Tetris: a training image generator for SGeMS

The recent developments of training image-based geostatistics have allowed the creation of stochastic models with greater realism than before. These algorithms aim at reproducing spatial patterns, such as connectivity, that are depicted in a training image. A training image contains the possible spatial configurations for any given geological object and relationships between objects. However, algorithms for generating training images are still incomplete and unsatisfactory. The proposed design overcomes a major hurdle of rigidity of pre-defined geological objects in the current training image generators. This paper presents a plugin to the SGeMS software that allows modelers to generate geological objects with complex geometries and the relevant interactions between these objects. First, a geological object is built either from pre-coded shapes, such as ellipsoid, cuboid, kernels, or user defined. These basic shapes are then assembled with geometrical operations (difference, union and intersection) to create new complex shapes. For added flexibility, any shape can be translated, rotated and sheared. A second set of parameters control the interactions between the geological objects. Each parameter used in the training image construction (e.g. size, the rotation angles, number of stacks) can vary in space. This locally varying parametrization allows the representation of trends in geological body geometry, interactions, and locations. ∗The auhors would like to thanks Marco Pontiggia, Sergio Nardon and Giuseppe Serafini at Eni for their support and valuable suggestions.