FD3D_TSN: A Fast and Simple Code for Dynamic Rupture Simulations with GPU Acceleration

2020 ◽  
Vol 91 (5) ◽  
pp. 2881-2889 ◽  
Author(s):  
Jan Premus ◽  
František Gallovič ◽  
Ladislav Hanyk ◽  
Alice-Agnes Gabriel
Keyword(s):  
Seismic Hazard Assessment ◽  
Gpu Acceleration ◽  
Staggered Grid ◽  
Source Inversion ◽  
Dynamic Rupture ◽  
Motion Modeling ◽  
Graphic Processing Units ◽  
Source Imaging ◽  
Core Solution ◽  
Dynamic Source

Abstract We introduce FD3D_TSN—an open-source Fortran code for 3D dynamic earthquake rupture modeling based on the staggered grid fourth-order finite-difference method employing a regular cubical spatial discretization. Slip-weakening and fast-velocity-weakening rate-and-state fault friction laws are combined with vertical planar fault geometry, orthogonal to a planar free surface. FD3D_TSN demonstrates good agreement with other methods in a range of benchmark exercises of the Southern California Earthquake Center and U.S. Geological Survey dynamic rupture code verification project. Efficient graphic processing units (GPU) acceleration using the OpenACC framework yields a factor of 10 speed-up in terms of time to solution compared to a single-core solution for current hardware (Intel i9-9900K and Nvidia RTX 2070). The software is fast and easy-to-use and suitable explicitly for data-driven applications requiring a large number of forward simulations such as dynamic source inversion or probabilistic ground-motion modeling. The code is freely available for the scientific community and may be incorporated in physics-based earthquake source imaging and seismic hazard assessment, or for teaching purposes.

Rupture parameters of dynamic source models compatible with NGA-West2 GMPEs

2020 ◽  
Author(s):  
Frantisek Gallovic ◽  
Lubica Valentova
Keyword(s):  
Ground Motions ◽  
Source Parameters ◽  
Strong Motion ◽  
Seismic Hazard Assessment ◽  
Response Spectra ◽  
Maximum Magnitude ◽  
Source Inversion ◽  
Dynamic Rupture ◽  
Heterogeneous Distribution ◽  
Dynamic Source

<p>Dynamic source inversions of individual earthquakes provide constraints on stress and frictional parameters, which are inherent to the studied event. However, general characteristics of both kinematic and dynamic rupture parameters are not well known, especially in terms of their variability. Here we constrain them by creating and analyzing a synthetic event database of dynamic rupture models that generate waveforms compatible with strong ground motions in a statistical sense.</p><p>We employ a framework that is similar to the Bayesian dynamic source inversion by Gallovič et al. (2019). Instead of waveforms of a single event, the data are represented by Ground Motion Prediction Equations (GMPEs), namely NGA-West2  (Boore et al., 2014). The Markov chain Monte Carlo technique produces samples of the dynamic source parameters with heterogeneous distribution on a fault. For all simulations, we assume a vertical 36x20km strike-slip fault, which limits our maximum magnitude to Mw<7. For dynamic rupture calculations, we employ upgraded finite-difference code FD3D_TSN (Premus et al., 2020) with linear slip-weakening friction law. Seismograms are calculated on a regular grid of phantom stations assuming a 1D velocity model using precalculated full wavefield Green's functions. The procedure results in a database with those dynamic rupture models that generate ground motions compatible with the GMPEs (acceleration response spectra in period band 0.5-5s) in terms of both median and variability.</p><p>The events exhibit various magnitudes and degrees of complexity (e.g. one or more asperities). We inspect seismologically determinable parameters, such as duration, moment rate spectrum, stress drop, size of the ruptured area, and energy budget, including their variabilities.  Comparison with empirically derived values and scaling relations suggests that the events are compatible with real earthquakes (Brune, 1970, Kanamori and Brodsky, 2004). Moreover, we investigate the stress and frictional parameters in terms of their scaling, power spectral densities, and possible correlations. The inferred statistical properties of the dynamic source parameters can be used for physics-based strong-motion modeling in seismic hazard assessment.</p>

