GNGTS 2013 - Atti del 32° Convegno Nazionale

not so different one form the other. Furthermore, it is worth noting that performances of each modes change with the g 0 value of concern. For longer return times (larger g 0 values) Mod 8 and Mod 6 are the best performing ones, while for intermediate values, Mod 30 and Mod 33 are the most performing. This is not in contradiction with the adopted approach. In fact, by Eq. (1) one can see that scoring should be performed for each g 0 value in order to obtain a comprehensive hazard curve. Tab. 1 – Scoring results for the time-independent PSHA models available in the DPC-INGV-S2 repository (https:// sites.google.com/site/ingvdpc2012progettos2/deliverables/d1-1 ). For coding please refer to this site. Longer average return time correspond to the part of the hazard curve relative to higher PGA values. Model Likelihood Counting Test Average Return Time MOD 8 0.1378 Compatible 484 MOD 8 0.2707 Compatible 984 MOD 6 0.1378 Compatible 484 MOD 6 0.2707 Compatible 984 MOD 3519 0.0839 Compatible 475 MOD 32 0.1419 Compatible 475 MOD 30 0.0197 Compatible 94 MOD 30 0.1678 Compatible 284 MOD 30 0.1971 Compatible 475 MOD 28 0.0003 Underestimates 2475 MOD 28 0.0001 Underestimates 974 MOD 28 0.0002 Underestimates 475 MOD 28 0.0716 Compatible 30 MOD 27 0.0182 Underestimates 475 MOD 27 0.0182 Underestimates 475 MOD 22 0.0289 Compatible 2475 MOD 22 0.0853 Compatible 101 MOD 33 0.2157 Compatible 475 Conclusions. A coherent forma procedure has been delineated to provide probabilistic seismic hazard estimates that include aleatory and epistemic uncertainty. A key aspect of this procedure is the scoring of available PSHA models by considering both ex-ante (e.g., logic- tree) and ex-post (empirical testing) approaches. An application of this procedure has been applied in the frame of the DPC-INGV-S2 seismological research project for empirical scoring of a number of PSHA models available for Italy. To this purpose, a set of accelerometric sites operated in then interval 1979-2004 was considered for testing. The preliminary results obtained indicate that the likelihood estimates accompanied by other testing procedures are able to provide useful indications about performances of competing models and could represent a basic tool for driving new researches devoted to a best practice for hazard assessment. Acknowledgements. Many thanks are due to A. Gorini (DPC) and to F. Pacor and R. Puglia (INGV-Milano) for their effective support in the key aspect of selecting accelerometric sites to be used for testing. 14 GNGTS 2013 S essione 2.1