How to use a MLA correctly

SEM-based image analysis (aka Automated Mineralogy) has become a standard method not only in mineral processing but also for petrological studies, e.g., for the determination of modal mineralogy, search for rare phases, or for fabric parameters such as grain size and mineral association. A very small aliquot (the surface of a polished epoxy block prepared from a few grams of ground material or the surface of a polished thin section) are used during analysis. All caution is needed to assure that this small aliquot is representative of the sample material to be examined – and that it yields the information of particular interest, e.g., the abundance and microfabric relations of rare phases such as PGEs or Au.
SEM-based image analysis is performed on surfaces, i.e. observations are obtained on two-dimensional surfaces, whereas the relevant attributes all related to the properties of three dimensional mineral grains/particles/rock volumes. For grain mounts, the appearance of a grain on the analyzed surface is controlled by the behavior of the grain during sample preparation and depends on grain size, shape and density of phases. Cutting effects usually cause an underestimation of grain size for isotropic grains, whereas phases forming grains with a shape preferred orientation (e.g. mica) may be over- or underestimated depending on the plane chosen for analysis. This causes also bias in modal mineralogy, grain size and shape or mineral association and liberation. At best, measurements can be precise but not accurate with reference to the entire sample.
To evaluate these effects, a systematic study was performed on grain mounts of a mineralogically rather simple greisen from the Erzgebirge (Germany), mainly constituted by zinnwaldite mica, topaz and quartz, i.e minerals with different density, mineral grain size and shape. Eight values of the ratio solid load/resin and analyses of two cutting planes (xy and xz) were performed, having 16 MLA measurements. Apparent modal mineralogy compositions were statistically compared, using bootstrapping techniques to estimate their uncertainty. Results showed a significant and constant difference of the topaz proportion between xy and xz measurements, as well as a significant systematic reduction in the ratio zinnwaldite/quartz with increasing solid load/resin ratio on xy measurements, a bias absent from xz measurements. In conclusion, sample preparation biases can be reduced by measuring on transverse planes (xz) or reducing the amount of resin added to the sample. For a first time the error in analyzing modal mineralogy was quantified. Other biases remain, like e.g. stereological distortion in estimating 3D quantities from 2D measurements. Such challenges are left for further research.