On the 12th of November 2017, a strong earthquake with a moment magnitude of 7.3 struck the west of Iran, causing massive damage in SarpoleZahab and surrounding areas. In this study, the Horizontal to Vertical Spectral Ratio (HVSR or H/V technique) of strong ground motion recorded at 60 stations is used for identifying of site response and subsequent site classification. Very clear amplification is observed at stiff and soft soil sites. Empirical equations from the literature relating frequency of HVSR peak to average shear wave velocity in the upper 30m(VS30), commonly used as a proxy for site classification, were found to be unreliable, making site classification difficult. To overcome this problem, a soft computing method based on the Generalized Regression Neural Networks (GRNN) algorithm is used. This algorithm relies on HVSR curves at the recording stations and uses pattern recognition against a set of pre-defined curves for different site types. In this approach, site characterization is based on the overall features of the HVSR curves rather than just the peak frequency. The stations are classified into four different site classes, and the results are consistent with local geotechnical information, where available. Rather than relying on VS30, which has been found deficient in many studies, and is known to have difficulty in reliably estimating when geotechnical data is lacking, site classification proposed in this study uses the overall characteristics of HVSR curves of response spectra of recorded strong motion data. This method is proposed as a more reliable and direct method for site characterization in seismically hazardous regions.
© 2020 Elsevier B.V.