S21D-01 INVITED
Seismic modeling of Earth's 3D structure: Recent advancements
Global models of Earth's seismic structure continue to improve due to the growth of seismic data sets, implementation of advanced wave propagations theories, and increased computational power. In my presentation, I will summarize seismic tomography results from the past 5-10 years. I will compare the most recent P and S velocity models, discuss model resolution and model interpretation, and present an, admittedly biased, list of research directions required to develop the next generation 3D models.
S21D-02 INVITED
Using finite-frequency methods to improve regional models
Finite-frequency sensitivity kernels provide a better representation of the true sensitivity of seismic
observables to Earth structure than simple ray-theory. However, given the limitations of all seismic datasets
and the necessary regularization of inversions, it is less clear when the use of finite-frequency kernels results
in a significant improvement to the accuracy of velocity models. We develop velocity models for the western
US using both ray-theoretical and finite-frequency (banana-doughnut) techniques applied to a single seismic
dataset. The teleseismic body-wave traveltime dataset is derived from over 800 stations across the western
US. Approximately half of the stations are provided by the Earthscope Transportable Array, and half are
provided by regional seismic networks. The dataset is inverted to construct our Dynamic North America
Models of P- and S-velocity structure (DNA08-P and DNA08-S). While the models are broadly similar they
differ in detail and we focus on these differences in an effort to identify when it is necessary to use finite
frequency methods. The most notable difference in the models is the amplitude of the recovered velocity
anomalies, which is greater when finite-frequency kernels are used. There are also significant differences in
the geometry and size of low-velocity anomalies. A notable example is the fact that the low velocity conduit
beneath the Yellowstone National Park is more continuous when finite-frequency kernels are used and
connects to a broad low-velocity feature in the lower mantle. We will present our comparison of these models
and hope to provide guidelines for when the use of finite-frequency methods are important.
http://dna.berkeley.edu
S21D-03 INVITED
Estimating the Uncertainty and Predictive Capabilities of a 3-D Velocity Model in Asia
In recent years many models of 3-D Earth structure in Eurasia have been developed, using a variety of techniques and data. Most of these models are not accompanied by quantitative estimates of uncertainty, either in the model parameters themselves (e.g. seismic velocities) or in geophysical observables predicted by the models (e.g. body-wave travel times). We have begun to address this issue by developing and applying metrics to robustly determine the uncertainty and predictive capabilities of 3-D seismic velocity models. Our 'test bed' for this effort is the new 3-D JWM (Joint Weston/MIT) model, a self- consistent P and S velocity model of the crust and upper mantle in a large region of southern and central Asia. The JWM model is the result of a nonlinear, joint body-wave/surface-wave inversion method applied to Pn travel times collected from the Engdahl, van der Hilst and Buland (EHB) bulletin and group- velocity measurements provided by the University of Colorado and Lawrence Livermore National Laboratory. Consistency between the P and S velocities is achieved by imposing bounds on Poisson's ratio and by invoking a regularization constraint that correlates variations in P and S velocity from an initial model. We have begun evaluating the predictive capabilities of our new model using data from a large database of ground-truth events, which were held out from the joint inversion. Our evaluation considers Pn and Sn travel-time residuals relative to JWM and other models, and event location errors incurred by relocating the ground-truth events using JWM travel-time predictions. We have also developed numerical algorithms to compute two types of uncertainty measures. The first measure comprises slices of the posterior velocity covariance function for selected points in the inversion model. The second uncertainty measure translates model uncertainty to travel-time prediction uncertainty in the form of the posterior variance/covariance matrix of the travel-time prediction errors along a set of paths, such as those between a fixed event location to a network of regional stations.
