Abstract : A three-component waveform is modeled as a sum of trigonometric functions using Dynamic Linear Modeling (DLM), a statistical extension of Kalman filtering. The DLM method converts a waveform into a multivariate time series of amplitude and phase-angle estimates. This time series is then transformed into a multivariate time series of descriptive features such as magnitude, direction-of-travel and dimensionality which are more closely related to the underlying seismic phases. There are many possible transformations, and several are considered here. This paper also explores the expected behavior of the dimensionality, magnitude and direction-of-travel time series over the duration of a phase. Dynamic Linear Modeling and parameter transformation is illustrated with an artificial waveform having compressional and Rayleigh phases.
[1]
T. Takanami.
A Study of Detection and Extraction Methods for Microearthquake Waves by Autoregressive Models
,
1991
.
[2]
Kevin K. Anderson,et al.
Dynamic Bayesian filtering for real-time seismic analyses
,
1994
.
[3]
Nozer D. Singpurwalla,et al.
Understanding the Kalman Filter
,
1983
.
[4]
M. West,et al.
Bayesian forecasting and dynamic models
,
1989
.
[5]
A. Jurkevics.
Polarization analysis of three-component array data
,
1988
.
[7]
John E. Vidale,et al.
Complex polarization analysis of particle motion
,
1986
.