Alteration zone mapping in tropical region: A comparison between data-driven and knowledge-driven techniques

Mapping alteration zones, a crucial step for mineral exploration, faces challenges in tropical areas. Dense vegetation hides important geological features, recent clay formation hides deeper alterations, and human activities like farming make it more complicated. However, alteration zones are crucial clues for specific ore deposits. We explore two approaches: one based on knowledge and the other on data. The knowledge-driven method involves experienced geologists analyzing GIS layers, including lineaments, drainage patterns, rock types, and topography. They use this data to identify key signs of ore-forming alterations. Translating this expert knowledge into spatial data helps us map alteration zones effectively. While this approach provides good approximations, it lacks direct evidence. The data-driven method involves advanced remote sensing tools like ASTER imagery. High-resolution data allows us to use image processing techniques to extract alteration information. However, conventional techniques face challenges in the tropics due to dense vegetation and human activity. To overcome this, we use machine learning algorithms trained on carefully selected samples. We found that among selected ASTER-derived products of conventional DIP techniques (reflectance, band ratio, PCA, DPCA), directed PCA alone is capable of demarcating alteration for the study area with a total accuracy of 81.41, 83.92, and 84.42% for LR, ANN, and RF, respectively. Besides, we used contextual geological evidence of alteration presence as another validation method. To validate results, we use the knowledge-driven approach again, employing Relative Alteration Indexes. All alteration indicative field and geological knowledge were weighted with the Analytical Hierarchy Process (AHP) and spatially integrated with three probability classes in the GIS platform. This combined strategy reveals that while Random Forest has the highest accuracy, Logistic Regression yields more geologically significant results. The high value of Relative Alteration Indexes representing highly altered zones indicates their successful mapping from both data and knowledge-driven techniques. This study shows the strengths of both approaches in understanding alteration zones in the tropics. By combining expert knowledge with advanced technology, we can pinpoint areas rich in valuable minerals, even in difficult-to-explore places. Our successful test in the South Purulia region suggests similar discoveries are possible in other unknown areas.