Preparing the PmP Dataset in Southern California Using PmPNet

The Moho-reflected PmP waves are of great importance for seismic imaging since their different trajectories through the crust from those of the first P-waves. The use of the PmP waves in seismic imaging can significantly increase the illumination of the lower crust and hence provide additional constraints on the heterogeneity and dynamics of the lower crust. Recently, automatic identification of seismic phases has become a popular topic. Despite some exciting success in automatic P- and S-waves picking, auto-identification of later seismic phases such as PmP-waves remains a major challenge in terms of matching the performance of experienced analysts in the seismological community. As of now, manual picking by experts, which is time-consuming, is still the common practice to prepare a PmP dataset. The main difficulty of machine identifying PmP waves lies in the fact that the identifiable PmP waves are extremely rare, which makes the problem of identifying the PmP waves from a massive seismic database inherently unbalanced. In this work, by utilizing a high-quality PmP dataset (10,192 manual picks) in southern California, we develop PmPNet, a deep-neural-network based algorithm to automatically identify PmP waves efficiently, by doing so, accelerate the process of identifying the PmP waves. PmPNet applies similar techniques in the computer science community to address the unbalancement of PmP data sets. The basic neural network architecture of PmPNet is ResNet-AutoEncoder. We conduct systematic research with field data, concluding that PmPNet can efficiently achieve high precision and high recall simultaneously for the automatic identification of PmP waves from a large seismic database. Applying the pre-trained PmPNet to the seismic database from January 1990 to December 1999 in southern California, we obtain three times more PmP picks than the original PmP dataset, which will provide valuable data for imaging the lower crust structures of southern California in later studies.