Automatic detection and classification of seismic events from glacier calving
This Master's project is available from the intake of autumn 2023. Note the prerequisites for this project. for more information, please contact the listed supervisors.
Glaciers generate seismic waves due to calving, fracturing, and crevassing. Falling and capsizing icebergs may also generate “mini-tsunamis”. Passive seismic recording in front of glaciers can be used to record these waves, either using hydrophones or geophones. The types of events and occurrence frequencies can be used to reveal information about the dynamics of the glacier. Such glacier monitoring may be important in a climate change context, but requires long-term data acquisition that lead to the generation of a huge amount of data.
Hypothesis (scientific problem):
Passively recording seismic receivers record all waves that reach their position, producing a large seismic dataset. Manual analysis of the dataset is time consuming and affected by the interpretation of the person carrying out the job.
In recent years, machine learning and other (semi-)automatic approaches for seismic event detection and classification have increased in popularity. Seismic waves from glacier calving clearly stand out from background noise in the time and frequency domain, and may potentially be a good dataset to use such approaches.
In this project, passive seismic data acquired during 8 days in October 2020 and 5 weeks in 2021 on the seabed and on land in front of a glacier on Svalbard will be used. Both geophone and hydrophone data are available. The dataset contains numerous seismic events with varying characteristics in the time and the frequency domain. The data have previously been manually investigated, but the detection and classification of events could potentially be improved by using an automatic workflow. The development of such an approach may also greatly reduce the time needed to analyze future similar datasets.
The work is to first review potential automatic picking and machine learning approaches for seismic event detection, and then to use these on the described dataset. The results should also be validated against manual analysis of the data. Finally, the project includes reviewing which source mechanisms may be responsible for the different recorded signals.
Bachelor in geophysics