Abstract

This work presents the design and evaluation of an architecture based on LSTM recurrent neural networks to create P and S wave identification models in volcanic earthquakes. The detection of these waves is a challenge in volcanic signals because, unlike tectonic seismicity, the distances between the seismic sources and the sensors are short. Nevertheless, it is an important stage for vulcanological monitoring because it can locate the origin of the seismic event and obtain physical parameters crucial to forecast the state of a volcano's activity. In general, this process is done manually by analysts in volcano observatories; however, due to the large number of volcanos monitored by the Observatorio Vulcanologico de los Andes Sur (OVDAS) in Chile, it must be automated. The article applies a methodology proposed in the literature to a currently active volcano in southern Chile, the Nevados de Chillan, achieving promising results, especially for the detection of S waves, which are more difficult to detect than P waves.