Item

Abstract

Subsurface geometries, such as faults and subducting slab interfaces, are often poorly constrained, yet they exert first-order control on key geophysical processes, including subduction zone thermal structure and earthquake rupture dynamics. Quantifying model sensitivity to geometric variability remains challenging for high-fidelity simulations that require generated meshes, due to the manual effort of mesh generation and the computational cost of exploring high-dimensional parameter spaces. We present a mesh morphing approach that deforms a reference mesh into geometrically varying configurations while preserving mesh connectivity. This enables the automated generation of large ensembles of geometrically variable meshes with minimal user input. Importantly, the preserved connectivity allows for the application of data-driven, non-intrusive reduced-order models (ROMs) to perform robust sensitivity analysis and uncertainty quantification. We demonstrate mesh morphing in two geophysical applications: (a) 3D dynamic rupture simulations with fault dip angles varying across a 40° range, and (b) 2D thermal models of subduction zones incorporating realistic slab interface curvature and depth uncertainties. The morphed meshes retain high quality and lead to accurate simulation results that closely match those obtained using generated meshes. For the dynamic rupture case, we construct ROMs that efficiently predict surface displacement and velocity time series as functions of fault geometry, achieving speedups of up to times relative to full simulations. Our results show that mesh morphing can be a powerful and generalizable tool for incorporating geometric uncertainty into physics-based modeling. The method supports efficient ensemble modeling for rigorous sensitivity studies applicable across a range of problems in computational geophysics.