High-resolution seismic scattering imaging for urban underground infrastructure mapping

Abstract

Urban redevelopment often encounters information gaps due to the lack of clear records on previously built, poorly documented infrastructure. Construction activities are especially vulnerable to unintentionally damaging small-scale underground structures (such as cables, pipes, and conduits) that are concealed within complex subsurface layers or masked by electrical and electromagnetic interference. This study integrates a series of data processing techniques with an unsupervised machine learning method, Gaussian Mixture Model (GMM), to accurately detect both the horizontal and vertical locations of underground openings. The results demonstrate that the Radon-transformed data combined with GMM clustering effectively capture the horizontal locations of underground openings and identify reference seismic sources, while the velocity semblance analysis and the cross-correlation method reliably determine their vertical positions. Additionally, the Devito simulation provides a clear interpretation of scattered wave generation and propagation, highlighting the challenges in determining the vertical locations due to scattered waves predominantly originating from the upper boundary of the openings. Our findings emphasize that selecting an appropriate frequency range is critical for reliably detecting small-scale openings. Finally, the method is applied to detect a jet fuel pipe, showcasing the performance of this method in detecting small-scale underground infrastructure.