A Deep Learning 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 inslab subduction earthquakes recorded in Chile. A total of ~7000 ground-motion records from ~1700 events are used to train the GGMM. Unlike common ground-motion models (GMM), which generally consider individual ground-motion intensity measures such as spectral acceleration at a given period, the proposed GGMM is a data-driven framework that coherently uses recurrent neural networks (RNN) and hierarchical mixed-effects regression to output a cross-dependent vector of 35 ground-motion intensity measures (IM). The IM vector includes geomean of Arias intensity, peak ground velocity, peak ground acceleration, and significant duration, and RotD50 spectral accelerations at 32 periods between 0.05 to 5 seconds (denoted as Sa(T)). The inputs to the GMM include six causal seismic source and site parameters. The statistical evaluation of the proposed GGMM shows that the proposed framework results in high prediction power with coefficient of determination R² > 0.7 for most IMs while maintaining the cross-IM dependencies. Furthermore, it is observed that the proposed GGMM leads to better goodness of fit for all periods of S_a(T) compared to two state-of-the-art Chilean GMMs (on average 0.2 higher R²).