3-D Frequency-domain Seismic Viscoelastic Anisotropic Wave Modelling and Full-waveform Inversion with an Efficient Solver for Imaging Subsurface Geological Structures

Abstract

Full waveform inversion (FWI) has emerged as a prominent research topic in recent decades due to its potential for high-resolution seismic imaging. The accuracy and reliability of FWI results depend heavily on the capability and efficiency of the forward modelling. However, so far most of previous studies rely on certain approximations, e.g., acoustic or viscoacoustic models. The reason for that is either because of failing to deal with the artificial boundary edge effect in anisotropic or viscoelastic anisotropic media in the frequency-domain or being stuck in the bottleneck of the huge computational load for solving the large, sparse linear equation system in the frequency and 3-D spatial domain. Thus, in this research an innovative numerical technique—the new Generalized Stiffness Reduction Method is first applied to the 3-D frequency-domain seismic wave modelling in heterogeneous, viscoelastic anisotropic media and the results have proven its validity. Parallel direct and iterative solvers have been applied and evaluated in terms of the modelling accuracy and computational efficiency as well as resources efficiency. Particularly, two fast parallel iterative solvers have been developed for multi-source seismic wave modelling. Thus, this research provides powerful computer software for fast 3-D frequency-domain seismic forward modelling in various media (elastic or viscoelastic, isotropic or anisotropic), with complex geological structures and free surface topography. More importantly, multiple sources can be simulated and solved simultaneously. Then the research has derived 3-D frequency-domain Fréchet sensitivity kernels of the observed full-waveform seismograms in viscoelastic isotropic and viscoelastic Transversely Tilted Isotropic (TTI) media. A fully parallel framework is developed for the fast computation of 3-D forward modelling, 3-D sensitivity kernels and data misfit minimization. Based on the FWI framework, Multiple FWI experiments have been performed and all the model parameters (including density, elastic moduli, Q-factors, azimuth and dip angles) have been reconstructed successfully, demonstrating the effectiveness of the methodology.

Date of Award	20 May 2025
Original language	American English
Supervisor	Aisha Alsuwaidi (Supervisor)