Earthquake forecasting and prediction have long and in some cases sordid histories but recent work has rekindled interest based on advances in early warning, hazard assessment for induced seismicity and successful prediction of laboratory earthquakes. In the lab, frictional stick-slip events provide an analog for earthquakes and the seismic cycle. Labquakes are ideal targets for machine learning (ML) because they can be produced in long sequences under controlled conditions. Recent works show that ML can predict several aspects of labquakes using fault zone acoustic emissions. Here, we generalize these results and explore deep learning (DL) methods for labquake prediction and autoregressive (AR) forecasting. DL improves existing ML methods of labquake prediction. AR methods allow forecasting at future horizons via iterative predictions. We demonstrate that DL models based on Long-Short Term Memory (LSTM) and Convolution Neural Networks predict labquakes under several conditions, and that fault zone stress can be predicted with fidelity, confirming that acoustic energy is a fingerprint of fault zone stress. We predict also time to start of failure (TTsF) and time to the end of Failure (TTeF) for labquakes. Interestingly, TTeF is successfully predicted in all seismic cycles, while the TTsF prediction varies with the amount of preseismic fault creep. We report AR methods to forecast the evolution of fault stress using three sequence modeling frameworks: LSTM, Temporal Convolution Network and Transformer Network. AR forecasting is distinct from existing predictive models, which predict only a target variable at a specific time. The results for forecasting beyond a single seismic cycle are limited but encouraging. Our ML/DL models outperform the state-of-the-art and our autoregressive model represents a novel framework that could enhance current methods of earthquake forecasting.
Forecasting fault failure is a fundamental but elusive goal in earthquake science. Here we show that by listening to the acoustic signal emitted by a laboratory fault, machine learning can predict the time remaining before it fails with great accuracy. These predictions are based solely on the instantaneous physical characteristics of the acoustical signal, and do not make use of its history. Surprisingly, machine learning identifies a signal emitted from the fault zone previously thought to be low-amplitude noise that enables failure forecasting throughout the laboratory quake cycle. We hypot…
Machine learning regression can predict macroscopic fault properties such as shear stress, friction, and time to failure using continuous records of fault zone acoustic emissions. Here we show that a similar approach is successful using event catalogs derived from the continuous data. Our methods are applicable to catalogs of arbitrary scale and magnitude of completeness. We investigate how machine learning regression from an event catalog of laboratory earthquakes performs as a function of the catalog magnitude of completeness. We find that strong model performance requires a sufficiently low…
During an earthquake, part of the released elastic strain energy is dissipated within the slip zone by frictional and fracturing processes, the rest being radiated away via elastic waves. Frictional heating thus plays a crucial role in the energy budget of earthquakes, but, to date, it cannot be resolved by seismological data. Here we investigate the dynamics of laboratory earthquakes by measuring frictional heat dissipated during the propagation of shear instabilities at typical seismogenic depth stress conditions. We perform, for the first time, the full energy budget of earthquake rupture a…
The successful prediction of earthquakes is one of the holy grails in Earth Sciences. Traditional predictions use statistical information on recurrence intervals, but those predictions are not accurate enough. In a recent paper, a machine learning approach was proposed and applied to data of laboratory earthquakes. The machine learning algorithm utilizes continuous measurements of radiated energy through acoustic emissions and the authors were able to successfully predict the timing of laboratory earthquakes. Here, we reproduced their model which was applied to a gouge layer of glass beads and…
We show that the mesoscopic and transport characteristics of networks follow the same trends for the same type of the shear ruptures in terms of rupture speed while also comparing the results of three different friction experiments.The classified fronts obtained from a saw cut Westerly granite fault regarding friction network parameters show a clear separation into two groups indicating two different rupture fronts. With respect to the scaling of local ruptures durations with the networks parameters we show that the gap is related to the possibility of a separation between slow and regular fro…
