← Back to Gallery

Session: Earthquake Characterization Using Fiber-optic Cables [Poster]

Type: Poster

Date: 10/7/2024

Time: 05:00 PM

Room: Stanley Park Ballroom

Unmasking Traffic Noise: Unsupervised Denoising for Distributed Acoustic Sensing (DAS) Data

Recent advances in sensing technologies, particularly Distributed Acoustic Sensing (DAS) have enhanced the collection and analysis of seismological data, and DAS has emerged as a powerful method for detecting vibrations from earthquakes, subsurface phenomena, and other esoteric sources. However, the vast amount of data produced by DAS necessitate the use of sophisticated analytical methods to differentiate between signals of interest, such as those originating from earthquakes and vehicles.

We introduce a novel approach by extending the Noise2Self framework of Batson and Royer to effectively remove unwanted, structured coherent noise, particularly traffic signals, from DAS data. By creating a set of masks based on the first-order characteristics of traffic signals, we isolate and preserve earthquake signals while maintaining the denoising performance of the original Noise2Self approach, which efficiently reduces noise without requiring clean reference data. For a comprehensive evaluation of our approach, we employed synthetic data, which was generated using seismic recordings from closely spaced seismometers. We then applied our approach to data gathered from a DAS array located near Haast, New Zealand, adjacent to the Alpine Fault. For a comprehensive evaluation of our approach, we employed synthetic data, which was generated using seismic recordings from closely spaced seismometers, to validate our method under controlled conditions and afterwards applied this technique to data gathered from a DAS array located near Haast, New Zealand, adjacent to the Alpine Fault.

The results demonstrate that our model successfully removes traffic noise, as well as other non-coherent noise while maintaining the integrity of seismic signals, leading to an improvement in both Signal-to-Noise Ratio and waveform coherence. Evaluations on real-world DAS data further substantiate the robustness of our method, positioning it as a valuable tool for the analysis of large-scale DAS datasets across a range of geoscientific applications, working towards near or real-time monitoring.

Presenting Author: Sebastian

Additional Authors