Article title

By Hugo Melo

MultiGaussian Kriging For Grade Domaining – An Uncertainty Based Approach

Authors

Author 1

Author 2

Author 3

Author 4

Grade domains are often considered, following the modelling of geologic domains, to further control the distribution of grades during resource estimation. This is usually achieved by wireframe modelling on sections displaying grade assays or composites, indicator kriging, and/or implicit modelling using radial basis functions. The latter approach is often facilitated by using commercial software such as Leapfrog, making it a fast, semi-automatic alternative to the more traditional manual approach to wireframing. Unfortunately, modellers must enter parameters that indicate spatial correlations, for which they either do not have the time and/or the expertise to suitably calculate or model the required variograms.

As an alternative to an implicit modelling approach, we propose the use of MultiGaussian (MG) kriging to establish grade domains. The method consists of estimating grades using MG kriging; however, instead of back transforming to obtain a grade estimate at each location, we determine the probability to exceed certain grade thresholds and categorise grade domains accordingly. This block model approach can then be imported into any commercial package to generate iso-probability contours for any grade thresholds of interest. This permits the uncertainty assessment of grade domains by post-processing for various grade thresholds and iso-probability shells, and permits the user to assess confidence in the grade shell for that threshold.

This method was applied to the Rainy River gold deposit in Northern Ontario, with visual comparisons to grade shells used for the previous resource model. The resource shells were manually generated with guidance from Leapfrog shells. While results showed the MG kriging approach to grade domains can yield shells comparable to existing implicit modelling methods, they also showed better conformity to the manually digitised wireframes accepted for resource modelling.

The MG approach provides a balance between the more time intensive manual approach to grade domaining and the fast, semi automatic modelling using implicit methods. It also provides greater fidelity to the data by requiring that spatial correlations are calculated and modelled. Finally, this alternative MG kriging approach is attractive because the method and model parameters are data-driven, tractable and repeatable, with the added benefit of quantifying a measure of confidence associated to the resulting grade shells.

 

A comparison of iso-probability shells for 0.90 gpt gold