Today, observations of earthquake precursors remain widely debated. While precursory slow slip is an important feature of earthquake nucleation, foreshock sequences are not always observed and their temporal evolution remains unconstrained. Here, we report on stick-slip experiments (laboratory earthquakes) conducted under seismogenic stresses in dry and fluid pressure conditions. We show that the precursory moment release scales with mainshock magnitude irrespective of the slip behavior (seismic or aseismic), the presence of fluid and the fault's slip history. Importantly, this observation is …
We report some new applications of functional complex networks on acoustic emission waveforms from frictional interfaces. Our results show that laboratory faults undergo a sequence of generic phases as well as strengthening, weakening or fast-slip and slow-slip leading to healing. Also, using functional networks, we extend the dissipated energy due to acoustic emission signals in terms of short-term and long-term features of events. We show that the transition from regular to slow ruptures can have an additional production from the critical rupture class similar to the direct observations of t…
Decades of seismological observations have highlighted the variability of foreshock occurrence prior to natural earthquakes, making thus difficult to track how earthquakes start. Here, we report on three stick-slip experiments performed on cylindrical samples of Indian metagabbro under upper crustal stress conditions (30-60 $MPa$). Acoustic emissions (AEs) were continuously recorded by 8 calibrated acoustic sensors during the experiments. Seismological parameters (moment magnitude, corner frequency and stress-drop) of the detected AEs ($-8.8 \leq Mw \leq -7 $) follow the scaling law between mo…
Over the last two decades, strain and GPS measurements have shown that slow slip on earthquake faults is a widespread phenomenon. Slow slip is also inferred from correlated small amplitude seismic signals known as nonvolcanic tremor and low frequency earthquakes (LFEs). Slow slip has been reproduced in laboratory and simulation studies, however the fundamental physics of these phenomena and their relationship to dynamic earthquake rupture remains poorly understood. Here we show that, in a laboratory setting, continuous seismic waves are imprinted with fundamental signatures of the fault's phys…
Earthquake size-frequency distributions commonly follow a power law, with the b value often used to quantify the relative proportion of small and large events. Laboratory experiments have found that the b value of microfractures decreases with increasing stress. Studies have inferred that this relationship also holds for earthquakes based on observations of earthquake b values varying systematically with faulting style, depth, and for subduction zone earthquakes, plate age. However, these studies are limited by small sample sizes despite aggregating events over large regions, which precludes t…
Data-driven machine-learning for predicting instantaneous and future fault-slip in laboratory experiments has recently progressed markedly due to large training data sets. In Earth however, earthquake interevent times range from 10's-100's of years and geophysical data typically exist for only a portion of an earthquake cycle. Sparse data presents a serious challenge to training machine learning models. Here we describe a transfer learning approach using numerical simulations to train a convolutional encoder-decoder that predicts fault-slip behavior in laboratory experiments. The model learns …
We monitor dynamic rupture propagation during laboratory stick-slip experiments performed on saw-cut Westerly granite under upper crustal conditions (10-90 MPa). Spectral analysis of high-frequency acoustic waveforms provided evidence that energy radiation is enhanced with stress conditions and rupture velocity. Using acoustic recordings bandpass filtered to 400-800 kHz (7-14 mm wavelength) and highpass filtered above 800 kHz, we back projected high-frequency energy generated during rupture propagation. Our results show that the high-frequency radiation originates behind the rupture front duri…
Nearly all aspects of earthquake rupture are controlled by the friction along the fault that progressively increases with tectonic forcing, but in general cannot be directly measured. We show that fault friction can be determined at any time, from the continuous seismic signal. In a classic laboratory experiment of repeating earthquakes, we find that the seismic signal follows a specific pattern with respect to fault friction, allowing us to determine the fault's position within its failure cycle. Using machine learning, we show that instantaneous statistical characteristics of the seismic sig…
