The process of full-waveform inversion (FWI) seeks a model of the Earth’s sub
surface that produces simulated data to fit the observed data. The resolution of the
model can be both complex and costly to meet such an objective. Resolving the
reservoir is even more challenging as it requires an accurate representation of the
physics throughout. Although FWI for diving waves has been successfully applied,
the reservoir located at depth requires FWI to take advantage of the reflections. How
ever, major issues are alongside the value of reflections, such as limited illumination,
difficulties in recovering lowwavenumbers of the model, trade-offs between the model
parameters, etc, which hinders its applications so far.
Recent studies on reflection waveform inversion (RWI) revealed the unique po
tential of reflections in illuminating the deep model building. RWI identifies the
nonzero-offset data mismatch and produces low-to-middle wavenumber model up
dates along the reflection wavepath, which brings unprecedented robustness to FWI.
FWI is therefore exposed to a better chance of resolving the deep targets within its
own framework. However, RWI makes FWI even more computationally intractable.
Alternatively, we introduce redatuming to FWI applications, aiming at retrieving
survey-sinking virtual data to focus our inversion on the target zones. It improves the
efficiency of entire loop of our inversion and, meanwhile, reduce the trade-offs of FWI
implemented on the entire model. Hereby, we split FWI into sequential optimization
problems consisting of overburden estimation, virtual data retrieval and target inver
sion. We exploit the advantages of the reflections produced by our datum modeling
5 to improve the robustness of the overburden inversion. The resulting macro model is
refined by follow-up FWI on relatively low-frequency bands to save the computation.
The virtual dataset is calculated using an extended imaging condition. We specially
summarize the reflection modeling and imaging process in terms of datuming. Higher
frequencies are involved in retrieving the virtual data that are substituted into the
target inversion, which allows adoption of a finer grid and some enhanced treatment
to satisfy the demand for high-resolution and multi-parameter delineation. The po
tential applications are demonstrated by examples, with limitations and future work
suggested in the last chapter.
Date of Award | Feb 2020 |
---|
Original language | English (US) |
---|
Awarding Institution | - Physical Sciences and Engineering
|
---|
Supervisor | Tariq Ali Alkhalifah (Supervisor) |
---|