Memory-efficient source wavefield reconstruction and its application to 3D reverse time migration

Geophysics ◽  
2021 ◽  
pp. 1-76
Author(s):  
Zhiming Ren ◽  
Qianzong Bao ◽  
Shigang Xu

Reverse time migration (RTM) generally uses the zero-lag crosscorrelation imaging condition, requiring the source and receiver wavefields to be known at the same time step. However, the receiver wavefield is calculated in time-reversed order, opposite to the order of the forward-propagated source wavefield. The inconvenience can be resolved by storing the source wavefield on a computer memory/disk or by reconstructing the source wavefield on the fly for multiplication with the receiver wavefield. The storage requirements for the former approach can be very large. Hence, we have followed the latter route and developed an efficient source wavefield reconstruction method. During forward propagation, the boundary wavefields at N layers of the spatial grid points and a linear combination of wavefields at M − N layers of the spatial grid points are stored. During backward propagation, it reconstructs the source wavefield using the saved wavefields based on a new finite-difference stencil ( M is the operator length parameter, and 0 ≤  N ≤  M). Unlike existing methods, our method allows a trade-off between accuracy and storage by adjusting N. A maximum-norm-based objective function is constructed to optimize the reconstruction coefficients based on the minimax approximation using the Remez exchange algorithm. Dispersion and stability analyses reveal that our method is more accurate and marginally less stable than the method that requires storage of a combination of boundary wavefields. Our method has been applied to 3D RTM on synthetic and field data. Numerical examples indicate that our method with N = 1 can produce images that are close to those obtained using a conventional method of storing M layers of boundary wavefields. The memory usage of our method is ( N + 1)/ M times that of the conventional method.

Geophysics ◽  
2015 ◽  
Vol 80 (2) ◽  
pp. S65-S77 ◽  
Author(s):  
Hu Jin ◽  
George A. McMechan ◽  
Bao Nguyen

We have developed a new method of extracting angle-domain common-image gathers (ADCIGs) from prestack reverse time migration (RTM) that has minimal intermediate storage requirements. To include multipathing, we applied an imaging condition for prestack RTM that uses multiple excitation image times. Instead of saving the full-source snapshots at all time steps, we picked and saved only a few of the highest amplitude arrivals, and their corresponding excitation times, of the source wavefield at each grid point, and we crosscorrelated with the receiver wavefield. When extracting the ADCIGs from RTM, we calculated the source propagation direction from the gradient of the excitation times. The result was that the source time snapshots do not have to be saved or reconstructed during RTM or while extracting ADCIGs. We calculated the receiver propagation direction from Poynting vectors during the receiver extrapolation at each time step and the reflector normal direction by the phase-gradient method. With a new strategy that uses three direction vectors (the source and receiver propagation directions as well as the reflector normal direction), we provided more reliable ADCIGs that are free of low-wavenumber artifacts than any two of them do separately, when the migration velocity model was near to the correct velocity model. The 2D and 3D synthetic tests demonstrated the successful application of the new algorithms with acceptable accuracy, improved storage efficiency, and without an input/output bottleneck.


Geophysics ◽  
2017 ◽  
Vol 82 (5) ◽  
pp. S359-S376 ◽  
Author(s):  
Chen Tang ◽  
George A. McMechan

