Modeling and data-driven isolation of two-way wavefield constituents

Patrick Elison,Dirk-Jan Van Manen,Johan O A Robertsson,Nele Börsing

doi:10.1190/geo2019-0394.1

Abstract

Synthesizing individual wavefield constituents (such as primaries, first-order scattering, and free-surface or internal multiples) is important in the development of seismic data processing algorithms, for instance, for seismic multiple removal and imaging. A range of methods that allow for the computation of such wavefield constituents exist, but they are generally restricted to relatively simple, horizontally layered media. For wave simulations on more complex models, a straightforward and performant alternative are finite-difference methods. They are, however, generally not perceived as being capable of delivering isolated wavefield constituents. Based on recent advances, we found how this can be achieved for (nonhorizontally) piecewise constant layered media. For example, we were able to accurately retrieve the isolated direct arrival of the transmission response (including tunneled waves), primary reflection data (without internal multiples), and all events related to a single (or multiple) interface(s) in a medium. Our methods required detailed knowledge of discretized medium parameters. Alternatively, if a medium is known only implicitly via recordings of reflection data, interface-related events can still be isolated through a combination of subdomain-related wavefields. We found how Marchenko redatuming can be used to derive these, which enables data-driven identification (and removal) of interface-related events from surface data.

Highlights

Full-waveform methods, such as finite-difference (FD) and finite-element (FE) methods, can be used for complex 2D and 3D wave propagation modeling
A range of methods that allow for the computation of such wavefield constituents exist, but they are generally restricted to relatively simple, horizontally-layered media
In the following we show how first-order reflected and transmitted wavefield constituents can be expressed analytically, and how their isolation is achieved using a sequence of MPS injection and recording steps

Summary

INTRODUCTION

Full-waveform methods, such as finite-difference (FD) and finite-element (FE) methods, can be used for complex 2D and 3D wave propagation modeling. We are using a transmission across the target interface that contains erroneous relative amplitudes when computing the data shown in Figure 10(f) (where the transmission is implicitly contained in the Marchenko-derived R2∪,2), but we are using the correct transmission response for the derivation of panel (g) Note that this problem cannot be resolved by comparing the surface data without the target interface [panel (g)] directly with the input reflection data [panel (a)], since the erroneous offset-dependent amplitudes in the Marchenko redatumed data are affecting all other wavefields used to synthesize the data in panel (g) (and other events would not cancel correctly when taking the difference). Note the reduced offset in these data with respect to the model in Figure 6 which owes to the absorbing boundaries at the lateral model sides

DISCUSSION

CONCLUSIONS