Initial processing of volcanic seismic signals using Hidden Markov Models: Nevado del Huila, Colombia

Abstract

Volcanic eruptions are commonly preceded by seismic signals that carry important information about the underlying physical processes driving the eruptive activity. The manual classification of this seismicity is often difficult during seismic crises and timely responses in volcanic surveillance depend upon on how fast this work is performed. Automatic classification programs are an excellent alternative because they can classify seismic signals in near real-time. Although these programs have been successfully tested for many volcanoes, they have not yet been applied to the monitoring of volcanic crises. We automated the initial processing of the seismic activity recorded at Nevado del Huila volcano in Colombia, using Hidden Markov Models. The technique consists of classifying seismic signals, extracting the main characteristics of each classified signal, and deriving important physical parameters that are important in volcanic surveillance. Continuous seismic data from 2007 to 2012 were used to develop this project. A total of 2142 events were selected to train the program. The best characteristic model obtained during the training had an accuracy of 91.25%, and a similar pattern in volcanic behavior to that from manual classification was modeled. The encouraging results demonstrate that automatic classification programs are a powerful tool that can be used in surveillance and monitoring of active volcanoes as they provide timely information for the proper management of volcanic crises and subsequent hazards mitigation.