Because receiver wavefields reconstructed from observed data are not as stable as synthetic source wavefields, the source-propagation vector and the reflector normal have often been used to calculate angle-domain common-image gathers (ADCIGs) from reverse time migration. However, the existing data flows have three main limitations: (1) Calculating the propagation direction only at the wavefields with maximum amplitudes ignores multiarrivals; using the crosscorrelation imaging condition at each time step can include the multiarrivals but will result in backscattering artifacts. (2) Neither amplitude picking nor Poynting-vector calculations are accurate for overlapping wavefields. (3) Calculating the reflector normal in space is not accurate for a structurally complicated reflection image, and calculating it in the wavenumber ([Formula: see text]) domain may give Fourier truncation artifacts. We address these three limitations in an improved data flow with two steps: During imaging, we use a multidirectional Poynting vector (MPV) to calculate the propagation vectors of the source wavefield at each time step and output intermediate source-angle-domain CIGs (SACIGs). After imaging, we use an antitruncation-artifact Fourier transform (ATFT) to convert SACIGs to ADCIGs in the [Formula: see text]-domain. To achieve the new flow, another three innovative aspects are included. In the first step, we develop an angle-tapering scheme to remove the Fourier truncation artifacts during the wave decomposition (of MPV) while preserving the amplitudes, and we use a wavefield decomposition plus angle-filter imaging condition to remove the backscattering artifacts in the SACIGs. In the second step, we compare two algorithms to remove the Fourier truncation artifacts that are caused by the plane-wave assumption. One uses an antileakage FT (ALFT) in local windows; the other uses an antitruncation-artifact FT, which relaxes the plane-wave assumption and thus can be done for the global space. The second algorithm is preferred. Numerical tests indicate that this new flow (source-side MPV plus ATFT) gives high-quality ADCIGs.


2014 ◽  
Vol 32 (4) ◽  
pp. 753 ◽  
Author(s):  
Rafael L. de Araújo ◽  
Reynam Da C. Pestana

ABSTRACT. Within the seismic method, in order to obtain an accurate image, it is necessary to use some processing techniques, among them the seismic migration. The reverse time migration (RTM) uses the complete wave equation, which implicitly includes multiple arrivals, can image all dips and, therefore, makes it possible to image complex structures. However, its application on 3D pre-stack data is still restricted due to the enormous computational effort required. With recent technological advances and faster computers, 3D pre-stack RTM is being used to address the imaging challenges posed by sub-salt and other complex subsurface targets. Thus, in order to balance processing cost and with image’s quality and confiability, different numeric methods are used to compute the migration. This work presents two different ways of performing the reverse time migration using the complete wave equation: RTMby interpolation and by the pseudo-analytical method. The first migrates the data with different constant velocities and interpolate the results, while the second uses modifications in the computation of the Laplacian operator inorder to improve the finite difference scheme used to approximate the second-order time derivative, making it possible to propagate the wave field stably even using larger time steps. The method’s applicability was tested by the migration of two-dimensional pre- and pos-stack synthetic datasets, the SEG/EAGE salt model and the Marmousi model. A real pre-stack data from the Gulf of Mexico was migrated successfully and is also presented. Through the numerical examples the applicabilityand robustness of these methods were proved and it was also showed that they can extrapolate wavefields with a much larger time step than commonly used.Keywords: acoustic wave equation, seismic migration, reverse time migration, pseudo-spectral method, pseudo-analytical method, pseudo-Laplacian operator. RESUMO. No método sísmico, a fim de se obter uma imagem precisa, faz-se necessário o uso de técnicas de processamento, entre elas a migração sísmica.A migração reversa no tempo (RTM) empregada aqui não é um conceito novo. Ela usa a equação completa da onda, implicitamente inclui múltiplas chegadas, consegue imagear todos os mergulhos e, assim, possibilita o imageamento de estruturas complexas. Porém, sua aplicação em problemas 3D pré-empilhamento continua endo restrita por conta do grande esforço computacional requerido. Mas, recentemente, com o avanço tecnológico e computadores mais rápidos, a migração 3D pré-empilhamento tem sido aplicada, especialmente, em problemas de difícil imageamento, como o de estruturas complexas em regiões de pré-sal. Assim, com o intuito de equilibrar o custo de processamento com a qualidade e confiabilidade da imagem obtida, são utilizados diferentes métodos numéricos para computar a migração. Este trabalho apresenta duas diferentes maneiras de se realizar a migração reversa no tempo partindo da solução exata da equação completa da onda: RTM por interpolação e pelo método pseudo-analítico. No método de interpolação, a migração é aplicada utilizando-se várias velocidades constantes, seguido de um procedimento de interpolação para obter a imagem migrada através da composição das imagens computadas a partir dessas velocidades constantes. Já no método pseudo-analítico, introduz-se modificações no cálculo do operador Laplaciano visando melhorar a aproximação da derivada segunda no tempo, que são feitas por esquemas de diferenças finitas de segunda ordem, possibilitando assim propagar o campo de onda de forma estável usando-se passos maiores no tempo. A aplicabilidadedas metodologias foi testada por meio da migração de dados bidimensionais sintéticos pré e pós-empilhamento, o modelo de domo de sal da SEG/EAGE e o modelo Marmousi. Um dado real bidimensional, adquirido no Golfo do México não empilhado, também, foi usado e migrado com sucesso. Assim, através desses exemplos numéricos, mostra-se a aplicabilidade e a robustez desses novos métodos de migração reversa no tempo no imageamento de estruturas complexas com os campos de ondas propagados com passos maiores no tempo do que os usados comumente.Palavras-chave: equação da onda, migração sísmica, migração reversa no tempo, método pseudo-espectral, método pseudo-analítico, operador pseudo-Laplaciano.


