ИСТИНА

Nowadays there are various tools and techniques to monitor volcanoes in order topredict their unrest and eruptions. One of the most effective approaches is based on interpretation of seismic observations. Volcanoes are very dynamic systems hosting a whole variety of processes reflected in diverse seismic signals recorded by seismic stations. Analysis and interpretation of these signals is the principal task of seismo-volcanic monitoring.This work is aimed at investigating the seismicity within the Klyuchevskoy volcano group (KVG) in Kamchatka. The KVG is one of the world’s largest and most active subduction zone volcanic clusters whose abundant seismo-volcanic activity make it a great “natural laboratory” to study and advance volcano seismology.In the first part, we study in details the deep long period (DLP) earthquakes that systematically occur near the crust-mantle boundary beneath the Klyuchevskoy volcano. This seismicity is believed to be one of the earliest manifestation of the volcanic unrest. However, the source mechanism of the DLP earthquakes remains poorly understood. To start with, the statistical analysis of DLP events was performed. This investigation was done using a detailed catalog obtained after processing almost two years of continuous data with the sensitive template matching algorithm. At the next step, we tried toreconstruct the source of DLP earthquakes using comparison of S-to-P waves amplitudes ratios instead the full waveform or polarity inversions. As the main result of the first part,it was shown that the source mechanism of the DLP seismicity is different from the oneof tectonic earthquakes. Also, the obtained observations turned out to be in agreementwith previously reported connection of the deep magmatic reservoir of the KVG with itsactive volcanoes.Second part of this thesis is dealing with methods of automatic signal analysis based on Machine Learning (ML) algorithms of classification and clustering. The ML exploration was started with the problem of separating two general classes of seismicity: tectonic and volcanic earthquakes. We show the results of events classification using the data recorded during the seismic crisis in 2018-2019. Different possible representations of signals and several ML algorithms were tested in order to obtain the best performance. Further, a possibility to identify various classes of volcanic seismicity from the simultaneously erupting volcanoes was considered. To do so, we clustered the seismic waveforms recordedduring October-November 2022 when the KVG reactivated and the Shiveluch volcano hasbeen active for several months. This work allowed to distinguish a particular cluster of seismicity with the location unusual for the Shiveluch volcano. The problems considered in this work are too broad and far from being solved within a thesis. However, the obtained results can contribute both in physical interpretation of observed signals and in development of the modern methods of volcano monitoring.