Revealing Glacier Dynamics by Hierarchical Clustering of Continuous Seismic Data Recorded on a Dense Seismic Array: Application to Argentiere Glacier, French Alps.

Machine learning algorithms have been widely used in seismic data analysis. However, most of the applications suggest analysis of the data obtained from one single seismic sensor. Such sort of analysis is nevertheless challenging when data is recorded on seismic arrays, particularly in case of glacier study. In this work, we address the unsupervised learning problem of detecting seismic events on signals recorded with a dense seismic array on Argentière glacier, French Alps. To do that, we perform the hierarchical cluster analysis where, as a preprocessing step, we extract interpretable features via computation of spectral width of covariance matrix. Then, hierarchical clustering is done on input data that represent coherence of seismic stations at a given time interval for a frequency band between 1Hz and 30Hz. We complete our analysis by investigating the clustering dendrogram and interpretation of obtained clusters by using complementary environmental and geospatial data.