Local Ranking of Geological Conceptual Models in Non-Stationary Settings using Multi-Point Geostatistics

In geomodeling, it is commonly accepted that the distribution of physical properties is controlled by the architecture of geological objects. However, insufficient data and the complexity of earth processes create an ill-posed problem where many architectures are plausible. Consequently, several geologists will produce different geological models for the same location. This contribution proposes a way to objectivize the ranking of those conceptual models by comparing them with hard data, both globally for the whole study region and locally for certain of its sectors. The idea is to extend the multi-point geostatistics direct sampling algorithm to be able to extract data events from different training images, representing several competing geological models, and to record the training image origin of values pasted on simulation grid cells. By tracking the frequency with which every training image is visited, we can rank the likelihood of each geological model. Histograms of the frequency of usage of each training image will provide a global ranking of the several conceptual models, while maps of these frequencies can be used to produce the local rankings. We demonstrate this method in two synthetic fluvial depositional environments where three distinct geological concepts are being proposed, with different abundances of hard data. Results indicate that the proposed method could be a useful tool in defining which geological concept dominates at a particular region and which is the frequency ranking for each training image on that region.