Advanced signal recognition methods applied to seismo-volcanic events from Planchon Peteroa Volcanic Complex: deep neural network classifier

Abstract

Advanced techniques in the recognition and classification of seismo-volcanic events are transcendental when studying active volcanoes, not only for their importance as an accurate real time seismic monitoring procedure but also for the use of their results in modeling the dynamics of the volcanic environment. It is well known that real time seismic monitoring deals with such a large amount of data that it would become an overwhelming job for an operator to do manually. Therefore the use of automatic detection and classification techniques based on the Machine Learning approach are suitable in meeting such a challenge. The aim of this work is to test the capability of the Deep Neural Network (DNN) by using different event parametrization as a confident classifier tool that could permit a reliable seismic catalog to be built in a new and un-analyzed volcanic scenario. We tested different configurations in order to build an approach that was as simple as possible to use this classifier with a limited number of events. In this regard, the feature space was explored in order to select the most significant parameters of the seismic signals. The data used for this analysis corresponds to the Planchon Peteroa Volcanic Complex (PPVC) located in the Transitional Southern Volcanic Zone (TSVZ) between Chile and Argentina, South America. The most significant result of this work was not only that it provided an analysis in terms of performance of this algorithm, especially when the training, validation and test dataset is reliable although definitely reduced, but it also gave an insight of into how an optimal event parametrization can significantly improve the automatic detection and classification of seismo-volcanic events.