Automatic low-frequency earthquake template generation using deep learning - Supplementary material

Tectonic tremor is the seismic manifestation of slow slip on the deep extensions of some major geological faults. It is composed of many repeating low-frequency earthquakes (LFEs). LFE analysis should provide valuable insights into plate motion and slow slip processes. However, LFEs have seismic amplitudes little above background noise levels, making them difficult to detect. Therefore, in most locations, they are detected only by targeted analysis using prior waveform templates. Using a 15-year catalog of more than 1 million LFEs along the San Andreas Fault, we implement and train a convolutional neural network (CNN) for LFE detection. We demonstrate that we detect new LFEs with low signal amplitude, even below the noise level, without a prior template, with an accuracy of 95.9\%. We cluster the detected events by waveform similarity to automatically characterize new templates. The described method is scalable and computationally efficient. We share the catalog of new events as supplemental material as a resource for future investigations