hydrogen sulfide — into the air. They also emit methane, a greenhouse gas roughly 28 times more potent than carbon dioxide over a 100-year period. The Interstate Oil and Gas Compact Commission estimated in 2021 that between 310,000 and 800,000 undocumented orphaned wells exist across the United States. The contiguous U.S. covers more than three million square miles, and without knowing where to look, finding these wells has historically been a matter of luck.
Old maps, new intelligence
The breakthrough came from an unlikely archive. Since 2011, the U.S. Geological Survey has digitized approximately 190,000 historical topographic maps produced between 1884 and 2006. Every map is georeferenced — each pixel corresponds to real-world coordinates that can be cross-checked against other data sources.
A specific window within that archive proved especially useful. Between 1947 and 1992, USGS quadrangle maps used a consistent symbol to mark oil and gas wells: a hollow black circle. Simple, standardized, machine-readable.
“For a human being, looking at this circle and recognizing it is extremely easy,” said Fabio Ciulla, a postdoctoral fellow at Lawrence Berkeley National Laboratory and lead author of the study published in Environmental Science & Technology. The problem, he noted, is that the manual approach does not scale across thousands of maps. Berkeley Lab researchers trained an AI to recognize these symbols across maps with varying terrain, print quality, and age — filtering out false positives like cul-de-sacs or characters such as “9” or “o.”
From algorithm to ground truth: 1,301 candidates found
Researchers tested the algorithm on four counties with substantial early oil production histories: Los Angeles and Kern counties in California, and Osage and Oklahoma counties in Oklahoma. The AI identified 1,301 potential undocumented orphaned wells across those areas.
Verification followed in stages. Twenty-nine candidates were confirmed using satellite imagery and historical aerial photographs; another 15 were verified through physical field surveys. The team deliberately designed the algorithm to favor false negatives over false positives, preserving the credibility of every flagged result. Verified wells were located an average of just 10 meters from where the algorithm predicted — precision that surprised even the researchers. “We think that the number of potential wells we’ve found is an underestimate,” said senior author Charuleka Varadharajan, a scientist at Berkeley Lab.
Into the field: magnetometers, drones, and a five-minute methane test
Once a candidate location is identified, verification moves from screen to soil. Researchers walk grid or spiral patterns over predicted sites carrying magnetometers, which detect disturbances in local magnetic fields caused by buried metal well casings. Drones extend that capability further — pre-programmed with flight routes, they carry methane sensors, hyperspectral cameras, and magnetometers suspended on a nine-foot cable to avoid interference from the drone’s own electronics.
Speed matters too. Berkeley Lab scientist Sebastien Biraud is developing a low-cost sensor array that can estimate a well’s methane leak rate in under five minutes. “We need to know if it’s not leaking, if it’s leaking between 10 and 100 grams per hour, or if it’s leaking kilograms per hour,” Biraud said. Rapid measurement is now a regulatory requirement both before and after plugging a well.
A multi-layer mission: the CATALOG consortium
The Berkeley Lab mapping work sits within a larger initiative called CATALOG — the Consortium Advancing Technology for Assessment of Lost Oil and Gas Wells. Led by Los Alamos National Laboratory, the consortium brings together five national laboratories, each contributing different tools and expertise toward a shared goal.
One key proving ground is the Osage Nation, whose territory spans nearly 1.5 million acres. Tribal partners provide direct feedback on field equipment and data accuracy. “Utilizing AI and state-of-the-art detection equipment has filled data gaps in records,” said Craig Walker, director of Osage Nation Natural Resources.
The broader ambition is a “multi-layer cake” approach — stacking historical maps, satellite data, production records, and drone surveys into a unified picture of where wells are and how urgently they need attention. Hundreds of thousands of wells still await discovery, and each one confirmed is a step closer to being properly plugged.
Carlos is an engineer with strong expertise in technical and industrial topics. He previously worked at international companies such as Siemens and speaks Spanish, German, English, and Italian.
