GNGTS 2013 - Atti del 32° Convegno Nazionale

approximately E-W directed polarisation, which is consistent with the indications about site response directivity, derived from seismic event recordings. This preferential concentration is just outlined in the total dataset (Fig. 2 a, d, g) and is more pronounced for Rayleigh and SH type polarisations. The presence of a certain dispersion in azimuth frequency distribution around the preferential orientation and the variation among results obtained for different campaigns suggest that polarisations controlled by noise source properties can alter polarisations depending on site response. This could explain: 1) the slight deviation of the preferential polarisation azimuth (e.g. the modal azimuth interval for the entire dataset was 80°-90° in 2007-2011 measurements and 90°-100° in 2013); 2) the occasional burst of a maximum in data polarisation distribution along a different azimuth at some frequencies (e.g. at azimuth of 160°-170° and frequency below 1 Hz for Rayleigh and SH-type packets in the 2011 recording: see Fig. 2 e, f). The latter case likely reflects the excitation of a strongly polarised signals by temporary noise source. thus, such signals cannot be considered representative of site response properties. The dependence of the CAR2 preferential polarisations on site effects is demonstrated by the comparison with the results obtained analysing noise recordings acquired at the same times at the nearby site CAR5 (located outside the 1989 landslide boundary). No common preferential polarisation is observed at different times and among differently classified signals (see Fig. 3). A possible preferential orientation of polarisation may be recognised for the 2013 campaign along an azimuth whose value is slightly larger than that observed at CAR2 in previous measurements (Fig. 3 g, h, i). However this is probably due to the presence of a polarisation controlled by sources acting on a wide frequency band: this can perhaps explain also the slight deviation of the modal azimuth interval observed at CAR2 for the 2013 measurements. It is also noteworthy that, in 2013 CAR5 measurements, H/V ratios larger than 10 3 appear starting from 5 Hz and become common to all azimuth at frequencies higher than 9 Hz, with values up to the order of 10 6 (Fig. 3 g). This appears due to the presence of strongly linearly polarised signals of SH-type, which show even higher H/V ratios with maximum up to the order of 10 7 (Fig. 3 i). Despite of the presence of abnormally high H/V values in the general dataset, Rayleigh type wave packets show H/V values in the same range as in the results of the previous campaigns (Fig. 3 h), confirming that H/V ratios derived from Rayleigh waves appear more coherent in different measurements and less influenced by peculiar condition of noise wavefield excitation. The presence of a preferential polarisation in the noise recording does not necessarily imply the existence of a directional resonance. Indeed, a high percentage value of data samples that, at a given frequency, show a common polarisation is not necessarily associated to high values of the H/V ratios (represented in Figs. 2 and 3 by the colour scale). Thus, for the recognition of directional resonance phenomena, both the presence of high H/V values and of a significant percentage of data samples showing a common orientation should be verified. A further support to resonance properties identification can derive from the observation of a moderate standard deviation of H/V values and by the consistency of results derived from measurement repetitions. Such a condition is observed, for example, at the site CAR2 for the peak values of H/V ratios found at 2.75 Hz for azimuth of 70°-90° (in quite good agreement with the observations on directional resonance from seismic event analysis), whose amplitude was relatively consistent among the results obtained from different campaigns, i.e. 7.0 ± 4.1 in 2007, 6.6 ± 4.8 in 2011 and 7.6 ± 4.7 in 2013. These values were quite higher than those (in the order of 3-5) derived from the HVNR technique (see Del Gaudio et al. , 2013). Conclusions. The analysis of ambient noise instantaneous polarisation offers interesting prospects as tool to investigate site response to seismic shaking, particularly in the case of complex site conditions, where resonance properties can be characterised by multiple anisotropic peaks. In particular one can select, in a noise recording, wave packets of consecutive data samples showing a Rayleigh-type polarisation, determining for them the particle motion ellipticity and the polarisation orientation. 233 GNGTS 2013 S essione 2.2