Real-Time Ground Motion Forecasting Using Front-Site Waveform Data Based on Artificial Neural Network

Real-time earthquake information made available by the Japan Meteorological Agency (JMA) publicly since October 2007 is intended to dramatically reduce human casualties and property damage following earthquakes. Its current limitations, however, such as a lack of applicability to near-source earthquakes and the insufficient accuracy of seismic ground motion intensity leave much to be desired. The authors have suggested that the forward use of front-site waveform data leads to improve accuracy of real-time ground motion prediction. This paper presents an advanced methodology based on artificial neural networks (ANN) for the forward forecasting of ground motion parameters, not only peak ground acceleration and velocity but also spectral information before S wave arrival using the initial P waveform at a front site. Estimated earthquake ground motion information can be used as a warning to lessen human casualties and property damage. Fourier amplitude spectra estimated highly accurately before strong shaking can be used for advanced engineering applications, e.g., feed-forward structural control. The validity and applicability of the proposed method have been verified using Kyoshin Network (K-NET) observation datasets for 39 earthquakes occurring in the Miyagi Oki area.