OBSU: a deep-learning seismic phase picker for OBS data using transfer learning and Unet

Abstract

Seismic phase identification and ariivals picking are two essential steps in the processing of seismic monitoring data. We propose OBSU, a transfer learning and Unet-based seismic phase picker for ocean earthquakes with multimodal inputs, using the land seismic dataset INSTANCE for pre-training to enhance performance. We test it using the ocean bottom seismometer (OBS) dataset and achieve mean absolute deviations of 0.17 s and 0.23 s for P-wave and S-wave, respectively. Meanwhile, the model results after using transfer learning are significantly better than before using it.