S13D-4508:
Wavelet Domain Source Inversion with Multi-Time-Window Parameterization for Examination of Frequency Dependence of Seismic Wave Radiation

Monday, 15 December 2014
Wataru Suzuki, Shin Aoi and Takashi Kunugi, NIED National Research Institute for Earth Science and Disaster Prevention, Tsukuba, Japan
Abstract:
Frequency-dependent characteristics of seismic wave radiation from earthquake sources are important subject for advancing the source physics and the strong-motion prediction. The importance is rediscovered from the 2011 Tohoku-Oki earthquake, for which the sources of the high- (>1 Hz) and low-frequency (<0.1 Hz) seismic waves seem totally different in depths. Our previous study examined the contribution of the notable slip events detected from the source inversion to the low-frequency (0.01-0.125 Hz) waveform synthesis. We found that the very-low-frequency (<0.02 Hz) and the higher-frequency waves seem to be radiated from the shallow and deep portions of the fault, respectively, even in this low-frequency band. However, the examination on the frequency dependence is somewhat indirect. We therefore develop a source inversion method that utilizes the wavelet coefficients as the target to fit. In order to flexibly describe slip velocity functions, we employ the multi-time-window scheme and linear inversion method. Thus, the derived moment rate is directly related to the waveforms in each octave band and we can discuss the frequency content straightforwardly generated by a specific slip event. The method has the other advantage that we can fit the synthetic waveform over the wide frequency bands including that with smaller amplitudes by using frequency-dependent weights in evaluating the fitness. To see the applicability of the developed method, we analyze the largest aftershock (Mw7.9) of the Tohoku-Oki event, whose rupture process seems simpler than the mainshock. The data set consists of 0.01-0.125 Hz strong-motion velocity waveforms, which are divided into four frequency bands. The fitness is evaluated as the residual of the wavelet coefficients normalized by the maximum value for each station and each frequency band. From the wavelet inversion, we estimate one large slip area located deeper than that derived from the waveform inversion. An examination of the waveform synthesis from the shallow and deep portions of the fault shows that the deep portion contributes more greatly to the highest frequency band than the shallow one whereas the two portions equally contribute to the lower-frequency bands. These results may suggest a depth dependence of the frequency contents of the radiated seismic waves.