Combining DAS Ambient Seismic Noise Interferometry and Unsupervised Machine Learning for 2D Imaging of an Urban Subsurface

Abstract Applying ambient noise interferometry to distributed acoustic sensing (DAS) data recorded along telecommunication networks offers a promising way to image the urban subsurface with high resolution at reduced costs. However, extracting coherent seismic signals from DAS in urban environments can be challenging due to diverse, unevenly distributed noise sources, which can distort interferometry results and produce spurious signals, complicating analysis and interpretation. To address these complexities, we have developed a modified ambient noise interferometry workflow that allows for the efficient selection of high-quality data. We analyze 15 days of continuous passive DAS data collected from a pre-existing 11 km long dark fiber-optic cable running along a major urban road in Berlin, Germany. After retrieval of virtual shot gathers (VSGs) using the cross-correlation method, we develop a selection strategy to identify high-quality data through unsupervised clustering. Next, we stack the VSGs within each cluster and choose the highest quality, stacked VSG based on the quality of the corresponding dispersion spectra. We initially test the clustering method on synthetic VSGs to ensure its effectiveness. The clustering results identify distinct groups of VSGs that exhibit consistent patterns in both synthetic and real VSGs. These distinct groupings offer valuable insights into the temporal variations in human activities and allow a better interpretation and identification of viable ambient noise signals for further processing. Thereafter, multichannel analysis of surface waves is utilized to obtain 1D shear-wave velocity models for consecutive array segments. A 2D subsurface velocity model is then constructed through merging the individual 1D velocity models derived from overlapping array subsections. This approach enables efficiently selecting highest-quality data within massive, noisy recordings, which ultimately results in enhanced dispersion measurements and thus improved images of the urban subsurface.