Dynamic Bayesian filtering for real-time seismic analyses

Abstract

State space modeling, which includes techniques such as the Kalman filter, has been used to analyze many non-stationary time series. The ability of these dynamic models to adapt and track changes in the underlying process makes them attractive for application to the real-time analysis of three-component seismic waveforms. The authors are investigating the application of state space models formulated as Bayesian time series models to phase detection, polarization, and spectrogram estimation of seismograms. This approach removes the need to specify data windows in the time series for time averaging estimation (e.g., spectrum estimation). They are using this model to isolate particular seismic phases based on polarization parameters that are determined at a spectrum of frequencies. They plan to use polarization parameters, frequency spectra, and magnitudes to discriminate between different types of seismic sources. They present the application of this technique to artificial time series and to several real seismic events including the Non-Proliferation Experiment (NPE) two nuclear tests and three earthquakes from the Nevada Test site, as recorded on several regional broadband seismic stations. A preliminary result of this analysis indicates that earthquakes and explosions can potentially be discriminated on the bass of the polarization characteristics of scattered seismic phases. However, the chemical (NPE) and nuclear explosions appear to have very similar polarization characteristics.