An Early Forecast of Long‐Period Ground Motions of Large Earthquakes Based on Deep Learning

Abstract

Long-period (LP) seismic waves (approximately 2-10 s), radiated from the sources of large earthquakes, propagate over several hundred kilometers without much weakening due to their long wavelength and produce large-amplitude and long-duration shaking in distant basins. Large-scale structures, such as skyscrapers and oil storage tanks, might resonate with this LP ground motion and suffer damage. During the 2003 Tokachi-oki earthquake (Mw 8.0), the floating roof of an oil storage tank in the Yufutsu basin of Hokkaido, 150 km from the epicenter, was sloshed by LP ground motion, resulting in damage and a fire (Koketsu et al., 2005). When the 2011 Off the Pacific coast of Tohoku earthquake (Tohoku-oki earthquake, Mw 9.0) occurred, high-rise buildings in the Kanto (Tokyo) and Osaka basins, more than 400-800 km away, were severely shaken, causing facility damages such as falling ceiling material and tangled elevator cables (Building Research Institute, 2012). For safety, a system to forecast and warn of the occurrence of LP ground motion is needed in large basins, using the lead time of the shaking arrival from distant large earthquakes.In this study, we applied deep learning technologies for immediate forecasting of distant LP ground motions from seismic observations near the epicenter, anticipating the feasibility of a computing facility to analyze real-time observation data. Previous studies based on deep learning have successfully predicted peak ground accelerations (PGA) and peak ground velocities (PGV) of distant target locations (e.g., Derras et al., 2012; Zhang et al., 2021) and the distribution of shaking intensity over a wide area (e.g., Kubo et al., 2020;Lilienkamp et al., 2022; Seo et al., 2022) using immediately estimated source parameters (magnitude (Mw), location, and depth) from observed waveforms. Research is underway to forecast the PGA, PGV (e.g., Jozinović et al., 2020Jozinović et al., , 2022, and velocity response spectrum (Taya & Furumura, 2022) of target locations from observed waveforms without estimating source parameters.

Recent advances in earthquake seismology using machine learning

“…Tamhidi et al (2022) proposed an approach where ground-motion time series at target sites are constructed from a set of observed motions using a Gaussian process regression, which treats the real and imaginary parts of the Fourier spectrum as random Gaussian variables. Furumura and Oishi (2023) developed a DL approach for early prediction of long-period ground motions at a target point far from the earthquake source based on waveform observations near the source. This approach could effectively forecast long-period ground motions of large earthquakes regarding amplitude, waveform envelope shape, spectral characteristics, and duration.…”