A mixed forward/inverse modeling framework for earthquake deformation problems

Subduction is responsible for the most powerful earthquakes and dangerous volcanic eruptions, resulting in significant human casualties and economic losses. However, the prediction of these natural events remains challenging due to an incomplete understanding of the underlying physics that govern these phenomena. Key questions persist regarding stress accumulation and dissipation, rock behavior under extreme pressure and temperature, and the influence of fluids and melt in these processes. Recent advancements in space geodesy and seismic networks have enabled the measurement of seismic responses and surface displacements, revealing the complex dynamics of subduction zones. To enhance our comprehension of these processes, computational modeling that integrates various types of observations and constraints is crucial. This can be achieved through forward modeling, where model parameters are adjusted to better fit the observations, or through inverse modeling, which extracts critical parameters and the underlying physical mechanisms directly from the data. However, a comprehensive numerical physics-based modeling framework that combines both forward and inverse capabilities, using adjoints, within a unified infrastructure is currently lacking. The objective of this dissertation is to address this gap by developing an open-source, flexible, transparent, and easily extendable framework capable of handling multi-physics coupled problems. This framework will also drive the advancement of innovative techniques for analyzing earthquake systems. It introduces a novel implementation of fault discontinuity within the finite-element model, an improved fault slip inversion method that does not require Green’s function computations, and a novel approach to infer material structure solely from surface displacement data, eliminating the need for seismic velocity analysis. Furthermore, by incorporating these approaches, it offers a novel joint inversion of surface geodetic data, facilitating the simultaneous recovery of subduction zone structure and coseismic slip distribution. This provides valuable insights into the interplay between heterogeneous material structure and fault processes. As a demonstration, the framework successfully recovers the coseismic slip distribution and subduction zone structure by inverting the coseismic surface displacements recorded during the 2011 M9 Tohoku-oki earthquake in Japan. The results reveal weaker material beneath several volcanoes in the same region where local coseismic subsidence was reported during the earthquake. Accounting for heterogeneity in fault slip inversions is crucial for accurately matching the surface displacement data, as suggested by previous studies. Overall, the proposed framework represents a significant advancement in subduction zone modeling, providing a comprehensive tool for understanding and analyzing these complex phenomena, thereby paving the way for improved hazard assessment and risk mitigation strategies.