LFW2BBP: Broadband Ground-Motion Parameters Estimation Using Physics-Based Simulated Low-frequency waveforms and Deep Learning

Accurate prediction of broadband ground motion parameters is important for earthquake disaster prevention and mitigation. Due to lack of high wavenumber components of the source rupture process and the velocity models, physics-Based ground motion simulation methods can only produce reliably low-frequency ground motions (<1 Hz). In this study, we developed a deep learning network, LFW2BBP, which maps physics-based simulated low-frequency ground motion waveforms to broadband ground motion parameters. LFW2BBP extracts features of low-frequency ground motion in time domain waveforms, time-frequency domain spectrum and spectrum acceleration, and integrates these features to establish a relationship with high-frequency ground motion parameters. Sensitivity tests are conducted to verify the stability and robustness of the LFW2BBP. Finally, we combined physics-based simulation and LFW2BBP to predict broadband ground motion parameters for the 2016 Mw 7.0 Kumamoto earthquake. The predicted results show good agreement with the observations.