Parameter calibration and broadband ground motion simulation based on bayesian optimization—a case study of the 2023 Türkiye earthquake doublet

Abstract

Effectively conducting rapid ground motion field simulations under the uncertainty of source and propagation path parameters following an earthquake poses a significant challenge for assessing earthquake disaster losses. This study presents an efficient broadband ground motion simulation method introducing Bayesian hyperparameter optimization into the stochastic finite fault model. This new method iteratively corrects and optimizes source and path parameters through Bayesian optimization to reduce parameter uncertainty in ground motion simulation. Therefore, it can obtain optimal ground-motion estimation while calibrating regional seismic parameters. Firstly, using suggested method, accelerations at 12 stations are simulated for the 2023 Türkiye earthquake (M_w7.5). Then, by comparing with the records of 12 stations, the effectiveness of the method is not only verified, but also the regional key seismological parameters are calibrated to reduce uncertainty, including stress drop and attenuation coefficient, etc. Finally, the calibrated parameters are used in the simulation of the earthquake in Türkiye (M_w7.8) and the ground-motion field is well reproduced. The results show that with the help of efficient black-box exploration capabilities of Bayesian optimization, the established new method can not only be used for the calibration of seismological parameters in areas lacking earthquake data, but also can efficiently simulate the ground motion field that conforms to the regional geological environment. Therefore, it is suitable for rapid assessment of post-earthquake disasters.