The movement and deformation of the Earth’s crust and upper mantle provide critical insights into the evolution of earthquake processes and future earthquake potentials. Crustal deformation can be modeled by dislocation models that represent earthquake faults in the crust as defects in a continuum medium. In this study, we propose a physics-informed deep learning approach to model crustal deformation due to earthquakes. Neural networks can represent continuous displacement fields in arbitrary geometrical structures and mechanical properties of rocks by incorporating governing equations and boundary conditions into a loss function. The polar coordinate system is introduced to accurately model the displacement discontinuity on a fault as a boundary condition. We illustrate the validity and usefulness of this approach through example problems with strike-slip faults. This approach has a potential advantage over conventional approaches in that it could be straightforwardly extended to high dimensional, anelastic, nonlinear, and inverse problems.
The Great Sumatran Fault system in Indonesia is a major right‐lateral trench‐parallel system that can be divided into several segments, most of which have ruptured within the last century. This study focuses on the northern portion of the fault system which contains a 200‐km‐long segment that has not experienced a major earthquake in at least 170 years. In 2005, we established the Aceh GPS Network for the Sumatran Fault System (AGNeSS) across this segment. AGNeSS observes large displacements which include significant postseismic deformation from recent large megathrust earthquakes as well as interseismic deformation due to continued elastic loading of both the megathrust and the strike slip system. We parameterize the displacements due to afterslip on the megathrust using a model based on a rate‐ and state‐dependent friction formalism. Using this approach, we are able to separate afterslip from other contributions. We remove predicted deformation due to afterslip from the observations, and use these corrected time series to infer the depth of shallow aseismic creep and deeper locked segments for the Great Sumatran Fault. In the northern portion of this fault segment, we infer aseismic creep down to 7.3 ± 4.8 km depth at a rate of 2.0 ± 0.6 cm/year. In the southwestern portion of the segment, we estimate a locking depth of 14.8 ± 3.4 km with a downdip slip rate of 1.6 ± 0.6 cm/year. This portion of the fault is capable of producing a magnitude 7.0 earthquake.
1 February 2000 in Earth and Planetary Science Letters