Next: About this document ... Up: Reproducible Documents

SEP174 -- TABLE OF CONTENTS

Waveform inversion

Instability of adjoint pseudo-acoustic anisotropic wave equations [

SRC]
Huy Le, Stewart A. Levin, and Biondo Biondi
We discover that the adjoint system of the second-order pseudo-acoustic anisotropic wave equations that we have implemented in previous work has unstable solutions and propose two possible alternative systems. The unstable solutions have magnitude that grows linearly with time and does not propagate spatially. This is not the kind of numerical instability that occurs when the propagation timestep does not satisfy the CFL condition. This instability is inherent to this particular type of wave equations. In fact, there is a family of unstable wave equations that all describe the same kinematics. The key to stability for this type of pseudo-acoustic anisotropic wave equations is not whether the system is self-adjoint but whether it reduces to the scalar wave equation in isotropic media.

Extended linearized waveform inversion with velocity updating [

SRC]
Alejandro Cabrales-Vargas
In this report I recast the objective function of linearized waveform inversion with ve- locity updating into a subsurface offset extended version. This approach prevents the corresponding Gauss-Newton Hessian from becoming non-positive definite. Both the- ory and numerical examples appear to support this asseveration, although some work is still required for proper precondition of the extended wave-equation migration velocity analysis operator. Additionally, the new implementation demands more computational resources as a consequence of the extension.

Full-waveform inversion problem using one-way wave extrapolation operators [

SRC]
Rustam Akhmadiev, Biondo Biondi, and Robert G. Clapp
We pose the problem of waveform inversion using one-way wave extrapolation and de- rive the modeling operator nonlinear with respect to slowness model and its linearized operator. We show that linearized operator is composed of three parts corresponding to upward and downward scattering (low-wavenumber components of slowness per- turbation) and perturbation in reflectivity (high-wavenumber components). Using the phase-shift method we simulate wave propagation using full nonlinear and linearized operators in simple 2D acoustic slowness models and discuss future work.

Signal and data processing

Wavefield component separation and debubble on a 3D OBN dataset [

SRC]
Taylor Dahlke, Alejandro Cabrales-Vargas, Rustam Akhmadiev
In this report we demonstrate the pre-processing steps performed on the OBN Gulf of Mexico data set provided to us by Shell in preparation for an FWI type workflow. We show the results of performing PZ-summation on the hydrophone data for up and downgoing wavefield separation, and follow with the extraction of the estimated source wavelet and creation of a shaping filter to remove the bubble and facilitate better phase matching with our synthetic data. We apply this filter to the remaining data and find we can effectively remove the bubble.

Earthquake detection through cross-correlation of the Stanford Fiber Seismic Observatory (SFSO) data [

SRC]
Siyuan Yuan, Biondo Biondi and Robert G. Clapp
Earthquakes occurring in similar locations can have repeatable waveforms. We show that with the Stanford Fiber Seismic Observatory (SFSO) earthquake recordings of Ladera and Felt Lake earthquakes as templates, we can detect other earthquakes oc- curred in similar locations through cross-correlation. As a benchmark, we applied cross-correlation to the recordings of the nearby seismometer stations. We found that the cross-correlation results of the SFSO recordings tend to have higher signal-to-noise ratios (SNR) and fewer false positives compared with the seismometer results. We also show that the SFSO is more sensitive to differences in earthquake mechanisms occurring in Ladera area compared to the two broadband stations. Through cross- correlation of the SFSO data, we identified two Felt Lake earthquakes that were not recorded in the United States Geological Survey (USGS) online catalog. With Felt Lake event recordings, we computed the amplitude ratios of the template to the four detected events. The ratios are similar to those computed with broadband stations for the two relatively large events, which shows the potential of magnitude estimation with SFSO data. For other two events, that are too weak to be included in the USGS catalog, SFSO tends to underestimate the ratios.

Prediction error interpolation in bathymetry [

SRC]
Stewart A. Levin and Christopher M. Castillo
We have implemented missing data interpolation for bathymetry under an ArcGIStm Python framework. We report on both stationary and nonstationary prediction error filter applications to synthetic and field data and the method compares favorably to several widely available interpolation methods.

Computation

Can we beat FFTs computational speed for one-way wavefield propagation? [

SRC]
Biondo Biondi
I compare the computational efficiency of one-way wavefield propagation by FFT-based methods and by implicit finite-differences methods. On modern computer architectures, the theoretical computational advantages of implicit finite-differences methods are not realized in practice. The speed tests that I present indicate that FFTs run much more efficiently than sparse linear solvers and thus they should be the foundation for efficient waveform inverse solutions based on one-way propagation.

Buffers: A library for fast parallel IO and compression [

SRC]
Robert G. Clapp
Data transfer speed improvements from different memory levels has continually lagged behind floating point operations per second (FLOPS) improvements. For seismic appli- cations these differing growth curves have made more and more applications IO bound. We created a library that breaks a dataset into blocks. These blocks can be read/writ- ten to disk in parallel, significantly reducing IO time when using parallel file system or object store. Each block can further be compressed, using multi-dimension compression schemes, reducing the amount of data that needs to be transferred between memory levels.

About this document ...

Next: About this document ... Up: Reproducible Documents

2018-10-07