Harnessing data science to understand Earth’s subsurface
The Stanford Natural Gas Initiative hosts the first big data workshop for students and industry leaders on data science techniques for better understanding and managing subsurface resources.
Data science, machine learning, AI – they are embedded in consumer products, medical devices, and security systems. But how can we harness them to manage Earth’s resources, especially those we cannot see or touch, thousands of feet below the surface? And how do we move cutting-edge data gathering and analytical techniques from academia to natural resource industry scientists so they can inform sustainable decision making?
These questions were the crux of conversations at the first Stanford Earth Subsurface Data Science Workshop held Oct. 7. The gathering drew more than 100 participants, including industry leaders from 54 companies as well as students who conduct research using big data. It was part of a week of activities hosted by the Stanford Natural Gas Initiative (NGI), an industry affiliate program of the School of Earth, Energy & Environmental Sciences (Stanford Earth) and the Precourt Institute for Energy.
The workshop was led by Stanford faculty who focus on harnessing information from the depths of our planet – everything from groundwater systems and earthquake hazards to oil and gas recovery. Speakers shared the latest techniques for capturing and modeling the streams of big data that cascade from a plethora of satellite and remote sensing systems now available, as well as the challenges of removing bias and reducing uncertainty. Like many other fields of research, information about Earth’s subsurface has exploded, shifting the culture about how to best acquire, process, model, and analyze data.
A new era
Stanford Earth Dean Stephan Graham kicked off the meeting with a history of geoscience at the school. “We are at the start of a new era,” he said, referring not only to data science and digital tools, but also to how the world is thinking about energy and bringing sustainability into it.
Graham discussed how Stanford scientists have been studying the subsurface for decades, noting the impact of professor Irwin Remson’s quantitative approach to groundwater resources and contamination in the 1960s, which expanded physical, geophysical, and chemical measurements and computational modeling at Stanford Earth. Today, 40 percent of the Stanford Earth faculty study subsurface science and engineering, from the solid Earth to the oceans.
“We’ve moved from data-supported to data-inspired to data-driven,” said energy resources engineering professor Margot Gerritsen, who spoke about the interface of data and Earth sciences.
From earthquakes to natural resources
Fourteen Stanford researchers presented their approaches to subsurface problems, from earthquake analysis to optimizing production in both conventional and unconventional oil and gas reservoirs. Geophysics professor Greg Beroza discussed his use of supercomputers to simulate ground motion for major earthquakes, while geophysics professor Rosemary Knight reviewed her integration of geophysical imaging with remote sensing data to evaluate and manage groundwater resources.
“The applications of data sciences to seismic imaging are here today and there are many different opportunities,” said geophysics professor Biondo Biondi. “We really need to understand the problem and we can use them to complement and enhance traditional seismic imaging.”
Accuracy and bias
Energy resources engineering professor Daniel Tartakovsky broached the idea that when it comes to his work using physics to quantify uncertainty – or the process of using a mathematical approach to determine the likelihood of different outcomes – the field is in a data scarcity rather than a data surplus. And that narrative tied into an idea that all the researchers emphasized: the importance of high-quality data and the detrimental impacts of bias.
“We as a community have to be able to deal with samples so they don’t become dusty on the shelves,” said presenter Allegra Hosford Scheirer, a research scientist with the Basin and Petroleum System Modeling Group at Stanford Earth. “We also have digital dust … we have to do better at collecting and archiving data.” Having shared standards for gathered and adequately cataloged data is essential to the accuracy of scientific calculations.
From academia to industry
The interactive event was part of an effort to create a community of data scientists that address high-level themes about the subsurface. While advanced computing techniques – a core skill for Stanford Earth students – have totally transformed the Earth sciences academically, that shift has not been entirely adopted by natural resource companies that also work with big data. NGI leadership saw an opportunity to share subsurface data analysis best practices with their industry members as well as to create an opportunity for graduate students to understand industry application needs.
“Our role is just to be the catalyst here – this transcends a lot of specific interests,” said NGI director and geophysics professor Mark Zoback, who also spoke about the application of machine learning and AI to reduce the hazards associated with earthquakes induced by hydraulic fracturing.
The workshop included roundtable discussions in which participants brainstormed how to foster collaborations with external organizations and retain data science talent. They debated the obstacles to finding value in big data, the importance of organizational alignment, and the role of academia in data science.
“I was very excited to see the new paradigm that’s put forward and I’m starting to think about how we, industry, can start to test those ideas,” said Xiaojun Huang, a subsurface digital transformation project executive at ExxonMobil. “But there are tremendous challenges ahead of us across the industry. Stanford is in a very unique position to help the industry make that transition.”
Some companies face challenges in communicating the value of new data science techniques to upper-level leadership and connecting new research with improved results. Discussions revolved around how to close that gap by attracting young talent to industry, and NGI managing director Naomi Boness emphasized the importance of having Stanford students involved in those conversations.
The last roundtable of the day focused on recruiting and retaining millennials to data science careers.
“Within the last decade, the way we’re taught has dramatically changed – the whole interaction with learning technology in the classroom has changed,” said attendee Capella Kerst, a PhD candidate in mechanical engineering. Along with advanced computer modeling techniques, students increasingly use drones and satellites to capture data from difficult locations.
Kerst attended the workshop along with about 20 other Stanford students using data analysis in geoscience applications. Their participation is an integral part of bringing technical expertise to challenges that impact our planet and how we take care of it into the future.
“I was so happy to see talks where the emphasis was on combining physics-based modeling and domain expertise with data science,” said Gerritsen, who is also senior associate dean of educational affairs at Stanford Earth and a senior fellow at Precourt. “We need a coherent message about the societal value of subsurface data science.”
Advanced computing techniques have completely transformed the Earth sciences, affecting not only how the field’s scientists and students acquire their data, but also how that data is processed, modeled, and analyzed. “We finally have the ability to explore Earth in its real complexity,” says Eric Dunham, an associate professor of geophysics.
Zoback is also a senior fellow at Precourt; Knight is an affiliate with the Stanford Woods Institute for the Environment; Biondi is a member of the Institute for Computational and Mathematical Engineering (ICME); Tartakovsy and Gerritsen are also members of Bio-X.