Improved subsurface imaging using urban ambient noise DAS recordings

Distributed Acoustic Sensing (DAS) technology can convert unused fiber-optic cables of existing telecommunication networks (dark fibers) into arrays of virtual seismic receivers. Moreover, the seismic waves generated by human activities recorded on these receiver networks can be used to seismically image the urban subsurface at high resolution with a small footprint. This capability can help to evaluate the potential of the urban subsurface for safe and sustainable utilization in numerous applications, such as groundwater management and also geothermal development of an area. However, extracting coherent seismic signals from the complex urban seismic noise remains challenging due the uneven distribution of urban noise sources and often uncertain deployment conditions and resulting coupling of dark fiber.

We present an enhanced ambient noise interferometry workflow designed to identify and enhance coherent surface waves in complex DAS urban seismic noise data. The workflow is applied to urban seismic noise, predominantly generated by traffic, recorded on a dark fiber located along a major urban road in Berlin, Germany. Our workflow comprises a standard interferometric approach based on cross-correlations to retrieve coherent seismic phases for each hour of recording (Virtual Shot Gathers, VSGs), followed by Multichannel Analysis of Surface Waves (MASW) to derive 1D velocity models along consecutive and overlapping portions of the array. The individual 1D velocity models are then merged into a pseudo-2D velocity model of the subsurface. Our results are improved by incorporating a scheme to select VSGs using clustering driven by unsupervised machine-learning. This approach effectively excludes transient and localized noise sources while retaining high-quality VSGs. Additionally, a coherence-based enhancement technique is applied to stacked VSGs to improve their signal-to-noise ratio and, consequently, enhance the quality of the resultant dispersion curves. Ultimately, the resultant 1D velocity models achieve an increased investigation depth and their interfaces correspond well with available lithologic information from boreholes and models for Berlin. Our enhanced workflow yields more reliable results requiring less data than conventional processing schemes, thus fostering reduced acquisition costs and thereby more efficient investigations of the urban subsurface.