Permeability prediction with geostatistical seismic inversion constrained by rock physics

Abstract

Summary Accurate predictions of the spatial distribution of permeability in the subsurface is fundamental in reservoir characterization for several tasks (e.g., CO2 injection and storage monitoring or natural resources characterization). Nonetheless, modelling permeability is particularly challenging, due to its strong variability, spatial anisotropy and dependency on several factors. The most common approaches for its modelling involve deterministic estimations from rocks’ porosity, using pre-calibrated rock physics models, or data-driven stochastic sequential simulation co-simulation to reproduce porosity-permeability experimental joint distributions. Both methods can be hence included in a geostatistical seismic inversion framework. However, rock physics models can strongly approximate the real variability of permeability, while purely data-driven simulation methods lack any physical constraints such as the rock physics information. To tackle such disadvantages, we propose an iterative geostatistical seismic inversion algorithm for facies and permeability predictions, where facies models are generated through Markov Chain Monte Carlo sampling, while stochastic sequential simulation and co-simulation generate permeability, porosity, and acoustic impedance from their experimental joint distributions. Rock physics modelling is used to derive a misfit between simulations of permeability and estimations form acoustic impedance. The method is illustrated herein through its application to a synthetic single-trace scenario and compared to pure data-driven and estimation approaches.