Abstract Earthquake‐related phenomena such as seismic waves and crustal deformation impact broad regions, requiring large‐scale 3‐D modeling with careful treatment of the boundaries of the model domain. A deep learning approach known as physics‐informed neural networks (PINNs) has potential for such large‐scale problems involving observational data. PINNs have been investigated for modeling coseismic crustal deformation in 2‐D structures. In this study, crustal deformation in 3‐D structures was analyzed using PINNs. To improve modeling accuracy, four neural networks were constructed to represent the displacement and stress fields in two subdomains divided by a fault surface and its extension. Forward simulations exhibited high accuracy for internal deformation but yielded errors for rigid motions, underscoring the inherent difficulty in constraining static deformation at a distant place. Furthermore, the fault slip distribution of the 2008 Iwate–Miyagi inland earthquake was estimated by incorporating surface topography and elastic heterogeneity structures. The results show fault slips consistent with previous studies, despite underestimation of magnitude. So, we demonstrate the capability of PINNs to model 3‐D crustal deformation, thereby offering a flexible approach for large‐scale earthquake modeling using real‐world observations and crustal structures.
Okazaki et al. (Wed,) studied this question.