Multivariate Simulation in Non-Stationary Domains: A Framework for Accurate Data Reproduction

Le résumé fourni par la source

Accurate multivariate Gaussian simulation is critical for resource assessment and mine planning, especially in polymetallic deposits where strong trends, data bias, and multivariate outliers introduce complexity. In this scenario, standard workflows applied to non-stationary domains may result in undesirable data statistics reproduction, especially the multivariate relationships between variables. This study proposes an enhanced simulation framework that integrates data standardization, multivariate outlier detection, trend modeling and removal, and a dual application of the Projection Pursuit Multivariate Transform (PPMT). The approach is demonstrated within a high-grade mineralized breccia domain of the Peñasquito deposit, utilizing data from diamond core and reverse circulation (RC) drill holes, including Au, Ag, Pb, and Zn. Bias in RC data was corrected using data standardization, and multivariate outliers were identified through the application of a robust Mahalanobis distance. Trend modeling was performed using a moving window average and was removed using the Gaussian Mixture Model and Stepwise Conditional Transform. PPMT was applied both before and after trend modeling in order to improve decorrelation and simulation performance. Results show improved data reproduction through histograms, variograms, and complex relationships, as well as correlation coefficients. Cross-validation confirms reduced bias and improved accuracy. This research highlights the importance of treating multivariate outliers and applying PPMT both before and after trend modeling. The study demonstrates that applying PPMT twice is more effective for managing persistent non-stationary features, especially in high-grade domains.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.