An extension of least-squares redatuming: Simultaneous reconstruction of overburden reflectivities and virtual data
Introduction
Redatuming is an operation applied to seismic data that translates the positions of sources or/and receivers and thus makes the reconstructed seismic traces appear to be acquired with virtual sources or/and receivers at a new datum [5], [40]. It can mitigate the effects of irregular topography and complex geological structures in the overburden such as strong heterogeneities [3], [45]. As a result, the redatumed data can be used to apply an imaging process where the wave propagation and scattering above the datum level can be completely neglected (e.g., [47]). Moreover, this can naturally lead to a target-oriented approach that alleviates the computational burden associated with least-squares migration (LSM) and full waveform inversion (FWI) by limiting the inversion process to only a portion of the subsurface where an accurate and high-resolution characterization is needed, for instance at reservoir level, e.g., Yuan et al. [46], Cui et al. [10], Guo and Alkhalifah [14].
Seismic redatuming generally requires an estimate of the transmission response of the overburden, which can either be computed with a macrovelocity model of the subsurface or retrieved from the recorded data. This leads to two types of methodology: model-based and correlation-based [31]. On the one hand, given a macrovelocity model, the acquired data can be extrapolated to the new datum with a Kirchhoff integral or wave-equation continuation operator [6], [20], [23], [2], [12], [45]. The model-based redatuming approaches effectively correct the seismic traces using traveltimes determined by the macrovelocity model. On the other hand, when receivers are deployed below the overburden (e.g., in a deviated or horizontal well) to measure the transmission response directly, correlation-based redatuming can be applied to construct virtual source/receiver gathers, without any prior knowledge of the medium [9], [30], [4], [38]. It requires applying weighted correlations of the traces with each other, followed by summation over all sources (or sometimes receivers), and thus use natural-phase information in the data to time shift the weighted seismic traces so that they appear to be excited by sources (or recorded by receivers) on a new datum. A theoretical overview of model- and correlation-based redatuming methods has been presented by Schuster and Zhou [31]. If buried receivers are not available, novel approaches based on the Marchenko method for wavefield focusing have been presented for redatuming of single-sided data. In this context, the Green's function is iteratively retrieved using only the reflection responses of the medium and an estimate of the first arrival at the recording surface from the virtual source [29], [7], [42], [27]. Alternatively, internal scattering can be imaged with the background Green's function from the surface to each subsurface location in a generalized internal multiple imaging procedure [48], [1]. More or less, all these methods are affected by recording aperture and spatial sampling.
In theory, redatuming can be formulated as an inverse modelling, in which the virtual data on the datum constitutes the model parameter to be estimated. Because the acquisition aperture is finite and the spatial sampling may be irregular, conventional redatuming methods only form an approximate inverse and produce many artifacts in the solution space. Ji [17] and Ferguson [13] suggested to suppress the artifacts and improve the reconstruction accuracy of wavefield extrapolation-based redatuming [6] through a least-squares optimization process. Accordingly, Xue and Schuster [43] implemented a least-squares interferometric redatuming scheme in the context of vertical seismic profile (VSP) experiment. Alternatively, Wapenaar and van der Neut [41] extended redatuming with a multidimensional deconvolution (MDD) based on point-spread functions (PSFs). Recently new representations using PSFs provide adaptation of the iterative Marchenko scheme for imperfectly sampled data [16].
Redatuming aims to completely remove the effects of a complex overburden. Even if surface and internal multiples are ignored (e.g., for deep datum levels) for the moment, one must properly account for transmission and primary reflections in the overburden media. The inaccuracy of the overburden macrovelocity model prevents a reasonable reconstruction of the virtual data. By formulating the redatuming operation with an extended Born representation, Guo and Alkhalifah [15] implemented a simultaneous inversion of the overburden long-wavelength velocity model and the virtual data at the datum. In fact, the overburden short-wavelength velocity structures (or reflectivities) could affect model-based least-squares redatuming (LSR). As we will demonstrate in this paper, imprints of the overburden reflections or/and diffractions generated very close to the datum could hinder the correct measurement of the data misfit and impede the iterative inversion. In the presence of complex and irregular geologic bodies (e.g., salt body and igneous rock) just above or cross the datum, removing these overburden reflections and diffractions, for instance with a delicate time window (e.g., [42]), is not trivial. To address this problem, we present a simultaneous inversion for the overburden reflectivities and the virtual data in LSR, given the overburden macrovelocity model as a prior.
