Felix Oghenekohwo

Using Common Information in Compressive Time-lapse Full-waveform Inversion

The use of time-lapse seismic data to monitor changes in the subsurface has become standard practice in industry. In addition, full-waveform inversion has also been extended to time-lapse seismic to obtain useful time-lapse information. The computational cost of this method are becoming more pronounced as the volume of data increases. Therefore, it is necessary to develop fast inversion algorithms that can also give improved time-lapse results. Rather than following existing joint inversion algorithms, we are motivated by a joint recovery model which exploits the common information among the baseline and monitor data. We propose a joint inversion framework, leveraging ideas from distributed compressive sensing and the modified Gauss-Newton method for full-waveform inversion, by using the shared information in the time-lapse data. Our results on a realistic synthetic example highlight the benefits of our joint inversion approach over a parallel inversion method that does not exploit the shared information. Preliminary results also indicate that our formulation can address time-lapse data with inconsistent acquisition geometries.

Compressed Sensing in 4-D Marine - Recovery of Dense Time-lapse Data from Subsampled Data without Repetition

We present an extension of our *time-jittered* marine acquisition for time-lapse surveys by working on more realistic field acquisition scenarios by incorporating *irregular* spatial grids without insisting on repeatability between the surveys. Since we are always subsampled in both the baseline and monitor surveys, we are interested in recovering the densely sampled baseline and monitor, and then the (complete) 4-D difference from subsampled/incomplete baseline and monitor data.

Economic time-lapse seismic acquisition and imaging - reaping the benefits of randomized sampling with distributed compressive sensing

This thesis presents a novel viewpoint on the implicit opportunities randomized surveys bring to time-lapse seismic which is a proven surveillance tool for hydrocarbon reservoir monitoring. Time-lapse (4D) seismic combines acquisition and processing of at least two seismic datasets (or vintages) in order to extract information related to changes in a reservoir within a specified time interval. The current paradigm places stringent requirements on replicating the 4D surveys, which is an expensive task often requiring uneconomical dense sampling of seismic wavefields. To mitigate the challenges of dense sampling, several advances in seismic acquisition have been made in recent years including the use of multiple sources firing at near simultaneous random times, and the adaptation of Compressive Sensing (CS) principles to design practical acquisition engines that improve sampling efficiency for seismic data acquisition. However, little is known regarding the implications of these developments for time-lapse studies. By conducting multiple experiments modelling surveys adhering to the principles of CS for 4D seismic, I propose a model that demonstrates the feasibility of randomized acquisitions for time-lapse seismic. The proposed joint recovery model (JRM), which derives from distributed CS, exploits the common information in time-lapse data during recovery of dense wavefields from measured subsampled data, providing highly repeatable and high-fidelity vintages. I show that we obtain better vintages when randomized surveys are not replicated, in contrast to standard practice, paving the way for an opportunity to relax the rigorous requirement to replicate surveys precisely. We assert that the vintages obtained using our proposed model are of sufficient quality to serve as inputs to processes that extract time-lapse attributes from which subsurface changes are deduced. Additionally, I show that recovery with the JRM is robust with respect to errors due to differences between actual and recorded postplot information. Finally, I present an opportunity to adapt our model to problems related to time-lapse seismic imaging where the main finding is that we can better delineate time-lapse changes by adapting the joint recovery model to wave-equation based inversion methods.

Improved Time-lapse Data Repeatability with Randomized Sampling and Distributed Compressive Sensing

Summary Recently, new ideas on randomized sampling for time-lapse seismic acquisition have been proposed to address some of the challenges of replicating time-lapse surveys. These ideas, which stem from distributed compressed sensing (DCS) led to the birth of a joint recovery model (JRM) for processing time-lapse data (noise-free) acquired from non-replicated acquisition geometries. However, when the earth does not change—i.e. no time-lapse—the recovered vintages from two non-replicated surveys should show high repeatability measured in terms of normalized RMS, which is a standard metric for quantifying time-lapse data repeatability. Under this assumption of no time-lapse change, we demonstrate improved repeatability (with JRM) of the recovered data from non-replicated random samplings, first with noisy data and secondly in situations where there are calibration errors i.e., where the acquisition parameters such as source/receiver coordinates are not precise.