MOVING ORDOVICIAN TEPHROCHRONOLOGY INTO THE 21ST CENTURY WITH LASER ABLATION AND MACHINE-LEARNING (ML) TECHNIQUES

K-bentonites have been widely used for correlation of Ordovician rocks at the local, regional, and global scale. For accurate correlation schemes it is critically important to develop an objective tephrochronological approach to distinguish individual K-bentonites. In recent years single crystal geochemistry of apatite has been used to differentiate tephra layers.

We present data from existing and newly acquired analytical datasets (electron probe micro-analyzer [EPMA] data and laser ablation ICP-MS [LA-ICP-MS] data, respectively) of apatite from the Deicke and Millbrig K-bentonites at the Big Ridge exposure near Gadsden, AL to test the use of machine-learning (ML) techniques to identify with confidence individual tephra layers based on unique geochemical signatures for these two different volcanic events.

Our results show that the decision tree based on EPMA data uses the elemental concentration patterns of Mg, Mn, and Cl, consistent with previous studies that emphasized the utility of these elements for distinguishing Ordovician K-bentonites. However, using LA-ICP-MS data gives a much more robust decision tree for identifying individual layers within each K-bentonite as elements (and elemental ratios) as many elements are below the level of detection of the EPMA (e.g., Ba, V, As). Overall, the ML model identified individual layers of multiphase eruptions and was able to distinguish K-bentonite layers, thus illustrating very well the great potential of applying ML techniques to tephrochronology.