Abstract

In this paper, a methodological approach to classify lithology features using random forest and multisensor satellite imagery is presented. The incorporation of remote sensing techniques into regional geology mapping is important in order to make possible the implementation of monitoring strategies aimed to enhance national capabilities to provide information for natural disasters management, sustainable exploitation of natural resources, etc. In this context, the use of both optical/infrared and multi-wavelength synthetic aperture radar satellite data was evaluated in order to train a random forest algorithm, taking as reference field data, to generate a lithological map. Preliminary results have shown the capability of this approach to lithological classification with accuracies values higher than 80% in some conditions to map sedimentary, igneous, and metamorphic lithological classes.