Geophysics ◽  
2017 ◽  
Vol 82 (2) ◽  
pp. S129-S141 ◽  
Author(s):  
Chen Tang ◽  
George A. McMechan ◽  
Deli Wang

Angle-domain common-image gathers (ADCIGs) obtained from reverse time migration are important for velocity and reflectivity inversion. Using the Poynting vector (PV) is an efficient way to calculate ADCIGs, but it suffers from inaccuracy and instability. A well-known reason is that a PV can give only one direction per grid point per time step, and thus it cannot calculate the individual directions of overlapping wavefields. This problem can be addressed by using a multidirectional PV (MPV), which decomposes the wavefields into several “approximate” directions and then calculates PVs for each decomposed wavefield. However, the MPV still suffers from another instability problem. The PV is the product of the time and space derivatives of the wavefield, and so it will be zero when the magnitude of the wavefield is at a local peak, which means that the directions are undefined. This leads to unstable points when the wavefields are close to a local magnitude peak, and it thus reduces the quality of the ADCIGs. We have developed two methods to stabilize the MPVs. The first method makes use of the property that the seismic wavelet has a short time duration, during which the propagation direction is stable. Thus, for each point in a decomposed wavefield, a time shift is used to locate the optimal PV during a short time duration, and the optimal location coincides with the local maximum magnitude of the time derivative. Therefore, there is a time shift between the wavefield and its corresponding PV. The second method combines the existing optical flow (OF) with the multidirectional scheme to produce a multidirectional OF (MOF). The MOF is iterative, and thus it has greater computational complexity. Numerical examples show that the time-shift MPV and MOF give more accurate ADCIGs than those using MPV only.


Geophysics ◽  
2019 ◽  
Vol 84 (3) ◽  
pp. R401-R410
Author(s):  
Hongwei Liu ◽  
Almomin Ali ◽  
Yi Luo

From a mathematical perspective, it is desirable to apply the adjoint operator for back propagating the receiver wavefield for migration of data residuals for inversion. For frequency-domain direct solvers, it is straightforward to apply the adjoint operator, whereas in most real applications of time-domain finite-difference (TDFD) stencils, the forward-propagation kernel function is reused for backward propagation for simplicity. However, when applying the exact adjoint operator, the migration result will be different, especially when strong variations exist in the velocity model. Actually, these three operators (forward, backward, and adjoint of the forward) can be expressed in matrix forms under the Born approximation for acoustic wave equations. These expressions linearly relate model perturbations to recorded seismic data. Every element in the matrices is well-defined, and all involved operations, such as the imaging condition and wavefield sampling, are included in the closed-form matrix expressions. These matrix expressions provide a platform for analyzing seismic modeling and inversion via mature linear algebra methodologies and provide clear strategies for developing computer algorithms. By analyzing the similarity of the matrix expressions, one can find that the time stepping approaches for all three operators are essentially the same. Based on this observation, a new time-marching stencil can be designed to realize the TDFD adjoint operator. Compared with traditional reverse time migration, the new method using the adjoint operator can provide better image quality, especially at sharp velocity boundaries.


