Artificial neural networks applied to mineral potential mapping for copper-gold mineralizations in the Carajas Mineral Province, Brazil

摘要

Motivated by recent successful results of artificial neural network modelling in a variety of problems related to the geosciences, we have applied a radial basis functional link net to a regional-scale mapping of the potential for Cu-Au mineralizations in the Carajas Mineral Province, northern Brazil. To derive the input feature vectors, we have used geological and both radiometric and magnetic geophysical data. A k-fold cross-validation method was employed in order to tune the parameters of the network and to select the best radial basis functional link net model amongst several others. Subsets of the available data set were used for training and validation and the estimated overall accuracy of the selected model is 91.7%. The plotting of a cumulative area versus favourability curve allowed us to define favourability zones of occurrences of Cu-Au mineralizations and to assess the efficiency and the predictive power of the model. A binary map showing high and low favourability sectors was produced for the study area as an end product that can be used to guide and support more detailed exploration efforts. Our results show that 4.18% of the study area has an extremely high potential to contain Cu-Au mineralizations, especially those of iron-oxide Cu-Au type, which are related to volcanic rocks and hydrothermal alteration.