Moment Tensors for Very Long Period Signals at Etna Volcano, Italy

Margaret Hellweg, Andrea Cannata (Universita Catania), Stefano Gresta (Universita Catania), Sean Ford, Guiseppi Di Grazia (INGV Sezione di Catania)


Very long period signals (VLP, 10 s - 30 s) associated with long period events (0.5 Hz - 5 Hz) were observed at Etna Volcano, Italy, during June-November 2005 (Figure 2.51). They are only recorded at the broadband stations nearest to Etna's craters, ECPN, EBEL, EPDN and EPLC. These stations are part of the permanent seismic network run by the Catania Section of the Istituto Nazionale di Geofisica e Vulcanologia (INGV). Although the signal-to-noise (S/N) ratio for these VLPs is in general only poor to fair, they seem to recur, and can be classified into two families. We attempt to improve the S/N by stacking and to determine moment tensors for the VLP events.

Figure 2.51: Example of long period (LP) associated very long period (VLP) events of Families I (top) and II (bottom). In each case, the upper rows show the broadband seismogram of the LP, while in the lower rows, the VLP signals are extracted by causal filtering.
\epsfig{file=hell08_1_1.eps, width=8cm}

Clustering and Stacking

The VLP events were clustered following Green and Neuberg (2006). For all events, (i) the vertical component of ECPN station (highest S/N) was lowpass filtered at 0.15 Hz to exclude the ocean microseisms; (ii) the correlation matrix was calculated for all events using 30 s of data; (iii) the cross correlation threshold was chosen to be 0.95 so that fairly dissimilar events would be classified as a single family, while members of one family would not be grouped with another; (iv) a master event was selected (the event with the most correlation values above the threshold); (v) the average family waveform was created as the stack of all events correlated with the master event; (vi) the stack waveform was then cross-correlated with the original seismic records, and all events with greater correlation than the threshold were grouped into a waveform family; (vii) the steps iv-vi were repeated until all events were classified into distinct groups; (viii) finally, a new correlation matrix was constructed with events sorted into families. In this procedure, no overlap was allowed between clusters; in fact, once an event was assigned to a group, it was removed from further correlation. We found that 85% of the events could be grouped into two main families of VLP events. Families I and II had 194 and 87 members, respectively. For further analysis, we used stacked seismograms for the two VLP families from the four summit station. The stacks were created using signals filtered with a 4-pole, causal Butterworth bandpass filter (0.033 Hz - 0.167 Hz). Using waveforms from all stations from two members of the family (12 waveforms for each event), we tried different time lags between the two sets of signals. For each time lag value, we evaluated the similarity between the events by averaging the cross correlation coefficients calculated for each pair of corresponding waveforms (for example ``ECPNz of event 1'' and ``ECPNz of event 2''). The time lag was chosen which gave the maximum average cross correlation coefficient and the signals for each component at each station were stacked. New events were compared with and then added to the stack event. Thus, all the events contributed to the stacked signals representing the two families (Figure 2.52). Although the stacking was performed on the Z, N and E components, we show the traces for Z, R and T with the optimum rotation determined using polarization analysis (Plesinger et al., 1986). The S/N is clearly improved in the stacks, where several important characteristics are apparent. (1) For Family I , the first motion on the Z component is positive at all stations. (2) For Family II, the first motion on the Z component is negative at all stations. (3) For both families, the amplitudes on all components are considerably larger at ECPN and EPDN, the stations to the SW and NE of the active craters, than at EBEL and EPLC, which lie to the SE and NW.

Figure 2.52: Stacked very long period (VLP) events of Family I (top) and Family II (bottom). The upper three traces show superimposed Z, R and T seismograms for all members at the 4 stations. Best azimuths were determined using polarization analysis. The lower traces show the stack seismograms.
\epsfig{file=hell08_1_2.eps, width=7cm}

Waveform Polarization

Note that there is little energy on the T-component at any of the stations (Figures 2.52 and [*]). In addition, the vertical and radial components are ``in phase'', with both Z and R being negative or positive at the same time. This is generally an indication for P-waves. The directions of particle motion of Family I and Family II stacks are very similar, differing at each station by only a few degrees.

Figure: From Chapter 2, Section 23: Particle motion for very long period (VLP) events of Family I (top) and Family II (bottom). The map view shows particle motion in the horizontal plane; the two cross sections (to the right of and below the maps) show particle motion across the crater from ECPN to EDPN and from EPLC to EBEL.
\begin{figure*}\centering\epsfig{file=hell08_1_3.eps, width=6.5cm}\end{figure*}

In the horizontal plane, the motion at ECPN and EPDN is polarized more or less toward the active craters. The motion at EPLC and EBEL is comparatively small and has both an element pointing toward the crater and a later segment of the motion oriented transversely. However, the S/N ratio at these two stations is poor.

In the vertical plane, both ECPN and EPDN (SW and NE of the craters, respectively) both point slightly downward toward the crater. On the plots of vertical motion, the particle motion diagrams are shown at the altitudes of the stations. Again, the motion for EBEL and EPLC are small, with a hint of ``downward toward the crater''.


The particle motion for these VLP signals is highly suggestive of either an initially opening (Family I) or initially closing (Family II) crack of fairly limited extent (little to no signal at EBEL and EPLC). There is a hint that the source is finite, as the directions of the horizontal particle motion change at ECPN and EPDN. We have calculated Greens functions for full moment tensors using a simple half space velocity model. Initial inversions using Minson and Dreger (2008) indicate that a volume change explains a large portion of the waveforms. We will perform further analysis using a grid of sources to determine both the best source mechanism and its location.


Green, D., and J. Neuberg, Waveform classification of volcanic low-frequency earthquake swarms and its implication at SoufriŠre Hills Volcano, Monserrat, J. Volcanol. Geotherm. Res., 153, 51-63, 2006. Minson, S. and D. Dreger, Stable Inversions for Complete Moment Tensors, In press Geophys. Journ. Int., 2008. Plesinger, A., M. Hellweg and D. Seidl, Interactive high-resolution polarization analysis of broad-band seismograms. J. Geophysics, 59, 129-139, 1986.

Berkeley Seismological Laboratory
215 McCone Hall, UC Berkeley, Berkeley, CA 94720-4760
Questions or comments? Send e-mail:
© 2007, The Regents of the University of California