Detecting seismic activity with a covariance matrix analysis of data recorded on seismic arrays

Abstract

Modern seismic networks are recording the ground motion continuously at the Earth's surface, providing dense spatial samples of the seismic wavefield. The aim of our study is to analyse these records with statistical array-based approaches to identify coherent time-series as a function of time and frequency. Using ideas mainly brought from the random matrix theory, we analyse the spatial coherence of the seismic wavefield from the width of the covariance matrix eigenvalue distribution. We propose a robust detection method that could be used for the analysis of weak and emergent signals embedded in background noise, such as the volcanic or tectonic tremors and local microseismicity, without any prior knowledge about the studied wavefields. We apply our algorithm to the records of the seismic monitoring network of the Piton de la Fournaise volcano located at La Réunion Island and composed of 21 receivers with an aperture of ~15 km. This array recorded many teleseismic earthquakes as well as seismovolcanic events during the year 2010. We show that the analysis of the wavefield at frequencies smaller than ~0.1 Hz results in detection of the majority of teleseismic events from the Global Centroid Moment Tensor database. The seismic activity related to the Piton de la Fournaise volcano is well detected at frequencies above 1 Hz.