Improving the Extraction of Seismic Signals from Ambient Noise Cross-correlations using CWT based Denoising

Abstract

Ambient noise tomography technique to image the shallow structure of earth has evolved significantly in the past two decades. Continuous ambient noise records are cross-correlated between a pair of recording stations for a long period about a year or so, to obtain the virtual seismogram. The virtual seismogram obtained in this manner is comprised of surface waves which are the signals of interest and also the noise. Despite the number of techniques developed in the literature, the extraction of a clean virtual seismogram from noise cross-correlation functions is still an open research issue. The quality of virtual seismogram affects the dispersion measurements (group velocity versus period) of the surface waves. These dispersion measurements further impacts the resolution of earth tomographic structure by probing into the depth-seismic wave velocity profile of earth. In this paper, an attempt is made to denoise the virtual seismogram using continuous wavelet transform (CWT) based denoising technique and improve its quality by enhancing signal to noise ratio (SNR). The virtual seismogram estimated from noise correlation functions is decomposed into CWT coefficients at different scales and these coefficients are thresholded at each scale. The threshold value is calculated at every scale using universal threshold method. The enhancement of surface waves present in the virtual seismogram is evaluated by signal to noise ratio (SNR) measurements. The SNR of denoised linearly stacked noise cross-correlations is compared with the SNR of linear and phase weighted stacks of noise cross-correlations.