GNGTS 2023 - Atti del 41° Convegno Nazionale

Session 2.2 GNGTS 2023 recovery technique that makes use of Voronoi tessellation to obtain a structured grid representation of sparse data. Each ensemble member is composed of four convolutional layers, with zero-padding, with 12 filters of size and a final convolutional layer with a single filter of 5 × 5 size , each using a rectified linear activation function. 5 × 5 The ensemble takes as input (left side of Fig. 1) the Voronoi tessellation of the GMP at the stations, a map of the active station locations, and a normalised VS30 map (Michelini et al., 2020) and produces as output the mean ground-shaking field and its uncertainty (right side of Fig. 1) with µ σ a resolution of : the former is computed as the mean of the ensemble members’ 0. 05 ◦ × 0. 05 ◦ reconstructions and is comparable to the GMPE expected value at each point; the latter is computed as the combination of an aleatoric and an epistemic term. The offshore area is masked as it is poorly constrained and of little interest for the initial stage of emergency management. The current ShakeRec implementation is able to update the results every , approximately. 2 Figure 1. ShakeRec flowchart: real‐time and static data are used as inputs to the ensemble that outputs the average reconstructed ground‐motion parameters (GMP) field and its associated uncertainty (from Fornasari, Pazzi, and Costa, 2022). RESULTS The events analysed have been selected as they represent two interesting study cases with opposite characteristics in terms of position relative to the network: the Bargagli earthquake is a event with epicentre inland and good azimuthal coverage; the Marche earthquake is a 4. 1 event with epicentre approximately offshore and a one-sided azimuthal coverage. 5. 7 30 The final peak ground acceleration (PGA) reconstructions obtained with ShakeRec (using the data available after for the Bargagli event and for the Marche event) have been compared 40 35 with the results of ShakeMap computed by INGV and are shown in Figure 2. The stations whose parameters contributed to the reconstructions are indicated on the maps: the black markers indicate stations whose parameters have been used in the reconstruction. The ground-shaking maps for the Marche event are in overall agreement: the differences, both in the extension of the affected area and in the values on the coast, are imputable to the different