Enhancing Event Detection in Distributed Acoustic Sensing Data through Comprehensive Denoising

• DOI: 10.5194/egusphere-egu25-11352
• Publication Date: 2025-03-14T23:40:12Z
• Authors: Johanna Zitt; Patrick Paitz; Andreas Fichtner; Fabian Walter; Josefine Umlauft

Distributed Acoustic Sensing (DAS) data often involve large volumes but may be characterized by low Signal-to-Noise Ratios (SNR) compared to traditional seismic point sensors when terrain conditions hinder appropriate sensor coupling. While the vast amount of data increases analysis capabilities, the low SNR may bury signals of interest under the high environmental noise level emitted by a multitude of near-surface processes - in turn hindering detection and analysis capabilities. Effective denoising techniques are therefore the essential preliminary to uncover buried signals of interest and further to significantly increase the number of event detections.We deployed a DAS system on Rhône Glacier, Switzerland, using a 9-km-long fibre-optic cable spanning the entire glacier from its ablation to its accumulation zone. With the intention to detect and analyse icequakes, we collected continuous records during one month in July 2020, comprising 14 TB of strain rate data at 1000 Hz sampling rate. We define signals of interest to be coherent signals originating from either surface events (e.g., crevasse formation) or basal events (e.g., stick-slip motion), while noise originates from incoherent sources like water flow or meteorological forcings. We demonstrate that a self-supervised machine learning denoising technique can significantly improve the SNR of signals of interest, facilitating event detection. By comparing our approach to traditional denoising methods and other self-supervised machine learning techniques, we highlight its advantages in achieving significantly enhanced SNR values. Further, we quantify the impact of comprehensive denoising by evaluating the event detection performance of the classical STA/LTA triggering algorithm applied to both raw and denoised datasets. Our results underscore the critical role of denoising in improving data interpretability and enhancing event detection capabilities in cryoseismological DAS studies.