Microseismic monitoring of rock glacier activity: a case study on Gran Sometta Rock Glaciers (NW Italy)

Fracturing, stick-slip movements, water flow, seepage and other natural events related to the glacial and periglacial environment release energy in the form of elastic waves, also known as microseismic events. Microseismic activity represents therefore a useful tool to study glacier dynamics as it can provide information about their mechanical and structural variations. For this reason, a network of passive seismometers has been installed on the Gran Sometta rock glacier (Aosta Valley, NW Italy) since July 2020 to continuously record ambient seismic noise and detect microseismicity. As a first step of this study, microseismic signals generated in and around the glacier have been detected in the continuous noise recordings through a short-time-average over long-time-average (STA/LTA) algorithm. It was applied directly on the raw signals after the setting of some key parameters (STA window length, LTA window length, pre-event window, post-event window etc.). The second part of the study was the classification of the detected events. It was performed in two ways: visual analysis of the event spectrograms and cluster analysis on four key time- and frequency-domain parameters. These are kurtosis, bracketed duration, 5-Hz frequency class containing the highest energy amount and peak frequency of the amplitude spectrum. The visual classification led to the recognition of six types of recurrent events: low frequency and long duration events (below 10 Hz, tens of seconds), earthquakes, rockfalls, rain-related noise events, snow-related events, high frequency and short duration events (in the range 10-40 Hz). Events characterized by very high frequencies and long durations were removed from the dataset since they were found to be mostly related to the rainfall seismic signatures and other events occurring in the rock glacier surroundings. This filtering procedure allowed to get a more stable clustering of the events likely related to the rock glacier internal processes. The automatic classification of the remaining events was then conducted through a k-means clustering algorithm. The latter was selected since the only required input is the number of desired clusters. Initially, four clusters were defined on the basis of the visual classification but through a trial-and-error procedure, the number of sub-clusters was increased to 8 to better describe the dataset. The seasonal trend of these classes was then computed and compared to meteorological parameters of the site (air temperature and precipitation) to get a more complete interpretation of the different processes involved. Similarities among the seasonal and hourly occurrence of these 8 classes were found and this allowed to combine the 8 sub-classes back into 4 main clusters, including i) low frequency deep quakes (likely related to the rock glacier basal movements); ii) high-frequency shallow quakes (likely icequakes and other processes related to the seasonal modifications in the rock glacier active layer); iii) snow-related events (likely due to a stress redistribution on the rock glacier due to the presence of a high snow cover) ; iv) residual hybrid events still related to rain and other high-frequency short duration quakes. The analysis of the temporal rate of the different clusters may provide a valuable monitoring tool for the rock glacier activity and internal processes.