The Hverir area in Iceland, known for its geothermal landscapes, is an important example of underground energy systems aimed at improving AI research, such as geothermal energy and Co2 storage. Photo/istock.
Environmental scientists have accumulated data on the surface of the Earth and the vastness of its atmosphere.
What about the underground world?
Not that much.
A new research project, co-led by Professor Yan Liu of USC Viterbi's Thomas Lord Computer Science Bureau, aims to better understand and predict how water, carbon dioxide (CO₂), and energy will move underground, which are important for improving safe Co₂ storage, water management and sustainable energy recovery.
Pi Yan Liu and Co-Pi Behnam Jafarpour.
For example, the findings from this study will help scientists tackle key challenges such as the safe underground storage of Co₂, a compound that promotes the shift in the Earth's energy balance.
CO₂ storage, also known as carbon capture and storage (CCS), is a process in which carbon dioxide (CO₂) emissions from industrial sources or power plants are captured before they enter the atmosphere and stored underground in underground layers, such as depleted oil and gas fields and deep-sea rich beds.
“CO₂ capture and storage are one of the epic challenges of geoscience, and our work could provide a major breakthrough solution for accurate prediction of CO₂ storage,” says Liu, the lead researcher for the study.
“Our work has the potential to provide a major breakthrough solution for accurate prediction of CO2 storage,” Yang Liu.
This research project can also support groundwater management and geothermal energy recovery. Added Liu, a professor of electrical and computer engineering and biomedical sciences, among other applications.
For example, geoscientists can better identify suitable storage reservoirs, predict reactions to development and operational strategies, and characterize important rock flow and transport characteristics, she said.
Take advantage of your strengths
Liu teams up with Behnam Jafarpour, a professor of chemical engineering and materials science, and co-researcher Behnam Jafarpour.
The research project employs machine learning tools to solve some of the mysteries that arise on the ground.
The three-year study, “Advanced underground flow and transport modeling using causal deep learning models based on physics,” is supported by the National Science Foundation as part of a collaboration between artificial intelligence and the Earth Science (CAIG) program.
“Cooperation between geoscientists and computer scientists is essential.” Behnam Jafarpour
“The collaboration between geoscientists and computer scientists is essential to advance underground flow and transport modeling by leveraging recent breakthroughs in AI and machine learning,” says Jafarpour. “The key is to seamlessly integrate trustworthy domain knowledge and physical principles with AI algorithms to develop innovative technologies that leverage the strengths of both fields.”
“Paradigm Shift”
Rocks, fractures, and liquids interact in complex ways beneath the Earth's surface, making it difficult to predict their behavior.
In particular, rock deposits form complex structures, and layers often exhibit complex fluid flow patterns in underground environments. Predicting the dynamics of emerging flow patterns in complex geological layers is of paramount importance for managing the development of underlying resources.
Liu and Jafarpour hope to create ways to better capture and predict underground flow and transport dynamics by combining the physical sciences and data generated by AI's deep learning models called Pincer (a causal deep learning model based on physics).
The study began in mid-September 2024 and is estimated to last until August 31, 2027.
“Pincer presents a paradigm shift from traditional data-driven approaches or model-based approaches to hybrid solutions that combine the benefits of both methods,” the study summary explains. “We will advance our research in geoscience by developing more efficient and robust modeling and predictions of fluid flow and transport processes in underground environments.”
Clearer image
As Liu explained, simulation systems have been using for decades to predict underground flow dynamics, but “these models have limitations,” she said. She explained that they rely heavily on very uncertain inputs and are based on a simplified explanation of the underlying physics.
The new AI tool will now build datasets from small amounts of data, she said.
A clearer understanding of underground dynamics, identifying suitable sites for underground stock storage, for example, will reduce the risk of accidental leakage due to unexpected movement of underground materials.
According to Jafarpour, standard AI tools rely heavily on large training data sets and can generate predictions that deviate from the dominant principles of underground systems.
“We hope that customized solutions like Pincer will help reduce these limitations by increasing physical consistency and reducing the data requirements of AI models,” he said.
Geoscience AI Techniques
Two other USC studies were funded in the NSF grant package. One involves paleoclimatology and another earthquake dynamics.
NSF aims to help develop and implement innovative AI technologies in geoscience to help you better understand extreme weather, solar activity, earthquake risks and more.
The CAIG grant, released in August 2024, calls for the cooperation of geoscientists, computer scientists, mathematicians and others.
Liu and Jafarpour had received seed funds for the USC Ershaghi Energy Transition to begin collaboration in this important field.
Released on July 9, 2025
Last updated on July 9, 2025
