GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 2.1 GNGTS 2024 Figure 1: Durations' Correlations for Geometric Mean of the Horizontal Components with GMMs explanatory variables Residuals were calculated, which represent the difference between the observed durations and the duration predicted by a specific model. It shows how well or poorly a model’s predictions are close to the observed duration. After observing the duration, the predicted duration was calculated according to the models given by Bommer et al., (2009), Afshari and Stewart (2016), and Tafreshi and Bora (2023). Bommer et al., (2009) model and study were based on a strong motion dataset extracted from the database compiled for the NGA-West project (Chiou et al., 2008), whose main purpose is to calibrate GMM for active crustal earthquakes in the Western US. The dataset used consists of 2406 records from 114 earthquakes with moment magnitudes in the range from 4.8 to 7.9. Afshari and Stewart (2026) model was based on the NGA-West2 project (Ancheta et al., 2014), and it was reduced to 11,284 duration pairs of duration parameters for two as-recorded horizontal components with a Mw range between 3.0-7.9 events. Tafreshi and Bora (2023) model was calibrated on an Iranian strong motion database. The dataset used consists of 1749 records from 566 events with a Mw range from 3 to 7.5, recorded at 338 stations from 1976 to 2020. The total residual was calculated for each record by subtracting the logarithmic of the observed duration from the logarithmic of the predicted duration. Logarithmic was used since the durations are log-normally distributed. The distribution showed that all the models on average underestimated the observed durations, but the model by Afshari and Stewart (2016) has a better median performance and lower associated variability (Figure 2). These results suggest that the functional form proposed by Afshari & Stewart (2016) is a good starting point for the calibration of a new predictive model for duration in Italy.