The sample preparation varies by sample type and depends on the material and the information of interest. This is based on the research projects that the samples stem from and for which the spectral and geochemical data were acquired.

The reflectance spectra for each sample were manually extracted from the processed hyperspectral image scenes by averaging a number of pixels over a central sample area. The resulting spectra were compiled in a spectral library. Thereby, each reflectance spectrum of a spectral library represents an average reflectance spectrum of the material, depending on the sample size and spectral homogeneity. The extraction of the reflectance spectra is explained in detail in each data description (Koellner et al., 2019; Koerting et al., 2019a, b).

The REO powders were measured in 100 % quartz glass petri dishes underlain by black cellular rubber; each powder was measured separately. Figure 2 shows the measurement setup of holmium-oxide powder as an example for the REO powders. The REE-bearing minerals were measured separately. Figure 3 shows the xenotime sample (brownish single crystal embedded in quartz) as an example for the REMin samples. The REMin samples were measured without sample preparation on black cellular rubber, as is shown for the copper-bearing minerals in Fig. 4. For all measurements, the final reflectance spectral analyses were spatially reduced to the center pixels of each identified REE-bearing mineral or a 5 pixel × 5 pixel average reflectance spectrum centered on the REO powder sample. Shadow effects from the sidewalls of the boxes could thus be minimized. One representative reflectance spectrum of every REMin and REO sample was collected for the spectral library (Herrmann, 2019).

The copper-bearing mineral samples were measured without any sample preparation as the variable surface of the minerals and the influence of the mineral structure were of interest. Figure 4 shows an example scan of some of the copper-bearing minerals. The full sample list including sample photos and the marked area of the geochemical sampling can be found in the technical report (Koellner et al., 2019). The area used to obtain the spectrum, averaging over a 5 pixel × 5 pixel window, was sampled afterwards for the geochemical analysis.

The Apliki mine samples were crushed and powdered so that ≥ 85 % of the sample was below 75 µm. Homogenized powders were measured as pressed powder tablets (Fig. 5). The area to obtain the sample's reflectance spectrum was chosen over a 5 pixel × 5 pixel window in the center of the powder tablet to minimize influences from the tablet's metal frame. The dark spots in each tablet were caused by previous measurements with a laser-induced breakdown spectrometer (LIBS). The hyperspectral sample spots were chosen in order to exclude the measurement points of the LIBS in the spectral footprint. In the case of broken powder tablets like 7d_Hem, the shadowed, rough surface areas were excluded from the spectral sampling, and an even powder surface was favored.

The HySpex VNIR-1600 and SWIR-320m-e (technical description available at http://hyspex.no/products/disc.php, last access: 18 June 2019) are two line-scanning cameras mounted in parallel. They cover the range of the visible to near infrared (VNIR, 414–993 nm) and the shortwave infrared (SWIR, 967–2498 nm) wavelength region. The sensors record an array line of 1600 pixels (VNIR) and 320 pixels (SWIR) (push-broom scanning). Every pixel contains a spectrum with a total spectral sampling number of 408 bands in total.

The HySpex cameras are provided with two acquisition modes: one for airborne data collection and one for laboratory measurements. In laboratory mode, the cameras are combined with a trigger pulse-moving sleigh (translation stage) of a definable frame period (depending on the integration time of every array-line acquisition). The configuration of the translation stage framework, the cameras and the light source (Halogen GX6.35, 2 × 1000 W, 45∘ illumination angle) are fixed, while the sleigh and the samples are moving through the focal plane (Rogass et al., 2017).

The reflectance level of a white reference panel, placed in line with the samples, is chosen according to the albedo of the samples. The higher the albedo of the sample, the higher is the diffuse reflectance factor of the white reference panel that is chosen. For the REE samples (REMin and REO), a white reference panel of 95 % reflectance was used because most of the REO samples were bright, white powders of a high albedo; this is based on test measurements of Bösche (2015), Herrmann (2019). The Apliki samples required a 50 % reflectance white reference panel, whereas the copper-bearing minerals were measured using a 20 % reflectance white reference panel. Both the geometrical setup and the heat-up time of the lamp influence the configuration of the light source. The maximum illumination was obtained with an angle of 45∘ between the incident light and the vertical plane. The distance between the lamp and the HySpex cameras was higher compared to the distance between the samples and the sensor to ensure diffuse illumination and to avoid thermal influence on the cameras and the samples. The integration time (meaning measurement time for each image line) was tested to be as high as possible to suppress the impact of signal uncorrelated Gaussian white noise and at the same time as low as needed to avoid detector saturation. For all measurements the integration time was chosen with respect to the sample albedo. The HySpex sensor characteristics are listed in Table 1. The settings used for the REMin and REOs are listed in Table 2, the settings for the copper-bearing minerals in Table 3 and the settings for the Apliki mine samples in Table 4. The laboratory is equipped with black-painted walls and doors, as well as black curtains to avoid reflected light from surfaces other than the sample; an example setup of the sensors, the translation stage and the samples can be seen in Fig. 6. The laboratory conditions were kept stable, the air temperature was regulated to 21 ± 0.5 ∘C and the humidity was below 70 % for all measurements. Black cellular rubber is used as a base material for all samples for hyperspectral data acquisition. It reflects less than 5 % on average of the incoming radiation.

