Hybrid geostatistics: Object-based simulations using MPS-generated meandering channels

Abstract

Object-based simulation methods are some of the most widely used approaches for modeling geological reservoirs and quantifying geological uncertainty. However, one major drawback of the object-based approach is its lack of flexibility. The objects are often defined with basic analytical shapes, and as such they may not reflect the complexity of the actual geobodies present in the subsurface. This results in models that lack realism, as well as difficulties in conditioning especially in the presence of dense datasets. To tackle these issues, we propose to stochastically generate objects using an analog-based approach. Instead of defining objects by a limited number of parameters, objects are produced by using a multiplepoint geostatistics (MPS) algorithm, which is a non-parametric process. Each new object is then drawn from an ensemble of possible members, all presenting similar statistical properties than a given reference object, namely a training image.