Marchenko focusing for target-oriented data processing and full-waveform inversion

Abstract

Measuring wave motion on the surface of an unknown medium can help us infer the physical properties of the medium through which waves propagate without intrusion. In many applications, multiple scattering and high computational cost prevent state-of-the-art data processing techniques from delivering a high-quality image of the medium, especially when it is complex. The recently developed Marchenko focusing technique has potential to overcome these challenges by performing data processing and imaging in a target-oriented fashion. Superior to conventional redatuming methods, Marchenko focusing retrieves multiply-scattered waves from single-sided reflection data with minimal a priori knowledge of medium properties. By iteratively re-emitting a time-reversed and time-windowed wavefield from a single side of the unknown medium, we can focus the wavefield at a prescribed virtual source position inside the medium. The redatumed wavefields in response to the virtual source can be used for imaging free from artefacts related to multiple scattering and interference from the medium surrounding the target. In this thesis, I work on practical implementation and applications of the Marchenko focusing technique. I first review the theory and methodology of the standard Marchenko scheme using 1D and 2D acoustic numerical examples. Then, I investigate its potential application to the hDVS (heterodyne distributed vibration sensing) technology, which records wavefields semi-continuously in space using an optical fibre. The hDVS signal is focused by the Marchenko method in the optical frequency regime, based on an analytical hDVS model. With a hypothesis of strong scattering inside the fibre, the redatumed hDVS signal may contribute to improving strain estimation. Next, I implement Marchenko focusing in a dissipative medium by both numerical modelling and laboratory implementation. To accommodate dissipation which is not accounted for in the standard Marchenko scheme, double-sided reflection data are required to derive an effectual reflection response. Physical focusing of the sound wave is achieved in a 1D variable-diameter tube. Finally, I conduct target-oriented FWI (full-waveform inversion) by using Marchenko focusing to extrapolate the wavefields from the acquisition surface to the vicinity of the target. Based on the representation theorem of the convolution type, a local forward modelling operator couples the target and the surrounding medium acoustically. I invert for the optimal target model, whose seismic response to the Marchenko retrieved areal sources best matches the Marchenko retrieved observed data. Given a sufficiently accurate initial velocity model, local FWI saves computational cost by a factor of ten in a 2D numerical test without sacrificing accuracy compared to conventional full-domain FWI. This thesis further establishes the applicability of Marchenko focusing to better characterize the physical properties of the medium of interest.