Evidential 3D swin transformer–CNN for uncertainty-aware blind gold prospectivity mapping: A case study from Siahcheshmeh, NW Iran

Articles in Press

Document Type : Research Paper

Authors

School of mining engineering, university of Tehran, Tehran, Iran.

10.22059/ijmge.2026.406125.595334

Abstract

Discovering blind gold deposits beneath post-mineral cover remains a major challenge in mineral exploration, where drilling decisions are high-risk and capital intensive. Conventional 2D prospectivity mapping and deterministic 3D machine-learning models fail to capture subsurface geological complexity and do not provide quantitative measures of predictive confidence. This study introduces an Evidential 3D Swin Transformer–CNN framework for uncertainty-aware blind gold prospectivity mapping. The hybrid architecture integrates multi-scale 3D convolutional feature extraction with hierarchical Swin Transformer self-attention to model local geological signatures and kilometre-scale structural controls. An evidential learning head under a Dirichlet prior enables analytical decomposition of aleatoric and epistemic uncertainties in a single forward pass, eliminating Monte Carlo overhead. The framework was applied to the Discovering blind gold deposits beneath post-mineral cover remains a major challenge in mineral exploration, where drilling decisions are high-risk and capital intensive. Conventional 2D prospectivity mapping and deterministic 3D machine-learning models fail to capture subsurface geological complexity and do not provide quantitative measures of predictive confidence. This study introduces an Evidential 3D Swin Transformer–CNN framework for uncertainty-aware blind gold prospectivity mapping. The hybrid architecture integrates multi-scale 3D convolutional feature extraction with hierarchical Swin Transformer self-attention to model local geological signatures and kilometre-scale structural controls. An evidential learning head under a Dirichlet prior enables analytical decomposition of aleatoric and epistemic uncertainties in a single forward pass, eliminating Monte Carlo overhead. The framework was applied to the Siahcheshmeh intrusion-related gold system (NW Iran) using twelve co-registered 3D evidence layers at 10 m voxel resolution. On an independent test set of 97,650 voxels from 64 previously undrilled boreholes, the model achieved ROC-AUC of 0.981 and PR-AUC of 0.967, while maintaining low epistemic uncertainty across the domain. High-confidence blind targets, defined by combined probability and evidence strength thresholds, coincide with fault intersections and phyllic-altered intrusive rocks, consistent with geological controls and drilling outcomes reported in the study area. These results demonstrate that integrating global structural modeling with uncertainty quantification enables risk-informed 3D prospectivity mapping in covered terranes.

