Estimation of Pressure-Saturation and Porosity Fields from Seismic AVA Data Using an Ensemble Based Method

Abstract

We propose an ensemble based seismic inversion framework to estimate static and dynamic reservoir parameters such as saturation, pressure and porosity fields using seismic data. The proposed method has certain novelties, in terms of the choice of seismic data, as well as uncertainty quantification of the estimates. Further, the method uses reservoir-engineering data as the prior information (such as pressure-saturation data) from reservoir simulation model to constrain the inversion process. Many conventional seismic inversion algorithms are deterministic in nature, and thus they pay less attention to uncertainty quantification. To quantify the uncertainties in the estimates, we adopt an iterative ensemble smoother as the inversion algorithm. Compared to the conventional deterministic inversion algorithms, this ensemble-based method is a derivative-free and non-intrusive approach, and has better capacity of uncertainty quantification. On the other hand, inverted seismic parameters, such as acoustic impedance, are often adopted as the data in inversion. In doing so, extra uncertainties may arise during the inversion processes. Here, we avoid such intermediate inversion processes by adopting amplitude versus angle (AVA) data. To handle the big-data problem in the AVA inversion process, we adopt a wavelet based sparse representation procedure (Luo et al., 2016). Precisely, we apply a discrete wavelet transform to the AVA data, and estimate noise in the resulting wavelet coefficients. We then use the leading wavelet coefficients above a certain threshold value as the data in inversion. We apply the proposed framework to a 2D synthetic model for a proof-of-concept study. This reservoir model consists of three phases (water, oil and gas), and is a vertical section of a 3D Norne field model. We also test the performance of the framework in the 3D Brugge benchmark case that consists of two phases (water and oil). The numerical results from both cases indicate that the proposed framework can integrate the reservoir-engineering data as prior knowledge with seismic data, while achieving reasonably good estimates of both static and dynamic reservoir variables.