next up previous [pdf]

Next: Method Up: Merzlikin et al.: Anisotropic Previous: Merzlikin et al.: Anisotropic

Introduction

Diffraction imaging is able to highlight subsurface discontinuities associated with channel edges, fracture swarms and faults. Since diffractions are usually weaker than reflections (Klem-Musatov, 1994) and have lower signal-to-noise ratio, robust diffraction extraction is of utmost importance for the imaging of subtle discontinuities. A number of diffraction imaging methods have been developed and can be classified based on the separation technique being employed. Methods based on optimal stacking of diffracted energy and suppression of reflections are described by Kanasewich and Phadke (1988), Landa and Keydar (1998), Berkovitch et al. (2009), Dell and Gajewski (2011), Tsingas et al. (2011) and Rad et al. (2014). Wavefield separation methods aim to decompose conventional full-wavefield seismic records into different components representing reflections and diffractions (Dell et al., 2019b; Schwarz and Gajewski, 2017; Taner et al., 2006; Papziner and Nick, 1998; Schwarz, 2019; Fomel et al., 2007). Decomposition can be carried out in the common-image gather domain (Reshef and Landa, 2009; Silvestrov et al., 2015; Klokov and Fomel, 2012). Other authors (Koren and Ravve, 2011; Kozlov et al., 2004; Popovici et al., 2015; Klokov and Fomel, 2013; Moser and Howard, 2008) modify migration kernel to eliminate specular energy coming from the first Fresnel zone and image diffractions only. For the methods involving migration, diffraction extraction quality becomes dependent on the velocity model accuracy. Numerous case studies show that diffraction images carry valuable additional information for seismic interpretation (Merzlikin et al., 2017b; Pelissier et al., 2017; Tyiasning et al., 2016; de Ribet et al., 2017; Sturzu et al., 2015; Moser et al., 2020; Foss et al., 2018; Zelewski et al., 2017; Glöckner et al., 2019; Klokov et al., 2017a; Burnett et al., 2015; Schoepp et al., 2015; Montazeri et al., 2020; Koltanovsky et al., 2017; Klokov et al., 2017b).

Consideration of diffraction phenomena in 3D (Keller, 1962; Klem-Musatov et al., 2008; Hoeber et al., 2010) requires taking into account edge diffractions. Due to lateral symmetry they kinematically behave as reflections when observed along the edge and as diffractions when observed perpendicular to the edge (Moser, 2011). Thus, edge diffraction signature is neither a ``pure'' reflection nor a ``pure'' diffraction but rather a combination of both and therefore requires a special processing procedure to be emphasized. Serfaty et al. (2017) separate reflections, tip and edge diffractions and noise using principal component analysis and deep learning. Klokov et al. (2011) and Bona and Pevzner (2015) investigate 3D signatures of different types of diffractors. Alonaizi et al. (2013) and Merzlikin et al. (2017a) propose workflows to properly process energy diffracted on the edge. Keydar and Landa (2019) propose a method for edge diffraction imaging based on time-reversal principle and the stacking operator directly targeting edge diffractions. Dell et al. (2019a) extract edge diffraction responses from full wavefield data by analyzing amplitude distribution along different azimuths on a 3D prestack time Kirchhoff migration stacking surface. Znak et al. (2019) develop a common-reflection-surface-based framework for distinguishing between point and edge diffractions and separating them from reflections.

Separation of reflections and diffractions can be done as a part of least-squares migration (Nemeth et al., 1999; Ronen and Liner, 2000). Harlan et al. (1984) pioneer in separating diffractions from noise by comparing observed data and data modeled from a migration image in a least-squares sense. Merzlikin and Fomel (2016) perform least-squares migration chained with plane-wave destruction and path-summation integral filtering and enforce sparsity in a diffraction model. Merzlikin et al. (2019) extend the approach and simultaneously decompose the input wavefield into reflections, diffractions and noise. Decker et al. (2017) denoise diffractions by applying semblance-based weights estimated in dip-angle gather (DAG) domain. Yu et al. (2016) utilize common-offset Kirchhoff least-squares migration with a sparse model regularization to emphasize diffractions. Yu et al. (2017a) extract diffractions based on plane wave destruction and dictionary learning for sparse representation. Yu et al. (2017b) use two separate modeling operators for diffractions and reflections and impose a sparsity constraint on diffractions. Sparse inversion is an efficient tool to perform extraction and denoising of diffractions since scatterers have spiky and intermittent distribution. However, a simple sparsity constraint does not account for the signature of the energy scattered on the edge, which is kinematically similar to a reflection when observed along the edge, and thus can distort it.

We combine sparsity constraints and structure-oriented smoothing in the form of shaping regularization (Fomel, 2007) to highlight edge diffractions and account for their signature. Structure-oriented smoothing performs smoothing along the edges emphasizing their continuity (Hale, 2009). Sparsity constraints imposed by thresholding in the model space force the model to describe the data with the fewest parameters and therefore denoise and emphasize edge diffraction signatures observed perpendicular to the edge. Thus, we properly account for edge diffraction kinematic behavior for both parallel and perpendicular to the edge directions.

For forward modeling we use a chain of operators introduced by Merzlikin and Fomel (2016). We extend this workflow to three-dimensions and modify reflection destruction operator to account for an edge diffraction signature by suppressing reflected energy perpendicular to edges. Edge orientations are determined through a plane-wave destruction based structure tensor (Merzlikin et al., 2017a). We start with a method introduction, then validate its performance on a synthetic, on a noisy marine field dataset and on a land field dataset by separating edge diffractions from reflections and noise.


next up previous [pdf]

Next: Method Up: Merzlikin et al.: Anisotropic Previous: Merzlikin et al.: Anisotropic

2021-02-24