Keywords

Main Subjects

References

[1]. Arndt, N. T., Fontboté, L., Hedenquist, J. W., Kesler, S. E., Thompson, J. F., and Wood, D. G. (2017). Section 3. Mineral exploration: discovering and defining ore bodies, Geochemical Perspectives, vol. 6, no. 1, pp. 52-85.
[2]. Reich, M. and Simon, A. C. (2024). Critical minerals, Annual Review of Earth and Planetary Sciences, vol. 53.  https://doi.org/10.1146/annurev-earth-040523-023316.
[3]. Davies, R. S., Trott, M., Georgi, J., and Farrar, A. (2025). Geosystems and Geoenvironment.  https://doi.org/10.1016/j.geogeo.2025.100361.
[4]. Schodde, R. (2025). Mineral Deposit Exploration—Discovery Trends: 1900–2023, SEG Discovery, no. 142, pp. 19-35.  https://doi.org/10.5382/Geo-and-Mining-28.
[5]. Behnia, P., Harris, J., Sherlock, R., Naghizadeh, M., and Vayavur, R. (2023). Mineral prospectivity mapping for orogenic gold mineralization in the rainy river area, Wabigoon subprovince, Minerals, vol. 13, no. 10, p. 1267.  https://doi.org/10.3390/min13101267.
[6]. Carranza, E., Hale, M., and Faassen, C. (2008). Selection of coherent deposit-type locations and their application in data-driven mineral prospectivity mapping, Ore geology reviews, vol. 33, no. 3-4, pp. 536-558.  https://doi.org/10.1016/j.oregeorev.2007.07.001.
[7]. Zuo, R., Yang, F., Cheng, Q., and Kreuzer, O. P. (2024). A novel data-knowledge dual-driven model coupling artificial intelligence with a mineral systems approach for mineral prospectivity mapping, Geology.  https://doi.org/10.1130/G52970.1.
[8]. Rodriguez-Galiano, V., Sanchez-Castillo, M., Chica-Olmo, M., and Chica-Rivas, M. (2015). Machine learning predictive models for mineral prospectivity: An evaluation of neural networks, random forest, regression trees and support vector machines, Ore Geology Reviews, vol. 71, pp. 804-818.  https://doi.org/10.1016/j.oregeorev.2015.01.001.
[9]. Zhang, Z. et al. (2023). Supervised mineral prospectivity mapping via class-balanced focal loss function on imbalanced geoscience datasets, Mathematical Geosciences, vol. 55, no. 7, pp. 989-1010.  https://doi.org/10.1007/s11004-023-10065-x.
[10]. Pourgholam, M. M., Adib, A., Afzal, P., Rahbar, K., and Gholinejad, M. (2025). Deep learning and fractal-wavelet techniques for magnetite-apatite exploration in Tarom Iran, Scientific Reports, vol. 15, no. 1, p. 31907.  https://doi.org/10.1038/s41598-025-16040-2.
[11]. Li, R., Tan, S., Zhang, M., Zhang, S., Wang, H., and Zhu, L. (2024). Geological Disaster Susceptibility Evaluation Using a Random Forest Empowerment Information Quantity Model, Sustainability, vol. 16, no. 2, p. 765.
[12]. Deng, H., Zheng, Y., Chen, J., Yu, S., Xiao, K., and Mao, X. (2022). Learning 3D mineral prospectivity from 3D geological models using convolutional neural networks: Application to a structure-controlled hydrothermal gold deposit, Computers & Geosciences, vol. 161, p. 105074.  https://doi.org/10.1016/j.cageo.2022.105074.
[13]. Pourgholam, M. M., Afzal, P., Adib, A., Rahbar, K., and Gholinejad, M. (2024). Recognition of REEs anomalies using an image Fusion fractal-wavelet model in Tarom metallogenic zone, NW Iran, Geochemistry, vol. 84, no. 2, p. 126093.
[14]. Chen, J. et al. (2024). 3D mineral prospectivity modeling using deep adaptation network transfer learning: A case study of the Xiadian gold deposit, Eastern China, Geochemistry, vol. 84, no. 4, p. 126189.  https://doi.org/10.1016/j.chemer.2024.126189.
[15].  Liu, Z. et al. (2024). 3D mineral prospectivity modeling in the Sanshandao goldfield, China using the convolutional neural network with attention mechanism, Ore Geology Reviews, vol. 164, p. 105861.  https://doi.org/10.1016/j.oregeorev.2023.105861.
[16]. Sillitoe, R. H. (2020). Gold deposit types: An overview.  https://doi.org/10.5382/SP.23.01.
[17]. Haghighi, A., Imamalipour, A., and Barak, S. (2025). Leveraging EDA, Fractal Analysis, SFA, and GMPI for Identifying Geochemical Anomalies in Siah Cheshmeh Ophiolite Area, NW Iran, Geopersia, vol. 15, no. 1, pp. 137-161.  https://doi.org/10.22059/geope.2024.380369.648769.
[18]. Sillitoe, R. H. (2010). Porphyry copper systems, Economic geology, vol. 105, no. 1, pp. 3-41.
[19]. Ebrahimi, S., Pan, Y., and Rezaeian, M. (2021). Origin and evolution of the Masjed Daghi Cu-Au-Mo porphyry and gold epithermal vein system, NW Iran: constraints from fluid inclusions and sulfur isotope studies, Mineralogy and Petrology, vol. 115, no. 6, pp. 643-662.  https://doi.org/10.1007/s00710-021-00761-z.
[20]. Mohammadi, A., Burg, J.-P., and Guillong, M. (2022). The Siah Cheshmeh-Khoy-Misho-Tabriz fault (NW Iran) is a cryptic neotethys suture: evidence from detrital zircon geochronology, Hf isotopes, and provenance analysis, International Geology Review, vol. 64, no. 2, pp. 182-202.  https://doi.org/10.1080/00206814.2020.1845992.
[21]. Goovaerts, P., Geostatistics for natural resources evaluation. Oxford university press, 1997.
[22]. Zhou, G., Chen, J., An, W., Liu, C., and Li, W. (2024). Three-dimensional mineral prospectivity mapping based on natural language processing and random forests: A case study of the Xiyu diamond deposit, China, Ore Geology Reviews, p. 106082.  https://doi.org/10.1016/j.oregeorev.2024.106082.
[23]. Szegedy, C., Ioffe, S., Vanhoucke, V., and Alemi, A., "Inception-v4, inception-resnet and the impact of residual connections on learning," in Proceedings of the AAAI conference on artificial intelligence, 2017, vol. 31, no. 1, doi: https://doi.org/10.1609/aaai.v31i1.11231.
[24]. He, K., Gkioxari, G., Dollár, P., and Girshick, R., "Mask r-cnn," in Proceedings of the IEEE international conference on computer vision, 2017, pp. 2961-2969.
[25]. Sensoy, M., Kaplan, L., and Kandemir, M. (2018). Evidential deep learning to quantify classification uncertainty, Advances in neural information processing systems, vol. 31.
[26]. Ulmer, D. (2024). On uncertainty in natural language processing, arXiv preprint arXiv:2410.03446.  https://doi.org/10.48550/arXiv.2410.03446.
[27]. Senge, R. et al. (2014). Reliable classification: Learning classifiers that distinguish aleatoric and epistemic uncertainty, Information Sciences, vol. 255, pp. 16-29.
 
 
International Journal of Mining and Geo-Engineering

Articles in Press, Accepted Manuscript
Available Online from 13 May 2026

Files

Share

How to cite

Statistics