Automatic Receiver Function Picking Using Fuzzy C-Means Clustering

Abstract

Computing receiver function (RF) from teleseismic records can be affected by noise present in the seismic waveforms, and therefore, visual inspection is still preferred for quality control purposes. However, human handpicking RF lacks consistency and requires a significant amount of time and human labor. From manually picked RF data sets, we have identified 4 features that can effectively separate the good and bad RFs. Using these selected features, we have developed a fuzzy clustering-based method to automate the classification of RFs into good or bad quality. This method has been applied to two RF data sets in China–computed from broadband arrays in the Tanlu fault zone and northeast China region. Compared to the hand-picked result, our clustering-based classifier achieves great recall and precision scores exceeding 93% and 83.4%, respectively. These robust classification results suggest that the 4 identified physical attributes could serve as a standard criterion for guiding RF picking. Furthermore, our efficient clustering-based automatic RF picking method holds significant promise for RF imaging with large numbers of seismic stations.