Identifying Orphaned Oil and Gas Wells Using Deep Learning
Author Information
Author(s): Fabio Ciulla, Andre Santos, Jordan Preston, Timothy Kneafsey, Sebastien C. Biraud, Charuleka Varadharajan
Primary Institution: Lawrence Berkeley National Laboratory
Hypothesis
Can a deep learning framework effectively identify undocumented orphaned oil and gas wells from historical maps?
Conclusion
The study successfully identified 1301 potential undocumented orphaned wells across four counties in California and Oklahoma using a deep learning framework.
Supporting Evidence
- 1301 potential undocumented orphaned wells were identified across more than 40,000 km2.
- The U-Net model achieved a precision of 0.98 and a recall of 0.88 in detecting well symbols.
- 29 UOWs were confirmed from satellite images and 15 from magnetic surveys.
- The method can be scaled to identify potential UOWs across the entire US.
Takeaway
Researchers used a computer program to find hidden oil wells on old maps, discovering over a thousand potential sites that might leak harmful gases.
Methodology
A U-Net deep learning model was trained on historical topographic maps to detect oil and gas well symbols, identifying potential undocumented orphaned wells based on their proximity to documented wells.
Potential Biases
Human errors in identifying well symbols on maps could lead to misclassification.
Limitations
The study may underestimate the number of orphaned wells due to potential errors in documented well locations and the limitations of historical maps.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website