The ambitious scopes of recent earthquake ground motion studies are generating a need for more high-quality ground motion records. As the number of deployed sensors is rapidly growing through improved accessibility and cost (e.g., ground motion stations, low-cost accelerometers, smart phones), an exponentially increasing amount of data are being generated. Previously, quality-assured ground motion data sets for engineering applications were generated using both manual and automated quality screening methodologies. More recently, new techniques have emerged that potentially offer both improved classification accuracy and computational expediency. This work presents a machine learning–oriented method to facilitate and accelerate the quality classification of ground motion records from small magnitude earthquakes. Feedforward neural networks are selected for their ability to efficiently recognize patterns and are trained on two New Zealand data sets. An application to physics-based ground motion simulation validation indicates that the proposed approach delivers results that are comparable with manual quality selection. Robust automatic ground motion quality screening allows a significant increase in data set size for development, calibration, and validation of ground motion models.

You do not currently have access to this article.