Recent laboratory studies of fault friction have shown that deep learning can accurately predict the magnitude and timing of stick-slip sliding events, the laboratory equivalent of earthquakes, from the preceding acoustic emissions (AE) events or time-lapse active-source ultrasonic signals. While there are observations that provide insight into the physics of these predictions, the underlying precursory mechanisms are not fully understood. Furthermore, these purely data-driven models require a large amount of training data and may not generalize well outside their training domain. Here, we present a physics-guided machine learning approach - by incorporating the relevant physics directly in the prediction model architecture - with the objectives of enhancing model predictions and generalizability as well as reducing the amount of required training data. We use data from well-controlled double-direct shear laboratory friction experiments on Westerly granite blocks exhibiting numerous regular and irregular stick-slip cycles. Simultaneously, AEs are recorded while the faults are also regularly probed by ultrasonic waves transmitted through the fault zone to monitor the evolution of the contact stiffness during shearing. Our physics-guided ML models take features extracted from AE time series or time-lapse active source ultrasonic signals and predict the shear stress history, which gives both the timing and size of the laboratory earthquakes. The models are constrained by friction laws as well as simplified physical laws governing ultrasonic transmission and AE generation. Our findings indicate that physics-guided ML models outperform purely data-driven models in important ways; they provide accurate predictions even with little training data and transfer learning is greatly enhanced when physics constraints are incorporated. These findings have important implications for earthquake predictions in the field, where training data are scarce.  

How to cite: Shokouhi, P., Borate, P., Riviere, J., Mali, A., and Kifer, D.: Physics-guided machine learning for laboratory earthquake prediction, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-15437, https://doi.org/10.5194/egusphere-egu23-15437, 2023.

Earthquake prediction relies on identification of distinctive patterns of precursory parameters that might precede a large earthquake. However, these patterns are typically not reliably observed in the field. Laboratory stick-slip experiments provide an analog of seismic cycle observed in nature in fully controlled conditions with the associated Acoustic Emission (AE) activity reproducing basic characteristics of seismicity preceding and following the large lab earthquake. Recent laboratory studies showed that the deployment of Machine Learning/Artificial Intelligence techniques has lead to new state of the art results in lab earthquake prediction on smooth faults while using simple statistical features derived from raw AE signals and AE-derived catalogs. However, not enough work has been done on explainability of earthquakes preparatory process on rough faults by leveraging deep learning techniques. In this work we attempt to mitigate this gap and analyze/grade a pool of  explainable seismo-mechanical features through the eyes of neural networks.

We used AE data from three laboratory stick-slip experiments performed in triaxial pressure vessel on Westerly Granite samples. Samples were first fractured at 75MPa confining pressure creating rough fault surfaces. The following stick-slip experiments were performed at constant displacement rate. The experimental procedure led to an extremely complex slip pattern composed of large and small slips of the whole surface, as well as the confined slips highlighted only with AE data and no externally measured slip. The AE catalog was used to extract temporal evolution of 16 seismo-mechanical and statistical features characterizing evolution of stress and damage in response to the axial stress change. The feature pool included clearly physically interpretable parameters such as AE rates, b-value, fractal dimension, AE localization, clustering and triggering properties, and features characterizing the variability of local stress field. 

We apply explainable AI techniques to identify what features are more important to forecast  axial stress and stress drop.  Our feature ranking and importance evaluation with the help of neural networks can serve as an indicator as to what research directions are more promising to take for further feature engineering efforts with an emphasis on explainability of earthquake phenomena.

How to cite: Caus, D., Grover, H., H. Goebel, T., Kwiatek, G., and Weigel, T.: Predicting fault stress level and stress drop using seismo-mechanical and statistical features derived from acoustic signals in laboratory stick-slip friction experiments and assesing feature importance via the derived models., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1967, https://doi.org/10.5194/egusphere-egu23-1967, 2023.