Detailed descriptions for the GFZ standard measurements and the process chain can be found in Rogass et al. (2017).

Each measurement run produces one VNIR and one SWIR 3D data cube. The three dimensions are the two spatial x and y dimensions and the spectral z dimension. The 3D image cubes are produced by moving a homogeneous reflecting white reference panel and the samples through the focal plane of the two sensors. The VNIR image cube is resized to the spatial dimensions of the SWIR data cube, co-registered and stacked with the SWIR data cube, resulting in a continuous image cube with the spectral range of 414–2498 nm. In order to produce a reflectance image, the image pixels that show the white standard were averaged to a one-line reference spectrum. The reflectance was calculated by dividing every image line spectrum by its reference spectrum from the reflecting white reference panel. The resulting reflectance data are scaled from 0–10 000. A detailed description for the laboratory setup and processing can be found in Rogass et al. (2017). The software HySpex Ground was used to perform the measurements, and the software HySpex Rad was used to perform the radiometric calibration on the image data.

Depending on the sample type, the geochemical analysis methods differ. The methods used for each sample type are listed in Table 5.

The REE-bearing minerals are listed in Table 8. Figure 8 shows a plot of the corresponding spectral library. Table 9 lists the rare-earth oxide powders, and Fig. 9 shows the plot of the spectral library.

The copper-bearing minerals are listed in Tables 10 (copper-bearing sulfides and native copper) and 11 (copper-bearing silicates, carbonates and sulfates). The plots of the corresponding spectral libraries are shown in Fig. 10 (copper-bearing sulfides and native copper) and 11 (copper-bearing silicates, carbonates and sulfates).

The Apliki mine samples are listed in Table 12, and the plots of the corresponding spectral library are shown in Figs. 12 and 13.

The REO powders were certified to contain at least 99.9 % of the corresponding REO. The certificates are listed in Koerting et al. (2019a). The REE mineral samples were geochemically analyzed using the Thermo Niton XL3t (Fisher Scientific, 2002) device. The resulting element concentrations and the measurement error (2σ) are provided in Koerting et al. (2019a). The validation for the copper-bearing minerals can be found in Koellner et al. (2019), and the Apliki mine sample validation, analyzed by BVM, can be found in Koerting et al. (2019b).

Systematic errors are discussed based on instrument drift, calibration and optimization of measurements. Initializing a warm-up phase of optical components, detectors and lamps reduced influences due to instrument drift. Additionally, laboratory conditions were monitored to ensure a stable temperature and humidity. The HySpex cameras and the reference standards are factory calibrated once per year. Measurements used for the final reflectance spectral library were collected within one calibration time span to ensure equal acquisition conditions. For HySpex, averaging multiple measurements minimizes variations in the data. An average (median) of 500 to 800 pixel reflectance spectrum was taken for the HySpex REE and REO reflectance spectra. This number relates to the maximum number of non-disturbed pixels per sample region of interest (e.g., pixels that were not shadowed from the sample holder side walls). For the copper-bearing minerals and the Apliki mine powders, a 5 × 5 average pixel window was chosen over the area of interest. For these samples using a smaller pixel number for the average was necessary, as the sampling of the copper-bearing minerals for geochemical validation occurred over a small area of the sample, and the Apliki mine powder tablets were too small to ensure a larger homogenous area.

Variations of measurements were not only based on instrument calibrations or drift. They can also occur due to the detector geometry or geochemical properties of the minerals. These variations may appear as a shift of the peak positions of the absorption bands. This means that different hyperspectral sensors will show variations in the spectrum of the same material. By only using one set of hyperspectral sensors, the HySpex VNIR and SWIR, these shifts will not appear in our data sets. They might show when comparing our reflectance spectra of a material with reflectance spectra taken from a different instrument. For the copper-bearing minerals, the sample reflectance spectra also differ when comparing different samples of the same mineral species (e.g., malachite) to each other. The spectral signal differs, for example, due to changes in geochemistry and physical appearance, e.g., crystallization and degree of weathering (Clark, 1999; Hunt, 1989; Hunt and Ashley, 1979). To avoid measurement variations caused by different sensors, imaging data from the same sensors as the spectral library should to be used for the analysis. An example for an application can be using the here provided spectral library of the Apliki mine samples for an analysis of the HySpex hyperspectral imaging data of the Apliki mine face to be published in 2021 (Koerting et al., 2021).

The XL3t is internally calibrated and provides an internal warm-up phase to guarantee stable measurement conditions. Unlike the spectrometer measurements, experimental error was only provided for the XL3t. In order to reduce the experimental error, a long duration measurement time of 120 s was set. The XL3t collects the emitted radiation from the sample using four different filters. While the sample was irradiated, each filter measures counts per second within a time span of 30 s. Next, the average counts per second were internally transformed to parts per million. The irradiation of, in total, 120 s per sample was empirically tested to enable short measurement duration in combination with the lowest achievable standard deviation of concentration level.