Application of multi-point statistics to irregular core sampling data in modeling of copper-porphyry deposits

The article discusses a procedure for adapting multi-point statistics to interval-based core sampling data obtained from a copper-porphyry system. The procedure uses irregular borehole sampling and generates local category-based patterns within an anisotropic ellipsoidal neighborhood of a point. The calculation includes the geological domain, lithological rock type, composite type, and aggregated class of the ore potential. Single-layer and multi-layer patterns, distance-weighted and balanced variants of multi-point statistics, indicative kriging, and hybrid schemes were compared for verification. The quality was assessed using a post-hoc validation against the borehole data, employing the proportion of the correct answers, the balanced proportion of the correct answers and the macro-averaged F1 score. For the geological domains, the best result was achieved by a hybrid variant combining multi-point statistics and indicative kriging: the macro-F1 score was 0.401, and the proportion of the correct answers was 0.495. For the lithological types, the best macro-F1 score was obtained by jointly considering the domain, lithological, and composite contexts. For the composite type and the aggregated ore potential class, the best macro-F1 scores were achieved using a distance-weighted approach with class balancing. The results show that the multilayer and distance-weighted templates improve the accuracy of category-based modeling and can be used to refine a geological model.