The generation of strong earthquakes is a long-debated problem in seismology, and its importance is increased by the possible implications for earthquake forecasting. It is hypothesized that the earthquake generation processes are anticipated by several phenomena occurring within a nucleation region. These phenomena, also defined as preparatory processes, load stress on the fault leading it to reach a critical state. In this paper, we investigate the seismicity preceding 19 moderate (Mw≥3.5) earthquakes at The Geysers, Northern California, aiming to verify the existence of a preparatory phase before their occurrence. We apply an unsupervised K-means clustering technique to analyze time-series of physics-related features extracted from catalog information and estimated for events occurred before the mainshocks. Specifically, we study the temporal evolution of the b-value from the Gutenberg-Richter (b), the magnitude of completeness (Mc), the fractal dimension (Dc), the inter-event time (dt), and the moment rate (M_r). Our analysis shows the existence of a common preparatory phase for 11 events, plus other 5 events for which we can guess a preparatory phase but with different characteristics of previous ones, indicating different possible activation behavior. The duration of the preparatory process ranges between about 16 hours and 4 days. We find that the retrieved preparatory process involves a decrease of b, Mc, and Dc, and an increase of M_r, as found by many authors. Finally, we show a clear correlation between events showing a preparation phase and the location of injection’s wells, suggesting an important role of fluids in the preparatory process.

How to cite: Iaccarino, A. G. and Picozzi, M.: Detecting the preparatory phase of induced earthquakes at The Geysers (California) using K-means clustering, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5398, https://doi.org/10.5194/egusphere-egu23-5398, 2023.

The Longitudinal Valley in eastern Taiwan, the arc-collision boundary between the Eurasian and Philippine Sea plates, is one the most seismic active areas in the world. On September 18, 2022, the Mw 6.9 Chihshang earthquake struck the south half of the valley and caused severe damage. Since November 2021, we have installed a five-station permanent broadband seismic array with station spacings of 10-20 km around the Chihshang area, and right after the Mw 6.5 foreshock occurred, we further installed a 46-station temporary dense array of nodal seismometers with station spacings of 2-5 km for 35 days. We use SeisBlue, a deep-learning platform/package, to extract the whole earthquake sequence including the Mw 6.5 foreshock, the Mw 6.9 main shock, and over 5,000 aftershocks from the broadband array, and to obtain over 40,000 aftershocks from the dense nodal array. With the high quality and quantity of P- and S-wave arrival times, we apply the finite difference travel time tomography, developed by Roecker et al. (2006). The improved resolution at the shallow part of the crust (at depth < 10 km) provides new constraints to get detailed (with grid spacing 1 km) and reliable Vp, Vs, and Vp/Vs velocity models at the local scale for the first time. Combined with the high-resolution velocity models and the much more complete seismicity, our results clearly depict not only the Central Range fault and the Longitudinal fault but also several local, shallow tectonic structures that have not been observed along the southern Longitudinal Valley.

How to cite: Sun, W.-F., Pan, S.-Y., Huang, C.-M., Guan, Z.-K., Yen, I.-C., and Kuo-Chen, H.: Deep Learning-based earthquake catalog and tomography reveal the rupture process of the 2022 Mw 6.9 Chihshang earthquake sequence, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-7112, https://doi.org/10.5194/egusphere-egu23-7112, 2023.

Fault zone properties can change significantly during the seismic cycle in response to stress changes, microcracking and wall rock damage. Lab experiments show consistent changes in elastic properties prior to and after lab earthquakes (EQ) and previous works show that machine learning/deep learning (ML/DL) techniques are successful for capturing such changes. Here, we apply DL techniques to assess whether similar changes occur during the seismic cycle of tectonic EQ. The main motivation is to generalize lab-based findings to tectonic faulting, to predict failure and identify precursors. The novelty is that we use EQ traces as probing signals to estimate the fault state.

