Statistical estimation of earthquake site response from noise recordings

Santa Barbara (California, USA)

Standard spectral ratio from earthquake recordings (SSR) is considered the reference empirical method for assessing site effects as a function of frequency. However, other estimates can be easily obtained from noise measurements (i.e., Horizontal-to-Vertical Spectral Ratio, HVN), even though their reliability in terms of amplitude is controversial.
In the framework of the ToK ITSAK-GR (2006-2010) EC project, Cultrera et al. (2010) analyzed recordings from 64 sites worldwide, founding that it is possible to have linear combinations of the HVN amplitudes significantly correlated to linear combinations of the SSR. In the present paper we show how to estimate the SSR spectral ratios when only noise measurements are available, using the results of the canonical correlation analysis between SSR and HVN recorded at several sites. The SSR evaluation has been tested by a cross validation procedure: the expected SSR at each validation site are in turn estimated by a weighted average of the SSR values measured at the other sites; the weights are properly set to account more for the sites with similar behavior in terms of the canonical correlation between HVN and SSR. To evaluate the goodness of the estimation, we compared all the inferred and original SSR, and we performed a critical analysis on the spectral characteristics of earthquake site response that can be easily recovered from noise measurements.