11:00 AM - 1:00 PM
[SSS06-P03] Improving depth resolution in crustal monitoring using ambient noise
Keywords:Ambient noise, S-wave velocity change, wavelet transform, Monitoring
Seismic velocity changes in the crust are caused by fault zones damage, changes in pore pressure and stress state, and the recovery process from earthquake damage. Therefore, estimating seismic velocity changes provides important information for predicting earthquakes and volcanic activities. In previous studies, monitoring of seismic velocity changes using cross-correlation of ambient noise has been actively conducted, but the depth resolution is limited to clarify depths where seismic velocity changes occurred. In this study, we aim to estimate the depth distribution of S-wave velocity changes by improving the depth resolution using the seismic velocity changes at different frequencies. We applied our approach to ambient noise data including the 2016 Kumamoto earthquake.
Firstly, we constructed the reference S-wave velocity structure from cross-correlation of ambient noise using frequency domain analysis (the zero-cross method). Then, we applied the stretching method to the coda of cross-correlation with the wavelet transform to obtain the phase velocity changes for each frequency. Finally, we estimated the S-wave velocity change at each depth by the inverse analysis using the reference S-wave velocity structure and the phase velocity changes at each frequency. In this study, we used two types of mother wavelet, the Morlet wavelet and the Paul wavelet, for the wavelet transform. The stretching method was applied to the time window of coda waves, and three types of window lengths, 10, 20 and 30 cycles of each frequency, were used for comparison.
The result of phase velocity changes from the stretching method shows a phase velocity decrease after the earthquake for any window length, but the window length of 20 cycles was the best for the following reasons. When the window length was 10 cycles, estimated velocity changes were fluctuated. On the other hand, after 20 cycles, velocity changes were stably estimated without clear fluctuation. Because it is known that the coda wave in later time includes the information of not only surface waves but also that of body waves, we determined the most appropriate window length as 20 cycles in our study.
We also compared the influence of the mother wavelet on the inverted depth distributions of S-wave velocity changes. When using the Morlet wavelet, the peak of S-wave velocity drops was clearly estimated in depth direction. However, the S-wave velocity changes were fluctuated at depth where velocity changes were small. On the other hand, when using the Paul wavelet, the resulting velocity changes were stable but their peaks were unclear. Thus, using the Morlet wavelet is better to clearly observe the peak of the S-wave velocity change, while using the Paul wavelet is better to estimate a stable velocity change.
In conclusions, the window length of 20 cycles is the best to estimate the S-wave velocity changes in our case. The Morlet wavelet is superior for improving the depth resolution of S-wave velocity change, while the Paul wavelet is better to estimate stable velocity change. To obtain stable S-wave velocity change at each depth, it is necessary to provide appropriate parameter settings and constrains for the inverse analysis of S-wave velocity change. In the future, to reduce the influence of the sensitivity kernel of the surface layer, the S-wave velocity change of the first layer can be fixed and first layer data can be divided from the existing data.
Firstly, we constructed the reference S-wave velocity structure from cross-correlation of ambient noise using frequency domain analysis (the zero-cross method). Then, we applied the stretching method to the coda of cross-correlation with the wavelet transform to obtain the phase velocity changes for each frequency. Finally, we estimated the S-wave velocity change at each depth by the inverse analysis using the reference S-wave velocity structure and the phase velocity changes at each frequency. In this study, we used two types of mother wavelet, the Morlet wavelet and the Paul wavelet, for the wavelet transform. The stretching method was applied to the time window of coda waves, and three types of window lengths, 10, 20 and 30 cycles of each frequency, were used for comparison.
The result of phase velocity changes from the stretching method shows a phase velocity decrease after the earthquake for any window length, but the window length of 20 cycles was the best for the following reasons. When the window length was 10 cycles, estimated velocity changes were fluctuated. On the other hand, after 20 cycles, velocity changes were stably estimated without clear fluctuation. Because it is known that the coda wave in later time includes the information of not only surface waves but also that of body waves, we determined the most appropriate window length as 20 cycles in our study.
We also compared the influence of the mother wavelet on the inverted depth distributions of S-wave velocity changes. When using the Morlet wavelet, the peak of S-wave velocity drops was clearly estimated in depth direction. However, the S-wave velocity changes were fluctuated at depth where velocity changes were small. On the other hand, when using the Paul wavelet, the resulting velocity changes were stable but their peaks were unclear. Thus, using the Morlet wavelet is better to clearly observe the peak of the S-wave velocity change, while using the Paul wavelet is better to estimate a stable velocity change.
In conclusions, the window length of 20 cycles is the best to estimate the S-wave velocity changes in our case. The Morlet wavelet is superior for improving the depth resolution of S-wave velocity change, while the Paul wavelet is better to estimate stable velocity change. To obtain stable S-wave velocity change at each depth, it is necessary to provide appropriate parameter settings and constrains for the inverse analysis of S-wave velocity change. In the future, to reduce the influence of the sensitivity kernel of the surface layer, the S-wave velocity change of the first layer can be fixed and first layer data can be divided from the existing data.