Geological Reliability in Deep Learning Mineral Prospectivity Mapping: A Review
Minerals · 16.10.2026 · 16 (10) · 1066
DOI: 10.3390/min16101066 ↗Kurz gefasst
Fachartikel in der Zeitschrift Minerals, online erschienen am 16.10.2026. Verfasst von Tayssir Mohamed und 5 weiteren Autor:innen. Die Arbeit nennt 132 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Guizhou Provincial Scientific and Technological Program, Guizhou Provincial Department of Science and Technology.
Abstract (Originalsprache)
Artificial intelligence (AI)-based mineral prospectivity mapping (MPM) has rapidly advanced through the integration of deep learning, multisource geoscience data, and increasingly complex representation-learning approaches. However, the reliability of AI-derived prospectivity predictions remains a critical challenge because high predictive performance does not necessarily indicate geological validity, spatial transferability, or exploration usefulness. The key challenge is therefore not model complexity, but aligning geological knowledge, data representation, validation, and uncertainty assessment to ensure reliable interpretation. This review provides a structured critical synthesis of geological reliability in deep learning-based MPM, focusing on the methodological factors that influence whether AI outputs can support defensible exploration decisions. Rather than comparing algorithms solely by predictive accuracy, we examine the interconnected roles of geological representation, prediction support, label construction, negative-sample strategies, validation design, uncertainty quantification, explainability, transferability, and independent geological verification. Assessment of representative AI-based MPM studies reveals inconsistent reporting of these reliability dimensions, limiting reproducibility and meaningful comparison across studies. The review identifies recurring reliability bottlenecks, including spatial leakage caused by inappropriate validation strategies, uncertainty introduced during evidence-layer construction, ambiguity associated with unlabelled geological regions, and limitations in interpreting model explanations as geological causality. To organize these challenges, we propose a hierarchical evidence framework consisting of eight levels: methodological feasibility, internally validated predictive performance, spatial or temporal transferability, geological coherence, calibrated uncertainty, independent geological evidence, prospective field or drilling evidence, and demonstrated decision value. This framework clarifies the strength of claims that can be supported by different levels of evidence and provides a basis for transparent reporting and reliability assessment. Future progress in AI-based MPM will require integration of predictive capability with geological knowledge, robust validation, uncertainty-aware modelling, interpretable predictions, and independent geological verification. The proposed framework aims to support the development of more reproducible, transferable, and decision-relevant AI-assisted mineral exploration workflows.
Förderung
- National Natural Science Foundation of China
- Guizhou Provincial Scientific and Technological Program
- Guizhou Provincial Department of Science and Technology
- Hubei Province for Basic Research Platforms
Zitieren
APA
Mohamed, T., Liu, G., Bakheit, A., Hussein, H., Taha, A. M., & Idris, E. (2026). Geological Reliability in Deep Learning Mineral Prospectivity Mapping: A Review. Minerals, 16(10), 1066. https://doi.org/10.3390/min16101066
Vancouver
Mohamed T, Liu G, Bakheit A, Hussein H, Taha AM, Idris E. Geological Reliability in Deep Learning Mineral Prospectivity Mapping: A Review. Minerals. 2026;16(10):1066. doi:10.3390/min16101066
BibTeX
@article{Mohamed2026,
author = {Mohamed, Tayssir and Liu, Gang and Bakheit, Aref and Hussein, Hashim and Taha, Abdallah Mohamed and Idris, Eltahir},
title = {Geological Reliability in Deep Learning Mineral Prospectivity Mapping: A Review},
journal = {Minerals},
year = {2026},
volume = {16},
number = {10},
pages = {1066},
doi = {10.3390/min16101066}
}
EUVIMED