Dynamic Rupture and Ground Motion Modeling of the 2016 M6.2 Amatrice and M6.5 Norcia, Central Italy, Earthquakes Constrained by Bayesian Dynamic Source Inversion

2020 ◽  
Author(s):  
Taufiq Taufiqurrahman ◽  
Alice-Agnes Gabriel ◽  
Frantisek Gallovic ◽  
Lubica Valentova
Keyword(s):  
Ground Motion ◽  
Dynamic Models ◽  
Central Italy ◽  
Foot Wall ◽  
Source Inversion ◽  
Dynamic Rupture ◽  
Highly Efficient ◽  
Starting Point ◽  
Dynamic Source ◽  
Strong Ground

<p>The complex evolution of earthquake rupture during the 2016 Central Italy sequence and the uniquely dense seismological observations provide an opportunity to better understand the processes controlling earthquake dynamics, strong ground motion, and earthquake interaction. </p><p>We here use fault initial stress and friction conditions constrained by a novel Bayesian dynamic source inversion as a starting point for high-resolution dynamic rupture scenarios. The best-fitting forward models are chosen out of ~10<sup>6</sup> highly efficient simulations restricted to a simple planar dipping fault. Such constrained, highly heterogeneous dynamic models fit strong motion data well. Utilizing the open-source SeisSol software (www.seissol.org), we then take into account non-planar (e.g., listric) fault geometries, inelastic off-fault damage rheology, free surface effects and topography which cannot be accounted for in the highly efficient dynamic source inversion. SeisSol is based on the discontinuous Galerkin method on unstructured tetrahedral meshes optimized for modern supercomputers. </p><p>We investigate the effects of including the realistic modeling ingredients on rupture dynamics and strong ground motions up to 5 Hz. Synthetic PGV mapping reveals that specifically fault listricity decreases ground motion amplitudes by  ~50 percent in the extreme near field on the foot-wall. On the hanging-wall shaking is increased by ~150 percent as a consequence of wave-focusing effects within 10 km away from the fault. Dynamic modeling thus suggests that geometrical fault complexity is important for seismic hazard assessment adjacent to dipping faults but difficult to identify by kinematic source inversions.</p>

Dynamic source inversion of the 2014 Mw6 South Napa, California, earthquake

2020 ◽  
Author(s):  
Jan Premus ◽  
Frantisek Gallovic
Keyword(s):  
Monte Carlo ◽  
Strong Motion ◽  
Inversion Method ◽  
Source Inversion ◽  
Dynamic Rupture ◽  
Dynamic Simulations ◽  
Simulation Code ◽  
Strong Motion Data ◽  
Motion Data ◽  
Dynamic Source

<p>Dynamic rupture modeling coupled with strong motion data fitting (dynamic source inversion) offers an insight into the rupture physics, constraining and enriching information gained from standard kinematic slip inversions. We utilize the Bayesian Monte Carlo dynamic source inversion method introduced recently by Gallovič et al. (2019), which, in addition to finding a best-fitting model, allows assessing uncertainties of the inferred parameters by sampling the posterior probability density function. The Monte Carlo approach requires running a large number (millions) of dynamic simulations due to the nonlinearity of the inverse problem. It is achieved by using GPU accelerated dynamic rupture simulation code FD3D_TSN (Premus et al., submitted) as a forward solver. We apply the inversion to the 2014 Mw6 South Napa, California, earthquake, employing strong motion data (up to 0.5 Hz) from the 10 closest stations. As an output, we obtain samples of the spatial distributions of dynamic parameters (prestress and parameters of the slip-weakening friction law). Regarding the rupture geometry, we consider two, presently ambiguous, fault planes (Pollitz et al., 2019), showing considerable differences in fitting seismograms in very close vicinity of the fault. We investigate properties of the rupture, especially in the region close to the free surface, and the viability of the model samples to explain the observed data in a broader frequency range (up to 5Hz).</p>

Comparison of local kinematic rupture joint inversion approaches for tsunami early warning: Examples of the 2017-2020 Mw > 6.3 East Aegean earthquakes

