AI Model Identifies Forgotten Oil and Gas Wells Across the U.S.
Felix Pinkston Jan 19, 2025 03:50
Researchers at Lawrence Berkeley National Laboratory have developed an AI model to locate potentially hazardous, undocumented orphaned wells (UOWs) to prevent environmental damage from leaking chemicals.
Researchers from the Lawrence Berkeley National Laboratory (LBNL) have developed an artificial intelligence (AI) model capable of identifying forgotten oil and gas wells across the United States. This development could play a crucial role in mitigating environmental risks associated with leaking wells, according to NVIDIA.
Challenges of Undocumented Orphaned Wells
The U.S. is home to an estimated 800,000 undocumented orphaned wells (UOWs), which are not listed in official records and lack a responsible legal entity for their maintenance. Many of these wells, drilled since the mid-1800s, pose a risk as they may leak toxic chemicals and greenhouse gases, such as methane, into the environment.
AI Model and Its Development
The team at LBNL trained a vision-language model known as U-Net on digitized maps of the U.S. from 1947 to 1992. These maps, aggregated and digitized by the U.S. Geological Survey, are consistent in their symbols and georeferencing, allowing for precise location identification of wellheads.
Fabio Ciulla, the lead author of the study, noted that these historical topographic maps enabled the researchers to investigate UOWs on a continental scale, a feat not previously accomplished.
Implementation and Testing
Utilizing the National Energy Research Scientific Computing Center (NERSC) supercomputer, powered by NVIDIA A100 Tensor Core GPUs, the researchers trained their model on maps of Los Angeles and Kern counties in California, historically significant for oil and gas production. The model's accuracy in identifying UOWs was tested through satellite imagery and on-site visits, showing a detection accuracy varying between 31% and 98%.
The AI model demonstrated strong transferability by successfully identifying potential UOWs in Oklahoma’s Osage and Oklahoma counties, despite not being specifically trained on maps from those areas.
Future Prospects
This study is part of a Department of Energy initiative aimed at assisting states in locating UOWs. Researchers plan to refine the AI model further, expanding its application to additional regions and collaborating with state authorities interested in adopting this technology to manage environmental risks associated with orphaned wells.
Image source: Shutterstock