We train DL model to distinguish foreshocks, aftershocks and time to failure of the Mw 6.5 2016 Norcia EQ in central Italy, October 30th 2016. We analyze a 25-second window of 3-component data around the P- and S-wave arrivals for events near the Norcia fault with M>0.5 and ±2 months before/after the Norcia mainshock. Normalized waveforms are used to train a Convolutional Neural Network (CNN). As a first task we divide events into two classes (foreshocks/aftershocks), and then refine the classification as a function of time-to-failure (TTF) for the mainshock. Our DL model perform very well for TTF classification into 2, 4, 8, or 9-classes for the 2 months before/after the mainshock. We explore a range of seismic ray paths near, through, and away from the Norcia mainshock fault zone. Model performance exceeds 90% for most stations. Waveform investigations show that wave amplitude is not the key factor; other waveform properties dictate model performance. Models derived from seismic spectra, rather than time-domain data, are equally good. We challenged the model in several ways to confirm the results. We found reduced performance in training the model with the wrong mainshock time and by omitting data immediately before/after the mainshock. Foreshock/aftershock identification is significantly degraded also by removing high frequencies (filtering seismic data above 25 Hz). We tested data from different years to understand seasonality at individual stations for the time period September to December and removed these effects. Comparing these seasonality effects defined from noise with our EQ results shows that foreshocks/aftershocks for the 2016 Norcia mainshock are well resolved. Training with data containing EQ offers a huge increase in classification performance over noise only, proving that EQ signals are the sole that enable assessing timing as a function of the fault status. To confirm our results and understand which stations are able to detect changes of fault properties we perform a further test cleaning the signals from the seasonality by confounding the DL with a shuffled noise (adversarial training).

We conclude that DL is able to recognize variations in the stress state and fracture during the seismic cycle. The model uses EQ-induced changes in seismic attenuation to distinguish foreshocks from aftershocks and time to failure. This is an important step in ongoing efforts to improve EQ prediction and precursor identification through the use of ML and DL.

How to cite: Laurenti, L., Paoletti, G., Tinti, E., Galasso, F., Franco, L., Collettini, C., and Marone, C.: Using Deep Learning to understand variations in fault zone properties: distinguishing foreshocks from aftershocks, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5810, https://doi.org/10.5194/egusphere-egu23-5810, 2023.

Seismic waves contain information about the earthquake (EQ) source and many forms of noise deriving from the seismometer, anthropogenic effects, background noise associated with ocean waves, and microseismic noise. Separating the noise from the EQ signal is a critical first step in EQ physics and seismic waveform analysis. However, this is difficult because optimal parameters for filtering noise typically vary with time and may strongly alter the shape of the waveform. A few recent works have employed Deep Learning (DL) model for seismic denoising, among which we have taken as a benchmark Deep Denoiser and SEDENOSS. These models turn the noisy trace into a  2D signal (spectrograms) within the model to denoise the traces, making the process pretty heavy. We propose a novel DL-powered seismic denoising algorithm based on Diffusion Models (DMs), keeping the signal in 1D. DMs are the latest trend in Machine Learning (ML), having revolutionized the application fields of audio and image processing for denoising (DiffWave), synthesis (Stable Diffusion), and sequence modeling (STARS). The training of DMs proceeds by polluting a signal with noise until the signal has completely vanished into noise, then reversing the process by iterative denoising, conditioned on the latent signal representation. This makes DMs the ideal tool for seismic traces cleaning, as the model naturally learns from seismic sequences by denoising, which aligns the ML training procedure and the final task objective. In a preliminary evaluation, we used the Stanford Earthquake Dataset (STEAD); our proposed Diffusion-based Seismic Denoiser (DiffSD) outperforms the state-of-the-art DL methods on the Signal Noise Ratio (SNR),  Scale-Invariant Source to Distortion Ratio (SI-SDR), and Source to Distortion Ratio (SDR) metrics. DiffSD also yields qualitatively pleasing EQ traces out of visual inspection in time and frequency. Finally, DiffSD proceeds from regenerating clean EQ signals from noise, which opens the way to data-driven EQ sequence generations, potentially instrumental to further study and dataset augmentations.

How to cite: Trappolini, D., Laurenti, L., Tinti, E., Galasso, F., Marone, C., and Michelini, A.: DiffSD: Diffusion models for seismic denoising, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-13811, https://doi.org/10.5194/egusphere-egu23-13811, 2023.

