Wansoo Ha

3D Laplace-domain full waveform inversion using a single GPU card

Abstract The Laplace-domain full waveform inversion is an efficient long-wavelength velocity estimation method for seismic datasets lacking low-frequency components. However, to invert a 3D velocity model, a large cluster of CPU cores have commonly been required to overcome the extremely long computing time caused by a large impedance matrix and a number of source positions. In this study, a workstation with a single GPU card (NVIDIA GTX 580) is successfully used for the 3D Laplace-domain full waveform inversion rather than a large cluster of CPU cores. To exploit a GPU for our inversion algorithm, the routine for the iterative matrix solver is ported to the CUDA programming language for forward and backward modeling parts with minimized modification of the remaining parts, which were originally written in Fortran 90. Using a uniformly structured grid set, nonzero values in the sparse impedance matrix can be arranged according to certain rules, which efficiently parallelize the preconditioned conjugate gradient method for a number of threads contained in the GPU card. We perform a numerical experiment to verify the accuracy of a floating point operation performed by a GPU to calculate the Laplace-domain wavefield. We also measure the efficiencies of the original CPU and modified GPU programs using a cluster of CPU cores and a workstation with a GPU card, respectively. Through the analysis, the parallelized inversion code for a GPU achieves the speedup of 14.7 – 24.6 x compared to a CPU-based serial code depending on the degrees of freedom of the impedance matrix. Finally, the practicality of the proposed algorithm is examined by inverting a 3D long-wavelength velocity model using wide azimuth real datasets in 3.7 days.

An efficient wavenumber–space–time domain finite-difference modeling of acoustic wave equation for synthesizing CMP gathers

The wavenumber–space–time domain wave equation is derived from the 2D time-domain wave equation in the horizontally stratified media by taking the Fourier transform of the equation with respect to one of the spatial variables. The derived wave equation is solved by the finite-difference method and analyzed in terms of stability and dispersion. Although there are lots of wavenumbers to be solved according to the sampling theorem, the modeling can be accelerated by limiting the range of wavenumber used for the inverse Fourier transform. The proposed method generates common mid-point (CMP) gathers directly without sorting process by applying the shifting theorem. We show that our algorithm is efficient and generates the same results as ones from the 2D modeling through two numerical examples. The proposed modeling algorithm can be used for the inversion of CMP gathers or amplitude-versus-offset inversion.

Laplace–Fourier-Domain Full Waveform Inversion of Deep-Sea Seismic Data Acquired with Limited Offsets

Laplace–Fourier-domain full waveform inversion is considered one of the most reliable schemes to alleviate the drawbacks of conventional frequency-domain inversion, such as local minima. Using a damped wavefield, we can reduce the possibility of converging to local minima and produce an accurate long-wavelength velocity model. Then, we can obtain final inversion results using high-frequency components and low damping coefficients. However, the imaging area is limited because this scheme uses a damped wavefield that makes the magnitudes of the gradient and residual small in deep areas. Generally, the imaging depth of Laplace–Fourier-domain full waveform inversion is half the streamer length. Thus, dealing with seismic data in the deep-sea layer is difficult. The deep-sea layer reduces the amplitude of signals and acts as an obstacle for computing an exact gradient image. To reduce the water layer’s effect, we extrapolated the wavefield with a downward continuation and performed refraction tomography. Then, we performed Laplace–Fourier-domain full waveform inversion using the refraction tomography results as an initial model. After obtaining a final velocity model, we verified the inversion results using Kirchhoff migration. We presented common image gathers and a synthetic seismogram of Sumatra field data to prove the reliability of the velocity model obtained by Laplace–Fourier-domain full waveform inversion. Through the test, we concluded that Laplace–Fourier-domain full waveform inversion with refraction tomography of the downward-continued wavefield recovers the subsurface structures located at depth despite a relatively short streamer length compared to the water depth.

Expanding domain method for 3D time-Laplace-domain hybrid modeling

Efficient wave propagation modeling is crucial for an efficient full waveform inversion (FWI). We adopt the expanding domain method to generate three-dimensional (3D) Laplace-domain wavefields. The Laplace-domain wavefields can be obtained by Laplace-transforming the time-domain wavefields. We accelerate the generation of Laplace-domain wavefields by applying the expanding domain method not only to the time domain wave propagation, but also to the Laplace transform by running integration. We compare three domain-expansion criteria and demonstrate that the absolute value criterion using is applicable to the Laplace domain. The method can be applied to 3D FWIs using time-domain modeling to generate Laplace or frequency domain wavefields.

A Performance Comparison between Coarray and MPI for Parallel Wave Propagation Modeling and Reverse-time Migration

Coarray is a parallel processing technique introduced in the Fortran 2008 standard. Coarray can implement parallel processing using simple syntax. In this research, we examined applicability of Coarray to seismic parallel processing by comparing performance of seismic data processing programs using Coarray and MPI. We compared calculation time using seismic wave propagation modeling and one to one communication time using domain decomposition technique. We also compared performance of parallel reverse-time migration programs using Coarray and MPI. Test results show that the computing speed of Coarray method is similar to that of MPI. On the other hand, MPI has superior communication speed to that of Coarray.

An efficient first arrival picking procedure for marine streamer data

Picking first breaks is a tedious and time-consuming process. We suggest a procedure for an efficient manual first arrival picking of marine towed-streamer data. The procedure contains three methods: we reduce the total number of manual picks by skipping shots and interpolating picks of each shot gather, reusing previous picks, and sorting three-dimensional (3D) shot gathers using offset. Two-dimensional and 3D simulations demonstrate that we can accelerate the picking process 11–117 times compared to that of a complete picking method. Although the efficiency varies depending on the data, the suggested procedure can save considerable manpower while maintaining the accuracy of manual picking.