Processing Methods for Extracting Subsurface Information from Ambient Seismic Noise

Abstract

The ambient seismic wave field carries information about the subsurface in the vicinity of the recording site. It also carries a lot of less useful information about anthropogenic activities nearby. Potentially interesting variations in the naturally occurring seismic background wave field happen at extremely low power spectral density levels (typically between -120 to -180 dB [w.r.t. 1 m/s]), and is oftentimes masked by noise of anthropogenic origin. In addition, the spatial, temporal and frequency-domain variability of cultural noise often exceeds the variations of the natural background wave field. It is therefore critical to remove any influence of cultural noise from the records before an attempt can be made at analyzing the ambient seismic noise level with respect to any meaningful signatures of subsurface variations. Here, we present methods to characterize the ambient seismic wave field recorded with broadband seismometers and propose methods for extracting subsurface-related information. The analysis includes the characterization of spectral signatures of different types of sources (both anthropogenic and natural origin), as well as spectral signatures that are indicative of the subsurface underneath the recording site. The characterization also drives the selection of certain noise-removal techniques. Due to the typically small amplitude of these subsurface-related variations, it is critical to remove as much of the surface energy as possible, and to correct for near surface effects in the recorded data. Analysis of the cleaned data then allows looking for special attributes that may carry subsurface information. A possibly important tool for achieving these goals is the ratio between vertical and horizontal components (V/H) in the frequency domain. In this domain, the receiver terms contain information about the shallow subsurface, but sometimes also information about fluid content in the deeper subsurface. Both pieces of information can be of use in practical applications. For illustration, we present examples from low-frequency passive seismic (LFPS) surveys with particularly strong anthropogenic noise contaminations. We show that, despite remnant contamination of the records by anthropogenic noise, we observe statistically significant variations of spectral attributes that can be used for subsurface characterization.