Regional Source-type Discrimination Using Nonlinear Alignment Algorithms

The discrimination problem in seismology aims to accurately classify different underground source types based on local, regional or teleseismic observations of ground motion. Typical discriminant approaches are rooted in fundamental, physics-based differences in radiation pattern or wave excitation, which can be frequency dependent and may not make use of the full waveform. In this paper, we explore a new method for event discrimination using phase and amplitude distances derived from dynamic time warping (DTW) and elastic shape analysis (ESA). We demonstrate the ability to distinguish underground point-sources using synthetic waveforms calculated for a 1-D Earth model and various source mechanisms. We then apply the method to recorded data from events in the Korean Peninsula, which includes declared nuclear explosions, a collapse event, and naturally occurring earthquakes. Phase and amplitude distances derived from DTW and ESA are then used to classify the event types via dendrogram and k-nearest neighbor clustering analyses. Using information from the full waveform, we show how different underground sources can be distinguished at regional distances. We highlight the potential of these nonlinear alignment algorithms for discrimination and comment on ways we can extend the framework presented here.