Stochastic simulation methods for characterization of lithoclasses in carbonate reservoirs

The construction of a reservoir model begins with the geology, specifically a geological conceptual model, sub-division into layers, top and bottom limits/surfaces, and characterization of rock types or facies within the layers. To model the geology in heterogeneous case studies, of which carbonate reservoirs are a good example, conceptual and deterministic models are unable to adequately represent the internal geometry and, for this reason, stochastic models are most, particularly used mainly for the spatial characterization of rock types.This paper presents a comparison of bi-point geostatistical simulation methods for characterization of lithoclasses, making use of a carbonated reservoir as a case study. To cope with the geological complexity, and to assist in understanding the internal distribution of properties, the lithoclasses were derived from cores based on their lithology, petrophysical properties and capillary pressure, and subsequently extended to log data.Five simulation methodologies, able to deal with categorical variables, are compared in a multi-phase framework: (1) truncated gaussian simulation, posterior conditioning and classification of simulated probabilities based on local and global proportions (TGSPC); (2) truncated gaussian simulation, posterior conditioning and classification using simulated annealing (TGSPC + SA); (3) sequential indicator simulation with correction for local probabilities (SIS); (4) simulated annealing post-processing of sequential indicator simulation images (SIS + SA); and (5) probability field simulation (PFS). Matched to experimental data, theoretical multi-phase variogram models, and proportions are observed, as well as the presence of artefacts. To evaluate the range of uncertainty of the simulated images, 30 realizations are generated, and entropy and average geobody volumes (connected blocks belonging to the same lithoclass) are computed and compared.