RMS and Hierarchical Agglomerative Clustering to Build Template Event Catalogs of Tectonic Tremors

Since their discovery in 2002, the detection of slow earthquakes and their precise locations in subduction zones have become important topics of interest in seismology. These earthquakes can potentially be linked to the generation of large megathrust earthquakes. Traditionally, tectonic tremors and low-frequency earthquakes (LFE) are identified using two different methods: envelope cross-correlation method and template matching, respectively. In particular, the template matching method identifies seismic events by means of quantifying the similarity between observations and master signals known as templates. Although powerful, it relies heavily on the template event catalog used. This method has not been used to detect tremor signals due to their stochastic nature and relatively low signal-to-noise ratio values. However, under the assumption that the tremor waveforms may be composed of several LFE, we may build tremor waveform catalogs. To address these points, recent research has been focusing on improving template matching. In this regard, we present the You Only Search Once (YOSO) method, which creates template catalogs based on automatic detection and classification of waveforms associated with seismic slow earthquakes. The YOSO method comprises three stages: identification of events using a root mean square (RMS)-based detector, estimation of the duration of each detected event by calculating the RMS amplitude decay curves, and their classification using a hierarchical clustering scheme. We applied the YOSO method to synthetic seismograms and observed seismograms from Hi-net broadband stations located in central Shikoku, Japan. YOSO retrieved tremor-related waveforms with durations ranging from 100 to 400 s in the 2–8 Hz frequency band. By determining first-order hypocenters and comparing them with existing catalogs, we confirmed that the waveforms correspond to tremor waveforms. This method provides a new scheme for template-based detection without the dependence on existent template catalogs.