SHAPE BASED METHODS FOR SEISMIC FULL-WAVEFORM INVERSION

Abstract

This thesis concentrates on a shape based approach for seismic full waveform inversion, especially a novel level set based shape estimation method for elas- tic waveform inversion. Full waveform inversion is a numerical data processing technique aiming at the reconstruction of subsurface structure of the earth using collected seismic reflective data. However, traditional techniques using an acous- tic or Helmholtz wave equation as forward model are faced with the limitation of using simulated non-elastic wave data to numerically fit elastic waveform data, which is physically incorrect and practically prone to obtaining wrong estimates; the correct scheme for modelling seismic waves is using an elastic wave equa- tion. We construct an elastic waveform inversion algorithm using a symmetric- hyperbolic scheme, and a time-reversal adjoint-state method; in addition, we introduce a Sobolev gradient method as a regularization method, with the goal to smooth the gradient function and thereby obtain more regular boundaries of the reconstructed shapes. However, the procedure of elastic waveform in- version is a multi-parameter estimation, which will lead to the numerical error of ‘cross-talk’; this phenomenon is particularly severe in high-contrast situations and irregular shape boundary reconstruction problems, as expected to face in salt dome estimation problems. Therefore we introduce a shape based method using a level set technique to tackle specific seismic reconstruction problems, instead of more traditional pixel-based schemes. We also introduce a stochastic gradient descent method as an alternative to traditional gradient line search techniques for level sets, in order to increase efficiency and to avoid certain local minima in large-scale inversion problems. In addition, we introduce an additional inte- grated internal-value reconstruction scheme; this will prove to be an interesting and possibly necessary expansion of the shape based approach in order to deal with more realistic 3D seismic reconstruction problems.

Date of Award	31 Dec 2019
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Oliver Dorn (Supervisor) & Sean Holman (Supervisor)