GNGTS 2017 - 36° Convegno Nazionale

288 GNGTS 2017 S essione 2.1 proper description of the aleatory variability and epistemic uncertainty, and the validation of the seismic hazard model. In the following we describe some key features and by-products of this activity. We have released the new historical earthquake catalogue that updates the previous versions. This catalogue, named CPTI15 (http://emidius.mi.ingv.it/CPTI15) , contains more than 4500 records, from 1000AD to 2014, that have been compiled with homogeneous criteria for location and magnitude of each event; for the pre-instrumental period, the moment magnitude has been assessed on the basis of the distribution of macroseismic intensities, now available for every earthquake in the Italian territory. This new catalog represents a major improvement because it has more realistic magnitude distribution with respect to the previous versions, and it removes some unrealistic temporal trend of the seismicity, which was observed in previous versions of the catalog. The scientific community has produced 12 new earthquake rate models that cover the national scale and 1 specific model for the Mt. Etna volcanic area. The 12 models have been grouped according to the (main) typology of seismogenic sources: areas (5 models), faults (2 models) and points (5 models). Among them we have also two innovative models mostly (or entirely) based on measures of ground deformation. At the moment of writing this document, modelers are exploring the epistemic uncertainty of their model through simple logic tree schemes. As regards the ground motion prediction equations (GMPEs), a first selection of the many GMPEs published in literature has been performed taking into account the following basic requirements: i) the GMPEs should uniformly cover the Italian territory and surrounding areas (with exception for volcanic zones); ii) rocky soil conditions and flat topography have to be considered (although hazard maps for different soil classes could also be released); iii) the hazard has to be estimated in terms of peak ground acceleration, velocity and displacement (PGA, PGV, PGD), response spectra in acceleration, velocity and displacement, macroseismic intensity and possible additional parameters, such as Housner or Arias intensity; iv) horizontal motion has to be represented by the geometrical mean of the horizontal components; v) GMPEs have to estimate 12 spectral periods, in the range 0.05-4 seconds. The forecasting performance of each GMPE has been evaluated through the comparison with accelerometric records available in the Italian (itaca.mi.ingv.it ) and European (esm.mi.ingv.it) strong-motion databases and with Italian macroseismic data (DBMI15; http://emidius.mi.ingv.it/DBMI15) . This step is essential to score each model in a quantitative way. Finally, we have introduced an innovative robust testing phase for the final model, and each of its components. In common practice, the reliability of a seismic hazard model is based on the consensus of the scientific community on the model’s outcomes. A more objective approach is based on testing the seismic hazard model, or its components such as the earthquake rate models and the GMPEs, using the available data. Despite the obvious importance of this step if we want to keep seismic hazard into a scientific domain, there is not yet a commonly accepted framework for testing seismic hazard models, and only a few approaches based on different assumptions have been proposed. The role of subjective expert opinion, the heterogeneous probabilistic approaches used to build the models, and the scarcity of independent data for evaluation make this issue extremely challenging, both from a theoretical and a practical point of view. We have used a recently develop probabilistic framework, which allows a meaningful testing and validation of a seismic hazard model, accounting for properly the epistemic uncertainty. The testing phase is composed by two steps: first, we evaluate the consistency of the model, i.e., if the model describes well the data; second, we score the models, i.e., we compare different models that form the new seismic hazard model to rank them according to their retrospective forecasting (hereafter pastcasting) performances. The goal of scoring is to facilitate the merging of models in a final seismic hazard assessment; in fact, the final weight of each component of the hazard model is assigned accounting for its scoring, and the outcomes of an experts’ elicitation session with independent national and international experts.