Hierarchical clustering using seismic waveform information: a strategy for automatic waveform identification

SUMMARY Seismic waveforms are essential for deciphering the subsurface structures of the Earth. Traditional methods for seismic waveform selection rely heavily on manual identification by experienced seismologists, which can be inconsistent and challenging when complex structures or huge amount of seismic data volumes are involved. Recent advancements in machine learning, particularly supervised learning techniques, have shown promising progress in addressing these challenges; however, their dependence on large labelled data sets limits their application to weak or rare seismic phases. In this study, we propose a new strategy using hierarchical clustering for seismic waveform identification, which does not need labelled data set and minimizes extensive parameter settings. Our strategy is especially powerful when dealing with multiple waveform phases that may shift according to epicentral distance or may be distorted due to attenuation or other factors. We apply our strategy to identify various seismic wave of both P and S phases, especially those sampling deep Earth such as SKS-SKPdS and ScP phases. The results show that the strategy performs excellently and can identify different anomalous signals. Our approach empowers researchers to conduct more detailed studies in previously overlooked regions or data sets, thereby leading to a better understanding of deep Earth’s structures.