Bridging the Gap in Gold Exploration: A Deep Learning Framework for Prospectivity Mapping

We are proud to announce the publication of our latest research, "Formulating gold prospectivity mapping as a constrained learning problem," featured in the proceedings of the Fifth International Meeting for Applied Geoscience & Energy. In this study, experts from Computational Geosciences Inc. (CGI)—including Frederick A. Jackson, Paulina Wozniakowska, and Eldad Haber—tackle the challenges of mineral targeting in sparse, complex environments. By leveraging Convolutional Neural Networks (CNNs), the team has created a framework that streamlines the identification of high-potential gold deposits.

The research compares traditional regression models against a novel distance-based formulation, using extensive datasets from the Yilgarn Craton in Western Australia. Our findings reveal that by incorporating airborne geophysics and remote sensing data into this new model, we can achieve much smoother predictions and superior generalization. This is particularly effective in under-sampled areas where data is limited, providing a more coherent spatial picture for exploration teams.

For our clients in the mining sector, this means more reliable targets and reduced exploration risk. At CGI, we continue to push the boundaries of geophysics and machine learning to deliver actionable insights. This paper underscores the importance of domain-informed constraints in training AI, ensuring that CGI remains at the forefront of digital transformation in mineral exploration.