Problems with a localized nature in exploration seismology

Abstract

A wide range of problems in seismology have a localized nature. Examples include inversion for time-lapse change, salt boundary position and Amplitude or Phase Versus Angle (AVA/PVA). With increasing computational power, many inversion procedures have transitioned away from forward models with simplified physics and now simulate synthetic wavefields on a model representation of the Earth. These simulated wavefields form the basis for the model updates in the inversion. Usually these wavefields are simulated on the entire computational model, even though the region of interest and the associated updates of the computational model are localized. In this thesis I explore ways of making use of the localized nature of the inversion by limiting the wavefield generation to the region of interest. This shrinking of the computational domain leads to faster solutions, reducing the time between the start and finish of the local inversion. This shorter turn-around time facilitates decision making and improves the interactivity of the inversion procedure. I first introduce a frequency domain local solver which can generate exactly the same wavefields as a full domain solver. I then apply this local solver to both a time-lapse problem and to salt boundary inversion. I then take a step back and look at time-lapse inversion outside the scope of local solvers. By looking at differences in the Full Waveform Inversion (FWI) gradient obtained from baseline and monitor datasets I compute a confidence map of locations likely to contain time-lapse change. This confidence map is then used to regularize a joint inversion for both the baseline and monitor datasets. I compare this methodology to other existing time-lapse algorithms on a synthetic North Sea model. In this study I notice the importance of post-critical reflections and become aware of their potential for reservoir characterization. Motivated by this observation I introduce a local solver for simulating elastic reflections in a region of interest. These elastic reflections can then be used for AVA and PVA to for instance invert for reservoir characteristics. The wavefield is efficiently propagated through the overburden using precomputed acoustic Green's functions obtained from a cheap acoustic solver.