Browsing by Author "Anzieta Reyes, Juan Camilo"
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item Metadata only Finding possible precursors for the 2015 Cotopaxi Volcano eruption using unsupervised machine learning techniques(2019) Anzieta Reyes, Juan Camilo; Ortiz Erazo, Hugo David; Ortiz Erazo, Hugo DavidCotopaxi Volcano showed an increased activity since April 2015 and evolved into its eventual mild eruption in August 2015. In this work we use records from a broadband seismic station located at less than 4 km from the vent that encompass data from April to December of 2015, to detect and study low-frequency seismic events. We applied unsupervised learning schemes to group and identify possible premonitory low-frequency seismic families. To find these families we applied a two-stage process in which the events were first separated by their frequency content by applying the k-means algorithm to the spectral density vector of the signals and then were further separated by their waveform by applying Correntropy and Dynamic Time Warping. As a result, we found a particular family related to the volcano’s state of activity by exploring its time distribution and estimating its events’ locations.Item Metadata only Non-supervised classification of volcanic-seismic events for Tungurahua-Volcano EcuadorAnzieta Reyes, Juan CamiloIn this paper we propose the use of self-organizing maps and archetypal analysis as an method of unsupervised classification of seismic signals. Using this method we analyzed the record of seismic events for Tungurahua-Volcano (Ecuador) for the year 2014, obtained by a permanent geophysical station from Instituto Geofísico EPN located at the volcano. In standard volcanic monitoring procedures there exists a classification for seismic events performed in a supervised manner (a human being assigns a class to each event based on perception and some fixed criteria). However, even if this classification yields some information on the possible ongoing volcanic processes inside a volcano, it is not determinant when used as a method to predict an actual volcanic eruption. The method proposed in this paper has several advantages over supervised classification by human or based on human classification of seismic signals, one is that it is fast and can be automatized without relying on human intervention, other is that correlates well with human classification for events that clearly mark a volcanic eruption, moreover it finds other cluster of events that could be examined further to established if they have a volcanic interpretation.