GNGTS 2013 - Atti del 32° Convegno Nazionale

respectively. The weak motion velocity data were differentiated to calculate acceleration time series. As a final step, the peak amplitude values of the ground acceleration (PGA) were read on the traces resulting from the geometric mean of the two horizontal components. Ground motion model. The ground-motion predictive equations (hereafter, GMPE) are calibrated with the formulation given by (e.g. Boore and Atkinson, 2008) (1) where Y is the ground-motion parameter to be predicted, M is the local magnitude, M ref is the magnitude to which the magnitude dependence of geometric spreading is referenced, R epi is the epicentral distance (km), h is the pseudo-depth (km), R ref is the distance at which near source predictions are pegged (following Boore and Atkinson (2008), we fixed M re f = 3.6 and R ref = 1 km), and S i are the dummy variables representing the influence of site class and assuming either the value 0 or 1 depending on EC8-soil classes. The grouping of the seismic stations into a particular soil class has been made following the results reported in Gresta and Langer (2002) for the area of the province of Catania. The authors created GIS based electronic maps exploiting the abundant seismic logs available in the province of Catania which allowed identifying the characteristic lithological units according to the EU building code EC8-soil classes. As shown in Fig. 2c, our recording sites are classified B, C, and D, with the class C poorly represented (only 16 data). In this study, the reference soil class is then B. Here, PGAs are expressed in cm/s 2 . The value of the coefficients of equation (1) have been determined through a nonlinear regression analysis of peak ground acceleration by using the Levenberg-Marquardt algorithm for nonlinear least squares. In particular, the procedure iteratively fits a weighted nonlinear regression, where the weights at each iteration are based on each observation’s residual from the previous iteration. These weights serve to down weight points that are outliers so that their influence on the fit is decreased. Iterations continue until the weights converge. We performed several inversions by changing the starting coefficient model. However, the solutions rapidly converged towards the same value of coefficients apart from the starting one. We have estimated the attenuation laws considering epicentral distances less than 30 km and less than 15 km. the latter analysis was carried out with the scope to focus on near-source effects on the GMPEs. For this we used a sub-set of 662 values of horizontal PGA. The obtained regression coefficients and their standard deviations are given in Tab. 1. The goodness of obtained fits is represented by the values of determination coefficients r 2 . The results of the analysis performed separately taking into account all the PGA data for both epicentral distance ranges ( R ep i < 30 km and R epi < 15 km) showed values of e C significantly less than zero, contrarily we expected for a soil class C. However, as said before, only 16 PGA values were available for this soil class, not enough to to warrant homogeneous sampling both with respect to magnitude and spatial sampling. Thus, we have decided to perform the following regressions omitting the contribution of the coefficient e C . The coefficients e D given in Tab. 1 show slightly negative values, as well. However, if we take into account the confidence interval, the positive values are within the range of uncertainties. For comparison, we have also estimated the GMPEs considering the PGA values measured just at recording sites on soil class B. On the other hand, about the 15% of total dataset is associated to soil class D. However, the results suggest that removing the peak ground parameters respective to the soil class D does not significantly change the goodness of fits, and produces a slight increase of the uncertainties of the single coefficient estimates. For this reason, in the following we focus on the results obtained for the site independent model, preferring the increase of significance level of the findings allowed by a bigger sample size. The fits of the data by the GMPEs are illustrated in Fig. 3a and 3b, for epicentral distances ( R epi ) less than 30 km and 15 km. Examining the figures reveals that, within the range of uncertainties, the PGA 155 GNGTS 2013 S essione 2.1