Friction is a complex phenomenon. This can be seen, for example, in laboratory experiments where stick-slip motion of various kind (i.e., slow and fast instabilities) can be produced when adapting the normal stress applied to the system. Similarly, natural earthquakes also produce
complex stick-slip behaviour. A first challenge in the description of friction comes from the potentially high number of degrees of freedom (dofs) involved in the description of the dynamics of the sliding surfaces. Nonetheless, it was shown that friction can be described with a reduced number of dofs or variables of the dynamics. These may include the shear stress, the relative sliding slip rate, and one or more variables that describe the state of the contact of the sliding surfaces. We investigate the possibility to extract directly from the data the governing equations of friction starting from a simplified synthetic example. We further study the laboratory data with the Hankel Alternative View Of Koopman (HAVOK) theory, a method rooted in dynamical system theory that leverages data driven techniques and produces a Reduced Order Model (ROM) to reconstruct a shadow of the attractor of a system from observational data. We finally compare the results obtained for the laboratory experiments with Cascadia slow earthquakes.

How to cite: Mengaldo, G., Gualandi, A., and Marone, C.: Data-driven slow earthquake dynamics, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12170, https://doi.org/10.5194/egusphere-egu23-12170, 2023.

Currently, the accuracy of synthetic seismograms used for Global CMT inversion, which are based on modern 3D Earth models, is limited by the validity of the path-average approximation for mode summation and surface-wave ray theory. Inaccurate computation of the ground motion’s amplitude and polarization as well as other effects that are not modeled may bias inverted earthquake parameters. Synthetic seismograms of higher accuracy will improve the determination of seismic sources in the CMT analysis, and remove concerns about this source of uncertainty. Strain tensors, and databases thereof, have recently been implemented for the spectral-element solver SPECFEM3D (Ding et al., 2020) based on the theory of previous work (Zhao et al., 2006) for regional inversion of seismograms for earthquake parameters. The main barriers to a global database of Green functions have been storage, I/O, and computation. But, compression tricks and smart selection of spectral elements, fast I/O data formats for high-performance computing, such as ADIOS, and wave-equation solution on GPUs, have dramatically decreased the cost of storage, I/O, and computation, respectively. Additionally, the global spectral-element grid matches the accuracy of a full forward calculation by virtue of Lagrange interpolation. Here, we present our first preliminary database of stored Green functions for 17 seismic stations of the global seismic networks to be used in future 3-D centroid moment tensor inversions. We demonstrate the fast retrieval and computation of seismograms from the database.

How to cite: Sawade, L., Ding, L., Peter, D., Gharti, H. N., Liu, Q., Nettles, M., Ekström, G., and Tromp, J.: A Preliminary Green Function Database for Global 3-D Centroid Moment Tensor Inversions, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16782, https://doi.org/10.5194/egusphere-egu23-16782, 2023.

Understanding the dynamics of microearthquakes is a timely challenge to solve current paradoxes in earthquake mechanics, such as the stress drop and fracture energy scaling with seismic moment. Dynamic modeling of microearthquakes induced by fluid injection is also relevant for studying rupture propagation following a stimulated nucleation. We study the main features of unstable dynamic ruptures caused by fluid injection on a target preexisting fault (50m x 50m) generating a M_w=1 event. The selected fault is located in the Bedretto Underground Laboratory (Swiss Alps) at ≈1000m depth. We perform fully dynamic rupture simulations and model seismic wave propagation in 3D by adopting a linear slip-weakening law. We use the distributed multi-GPU implementation of SeisSol on the supercomputer Marconi100.

 Stress field and fault geometry are well constrained by in-situ observations, allowing us to minimize the a priori imposed parameters. We investigate the scaling relations of stress drop, slip-weakening distance (D_c) and fracture energy (G_c) focusing on their role in governing dynamics of rupture propagation and arrest for a target M_w=1 induced earthquake. We explore different homogenous conditions of frictional parameters, and we show that the spontaneous arrest of the rupture is possible in the modeled stress regime, by assuming a high ratio between stress excess and dynamic stress drop (the fault strength parameter S), characterizing the fault before the fluid pressure change. The rupture arrest of modeled induced earthquakes depends on the heterogeneity of dynamic parameters due to the spatially variable effective normal stress. Moreover, for a fault with high S values (not ready to slip), small variations of D_c (≈0.5÷1.2 mm) can drive the rupture from self-arrested to run-away. Studying dynamic interactions (stress transfer) among slipping points on the rupturing fault provides insight on the breakdown process zone and shear stress evolution at the crack tip leading to failure. The inferred spatial dimension of the cohesive zone in our models is nearly ~0.3-0.4m, with a maximum slip of ~0.6 cm. Finally, we compare stress drop and fracture energy estimated from synthetic waveforms with assumed dynamic parameters. Our results suggest that meso-scale processes near the crack-tip affect rupture dynamics of micro-earthquakes.

