From Prospectivity Maps to Responsible Drill Decisions: Explainable, Uncertainty-Aware Multimodal AI for Sustainable Mineral Exploration

کد مقاله : 1015-IMEC15 (R2)

نویسندگان

رضا معزی نسب * دانشگاه صنعتی شاهرود

چکیده مقاله

Mineral exploration is increasingly data-rich but decision-poor: modern campaigns can generate large volumes of geological, geochemical, geophysical, remote-sensing, and drilling information, yet the decisive operational question remains which target should be tested next and why. Recent artificial-intelligence (AI) research has substantially improved mineral prospectivity mapping (MPM), but high predictive accuracy alone is insufficient for high-cost, environmentally consequential drilling decisions. This critical review synthesizes developments through 15 August 2026 at the intersection of multimodal geoscience data fusion, geologically informed machine and deep learning, explainable AI (XAI), uncertainty quantification, and sequential value-of-information (VOI) decision making. Particular attention is given to sparse positive labels, class imbalance, workflow-induced uncertainty, spatial transferability, semi-supervised learning, graph-based and foundation-model approaches, and the need to distinguish model confidence from geological plausibility. We argue that the next practical step is to move from static prospectivity maps toward adaptive, auditable decision systems that rank actions rather than pixels. To operationalize this transition, we propose the TRUST-MPM framework, in which mineral targeting is Transparent, Risk-aware, Uncertainty-calibrated, Sustainable, and Theory-guided. The framework combines calibrated prospectivity, uncertainty, geological consistency, expected information gain, cost, logistics, and environmental constraints in a human-in-the-loop decision cycle. A minimum reporting standard and an implementation roadmap are provided for exploration teams seeking to deploy AI without creating opaque or overconfident target lists.

کلیدواژه ها

Mineral Prospectivity Mapping, Explainable Artificial Intelligence, Uncertainty Quantification, Multimodal Geoscience Data, Value Of Information.

وضعیت: پذیرفته شده برای ارائه شفاهی