Application of mid-infrared (MIR) spectroscopy to identify and quantify minerals in New Zealand soils

The analysis and characterization of soil minerals rely on the availability and capability of experts, therefore greatly influencing the efficiency and accuracy of mineral analysis. Mid-infrared (MIR) spectroscopy offers a low-cost, high-throughput alternative, as it instantly records the spectral signatures of soil minerals through the absorption of infrared radiation. This study investigates the potential of MIR spectroscopy to identify and quantify the most common minerals in the clay (<2 jim) fraction of New Zealand soils. We analysed 3,097 samples and developed partial least squares regression (PLSR) models to quantify the concentration for each of the 11 mineral types in the collated database. In addition, we developed a partial least squares discriminant analysis (PLS-DA) model to identify the dominant mineral of the analysed soil samples. Most models used for quantifying mineral compositions indicate high accuracy. The highest values of R2 (0.81), Lin's concordance correlation coefficient (0.90), RPIQ (4), and the lowest bias (-0.26 %) were obtained for the quantification of mica on the test set. For the identification of the dominant mineral, the overall accuracy was 75 % on the test set. In addition, we could correctly classify 94 % of the samples dominated by mica, 86 % of the samples by volcanic glass/amorphous silica, 82 % of the samples by allophane/imogolite, and 77 % of samples dominated by kaolin-smectite. Thus, MIR spectroscopy offers a valuable solution for mineral quantification and identification, especially in areas where quantified mineral databases are not readily available and human expertise is lacking.