Tight Gas Reservoir Properties Inference from AVO and Rock Physics Analysis

Loures, Luiz¹
 Rostirolla, Sidnei²
 Pereira, Edinei¹
 Zagotto, Eric²

We present a quantitative inference study on a tight gas sand reservoir, focused on porosity, facies and fluid content properties from pre-stack seismic data. The methodology follows a Bayesian approach used in combination with AVO and rock physiscs. The seismic data is inverted to yield P and S impedances and others seismic attributes, which are related to reservoir properties by rock physics relations. The Bayesian formulation captures the uncertainties from seismic data to reservoir properties. However, to quantitatively estimate all parameters of interest from seismic attributes is an ill posed inversion problem. Thus we break the solution into parts: facies, followed by fluid content discrimination.

Initially, following a Bayesian approach, a neural network is applied to define facies; for training and for facies classification on a seismic volume. At the next step, porosity is inferred using rock physics models, which are independent of fluid saturation. Because seismic impedances are highly sensitive to porosity, we are able to make quantitative predictions on porosity. As a result we obtain a probability indicator for the reservoir rock on a target interval.

Fluid discrimination from seismic data presents a more difficult problem, due to the low sensibility of the seismic response to fluid saturation and ambiguity with respect to variations such as pressure, net-to-gross ratio and porosity. We rely on our improved porosity and facies models to be able to determine fluid content. We apply this method to characterize a tight gas sand reservoir, offshore in