Lithology Mapping with Satellite, Fieldwork-Based Spectral Data, and Machine Learning : The Case Study of Beiras Group (Central Portugal)

Geological remote sensing has been an invaluable means to obtain data to perform geological mapping objectively and with high accuracy. The lack of geological cartography decreases the knowledge of geological resources and land management in regions that benefit greatly from this information. There is a large gap of geological cartography information at 1:50 000 scale throughout the territory in mainland Portugal. Therefore, this work intends to address this issue by applying techniques to complement classic geological mapping methods, using the Beiras Group (mainland Portugal) as the case study area. The methodology used various data sources to increase the accuracy of the results obtained. Multispectral analysis was carried out in two Landsat-8 images of 2015, one in the dry season and the other in the rainy season. Hyperspectral data were obtained using a 400-1000 nm spectroradiometer applied to 24 rock samples collected in two field campaigns. Geochemical data was carried out using X-Ray Fluorescence of glass beads with rock powder (<125 μm). All these methods validated the spectral results obtained. There are significant spectral and geochemical differences between two lithological main units: plutonic (Granito de Coentral and Granito de Vila Nova) and metamorphic (Boque Serpins Formation and Caneiro Formation). Machine learning models were created using Landsat-8 reflectance data and showed a high success rate in both confusion matrixes