Statistical Analysis of Seismic Events Occurring on Piton de la Fournaise Volcano, La Réunion : Bringing Out Eruptive Precursors

Abstract:

On Piton de la Fournaise volcano, La Reunion island, continuous seismic recordings allow to extract signals associated to rockfalls occuring inside the Dolomieu crater. Using the OVPF catalog, we have investigated these seismic signals in order to find how their characteristics could relate to physical characteristics of rockfalls. Here, we analyze such seismic signals from a statistical viewpoint. Data are first taken on an 8 month period that includes an eruption, from January to August 2014. For all the seismic signals associated to rockfalls in this period, 14 seismic and physical attributes are retrieved, allowing to perform a statistical method known as Principal Component Analysis (PCA). It is processed three times in this study : firstly with the 14 attributes, to highlight the main features that outline the data – which we find to be duration and seismic energy, as well as waveform and frequency content. The second PCA, based on 6 attributes, leads to the definition of several physical types of rockfalls, based on the k-means clustering method. Lastly, PCA and k-means clustering are done on 6 different, easily-computed seismic attributes, and reveal a specific behavior of one cluster of events just before the June 20^th eruption. Based on this finding, 15 easily-retrievable numerical attributes are defined from the specific cluster pointed out in the 2014 study, and tested on 2 other datasets : from January to July 2015, they detect the approach of the February 4^th, May 17^th and July 31^st eruptions. From August to December 2010, our attributes show precursory variations a few days before the October 14^th and December 9^th eruptions. We highlight the increase in a specific type of seismic events shortly before an eruption ; we believe they rather have a volcano-tectonic source but hardly distinguish themselves from rockfalls on seismic signals.