CHARACTERIZATION OF MARINE MUDSTONES: TOWARDS A STATISTICAL APPROACH TO INTERPRET XRF DATA

Portable X-ray fluorescence (pXRF) spectroscopy has led to an abundance of bulk rock elemental concentration data of marine mudstones. It is expected that statistical analysis of these datasets should reveal stratigraphic trends hidden to subjective observations based on elemental cross-plots and geochemical logs. This study applied several dimensionality reduction methods on pXRF data from drill cores of the Ordovician Utica Shale and Trenton Group to test this hypothesis. K-means clustering, principal component analysis (PCA), factor analysis (FA), and positive matrix factorization (PMF) where applied to a pXRF dataset of 2,838 samples collected at a 3-inch spacing from the 74NY-5 core of central New York State. The studied interval contains the Flat Creek and Indian Castle members of the Utica Shale and the Dolgeville Formation of the Trenton Group. K-means clustering places each sample into only one cluster allowing for indirect mineralogical interpretation using ternary diagrams. PCA and FA reduce dimensionality into factors with the most variance and output data in both positive and negative directions away from a center point. PMF can unmix each individual sample into a percent contribution in only positive space, allowing for more robust solutions and a level of interpretability surpassing that which is obtained through qualitative methods by revealing details regarding sediment provenance and depositional processes previously hidden to traditional techniques. Results from PMF analyses were compared to interpretations made using traditional sedimentology methods. They match the overall stratigraphy, but also reveal new trends. For example, the carbonate and siliciclastic lithologies can be speciated into four and three individual factors, respectively. The speciated factors reveal details about fluctuations in sediment provenance and differentiate between biogenic and detrital silica. XRF can only detect elements and cannot directly differentiate between biogenic and detrital grains. XRD analysis was used to address mineralogy complications due to pXRF detection limitations and SEM analysis was used to visually confirm detrital and biogenic silica.