2021 ◽  
Author(s):  
Malte Metz ◽  
Marius Isken ◽  
Rongjiang Wang ◽  
Torsten Dahm ◽  
Haluk Özener ◽  
...  
Keyword(s):  
Early Warning ◽  
Joint Inversion ◽  
Moment Tensor ◽  
Strong Motion ◽  
Aegean Sea ◽  
High Rate ◽  
Source Inversion ◽  
Dynamic Rupture ◽  
Centroid Moment Tensor ◽  
Tsunami Early Warning

<p>The fast inversion of reliable centroid moment tensor and kinematic rupture parameters of earthquakes occurring near coastal margins is a key for the assessment of the tsunamigenic potential and early tsunami warning (TEW). In recent years, more and more multi-channel seismic and geodetic online station networks have been built-up to improve the TEW, for instance the GNSS and strong motion networks in Italy, Greece, and Turkey, additionally to the broadband seismological monitoring. Inclusion of such data for the fast kinematic source inversion can improve the resolution and robustness of its’ solutions. However, methods have to be further developed and tested to fully exploit the potential of such rich joint dataset.</p><p>In this frame, we compare and test two in-house developed, kinematic / dynamic rupture inversion methods which are based on completely different approaches. The IDS (Iterative Deconvolution and Stacking, Zhang et al., 2014) combines an iterative seismic network inversion with back projection techniques to retrieve subfault source time functions. The pseudo dynamic rupture model (Dahm et al., in review) links the rupture front propagation estimate based on the Eikonal equation with the dislocation derived from a boundary element method to model dislocation snapshots. We used the latter in both a fast rupture estimate and a fully probabilistic source inversion.</p><p>We use some Mw > 6.3 earthquakes that occurred in the coastal range of the Aegean Sea as an example for comparison: the Mw 6.3 Lesbos earthquake (12 June 2017), the Mw 6.6 Bodrum earthquake (20 July 2017), and the recent Mw 7.0 earthquake which occurred at Samos on 30 October 2020. The latter earthquake and the resulting tsunami caused fatalities and severe damage at the shorelines of Samos and around the city of Izmir, Turkey.<br>The fast estimates are based on only little data and/or prior information obtained from the regional seismicity catalogue and available active fault information. The large number of seismic (broadband, strong motion) and geodetic (high-rate GNSS) stations in local and regional distance from the earthquake with good azimuthal coverage jointly inverted with InSAR data allows for robust inversion results. These, and other solutions, are used as a reference for the comparison of our fast source estimates.<br>Preliminary results of the slip distribution and the source time function are in good agreement with modelling results from other authors.</p><p>We present our insights into the kinematics of the chosen earthquakes investigated by means of joint inversions. Finally, the accuracy of our first fast source estimates, which could be of potential use in tsunami early warning, will be discussed.</p>

Multicycle Simulation of Strike-Slip Earthquake Rupture for Use in Near-Source Ground-Motion Simulations

2021 ◽  
Author(s):  
Percy Galvez ◽  
Anatoly Petukhin ◽  
Paul Somerville ◽  
Jean-Paul Ampuero ◽  
Ken Miyakoshi ◽  
...  
Keyword(s):  
Ground Motion ◽  
Stress Drop ◽  
Slip Rate ◽  
Rupture Process ◽  
Earthquake Rupture ◽  
Rupture Velocity ◽  
Source Inversion ◽  
Dynamic Rupture ◽  
Wide Range ◽  
Source Scaling

ABSTRACT Realistic dynamic rupture modeling validated by observed earthquakes is necessary for estimating parameters that are poorly resolved by seismic source inversion, such as stress drop, rupture velocity, and slip rate function. Source inversions using forward dynamic modeling are increasingly used to obtain earthquake rupture models. In this study, to generate a large number of physically self-consistent rupture models, rupture process of which is consistent with the spatiotemporal heterogeneity of stress produced by previous earthquakes on the same fault, we use multicycle simulations under the rate and state (RS) friction law. We adopt a one-way coupling from multicycle simulations to dynamic rupture simulations; the quasidynamic solver QDYN is used to nucleate the seismic events and the spectral element dynamic solver SPECFEM3D to resolve their rupture process. To simulate realistic seismicity, with a wide range of magnitudes and irregular recurrence, several realizations of 2D-correlated heterogeneous random distributions of characteristic weakening distance (Dc) in RS friction are tested. Other important parameters are the normal stress, which controls the stress drop and rupture velocity during an earthquake, and the maximum value of Dc, which controls rupture velocity but not stress drop. We perform a parametric study on a vertical planar fault and generate a set of a hundred spontaneous rupture models in a wide magnitude range (Mw 5.5–7.4). We validate the rupture models by comparison of source scaling, ground motion (GM), and surface slip properties to observations. We compare the source-scaling relations between rupture area, average slip, and seismic moment of the modeled events with empirical ones derived from source inversions. Near-fault GMs are computed from the source models. Their peak ground velocities and peak ground accelerations agree well with the ground-motion prediction equation values. We also obtain good agreement of the surface fault displacements with observed values.

