GNGTS 2019 - Atti del 38° Convegno Nazionale

476 GNGTS 2019 S essione 2.2 ergodic standard deviation (i.e. the spatial variability at many sites is assumed identical to the variability at a single site; Anderson and Brune, 1999). This assumption implies higher level of uncertainty associated to the model predictions, due to the fact that ergodic GMMs are calibrated on geographical areas where more records are available, thus neglecting region- specific features of ground motion behavior. However, in case of site-specific PSHA purposes or engineering applications at local scale (i.e. loss assessment and risk analyses of structures and infrastructures) there is the need to improve the ground motion prediction performance. In these cases, the simplified assumption of ergodicity may be not reliable to describe regional properties of the ground shaking (i.e. magnitude scaling, distance attenuation characteristics and site effects). More accurate predictions can thus be computed by relaxing the ergodic assumption in favor of non-ergodic approaches, in which the repeatable terms of variability due to source-, path- and –site effects are used to provide region-specific corrections of the median predictions, as well as to transfer part of the aleatory variability into epistemic uncertainty (e.g., Rodriguez-Marek et al. , 2013; Villani and Abrahamson, 2015; Baltay et al. , 2017; Lanzano et al. , 2017). Following this concept, we propose a methodology for generating empirical shaking maps of the acceleration spectral ordinates based on a non-ergodic GMM calibrated on Central Italy, in which the systematic contributions of the variability are decomposed. The obtained corrective terms are then mapped by means of spatial correlation models to provide the local adjustments of the ground shaking, following the approaches proposed for California (Landwear, 2019; Sahakian et al. , 2019) and Emilia region in Italy (Sgobba et al. , 2019). We finally simulate the ground shaking empirically, by adding up the ground motion intensity field predicted by the GMM and the spatially correlated fields of the site (dS2S), source region (dL2L) and path (dP2P) effects computed at any point of a regular grid. Implementation of such a modelling clearly requires a dense dataset in order to compute robust estimation of the repeatable terms. For this reason, we focus our study on Central Italy, where a huge quantity of high-quality strong-motion records (more than 30.000 waveforms) has become available after the occurrences of significant events in the last 10 years. Results show peculiar spatial patterns of the site and path effects in the region, that can be related to physical aspects not fully captured by the GMM. The impact of the corrections on the shaking pattern and spectral intensity amplitudes is also shown through empirical simulations of the ground motion scenarios related to past earthquakes. Dataset. Dataset is composed by accelerometric and velocimetric earthquake signals, recorded by stations and events located in Central Italy since 2008 and including data of the 2009 L’Aquila and the 2016-2017 Central Italy sequences. It is composed by more than 30,000 records of about 450 earthquakes in the magnitude range 3.4 – 6.5 and more than 460 stations within 250 from the epicenters. The huge number of waveforms is due to the great effort made during the last seismic sequences in 2016 (Amatrice Mw 6.0 - VissoMw 5.9 - Norcia Mw 6.5) by the Italian Department of Civil Protection (DPC), the Istituto Nazionale di Geofisica e Vulcanologia (INGV) and several academic and research institutions that installed more than 100 stations with the aim of improving the event locations and monitoring the site effects (Priolo et al. , 2019; Cara et al. , 2019). Region-specific GMM. In order to calibrate an ad-hoc non-ergodic GMM in the study area, a mixed effect model is applied similarly to the one recently proposed for Italy (ITA18, Lanzano et al. , 2019), where the “fixed” effects account for magnitude scaling and for geometric and anelastic decay with distance, while the “random” effects are related to event, site, source and path. The median prediction of this GMM is assumed to be referred to the ground motion intensity level predicted for the reference sites (i.e. the sites characterized by a flat site response and an amplification factor around unity) identified according to the procedure by Felicetta et al., 2018.