How to cite: Mosconi, F., Tinti, E., Casarotti, E., Gabriel, A., Dorozhinskii, R., Dal Zilio, L., Rinaldi, A. P., and Cocco, M.: Modeling dynamic ruptures on extended faults for microearthquakes induced by fluid injection, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12392, https://doi.org/10.5194/egusphere-egu23-12392, 2023.

Recent advances in machine learning are having a revolutionizing effect on our understanding of the Solid Earth, in particular in the automatic detection of geophysical events and objects (such as volcano movements in InSAR [Anantrasirichai et al. 2018], landslides in optical satellite imaging [Mohan et al. 2021], and earthquakes in seismic recordings [Zhu et al. 2019]). Yet, the understanding of geophysical phenomena requires us to be able to accurately characterize them: automatizing such tasks by machine learning is the new challenge for future years. One main difficulty resides in the availability of a high quality labeled database, that is a database with both input data (such as remote sensing acquisitions) together with their ground truth (what we are looking for). In this context, the problem of ground deformation estimation by sub-pixel optical satellite image registration (or correlation) is a good example.

Precise estimation of ground displacement at regional scales from optical satellite imagery is fundamental for the understanding of earthquake ruptures. Current methods make use of correlation techniques between two image acquisitions in order to retrieve a fractional pixel shift [Rosu et al. 2014, Leprince et al. 2007]. However, the precision and accuracy of image correlation can be limited by various problems, such as differences in local lighting conditions between acquisitions, seasonal changes in image reflectance, stereoscopic and resampling artifacts, which can all bias the displacement estimate, especially in the sub-pixel domain.

Image correlation is a valuable and unique source of information on the coseismic strain particularly in the near-field of earthquake ruptures, where InSAR can often decorrelate. However, the correlation process can be limited by the underlying assumption of a locally homogenous displacement within the correlation window (typically 3x3 to 32x32 pixels wide), leading to a bias when the correlation window crosses a fault discontinuity. Data-driven methods may provide a way to overcome these errors. Yet, no ground truth displacement field exists in real world datasets. From the generation of a realistic simulated database based on Landsat-8 satellite image pairs, with added simulated sub-pixel shifts, we developed a Convolutional Neural Network (CNN) able to retrieve sub-pixel displacements. In particular, we show how to specifically design discontinuities in the training set in order to reduce the near-field bias where the correlation window crosses the fault. Comparisions are made with state-of-the-art correlations methods both on synthetic (and realistic) data, and on real images from the Ridgequest area.
This preliminary study provides an example of how to use realistic synthetic data generation (combining real data with synthetic numerical approaches) for training a machine learning model able to estimate fault displacement fields. Such an approach could be applied to other characterization tasks, e.g. when realistic numerical simulation data is available, while sufficient ground truth data is not.

How to cite: Giffard-Roisin, S., Montagnon, T., Pathier, E., Dalla Mura, M., Marchandon, M., and Hollingsworth, J.: Can deep learning help understand and characterize earthquakes? An example with deep learning optical satellite image correlation., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-7571, https://doi.org/10.5194/egusphere-egu23-7571, 2023.

Fault rupture has various complexity in space and time. The temporal complexity could be expressed by the radiated energy enhancement factor (Ye et al., 2018), which is the ratio of the radiated energy to its theoretical minimum value. Regarding the spatial complexity, the rupture directivity has been well investigated from small scales (Boatwright 2007; Kane et al., 2013; Ross and Ben-Zion, 2016) to large scales (e.g. Ide and Takeo, 1997; Ruiz et al., 2016) by investigating the azimuthal dependency of dominant frequency or estimating the source process. While these studies focus on the spatiotemporal complexity of the entire fault rupture, the mesoscopic scale rupture complexity also exists through the rupture propagation, which is an important perspective of the rupture mechanics. Specifically, we can classify the rupture propagation into two endmembers: mode-II and -III ruptures. In this regard, we focused on the rupture propagations at the scale of subfault extracted from the waveform inversion.

