Phase curves may be constructed in either of the ways documented in Chapter 4. To ensure that the code written by the author was, in fact, performing as it should, an impulsive test signal with predetermined dispersion was generated and the code was able to successfully reproduce the phase velocity dispersion curve. An example of the resultant plot from one good window of station pair AKLV-AKUT at 0.7576 Hz as calculated by the cross correla-
−1000 −800 −600 −400 −200 0 200 400 600 800 1000 −1.5 −1 −0.5 0 0.5 1 1.5 2x 10
8 xcorr for first good window, Z channels of 1 and 2start time 18770972.52
Lags in Number of Samples
Amplitude(Not Normalized)
Figure 5.6: Cross correlation for the two northernmost stations, AKGG and AKLV.
tion method discussed in Chapter 4 is shown in Figure 5.7. The inversion of resultant phase velocity curves may be performed in the manner favored by the user.
To ensure that the results shown in Figure 5.7 were repeatable, a second data set consist- ing of another seismically quiet period of time was tested using the code. For the twenty minute period beginning at midnight on July 6, 2008, using stations AKRB and AKLV, dispersion curves were extracted for shared good windows of time. As with the station pair AKUT-AKLV, it was observed that some of the dispersion curves were not physically realistic, whereas others were reasonable. A plot of a reasonable dispersion curve for the second set of data is shown in Figure 5.9. Some of the irregularity in the unwrapped phase is captured in the very low frequencies of the curve, but then the plot is more regular and there is the suggestion of a higher mode at higher frequencies. It is worth noting that dis- persion curves from any set of real data are generally irregular at the lowest frequencies, but to test the robustness of the results a physically realistic forward model was also devel- oped to compare to the extracted results. Values of layer thickness and shear velocity were
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 500 1000 1500 2000 2500 3000 3500
real data phase velocity dispersion curve
frequency (Hz)
velocity (m/s)
Figure 5.7: Phase velocity dispersion curve from the first of the shared good windows in the time series from station AKUT and AKLV. The filter used for the PCA analysis had a center frequency of 0.7576 Hz and produced several good windows. Note that there is strong dispersion. At frequencies below about 0.2 Hz, there is little signal, which explains the steep dropoff to unreasonable slow velocities.
chosen based on the published values of compressional velocity from the Alaska Volcano
Observatory (Dixonet al., 2002). The plot of the forward model was for the fundamental
mode only, but produced a shape quite similar to the curves extracted for both data sets. Despite the existence of sensible phase velocity dispersion curves for several “good” windows produced by measuring the angle of energy arriving at 0.7576 Hz, it was apparent that there was a need for additional selection criteria because there also existed several windows at this frequency with nonsensical dispersion curves. An example of an unrealistic dispersion curve is seen in Figure 6.1. A discussion of ideas for improvements to the selection criteria is provided in Chapter 6, Section 6.5.
0 500 1000 1500 2000 2500 3000 3500 4000 0 0.2 0.4 0.6 0.8 1 Velocity (m/s) Frequency (Hz) Fundamental Mode Dispersion
Figure 5.8: Forward model produced for the case of a layer over a half-space. Top layer shear velocity was set to 500 m/s (appropriate for unsaturated to saturated sandy material) with a thickness of 600 m. The half-space velocity was set to 4000 m/s to represent bedrock.
0 500 1000 1500 2000 2500 3000 3500 4000 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Velocity (m/s) Frequency (Hz)
Julian Day 188, 2008 - Phase Velocity Dispersion Curve
Figure 5.9: Phase velocity dispersion curve from shared good window for station pair AKRB-AKLV on Julian Day 188 (July 6), 2008 from 12:00 am to 12:20 am. Note possible arrival of a second mode above 0.6 Hz.
Chapter 6
DISCUSSION
6.1
Introduction
The current body of research on the analysis of ambient seismic surface waves shows little overlap between the works of scientists who are interested in characterizing the source of the noise (see, for example Tanimoto (2007), Webb (2008), or Yang and Ritzwoller (2008)) and those who use the noise for inversion of velocity dispersion curves for imaging the
subsurface (references include Draganov et al. (2006), Nakata et al. (2011), and works
cited in previous chapters). In the field of source parameterization, there generally are not attempts to use the findings about the source characteristics to inform inversion models. In the field of subsurface modeling, variations in the behavior of the source are generally not incorporated into analysis. Including information on the direction of arrival to produce dispersive velocity curves as in Chapter 5 ensures that the correct travel time between stations is used. Since the validity of the subsurface model produced from inversion of dispersion curves is reliant on the validity of the travel time measurements between stations, times with strongly directional energy should be isolated and analyzed to determine the correction factor for an angled approach. This section is dedicated to exploring the meaning
of the results in Chapter 5 and hypothesizing about the possible effects of incorporating the information learned in this study into future inversion routines.