Babak Ghane, PhD

Gold Prospectivity Mapping in Southwestern New Brunswick Using Airborne Geophysical Data and XGBoost Machine Learning Method

B. Ghane1, D. Lentz1, K. Thorne2 
1Department of Earth Sciences, University of New Brunswick, Fredericton, New Brunswick, Canada
2New Brunswick Department of Natural Resources and Energy Development, Fredericton, New Brunswick, Canada 

Airborne geophysical surveys, such as magnetic and radiometric methods, play a pivotal role in mineral exploration by providing systematic spatial coverage and high-resolution insights into subsurface geological features. These techniques are particularly valuable in regions with vegetation or sedimentary cover. Airborne geophysical datasets are extensively employed in mineral prospectivity mapping (MPM) to delineate areas with high mineralization potential. Recent advancements in machine learning have further enhanced the effectiveness of MPM, offering improved prediction accuracy and interpretability to aid in prioritizing exploration targets. 

This study focuses on gold mineralization within a 1,500 km area in southwestern New Brunswick, a geologically complex yet underexplored region. Southwestern New Brunswick has a long and complex tectonic history that has given rise to a great variety of mineralized systems. Precambrian and Silurian stratiform base-metal sulfide deposits were intensely deformed, and in part remobilized, from Early to late Devonian (Acadian orogeny) and/or Early Carboniferous deformation events. Numerous deposits formed as a result of late- to post-Acadian tectonic activity and include both epigenetic and stratiform deposits. 

A suite of complementary geophysical predictor layers was developed and integrated into a predictive model for gold mineralization in the study area. In this study, the Centre for Exploration Targeting (CET) grid analysis is used to delineate the structural complexities by extracting and digitizing magnetic ridges or troughs in the region. The Contact Occurrence Density (COD) and Orientation Entropy (OE) maps are produced using CET analysis. Additional edge detection techniques, such as the First Vertical Derivative (FVD), Tilt Angle Derivative (TDR), and Total Horizontal Derivative (THDR), are valuable for emphasizing the role of structural features in the precipitation of gold mineralization. Furthermore, identifying the source of magma that feeds mineralization is a critical aspect of exploration targeting. Tools like the Analytical Signal (AS) method, Upward Continuation filters, and reduction to pole (RTP) map, are used for this purpose. Also, radiometric layers including radiometric ratio maps of eTh/K, eU/K, and eU/eTh are produced to identify hydrothermally altered zones related to mineralization. In total, 11 geophysical layers were created for delineating the high potential zones for gold mineralization. These geophysical layers were integrated into a predictive framework using the XGBoost machine learning method, a powerful tool for modeling mineral prospectivity. The model was trained and validated using 64 known gold occurrences, ensuring its reliability. Validation metrics demonstrated the robustness of the framework and its ability to accurately predict high-potential gold mineralization zones.