In this study, we analyzed multiple M6-class inland earthquakes in Japan using waveform inversion with the radiation-corrected empirical Green’s functions (Shibata et al., 2022), which enable us to estimate slip distributions with slip directions by synthesizing the EGF waveforms for any focal mechanisms. Then, we introduced rupture-mode intensity to evaluate the rupture-mode preferences by comparing the rupture propagation direction with the slip direction for each earthquake. As a result, we confirmed that the rupture preferentially propagated parallel (mode II) or perpendicular (mode III) to the slip direction, which is expected from the fracture mechanics. In addition, the characteristic of rupture propagation at the early stage was similar to that during the entire rupture, implying that most rupture characteristics are determined at the early stage.

How to cite: Shibata, R. and Aso, N.: Rupture-mode preferences of crustal earthquakes in Japan, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-242, https://doi.org/10.5194/egusphere-egu23-242, 2023.

The Denali Fault earthquake was one of the largest strike-slip earthquakes with significant surface ruptures that occurred in 2002 in Alaska, United States. Probabilistic fault displacement hazard assessment (PFDHA) plays an important role in post-earthquake disaster management. This is because critical facilities and infrastructures in the vicinity of active faults are prone to major damage, leading to suspension of service due to fault displacement. Hence, in the present study, a PFDHA due to the Denali earthquake is conducted using a new methodology of stochastic source-based fault displacement hazard analysis. In this method, the surface rupture can be evaluated by applying Okada equations to simulated earthquake source models. The main differences between the methodology of the present study with conventional fault displacement assessment practices are to utilize the stochastic source models instead of the empirical predictive relationship of surface fault displacement and to calculate the distribution for surface fault displacement at a site of interest using the Okada equations. The new methodology is more versatile than the existing methods in several characteristics. First, it is applicable to all faulting mechanisms (e.g., strike-slip, normal, and reverse) by specifying different rake angles of the ruptured fault. Second, it has the ability to consider multi-segment fault rupture. Third, the calculation of three translational displacements by the Okada equations for a given location is available. Lastly, it provides physically consistent fault displacement modelling at two locations for a given earthquake scenario, allowing estimating of the differential fault displacement at two sites. Then the capability of the method is evaluated by applying it to the historical case of the 2002 Denali Fault earthquake. The satisfactory match of the modelled fault displacement and the observations, such as surface offset, Global Positioning System (GPS), and Interferometric Synthetic Aperture Radar (InSAR) data, is achieved based on calculating a performance metric. Therefore, more realistic ground deformation assessments can be carried out. Importantly, the obtained results significantly contribute to the hazard in earthquake-prone areas and reduce potential fatality and casualty risks as well as the post-earthquake damage repair costs of the built environment.

How to cite: Shoaeifar, P. and Goda, K.: Stochastic Source Modelling of the 2002 Denali Earthquake for Fault Displacement Hazard Assessment, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1726, https://doi.org/10.5194/egusphere-egu23-1726, 2023.

Seismological data from almost 100 broadband stations (70 < Δ < 420 km) from Croatia, Slovenia, Hungary, Italy, Austria, Bosnia and Hercegovina, Montenegro, and Slovakia have been used in the rupture analysis of the Petrinja (Croatia) M_W6.4 earthquake, that occurred on the 29th of December 2020. Several foreshocks and aftershocks have been used as empirical Green’s function (EGF) to isolate source effects from propagation and local soil effects. First, P-wave mainshock seismograms are deconvolved from the EGF seismograms in the frequency domain to obtain the corner frequency (f_c). Assuming Brune’s source model, the spectral analysis results in a large stress drop of 25 MPa. Second, using time-domain deconvolution of the Love wave time windows, apparent source time functions (ASTFs) have been computed and indicate an average source duration of 5 seconds. No significant directivity effects can be seen in both the f_c values and source durations, whose weak variability suggests a bilateral rupture. Lastly, physical rupture parameters, such as rupture velocity, rupture dimensions, slip model and rise time, have been extracted from the ASTFs by two different techniques: (1) the Bayesian inversion method (Causse et al. 2017) and (2) the backprojection of the ASTFs on the isochrones (Király‐Proag et al. 2019). Both techniques indicate a slow rupture velocity (about 50% of the shear-wave velocity) and a rather short rupture length for an M_W6.4 event (about 8 km), consistent with the obtained large seismological stress drop. Such features may be explained by the relatively complex and segmented fault system, typical of immature fault contexts.

