The episodic transient fault slips called slow slip events (SSEs) have been observed in many subduction zones. These slips often occur in regions adjacent to the seismogenic zone during the interseismic period, making monitoring SSEs significant for understanding large earthquakes. Various fault slip behaviors, including SSEs and earthquakes, can be explained by the spatial heterogeneity of frictional properties on the fault. Therefore, estimating frictional properties from geodetic observations and physics-based models is crucial for fault slip monitoring. In this study, we propose a Physics-Informed Neural Network (PINN)-based new approach to simulate fault slip evolutions, estimate frictional parameters from observation data, and predict subsequent fault slips. PINNs, which integrate physical laws and observation data, represent the solution of physics-based differential equations. As a first step, we validate the effectiveness of the PINN-based approach using a simple single-degree-of-freedom spring-slider system to model SSEs. As a forward problem, we successfully reproduced the temporal evolution of SSEs using PINNs and obtained implications on how to choose the appropriate collocation points by analyzing the residuals of physics-based differential equations. As an inverse problem, we estimated the frictional parameters from synthetic observation data and demonstrated the ability to obtain accurate values regardless of the choice of first-guess values. Furthermore, we discussed the potential of the predictability of the subsequent fault slips using limited observation data, taking into account uncertainties. Our results indicate the significant potential of PINNs for fault slip monitoring.

Monitoring and predicting fault slip behaviors in subduction zones are essential for understanding earthquake cycles and assessing future earthquake potential. We developed a data assimilation (DA) method for fault slip monitoring and short-term prediction of slow slip events (SSEs), which was applied to the 2010 Bungo Channel SSE in southwest Japan. The observed geodetic data were quantitatively explained using a physics-based model with DA. We investigated short-term predictability by assimilating observation data with limited periods. Without prior constraint on fault slip style, observations solely during slip acceleration predicted the occurrence of a fast slip; however, the inclusion of slip deceleration data successfully predicted a slow transient slip. With prior constraint to exclude unstable slip, the assimilation of data after the SSE occurrence predicted a slow transient slip. This study provided a tool using DA for fault slip monitoring and prediction based on real observation data.

Methods for modeling crustal deformation related to earthquakes and plate motions have been developed to incorporate complex crustal structures and multi-fidelity observations. A machine learning approach called physics-informed neural networks (PINNs), which can solve both forward and inverse problems of physical systems, was proposed and applied to the forward simulations of antiplane deformation. Here, we aimed to extend the PINN approach to crustal deformation in two directions: (1) inplane deformation, which is typically used for modeling subduction zones, and (2) inversion analysis of fault slips from geodetic observations. We verified the performance of PINNs on these problems and suggested that formulations in Cartesian and polar coordinates are suitable for forward and inverse modeling, respectively. Furthermore, PINNs yielded stable inversion results without explicit regularization terms, implying that solving the governing equations with PINNs implicitly imposes regularization based on the physical requirements. This may elucidate the distinctive properties of PINNs and provide insights into inversion analyses in geophysics and other fields.

Slow slip events (SSEs) have been observed in many subduction zones and are understood to result from frictional unstable slip on the plate interface. The diversity of their characteristics and the fact that interplate slip can also be seismic suggest that frictional properties are heterogeneous. We are however lacking methods to constrain spatial distribution of frictional properties. In this paper, we employ Physics-Informed Neural Networks (PINNs) to achieve this goal using a synthetic model inspired by the long-term SSEs observed in the Bungo channel. PINN is a deep learning technique which can be used to solve the physics-based differential equations and determine the model parameters from observations. To examine the potential of our proposed method, we execute a series of numerical experiments. We start with an idealized case where it is assumed that fault slip is directly observed. We next move to a more realistic case where the synthetic surface displacement velocity data are observed by virtual GNSS stations. The geometry and friction properties of the velocity weakening region, where the slip instability develops, are well estimated, especially if surface displacement velocities above the velocity weakening region are observed. Our PINN-based method can be seen as an inversion technique with the regularization constraint that fault slip obeys a particular friction law. This approach remediates the issue that standard regularization techniques are based on non-physical constraints. These results of numerical experiments reveal that the PINN-based method is a promising approach for estimating the spatial distribution of friction parameters from GNSS observation.

Short-term slow slip events (S-SSEs) intensively occur at the transition zone along the Nankai subduction zone, southwest Japan. Because crustal deformation due to a single S-SSE is small, the source fault is often represented using a planar uniform single-fault slip model, resulting to little constraint on the spatial heterogeneity in amounts of fault slip. To comprehensively investigate the detailed cumulative spatial distribution of S-SSEs in the entire Nankai subduction zone, we adopted a stacking approach of Global Navigation Satellite System (GNSS) data using low-frequency earthquakes as reference. We extracted cumulative displacements due to a series of S-SSEs from 2004 to 2009; coherent signals in almost opposite direction of plate subduction were obtained. The inverted slip indicated significant slip patches laterally elongated along the transition zone at ~30–35 km depth, and small patches in the shallow portions at ~15–20 km and ~10–15 km depth in eastern Shikoku and in Tokai as well as western Shikoku, respectively. The shallow patches in Shikoku were located on the downdip edge of the coseismic slip area of the 1946 Nankai earthquake, while the Tokai small slip was located on the shallower side of the anticipated source area of a large earthquake. Large slip patches of S-SSEs were complementary to the spatially dense low-frequency earthquake areas; in major S-SSE areas, the number of low-frequency earthquakes is small. This spatial dependence of fault slip style even within the transition zone provides new insights regarding the generation mechanism of slow earthquakes.

The discovery of slow slip events (SSEs) based on the installation of dense geodetic observation networks has provided important clues to understanding the process of stress release and accumulation in subduction zones. Because short-term SSEs (S-SSEs) do not often result in sufficient displacements that can be visually inspected, refined automated detection methods are required to understand the occurrence of S-SSEs. In this study, we propose a new method based on which S-SSEs can be detected in observations derived by a Global Navigation Satellite System (GNSS) array by using l1 trend filtering, a variation of sparse estimation, in conjunction with combined -value techniques. The sparse estimation technique and data-driven determination of hyperparameters are utilized in the proposed method to identify candidates of S-SSE onsets. In addition, combined -value techniques are used to provide confidence values for the detections. The results of synthetic tests demonstrated that almost all events can be detected with the new method, with few misdetections, compared with automated detection methods based on Akaike’s information criteria. The proposed method was then applied to daily displacements obtained at 39 GNSS stations in the Nankai subduction zone in western Shikoku, southwest Japan. The results revealed that, in addition to all known events, new events can be detected with the proposed method. Finally, we found the number of low-frequency earthquakes in the target region increased around at the onsets of potential events.