Hyperspectral remote sensing of carbonatite hosted REE-deposits in Namibia

In this thesis, two case-studies for exploration of carbonatite hosted REE-deposits in Namibia using hyperspectral remote sensing data are presented. REEs have enormous economic importance because they are the major constituents of high-tech electronic devices and are non-substitutable.
The two areas of interest in Namibia are Epembe/Kunene region and Lofdal/Bergville Farm close to Khorixas. Namibia is convenient because of less vegetation, a good data coverage, and a good geological knowledge.
Multi- and hyperspectral data of multiple source and scale in combination with tectonic geomorphology were used for data analysis. In a first step, spectral signatures and geomorphic indices were extracted, followed by expert classification made by a Decision Tree approach. The detailed processing chain is shown in figure 1.
Standard procedures for spectral unmixing (e.g. Spectral Feature Fitting, Matched Filtering) are well suited to map the abundances of carbonate minerals. However, calcretes, meta-sedimentary units, and amphibolites have similar spectral signatures in the SWIR range and are difficult to distinguish.
More advantageous approaches, like Self-Organizing maps and Decision Trees, were used to discriminate the carbonatite from surrounding rocks with similar signatures. The carbonatites are associated to specific textural, geomorphic and structural patterns. Incorporating these features into hyperspectral data analysis improve classification results and reduce misclassifications for carbonatite.