3D Mineral Prospectivity Mapping with Random Forests: A Case Study of Tongling, Anhui, China

In the past few decades, a variety of data-driven predictive modeling techniques has led to a dramatic advancement in mineral prospectivity mapping (MPM). The random forests (RF) algorithm, a machine learning method, has been applied successfully to data-driven MPM. However, there are two main challenges that need to be examined. Firstly, whether RF modeling can be used for the 3D MPM. The voxel (in 3D) has replaced the pixel (in 2D) to represent geological features, and so the capability of the RF model should be tested. Secondly, when we conduct regional-scale MPM, building a suitable conceptual model has a significant influence on the results; however, mineral deposit models often focus on just deposit-scale features. These two challenges were encountered in the case study in the Tongling ore cluster, which is the most representative skarn ore-concentrated area in the Middle-Lower Yangtze River Valley Metallogenic Belt in Eastern China. Thus, 3D geological models of the Tongling ore cluster were constructed from the multiple geological datasets. Then, a conceptual model was translated into 3D predictor layers. Finally, we tested and compared the MPM capabilities of the RF and compared it with weights-of-evidence (WofE) modeling. The results indicate that RF modeling not only outperforms WofE modeling in 3D MPM, but it also has capability to assess the relative importance of different predictor layers. Further testing of this method is warranted in other areas with different scales or metallogenic model to investigate fully its efficiency.