Dense Array and Machine Learning Reveal Detailed Relationship between Seismicity and Volcano Magmatic Activity beneath Klyuchevskoy Volcanic Group, Kamchatka

Klyuchevskoy Volcanic Group (KVG) is one of the World’s largest and most active clusters of subduction-zone volcanoes and hosts a large and very active trans-crustal magmatic system. In this study, we applied machine-learning–based detection to the data of the KISS temporary seismic experiment operated in 2015-2016 in order to obtain a detailed catalog of earthquakes associated with the KVG volcano-magmatic activity. Our approach resulted in more than 11,000 detections, approximately ten times more than the previous catalog based on manual picking.

The detected seismic activity is clustered in time and space with many earthquakes occurring in spatially localized swarms. Three main earthquakes clusters are clearly associated with major active volcanoes: Klyuchevskoy, Tolbachik, and Ushkovsky. We automatically classified earthquakes into volcano-tectonic (VT) and long-period (LP) events based on differences in their frequency content. All three clusters mentioned above are dominated by LP events. The largest cluster beneath Klyuchevskoy corresponds to the well-known KVG deep long-period (DLP) seismic activity. It is located at approximately 30 km below the surface (i.e., at the crust-mantle boundary) and contains more than 4,000 events that are strongly clustered in time. Two largest DLP bursts precede the re-activation of Klychevskoy in January 2016 and its eruption in April 2016. Two smaller clusters beneath Tolbachik, and Ushkovsky contain earthquakes located in the crust above 20 km depth. 

We also computed the frequency–magnitude distributions for each of these volcanic LP earthquake clusters and found that they differ from the Gutenberg–Richter power law typical for regular tectonic earthquakes. Volcanic LP earthquakes are deficient in larger-magnitude events and exhibit a much steeper decay in their magnitude distributions. These deviations likely reflect differences in source processes and mechanisms between volcanic and tectonic earthquakes.

We also compared our results with the previously established catalog of seismo-volcanic tremors and found that detections of earthquakes and tremors at first order are anti-correlated in time. Therefore, we suggest that a complete characterization of seismic response to re-activation of a trans-crustal magmatic system requires simultaneous analysis of “discrete” earthquakes and “continuous” tremors with the former providing a very detailed illumination during periods of “quiescence” and the latter containing the information during the periods of significant activity within the plumbing system.

Overall, our study demonstrates the potential of AI-based workflows to efficiently process seismic records from dense seismo-volcanic networks recording simultaneously occurring various types of seismo-volcanic events.