Volcano-Seismic Signal Detection and Classification Processing Using Hidden Markov Models - Application to San Cristóbal and Telica Volcanoes, Nicaragua

The identification and classification of seismic signals is one of primary work that a volcano observatory must do, and this work should be done in a period of time as close as possible to the real time”. Each seismic event is related to a different source process, and its time and spatial distribution could be used as elements of an early warning system of volcanic eruption (see for example Chouet el el 1996 [1, 2], or Chouet 2003 [3]). Moreover, the recognition system is based on the HMM theory, published in the late 60s by Baum et al. (1966) [4] and Baum et al. (1970) [5]. Nowadays, the Hidden Markov Models technique is the more effective one to implement voice recognition systems. Over the past years, Hidden Markov Models have been widely applied in several models like pattern [6, 7], pathologies [8] or speech recognition [9, 10], and DNA sequence analysis [11, 12]. On the other hand, previous works [13, 14, 15, 16, 16a, 16b] have probed the parallelism among speech and volcano-seismic events in terms of signal complexity and real time requirements. At the present many observatories perform this work observing on the screen or in paper the seismograms, and a technician decides the type of event under their appearance in the time domain and their experience. This work is in many cases difficulty if the rate of occurrence of seismic events per hour is high, or if weather or local conditions increase the level of seismic noise. A more detailed analysis, including for example spectral characteristics, would be too time-consuming to be carried out in real time. Furthermore, in a crisis situation, there is a need to make fast decisions that can affect the public safety. This is the reason because many researchers are focussing their efforts in the development of a robust automatic discrimination algorithm of seismic events, enabling technicians to focus their efforts in the interpretation of the situation or to analyze only a reduced number of signals. Recently Del Pezzo et al.[17] and Scarpeta et al.[18] have presented the application of neuronal networks for discrimination and classification of volcanic and artificial signals at Vesuvius Volcano and Phlegraean Fields (Italy). These methods have been successfully applied to discriminate signals for local and volcanic seismicity. However, a limitation of