Post-processing of sequential indicator simulation realizations for modeling geologic bodies

Abstract

Sequential indicator simulation realizations contain unavoidable artifacts that are geologically unrealistic. This happens because unlikely types can be drawn randomly from the cumulative distribution and be assigned to a cell in the simulated model. This cell may then be used as previously simulated data when a cell in its neighborhood is visited during a random walk. The sequential process sometimes results in geologically unrealistic realizations. However, different realizations can reveal hidden features. Each realization contains both reliable geological information and noise that is displayed as unlikely types. This paper proposes applying the averaging filter that is commonly used in seismic reflection data to improve the signal to noise ratio. After applying this filter, all L realizations will be condensed into a single geological model that contains certain and uncertain cells. This average model is then exhaustively sampled for the certain cells, and this new sample is used to post-process the uncertain cells to reduce the uncertainty. This resampling and post-processing procedure can be repeated until the final model is considered to be good enough. The proposed method is tested with a model of a dike that crosscuts two sedimentary units. The synthetic geologic model was sampled with 24 drill holes. The resultsshow that the final geological model with reduced uncertainty reproduces very well the sedimentary units and the orientation of the dike as well. The dike shape is not fully reproduced and still presents uncertainties because of lack of neighbor data.