Enhancing High-frequency Ambient Noise for shallow subsurface imaging using urban ambient noise DAS recordings

Distributed Acoustic Sensing (DAS) enables unused fiber optic cables in existing telecommunication networks, known as dark fibers, to function as dense arrays of virtual seismic receivers. Seismic waves generated by human activities and recorded by dense sensor networks provide an abundant, high-frequency energy source for high-resolution, non-invasive imaging of the urban subsurface. This approach enables detailed characterization of near-surface soils, sediments, and shallow geological structures with minimal surface impact, supporting applications such as groundwater management, site response and seismic amplification analysis, seismic hazard assessment, geothermal development, and urban planning. However, extracting coherent seismic signals from complex urban noise is challenging due to uneven source distribution, uncertain fiber deployment conditions, and variable coupling between the fiber and the ground. In particular, high-frequency range signals (e.g., above 4 Hz), needed to resolve shallow subsurface structures, are particularly difficult to recover. Two strategies can be used to address some of these challenges, by discarding poor quality seismic noise segments or by focusing on particularly favorable noise sources. In this study, we adopt the second approach and use vibrations generated by passing vehicles, particularly trains which are energetic sources that contain valuable high frequency information . Capturing and exploiting the seismic waves generated by these vehicles offers unique opportunities for efficient and high resolution urban seismic imaging.

We present an enhanced ambient noise interferometry workflow designed to exploit noise sources that are particularly favorable to the fiber geometry, i.e. transient and strong sources occurring at the edge of the fiber segment to be analyzed. The workflow is applied to traffic-dominated seismic noise recorded on a dark fiber deployed along a major urban road in Berlin, Germany. First, we select short seismic noise segments that contain signals from passing trains and then apply a frequency–wavenumber filter to isolate the targeted train-generated surface waves while suppressing other wavefield contributions. The filtered data is then processed using a standard interferometric approach based on cross-correlations to retrieve coherent seismic phases from ambient noise, producing virtual shot gathers. Finally, Multichannel Analysis of Surface Waves is applied to derive one dimensional velocity models. This workflow targeted on specific transient sources reduces computational cost while enhancing dispersion measurements particularly at higher frequencies. By stacking the responses from tens of tracked vehicles, enhanced virtual shot gathers can be obtained and inverted to improve shallow subsurface models. This can be achieved with only a few hours of seismic noise recording, which is challenging using conventional ambient noise interferometry workflows.