The structure of this paper is given as follows: First, we describe the forward problem of this double-parameter LSR based on the convolution-type reciprocity theorem and the first-order scattering theory. Then, instead of solving the normal equation that explicitly incorporate the inverse Hessian, we simultaneously update the two classes of model parameter with a gradient-type iterative inversion based on the L2-norm misfit function. In this context, the functional gradients are derived using the adjoint-state method and constructed through correlation of the forward and adjoint wavefields. To enhance the robustness and suppress the numerical artifacts, we apply sparsity constraints on the overburden reflectivities and the virtual data using a split Bregman method. Finally, synthetic and field data examples are presented to demonstrate the effectiveness of the proposed approach to tackle irregular sampling and overburden effects.
Section snippets
The forward problem
In this study, we focus on redatuming of a scalar wavefield using a constant-density acoustic wave equation. For simplicity, we neglect free-surface effects and internal multi-scattering for the time being. As illustrated in Fig. 1, a datum plane divides the subsurface medium into the overburden part () and the underlying part (), and both of them can be split into a smooth background (macrovelocity) and a rough perturbation (reflectivity). We assume that the seismograms at the surface
Numerical examples
We present two synthetic and one field data examples to demonstrate the proposed method. The first example reduces to a single-parameter LSR problem, because the toy model includes no short-wavelength perturbation in the overburden media. This helps us to illustrate how the constrained optimization can handle irregular and limited observation, and mitigate the ringing artifacts during the least-squares iterations. The second example on SEG Marmousi model demonstrates the benefit of
Discussion
Compared with the single-parameter LSR, the double-parameter LSR only requires a very small amount of additional computation cost, because we can update the overburden reflectivity model using the forward and adjoint wavefields that have been computed for redatuming. Taking the overburden LSM as a reference, double-parameter LSR requires non-zero lag correlation of the forward and adjoint wavefields to update the virtual data, of which the computational time is negligible compared with the cost
Conclusion
We have presented an extended LSR approach that simultaneously updates the overburden reflectivities and the virtual data at the datum. Compared with the conventional LSR method that only retrieves the virtual data, this extension adds a little more computation cost but effectively suppresses the spurious events or/and contamination from the overburden reflections or/and diffractions very close to the datum. As a by-product, it produces a least-squares migration for the overburden reflections.
CRediT authorship contribution statement
Feng Zhu: Conceptualization, Methodology, Software, Validation, Visualization, Writing – original draft. Jiubing Cheng: Conceptualization, Data curation, Funding acquisition, Project administration, Supervision, Writing – review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (48)
- et al.
Subsurface wavefields based on the generalized internal multiple imaging
Geophys. J. Int.
(2019) - et al.
Straight-rays redatuming: a fast and robust alternative to wave-equation-based datuming
Geophysics
(2006) - et al.
Virtual source: new method for imaging and 4D below complex overburden
- et al.
The virtual source method: theory and case study
Geophysics
(2006) Wave-equation datuming
Geophysics
(1979)Wave-equation datuming before stack
Geophysics
(1984)- et al.
Connection of scattering principles: a visual and mathematical tour
Eur. J. Phys.
(2012) - et al.
Data-driven wavefield focusing and imaging with multidimensional deconvolution: numerical examples for reflection data with internal multiples
Geophysics
(2014) Synthesis of a layered medium from its acoustic transmission response
Geophysics
(1968)- et al.
Target-oriented full-waveform inversion using marchenko redatumed wavefields
Geophys. J. Int.
(2020)
Source-receiver wave field interferometry
Phys. Rev. E
Fast 3D target-oriented reverse-time datuming
Geophysics
Regularization and datuming of seismic data by weighted, damped least squares
Geophysics
Target-oriented waveform redatuming and high-resolution inversion: role of the overburden
Geophysics
Datum-based waveform inversion using a subsurface-scattering imaging condition
Geophysics
Adaptation of the iterative Marchenko scheme for imperfectly sampled data
Geophys. J. Int.
Least squares imaging, datuming, and interpolation using the wave equation
Iterative asymptotic inversion in the acoustic approximation
Geophysics
An iterative inversion of back-scattered acoustic waves
Geophysics
Wavefield extrapolation by nonstationary phase shift
Geophysics
Improving the virtual source method by wavefield separation
Geophysics
Understanding subsalt illumination through ray-trace modelling, Part I: Simpe 2-D salt models
Geophysics
Rigorous redatuming
Geophys. J. Int.
An iterative regularization method for total variation-based image restoration
Multiscale Model. Simul.
Cited by (3)
Dynamic seismic imaging driven by logging while drilling in ultra-deep targets
2023, Acta Geophysica SinicaElimination of free-surface multiples using least-squares redatuming, Part I: Ocean-bottom seismic data
2022, Acta Geophysica Sinica