Vertical ground motion model for the NGA-West2 database using deep learning method

Vertical-component of ground motions (GM) plays a significant role in seismic hazard analysis, especially for long-span structures and high-rising buildings. The former is usually predicted by empirical ground motion models (GMMs) that are developed on the basis of a preset function form and thus intensely depend on re-searchers' choices and prior knowledge. To overcome this issue, a deep learning-based GMM to predict the vertical component of GMs' IMs is developed in this study. 20,651 GM recordings are selected and divided into training, validation, and testing dataset based on the Next Generation Attenuation-West2 Project (NGA-West2). Comparative assessments with existing models are introduced on predicting performance indicators, IMs' dis-tribution with respect to seismic parameters, residuals, and variabilities. It can be concluded that the proposed model possesses better predictive power than the compared models. Meanwhile, sound physical features (e.g., magnitude scaling effects and near-fault saturation) can be observed.

Automated summary

A deep learning-based ground motion model is developed in this study to predict the vertical component of ground motions. Comparative assessments with existing models show that the proposed model has better predictive power and captures important physical features.