DENVER, Oct. 24, 2023 /S&P Global/ — With market and regulatory pressures pushing the oil and gas sector to cut methane emissions, drillers and pipeline operators are turning to artificial intelligence to fill the gaps in their technical expertise and determine the severity of emissions events to achieve lower-emission gas production and transport.
Oil and gas producers have their own set of emissions challenges at well sites, as well as their own set of in-the-works regulations that target methane emissions, which data companies such as Project Canary are out to address with continuous monitoring systems and machine learning.
Project Canary got its start in the market as a third-party gas certifier but has since pivoted to mainly emphasize providing more specialized emissions data, with its new commercial modus operandi being to “differentiate” gas with its data services rather than to certify it as “responsibly sourced.”
The company has close to 60 customers across US shale basins utilizing its continuous monitoring technologies, which feed data retrieved at well sites and other facilities through machine learning models.
Natural gas produced with lower emissions can be described in several ways: differentiated, certified, low-carbon or responsibly sourced. They all confer a desirable environmental attribute that has appeal in multiple markets: gas utilities that wish to improve their environmental profile; industries that wish to appeal to environmentally conscious consumers; and global buyers of liquefied US gas in Europe and Asia, where “cleaner” gas can be highly prized.
Using AI enables Project Canary to generate a more comprehensive picture of the emissions at a given monitoring site and to isolate which emissions events diverge from expectations, co-CEO Will Foiles said in an interview. This information can then be used to determine the methane intensity of a given operator’s gas supply.
“When a human would look at a chart, what you’re really trying to figure out is was that level of methane that you saw at the facility — was that something that you expected to see? Machine learning models are pretty good at coming up with a rough indication of expectations,” Foiles said.
In the Permian Basin, where Project Canary has issued low-methane certifications for operators, such as US major Chevron, methane emissions have ballooned over the last decade alongside the exponential rise in oil and gas production.
Annual average volumes of vented and fugitive methane in the Permian, which peaked over the last decade in 2019, rose by close to 580% between 2011 and 2022, according to data from S&P Global Commodity Insights.
Project Canary is able to use machine learning make sense of data from these various emissions streams to inform mitigation activities, Foiles said.
“[Our model] takes in all of the existing data and all the ambient conditions, both forecasted and historical, and because these well sites have emissions largely around the clock — there’s lots of operational emissions — the ambient conditions influence what methane makes its way to a monitor,” he added.
Foiles also offered the example of an operator’s morning water hauling routine as a data point that could be fed into the model to quantify emissions: “If you put that in the training data set for a machine learning model, roughly speaking, it will come to predict that between those hours and under these conditions, whatever they may be, this is the emission level.”
AI in response to rulemakings
US regulators are accelerating this trend, strengthening the methane reduction imperative by way of new rules requiring operators to implement technologies to slash emissions.
The US Pipeline and Hazardous Materials Safety Administration is in the process of finalizing a suite of new methane performance regulations for pipeline operators. In addition to tightening leak detection and repair obligations, PHMSA’s proposed rule would require operators to develop advanced leak detection programs and certify them within six months of the rule’s finalization.
The proposed rule does not explicitly require use of AI; however, making best use of the millions of data points gleaned from advanced leak detection methods, such as satellite imaging and continuous ground-level sensing, necessitates the use of AI virtually by default, said Lauren Crowe, director of gas business transformation and improvement at Duke Energy.
“I think most [advanced leak detection] at its core has some form of AI, whether it’s satellite, whether it’s a vehicle-based technology,” Crowe said in an interview, detailing Duke’s plans to carry out a pilot program with Picarro, a company that uses vehicle-based monitoring and machine learning to estimate leaks.
“Most of the vendors we’ve talked to and have interest in trying and testing have some form of AI and [machine learning] backing their field collection and measurements,” Crowe said.
AI in action
Machine learning can be used to digest vast amounts of data retrieved through measurement sources and produce conclusions that can inform mitigation activities and improve efficiencies, according to Crowe and other industry players.
Duke, for its part, turned to satellites for “top-down” emissions monitoring along its asset base of 30,000 miles of distribution pipeline and 3,000 miles of transmission pipeline and has begun to rely on AI to inform repair operations.
Commercial AI software provider Satelytics takes satellite imagery of Duke assets and then produces alerts to the utility to indicate the presence of a high-, medium- or low-level emissions event based on methane concentration and an estimated flow rate, Crowe explained.
Duke then applies its own internal data to complement the alert to determine how to prioritize the event.
“If we found a large plume over a transmission pipe, that’s obviously in our algorithms going to bump to the top of the list of response,” Crowe explained, but the imagery itself is “just raw imagery.”
“The AI and the machine learning algorithms are really the secret sauce,” Crowe said. “While Duke has a strong presence in the space of building AI and ML internally, Satelytics had already cracked that nut of being able to take that raw imagery and translate it into understanding absorption and being able to essentially parse out the presence of methane within that imagery.”
The upstream oil and gas industry is no stranger to AI in other aspects of its business. For instance, producer Coterra Energy’s CEO, Tom Jorden, credits AI for helping to predict well performance better than reservoir engineers can. “Machine learning lets you explore all the different branch pads you might have taken… and lets you select the most capital-efficient outcome that you could have taken with thousands of different options,” Jorden said at an industry conference Sept. 6.
On the emissions side, Duke Energy’s Crowe said she expects application of machine learning for methane mitigation activities will become more advanced as the models ingest more data and become more refined.
While advanced leak detection and AI will have incremental costs, they could generate savings over time when they enable greater efficiency, according to the utility.
“Imagine where we have years of below-ground leaks and how they express themselves in the atmosphere — the AI can start picking up on those patterns and start to say, ‘hey, we’ve seen things like this in the past and that ended up being associated with a below-ground leak on a valve.’ That’s where I think we’ll be, where today, essentially, we are establishing a priority, but it’s very, in my mind, basic and rudimentary,” Crowe said. “Where I think we’ll be in three to five years is AI being able to predict that risk.”