3D mineral prospectivity modeling for concealed gold deposits based on 3DCNN and numerical modeling of ore-forming process in the Wuhe area, Anhui Province, China

Here we present a 3D mineral prospectivity modeling approach for targeting concealed gold deposits in the Wuhe area, integrating 3D convolutional neural network (3DCNN) and numerical modeling of ore-forming process. The results of this study not only enhance the understanding of the structural controls on lode gold deposits and delineate four targets for further mineral exploration in the Wuhe area, but also indicate that the MSAM-3DCNN model has significant advantages in the integration of multi-source predictive maps, particularly when incorporating additional 3D predictive maps generated by numerical modeling of ore-forming process • The numerical modeling of ore-forming process can enhance the understanding of the structural controls on gold mineralization. • The MSAM-3DCNN model can effectively integrate 3D predictive maps generated by numerical modeling. • 3D mineral prospectivity modeling is a powerful tool for targeting concealed lode gold deposits. The Wuhe area is located in the southeastern margin of the North China Craton (NCC), adjacent to the Dabie–Sulu orogenic belt. It is considered an extension of the Zhaoyuan–Yexian gold belt to the west of the Tan–Lu Fault Zone. However, only a limited number of small- to medium-sized deposits have been discovered thus far in the Wuhe area. Recently, Three-dimensional mineral prospectivity modeling (3DMPM) has become a powerful tool for targeting deep concealed deposits. The generation of 3D predictive maps associated with mineralization is a crucial step in 3DMPM. The numerical modeling of ore-forming process has gradually been integrated into 3DMPM. Numerical modeling can provide useful predictive information for 3DMPM. However, better ways to integrate these additional 3D predictive maps during 3DMPM require further exploration and research. Besides, the 3D convolutional neural network (3DCNN) model has demonstrated excellent data fusion capabilities when applied to 3DMPM for Skarn-type ore deposits. However, the application of 3DCNN in mineral prospectivity modeling for lode gold deposits has not yet been explored. Therefore, this study initially adopts the implicit 3D geological modeling method to build 3D geological models. Subsequently, 3D spatial analysis and numerical modeling are employed to generate 3D predictive maps. Finally, a 3DCNN model incorporating a spatial attention mechanism and an Inception module (MSAM-3DCNN) is employed to integrate the 3D predictive maps and delineate several highly prospective areas for future exploration. The high prospectivity areas not only delineate the data-driven deposits but also include the other deposits in the Wuhe area which demonstrates that the methods employed in this paper can map the mineral prospectivity of lode gold deposits efficiently. Numerical modeling can provide a series of 3D maps that significantly enhance the effectiveness of the 3DMPM, especially in modeling the potential mineralization of the lode gold deposits, where 3D predictive maps are limited. The MSAM-3DCNN model has significant advantages in multi-source predictive data integration compared to the Logistic Regression (LR) and Radom Forest (RF) models, particularly by integrating more 3D maps generated by numerical modeling.