Investigation of uncertainties in data imputation through application of sequential co-simulation

Abstract

The main principle of mineral resource estimation process is to quantify a mineral grade in target places. However, in many cases, available dataset show heterotopic sampling pattern that makes the current practice of mineral resource estimation challenging. In order to cope this difficulty, one alternative consists of removing the sample locations that only one variable is available. Despite an attractive simplicity of this method, one may mention that such excluding of valuable data may lead to biased results for resource modeling. Another solution for this impediment is using imputation algorithms, for which the data at sampled location is imputed by stochastic techniques. This is beneficial not only because one can keep the co-variate data, but also the uncertainty can be quantified. In order to show the capability of the imputation method, in this study, a homotopic dataset from a limestone deposit located in south of Kazakhstan is selected as the actual case study accompanying with another homotopic dataset from Chile to evaluate the proposed approach of imputation. Then, to investigate uncertainties in data imputation process, a technique called Sequential Gaussian Co-simulation was utilized. For this purpose, homotopic dataset was transformed to heterotopic dataset. This allows performing validation of the proposed technique. Lastly, sensitivity analysis was carried out to observe the effect of amount of missing data. Overall results of imputation techniques showed satisfactory results, and it can be concluded that derived simulated grade values for drillholes can be incorporated to the ore body modeling.