Scattered across the United States are remnants from almost 170 years of commercial drilling: hundreds of thousands of forgotten oil and gas wells. These undocumented orphaned wells (UOWs) are not listed in formal records, and they have no known (or financially solvent) operators. They are often out of sight and out of mind—a hazardous combination.
If the wells weren’t properly plugged, they can potentially leak oil and chemicals into nearby water sources or send toxic substances such as benzene and hydrogen sulfide into the air. They can also contribute to climate change by emitting the greenhouse gas methane, which is about 28 times as potent as carbon dioxide at trapping heat in our atmosphere on a hundred-year timescale (with even higher global warming potential over shorter periods).
To find UOWs and measure methane emissions in the field, researchers are using modern tools, including drones, laser imaging, and suites of sensors. But the contiguous United States covers more than 3 million square miles. To better predict where the undocumented wells might be, researchers first pair the new with the old: modern artificial intelligence (AI) and historical topographic maps (Fig. 1).
“While AI is a contemporary and rapidly evolving technology, it should not be exclusively associated with modern data sources,” said Fabio Ciulla, a postdoctoral fellow at the Department of Energy’s Lawrence Berkeley National Laboratory and lead author of a case study on using artificial intelligence to find UOWs published in the journal Environmental Science & Technology. “AI can enhance our understanding of the past by extracting information from historical data on a scale that was unattainable just a few years ago. The more we go into the future, the more you can also use the past.”
Since 2011, the United States Geological Survey has uploaded 190,000 scans of historical USGS topographic maps made between 1884 and 2006. Crucially, the maps are geotagged, meaning each pixel corresponds to coordinates that can be easily referenced.
Ciulla pulled together quadrangle maps, rectangular maps that cover a set amount of latitude and longitude and were mapped at a scale where 1 in. represents 2,000 ft. Between 1947 and 1992, these maps also used consistent symbols for oil and gas wells: a hollow black circle.
“For a human being, looking at this circle and recognizing it is extremely easy,” Ciulla said. “Until recently, this was the only available method to extract information from these maps. But that strategy does not scale well if we want to apply it to thousands of maps. This is where artificial intelligence comes into play.”
For this approach to work, the Berkeley Lab research team needed to teach the AI how to identify the correct symbols amid all the other visual information. It also needed to work on maps with different terrain and colors, as well as maps in different conditions (old, new, stained, pristine).
“This problem is equivalent to finding a needle in a haystack, since we are trying to find a few unknown wells that are scattered in the midst of many more documented wells,” said Charuleka Varadharajan, a scientist at Berkeley Lab and senior author of the study.
Researchers used a digital tool to manually mark oil wells on nearly 100 maps from California and create a training set for the AI. Once taught to find the hollow circles and to ignore false positives (such as cul-de-sacs or symbols with circular patterns, like the number 9 or letter “o”), the algorithm could be applied to any of the USGS maps with the same symbols. And because the maps were georeferenced, the algorithm could take the coordinates for the oil wells marked on the map and compare them with coordinates for documented wells.
They also built a novel tool that lets a human quickly vet what the algorithm finds, double checking that the AI is correctly interpreting the symbols on the map.
Researchers used the AI algorithm to scour four counties of interest that had substantial early oil production—Los Angeles and Kern counties in California and Osage and Oklahoma counties in Oklahoma—and found 1,301 potential undocumented orphaned wells. So far, researchers have verified 29 of the UOWs using satellite images and another 15 from surveys in the field; additional investigation on the ground will be needed to confirm other potential wells.
“With our method, we were conservative about what would be considered as a potential undocumented orphaned well,” Varadharajan said. “We intentionally chose to have more false negatives than false positives, since we wanted to be careful about the individual well locations identified through our approach. We think that the number of potential wells we’ve found is an underestimate, and we might find more wells with more refinement of our methods.”
From the Map to the Field
The first pass at verifying an undocumented well happens remotely. Researchers consult satellite images and historical aerial photos, looking for features such as oil derricks and pump jacks (or their shadows), lifting equipment, oil pads, storage tanks, or disturbed ground.
In many cases, wells were capped at or below the surface level, leaving no obvious sign in reference images. Instead, researchers need to head into the field with equipment to confirm whether a well exists.
At a predicted well location, researchers look for any surface well structures. If there aren’t any, they walk in a grid or spiral pattern carrying a magnetometer, which measures magnetic fields. Buried metal well casings disturb the magnetic field, allowing researchers to home in on the well. Once they finish surveying the area, researchers save the magnetometer file, record whether a well was found, and, if so, take a picture of the site, record GPS coordinates, and check for methane leaks.
For the wells they could verify, the Berkeley Lab team found the UOWs were located an average of 10 meters from where the algorithm and map predicted. They believe the AI approach is the first that can identify the precise locations for potential UOWs at county scales. And with the bounty of maps covering the United States, the technique can be scaled up and translated to other regions of interest.
The AI mapping and verification effort is part of a much larger project to address UOWs: the Consortium Advancing Technology for Assessment of Lost Oil and Gas Wells (CATALOG). The program is led by Los Alamos National Laboratory and includes research teams from Berkeley Lab, Lawrence Livermore National Laboratory, the National Energy Technology Laboratory, and Sandia National Laboratories.
It’s a big collaboration to address an equally sprawling problem: The Interstate Oil and Gas Compact Commission estimated in 2021 that there are somewhere between 310,000 and 800,000 undocumented orphaned wells across the United States.