Bayesian inversion of 4D seismic waveform data for quantitative integration with production data

With the development of new innovative seismic methods, it should be possible to further increase oil recovery from existing fields. This makes sense from an environmental and an economic perspective. Seismic methods generally make it possible to image sedimentary basins and reservoirs (related to petroleum production, CO2 storage or geothermal energy) based on essentially the same principles as in medical imaging. The term 4D seismic refers to repeated seismic collections over time. Using 4D seismic, you can "hear" the dynamic behavior of the reservoir and get help with the reservoir control. The greatest value of 4D seismic is the additional information you get to update a model of the reservoir, as well as help in locating undrained oil and flow barriers, which is important for well planning. In the geophysical community, full seismic waveform inversion (FWI) has emerged as the ultimate and ultimate solution to the Earth's dissolution and imaging problem. FWI can potentially contain a complete numerical simulation of the real wave field, where observed waveform data is reproduced using an inverted model of the subsurface. But FWI faces many challenges, including the sensitivity of the results to the choice of starting model and its enormous calculation cost. In this project, we have addressed these challenges and developed scattering theoretical methods for Bayesian FWI that provide information about the uncertainty as well as the most likely values of the changes in (directional) elastic parameters that determine 4D seismic response to a petroleum reservoir in production. Also, we have developed different strategies for inversion of 4D seismic waveform data and performed research on effects of elasticity and anisotropy. All this is useful for a quantitative integration of 4D seismic waveform and production data, which can provide better reservoir management. During the project period, we have published more than 15 peer-reviewed articles in international journals. In addition, we have published a similar number of conference papers and given many oral presentations at international meetings. We have also interacted with users from industry to obtain feedback to the development of our seismic methods. A PhD candidate was given the opportunity to work on this project in collaboration with the project manager from the University of Bergen, two reservoir mathematicians from NORCE and leading international partners at the University of California and the Czech Academy of Sciences. PhD candidate Xingguo finished his doctorate six months ahead of schedule and received the award for best Chinese international student. As a result of this project, Xingguo now works as a full professor of geophysics at a leading Chinese university. The project has also contributed to the education of several master's degree candidates who either work with seismic or related topics such as acoustics.

This project, which represents a collaboration between the University of Bergen and NORCE, focused on the development of methodology for quantitative integration of 4D seismic waveform and production data; which is important for optimal reservoir management. The project was associated with the National IOR Centre of Norway and helped strengthening the geophysical activities within the IOR Centre as well as at NORCE. This project also helped to consolidate our international contacts at the University of California, the Czech Academy of Sciences and (recently) TUDelft, which can be very useful for future projects. The project also had a positive impact on the training of research students within exploration geophysics at the University of Bergen. Some of the results from this project will be useful for a new VISTA Centre for modeling of coupled subsurface processes as well as a new centre for research based innovation called SFI DigiWell; involving the project leader.

The principal aim of this project is to develop methods for Bayesian inversion of 4D seismic waveform data that provides information about the uncertainty as well as the most likely values of the inverted seismic or elastic parameters. The main idea behind the project is that we can reduce the computational cost of a Bayesian FWI in time-lapse mode by using a target-oriented scattering-integral approach which is based on an explicit representation of the data sensitivity matrix and compatible with the use of general domain decomposition and renormalization methods for strongly scattering media. We have already significant experience from the development of the scattering-integral approach within the framework of Tikhonov regularization, but more work related to the Bayesian (statistical) approach is needed. The methods we propose to develop will be relevant for both seismic exploration and production monitoring. However, our numerical examples will be biased towards applications within dynamic reservoir characterization. Therefore, we shall also investigate different strategies for dealing with time-lapse seismic waveform data and rock physics models within this scattering-based framework for FWI and seismic history matching.