Seismic Event Parameter Inference from Sparse Early Warning Stations Using Bayesian Networks

Session: Earthquake Early Warning: Current Status and Latest Innovations
Type: Oral
Date: 4/29/2020
Time: 09:30 AM
Room: 115
Description: 

Inference of the seismic event parameters is one of the primary goals of the earthquake early warning systems. We calculate magnitude and epicentral distance from the measurements on a sparse regional seismic network by means of Bayesian network analysis, which allows us to incorporate the uncertainties in the parametric estimates.

The P wave is identified in the real-time triaxial seismic records via variance maximization and covariance matrix eigenvalue ratio method. Afterwards, we extract a set of features associated with the body-wave and the duration-based magnitudes. We use robust methods for instantaneous frequency estimation of the seismic signal to identify the dominant period corresponding to the peak amplitude of compressional wave, and perform least-squares analysis of the envelope function to extract body wave coda parameters. As well, station-correction filters are applied to remove the influence of local effects.

We perform diagnostic inference using Bayesian network analysis to calculate distributions of the event parameters, while our numerical approach employs rejection sampling Markov Chain Monte Carlo method. Since functional relationships between features and event parameters are non-linear, the resulting joint distributions of magnitude and epicentral distance are strongly non-Gaussian.

The efficiency of the algorithm is illustrated using a set of regional earthquakes recorded by our sparse seismic network. We emphasize the need for the station correction by analyzing earthquake records obtained in dissimilar conditions.

Presenting Author: Anton G. Zaicenco

Authors