Spatial Interpolation of Seismic Waveforms Based on Gaussian Process Regression of Wavelet Coefficients

Abstract

Understanding the spatial distribution of seismic ground motion characteristics swiftly after major earthquakes is crucial for estimating the extent of damage in affected areas. In the immediate aftermath, when sufficient information about the seismic source process is lacking, the spatial distribution of seismic ground motion characteristics is assessed in the form of scalar measures, such as seismic intensity or maximum velocity, by interpolating the observed seismic motion data. However, for detailed damage assessment and safety evaluations of structures with complex features, such as railway bridges, it is desirable to evaluate dynamic oscillations using seismic waveforms rather than scalar measures. Yet, methods to estimate the spatial distribution of such seismic waveforms immediately after an earthquake have not been established. In this paper, we propose a method for spatial interpolation of seismic waveforms, using wavelet transforms of observed seismic motion and Gaussian process regression in the time-frequency domain. Additionally, we validate the effectiveness of this approach by applying it to past records of actual seismic events.