Automatic Detection of Short Duration Events (SDEs) in Fumarolic Tremor of Campi Flegrei Caldera: Preliminary Results
摘要
The aim of this paper is the automatic detection of Short Duration Events (SDEs) recorded in the fumarolic seismic tremor of Campi Flegrei caldera (Italy). In recent times, the main fumarole in the Solfatara-Pisciarelli area has shown episodes of sudden increase in hydrothermal activity during which SDEs events have occurred, due to an increase in the boiling of the mud pool. In this work we examined the SDEs occurred on December 1, 2018, when the amplitude of the fumarolic tremor showed a sudden increase, simultaneously with an enlargement of the gas and sludge emission areas at Pisciarelli. Furthermore, we considered November 15, 2018, as an example of a day characterized by typical fumarolic tremor, without SDEs. To detect the SDEs we used the Multi-layer Perceptron (MLP) neural network. The application of Neural Networks in different research fields of Earth Science, such as seismology and volcanology, has greatly increased in recent years. We tested the MLP network on a dataset of 600 seismic signals, 300 containing SDEs from December 1, 2018, and 300 signals without SDEs from November 15, 2018. Before applying the MLP, we performed a preprocessing phase in order to uniquely characterize the data and encode them through a compact representation. We used a specific procedure, proposed in previous experiments conducted adopting the Self-Organizing Map, to extract information on the frequency content and waveform of the signals. The final feature vector had dimension 52 corresponding to 34 spectral coefficients plus 18 related to the waveform. We used 5/8 of the total dataset for the neural network training and the remaining files for the testing. The results show an average performance of about 99% correct classification of the MLP on the testing set. This means that the suggested automatic procedure well discriminates the signals containing SDEs from those representing the typical fumarolic tremor. These results obtained in this work and the proposed method can be exploited to recognize anomalies in the hydrothermal activity of the Campi Flegrei caldera, which is showing an escalating unrest.