This study explores the use of geochemical and remote sensing data to predict areas with high mineralization potential in Mississippi Valley-Type (MVT) Pb–Zn deposits in western Semnan, Iran. Two methods were applied: 1DCNN (1D Convolutional Neural Network) and PRA (Predictive Raster Averaging). The 1DCNN model, known for its capacity to learn complex spatial patterns from high-dimensional data, was trained on satellite imagery and geochemical data, producing a probability map indicating potential mineralized zones. The simpler PRA method combined multiple raster layers through averaging, providing a computationally efficient alternative for generating prospectivity maps. Evaluation metrics, including Weighted Class Distribution (WCD) Method, ROC-AUC, and Feature Importance, were used to assess the performance of both methods. The WCD method was applied to evaluate the distribution of known mineral deposits across different probability classes, while ROC-AUC was used to measure the 1DCNN model's ability to distinguish between mineralized and non-mineralized areas. Feature Importance analysis helped identify which geochemical and remote sensing features contributed most to the predictions made by the 1DCNN model. Results showed that the 1DCNN model had higher accuracy in identifying the most promising mineralization zones, with 68.75 % of known deposits falling into the highest probability class, occupying 8 % of the study area. The PRA model, though less refined, successfully highlighted regions with mineral potential, albeit with a broader spread across probability classes. This research contributes to understanding the trade-offs between artificial intelligence-based and traditional mapping methods for mineral exploration, providing valuable insights for future studies in resource prospectivity.
扫码关注我们
求助内容:
应助结果提醒方式:
