GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 2.2 GNGTS 2024 original space of the intensity measures such that the output, for each target point, is a condiConal distribuCon. The input and output of the ConvCNP are expressed in log-units and thus assumed to be corrected for site effects: the effects of local geology are removed from the IMs recorded at the staCons and reintroduced into the esCmated IMs at the target points using the corresponding amplificaCon factor by Falcone et al. (2021). The choice of using the amplificaCon factors (instead of, for example, a Vs30-based approach) to address the site effects is double-fold: on one hand, it simplifies the ConvCNP process by operaCng on uniform inputs and outputs (which is especially useful since the encoder and decoder can seamlessly handle input and output points affected by different local effects); on the other hand, it improves the interpretability of the results by separaCng the contribuCon of local geology to the final results. The implemented MVN is based on the formulaCon by Worden et al. (2018) and the correlaCon funcCon by Loth and Baker (2013) is adopted: the choice of a correlaCon funcCon independent of the epicentral distance and the event magnitude is required to obtain a workflow no longer dependant on the evaluaCon of the source parameters. A flowchart of the hybrid method is shown in Fig. 1: Fig. 1 – ShakeRec-hybrid flowchart: real-Cme data are corrected for the site effects with an amplificaCon factor and passed to the ConvCNP (defined by the dashed line). IM values are esCmated both at the staCon locaCons and the target points (here represented by a regular grid) and the soil effect is reintroduced by the corresponding amplificaCon factors. These outputs and the real-Cme original inputs are then passed to the MVN to compute the ground-shaking maps.