Decision tree algorithms applied to detect mineral alteration related to cobalt-nickel mineralisation in Asturias: insights of S34I project

The S34I (Secure Sustainable Supply of Raw Materials for EU Industry) project aims to enhance European autonomy in raw materials production by prototyping Earth Observation (EO) methods, which can support the multiple-phase approach of the exploration and mining industry. The present study explores two ensemble machine learning algorithms based on decision trees, Extreme Gradient Boosting (XGBoost) and Random Forest (RF), to predict alteration zones associated with Copper-Cobalt-Nickel mineralisation in northwest Spain. The study site is located in Asturias, Spain, within the Saint Patrick License area, which encompasses the Aramo Plateau adjacent to the historical AramoTexeo Mine. The training dataset was extracted using bands from the Landsat-9 and PRISMA satellites referencing lithogeochemical alteration zones and associated anomalous mineralisation previously identified during active exploration programmes conducted by Aurum Global Exploration. Independent Component Analysis (ICA) was applied to the satellite bands to reduce the dimensionality and increase computational efficiency. As a result, the pixels of the image have been classified as either host rock or alteration zone. The RF algorithm achieved a mean classification accuracy of 0.97 for the PRISMA image. The accuracy for the Landsat-9 image was at 0.90. The XGBoost algorithm demonstrated an accuracy of 0.95 for the PRISMA image and 0.82 for the Landsat 9 image, indicating reduced overfitting. The results enable the creation of predictive mineral maps that can support exploration programmes for CRMs, establish the resource potential of new areas in Europe, and ultimately lead to sustainable and ethical European-based mining practices.