New Candidate Ultralow-Velocity Zone Locations from Highly Anomalous SPdKS Waveforms

Abstract

Ultralow-velocity zones (ULVZs) at the core–mantle boundary (CMB) represent some of the most preternatural features in Earth’s mantle. These zones most likely contain partial melt, extremely high iron content ferropericlase, or combinations of both. We analyzed a new collection of 58,155 carefully processed and quality-controlled broadband recordings of the seismic phase SPdKS in the epicentral distance range from 106° to 115°. These data sample 56.9% of the CMB by surface area. From these recordings we searched for the most anomalous seismic waveforms that are indicative of ULVZ presence. We used a Bayesian approach to identify the regions of the CMB that have the highest probability of containing ULVZs, thereby identifying sixteen regions of interest. Of these regions, we corroborate well-known ULVZ existence beneath the South China Sea, southwest Pacific, the Samoa hotspot, the southwestern US/northern Mexico, and Iceland. We find good evidence for new ULVZs beneath North Africa, East Asia, and north of Papua New Guinea. We provide further evidence for ULVZs in regions where some evidence has been hinted at before beneath the Philippine Sea, the Pacific Northwest, and the Amazon Basin. Additional evidence is shown for potential ULVZs at the base of the Caroline, San Felix and Galapagos hotspots.

The Most Parsimonious Ultralow‐Velocity Zone Distribution From Highly Anomalous SPdKS Waveforms

“…Waveforms identified as anomalous all show two arrivals, with the second arrival amplitude approximately equal to or greater than the first arrival. Non-anomalous waveforms in the azimuthal range from 0° to 180° are similar in character to the waveforms labeled as highly anomalous, but do not get labeled as such due to the strict rules imposed in Thorne et al (2020). For example, records at distances approximately less than 110° (Figure 2c) do not have a distinct second arrival, although a second arrival is clearly emerging.…”

“…These synthetics are computed for a ULVZ with identical properties located on either the source-side ( Figure 3) or receiver-side ( Figure 4) of the path. Using the classification scheme as defined in Thorne et al (2020), highly anomalous recordings are only found for this ULVZ model in the epicentral distance range from 111° to 113°.…”

“…In this test, we allowed all arrivals with amplitudes greater than 0.75 to be considered for classification as highly anomalous. Thorne et al (2020) had originally set this number to 0.8. This search yielded 2,425 highly anomalous records (relative to 2,222 previously found).…”

“…We do this by first taking the observations of highly anomalous SPdKS waveforms published by Thorne et al. (2020) and calculating the most parsimonious distribution of CMB heterogeneities that could explain these waveforms. This distribution is determined by finding the smallest number of heterogeneities that can be used to explain the observed anomalous waveforms within a predetermined level of misfit.…”

Automatic Slowness Vector Measurements of Seismic Arrivals with Uncertainty Estimates using Bootstrap Sampling, Array Methods and Unsupervised Learning

“…Machine learning methodologies are becoming more prevalent in the geosciences (for a review see: Bower et al, 2013) and seis-mology (for a review see: Kong et al, 2019) with methods used to automate data selection (e.g. Valentine & Woodhouse, 2010;Thorne et al, 2020) and extracting properties from data by mapping seismograms to lower dimensional space using autoencoders (Valentine & Trampert, 2012) or sequence seismograms and identify features such as the precense of seismic scatterers (Kim et al, 2020). Here we use an unsupervised learning algorithm as part of our automation technique.…”