GNGTS 2019 - Atti del 38° Convegno Nazionale

416 GNGTS 2019 S essione 2.2 applications, we have applied the “ANN2BB” approach presented in Paolucci et al. (2018) based on the training and use of Artificial Neural Networks (ANN). Such an approach has been demonstrated be able to preserve the full spatial correlation of ground motion, incorporating important physical features, such as directivity effects, radiation pattern, and 3D complex site effects also beyond the frequency limit of the numerical model. Fig. 2 shows and compares the results according the different PSHA studies, i.e. Physics- based against GMMs-based. More specifically the Peak Ground Velocity (PGV) hazard maps are presented and PGV hazard curves are obtained at some selected sites in order to highlight the main differences. The Chiou and Young (2014) empirical model, referred hereinafter as CHYO14, has been used to this end. It is worth noting that both the GAF and CHYO14 peak ground velocity maps have been obtained integrating over 3σ. Observing the Figures, it is evident, as expected, that the footprint and GAF approaches produce relatively consistent results while they differ significantly from the CHYO14-based estimates. Finally, under the simplifying assumption of a homogeneous buildings portfolio consisting of reinforced concrete low-rise buildings, a simple risk seismic analysis was carried out in order to evaluate the effects of the different PSHA approaches considered. The vulnerability curve is the one proposed by Özcebe et al. (2014). The results, presented in terms of relative loss with respect to the total sum insured, have been obtained assuming: (i) the entire portfolio concentrated at a single point (Fig. 3,a) and (ii) a distributed portfolio over the entire urban area (Fig. 3,b). Such a distinction allows us to infer how the different physics-based PSHA approaches take into account the spatial correlation, which plays a key role when a distributed portfolio of assets is considered. Indeed, in the first case (i) in which the entire portfolio is concentrated in a single point, the spatial correlation is totally neglected and, as intuitively, the trend of the loss curves is the same of the hazard ones (Fig. 2) in which the footprint and GAF results are consistent. The situation deeply changes in the second case (ii), where a distributed portfolio is assumed. In this case the GAF and footprint curves are different due to the fact that the formulation of the footprint PSHA methodology allows to naturally take into account the spatial correlation features of the ground motion differently from the GAF that is not able to incorporate it unless it is supplemented by a proper spatial correlation structure. Fig. 3 – (a) Loss curves assuming a portfolio concentrated at the site I1 indicated in Fig. 2 and (b) loss curves considering a distributed portfolio of building. The portfolio distribution is indicated in the top left panel. Acknowledgments. The authors are grateful to their co-workers Ilario Mazzieri and Ali Güney Özcebe. Part of this work has been carried out in the framework of the 2015-2017 agreement between Munich Re and Politecnico di Milano with the contribution of Alexander Allmann.