ASSESSING TEMPORAL VARIABILITY OF COASTAL DEFORMATION DURING MEGATHRUST EARTHQUAKES AT THE CASCADIA SUBDUCTION ZONE USING A DIATOM BAYESIAN TRANSFER FUNCTION
We developed a new, diatom-based Bayesian transfer function (BTF) to quantitatively reconstruct coseismic subsidence. The modern diatom dataset consists of 366 species from 94 samples collected from the subtidal to forested upland environments of Willapa Bay, WA, USA. To improve the dissimilarity between modern and fossil samples, we used a hierarchical clustering method to identify diatom groups with similar abundance and distributions with elevation. The diatom BTF integrates the grouped species response curves (positive and negative linear, unimodal, bimodal and multimodal) and prior information (lithologic context) about each sample to improve the precision of reconstructed earthquake subsidence. A 10-fold cross-validation provides an assessment of the predictive performance of the diatom BTF showing a strong relationship between observed and predicted elevations. We apply the BTF to fossil diatom assemblages from a stratigraphic sequence at the Redtail locality of Willapa Bay where six, mud-over-peat contacts have been inferred to record subsidence during megathrust earthquakes over the past 3500 years. These new estimates will aid in constraining models of earthquake deformation as part of earthquake hazard assessments for the Cascadia subduction zone.