Gleaning Insights from Sequencing Geophysical Timeseries

Session: Applications and Technologies in Large-Scale Seismic Analysis [Poster]
Type: Poster
Date: 4/28/2020
Time: 08:00 AM
Room: Ballroom
Description: 

Gleaning insights into earth structures and processes from massive geophysical datasets requires approaches flexible enough to detect robust patterns with little-to-no user supervision within and across different types of measurements. Often, observed phenomena are driven by a leading effect or parameter. In such a case, there should exist an order in which the data should be optimally viewed. This order reflects a one-dimensional manifold representing the underlying trend, even in the presence of complex behavior. Here, we apply a graph-based manifold learning algorithm called the Sequencer to reveal this leading trend from complex data, which is especially useful in the absence of theoretical guidance. We use the Sequencer to identify trends and anomalous signals across a diverse set of geophysical time-series data. Specifically, we analyze receiver functions and seismic waveforms to constrain structures in the crust and deep mantle and geodetic timeseries to map out surface processes. We discuss the utility of sequencing and its performance compared to more commonly-used approaches such as tSNE and k-means clustering.

Presenting Author: Vedran Lekic

Authors