Li, Xiaohui, Xue, Chen, Chen, Yuheng, Yuan, Feng, Li, Yue, Zheng, Chaojie, Zhang, Mingming, Ge, Can, Guo, Dong, Lan, Xueyi, Tang, Minhui, and Lu, Sanming
In recent years, many concealed skarn-type and porphyry-type deposits have been discovered in Chating area, which shows very good prospectivity to host hydrothermal deposits. In this paper, we carried out the 3D mineral prospectivity modeling for porphyry and skarn-type mineralization of the Chating area. During this study, implicit geological modeling and the "total litho-inversion" method were used to build the 3D geological models and optimize the 3D geological models, various 3D predictive maps were generated by multiple 3D methods according to the exploration model. Finally, we use a 3D convolutional neural network (3D CNN) model to integrate all of the 3D predictive maps and delineate the concealed targets. The results show that the high- prospective areas delineated based on the 3DMPM not only include the deposits which are used as training data, but also the deposits that have already been discovered in the study area. It means that the 3DMPM approach used in this paper has good prediction and generalization ability. Compared with the Logistic Regression model (LR), Support Vector Machines (SVM), and Radom Forest (RF) model, the 3D CNN model can efficiently capture the correlation between various deposit types and 3D predictive information, and has better commendable predictive capabilities. The results of this paper can not only provide some new highly prospective areas for further mineral exploration in Chating area but also can promote the research on 3DMPM with deep learning methods. [Display omitted] • 3D Convolutional Neural Network would provide a new and powerful tool for 3D mineral prospectivity modeling. • The "total litho-inversion" method is a viable tool for the optimization of the 3D geological models which can provide more reliable data for 3D mineral prospectivity modeling. • 3D mineral prospectivity modeling can effectively identify not only the known deposits (not training data), but also delineate several new favorable exploration areas within Chating area at depth. The Chating area is situated within the Middle-Lower Yangtze River Metallogenic Belt, China. Several concealed skarn and porphyry-type deposits have been discovered in this area, indicating high potential for hosting hydrothermal deposits. However, due to the complex geological structure, exploration risks significantly increase with increasing depth. To overcome this challenge, three-dimensional mineral prospectivity modeling (3DMPM) has begun to be widely applied for mapping the prospectivity of deep-seated and concealed mineralization. However, most previous studies on 3DMPM were based on shallow supervised machine learning models and dimensionality-reduced 3D predictive maps. Although these models have shown good results, they may lose spatial correlation within the 3D predictive maps and fail to explore nonlinear correlations between the 3D predictive maps and mineralization. Meanwhile, 3D geological models are the most important basis of the 3DMPM, however, in the past, few studies have incorporated the optimization of the 3D geological models into the process of 3DMPM. Therefore, this paper initially builds and optimizes 3D geological models through implicit 3D geological modeling and "total litho-inversion" approach. Subsequently, the 3D predictive maps are generated by employing various 3D methods, which are further integrated using a 3D convolutional neural network (3D CNN) model to identify highly prospective areas for mineralization. The results show that the highly prospective areas identified by the 3DMPM include not only the training data but also other mineral deposits that have previously been discovered within the study area. In addition, compared with the Logistic Regression model (LR), Support Vector Machines (SVM), and Radom Forest (RF), the 3D CNN performs better prediction capabilities due to its enhanced ability to capture the correlations between 3D predictive maps and multiple types of mineral deposits. It suggests that the 3DMPM based on the 3D CNN model has commendable predictive capabilities in identifying prospective mineralization areas, and some new highly prospective areas can be considered as priority areas for future exploration of concealed mineralization within the Chating Area. [ABSTRACT FROM AUTHOR]