Association of array processing and statistical modelling for seismic event monitoring

We associate an array processing method, called progressive multi-channel correlation (PMCC), and statistical modelling, to detect and classify seismic events. PMCC detects any co herent wavefront crossing an array of seismometers, including the wavefronts not generated by actual seismic events. We use machine learning techniques to classify the PMCC detections between "events" and "noise". These techniques are based on the statistical modelling of features extracted from the seismic signal. The features we select combine features computed directly from the raw signal and features re trieved by the PMCC detector. We apply our method on a real data set from the Songino seismic station, in Mongolia. We compare the performance of fours classifiers: Gaussian naive Bayes classifier, logistic regression, Gaussian mixture mod els (GMM), and hidden Markov models (HMM). In our case study, the GMM and the HMM yield the highest performance.