Classification of minerals and the assessment of lithium and beryllium content in granitoid rocks by laser-induced breakdown spectroscopy
This study demonstrates that LIBS mapping and spatially resolved local analysis is an efficient and practical approach for the classification of mineral grains (quartz, feldspar, biotite, amphibole) and for prospecting of technologically relevant, low-Z elements (e.g. Be and Li) in granitoid rock samples. We tested three statistical approaches (classification tree (CT) based on indicator elements, linear discriminant analysis (LDA) and random forest (RF)) for the classification of the mineral grains and found that each of the three methods provides fairly similar, very good classification accuracies. RF and LDA provided better than 92% accuracy for all minerals, whereas CT showed a somewhat poorer (around 80%) accuracy for quartz in particular. Our results also demonstrate that using multiple analytical locations within each grain and resting the classification on the majority vote of these individual analysis gives more reliable discrimination (grain-based accuracy is better than location-based accuracy). We also demonstrated that LIBS elemental mapping can provide valuable information about the distribution of chemical elements among the minerals, especially if it is combined with matrix-matched calibration of emission intensity data. We illustrated this by the successful assessment of ng to μg amounts of Be and Li in the studied mineral grains. Our results suggest that mining for Be and Li in granitoid rocks should be aiming for biotite and amphibole grains.