Kinematic constraining of the multi-fault rupture dynamics of the Norcia, Mw 6.5, 30 October 2016, Central Italy earthquake

2020 ◽  
Author(s):  
Elisa Tinti ◽  
Emanuele Casarotti ◽  
Alice-Agnes Gabriel ◽  
Taufiqurrahman Taufiqurrahman ◽  
Thomas Ulrich ◽  
...  
Keyword(s):  
Central Italy ◽  
Source Inversion ◽  
Fault Rupture ◽  
Kinematic Modeling ◽  
Dynamic Rupture ◽  
Multiple Faults ◽  
Initial Strength ◽  
Central Italy Earthquake ◽  
Kinematic Models ◽  
Geological Constraints

<p>The 2016 Central Italy sequence showed a remarkable complexity involving multiple faults. Highly heterogeneous slip distributions were inferred from kinematic finite source inversions. The coverage and quality of the geodetic and seismic data allow resolving high-resolution details of rupture kinematics of the largest event of the sequence, the Mw 6.5 30 October 2016 Norcia earthquake. Composite fault rupture models suggest that two fault planes may have slipped simultaneously. Nevertheless, kinematic modeling cannot assess the mechanic viability of such multiple fault plane models.</p><p>Using SeisSol, a software package for simulating wave propagation and dynamic rupture based on the arbitrary high-order accurate derivative discontinuous Galerkin method, we therefore try to generate spontaneous dynamic ruptures models compatible with the two fault planes constrained by kinematic inversions. To this end, we adopt a simple slip-weakening friction law with spatially variable dynamic friction and initial strength parameters along multiple faults, compatible with the slip distributions found in the literature. Although we do not to aim to explore the full parameter space, our approach allows testing the feasibility of kinematic models in conjunction with successfully generating spontaneous dynamic rupture scenarios matching seismic and geodetic observations with geological constraints. Such linking enhances and validates the physical implications of kinematic earthquake source inversion.</p>

Modeling dynamic rupture in a 3D earthquake fault model

1998 ◽  
Vol 88 (5) ◽  
pp. 1182-1197 ◽  
Author(s):  
R. Madariaga ◽  
K. Olsen ◽  
R. Archuleta
Keyword(s):  
Boundary Conditions ◽  
Fault Zone ◽  
Spontaneous Rupture ◽  
Grid Method ◽  
Three Dimensions ◽  
Initial Stresses ◽  
Staggered Grid ◽  
Dynamic Rupture ◽  
Plane Shear ◽  
Frictional Instability

Abstract We propose a fourth-order staggered-grid finite-difference method to study dynamic faulting in three dimensions. The method uses an implementation of the boundary conditions on the fault that allows the use of general friction models including slip weakening and rate dependence. Because the staggered-grid method defines stresses and particle velocities at different grid points, we preserve symmetry by implementing a two-grid-row “thick” fault zone. Slip is computed between points located at the borders of the fault zone, while the two components of shear traction on the fault are forced to be symmetric inside the fault zone. We study the properties of the numerical method comparing our simulations with well-known properties of seismic ruptures in 3D. Among the properties that are well modeled by our method are full elastic-wave interactions, frictional instability, rupture initiation from a finite initial patch, spontaneous rupture growth at subsonic and supersonic speeds, as well as healing by either stopping phases or rate-dependent friction. We use this method for simulating spontaneous rupture propagation along an arbitrarily loaded planar fault starting from a localized asperity on circular and rectangular faults. The shape of the rupture front is close to elliptical and is systematically elongated in the inplane direction of traction drop. This elongation is due to the presence of a strong shear stress peak that moves ahead of the rupture in the in-plane direction. At high initial stresses the rupture front becomes unstable and jumps to super-shear speeds in the direction of in-plane shear. Another interesting effect is the development of relatively narrow rupture fronts due to the presence of rate-weakening friction. The solutions for the “thick fault” boundary conditions scale with the slip-weakening distance (D0) and are stable and reproducible for D0 greater than about 4 in terms of 2Tu/μ × Δx. Finally, a comparison of scalar and vector boundary conditions for the friction shows that slip is dominant along the direction of the prestress, with the largest deviations in slip-rate direction occurring near the rupture front and the edges of the fault.