A major part of the seismicity striking the Mediterranean area and other regions worldwide is hosted in carbonate rocks. Recent examples are the destructive earthquakes of L'Aquila Mw 6.1 2009 and Norcia Mw 6.5 2016 in Central Italy. Surprisingly, within this region, fast (\approx 3km/s) and destructive seismic ruptures coexist with slow (maximum 10 m/s) and non-destructive rupture phenomena. Despite of its relevance for seismic hazard studies, the transitions from fault creep to slow and fast seismic rupture propagation are still poorly constrained by seismological and laboratory observations…
Rice University（莱斯大学）；Pennsylvania State University（宾夕法尼亚州立大学）；Los Alamos National Laboratory（洛斯阿拉莫斯国家实验室）；University of Nevada, Reno（内华达大学雷诺分校）；Sapienza Universitá di Roma（罗马一大（萨皮恩扎大学））
Rocks under stress deform by creep mechanisms that include formation and slip on small-scale internal cracks. Intragranular cracks and slip along grain contacts release energy as elastic waves termed acoustic emissions (AE). AEs are thought to contain predictive information that can be used for fault failure forecasting. Here we present a method using unsupervised classification and an attention network to forecast labquakes using AE waveform features. Our data were generated in a laboratory setting using a biaxial shearing device with granular fault gouge intended to mimic the conditions of t…
When a dynamic earthquake rupture propagates on a fault in the Earth's crust, the medium around the fault is dynamically damaged due to stress concentrations around the rupture tip. Recent field observations, laboratory experiments and canonical numerical models show the coseismic off-fault damage is essential to describe the coseismic off-fault deformation, rupture dynamics, radiation and overall energy budget. However, the numerical modeling of "localized" off-fault fractures remains a challenge mainly because of computational limitations and model formulation shortcomings. We thus developed…
Earthquake ruptures dynamically activate coseismic off-fault damage around fault cores. Systematic field observation efforts have shown the distribution of off-fault damage around main faults, while numerical modeling using elastic-plastic off-fault material models has demonstrated the evolution of coseismic off-fault damage during earthquake ruptures. Laboratory scale micro-earthquake experiments have pointed out the enhanced high-frequency radiation due to the coseismic off-fault damage. However, the detailed off-fault fracturing mechanisms, subsequent radiation and its contribution to the o…
Fluid pressure and flow in the crust is a key parameter controlling earthquake physics. Since earthquake slip is linked to spatio-temporal localisation of deformation, it is expected that the localised fluid pressure around the fault plane could potentially impact the dynamic strength of the slipping fault zone. Coseismic fluid pressure drops have been inferred from field studies, notably in gold deposits which are thought to be formed by this process, but reliable quantitative predictions are still lacking. Here, experimental results are presented where local on- and off-fault fluid pressure …
Recent advances in large ring laser gyroscopes (RLG) technologies opened the possibility to observe rotations of the ground with sensitivities up to $10^{-11}$ $\frac{rad}{s}$ over the frequency band of seismological interest (0.01-1Hz), thus opening the way to a new geophysical discipline, i.e. rotational seismology. A measure of rotations in seismology is of fundamental interest for (a) the determination of all the six degrees of freedom that characterize a rigid body motion, and (b) the quantitative estimate of the rotational motions contaminating ground translation measurements obtained fr…
Using 9-sets of different laboratory earthquake tests, we examined the nature of cracking under true triaxial stress conditions in the lithosphere . We found that 3D stress state can induce oblique nucleation of many fractures, forming final plane of complex polymodal faults. Our fully 3D experiments indicate unconventional fault nucleation with 2-3 times faster slip phase, implying a new slip-weakening mechanism for earthquakes in upper crust. In addition, We compared our observations of irregular cracks with typical anti-cracks signals in multi-anvil High Pressure and Temperature test. For t…
We study experimentally the fluctuations of deformation along a shear fault naturally emerging within a compressed frictional granular medium. Using laser interferometry, we show that the deformation inside this granular gouge occurs as a succession of localized micro-slips distributed along the fault. The associated distributions of released seismic moments, the memory effects in strain fluctuations, as well as the time correlations between successive events, follow exactly the empirical laws of natural earthquakes. Using a methodology initially developed in seismology and social science, we …
Convergent representations of elastic-wave structure emerge across deliberately distinct seismic training routes
