Correlative Raman microscopy, SEM and EDS – The combined evaluation of a whole sample mapping of a Chelyabinsk meteorite fragment

The correlation of different microscopic techniques has seen increased interest in recent yearsdue to the possibility of combining the strengths of multiple techniques. In addition to the practicalchallenges with regard to sample preparation, instruments design and the need for operatorsexperienced in multiple techniques, unique data treatment challenges arise when combining datasets with different resolutions and contrast mechanism. Two key questions arise. How can a SEMimage with a pixel resolution of 30 nm, an EDS mapping with a pixel resolution of 100 nm and aRaman mapping with a pixel resolution of 1 μm that are distorted against each other (differentcontrast mechanism) be combined into a single map? How can we evaluate the resulting map thatconsist of Raman bands, EDS-elemental concentrations and SEM contrast values? We want toaddress these questions (on the example of a whole sample Raman-SEM-EDS-mapping of theChelyabinsk meteorite), but please note that these approaches generalize to other combinations.The basic approach of our workflow is shown in figure 1. Starting from the separated mappingdata, we start with an evaluation of the separated data as far as possible. This serves the purposeof both reducing the amount of data as well as preserving the advantages of establishedevaluation routines as far as possible. The second step is to correlate the mappings based oncommon features visible in all mappings and interpolating everything to the resolution of the mosthigh-resolution technique (BSE in this case). This way a combined “super-spectral-map” isgenerated that contains all the relevant analytical information of the separated mappings. In thefinal step a variety of option are available to evaluate the combined data. In this case we opted fora random forest algorithm for the classification of the phases of the meteorite fragment (figure 2a).Note that none of the initial techniques (SEM, EDS, Raman) is capable of differentiating all of thephases on its own, which is one of the main benefits of correlative microscopy. In a secondevaluation step we used the combination of the random forest classification and theat-%-quantification from the EDS for a further analysis of the phase. The example of the pyroxenephase, which divides into two compositional clusters is shown in figure 2 (b,c).To sum up we aim to provide a workflow for correlating and combining large datasets of variousmicroscopic techniques, whilst also pointing out some of the best options for evaluating thosecombined mappings, using the fascinating example of a meteorite fragment.