Scholarlink

<p>Orogenic gold deposits account for roughly 30% of the world’s gold resources, yet remain challenging to explore due to their complex tectonic–fluid–alteration systems and the high cost of subsurface characterization. Surface geochemical anomalies—generated by long-range migration, enrichment, and depletion of mineralization-related elements—offer subtle but detectable clues to underlying alteration structures. However, existing qualitative approaches struggle to quantify the intricate, long-range coupling between geochemical signatures and alteration intensity. In this paper, we propose TransAlter, a Transformer-based framework for three-dimensional (3D) inference and modeling of deep-seated alteration zones in orogenic gold systems. Our basis is a conceptual model that formalizes the theoretical link between surface geochemical distributions and mineralized alteration. A 3D fault-constrained geological model was then constructed to spatially bound alteration zones, and a paired dataset of surface geochemistry and drillhole-derived alteration thickness measurements was assembled. TransAlter uses trainable positional encodings to embed spatial context and an end-to-end Transformer architecture to learn both local and global associations between surface element concentrations and alteration zone thickness. To bolster robustness against limited training samples, we integrate a self-supervised contrastive learning strategy that enforces prediction consistency across differently masked inputs. Finally, by extending TransAlter to deeper subsurface targets, we demonstrate its ability to reconstruct fully three-dimensional distributions of alteration thickness. The proposed method is validated through a case study of the Canzhuang gold deposit, located in the world-class Jiaodong gold province of eastern China. The results show that TransAlter not only improves the accuracy of thickness prediction but also produces geologically plausible 3D alteration models. This approach promises to enhance the targeting and characterization of concealed orogenic gold deposits, offering a scalable, data-driven tool for modern mineral exploration.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Toward 3D Inference and Reconstruction of Deep-Seated Alteration Zones in Orogenic Gold Deposits: Linking Surface Geochemical Data and Undercover Alteration Thickness with an End-to-End Transformer

  • 期刊论文

  • 发表日期：

    2026年1月20日

查看全文
  • Jin Chen,
  • Nanxin Liang,
  • Hao Deng,
  • Shuyan Yu,
  • Zhankun Liu,
  • Xiancheng Mao

摘要

Orogenic gold deposits account for roughly 30% of the world’s gold resources, yet remain challenging to explore due to their complex tectonic–fluid–alteration systems and the high cost of subsurface characterization. Surface geochemical anomalies—generated by long-range migration, enrichment, and depletion of mineralization-related elements—offer subtle but detectable clues to underlying alteration structures. However, existing qualitative approaches struggle to quantify the intricate, long-range coupling between geochemical signatures and alteration intensity. In this paper, we propose TransAlter, a Transformer-based framework for three-dimensional (3D) inference and modeling of deep-seated alteration zones in orogenic gold systems. Our basis is a conceptual model that formalizes the theoretical link between surface geochemical distributions and mineralized alteration. A 3D fault-constrained geological model was then constructed to spatially bound alteration zones, and a paired dataset of surface geochemistry and drillhole-derived alteration thickness measurements was assembled. TransAlter uses trainable positional encodings to embed spatial context and an end-to-end Transformer architecture to learn both local and global associations between surface element concentrations and alteration zone thickness. To bolster robustness against limited training samples, we integrate a self-supervised contrastive learning strategy that enforces prediction consistency across differently masked inputs. Finally, by extending TransAlter to deeper subsurface targets, we demonstrate its ability to reconstruct fully three-dimensional distributions of alteration thickness. The proposed method is validated through a case study of the Canzhuang gold deposit, located in the world-class Jiaodong gold province of eastern China. The results show that TransAlter not only improves the accuracy of thickness prediction but also produces geologically plausible 3D alteration models. This approach promises to enhance the targeting and characterization of concealed orogenic gold deposits, offering a scalable, data-driven tool for modern mineral exploration.