Array beamforming on ambient seismic noise correlations reveals repeating direct waves in the coda

Over the past two decades, the ambient seismic noise correlation method has revolutionized our ability to investigate the Earth's subsurface structure. Perfect reconstruction of the cross correlation wavefields towards Green's function demands some strong hypotheses about uncorrelated and spatially uniform seismic source distributions. In reality, violation of this assumption impacts the accuracy of the Green’s function estimate, which may bias applications in further studies. As this technique has become a standard method for subsurface imaging and monitoring, a methodology that identifies the effect of seismic source distributions plays an essential role in removing their contribution to achieve less-biased signals. Seismic array beamforming is commonly applied in estimating the direction of seismic waves crossing the array.

In this study, we use a new strategy based on beamforming of noise correlation signals. We consider several seismic stations surrounding the Gräfenberg array throughout Europe as virtual sources. We process two years of vertical component continuous noise recording from the Gräfenberg array in Germany and virtual sources in Poland, Italy, Portugal, France and Finland. Using the noise correlation-based beamforming method, we detect source directions for direct and coda waves for the primary (0.05-0.1 Hz) and secondary (0.1-0.4 Hz) microseism frequency bands. The source directions for the direct waves correspond to the converging and diverging part of the correlation wavefield. Throughout the coda, however, we detect the dominant noise source directions, i.e., surface waves generated by ocean microseisms in the Northern Atlantic during winter months and body waves from the Southern Pacific during summer months. This suggests that the coda of the correlation functions contains repeating direct waves from the dominant source regions, which may lead to incorrect estimates and interpretation of velocity variations if not accounted for. Knowledge of the ambient noise source origins and their spatiotemporal distribution is required to correctly interpret velocity variations.