GNGTS 2013 - Atti del 32° Convegno Nazionale

satisfy the following requirements: for Rayleigh waves, a maximum deviation by 5° of the planarity vector from horizontality and, simultaneously, of and from horizontality and verticality (or viceversa), respectively; for SH-type waves, a maximum deviation of from horizontality by 5° and a minimum rectilinearity value [Eq. (5)] of 0.85. For each filtered signal, following the procedure described in the previous section, the H/V ratio amplitudes and orientations of all the data samples were calculated, then grouping together those falling into azimuth intervals of 10°. The percentages of data samples with polarisation comprised within each azimuth interval were calculated together with the average and the standard deviation of their H/V ratios. Percentages and H/V mean ratios were then represented as function of azimuth and filtering central frequency. This calculation was carried out for all the data samples and, separately for only those satisfying the criteria of identification of Rayleigh and SH-type wave packets. A synthetic representation of results was adopted making use of 3D histograms where, as function of filtering central frequencies, columns are reported with height proportional to the percentage of data samples falling into each azimuth bin, and colour assigned according to a chromatic scale to represent the corresponding H/V mean ratio. Results. A comparison of the mean H/V ratios obtained following the HVNR technique with those obtained from instantaneous polarisation analysis (hereafter, HVIP) was first extended to all the recording data samples. HVNR and HVIP show a similar pattern of relative maximum distribution as function of frequency and azimuth (Fig. 1), but the amplitude of the HVIP values are generally quite larger (e.g. 7.3 against 4.2 for the CAR2 H/V peak value). Standard deviation of HVIP is much larger as well, but this reflects the fact that, for the HVNR values, the averaging of spectral ratios calculated on time windows, typically of some tens of second, implies a minor variability of results in comparison to the averaging of instantaneous H/V estimates. The application of criteria for the identification of Rayleigh or SH-type wave packets leads to select a much smaller subset of data samples. Starting from data sets including from a minimum of about 100000 to a maximum of almost 500000 data, samples classified as Rayleigh type were at most in the order of 1% and were arranged in packets from less than 10 to several tens of data. A larger number of samples were classified as SH-type (in some cases and for some frequencies up to 30 %) arranged in packets including from about 10 to several hundreds of consecutive data samples. For both the polarisation types, longer wave packets tends to appear at lower frequencies, whereas the highest percentages of classified data can be found at variable frequencies. These results indicate that only a small fraction of the recordings shows a well defined polarisation of Rayleigh or SH type: this is probably due to the fact that, for most of the noise recording, there is an overlapping of signals of different polarisation. Thus, the instantaneous polarisation analysis allow to select wave packets only when signals with a specific kind of polarisation are dominant. The values of H/V mean ratios are obviously much higher for SH-type than for Rayleigh type wave packets. However, although the former kind of polarisation appear more common than the latter, the mean H/V ratios for the entire recording are generally of the same order of those of Rayleigh type packets. This is probably due to the fact that a large majority of data samples not presenting a well defined polarisation has low values of H/V ratios, thus balancing, on the average, the very high values of the SH-type data samples. Furthermore standard deviation of H/V ratios for Rayleigh type packets, are generally lower than for the SH-type or for the entire signals, which indicates that Rayleigh waves tend to show more coherent properties. Interesting observations can be derived from the examination of 3D histogram that synthesise information on polarisation distribution and H/V mean values for different azimuth intervals, at different frequencies. In the case of the site CAR2, characterised by site response directivity (Fig. 2), it is apparent the preferential concentration of data samples having an 232 GNGTS 2013 S essione 2.2