GNGTS 2017 - 36° Convegno Nazionale

284 GNGTS 2017 S essione 2.1 Tab. 1 - Final rank of GMPEs for ASCRs. ITA10: Bindi et al. (2011); ASB14_Rjb and ASB14_Rhypo: Akkar et al. (2014); AB10: Akkar and Bommer (2010); BSSA14: Boore et al. (2014); BND14_Rjb_EC8 and BND14_Rhypo_ EC8: Bindi et al. (2014); KS15: Kuehn and Scherbaum (2015); DBC14: Derras et al. (2014); CZ15_800: Cauzzi et al. (2015); ZHA06: Zhao et al. (2006); ASK14: Abrahamson et al. (2014); IDR14: Idriss (2014); CY14: Chiou and Youngs (2014); CB14: Campbell and Bozorgnia (2014). Additional details on selected models can be found at http://www.gmpe.org.uk . GMPE s All EC8-A MR After 2009 tot ITA10 47 48 45 44 184 BND14 Rhypo 39 40 42 40 161 ASB14 Repi 41 40 22 42 145 ASB14 Rhypo 43 36 21 44 144 BND14 Rjb 35 27 47 35 144 ASB14 Rjb 36 32 26 37 131 CZ15 28 39 36 26 129 DBC14 31 29 32 30 122 KS15 20 24 33 18 95 CY14 21 21 15 20 77 ASK14 16 21 8 19 64 BSSA14 17 15 15 17 64 CB14 15 15 4 16 50 IDR14 8 12 17 10 47 AB10 7 5 20 6 38 ZHA06 4 4 25 4 37 In particular, from many available models at global, European and regional/local scale, we select: i) 13 GMPEs for shallow active crustal regions; ii) 4 for subduction zones to be applied to the Calabrian arc, and iii) 2 for volcanic areas, specifically for Mount Etna. The most popular ranking schemes for GMPEs are based on residuals, calculated as the logarithmic differences between observations and predictions. Several scoring techniques have been applied, included the renowned method based on the log-likelihood value (Scherbaum et al. , 2009) and two novel methods commonly used for evaluating general probabilistic forecasts, such as the gambling score (Zechar and Zuang, 2014) and the scoring rule for quantiles (Gneiting and Raftery, 2005). The latter method has been applied for the first time to GMPEs and seismological data in general. Results for Active Crustal Regions (ACRs). The dataset has been derived starting from those used to derive the GMPEs by Bindi et al. (2011), excluding the analog records of low sampled events or without style of faulting and adding the events with Mw≥5 since 2009 and the well-sampled events (more than 15 records) with 4≤Mw≤5 since 2009 uniformly distributed on the national territory. We increased the data set in order to match the maximum magnitude foreseen for MPS16 with well-sampled foreign events with Mw up to 7.5 and known fault geometry. The dataset includes more than 4,000 waveforms recorded by more than 1,000 strong motion stations relative to 137 events. The location of the events and the magnitude-distance distribution of the dataset are shown in Fig. 1. For ACRs, we calculated the scores for 4 sets of data, derived by initial one: 1. ������� �������� ������ ����� dataset complete (named All); 2. ������� �� �������� ���������� �� � ��������� �� ��� ����� � dataset of stations classified as A according to EC8 (with V S30 larger or equal than 800 m/s, CEN 2003), named EC8-A; 3. ������� �� ������ ���� ��������� ������ ���� ��� ��� �������� ������� �� ������������ dataset of events with magnitude larger than 5.0 and stations located at Joyner-Boore distances lower than 50 km (named MR);