Worldwide, there are numerous abandoned and undocumented oil and gas wells, posing a significant environmental threat and impacting health. Artificial Intelligence (AI) is proving instrumental in their detection

AI is invaluable in pinpointing these abandoned oil wells, which are major environmental and health hazards. The issue is so acute in the United States that the Biden Administration has earmarked $4.7 billion through the Bipartisan Infrastructure Law to catalogue and systematically seal these wells. These wells have the potential to pollute water supplies, disrupt ecosystems, and emit methane along with other atmospheric contaminants, including carcinogens, adversely affecting surrounding areas and their inhabitants. It is estimated that around 14 million people reside within a mile of 80,000 documented wells.

The primary challenge lies in locating these wells, particularly the undocumented ‘orphan’ wells which are abundant. Various AI techniques are being employed to locate them with high precision, which is vital given the current lack of clarity. The Interstate Oil and Gas Compact Commission has estimated that the number of these abandoned wells in the United States alone is between 310,000 and 800,000, a figure which suggests limited understanding. These estimates are conservative, especially compared to the U.S. Environmental Protection Agency’s count of over 3.2 million. The agency also estimates that methane emissions from more than 2 million inactive and unconnected wells, which include the documented orphan wells, range between 7 and 20 million tonnes of CO2 equivalent annually.

---

TAKEAWAYS

---

Abandoned Oil Wells: The Reasons Behind the ‘Orphan’ Label and Their Numbers

Abandoned oil wells are often referred to as ‘orphans’ because they typically lack a documented or living owner, a situation compounded by the oil and gas sector’s 170-year history. The extent of the problem is considerable, particularly in the United States, where their numbers range from several hundred thousand to multiple millions. Research by the Environmental Defense Fund and McGill University, analyzing 80,000 documented orphan wells across 30 states and utilizing census data, revealed that approximately 14 million people live within a mile of these wells.

The federal government’s commitment to addressing this long-neglected issue signifies a historic shift, potentially leading to profound impacts. Allocating billions of dollars to tackle the country’s most problematic wells could significantly reduce toxic substances, such as arsenic, contaminating groundwater sources.

Addressing the issue of abandoned oil and gas wells is vital for other reasons too. A study published in the ACS Omega journal identified the release of volatile organic compounds (VOCs) from these wells, including benzene, a recognized carcinogen.

Nevertheless, some experts, like Mary Kang, a McGill University professor and the lead author of an article published in Environmental Research Letters, co-authored by twenty other scientists, argue that the $4.7 billion might be insufficient. They estimate that:

“the cost of plugging the documented orphan wells in the United States could exceed this figure by 30-80%, if not more”.

AI, Edge Computing, and Drones: Pinpointing Orphan Wells with Precision

The 2021 Bipartisan Infrastructure Law set aside $30 million to support the formation of a research consortium dedicated to developing technologies and best practices for identifying orphan oil wells, measuring their methane emissions, and prioritizing their closure.

This funding recently enabled a U.S. Department of Energy research team to embark on investigative work, employing drones, electromagnetic field detectors, and other remote sensing technologies to search for ‘ghost’ oil wells on public lands, stretching from Pennsylvania to Oklahoma. The use of AI techniques was likely integral to managing, analysing, and making informed decisions based on the amassed data.

A pivotal role for AI was established in a recent innovation by a team at Los Alamos National Laboratory. They devised a method to reconstruct extensive datasets from a limited number of sensors deployable in the field, utilizing edge computing.

The team developed a neural network that efficiently represents a large system with less computational demand than the latest convolutional neural network architectures. This makes it apt for field deployment on drones, sensors, and other edge computing applications that bring computational power closer to the point of use.

Featured in a Nature Machine Intelligence article, their research introduced a novel AI technique named Senseiver, grounded in the Perceiver IO AI model developed by Google. This model applies natural language processing techniques, similar to those in ChatGPT, to reconstruct information across vast areas like oceans, based on relatively few measurements.

The model assimilates various measurements taken over decades by satellites and ship sensors. From these scattered data, it predicts ocean temperatures, offering valuable insights for global climate models.

The potential of Senseiver in locating orphan wells is particularly notable. It’s considered suitable for a range of projects and research areas relevant to Los Alamos, especially the lab’s efforts in identifying and characterizing orphan wells. The lab leads the Consortium Advancing Technology for Assessment of Lost Oil & Gas Wells (CATALOG), a DoE-funded federal program tasked with identifying and characterizing undocumented orphan wells and measuring their methane emissions.

AI in the Field and the Future of ‘Renewable’ Opportunities

In the quest to identify abandoned oil wells, various research initiatives are underway to paint as precise a picture as possible. Experts at the National Energy Technology Laboratory of the U.S. Department of Energy are utilizing machine learning and AI models to analyze information from diverse data streams, including historical documents, reports from citizen scientists, field data collections, and other less obvious sources, to pinpoint orphan wells.

Simultaneously, NETL experts are processing well integrity test logs to glean insights into the construction, design, and operational history of these wells. By employing supervised learning to decipher the signs and symbols used to denote wells on such maps, they can create images to infer potential well locations. This involves juxtaposing historical data with ‘resurrected’ sites in contemporary documents to determine whether they represent undocumented wells.

Deloitte is also contributing to this field. The research firm has developed a methane emission quantification solution using Google Earth Engine, combining an AI-based geospatial system with a machine learning model. This enables organizations to monitor, quantify, and prioritize the sealing of oil wells.

Further, the University of Cambridge has introduced research based on a popular machine learning algorithm, the random forest, to devise a methodology for identifying wells most likely to experience severe fluid leaks in production fields.

Once the issue is addressed and resolved, promising opportunities arise for repurposing the lands housing these wells. Researchers from McGill University have suggested that dismantling these oil wells could pave the way for environmental sustainability in various forms. These include the underground storage of carbon dioxide and hydrogen and the initiation of geothermal energy systems. Additionally, many of these lands could be ideally suited for wind power installations.