Multi-Point Statistics for Tailings Deposits

Technical and economic evolution of the mineral industry resulted in a new view of mining tailings. Formerly tailings are considered not valuable but now they become new resources that have promising economic values. The spatial estimation of mineral distribution is essential for optimally exploiting tailings, but this faces several issues such as non-stationarities, complex and artificial structures, and limited historic information on the feed streams and spilling points. Multi-Point Statistics methods are capable of reproducing complicated structures more appropriately as compared to two-point statistics methods. This paper proposes a new framework for performing Multi-Point Statistics on tailings deposits. Instead of using one big training image, we used several training images. In this way we can use different joint distributions at different locations to cope with the nonstationarity of tailings deposits. By providing and eventually weighting training images generated with different forward modelling parameters we can handle the uncertainty about the history of the deposit, while still exploiting available historic information. The framework is illustrated through a test on a synthetic tailings model. The synthetic truth and the training images are generated using Delft3D-Flow, an open source process-based modelling program that can also perform stratigraphic forward modeling in deltaic depositional environments. The MPS analysis is based on a new implementation with advanced capabilities.