Geophysics ◽  
2019 ◽  
Vol 84 (2) ◽  
pp. S41-S55 ◽  
Author(s):  
Yongchae Cho ◽  
Richard L. Gibson, Jr.

Reverse time migration (RTM) is widely used because of its ability to recover complex geologic structures. However, RTM also has a drawback in that it requires significant computational cost. In RTM, wave modeling accounts for the largest part of the computing cost for calculating forward- and backward-propagated wavefields before applying an imaging condition. For this reason, we have applied a frequency-adaptive multiscale spatial grid to enhance the efficiency of the wave simulations. To implement wave modeling for different values of the spatial grid interval, we apply a model reduction technique, the generalized multiscale finite-element method (GMsFEM), which solves local spectral problems on a fine grid to simulate wave propagation on a coarser grid. We can enhance the speed of computation without sacrificing accuracy by using coarser grids for lower frequency waves, while applying a finer grid for higher frequency waves. In the proposed method, we can control the size of the coarse grid and level of heterogeneity of the wave solutions to tune the trade-off between speedup and accuracy. As we increase the expected level of complexity of the wave solutions, the GMsFEM wave modeling can capture more detailed features of waves. After computing the forward and backward wavefield on the coarse grid, we reproject the coarse wave solutions to the fine grid to construct the RTM gradient image. Although wave solutions are computed on a coarse grid, we still obtain the RTM images without reducing the image resolution by projecting coarse wave solutions to the fine grid. We determine the efficiency of the proposed imaging method using the Marmousi-2 model. We compare the RTM images using GMsFEM with a fixed coarse mesh and a multiple frequency-adaptive coarse meshes to indicate the image quality and computational speed of the new approach.


2019 ◽  
Author(s):  
Qiancheng Liu ◽  
Jianfeng Zhang

Abstract. Least-squares reverse-time migration (LSRTM) attempts to invert for the broadband-wavenumber reflectivity image by minimizing the residual between observed and predicted seismograms via linearized inversion. However, rugged topography poses a challenge in front of LSRTM. To tackle this issue, we present an unstructured mesh-based solution to topography LSRTM. As to the forward/adjoint modeling operators in LSRTM, we take a so-called unstructured mesh-based “grid method”. Before solving the two-way wave equation with the grid method, we prepare for it a velocity-adaptive unstructured mesh using a Delaunay Triangulation plus Centroidal Voronoi Tessellation (DT-CVT) algorithm. The rugged topography acts as constraint boundaries during mesh generation. Then, by using the adjoint method, we put the observed seismograms to the receivers on the topography for backward propagation to produce the gradient through the cross-correlation imaging condition. We seek the inverted image using the conjugate gradient method during linearized inversion to linearly reduce the data misfit function. Through the 2D SEG Foothill synthetic dataset, we see that our method can handle the LSRTM from rugged topography.


2021 ◽  
Vol 9 ◽  
Author(s):  
Chuang Xie ◽  
Peng Song ◽  
Xishuang Li ◽  
Jun Tan ◽  
Shaowen Wang ◽  
...  

A gradient preconditioning approach based on transmitted wave energy for least-squares reverse time migration (LSRTM) is proposed in this study. The gradient is preconditioned by using the energy of “approximate transmission wavefield,” which is calculated based on the non-reflecting acoustic equation. The proposed method can effectively avoid a huge amount of calculation and storage required by the Hessian matrix or approximated Hessian matrix and can overcome the influence of reflected waves, multiples, and other wavefields on the gradient in gradient preconditioning based on seismic wave energy (GPSWE). The numerical experiments, compared with that using GPSWE, show that LSRTM using the gradient preconditioning based on transmitted wave energy (GPTWE) can significantly improve the imaging accuracy of deep target and accelerate the convergence rate without trivial increased calculation.


Sign in / Sign up

Export Citation Format

Share Document