HR: 08:55h
AN: S31C-04    [Abstracts]
TI: Discrete Wavelet Packet Transforms and volcanic tremor: method and application to Erta 'Ale, Ethiopia
AU: * Jones, J P
EM: josh@ess.washington.edu
AF: Dept. of Earth and Space Sciences, Box 351310 University of Washington, Seattle, WA 98195-1310 United States
AU: Carniel, R
EM: roberto.carniel@uniud.it
AF: Dipartimento di Georisorse e Territorio, via Cotonificio, 114 Universit… di Udine, Udine, 33100 Italy
AU: Malone, S
EM: steve@ess.washington.edu
AF: Dept. of Earth and Space Sciences, Box 351310 University of Washington, Seattle, WA 98195-1310 United States
AB: The time-varying properties of volcanic tremor demand advanced techniques capable of analyzing changes in both time and frequency domains. Specifically, rapid data preprocessing techniques with the ability to distinguish signal from noise are especially valuable in analyzing the temporal, spatial, and spectral properties of these signals. To this end, we use the Discrete Wavelet Packet Transform and the Best Shift Basis algorithm to select an orthonormal basis for continuous volcanic tremor data, then apply a simple statistical test to eliminate frequency bands that primarily consist of Gaussian white noise. We then use the Maximal Overlap Discrete Wavelet Packet Transform to compute and analyze features in the detail coefficients of each "signal" band. Because MODWPT detail coefficients are equivalent to a time series convolved with a zero phase filter, we apply standard polarization and amplitude-based location techniques to each frequency band's detail coefficients to analyze possible source locations and mechanisms. To demonstrate the usefulness of these techniques, we present a sample analysis of data from Erta 'Ale volcano, Ethiopia, recorded on a temporary network in November 2003. Data were sampled at 100 Hz and the DWPT was computed with the LA(16) wavelet to a maximum level of j = 7. The optimal basis for this data set consists of 54 frequency bands, but only 9 contain meaningful "signal" energy. We identify two frequency bands whose locations suggest a distributed source; three frequency bands whose signals may come from the lava lake itself; three high-frequency bands of scattered energy; and one very high frequency band of non-Gaussian instrument noise. Finally, we discuss optimization efforts, computational efficiency, and the feasibility of using similar wavelet methods to preprocess data in real time or near real time.
DE: 7280 Volcano seismology (8419)
DE: 7299 General or miscellaneous
DE: 8419 Volcano monitoring (7280)
DE: 8494 Instruments and techniques
DE: 8499 General or miscellaneous
SC: Seismology [S]
MN: Fall Meeting 2005