GNGTS 2013 - Atti del 32° Convegno Nazionale

amplification factors cannot be derived from H/V ratio amplitude because of the uncertain composition of noise wavefield, including a mix of contributions from body and surface waves (Bonnefoy-Claudet et al. , 2006). In principle, the H/V curve could be inverted to derive shear wave velocity vertical profiles (Castellaro and Mulargia, 2009), which, in turn could be used to derive amplification factors from numerical simulations. However, data interpretation would require a correct identification of the nature of noise wavefield, considering that H/V ratio amplitude varies according to the proportion of Rayleigh, Love or body waves present in the ambient noise (cf. Albarello and Lunedei, 2009). The standard application of the HVNR technique, for the calculation of the H/V ratios, makes use of an averaging of the horizontal components, which, for the case of flat horizontal layering, is justified by the assumption of site response isotropy. However, under more complex site conditions (e.g. on slopes affected by landslides) resonance often shows pronounced directional variations (e.g. Del Gaudio and Wasowski, 2007) which can be revealed by an analysis of H/V azimuthal variations (Del Gaudio et al. , 2008). Among noise wave trains, Rayleigh waves appear to better reflect site response directivity (Del Gaudio et al. , 2013), but, on geological bodies having a 3D geometry, noise wavefield generally presents a mix of different kinds of waves. The energy distribution among such wave types depends on measurement site characteristics (e.g. presence of a more or less large impedance contrast, deviation of interface geometry from 1D layering) and on noise source distance (cf. Haghshenas et al. , 2008). On the whole, for a reliable determination of site response properties from noise analysis, it would be desirable to distinguish different kinds of noise wave packets, possibly isolating the Rayleigh wave contribution. Some workers (e.g. Fäh et al. , 2001; Poggi et al. , 2012) have made attempts in this direction, using a time-frequency analysis to identify signal portions for which the presence of a noteworthy energy on the vertical component leads to exclude a significant contribution from SH-type polarised signals (Love or body waves). In this paper a different approach is presented, which is based on a technique of analysis of instantaneous polarisation of the noise signal. After a brief description of the methodology, the results of some preliminary tests are presented and discussed. These tests were carried out on data recorded on slopes affected by a landslide, where the presence of site response directivity had been revealed by previous analysis of seismic events and noise recordings. Thus, the pre- existing available data provides a possibility of validating the outcome of this new approach. Methodology. Polarisation properties of ambient noise recording can be estimated instant by instant through its analytic representation, which, for a single component signal u(t) , is given by (1) where j is the imaginary unit and û(t) is the Hilbert transform of u(t) . The Eq. (1) allows to represent the signal as a complex quantity characterised by a complex amplitude A ( t ) and a real phase Ф( t ) (both changing instant by instant), i.e. as a vector in a complex space, whose real component coincides with the observed instantaneous signal. Considering i) the combination of the two horizontal components or ii) of all the three spatial components, their analytic representation allows to describe the ground motion as a vector in a multidimensional complex space. Its real components at each instant describe in the real space a time-variant elliptical trajectory lying, in the case i), in the horizontal plane and, in the case ii), in a plane of the three-dimensional space (Morozov and Smithson, 1996). With regard to these “instantaneous” ellipses, one can determine modulus and orientation of their major and minor axes by finding the phase angle Ф o ( t ) that, subtracted from the phases of the components of the analytic signal, maximises the modulus of the vectorial composition of the real components. Ф o ( t ) is quite simply obtained by the equation (Morozov and Smithson, 1996): 227 GNGTS 2013 S essione 2.2