Bayesian approach for detecting dynamically triggered very low-frequency earthquakes in the Nankai subduction zone and application to the 2016 Mw5.9 off-Kii Peninsula earthquake, Japan

A correction has been published: Geophysical Journal International, Volume 218, Issue 2, August 2019, Page 977, https://doi.org/10.1093/gji/ggz218.Remote triggering of very low-frequency (VLF) earthquakes in the Nankai subduction zone by surface waves from a moderate to large, distant earthquake was examined using a Bayesian approach. The triggering of another type of tectonic slow earthquake/tremor and slow slip was previously discovered using simple signal processing techniques that exploit the different characteristic periods of the triggered and triggering event waveforms. Meanwhile, detecting low-amplitude VLF earthquakes embedded in high-amplitude seismic waves is challenging when they have similar characteristic periods, as established detection algorithms are inapplicable in such cases. Here we use a particle filter/smoother for time-series analysis combined with Markov chain Monte Carlo methods, to detect seismic signals and estimate the maximum likelihood source parameters of VLF events that might have been remotely triggered by surface waves from a moderate to large, distant earthquake, in data where the surface waves and target signals both have predominant periods between 10 and 100 s. This approach was applied to seismograms of the 2016 Mw 5.9 off-Kii Peninsula earthquake (Japan), recorded by the KiK-net borehole array in western Shikoku, Japan, where VLF events were previously reported. The borehole array is used because comparing surface and borehole records can identify vertically incident body waves based on depth-dependent phase differences, which do not appear in records of horizontally propagating surface waves. In particle filtering/smoothing, waveforms of a probable VLF event at each seismic station are predicted by a full-wavefield simulation in a 3-D structure model, and seismic interferometry that estimates traveltimes between surface and borehole sensors. A maximum-likelihood approach is used to estimate source parameters that best explain the surface observations predicted by the borehole observations and probable VLF signals. We found that six VLF events with magnitudes 3.4 ≤ Mw ≤ 4.3 likely occurred in the seismogenic regions of ambient low-frequency earthquakes/tremors (LFEs), and were likely triggered by the arrivals of high-amplitude surface waves. The triggering stress changes are estimated from 0.4 to 1.5 kPa, similar to the values for triggered LFEs reported previously. Since there were no triggered LFEs in western Shikoku following the 2016 Mw 5.9 earthquake, these VLF earthquakes are more sensitive to stress changes caused by surface waves from large earthquakes. This is the first reported example of dynamic triggering of VLF earthquakes