Geodetic inversion and optimization using physics-based FEMs models and AI
PI: Christelle Wauthier, Department of Geosciences, EMS & Associate Director, ICDS
Other team members: Romit Maulik, IST
Level of effort appropriate for the proposed project: 1 year of graduate student funding postCOMPS (stipend only) in FY26.
Description of the proposed project: Elastic continuum mechanical models are essential tools for estimating deformation resulting from pressure changes in subsurface cavities, such as magma reservoirs or intrusions. While analytical solutions offer rapid evaluations, they often suffer from reduced accuracy due to their inability to satisfy boundary conditions for complex geometries or very shallow magma sources. Conversely, numerical models—though accurate—are computationally intensive and typically require specialized software and expertise, limiting their accessibility and scalability. To address these limitations, we propose a novel surrogate modeling approach using supervised machine learning, specifically parallel partial Gaussian process emulators. These emulators are trained to replicate the output of high-fidelity finite element simulations with approximately 1,000-fold improvement in computational efficiency. The framework is built upon generalized, nondimensional forms of the governing equations for finite, non-dipping spheroidal cavities or intrusions within elastic half-spaces. This computational efficiency enables transformative advances in geophysical modeling. The emulators make Bayesian inference techniques such as Markov chain Monte Carlo sampling tractable, allowing for more robust uncertainty quantification in data inversions. Furthermore, they provide a platform for systematically exploring the influence of cavity geometry and material properties, offering insight into the limitations of conventional analytical models. Our open-source implementation will deliver the accuracy of numerical simulations at the speed of analytical approximations. It eliminates the need for finite element software, supports a wide range of subsurface geometries and depths, includes emulation uncertainty estimates, and can be extended to new source configurations through additional training. This work not only advances scientific modeling capabilities but also democratizes access to high-fidelity geophysical tools.
List of specific areas of computational and/or data science expertise or skills: AI, surrogate approaches, Bayesian optimization, FEM.
List of specific objectives for work supported by this call: The method will be developed and tested for a few volcanoes where Wauthier’s geodetic modeling studies show the presence of shallow magma reservoirs inducing boundary limitations with analytical solutions (e.g., Ol Doinyo and Nyiragongo volcanoes in the East African Rift). The results will then support grant proposals to NSF, NASA, and other relevant opportunities (for example to the NASA ESI 2026 call when it opens). We will submit at the minimum two papers showcasing the models obtained for the two volcanic case studies envisioned.
Connection of the project to ICDS’s mission: we will develop and apply AI and computational modeling methods to volcanic processes that will have broader impacts on forecasting