Spatial correlation analysis for ANN generated physics-based broadband response spectra: A case study for 2023 Turkey events

For the risk analysis of spatially distributed structures, the joint prediction of ground motion intensities at multiple sites is required. Therefore, many researchers have come up with spatial correlation models for seismic intensity measures (IMs). The spatial correlation model requires site-specific non-ergodic ground motions. One of the chosen approaches for correlation studies is physics-based simulations (PBS), which account for the complexities related to earthquake source, path and site. The present study evaluates the efficiency of the artificial neural network-based broadband response spectra generator (BBANN) of Sharma et al. (2023) in terms of residual analysis and spatial correlation. The Turkey earthquakes of 6th February 2023 are taken as the case study. The model-predicted values corresponding to a short period are compared with the recorded values for the events. We found that the model was able to capture ground motion trends without any bias in residuals. Also, in the spatial correlation scale, the model predictions are comparable with recorded values. The results highlighted the efficiency of the BBANN model in effectively capturing the spatial pattern of ground motion intensity measures. The study draws attention to the ability of PBS to generate non-ergodic ground motion and, hence, can be useful in seismic hazard and risk frameworks.