Improving the quality of high-frequency surface waves retrieved from ultrashort traffic-induced noise based on eigenvalue selection

Stacking cross-correlations of time windows from continuous long-duration noise data is an effective solution to improve the quality of retrieved high-frequency (>1 Hz) surface waves and the accuracy of dispersion energy. The observation duration, however, is usually limited due to traffic control, making it difficult for ambient noise sources to fulfill the requirement of uniform distribution. Additionally, strong human-related noise sources exist near survey lines deployed along urban roads, which often act as interfering sources, such as local noise sources located in the non-stationary-phase zones. Local noise sources cause spurious arrivals in cross-correlations, degrade signal-to-noise ratio (SNR) of retrieved surface waves and distort their dispersion energy. To attenuate these adverse effects and improve the quality of surface waves retrieved from ultrashort noise data, we perform the eigendecomposition technique on the cross-spectral density matrix (CSDM) and apply a Wiener filter on the decomposed eigenvectors. The correct eigenvalues and the corresponding filtered eigenvectors are selected to reconstruct the CSDM related to stationary-phase sources based on the matched-field processing outputs. This procedure significantly suppresses the backpropagated signals and efficiently recovers surface waves by improving the contribution of the stationary-phase sources. We validate our scheme on a synthetic test and two practical applications and show that we obtain higher-SNR virtual shot gathers and higher-quality surface-wave dispersion images compared to seismic interferometry. Our scheme can be a new alternative technique to conduct passive seismic surveys in densely populated urban environments without being affected by local noise sources.