The potential of drones in multi-scale hyperspectral imaging for mineral exploration: Examples from Southern Africa

Abstract

The transition towards a green economy is paradoxically the main driver to the in- creased demand for resources. Recycling alone is not enough to sustain a purely circular economy, thus mineral exploration will still be required. Traditional exploration tech- niques are primarily based on extensive field work that are supported by geophysical sur- veying and drilling. These techniques can be restricted by field accessibility, financial status, area size and climate. Furthermore, these methods typically have a considerable footprint on the environment, upsetting the surrounding community and resulting in mis- trust in the exploration sector. This PhD introduces a novel multi-scale remote sensing ap- proach that incorporates state-of-the-art methods focused on the applicability of uncrewed aerial vehicles (UAVs)- and ground-based hyperspectral imaging (HSI) for mineral explo- ration. The aim of this approach is to improve eciency, reduce costs and increase the safety of field personnel. The work also shows the benefits by promoting non-invasive, innovative remote sensing methods to foster social acceptability. This approach is exemplified by the exploration of economically important commodi- ties that contribute to the development of green technologies. Deposits and geological bodies containing tin (Sn), lithium (Li) and rare earth elements (REEs) were used as case studies in order to test the multi-scale approach. These sites comprise the Li-bearing pegmatites of the Uis tin mine in Namibia, the REE-bearing carbonatite complexes of Marinkas Quellen and Lofdal in Namibia and finally the Zaaiplaats historic tin mine in South Africa. In the proposed scheme, satellite and aeroplane-based data constitute the first level of exploration. The second level consists of the improvement of field work by making use of hyperspectral imaging, both with ground-based and UAVs surveys. Each level of data acquisition comes with its own set of advantages and limitations; i.e., satellites can cover a large area extent, but the data are usually at a relatively low spatial resolution, while UAV-based data can have a high spatial resolution, UAVs themselves can only cover a limited area. By using a multi-scale approach, we can minimize each platform’s limita- tion and exploit their advantages. As this approach is innovative and yet untested, the core of the project focuses on the numerous aspects of the acquisition, processing and valida- tion of UAV-based hyperspectral data. In order to adequately demonstrate its relevance, this work highlights specific aspects of the multi-source, multi-scale approach adapted to the peculiarities of the case studies. Spectral tools and machine learning techniques were adapted to process satellite and plane-base data in order to locate areas of interests. In the present case, these methods were used to map the regional geology and identify potential zones of mineralisation for further investigation at the Sn-hosted granites of Zaaiplaats and the REE-bearing carbonatites of Lofdal. Following this, UAV-based hyperspectral data can vastly improve the accuracy of field mapping in mineral exploration. UAV-based measurements can supplement and di- v rect geological observation immediately in the field and therefore allow better integration with in-situ ground investigations. A hyperspectral camera was attached to a multi-copter to acquire data from the visible (VIS) to the near-infrared (NIR) range of the electromag- netic spectrum. The acquired data was then corrected for radiometric and geometric dis- tortions. In addition, high resolution digital surface models (DSM) and orthomosaics were generated using photogrammetry. The corrected data provides information on the spectral signatures of outcropping lithologies to the field geologists and the exploration teams. In the cases of Marinkas Quellen and Lofdal, this was achieved by using end-member mod- elling and classification techniques such as non-linear machine learning algorithms, e.g., spectral angle mapper (SAM) or minimum wavelength mapping (MWM). Ground truth points, which were collected during field work, were used as referencing. Furthermore, this work also shows a novel approach whereby 3D virtual outcrops are used to map vertical mine walls. Heavy sensors that cannot be mounted to a UAV such as short-wave infrared (SWIR) or long-wave infrared (LWIR) sensors can be used to image outcropping rocks in the field, or as validation tools in the laboratory. In the case study of the Li-bearing pegmatites at Uis, ground-based hyperspectral SWIR data was used in uni- son with photogrammetric 3D models to produce georeferenced 3D virtual models with hyperspectral attributes in order to map Li mineralisation by using MWM. To verify the ground-based data from Uis, decision trees allowed for the determination of the miner- alogy and mineral associations in hand samples and drill-cores. Lastly, to validate all the remote sensing data, in-situ measurements were taken with portable devices such as a handheld X-ray fluorescence (XRF), laser-induced breakdown spectroscopy (LIBS) and spectroradiometer. Geochemical analyses such as XRF, X-ray diraction (XRD) and in- ductively coupled plasma mass spectrometry (ICP-MS) were performed on field samples. Additional validation was performed in laboratories such as thin section microscopy, and advanced spectroscopic techniques such as mineral liberation analysis (MLA) and laser induced fluorescence (LiF). The results indicate that UAV-based surveying has a very high potential in fundamen- tally lowering the acquisition costs and increasing the amount of valuable information captured in the field. Together with ground-based hyperspectral imaging, these techniques can be seamlessly used in a multi-scale approach in the exploration of other deposits as well. Furthermore, innovative, non-invasive and more sustainable techniques in explo- ration, such as the one proposed in this work, encourage social acceptability in the com- munity at large and promote the use of greener technology. Finally, this work has laid the foundation for future research in the integration of UAV- based hyperspectral and geophysical data. This would allow us to gain information not just on the surface, but fuse surficial with sub-surface information. Additionally, by incor- porating a time component with repeated acquisitions, 4D models can be made that would allow to monitor the evolution of industrial or environmental targets.