A connectionist approach for automatic labeling of regional seismic phases using a single vertical component seismogram

Abstract

We present an automatic system for regional seismic phase identification from monocomponent single station records. It is based on a neural network study of the spectrogram. A large dataset of regional events checked by experts has been used for the training step. A sophisticated neural network design allows the system to take into account the variability of the different regional seismic phases in a wide magnitude and distance range. On the training and test sets respectively, more than 85% and 70% of the data are correctly classified.