In the continental United States, evidence of our extensive oil consumption is abundant. Over the years, approximately 3.5 million oil and gas wells have been drilled across the country since the 1850s. Many of these wells were abandoned when the companies operating them went out of business or ceased operations. These neglected wells, known as “undocumented orphan wells” (UOWs), pose a significant environmental risk as they often remain unplugged, allowing dangerous substances like methane, oil, and chemicals to leak into the air and potentially contaminate nearby water sources. The Bureau of Land Management estimates that there are still around 130,000 unplugged orphan wells in the US, while industry organizations like the Interstate Oil and Gas Compact Commission suggest that the number could be as high as 740,000.
The task of locating and sealing these orphan wells is a challenging and time-consuming process. However, researchers from the US Department of Energy’s Berkeley Lab have developed a new artificial intelligence (AI) model trained on decades worth of maps that can identify previously undiscovered orphan wells within a 10-meter radius of their actual locations. The AI model has already identified 44 of the 1,301 potential wells in California and Oklahoma. Scaling up this AI-driven approach could significantly accelerate the process of decommissioning these dormant wells.
AI model was trained on thousands of topographical maps
Researchers detailed their methodology for training the AI in a recent article published in the journal Environmental Science & Technology. The AI model was specifically trained to recognize a symbol resembling a hollow black circle commonly used to indicate oil and gas wells on topographical maps. After manually identifying several examples of these symbols, the AI model was trained using this dataset. The researchers had to account for other circular symbols on the maps that could potentially lead to false positives. This training process was likened to “finding a needle in a haystack” by Berkeley Lab scientist Charuleka Varadharajan.
Once the AI was trained to recognize the well symbols, the researchers applied it to thousands of maps in four oil-rich counties in California and Nevada. The AI identified 1,301 potentially undocumented orphan wells, 29 of which were confirmed using aerial and satellite images from Google Earth. In cases where the wells were not visible from above, researchers conducted field tests using magnetometers to detect buried metal pipes, confirming an additional 15 wells.
However, not all abandoned wells are visible through aerial imagery, as many are located below the surface. In such cases, researchers conduct field tests using magnetometers to detect magnetic anomalies indicating buried metal pipes. Using this method, 15 more wells were verified by the researchers.
The researchers intentionally prioritized minimizing false negatives over false positives to ensure accuracy in their findings. They believe that the number of potential wells identified is underestimated and that further refinement of their methods could lead to the discovery of more wells.
AI predictions can work in tandem with well-detecting drones
The researchers aim to combine the predictive capabilities of AI with modern technologies like sensor-equipped drones to expedite the detection and sealing of potentially hazardous wells. Drones equipped with magnetometers could be deployed to areas where aerial detection is not feasible, while drones with methane sensors could detect air leakage. Additionally, drones with hyperspectral cameras could scan for methane plumes undetectable by the human eye.
Lawrence Berkeley National Laboratory postdoctoral fellow Fabio Ciulla stated, “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.”
Win the Holidays with PopSci’s Gift Guides
