Loading...
Full waveform inversion with reflected waves for acoustic 2D VTI media
Pattnaik, Sonali
Pattnaik, Sonali
Citations
Altmetric:
Advisor
Editor
Date
Date Issued
2017
Date Submitted
Keywords
Collections
Research Projects
Organizational Units
Journal Issue
Embargo Expires
Abstract
With the recent advances in seismic data acquisition, such as wide-azimuth, long-offset surveys and low-frequency sources, full-waveform inversion (FWI) has become an efficient tool in building high-resolution subsurface models. Conventional FWI relies mainly on diving waves to update the low-wavenumber components of the background model. However, such FWI algorithms may fail to provide a satisfactory model update for regions probed primarily by reflected waves. This typically occurs for deep target zones where the conventional FWI updates mostly the high-wavenumber model components due to the absence of diving waves. Reflection waveform inversion (RWI) has been developed to retrieve the intermediate-to-long wavelength model components in those deeper regions from reflection energy. In this thesis, I highlight the limitations of conventional waveform inversion when applied to reflections-dominated seismic data and propose a new implementation of RWI for acoustic VTI (transversely isotropic with a vertical symmetry axis) media. I extend the idea of scale separation between the background and perturbation models to VTI media and use an optimized parameterization to mitigate parameter trade-offs in RWI. The proposed workflow repeatedly alternates between updating the long-wavelength model components by fixing the perturbation model and the shorter-wavelength, migration-based reflectivity update. I develop an hierarchical two-stage approach that operates with the P-wave zero-dip normal-moveout velocity $V_{\rm nmo}$ and anisotropy coefficients $\delta$ and $\eta$. At the first stage, $V_{\rm nmo}$ is estimated by applying the Born approximation to a perturbation model in $\delta$ to compute the corresponding reflection data. Although the algorithm does not invert for $\delta$, this parameter helps improve the amplitude fit for the employed acoustic model that ignores the elastic nature of the subsurface. At the second stage, the parameter $\eta$, which can be constrained by far-offset data, is estimated from the obtained perturbation model in $V_{\rm nmo}$. The proposed 2D algorithm is tested on a horizontally layered VTI medium and the VTI Marmousi model. Application of a temporal correlation-based objective function significantly improves recovery of the long-wavelength $\eta$-component, as demonstrated on the Marmousi model.
Associated Publications
Rights
Copyright of the original work is retained by the author.