A recurrent-neural-network-based generalized ground-motion model for the Chilean subduction seismic environment

This paper proposes a deep learning-based generalized ground motion model (GGMM) for interface and intraslab subduction earthquakes recorded in Chile. A total of ∼7000 ground-motion records from ∼1700 events are used to train the proposed GGMM. Unlike common models (GMMs), which generally consider individual intensity measures such as peak acceleration spectral accelerations at given structural periods, GGMM is based on data-driven framework that coherently uses recurrent neural networks (RNNs) hierarchical mixed-effects regression output cross-dependent vector 35 (denoted IM). The IM includes geometric mean Arias intensity, velocity, acceleration, significant duration Iageom, PGVgeom, PGAgeom, D5-95geom, respectively), RotD50 31 periods between 0.05 5 s % damped oscillator Sa(T)). inputs include six causal seismic source site parameters, including fault slab mechanism, moment magnitude, closest rupture distance, Joyne-Boore soil shear-wave hypocentral depth. statistical evaluation shows high prediction power with R2 > 0.7 most IMs while maintaining cross-IM dependencies. Furthermore, carefully compared against two state-of-the-art Chilean GMMs, showing leads better goodness fit all Sa(T) considered GMMs (on average 0.2 higher R2). Finally, implemented select hazard-consistent motions nonlinear time history analysis sophisticated finite-element 20-story steel special moment-resisting frame. Results this statistically those selected conditional spectrum (CMS) approach. In general, it observed drift demands computed using approaches cannot be similar demands.