Predictive modelling of gold mineral system in northern New Brunswick: Insights from the analysis of multiple geoscience data
Loading...
Date
2025-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of New Brunswick
Abstract
The study area lies within the Tobique–Chaleur Zone of northern New Brunswick, specifically in the Chaleur Bay Synclinorium. The synclinorium contains Upper Silurian to Early Devonian volcanic and sedimentary rocks that host several epithermal gold occurrences spatially linked to the trans-crustal Rocky Brook–Millstream Fault system. This research advances the understanding and prediction of epithermal gold mineralization through a multidisciplinary and multi-scale approach that integrates geoscientific modeling, machine learning, and geochemical analysis. First, the study addresses the need for geochemical validation by employing portable X-ray fluorescence (pXRF) and micro-XRF (µXRF) spectrometry to characterize pathfinder elements, indicator minerals, and alteration assemblages across four key gold occurrences. Multivariate compositional data analysis (CoDA), including biplots and clustering, reveals geochemical signatures associated with gold-bearing quartz veins. These veins are dominantly hosted in felsic to intermediate volcanic and intrusive rocks and are characterized by sericitic (illitic), potassic, and silica alteration, with gold commonly associated with sulfide phases, such as pyrite, arsenopyrite, and chalcopyrite. The second study applies a mineral systems framework to translate ore-forming processes into mappable criteria using both knowledge-driven methods (fuzzy logic and geometric average) and a data-driven approach (logistic regression). The integration of these models, reclassified through a concentration–area fractal method, allows segmentation of the region into distinct favourability zones, with the fuzzy logic model yielding the highest predictive success. The consistent alignment of high-potential zones with known gold occurrences validates the hybrid modeling approach.
Complementing this, the third study develops mineral prospectivity maps (MPM) using supervised machine learning algorithms Random Forest (RF), Support Vector Machine (SVM), and XGBoost, based on 24 evidence layers derived from geological, geophysical, and geochemical datasets. These models, evaluated through Receiver Operating Characteristic (ROC) and prediction–area (P–A) analyses, demonstrate high predictive accuracy, with ensemble modeling and confidence indexing further enhancing robustness and interpretability. Together, these integrated studies establish a scientifically rigorous and spatially coherent framework for predictive targeting and mineral exploration in covered and structurally complex terrains. The combination of machine learning, geological knowledge, and geochemical validation provides a replicable model for gold exploration in analogous regions worldwide.
Description
Keywords
gold mineralization, pathfinder elements, mineral indicators, pXRF, μXRF, CODA, Mineral prospectivity mapping, northern New Brunswick, Support Vector Machine, XGboost, epithermal gold mineralization