Testing three different approaches of extracting empirical Green's functions using ambient noise correlation analysis with a limestone mine induced event in South Korea

Ambient noise cross-correlation analysis is applied to simulate ground motions from a Mw 4.2 seismic event, excited by a limestone mine collapse in South Korea on January 31, 2015. The moment tensor solution shows that the mine-induced event can be interpreted as a downward single force at the Earth surface. A semi-broadband station (MGR) was installed at the epicenter location of the mine-induced event in order to verify that the ambient noise cross-correlation is applicable to retrieve both amplitude and phase information of Green's functions between station pairs. Six months of continuous ambient noise data with three components were collected at broadband stations in South Korea and employed for the cross-correlation analysis.We tested three different approaches: (1) The first approach is a deconvolution based one, i.e., cross-correlating two ambient noise data in the frequency domain and normalizing it by the square of the amplitude spectrum of source station data (Prieto and Beroza, 2008; Denolle et al., 2013). (2) The second approach is a cross-correlation method with one-bit normalization and spectral whitening, commonly used to extract surface wave dispersion curves from ambient noise data (Bensen et al., 2007; Shapiro & Campillo, 2004). (3) Lastly cross-correlation is solely applied to retrieve Green's functions without any spectral treatment (Wapenaar, 2004). Three approaches adopted in this study have differences in terms of spectral stabilization.Our results show that both amplitude and phase information of ground motions are well retrieved from the first and second methods. In particular, the ground motions obtained by the deconvolution-based method shows the best fitting with the observed ground motion records from the mine induced event. It is also surprising to see that the second method also yields relatively good waveform fitting in spite of the one-bit normalization and spectral whitening.