S21D-04 INVITED
Regional-Scale Differential Time Tomography Methods: Development and Application to the Sichuan, China, Dataset
We extended our recent development of double-difference seismic tomography [Zhang and Thurber, BSSA, 2003] to the use of station-pair residual differences in addition to event-pair residual differences. Tomography using station- pair residual differences is somewhat akin to teleseismic tomography but with the sources contained within the model region. Synthetic tests show that the inversion using both event- and station-pair residual differences has advantages in terms of more accurately recovering higher-resolution structure in both the source and receiver regions. We used the Spherical-Earth Finite-Difference (SEFD) travel time calculation method in the tomographic system. The basic concept is the extension of a standard Cartesian FD travel time algorithm [Vidale, 1990] to the spherical case by developing a mesh in radius, co-latitude, and longitude, expressing the FD derivatives in a form appropriate to the spherical mesh, and constructing"stencil" to calculate extrapolated travel times. The SEFD travel time calculation method is more advantageous in dealing with heterogeneity and sphericity of the Earth than the simple Earth flattening transformation and the"sphere-in-a-bo" approach [Flanagan et al., 2007]. We applied this method to the Sichuan, China data set for the period of 2001 to 2004. The Vp, Vs and Vp/Vs models show that there is a clear contrast across the Longmenshan Fault, where the 2008 M8 Wenchuan earthquake initiated.
S21D-05
Adjoint tomography for the Middle East
The complex geology and tectonic setting as well as the sparse coverage of available seismic data complicates seismic monitoring of the Middle East at regional distances. Several studies investigated the seismic structure in the region based on different observations and measurement types, leading to diverging interpretations of crustal and underlying upper mantle structure. We evaluate some of these 3D models by computing full waveforms for several regional earthquakes by a spectral-element method. We measure traveltime shifts between observed broadband data and synthetic seismograms for distinct seismic phases within selected time windows using a recently developed automated measurement algorithm. Based on the measurements for our study region, we select a best model to feature as the starting seismic model for a fully numerical 3D seismic tomography approach. In order to improve this initial 3D seismic model, the sensitivity to seismic structure of the traveltime measurements for all available seismic network recordings is computed. As this represents a computationally very intensive task, we take advantage of a fully numerical adjoint approach by using the efficient software package SPECFEM on a dedicated cluster. We show examples of such sensitivity kernels for different seismic events and use them in a conjugate gradient approach to update the 3D seismic model. We propose various improvements of the initial seismic structure in order to better fit regional seismic waveforms in the Middle East.
S21D-06
Toward Improved Nuclear Explosion Monitoring With Complete Waveform Simulations Using Three-Dimensional Models and Parallel Computing
Next generation methods for lowering seismic monitoring thresholds and reducing uncertainties will likely rely on complete waveform simulations using three-dimensional (3D) earth models. Recent advances in numerical methods for both non-linear (shock wave) and linear (anelastic, seismic wave) propagation, improved 3D models and the steady growth of parallel computing promise to improve the accuracy and efficiency of explosion simulations. These methods implemented in new computer codes can advance physics-based understanding of nuclear explosions as well as the propagation effects caused by path-dependent earth structure. This presentation will summarize new 3D modeling capabilities developed to improve understanding of the seismic waves emerging from an explosion. Specifically we are working in three thrust areas: 1) computation of regional distance intermediate-period (50-10 seconds) synthetic seismograms in 3D earth models to assess the ability of these models to predict observed seismograms from well-characterized events; 2) coupling of non-linear hydrodynamic simulations of explosion shock waves with an anelastic finite difference code for modeling the dependence of seismic wave observables on explosion emplacement conditions and near-source heterogeneity; and 3) implementation of surface topography in our anelastic finite difference code to include scattering and mode-conversion due to a non-planar free surface. Current 3D continental-to-global scale seismic models represent long-wavelength (greater than 100 km) heterogeneity. We are investigating the efficacy of current 3D models to predict complete intermediate (50- 10 seconds) waveforms for well-characterized events (mostly earthquakes) using the spectral element code, SPECFEM3D. Intermediate period seismograms for crustal events at regional distance are strongly impacted by path propagation effects due to laterally variable crustal and upper mantle structure. We are also modeling shock wave propagation from buried explosions with our Eulerian Godunov finite difference code hydrodynamic code, GEODYN. This code includes complete thermodynamic treatment of the near-source region (e.g. pore compaction, material damage, compression and tensile failure). In order to efficiently propagate motions to seismic distances, we are passing GEODYN motions to our anelastic finite difference code, WPP. We are improving WPP to accurately represent the boundary condition for non-planar free- surface topography. This presentation will summarize recent progress in development of these new capabilities for 3D seismic wave modeling at LLNL for treaty monitoring and improving understanding of nuclear explosion phenomenology.
