Caig: testing geological hypotheses of the yellowstone caldera using deep generative models

The National Science Foundation Office of Integrative Activities awarded the University of Houston System $662,871 on October 1, 2026, under the Geosciences program (CFDA 47.050) to test geological hypotheses of the Yellowstone Caldera using deep generative models.

The research will create high-resolution, three-dimensional electrical conductivity models of Yellowstone's hydrothermal plumbing system by applying unsupervised, probabilistic deep-learning inversion to airborne electromagnetic and magnetotelluric data. The team will develop a joint inversion framework based on deep generative models to quantify uncertainty and test competing hypotheses for the magmatic system, including stacked sills, vertical dikes, localized lenses, and isolated melt pockets. Rather than producing a single image, the framework will estimate relative support for alternative magmatic melt architectures and constrain melt fraction, connectivity, and volume with quantified uncertainty. The research will improve understanding of how heat, water, fractures, and magma affect geysers, hot springs, ground motion, and other signs of change at Yellowstone. The findings will support volcano and steam-driven hazard assessment and strengthen monitoring and early warning capabilities, and advance methods applicable to mineral exploration for modern technologies and infrastructure.

Work is performed in Houston, Texas through September 30, 2029. The project integrates Earth science with artificial intelligence to train researchers across fields and strengthen the nation's STEM workforce.

Abstract

The project will study yellowstone, one of the world's best-known volcanic regions. IT will create clearer images of hot water and molten rocks deep underground. IT will test which fracture shapes in the subsurface best explain data collected on the ground and from aircraft. The findings will show how heat, water, cracks, and magma affect geysers, hot springs, ground motion, and other signs of change. This new knowledge supports the national interest by promoting assessment of volcano and steam-driven hazards and strengthening monitoring and early warning. The new methods will further benefit society in the search for minerals needed for modern technologies and infrastructure. By linking earth science with artificial intelligence, the project will train researchers across fields and strengthen the nation's stem workforce. The research team will improve imaging of yellowstone?s interconnected hydrothermal and magmatic systems, which remain poorly constrained because of the inherent non-uniqueness of geophysical inversions and the substantial challenge of integrating seismic and electromagnetic measurements. The team will construct high-resolution, three-dimensional electrical conductivity models of yellowstone's hydrothermal plumbing system by applying a novel, unsupervised, probabilistic deep-learning inversion to airborne electromagnetic and magnetotelluric data. The team will develop a joint inversion framework based on deep generative models to quantify uncertainty and test competing hypotheses for the magmatic system, including stacked sills, vertical dikes, localized lenses, and isolated melt pockets. Rather than producing a single blurred image, the framework will estimate the relative support for alternative magmatic melt architectures and constrain melt fraction, connectivity, and volume with quantified uncertainty. The research will advance fundamental understanding of interactions among heat, fluids, fractures, and magma and will establish a diagnostic use of artificial intelligence for geological hypothesis testing. The resulting methods and interdisciplinary training will broaden geophysical imaging capabilities for volcanic hazards and critical-mineral systems worldwide, where the interplay among fractures, fluids and heat in the subsurface is complex. 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.