Institute of Geophysics, China Earthquake Administration（中国地震局地球物理研究所）；University of Chinese Academy of Sciences（中国科学院大学）；Laboratory of Seismology and Physics of Earth’s Interior, School of Earth and Space Sciences, University of Science and Technology of China（中国科学技术大学地球和空间科学学院地球内部地震学与物理学实验室）；Institute of Advanced Technology, University of Science and Technology of China（中国科学技术大学先进技术研究院）
A Discontinuous Galerkin Method for Simulating 3D Seismic Wave Propagation in Nonlinear Rock Models: Verification and Application to the 2015 Mw 7.8 Gorkha Earthquake
We consider two statistical regularities that were used to explain Omori's law of the aftershock rate decay: the Levy and Inverse Gaussian (IGD) distributions. These distributions are thought to describe stress behavior influenced by various random factors: post-earthquake stress time history is described by a Brownian motion. Both distributions decay to zero for time intervals close to zero. But this feature contradicts the high immediate aftershock level according to Omori's law. We propose that these statistical distributions are influenced by the power-law stress distribution near the eart…
Deep-focus earthquakes that occur at 350-660 km, where pressures 12-23 GPa and temperature 1800-2000 K, are generally assumed to be caused by olivine-spinel phase transformation (PT). However, there are many existing puzzles: (a) What are the mechanisms for jump from geological 10^-17-10^-15 1/s to seismic 10-10^3 1/s strain rates? Is it possible without PT? (b) How does metastable olivine, which does not completely transform to spinel at high temperature and deeply in the region of stability of spinel for over the million years, suddenly transforms during seconds and generates seismic strain …
Heating during frictional sliding is a major component of the energy budget of earthquakes and represents a potential weakening mechanism. It is therefore important to investigate how heat dissipates during sliding on simulated faults. We present results from laboratory friction experiments where a halite (NaCl) slider held under constant load is dragged across a coarse substrate. Surface evolution and frictional resistance are recorded. Heat emission at the sliding surface is monitored using an infra-red camera. We demonstrate a link between plastic deformations of halite and enhanced heating…
This study illustrates presents a set of the Long-Term Geoelectric Potential (LTGP) measurements that are collected for experimental investigation in Western Greece during a five-year period (1993-1997). During this period, many major destructive earthquake events occurred that caused human casualties and extended material damages. The collection and processing of geoelectric measurements was done by an automated data acquisition system at the Seismological Laboratory of the University of Patras, Greece. This novel study considers seismic activity of this area as a typical linear dynamic syste…
Understanding rock shear failure behavior is crucial to gain insights into slip-related geohazards such as rock avalanches, landslides, and earthquakes. However, descriptions of the progressive damage on the shear surface are still incomplete or ambiguous. In this study, we use the hybrid finite-discrete element method (FDEM) to simulate a shear experiment and obtain a detailed comprehension of shear induced progressive damage and the associated seismic activity. We built a laboratory fault model from high resolution surface scans and micro-CT imaging. Our results show that under quasi-static …
We report on slow earthquakes in Northern Cascadia, and show that continuous seismic energy in the subduction zone follows specific patterns leading to failure. We rely on machine learning models to map characteristic energy signals from low-amplitude seismic waves to the timing of slow slip events. We find that patterns in seismic energy follow the 14-month slow slip cycle. Our results point towards a recurrent build-up in seismic energy as the fault approaches failure. This behavior shares a striking resemblance with our previous observations from slow slips in the laboratory.
Shear banding is widely observed in natural fault zones as well as in gouge layers in laboratory experiments. Understanding the dynamics of strain localization under different loading conditions is essential for quantifying strength evolution of fault gouge, energy partitioning during earthquakes and characterizing rheological transitions and fault zone structure changes. To that end, we develop a physics-based continuum model for strain localization in sheared granular materials. The grain-scale dynamics is described by the Shear Transformation Zone (STZ) theory, a non-equilibrium statistical…
Physics-informed deep learning links geodetic data and fault friction
Stanford University（斯坦福大学）；Disaster Prevention Research Institute, Kyoto University（京都大学防灾研究所）；Japan Agency for Marine-Earth Science and Technology（日本海洋地球科学技术机构）；RIKEN Center for Advanced Intelligence Project（理化学研究所先进智能项目中心）；Kagawa University（香川大学）；Graduate School of Science, Kyoto University（京都大学理学研究科）
