Characterizing Seismic Activity From a Rock Cliff With Unsupervised Learning

Abstract

Passive seismic monitoring (PSM) is emerging as a tool for detecting rockfall events and pre-failure seismicity. In this paper, the potential of PSM for rockfall monitoring is assessed through a case study carried out in Gros-Morne, Eastern Québec, in a region with prominent roadside cliffs, where more than 500 fallen rocks are found on the main regional road each year. The proposed method relies on using sensitive STA-LTA windows to detect a very large number of seismic events and build a comprehensive catalog. In total, more than 70,000 seismic events were detected over one year. Gaussian mixtures are used to partition the data set. Based on visual inspection of the data, a main working hypothesis is that the seismic events can be clustered into three groups. After analyzing the spatio-temporal distribution of the events in each group, we find that the events of one cluster can be associated with anthropogenic activity. The frequency of occurrence of the events of the different clusters and their link with meteorological data is also examined through a regression exercise, to assess the importance of the meteorological variables as explanatory variables. The results allow us to postulate on the physical origins of the signals in the different clusters, attributing them to rockfall activity and wind-induced seismic noise.