Multiscale hyperspectral remote sensing of REE deposits in Southern Africa

Abstract

Traditional exploration techniques are mainly based on extensive field work, supported by geophysical surveying and drilling. These techniques can be restricted by field accessibility, financial status, area size and climate. Additionally, they usually have a considerable footprint on the environment. This work aims to demonstrate that multi-scale, multi-source remote sensing can 1) improve the efficiency of modern exploration, 2) reduce the costs and increase the safety of the personal involved and, most important, 3) improve the acceptance of exploration by using non-invasivetechniques. This approach is exemplified by the exploration of Rare Earth Elements in Southern Africa. Both spectral and geometrical data were implemented in the processing. Additional information such as geological maps were used when available. In the proposed scheme, satellite and airborne data, when available, constitute the first level of exploration. Machine learning algorithms such as Support Vector Machines are used to locate area of interests. In the present case, carbonatites, as main hosts for REEs were identified with good accuracy. Known and poorly known carbonatite complexes were identified and characterized. The study areas chosen for the demonstration of the proposed method were the carbonatite-hosted REE deposits of Lofdal, Namibia, and Zandkopsdrift, South Africa as well as the carbonatite complex of Marinkas Quellen in Namibia. The second level consists of the improvement of field work by making use of hyperspectral imaging, bothground-basedand,more importantly, unmannedaerial vehicles (UAV) were used. As this approach is innovative, the core of the project focusses on the numerous aspects of the acquisition, processing and validation of UAV-basedhyperspectraldata.