GNGTS 2022 - Atti del 40° Convegno Nazionale

GNGTS 2022 Sessione 2.1 201 ANALYSIS OF RESIDUALS FROM A SET OF GROUND MOTION MODELS IN THE NEAR SOURCE A. Chiecchio, R. Paolucci, M. Vanini Department of Civil and Environmental Engineering (DICA), Politecnico di Milano, Milano, Italy Introduction. In a probabilistic framework, ground motion prediction models (GMMs) use datasets of recorded groundmotion parameters at multiple stations fromdifferent earthquakes in various source regions to generate equations that are used to predict site-specific ground motions (Al-Atik et al. , 2010). Typically, these GMMs are based on observed records of past earthquakes, which describe the median and the standard deviation of the ground motion as a function of magnitude, distance, site, and other parameters (Villani and Abrahamson, 2015). During the last couple of decades, thanks to the advances made in terms of computing power and the development of newly published datasets, ground motion modelling had experienced a significant growth, with a range of predictor variables that are now used to describe the source, path, and site effect in detail. As a result, with an enormous number of GMMs being released in the recent years, some authors recommended criteria for the selection of models and suggested that these criteria could be used as a step to guide the publication of new ones (Douglas and Edwards, 2016). Furthermore, GMMs are a key pillar of a probabilistic seismic hazard analysis, and their true performance can be understood completely just when they are used in practice. In order to provide an overview of the performance of several GMMs, this paper has tested a selection of models against a recently published dataset of near-source strong motion records: NESS2, which is the second version of the Near-Source Strong motion flat-file compiled by Sgobba et al. (2021) and available at http://ness.mi.ingv.it/ (Last access: 27/04/22). The analysis was made to consider the records in those conditions that have been demonstrated to dominate the hazard in the high seismically active regions of Italy, for the return period of upmost interest in seismic design (see Barani et al. , 2009), therefore testing the performance of GMMs in a range of magnitudes and distances typically poorly constrained by calibration datasets. For a detailed description of the method used and of related results, reference can be made to Paolucci et al., 2022. Data and analysis. The GMMs selected for the analysis are largely used ones, from the pioneering ground motion model of Ambraseys et al. (1996), to the models calibrated in the Italian context (e.g., Bindi et al. , 2011) and to the most recent GMM proposed by Kotha et al. (2020). Table 1 shows the GMMs selected, with their acronym, their distance metric (R metric) and their range of applicability (M-R). Note that wherever not specified, M is as a moment magnitude. The NESS2 dataset consists of manually processed waveforms collected according to specific criteria, with a M w ≥ 5.5 and containing information about events, source parameters and high-quality metadata. The records of NESS2 used in this work were selected according to these conditions: (i) a Joyner and Boore source-to-site distance within 20km ( R jb ≤ 20km), (ii) a moment magnitude ranging from 5.4 to 7.1, (iii) a soil class classified as EC8 class B (still soil conditions, CEN 2004). Fig. 1 shows the (M w – R) distribution of the NESS2 dataset and the ranges of the records used in this study, highlighted with the blue rectangle. With the purpose of making a quantitative comparison between records and GMmodels, a residual analysis was performed at four selected periods of interest in seismic design: 0s (PGA), 0.5s, 1s and 2s. According to their definition by Strasser et al. (2008) and Al-Atik et al. (2010), total residuals were defined as the difference between the natural logarithm of records and the corresponding prediction from a specific GMM. In order to have a better understanding of the possible misfit of a given GMM with respect to records, residuals were then split into a