Cryoseismic Event Analysis on Distributed Strain Recordings Leveraging Statistical Learning Methods

Cryoseismic Event Analysis on Distributed Strain Recordings Leveraging Statistical Learning Methods

Grimm, Julius (TU Delft Civil Engineering and Geosciences)

Paitz, Patrick (mentor)
Martin, Eileen (mentor)
Edme, Pascal (mentor)
Walter, Fabian (graduation committee)
Fichtner, Andreas (graduation committee)

Delft University of Technology
ETH Zürich
RWTH Aachen University

Applied Geophysics | IDEA League

2021-08-25

In the summer of 2020, ETH researchers installed a 9 km long fiber-optic cable on the Rhonegletscher (Switzerland), covering the whole glacier extent from accumulation to ablation zone.
The fiber was then interrogated with a DAS system, turning it into a distributed seismic antenna. The DAS system recorded continuously for one month on more than 2000 channels at 1 kHz sampling rate. The large data volume (~18 TB) renders manual event picking and categorization practically unfeasible. For such DAS monitoring experiments, automatic event detection and classification might become indispensable.
Different types of icequakes and meteorological effects are visible on the raw strain-rate recordings. Most cryoseismic events (stick-slip icequakes and surface crevassing) are of short-time duration and small spatial extent. Since most events are only visible on fractions of the array, the data is divided into smaller sub-windows in time and space.
Different array processing techniques are discussed to obtain a low-dimensional feature representation of each sub-window. The output from the array processing is used as input for an unsupervised clustering algorithm assigning a class-membership to each sub-window. This is applied to parts of the Rhonegletscher dataset. Meaningful clusters are returned that corre-
spond to seismic events or noise.
The analysis yields a preliminary overview of signal types contained in the Rhonegletscher dataset and their spatio-temporal distribution. This method can be used to obtain an event catalogue of clustered signals. This could be used as basis for further analyses such as seismic imaging, template matching or supervised machine learning.

Distributed Acoustic Sensing
Array Processing
Event Detection
Unupservised Classification
Clustering

http://resolver.tudelft.nl/uuid:b98362cd-ab70-4158-9055-733e86d29b13

Student theses

master thesis

© 2021 Julius Grimm