AI models help map hidden underground water and oil flows

For two decades, Professor Arshad Kudrolli has studied how fluids and gases reshape the Earth's interior. His work supports efforts to find new energy sources, prevent induced earthquakes, and store carbon. Now, with a $125,000 grant from the American Chemical Society’s Petroleum Research Fund, his team is integrating artificial intelligence into this physics-based research. The goal is to create more accurate models of subsurface fluid movement without relying solely on extensive physical drilling.

The project introduces physics-informed neural networks, a type of AI that learns patterns from data. Traditional neural networks often ignore physical laws, leading to unreliable predictions in complex geological settings. By embedding established equations like Darcy’s Law, which describes flow through porous materials, the researchers ensure the AI respects the underlying physics. This approach allows the model to reconstruct entire underground basins from limited data points, potentially reducing the cost and environmental impact of exploration.

Embedding physical laws into neural networks

Standard machine learning models can make wild guesses when faced with vast amounts of geological data because they do not inherently understand the physical constraints of the environment. Kudrolli’s team addresses this by forcing the network to align with known physical principles. This method acts as a filter, discarding nonsensical scenarios that pure data-driven models might generate. The result is a tool that can predict changes in porous rocks and sand with greater precision, even in areas where direct measurement is impossible.

Reducing invasive drilling for resource exploration

Currently, surveyors often drill multiple holes to locate oil, gas, or water, which is expensive and disrupts the land. The new AI-driven modeling offers a less invasive alternative. By providing the network with a few reference points, such as flow speeds at specific locations, the system can infer the structure of the entire basin. This allows surveyors to target their drilling efforts more accurately. According to reports from GN technics/ai (en-US), this shift toward computational prediction could significantly lower the environmental footprint of subsurface exploration projects.

Prioritizing water security over petroleum

While the technology could help identify new hydrocarbon pockets, Kudrolli emphasizes the growing importance of water. As global demand for clean water rises, understanding subsurface flow becomes critical for locating new sources and tracking pollutant spread. The research aims to answer how the underground structure dictates water movement. By predicting these shifts, the team hopes to help society adapt to changing geological conditions. Kudrolli notes that humanity currently knows more about distant galaxies than about the Earth just a few miles below the surface.