Development of an empirical model to predict mineral composition for sample mixtures of chlorite, goethite, hematite, and quartz in a calcite matrix using NUV/VIS/NIR spectra

The goal of this research was to construct a model to predict mineral composition from spectra utilizing wavelengths between 250 and 850 nm (NUV/VIS/NIR). Sample mixtures of five common minerals found in deep-sea North Atlantic sediments were made using mineral standards and constructed in both combinations and proportions which occur naturally. The sampling design was a complete 54 factorial comprised of chlorite, goethite, hematite, and quartz with calcite as the filler matrix material. Attempts to fit the data using a wide variety of statistical models met with limited success. The MARS algorithm, designed by Friedman (1991) as a curve-fitting technique, was used to develop the most successful model. Predictions for chlorite, hematite, and to a lesser degree, goethite, were moderately effective, whereas quartz predictions were less satisfactory. The relative effectiveness of the model for these minerals is apparently due to the fact that all three contain Fe cations; and iron has absorption bands in the part of the spectrum under investigation. It is well documented in the literature that silicon-oxygen bonds are virtually featureless over this interval but instead have absorption bands in the infrared region. Results from the literature suggest that utilizing a wider range of spectral values probably would have yielded a more effective model.