GNGTS 2015 - Atti del 34° Convegno Nazionale

waves both for the case of directionally and isotropically polarized signals. Selecting the analysis parameters that maximize precision (minimizing the scatter rms sc ), the resulting HVIP curves are coincident with (for surf100) or very close to (for surf100i) that minimizing errors. The rms err value obtained for these curves are 0.18 and 0.21, respectively for the directional and isotropic cases, whereas the scatter rms sc is 0.42 and 0.41, respectively. The HVIP curves correctly identifies the peak frequency, estimates H/V peak value with a 13-17% error and reproduces well the general shape of the H/V curve, including the lower part where the H/ V ratios are below 1. Comparatively, the outcome of Nakamura technique provides a better estimate of the peak value for the directional case (3% error), but larger deviations at higher frequencies, so that rms err is higher (0.39). In the case of isotropic signal (surf100i), the HVNR calculated through the Euclidean average gives a better estimate of the peak value (5% error), whereas geometric average better fits the lower part of the H/V curve, but both show overall a higher rms err than HVIP (0.41 and 0.35, respectively). Fig. 2 – Comparison of actual H/V ratio curve of Rayleigh waves present in the synthetic signals (black solid line) with estimates obtained from HVIP analysis and from Nakamura’s technique, for different synthetic signals: a) surf100, b) surf100i, c) surf100sn3, d) surf100sn3i, e) surfsn1, f) surf100sn1i. Grey and dotted lines represent the best HVIP estimate in term of accuracy and precision, respectively; red line for the case of directionally polarized signal, represent the HVNR curve obtained for the azimuth (35°) characterized by the highest spectral ratio values; red and yellow curves, for the case of isotropic signals, represent the HVNR values obtained from North and Est signal components through euclidean and geometric averages, respectively. Vertical bars represent the scatter of HVIP estimates at each analysed frequency. GNGTS 2015 S essione 2.2 79