Least-squares full-wavefield migration

Abstract

We present a least-squares solution for depth migration of the full reflected wavefield. The algorithm combines primary and high-order reflected energy and significantly enhances the image illumination and resolution compared to those of conventional migration. Least-squares full-wavefield migration (LS-FWM) directly computes the earth's reflectivity, thereby avoiding crosstalk noise often observed in imaging using high-order reflections. Iteratively solving an inversion problem is computationally intensive and can suffer from instability issues; however, we develop an efficient least-squares procedure by combining a fast and accurate one-way wave-equation propagator with an effective linear inversion solver. An advanced regularization method is also employed to stabilize the inversion by controlling the over-fitting problem. Successful applications to both synthetic and field data examples demonstrate that LS-FWM greatly improves the imaging illumination and resolution compared to conventional migration.

Full-wavefield migration by inversion

“…Each inversion iteration consists of one Born modeling (Cohen et al, 1966) and one migration. Advanced regularization techniques are also employed to composite an efficient and stable migration inversion algorithm (Lu et al, 2018).…”

“…However, both strategies are sub-optimal, as the crosstalk in FWM is created by the interference in the blended migration algorithm. Conversely, the inversion algorithm solves a Least-Squares minimization problem, which excludes the interference terms existing in a one-step migration (Davydenko and Verschuur, 2014;Tu, 2015;Lu et al, 2018). Least-Squares Full Wavefield Migration (LS-FWM) computes a reflectivity image that explains the observed data without involving interference noise, and automatically balances the contribution from each component (primaries and all orders of multiple reflections).…”

Crosstalk attenuation for imaging of multiples based on angle gather residual moveout analysis

“…Many methods have been proposed to attenuate crosstalk for imaging of multiples. One category is to remove crosstalk during the migration, such as the least-squares migration (LSM) (Berkhout, 2014;Ordoñez et al, 2014;Zhang and Schuster, 2014;Tu and Herrmann, 2015;Wong et al, 2015;Lu et al, 2018;Lu et al, 2021) and imaging using controlled-order multiples (Liu et al, 2016). However, such methods usually require a huge amount of computation or complete separation of different orders multiples.…”

An Efficient and Stable High‐Resolution Seismic Imaging Method: Point‐Spread Function Deconvolution

“…Currently, the data‐domain LSM has been implemented using a variety of different ray‐based and wave equation migration methods (Dai et al., 2011, 2012; Xue et al., 2016; Yang & Zhang, 2019; Yang et al., 2018; Zhang & Schuster, 2014), and extended to (an)elastic and anisotropic media (Dutta & Schuster, 2014; Feng & Schuster, 2017; Ren et al., 2017; Yang et al., 2020). These developments make LSM a robust imaging method and so has been gradually applied to field data processing (Latter et al., 2018; Lu et al., 2018; Wong et al., 2011; Zhang et al., 2015), which produces superior imaging results in comparison to traditional adjoint migration. However, many iterations of data‐domain LSM require a large number of forward simulations and adjoint migrations to converge to a good solution, which is still expensive and difficult to be widely used for large‐scale models in production under current computational capacity.…”