Monitoring reservoir production and injection by time-lapse seismic surveys has become an essential method in industry applications. By solving the wave equation and fitting the baseline or time-lapse seismic data with iterative optimization algorithms, full waveform inversion (FWI) can deliver velocity models with a resolution finer than one wavelength (=λ/4). However, non-repeatability (NR) can contaminate the data preventing reliable estimates of time-lapse (4D) changes. We modify the SEAM time-lapse pilot models to mimic sea water velocity changes in warm versus cold currents, and include realistic NR measurements inspired by a 4D OBN survey in the GOM. We show the effects of source-receiver position errors and water velocity changes on the 4D OBN seismic data, and test the robustness of different 4D FWI algorithms against these effects. We find that ignoring these NR effects in the inversion can cause strong artifacts in the estimate of velocity changes, and thus should be addressed before or during inversion.
We show that forward bootstrap FWI is prone to artifacts preventing reliable 4D estimates, although it shows higher resolution than parallel FWI. Interestingly, the reverse bootstrap implementation (i.e. monitor inversion before baseline inversion) can regenerate these forward bootstrap artifacts with a phase reversal. Thus, we propose a two-step central-difference FWI method in which the forward and reverse bootstrap FWI results are averaged to attenuate inversion artifacts, while the high resolution is retained in the 4D estimate of velocity changes.
Additionally, FWI mainly relies on far-offset diving waves to estimate large-scale (low-wavenumber) variations of velocity structures, and thus may not penetrate deeper reservoirs. Alternatively, joint full waveform inversion (JFWI) assumes a scale separation in the model space such that low wavenumbers can be recovered from both diving and reflected waves, in addition to high-wavenumber reflection impedance perturbations. The SEAM 4D model example shows that classical 4D FWI methods, including central-difference FWI, cannot properly estimate the 4D changes in a deep reservoir layer that is not penetrated by diving waves due to limited source-receiver offset apertures, whereas a workflow of 4D JFWI followed by 4D FWI leads to more accurate 4D estimates and higher resolutions, as well as an improved data fit.
Speaker Bio
Dr. Wei Zhou
Dr. Wei Zhou is a researcher at University Grenoble Alpes within the framework of Seiscope consortium. Before that, he was a research associate in the Geophysics Department at the University of Texas at Dallas. His research interests include full waveform modeling and inversion, high-resolution time-lapse imaging, and high performance computing. Wei did his thesis on reflection full waveform inversion and received his PhD degree at University Grenoble Alpes in 2016. Prior to that, his did his Master study on one-way wave equation migration in the Institute of Geophysics Geology, Chinese Academy of Sciences.