Compositional Multi-Point Geostatistics for Tailings Deposits - A Synthetic Case Study

Currently tailings deposits have become new resources that are challenging and valuable to exploit. To properly exploit them, we require a 3D spatial characterization of their mineral content. In a natural deposit, this is achieved by sampling at several locations and applying geostatistics to estimate block values. Certain characteristics of tailings deposits make them not amenable to conventional geostatistics. In particular, it is important to consider both valuable and gangue minerals, thus we need to take the compositional nature of our variables into account. In addition, the interplay of erosional and depositional processes creates structures with certain continuity patterns that cannot be modelled by conventional variogram-based methods such as kriging.
Therefore, we use a Multi-Point Geostatistics method, Direct Sampling (DS). DS is based on selecting the event from a training image, the conceptual spatial arrangement of a variable, with the shortest distance to the data event from the simulation domain. To account for the compositional nature of our variables, the Aitchison distance is calculated. We use numerical stratigraphic modelling to obtain a variety of training images, which we feed into a modified DS to deal with multiple training images at once.
We tested the proposal on a multi-source synthetic tailings deposit produced by numerical stratigraphic processes. Each grid of the model contains information about the content of several sediment species summing to 100%. Hard data are sampled on the model at certain locations and along with several unique training images we recreate the full 3D spatial distribution of the properties.