MULTIVARIATE MODELING OF GLACIMARINE LITHOSTRATIGRAPHY USING SCANNING XRF AND MULTISENSORY CORE PROPERTIES VALIDATED WITH CT IMAGERY

Lithofacies description and analysis is an integral part of using sediment cores to interpret depositional environments. In this study, we compare core lithofacies identified from visual analysis and CT-scan imagery from IODP Exp. 341 Site U1419 to those modeled through multivariate hierarchical and modified mixture-model clustering (Templ et al., 2008; doi: 10.1016/j.apgeochem.2008.03.004; Ellefsen et al., 2014; doi: 10.1016/j.apgeochem.2014.10.011). Site U1419 contains fine-grained, glacigenic sediments, covering MIS 3 to the present (Gulick et al., 2015; doi: 10.1073/pnas.1512549112). Cluster model inputs are standard non-destructive datasets collected shipboard on most IODP cores (bulk density, magnetic susceptibility (MS), color, natural gamma-ray (NGR)) and post-cruise scanning XRF elemental data. Scanning XRF model data are restricted to elements sensitive to grain size (e.g., Al, Rb). Models were run with combinations of physical property and/or elemental inputs. Comparison between visual core descriptions and CT-scans is used to validate cluster model results. Six distinctive lithofacies were visually identified ranging from massive mud with and with out lonestones, to massive and stratified diamict. Lithofacies fall into two main groups: glacial (massive/stratified diamict, laminations and mud with lonestones); interglacial (mud and laminations with out lonestones). Model results show that both types of cluster methods, incorporating both physical property and elemental data, identify the change from glacial to interglacial lithofacies, with 99% success. However, hierarchical clustering provides a slightly better model fit. Glacial lithofacies are defined by higher MS, NGR and interglacial lithofacies are defined by higher b*(color), Rb, K. Cluster models using only physical property data capture main glacial-interglacial changes, but element-only models do not fit the observed lithofacies and are much more difficult to interpret. Scanning XRF elemental data alone are not as useful for lithofacies cluster models as physical property data, and hierarchical clustering provides better models of glacimarine lithofacies.