GNGTS 2023 - Atti del 41° Convegno Nazionale

Session 2.2 GNGTS 2023 Data and methodology. To extract the Macroseismic Data Points (MDPs) from DBMI15, we used a web service, which enables us to download intensity data encoded in a ShakeMap-compatible format. The dataset includes earthquakes that occurred between January 1117 and January 1968 (6.0 ≤ M ≤ 7.3). For each earthquake, to generate the shakemap set (Worden et al., 2020), we used the GMICEs by Oliveti et al. (2022) and the Ground Motion Model (GMM) by Bindi et al. (2011) proposed by Michelini et al. (2020). The selected events are illustrated in Fig. 1. We implemented an iterative cross-validation procedure to select the most appropriate between the Virtual IPE (VIPE hereafter) used by default in ShakeMap and the Italian-derived IPE developed by Pasolini et al. (2008) and subsequently recalibrated by Lolli et al. (2019) (Pea08 hereafter). The validation analysis (also known as leave-one-out analysis; Tomczak, 1998; Hofierka et al., 2007; Worden et al., 2010) selects a target earthquake and, iteratively, for each MDP, does the following: • removes the MDP from the dataset; • uses the ShakeMap procedure to predict the intensity at the removed MDP (i.e., while keeping all the others); • takes the difference between observed and predicted intensity values at the removed MDP. This procedure has been repeated for all the earthquakes included in the dataset. Then, the shakemaps have been computed using Pea08 and VIPE separately as input. For both IPEs, we used the same values for the source parameters, and the same configuration for GMMs, site effects and GMICEs. Results The results of the leave-one-out cross-validation for all the earthquakes are presented in Fig. 2 as residuals (i.e., differences between observed and predicted intensity values) obtained with VIPE and Pea08, respectively. In Fig. 2, we note that the median value of the residuals for both IPEs is around zero, whereas the standard deviation calculated using VIPE is lower than that achieved adopting Pea08. That is, the boxplots show that both configurations accurately predict the intensity at the observed points, but the residuals of Pea08 show more scatter of the data. When the data are grouped according to distance (Fig. 3a) and intensity value (Fig. 3b), we observe overall the same behaviour described for the entire dataset. However, if we focus our attention to intensities less than 4-5 for both VIPE and Pea08, the median value of the residuals leads us to conclude that ShakeMap slightly overestimated intensity values when compared to observed data. This behaviour is expected given that our dataset was compiled for earthquakes with a magnitude of at least 6.0, with very few data with intensity less than 4-5, principally located at large epicentral distances. Other factors, however, related to the calibration of Pea08 and the GMM by Bindi et al. (2011) to account for low shaking, cannot be excluded.