17:15 〜 18:45
[SSS07-P05] Influence of incident wave and taper process on receiver function analysis
キーワード：デコンボリューション、スペクトルホール、テーパ
Receiver function analysis has been used to estimate S-wave velocity structure [Langston, 1979]. Receiver functions are computed by deconvolving the horizontal component by the vertical component of seismic waveforms. Frequencies with extremely small values in a power spectrum are called spectral holes, and ideal receiver functions derived from the S-wave velocity structure are not computed by deconvolution when the power spectrum of the vertical component has spectral holes. In order to reduce spectral holes, the multitaper method has been suggested [Park and Levin, 2000]. The multitaper method applies multiple tapers to seismic waveforms to smooth the spectra and reduce spectral holes. However, the taper process has a nonlinear change in the power spectral, and the effect of tapers on the receiver function analysis could be clearer.Furthermore, the calculation formula for the multitaper method is different from the straightforward deconvolution in the frequency domain because it applies multiple tapers and performs an operation to sum the spectra of signals applied to each taper. Park and Levin [2000] conducted the spike deconvolution test to evaluate the accuracy of the multitaper method in receiver function analysis. According to the spike deconvolution test, although almost the ideal receiver function is extracted by the multitaper method, the calculated receiver function contains computational stationary noise, and the relationship between the stationary noise and the taper process has yet to be considered. In this study, we created several synthetic waveforms by convoluting incident waves and Green functions and conducted receiver function analysis by deconvolution in the frequency domain after taper processing on the synthetic waveforms. Then, comparing the calculated receiver function with the ideal one, we considered the relationship between the receiver function and the taper process.We assumed a multi-layered structure and almost perpendicular P-waves incident on the lowest layer of the layer boundary under consideration. We assumed an impulse and a triplet of impulses as the vertical and horizontal Green function, respectively, corresponding to the direct P-wave and the P-S conversion wave by each layer. We computed deconvolution to obtain receiver functions in three ways: without tapering, with a single taper, and with multiple tapers. We employed the Slepian sequence as tapers [Park et al., 1987]. The receiver function is ideally equal to the ratio of the horizontal and vertical components of the Green function by removing the influence of the source time function. Therefore, we compared the calculated receiver function and the ideal one.We assumed three types of incident waves: Hanning, the asymmetric function without spectral holes, and P-wave of natural seismic waveform including P-wave coda. We created the synthetic wave by convoluting an N-second Hanning as the incident wave and the Green function. The deconvolution without taper processing reproduced the ideal receiver function well, whereas the deconvolution with taper processing computed equidistant spike noise and stationary noise that did not exist in the ideal receiver function. The spike noise occurred at intervals of N seconds from the peaks of the delta functions of the ideal receiver function. When the P-wave and its coda extracted from a natural seismic waveform was used as the incident wave, spike noise was not generated, and stationary noise was larger. When the asymmetric function without spectral holes was used as incident waves, stationary noise was smaller, and spike noise was not generated. We conclude that the spike noise and stationary noise generated by the deconvolution depend on the assumed incident wave and taper process.