GNGTS 2015 - Atti del 34° Convegno Nazionale

smoothed using a Konno-Ohmachi filter, of the S-wave time window were compared with the pre-event noise, in order to select good quality data on the basis of the signal/noise ratio. Only those signals with s/n ≥ 3 were considered for analysis. The spectral ratio were evaluated at each station for the selected events and a geometric mean of all spectral ratios were computed to obtain the mean HVSR curve and the corresponding standard deviation. Ambient noise recordings were selected from the stations of the INGV-Osservatorio Etneo. The signal length was of about 1 hour at each stations. To check the stability of the HVNR the signals were selected in different days, taking into account both days with low level of noise and with an high level of noise, due to either weather conditions or to the presence of lava fountains. The microtremor recordings were de-trended, band-pass filtered and subdivided in 30 s, 5% cosine tapered, time windows. Through an anti-trigger algorithm based on STA/LTA (Short Time Average over Long Time Average) only the most stationary parts were selected and transients associated to very close sources were excluded. �� ����������� ��� ��� ������� �� �� In particular, STA was settled to 1s and LTA to 20 s. ����� ��� ��������� ��������� �� ������ �������� ���� ���������� ��� �������� After the necessary processes of signal cleaning were completed, the spectral ratio technique was applied to obtain the mean spectral ratios and corresponding standard deviation. Once the HVSR and HVNR were performed, they were overlapped ��� �� ��� �������� � and it was observed a strong similarity between the obtained results ����� ��� ��������� ���������� ��� ����������� �� ��� (Fig. 1), therefore confirming the reliability of the peak frequency found through both the techniques. Numerical modelling. HVSR inversion. The ModelHVSR Matlab routines (Herak, 2008) were adopted to compute theoretical HVSR in homogeneous and isotropic layers. The soil model consists of a number of visco-elastic layers, stacked over a half-space, each of them being defined by the thickness (h), the velocity of the body waves (V P and V S ), the density (ρ), and the Q-factor, which controls the inelastic properties. The incoming waves are assumed to be travelling vertically and ������� ������������� �� ��� ������� �� ��� ���������� ��� �������� without ������������� �� ��� ������� �� ��� ���������� ��� �������� amplification at the bedrock of the horizontal and vertical motions. The HVSR at the surface is then obtained as the ratio between the theoretical transfer functions of S and P waves. In particular, the observed HVSR is inverted through a Monte Carlo simulation aiming to find soil models that minimize the misfit function with theoretical HVSR. The Monte Carlo search is started with an initial model whose parameters are then randomly perturbed within the bounds defined by the user. In our case, the number of random tries was settled to 10,000 and the initial model of body wave velocities (V P and V S ) and layers thickness was perturbed by ±5% and ±25% respectively, to obtain a good fit with the experimental results. For all the considered seismic stations of Mt. Etna, stratigraphic sequences were made by integrating observations and literature information (Branca et al. , 2011; Branca and Ferrara, 2013). Such procedure allowed us to estimate the thicknesses of the layers and then, by using geotechnical and geophysical literature data (Azzaro et al. , 2010; Priolo, 1999), a initial velocity model as well as density, Qp and Qs were assigned to each layer. Site amplification functions. The amplifications were computed from frequency-domain calculations, using the programs Site_amp and Nrattle (Boore, 2003). In particular, using as input parameters V S , density (��� ��� �� ��� ������� ������� ������� ���������� ��� �������� ρ ������� ������� ���������� ��� �������� ), and Q, the Nrattle Fortran routine calculates the Thomson- Haskell plane SH-wave transfer function for horizontally stratified constant velocity layers at a specific incidence angle, within a uniform velocity halfspace settled equal to the deepest measured layer. The code compute amplifications at specified frequencies or at frequencies corresponding to the “breakpoints” in the velocity model. To compile the input file for the Site_ amp and Nrattle codes we used the output model of HVSR obtaining as input the amplification function (AF) for 14 seismic stations, showing the best convergence between experimental and theoretical models. We adopted a classification scheme different from those based on Vs 30 , so that following the one proposed by many authors (Luzi et al., 2011; Di Alessandro et al. , 2012; Zhao et al. , 2006), we used the predominant soil period to discriminate among classes. The following Amplification Function classes were obtained taking into account the dominant frequency peak in the HVSRs: 144 GNGTS 2015 S essione 2.2