| Abstract: | The authors examine the reliability of site response estimations obtained by the horizontal to vertical (H/V) spectral ratios of microtremors by means of cross‐validation with the ratio of the horizontal spectra of earthquake motion with respect to reference sites. The data comprise microtremor and ground motion records recorded at 150 sites of Yokohama strong motion array. The use of non‐supervised pattern recognition techniques aims to group the sites with more objectivity. Attributes defining the overall shape of the amplification spectra serve as input in the computation of Euclidean distance similarity coefficients amongst sites. The implementation of the Ward clustering scheme leads to the attainment of a meaningful tree diagram. Its analysis shows the possibility of summarizing the results into six general patterns. A good coincidence of site effects estimates at 80 per cent of the sites becomes apparent. However, this coincidence appears poor for sites characterized by H/V amplification ratios around 2 or smaller and predominant periods longer than 0.5 s. In such cases, the presence of stiff, sandy sediments in the soil profile proves common. To proscribe H/V estimations, relying solely on the small spectral ratios criterion seems inadequate. Copyright © 2001 John Wiley & Sons, Ltd. |
