Creating a Comprehensive Cryoseismic Catalog at Rhonegletscher: A Scalable Approach Using Distributed Acoustic Sensing and Machine Learning

Monitoring glacier dynamics is essential for understanding climate change impacts, safeguarding water resources, and protecting communities from related hazards. Distributed acoustic sensing (DAS) provides a unique opportunity to observe these dynamic environments with high spatial and temporal resolution. However, creating comprehensive seismic event catalogs from DAS data requires the development of efficient, automated tools. In this study, we used DAS technology, machine learning, and cloud computing to generate a detailed catalog of cryoseismic events at Rhonegletscher, Switzerland. We developed a robust preprocessing pipeline to address challenges posed by noise, coupling inconsistencies, and large data volumes. Our feature extraction is based on covariance matrix analysis, which allowed us to characterize wavefield properties using the first eigenvalue, coherency function, and eigenvalue variance. We compared unsupervised and supervised approaches, evaluating their relative effectiveness in detecting cryoseismic events across noisy DAS datasets. Although unsupervised methods provided valuable insights into inherent data patterns, their performance was hindered by dataset imbalance and noise. In contrast, supervised methods, trained on manually labeled data, demonstrated higher classification accuracy and reliability, with random forest emerging as the top performer. By combining cloud and parallel computing, we develop a scalable framework that streamlines DAS data analysis and supports future operational monitoring and research on glacier dynamics. The resulting comprehensive catalog of cryoseismic events provides a valuable resource for scientists, fostering advancements in cryosphere monitoring and hazard assessment within the context of a changing climate.