Unsupervised Learning of Seismic Wavefield Features: Clustering Continuous Array Seismic Data During the 2009 L'Aquila Earthquake

Abstract

AbstractWe apply unsupervised machine learning to 3 years of continuous seismic data to unravel the evolution of seismic wavefield properties in the period of the 2009 L'Aquila earthquake. To obtain sensible representations of the wavefield properties variations, we extract wavefield features (i.e., entropy, coherency, eigenvalue variance, and first eigenvalue) from the covariance matrix analysis of the continuous wavefield data. The defined wavefield features are insensitive to site‐dependent local noise, and inform the spatiotemporal properties of seismic waves generated by sources inside the array. We perform a sensitivity analysis of these wavefield features, and track the evolution of source properties from the unsupervised learning of the uncorrelated features. By clustering the wavefield features, our unsupervised analysis avoids explicit physical modeling (e.g., no requirement for event location and magnitude estimation) and can naturally separate peculiar patterns solely from continuous seismic data. Our model‐free unsupervised learning of wavefield features reveals distinct clusters well correlated with different periods of the seismic cycle, which are consistent with previous model‐dependent studies.