Affiliation:
1. 1Laboratory of Seismology and Physics of Earth’s Interior, School of Earth and Space Sciences, University of Science and Technology of China, Hefei, China
2. 2CAS Center for Excellence in Comparative Planetology, Hefei, China
3. 3National Geophysics Observatory, Mengcheng, China
Abstract
Abstract
Accurately defining slab geometry is crucial for unraveling the seismogenic mechanism and subduction dynamics. Guided wave, generated from deep earthquakes with a focal depth greater than 100 km, efficiently propagates along a continuous slab and offers an effective way to image the slab geometry. However, it is challenging to manually identify slab guided waves through a large dataset, hindering their application in determining slab geometry. We propose the use of a deep embedding clustering algorithm for identifying slab guided waves. Using waveform data for deep earthquakes within the northwestern Pacific slab recorded by the F-net in Japan, we first employ spectra clustering analysis to determine three classification types. Subsequently, we perform clustering analysis on the spectrogram, efficiently featuring guided wave characteristics by enhancing the high-frequency energy. Then, using the sampled region by slab guided wave as a proxy, we map out the boundaries of the northwest Pacific slab at different depths, particularly within the depth range of 200–400 km. Our inferred slab boundaries correlate well with those derived from other methods, validating the accuracy and efficiency of our clustering analysis. Evaluation of our proposed workflow on smaller earthquakes with a lower signal-to-noise ratio underscores its great potential in determining slab geometry, particularly in less-studied regions.
Publisher
Seismological Society of America (SSA)