**5. Discussion**

The presented results and images of SEM-MLA and Raman measurements of magnetite and hematite-bearing ores from apatite iron oxide deposits in the Kiruna area depend, in both cases, mainly on the correct classification of the present minerals. For SEM-MLA, a comprehensive mineral database containing all present minerals that are correctly distinguished by their chemical information is inevitable. Raman imaging does not necessarily require any preexisting knowledge about the existing minerals. Therefore, online data processing is not necessary and data analysis happens solely o ffline after data acquisition, e.g., using unsupervised multivariate data analysis to identify di fferent minerals.

When it comes to mineral classification, Raman shows some major advantages, especially in the discrimination of magnetite and hematite due to their distinct Raman spectral signatures [25–30]. A diagnostic Raman peak for magnetite is 663 cm<sup>−</sup><sup>1</sup> (T2g), whereas hematite is dominated by strong peaks at 293 (Eg), 410 (Eg), 615 (Eg) and 1320 cm<sup>−</sup><sup>1</sup> (Figure 4). In comparison, Figueroa et al. [13] showed that using the SEM-based MLA method discrimination of magnetite and hematite is possible but needs significant adjustments prior to and during measurements. It was stated, that measurement times did not increase significantly but the process of time-consuming o ffline modification and the effect of fluctuations in beam current was neglected. The ability of Raman spectroscopy to intrinsically discriminate clearly between magnetite and hematite shows grea<sup>t</sup> potential for the characterization of iron oxide deposits. However, it needs to be noted that Fe(II)-containing minerals can be easily converted into Fe(III)-oxides like hematite by high laser powers [25,29,30]. In order to avoid conversion, it is crucial to optimize spectral acquisition parameters prior to mapping (see experimental details).

**Figure 4.** Characteristic Raman spectra of magnetite, hematite, and Ti-bearing hematite. Note that the peak width and intensity for both hematites vary according to elemental changes.

Another grea<sup>t</sup> advantage of Raman spectroscopy is its ability to not only identify, but also quantify incorporated foreign elements. Substitution of iron with aluminum [31–33], manganese/chromium [34,35], or titanium [36] results in a shift of Raman peak positions due to di fferences in mass. In a similar manner, carbonates show changes in peak location and shape depending on the amount of magnesium, iron, and manganese replacing calcium [37–40]. Ti-bearing hematite was classified by both methods because of the incorporation of ca. 3 wt % Ti. In Raman, we observed a change in the peak intensity at 295 cm<sup>−</sup><sup>1</sup> and 660 cm<sup>−</sup><sup>1</sup> (LO Eu) and an increased peak width, especially at 660 cm<sup>−</sup><sup>1</sup> and 1320 cm<sup>−</sup>1. A shift to higher wavenumbers, as described by Varshney et al. [36] was not observed most likely due to lack of spectral resolution. The IR-active *Eu* (LO) at 660 cm<sup>−</sup><sup>1</sup> is theoretically not allowed (Raman-inactive). The intensity increase at about 660 cm<sup>−</sup><sup>1</sup> is a common indication for the incorporation of foreign elements. It becomes Raman-active due to structural disorder induced by surface defects or stress [34–36].

Please note that Raman imaging can be compromised by fluorescence excited by the laser source. The use of di fferent laser wavelengths, bleaching, or fluorescence correction algorithms can e fficiently counter the possible fluorescence of geological samples. If necessary, fluorescent areas can also be visualized using the Raman system.

Raman spectroscopy o ffers a variety of additional possibilities to characterize inorganic materials in terms of polymorphism, crystal orientation, crystallinity, phase, stress, and strain [41,42]. The combination of the strong and distinct Raman signatures with its nondestructive nature makes Raman spectroscopy a powerful tool for fine-scale identification and characterization of iron oxides. A3.

summary of the pros and cons of SEM-MLA and Raman imaging is given in Table


**Table 3.** Pros and cons of MLA and Raman imaging for characterization of iron ores.

Quantification of the mineral abundance is well established using the MLA software (FEI, Brisbane, Australia) [12]. For Raman imaging, quantification using a similar algorithm as described by Fandrich et al. [12] and FEI [21], needs to be implemented for geological applications. For this reason, we decided to compare area % in Table 2, which are easily accessible by both methods. For mineral quantification of the Raman map, we relied solely on the results of linear combination of reference spectra, not taking into account any microscopic data. The results also rely on color thresholds set in the SpectralImaging software. Therefore, it should be noted that the error of the Raman quantification method described here is larger than that of MLA quantification. Nevertheless, both methods reveal similar quantitative analyses of mineral abundance in the range of 1.5% with the exception of Ti-bearing hematite. The discrepancy of >3% is assumed to be caused by the fact that Raman relies on a fixed mapping grid neglecting mineral boundaries visible in the BSE image. Therefore, in some areas the step size of 30 μm is too big to resolve fine structures smaller than 30 μm size. Furthermore, the increased Raman intensity at 660 cm<sup>−</sup><sup>1</sup> for Ti-bearing hematite could, at least partially, result in a misassignment of magnetite using the linear combination approach.

Another important criterion for institutions and companies working on iron ores is the cost of analysis when making feasible assumptions according to improved characterization and extraction of the ore mineral. In purchase and maintenance, the financial advantage is on the Raman side, as the

acquisition of an SEM-MLA (~800.000€) is roughly three times higher than a Raman imaging device including software (~250.000€). Furthermore, operating an SEM requires high vacuum pumps and nitrogen supply. Both techniques require a skilled operator to efficiently adjust measurements and they should be operated in an air-conditioned laboratory. In this study, the analyses were conducted on prepared thin sections for both methods. However, the advantage of Raman on the cost side is further supported by the fact that no sample preparation is needed, whereas SEM-MLA requires preparation and carbon coating of the samples.

Taking into account the limits and potentials of both methods, it is essential to define the right questions to a problem in order to find the most suitable analytical solution. Samples that contain very fine-grained aggregates or minerals that have fine intergrowths may need to be resolved in higher resolution. These detailed maps with a spot size in nm range (up to around half the laser wavelength) are one of the major advantages of the Raman technique, whereas the SEM is limited to μm range (~1 μm) even with the best operating conditions. It should be noted that, with higher resolution and smaller spot, size measurements become more time-consuming. In this study, Raman measurements on the same sample area with the same spot size took at least 10 times longer than MLA mapping, thus they are not nearly as time-efficient. Although Ramanaidou and Wells [43], suggested Raman spectroscopy as a potential method for large volume or bulk analysis, this seems not applicable with stationary Raman spectroscopes at this stage. Transportable Raman devices may offer fast field analysis by point measurements, but the loss of spectral information due to the downsized equipment needs to be compensated. Future instrumental research and development may lead to Raman spectroscopy and imaging as the first order application in process mineralogical analysis of iron ore. However, solely imaging will not solve the need for precise characterization of these complex ores, thus parameters, such as mineral liberation, association or locking, especially from processed ore, need to be extracted. Raman manufacturers start to offer software solutions for particle analysis, e.g., ParticleScout (WITec GmbH, Ulm, Germany), applicable to large sample areas typical for geosciences. The authors of this study are currently working on the development of these functions too, to further enhance the application of Raman imaging as a modern tool for analysis, especially for iron ore.
