Developing Seismo-tectonic Regime Wise Ground Motion Prediction Models for Bangladesh

Being surrounded by the Himalayan Frontal Thrust to the north, Dauki and Naga system to the north-east, Arakan Megathrust, and associated Chittagong Fold Belt to the east, the seismo-tectonic activity of Bangladesh is much prominent. Due to the lack of regionally established Ground Motion Prediction (GMP) Models, global GMP models are practiced for seismic hazard analysis. As there are regional differences in source, site and attenuation characteristics, the exact scenario might not get manifested with global GMP models. This study aims to develop seismo-tectonic regime-wise Ground Motion Prediction models for Bangladesh. Due to insufficient strong-motion recordings, developing GMP models was not an easy task. Therefore, diverting from the traditional approaches, this study intends to utilize USGS earthquake catalog data for ground motion modeling applying machine learning algorithms. However, GMP models have been developed for the four significant seismo-tectonic regimes of Bangladesh; Himalayan Frontal Thrust (HFT) Zone, Dauki Fault Zone, Chittagong-Tripura Fold Belt (CTFB), and Stable Continental Crust (SCC) Zone. Artificial Neural Network with Bayesian Regularization function has been designated for developing the GMP models, as it provided more accurate prediction among all the machine learning algorithms, when validated with strong motion data. The developed GMP models are compared with the global GMP models for a particular earthquake scenario of each zone. All the proposed models showed a deviation of prediction with the global models at a shorter distance and have been reduced as distance increased, which can be explained by regional parameter characteristics and model uncertainty. For HFT Zone, the proposed model provided a close approximate prediction with Abrahamson et al. (2014) among the five NGA West-2 GMP models for the 2018 Gorkha earthquake. In the Dauki zone, our model showed well harmonization with Campbell & Bozorgnia (2014) for the 1930 Dhubri earthquake. For the CTFB regime, our model followed the trend of a smooth decline in PSA with higher distance like the model of Atkinson & Boore (2003), but the deviation of prediction is lower with Zhao et al. (2006). Therefore, our proposed models are able to predict the ground motions considering the regional tectonic attributes of Bangladesh.