Probabilistic seismic inversion based on rock-physics models for reservoir characterization

This dissertation addresses recurrent questions in hydrocarbon reservoir characterization. What properties of rock stand behind recorded seismic reflections? How and to what extent can these rock properties be quantified from these reflections? To answer these questions, I link reservoir thickness, lithology, porosity, and saturation to seismic data by coupling deterministic rock-physics relationships and Bayesian statistics. The key innovation is a seismic-inversion method that functions on the principles of deterministic physics and probabilistic variations of rock properties for a potential reservoir. This is intended for use in practical reservoir characterization. Within it, I engage into the fundamental issue of reconciling spatial scale differences between seismic imaging, geology and stratigraphy, and rock physics.; Seismic reservoir characterization is a non-unique problem. This ambiguity necessitates using statistical methods to constrain the problem and quantify any associated uncertainty. This dissertation contributes to both deterministic and statistical seismic-based reservoir characterization. Complementary, I examined empirical velocity-porosity transforms to determine their consistencies with fluid-substitution equations. Results indicated that some transforms require a fluid-substitution step to provide accurate porosity values when brine is not the in situ fluid. The deterministic reservoir-characterization work combined multiple elastic properties to simultaneously predict lithology and porosity for constant fluid saturation. By establishing a relationship between lithology and porosity, the problem reduced to allow estimations of lithology and porosity from a combination of P-impedance and Poisson's ratio.; The probabilistic technique simultaneously predicts four reservoir properties and associated uncertainty corresponding to a reflection from a potential reservoir unit. This involves exhaustive forward modeling of the prior model and a hill-grid search in the inverse problem. Seismic and well data from offshore Africa provided the basis to develop and perform the inversion. Results included predictions of combinations of reservoir thickness, porosity, and saturation that matched well data within the associated uncertainty. Lithology predictions for the data set, however, were too uncertain to provide a fill solution. An additional application to North Sea data demonstrated that if seismic resolution was fine enough, then the inversion provided independent predictions of thickness and porosity. Given these results, the inversion decomposes a seismic reflection into combinations of probabilistic reservoir properties.