References:

Causse, M., Cultrera, G., Moreau, L., Herrero, A., Schiappapietra, E. and Courboulex, F., 2017. Bayesian rupture imaging in a complex medium: The 29 May 2012 Emilia, Northern Italy, earthquake. Geophysical Research Letters, 44(15), pp.7783-7792.

Király‐Proag, E., Satriano, C., Bernard, P. and Wiemer, S., 2019. Rupture process of the M w 3.3 earthquake in the St. Gallen 2013 geothermal reservoir, Switzerland. Geophysical Research Letters, 46(14), pp.7990-7999.

How to cite: Lončar, I., Causse, M., Vallée, M., and Markušić, S.: Slow rupture propagation and large stress drop during the 2020 Mw6.4 Petrinja earthquake, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4934, https://doi.org/10.5194/egusphere-egu23-4934, 2023.

The 2015 Mw=8.3 Illapel earthquake is one of the largest megathrust earthquakes that has been recorded along the Chilean subduction zone. Given its magnitude, different rupture scenarios have been obtained. Previous studies show different amounts of shallow slip with some results showing almost no slip at the trench and others showing significant slip at shallow depths, up to 14 meters. In this work, we revisit this event by assembling a comprehensive data set including continuous and survey GNSS measurements corrected for post-seismic and aftershock signals, ascending and descending InSAR images of the Sentinel-1A satellite, tsunami data along with high-rate GPS, and doubly integrated strong-motion waveforms. We follow a Bayesian approach using the AlTar algorithm, in which we aim to obtain the posterior PDF of the joint inversion problem. In addition, we explore a new approach to account for forward problem uncertainties using a second-order perturbation approach. 

Results show a rupture with two main slip regions, and with significant slip at shallow depth that correlates with outer-rise aftershocks. Furthermore, kinematic models indicate that the rupture is encircling two regions updip of the hypocenter that remain unbroken during the mainshock and its aftershocks. These encircling patterns have been previously suggested by back-projection results but have not been observed in finite-fault slip models. We propose that the encircled regions correspond to barriers that can potentially be related to secondary fracture zones in the Chilean subduction zone.

How to cite: Duputel, Z., Caballero, E., Twardzik, C., Rivera, L., Klein, E., Jiang, J., Liang, C., Zhu, L., Jolivet, R., Fielding, E., and Simons, M.: Revisiting the 2015 Mw=8.3 Illapel earthquake: Unveiling complex fault slip properties using Bayesian inversion., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-6666, https://doi.org/10.5194/egusphere-egu23-6666, 2023.

We produced a comprehensive stress drop catalog for northern Chile. To improve reliability, we applied a combination of two different stress drop estimation approaches. The result is a mapped stress drop distribution for more than 30,000 events covering the subduction zone from the trench to a depth of about 150 km. The stress drops were computed on the basis of a recently updated version of the IPOC seismic catalog, now spanning the years 2007 to 2021, using the spectral stacking technique as well as the spectral ratio technique.
The resulting distribution reveals a segmentation of median stress drop values for different seismogenic parts of the subduction zone: We find the lowest stress drops for interface events and slightly increased values for the two parallel bands of seismicity below, which lie inside the subducting plate. The upper plate events, show higher stress drops and the intermediate depth events bear the highest median stress drop. The variation of the median stress drops between classes is small: from 1.3 MPa for interface events to about 3.2 MPa for intermediate depth events. This being the values of the spectral ratio results. Using spectral ratios we find the exact same order of median stress drops between the classes with a range of 2.0 MPa to 5.8 MPa for interface and intermediate depth events, respectively. Interestingly, there is no stress drop increase with dept in the uppermost ~80 km, i.e. within each of the classes except for the intermediate depth events.
Additionally, we observe spatial stress drop variability, a noticeable increase with distance from the plate interface, and temporal variability connected with the two megathrust events in the study region, the Mw7.6 2007 Tocopilla event and the Mw 8.1 Iquique event. 

