Using Three-dimensional Modeling and Random Forests to Predict Deep Ore Potentials: A Case Study on Xiongcun Porphyry Copper–Gold Deposit in Tibet, China

Abstract

The chances of discovering hidden deposits are higher when exploring deeper into known deposits or historic mines, compared to broad-scale regional exploration. Machine learning algorithms and three-dimensional modeling can effectively identify deep targets and provide quantitative predictions of potential resources. This research paper presents a proposed workflow that utilizes random forest algorithms and a three-dimensional model incorporating geological factors such as strata, lithology, alteration, and primary halo to enhance the accuracy of exploration predictions. The study involved collecting 7949 rock samples from 34 boreholes in eight exploration lines at the Xiongcun No. 2 deposit, and performing geochemical analysis calculations on 18 elements. The methodologies employed can be summarized as follows: (1) establishing and preprocessing the geological dataset of the Xiongcun No. 2 deposit, followed by multivariate statistical analysis, (2) delineating primary halo zoning sequences to identify potential mineralization at greater depths, (3) constructing three-dimensional models incorporating geological and geochemical mineralization information, and (4) utilizing the random forest algorithm to extract exploration criteria and quantitatively predict deep exploration targets. The results indicate a significant mineralization located 300 m to the west–northwest of the No. 2 deposit, within the downward extension of the control depth. The three-dimensional model of the target volume reveals the presence of approximately 0.33 million tons of copper (Cu), 7.6 tons of gold (Au), and 22.8 tons of silver (Ag).