Applied 3D full-waveform inversion: increasing the resolution and depth penetration
File(s)
Author(s)
Silverton, Akela Tian Theresa
Type
Thesis
Abstract
High-resolution velocity models, at near surface and deeper reservoir depths are produced
with three-dimensional, acoustic, anisotropic, full-waveform inversion. Industry experts show
eager interest in the development of this technology with a drive to push its application to
reflection-dominated streamer datasets as well as ocean-bottom-node datasets in geologically
complex environments. Here, a robust methodology employing the use of key strategies to
address the inversion of such datasets, attaining increased resolution and depth penetration, is
explored.
Synthetic tests were undertaken to exploit the use of reflected energy. Key strategies: muting
of direct arrivals, time windowing, and layer-stripping, all produced highly resolved, full waveform
inversion models. These strategies have been incorporated into inversion schemes
focusing solely on reflection targets. Strategies to further improve model resolution for field
datasets were then investigated. Close examination of a full-waveform inversion model for a
shallow-water ocean-bottom-node dataset, revealed a systematic mismatch between the
observed and predicted data. After conducting a series of tests, it was illustrated that
systematic errors in the starting model, source wavelet, incomplete convergence, or an
inadequate finite-difference mesh did not cause the mismatch. Instead, inadequacies in the
physics used during inversion are believed to be the cause. The introduction of an offsetvariable
density scheme during inversion, compensated efficiently and heuristically for these
inaccuracies, removing the mismatch and increasing the model resolution.
The sensitivity of full-waveform inversion to local minima, where the computed model is
stuck away from the real global-minimum solution and further iterations of the optimisation
bring no reward, was kept in mind during the inversion of two deep-water ocean-bottom node
datasets. Thus, full-waveform inversion was undertaken using conditioned data
obtained through adaptive matching, incorporating higher frequencies and a greater weight
on reflected energy, valuably pushing the limits of resolution and depth penetration of the
update. The use of all these robust methodologies improved the travel-time match; better
flattened common-image gathers giving a closer fit to well logs and an improvement in the
pre-stack depth-migrated image. Effectively, the reflectivity was non-linearly migrated into
the velocity model via the inversion acting on raw unprocessed waveforms. Thus, full waveform
inversion can eventually replace conventional processing and migration – all that
is needed, is a full-bandwidth velocity model.
with three-dimensional, acoustic, anisotropic, full-waveform inversion. Industry experts show
eager interest in the development of this technology with a drive to push its application to
reflection-dominated streamer datasets as well as ocean-bottom-node datasets in geologically
complex environments. Here, a robust methodology employing the use of key strategies to
address the inversion of such datasets, attaining increased resolution and depth penetration, is
explored.
Synthetic tests were undertaken to exploit the use of reflected energy. Key strategies: muting
of direct arrivals, time windowing, and layer-stripping, all produced highly resolved, full waveform
inversion models. These strategies have been incorporated into inversion schemes
focusing solely on reflection targets. Strategies to further improve model resolution for field
datasets were then investigated. Close examination of a full-waveform inversion model for a
shallow-water ocean-bottom-node dataset, revealed a systematic mismatch between the
observed and predicted data. After conducting a series of tests, it was illustrated that
systematic errors in the starting model, source wavelet, incomplete convergence, or an
inadequate finite-difference mesh did not cause the mismatch. Instead, inadequacies in the
physics used during inversion are believed to be the cause. The introduction of an offsetvariable
density scheme during inversion, compensated efficiently and heuristically for these
inaccuracies, removing the mismatch and increasing the model resolution.
The sensitivity of full-waveform inversion to local minima, where the computed model is
stuck away from the real global-minimum solution and further iterations of the optimisation
bring no reward, was kept in mind during the inversion of two deep-water ocean-bottom node
datasets. Thus, full-waveform inversion was undertaken using conditioned data
obtained through adaptive matching, incorporating higher frequencies and a greater weight
on reflected energy, valuably pushing the limits of resolution and depth penetration of the
update. The use of all these robust methodologies improved the travel-time match; better
flattened common-image gathers giving a closer fit to well logs and an improvement in the
pre-stack depth-migrated image. Effectively, the reflectivity was non-linearly migrated into
the velocity model via the inversion acting on raw unprocessed waveforms. Thus, full waveform
inversion can eventually replace conventional processing and migration – all that
is needed, is a full-bandwidth velocity model.
Version
Open Access
Date Issued
2015-10
Date Awarded
2016-03
Copyright Statement
Attribution NoDerivatives 4.0 International Licence (CC BY-ND)
Advisor
Warner, Michael
Morgan, Joanna
Publisher Department
Earth Science & Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)