This study aims to present a relatively short list of interim induced proxy ground‐motion models (GMMs) most suitable for induced‐seismicity application in central and eastern North America (CENA). Induced proxy GMMs are models not established from datasets strictly made of induced events but can be used to predict ground motions from such events. For this purpose, we test the predictive power of a long list of GMMs against a dataset of induced earthquakes using the popular log‐likelihood (LLH) method of Scherbaum et al. (2009) and its natural extension, known as the multivariate logarithmic score of Mak et al. (2017). Our dataset is a subset of data provided by Rennolet et al. (2017) and is composed of 2414 time histories from 384 CENA induced events with hypocentral distances below 50 km and moment magnitudes from 3.5 to 5.8. Candidate GMMs are from two categories, including purely empirical models developed from the Next Generation Attenuation‐West2 (NGA‐West2) database and indigenous models of CENA. The NGA‐West2 database contains a large number of shallow small‐to‐moderate magnitude events from California that may approximate characteristic features of induced events in CENA. Some of the CENA models have considered near‐distance saturation for small‐to‐moderate magnitude range and/or have explicitly modeled source parameter as a function of focal depth that may make them reasonable induced proxy GMMs.
Some models performed better in certain frequencies than others, and not a single model performed the best over the entire frequency range. Overall, three models including Abrahamson et al. (2014), Chiou and Youngs (2014), and Atkinson (2015) GMMs outperformed other models. These models are not specifically established for CENA but are properly modeled for magnitude and depth scaling. In addition, stochastic models favored in the low‐seismicity region of CENA appear not to perform better than models developed based on conventional statistical and empirical approaches for induced‐seismicity applications. The result of this study can be useful in selecting a suite of appropriate GMMs for performing probabilistic seismic hazard assessment.