How to cite: Folesky, J., Hofman, R., and Kummerow, J.: Stress Drop Segmentation in the Northern Chilean Subduction Zone: from Interface to Deep Seismicity., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8901, https://doi.org/10.5194/egusphere-egu23-8901, 2023.

The 2019 Ridgecrest earthquake sequence is an exceptionally well-studied event, captured in nearly unprecedented geophysical detail. Three horizontal tensor borehole strainmeters (BSMs), ranging from ~2 to 30 kms near the trace of the rupture, offer a less-conventional and more sensitive measure of coseismic and postseismic deformation for the event. Historically, these instruments are noted as unreliable for measurements of coseismic strain because they are sensitive to small-scale, near-instrument heterogeneities, such as additional offsets triggered by dynamic strains or pore pressure effects. However, many studies compare the strains with pre-constrained forward models of slip. Our preliminary investigations show that we can better match the observed strains if we include BSM measurements in a joint inversion with GPS displacements for coseismic slip. Postseismically, the strainmeters record rapid, non-monotonic deformation that likewise does not match existing afterslip models with a single decay time. We present a new interpretation of co- and post-seismic deformation using BSM strains and GPS displacements at discrete time intervals marked by a change in sign or rate of strain accumulation. Our joint analysis resolves details of the co- and post-seismic slip history that remain otherwise hidden with more common satellite-based inversions from GPS and InSAR alone. 

How to cite: Hanagan, C. and Bennett, R.: Co- and post-seismic deformation resolved from joint inversion of GPS and borehole strainmeter measurements during the 2019 Ridgecrest earthquake sequence, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9037, https://doi.org/10.5194/egusphere-egu23-9037, 2023.

Mechanical models of slip development on geological faults and basal slip development in landslide or ice-sheets generally consider interfacial strength to be frictional and deformation of the surrounding medium to be elastic. The frictional strength is usually considered as sliding rate- and state-dependent. Their combination, elastic deformation due to differential slip and rate-state frictional strength, leads to nonlinear partial differential equations (PDEs) that govern the spatio-temporal evolution of slip. Here, we investigate how (synthetic) data on fault slip rate and traction can find the system of PDEs that governs fault slip development during the aseismic rupture phase and the slip instability phase. We first prepare (synthetic) data sets by numerically solving the forward problem of slip rate and fault shear stress evolution during a seismic cycle. We now identify the physical variables, for example, slip rate or frictional state variable, and apply nonlinearity identification algorithms within different time durations. We show that the nonlinearity identification algorithms can find the terms of the PDE that governs the slip rate evolution during the aseismic rupture phase and subsequent instability phase.

In particular, we use nonlinear dynamics identification algorithms (e.g., SINDy, Brunton et al., 2016) where we solve a regression problem, Ax=y. Here, y is the time derivative of the variable of interest, for example, slip rate. A is a large matrix (library) with all possible candidate functions that may appear in the slip rate evolution PDE. The entries in x, to be solved for, are coefficients corresponding to each library function in matrix A. We update A according to the solutions x so that A's column space can span the dynamics we seek to find. To find the suitable column space for A, we encourage sparse solutions for x, suggesting that only a few columns in matrix A are dominant, leading to a parsimonious representation of the governing PDE. 

We show that the algorithm successfully recovers the PDE governing quasi-static fault slip and basal slip evolution. Additionally, we could also find the frictional parameter, for example, a/b, where a and b, respectively, are the magnitudes that control direct and evolution effects. Moreover, the algorithm can also determine whether the associated state variable evolves as aging- or slip-law types or their combination. Further, with the data set prepared from distinct initial conditions, we show that the nonlinear dynamics identification algorithm can also determine the problem parameters’ spatial distributions (heterogeneities) from fault slip rate and shear stress data. 

How to cite: Biswas, P. and Ray, S.: Finding governing PDEs of quasistatic fault slip and basal slip evolution from (synthetic) slip rate and shear traction data., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16374, https://doi.org/10.5194/egusphere-egu23-16374, 2023.