Predicting Off‐Fault Deformation From Experimental Strike‐Slip Fault Images Using Convolutional Neural Networks

Abstract

Crustal deformation occurs both as localized slip along faults and distributed deformation off of faults. While there are few robust estimates of off‐fault deformation in nature, scaled physical experiments simulating crustal strike‐slip faulting allow direct measurement of the ratio of fault slip to regional deformation, quantified as kinematic efficiency (KE). We offer an approach to predict KE using a 2D convolutional neural network (CNN) trained directly on fault maps produced by physical experiments. Experiments with different loading rates and basal boundary conditions generate the fault maps throughout the evolution of strike‐slip faults. Strain maps allow us to directly calculate KE and its uncertainty, utilized in the loss function and performance metric. The trained CNN achieves 91% custom accuracy in the KE prediction of an unseen data set. Although the CNN model is trained on scaled experiments, it can predict off‐fault deformation of crustal faults that matches available geologic estimates.

Inverting Geodetic Strain Rates for Slip Deficit Rate in Complex Deforming Zones: An Application to the New Zealand Plate Boundary

“…Whether the residual strain rate field is in reality accommodated by slip deficit on faults not in the model, or as distributed inelastic off‐fault strain throughout the crust is not known from this study. Laboratory studies (e.g., Chaipornkaew et al., 2022), field studies (e.g., Goren et al., 2015; Gray et al., 2017; Shelef & Oskin, 2010), and numerical models (e.g., Herbert et al., 2014; Bird, 2009a) indicate that a sizable portion (1̃0%–30%) of deformation can occur off main faults. Thus it is possible that a portion of the geodetic moment rate is not released by slip on faults and instead manifests as distributed inelastic deformation through the crust.…”

Deep learning for laboratory earthquake prediction and autoregressive forecasting of fault zone stress

“…The active source signals record changes in fault properties prior to failure and thus offer the possibility of incorporating physics-based models (i.e, of asperity contact mechanics) in the ML/DL algorithm. Other studies show that ML can be used to connect lab AE events with fault zone microstructure (Chaipornkaew et al, 2022;Trugman et al, 2020;Ma et al, 2021) and that ML methods can be augmented by directly incorporating physics into the prediction models (Raissi et al, 2019). Here, we build on these works to further the basis for ML/DL work in earthquake physics and forecasting.…”

Recent advances in earthquake seismology using machine learning

“…They investigated the relationship between macroscopic stress changes and microslips between particles using the gradient boosting method (XGBoost) and performed feature importance analysis, revealing that the local spatial autocorrelation of microslips is the most important parameter. Chaipornkaew et al (2022) established an ML-based method to evaluate kinematic efficiency (ratio of fault slip to regional deformation), an indicator of offfault deformation. They conducted an analog experiment in which the temporal evolution of a strike-slip fault could be tracked and trained a CNN model based on the obtained fault maps.…”