Ground motion model for acceleration response spectra using conditional-generative adversarial network

The present study focuses on developing a ground motion model (GMM) for 5%-damped spectral acceleration (S_a) using a Conditional Generative Adversarial Network (C-GAN). Unlike traditional methods, the model incorporates the physics of source, path, and site characteristics into the adversarial training process between the generator and discriminator. The model is trained on a comprehensive dataset comprising 23,929 ground motion records from both horizontal and vertical directions, sourced from 325 shallow crustal events in the updated NGA-West2 database. The input parameters include the moment magnitude (M_w), Joyner-Boore distance (R_JB), the focal mechanism (F), hypocentral depth (H_d), average shear-wave velocity up to 30 m depth (V_s30), and the direction of S_a (dir). To ensure the model’s integrity, an inter-event and intra-event residual analysis is conducted, validating its robustness and unbiasedness. Additionally, the model’s performance is evaluated against established GMMs relevant to similar seismo-tectonic backgrounds. Moreover, the applicability of the developed model is demonstrated through the estimation of site-specific response spectra for Chi-Chi, Taiwan and Loma Prieta. Thus, the study contributes to advancing ground motion modelling techniques applicable in seismic hazard assessment and structural engineering practices.