GNGTS 2016 - Atti del 35° Convegno Nazionale

284 GNGTS 2016 S essione 2.1 PHV values without including site effect terms. In this case, apart from fluctuations affecting averages calculated on a small number of data, residuals tend to increase systematically passing from B to C and D sites, regardless of magnitude and distance range. This explains the better performance, in PHV predictions, of GMPE accounting for site amplification and, in this case, modelling the site effect with dummy variables appears adequate. The predictive performances of Eqs. (2) and (3) were also compared with those obtained in previous studies, i.e. Danciu and Tselentis (2007), relative to the Greek region, and Akkar and Bommer (2007a, 2007b) relative to the Europe - Middle East region (including Greece). Since these equations were calibrated on dataset whose recordings were within a different range of magnitude and distances, their predictive performances were compared with those of Eqs. (2) and (3) by using subsets of the validation datasets complying with the applicability limits of the matched equations. In terms of root mean square of errors of logarithmic estimates ( rmsl ), the present study equations provided better results than both Danciu and Tselentis (2007) ( rmsl = 0.321 and 0.350 for PHA and PHV, respectively, against 0. 377 and 0.385) and Akkar and Bommer (2007a, 2007b) ( rmsl = 0.311 and 0.355 for PHA and PHV, respectively, against 0. 328 and 0.419). This result was obtained despite the use of simpler functional forms not including focal mechanism terms and, for PHA, even site effect terms, present in the previously published equations. Although, overall, residuals do not show a correlation with magnitude and distance when dependence on these parameters is separately considered, plotting logarithmic error as function of observed PHA and PHV, an inverse correlation was found, which implies ground motion underestimates for magnitude-distance combinations responsible for larger ground motions, Fig. 1 – Mean residuals of the best PHA predictor when applied to the validation dataset. Residuals were averaged over a range of recording distances (top) or source magnitudes (bottom) from minimum to increasing upper bounds (to the left) or from increasing lower bounds and maximum (to the right). Different symbols are used for residuals relative to recordings acquired under different site conditions (class B, C and D), according to the legend.