Leveraging deep learning for denoising DAS recordings in urban volcanic areas

Distributed Acoustic Sensing (DAS) has emerged as a transformative technology in the field of geophysics. Among its notable advantages stands out the ability to leverage existing fibre-optic telecommunications infrastructure to obtain high-quality seismic recordings with unprecedented spatial and temporal resolution. This feature makes the DAS system particularly well-suited to densely populated urban areas, where the deployment of traditional seismic arrays is often hindered by prohibitive costs and logistical complexities. However, the proximity of commercial cables to human activity introduces significant challenges, as anthropogenic noise—arising from transportation, industrial, and construction activities—frequently masks target seismic-volcanic signals, severely degrading the signal-to-noise ratio.

With the purpose of cleaning the DAS signal, we propose a deep learning-based approach for denoising of DAS data. We have developed a specialized neural network architecture and an ad-hoc training strategy designed in order to remove man-induced interference while preserving the essential characteristics of the seismic-volcanic signal waveforms.

The findings demonstrate the efficacy of the proposed model in enhancing signal clarity, underscoring its potential as a robust pre-processing tool to facilitate and refine subsequent DAS signal analysis in complex, noise-rich environments.