Author(s): La Rosa, R ; Steffen, M ; Storch, I ; Knobloch, A ; Cardoso Fernandes, J ; Carvalho, M ; Barrios, MS ; Sánchez Migallón, JM ; Nygren, P ; Williams, V ; Teodoro, AC
Date: 2025
Persistent ID: https://hdl.handle.net/10216/166219
Origin: Repositório Aberto da Universidade do Porto
To meet the European Unions growing demand for critical raw materials in the transition to green energy, this study presents a novel, cost-effective, and non-invasive methodology for mineral prospectivity mapping. By integrating hyperspectral data from satellite, airborne, and ground-based sources with deep learning techniques, we enhance mineral exploration efficiency. We employ Bayesian Neural Networks (BNNs) to predict mineral prospective areas while providing uncertainty estimates, improving decision-making. To address the challenge of obtaining reliable negative labels for supervised learning, Self-Organizing Maps (SOMs) are used for unsupervised clustering, identifying barren areas through co-registration with known mineral occurrences. We illustrate this approach in the Aramo Unit in Spain, a geologically complex region with Cu-Co-Ni mineralized veins. Our workflow integrates local geology, mineralogy, geochemistry, and structural data with hyperspectral data from PRISMA, airborne Specim AisaFenix, LiDAR and ground-based spectroradiometry. By leveraging learning techniques and high-resolution remote sensing, we accelerate exploration, reduce costs, and minimize environmental impact. This methodology supports the EUs S34I project by delivering high-value, unbiased datasets and promoting sustainable, cutting-edge mineral exploration technologies. (c) 2025 by SCITEPRESS - Science and Technology Publications, Lda.
