GNGTS 2022 - Atti del 40° Convegno Nazionale

GNGTS 2022 Sessione 2.2 311 BB-SPEEDSET: AN OPEN-SOURCE DATASET OF NEAR-SOURCE BROADBAND EARTHQUAKE GROUND MOTION FROM 3D PHYSICS-BASED NUMERICAL SIMULATIONS R. Paolucci, C. Smerzini, M. Vanini Department of Civil and Environmental Engineering, Politecnico di Milano, Milano, Italy Introduction. Empirical Ground Motion Models (GMM) (Douglas and Edwards 2016) and ShakeMaps (Wald et al., 2021) represent the reference tool of choice for seismic hazard and risk analyses, but, in spite of their considerable growth and ease of use within the performance- based earthquake engineering framework, they still have some drawbacks. First, the paucity of ground motion recordings in the proximity of the earthquake source still persists and prevents an accurate and robust prediction of the ground motion (and of its uncertainty) in the variety of source-to-site conditions which are typically found in the near-source region (Paolucci et al., 2022). Second, empirical GMMs typically reproduce the spatial correlation of ground motion through simplified approaches based on the stochastic simulation of spatially correlated random fields under the hypotheses of ergodicity, isotropy and stationarity. Such assumptions are hardly found in near-source conditions and may not be suitable to reproduce scenario- and region-specific features of ground motion spatial correlation and cross-correlation (Chen and Baker 2019; Infantino et al., 2021; Schiappapietra and Smerzini 2021). As in most fields of science, when the experimental observations are limited, numerical modeling may be employed to fill in the gap and gain insight where information from nature is scarce. In this perspective, with the advancement of high-performance computing (HPC) resources, physics-based numerical simulations (PBS) have started to play a role in providing region- and site-specific predictions of seismic shaking and they can be employed, in addition to or in place of observations, to shed light on the physics of near-fault ground shaking and on its spatial variability (McCallen et al., 2021). Based on the rigorous solution of the elastodynamics equation, PBS provides ground motion time histories reflecting the physics of the whole seismic wave propagation problem, from the fault rupture to the propagation path and local site response at shallow geology. With a long-lasting expertise gained (i) in the development of the open-source spectral element code SPEED (Mazzieri et al., 2013), (ii) in the advancement of techniques to enrich at high frequencies the PBS results (Paolucci et al. 2018), (iii) in the validation of PBS results against near-source ground motions recorded from different earthquakes in Italy and worldwide and in the application to several scenario case studies (e.g. Paolucci et al., 2015; 2021a; Smerzini et al., 2017, amongst others), the research group of Politecnico di Milano has constructed the BB-SPEEDset, an open-source dataset of near-source broadband accelerograms obtained from the source-to-site 3D numerical simulation of seismic wave propagation (Paolucci et al., 2021b). Overview of BB-SPEEDset. BB-SPEEDset has been constructed by assembling a large set of waveforms simulated by SPEED, in most cases validated against recordings of real earthquakes, and post-processed with an effective workflow apt to generate broadband accelerograms. The generation of broadband time histories starting from low-frequency SPEED results makes uses of a technique based on Artificial Neural Networks – ANN2BB (Paolucci et al., 2018; 2021b), trained on strong motion records. The dataset includes a total of 12058 three-component simulated waveforms from earthquake scenarios with moment magnitude (M w ) from 5.5 to 7.4 and Joyner-Boore distances (R jb ) up to 80 km, see Fig. 1. Strike-slip, normal and thrust events are included in the dataset. Most records refer to normal (50%) and strike-slip (41%) focal mechanisms, while only 9% is from thrust earthquakes. As regards the distribution of BB-SPEEDset with respect to site conditions,