GNGTS 2015 - Atti del 34° Convegno Nazionale

respectively) and the other characterized by a fixed polarization direction, simulating a site with directional resonance. Using a purposely written code, six synthetic signals were generated, named: 1) surf100, derived from 50 sources of Rayleigh type and 50 of Love type, both with a ground motion polarized along an azimuth of 37°, summed with an incoherent background noise kept at a very low amplitude level (of the order of 1/1000 of Rayleigh waves); 2) surf100i, differing from surf100 for the variability of the azimuth of surface wave polarizations, depending on the wave propagation direction; 3) surf100sn3, obtained from surf100 by increasing the level of random noise so to reduce the signal-to-noise amplitude ratio ( snr ) to 3; 4) surf100sn3i, derived from surf100i by increasing noise as for surf100sn3; 5) surf100sn1, obtained from surf100 by further increasing noise so that snr is equal to 1; 6) surf100sn1i, derived from surf100i with the same noise as surf100sn1. These signals were analyzed according to the procedure outlined in the previous section to obtain HVIP values calculated at 23 frequencies, spaced by 0.25 Hz, between 0.5 and 6 Hz. The HVIP values were compared with the actual H/V ratios of the Rayleigh waves present in the synthetic signal. During these tests, two aspects were mainly examined, i.e. accuracy and precision of the H/V estimates derived from HVIP values. Accuracy was evaluated from the root mean square of errors rms err (HVIP deviations from actual H/V ratios of Rayleigh waves), precision from the scatter, expressed through the root mean square rms sc of deviations of instantaneous H max /V values from the mean HVIP values. The code used for the instantaneous polarization analysis calculates both HVIP values averaged among samples polarized in any direction and among samples grouped into azimuth bins spaced by 10°. For signals with equally polarized surface waves (surf100, surf100sn3, surf100sn1), the actual H/V values were compared with HVIP averaged on samples polarized within the azimuth bin 30°- 40°, whereas, for randomly polarized signals (surf100i, surf100sn3i, surf100sn1i), with the HVIP averaged over all the azimuths. These two situations are representative of the cases of directional and isotropic site response, respectively. Results. A preliminary series of test was aimed at selecting an optimal filtering type to reconstruct the H/V curve as function of frequency. The best results in terms of accuracy of H/V curve estimation were obtained by adopting a Gaussian filtering multiplying spectral amplitudes by a function (3) where ν c is the central frequency and β is a fixed parameter governing the filtering bandwidth. Using this kind of filtering, each of the six synthetic signals was analyzed adopting different combinations of values for: 1) the filtering parameter β ; 2) the threshold diplim of angular deviations of a →  ( t ), b →  ( t ) and p →  ( t ) from horizontal/vertical directions; 3) the rectilinearity limit rllim distinguishing Rayleigh from Love waves; 4) the minimum number nmin of consecutive data samples that identifies a coherent wave packet. The results of these tests show that the accuracy in H/V curve estimate is better for the directional case than for the isotropic one and deteriorates as snr decreases (more rapidly for the isotropic than for the directional one). Error minimization was generally obtained adopting a relatively small threshold for diplim (5°) and for rllim = 0.80, combined with the highest nmin values among those tested (15-20). All these thresholds corresponds to more restrictive criteria in Rayleigh wave packet identification, taking however into account that a too restrictive criterion can excessively limit the number of Rayleigh type signals over which HVIP values are calculated. This limitation can have a negative effect on the relation between accuracy and precision. In general, it is desirable to have a good correlation between accuracy and precision so that, in field observations, the latter can give indication about the former. From the tests carried out, GNGTS 2015 S essione 2.2 77