Collaborative research: caig: physics-informed conditional generative inversion of fracture fields

The National Science Foundation Office of Integrative Activities awarded $499,863 to the Desert Research Institute (Research Administration Services and Compliance Division of the Nevada System of Higher Education) on October 1, 2026, for physics-informed conditional generative deep learning research on fractured rock systems. The three-year project, running through September 30, 2029, develops artificial intelligence tools to create accurate maps of underground fracture networks by combining hydraulic and tracer tomography with physics-informed deep learning frameworks that embed governing physical laws through numerical simulators. The research will estimate fracture network geometry, hydraulic properties, and associated uncertainties, with methods validated using synthetic datasets, 3D-printed fractured rock experiments, and field observations. Work is performed in Reno, Nevada. The project supports groundwater and subsurface resource management, reduces risks in subsurface engineering, and includes hands-on learning experiences for K–12 students and workforce training for the next generation of scientists and engineers. This award funds fundamental research under the Geosciences program (CFDA 47.050). No subawards are planned.

Abstract

Understanding how water moves through fractured underground rocks is essential for management of groundwater and other subsurface resources. This project will develop new artificial intelligence tools to create more accurate maps of underground fracture networks. The project will improve our ability to manage water and energy resources and reduce risks in subsurface engineering. IT will help train the next generation of scientists and engineers and provide hands-on learning experiences to k?12 students. This project develops a physics-informed conditional generative deep learning framework for inverse characterization of fractured rock systems using hydraulic and tracer tomography. The research will develop and evaluate approaches to efficiently estimate fracture network geometry, hydraulic properties, and associated uncertainties. Governing physical laws will be incorporated through embedded numerical simulators to improve learning robustness and generalization under sparse observations. The methods will be validated using synthetic datasets, 3D-printed fractured rock experiments, and field observations. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are not planned for this award.