S21D-07
One Dimensional versus Three-Dimensional Crustal Velocity Models for the Korean Peninsula
The current study is intended to determine the crustal velocity structure of the Korean peninsula to help obtain precise earthquake locations. To achieve these goals, travel-time analysis of Korean Meteorological Administration (KMA) waveform data was performed for data from 2001 through 5/2008. Waveform data from 270 local and regional earthquakes published by KMA were used in the current study. The earthquakes were recorded by 119 stations (velocity and accelerometers) located throughout the southern Korean peninsula. The lack of seismicity is typical for the metamorphic crust of the Korean peninsula which produces and average of 30-35 events per year. Nevertheless, ray coverage was sufficient to perform 3-D joint inversion for velocity structure and hypocenter locations. All phase data were hand-picked and preliminary hypocenter locations were estimated using a 1-D velocity model (IASPEI 91). Additionally, regional 1-D velocity models for crust and upper mantle were derived for P- and S-wave velocities and their uncertainties determined. The standard deviations were of magnitude 0.01 or less indicating a good fit of a 1-D model for the Korean peninsula. Static corrections for each seismic station were subsequently calculated to remove any bias introduced by near-surface inhomogeneities in the vicinity of the stations. Although the three-dimensional velocity inversion reduced the variance between observed and modeled travel-times by more than 70 %, the resulting velocity structure reveals a homogeneous distribution which appears, to a large degree, uncorrelated to the tectonics of the peninsula. Crustal velocities of 6.2 km/s and 3.6 km/s for P- and S-waves, respectively, were obtained for the upper crust, while these values increase beneath 25 km to 7 km/s for P- and 4.3 km/s for S-waves. Mantle velocities are encountered below 32 km depth reaching 8.5 km/s and 4.8 km/s for P- and S-waves, respectively. Comparison of hypocenter locations based on 1-D and 3-D inversions indicate shifts to slightly deeper locations for the latter method. Despite these slight differences, the overall result corroborates an acceptable fit of a 1-D velocity model to the crustal structure of the Korean peninsula.
S21D-08
Three-dimensional modeling of explosions and propagation using the representation theorem
In spite of extensive research on the generation of seismic waves by underground nuclear explosions, it is still not possible to provide a complete explanation for the observed wavefields, particularly at regional distances. Spherically symmetric explosion models embedded in layered elastic media effectively model the P phases generated by explosions and the major characteristics of reflected and transmitted phases. Nonlinear axisymmetric finite difference calculations of explosions including gravity can model a more realistic explosion source that directly generates shear waves. These models explain more characteristics of explosion-generated seismic waves, including some aspects of regional shear phases. However, linear and nonlinear near-source 3D effects are clearly important in many cases. SH waves are commonly observed within a few kilometers of explosions, too close to have been generated by conversion of vertical and radial components, and often larger than those components. Furthermore, it has not been established what impact 3D effects have on discriminants and on explosion yield estimates. It is important, therefore, to be able to model and understand how 3D source and source region heterogeneity affect the seismic wavefield and what impact this has on parameters used for nuclear monitoring. To investigate the impact of 3D near-source effects on regional and teleseismic phases, focusing in particular on the generation of SH phases by explosion sources, we developed a 3D hybrid method for full- waveform computations. The method is to perform 3D explosion source region calculations, and then to invoke the representation theorem to propagate the wavefield to local, regional and teleseismic distances using layered earth Green's functions. We are interested in the effects of near-source heterogeneities in both the nonlinear and linear regimes, and therefore require both nonlinear and linear 3D codes to model the source region. We use three codes: TRES3D, a linear elastic code, STELLAR, an Eulerian code well suited to performing near-source calculations including decoupled explosions, and a new 3D Lagrangian code capable of modeling the transition zone between the vaporized explosion cavity and linear elastic region, including the effects of gravity and the free surface. An initial calculation shows that detonation in a rectangular cavity generates SH waves. A more fundamental question is whether the additional freedom of motion in 3D relative to axisymmetric geometry causes SH waves to be generated more easily by small amounts of heterogeneity.