A High-Order Finite-Difference Method on Staggered Curvilinear Grids for Seismic Wave Propagation Applications with Topography

2021 ◽  
Author(s):  
Ossian O’Reilly ◽  
Te-Yang Yeh ◽  
Kim B. Olsen ◽  
Zhifeng Hu ◽  
Alex Breuer ◽  
...  
Keyword(s):  
Wave Propagation ◽  
Finite Difference ◽  
Finite Difference Method ◽  
Difference Method ◽  
Vertical Axis ◽  
Cartesian Grid ◽  
Staggered Grid ◽  
Graphic Processing Units ◽  
Curvilinear Grids ◽  
Curvilinear Grid

ABSTRACT We developed a 3D elastic wave propagation solver that supports topography using staggered curvilinear grids. Our method achieves comparable accuracy to the classical fourth-order staggered grid velocity–stress finite-difference method on a Cartesian grid. We show that the method is provably stable using summation-by-parts operators and weakly imposed boundary conditions via penalty terms. The maximum stable timestep obeys a relationship that depends on the topography-induced grid stretching along the vertical axis. The solutions from the approach are in excellent agreement with verified results for a Gaussian-shaped hill and for a complex topographic model. Compared with a Cartesian grid, the curvilinear grid adds negligible memory requirements, but requires longer simulation times due to smaller timesteps for complex topography. The code shows 94% weak scaling efficiency up to 1014 graphic processing units.

Near-Source Ground Motions and Their Variability Derived from Dynamic Rupture Simulations Constrained by NGA-West2 GMPEs

2021 ◽  
Author(s):  
Ľubica Valentová ◽  
František Gallovič ◽  
Sébastien Hok
Keyword(s):  
Ground Motion ◽  
Ground Motions ◽  
Synthetic Data ◽  
Ground Motion Prediction Equations ◽  
Dynamic Rupture ◽  
Motion Modeling ◽  
Mixed Effect ◽  
Effect Model ◽  
Stress Drops ◽  
Strong Ground

ABSTRACT Empirical ground-motion prediction equations (GMPEs) lack a sufficient number of measurements at near-source distances. Seismologists strive to supplement the missing data by physics-based strong ground-motion modeling. Here, we build a database of ∼3000 dynamic rupture scenarios, assuming a vertical strike-slip fault of 36×20  km embedded in a 1D layered elastic medium and linear slip-weakening friction with heterogeneous parameters along the fault. The database is built by a Monte Carlo procedure to follow median and variability of Next Generation Attenuation-West2 Project GMPEs by Boore et al. (2014) at Joyner–Boore distances 10–80 km. The synthetic events span a magnitude range of 5.8–6.8 and have static stress drops between 5 and 40 MPa. These events are used to simulate ground motions at near-source stations within 5 km from the fault. The synthetic ground motions saturate at the near-source distances, and their variability increases at the near stations compared to the distant ones. In the synthetic database, the within-event and between-event variability are extracted for the near and distant stations employing a mixed-effect model. The within-event variability is lower than its empirical value, only weakly dependent on period, and generally larger for the near stations than for the distant ones. The between-event variability is by 1/4 lower than its empirical value at periods >1  s. We show that this can be reconciled by considering epistemic error in Mw when determining GMPEs, which is not present in the synthetic data.

Sign in / Sign up

Export Citation Format

Share Document