Amplitude source location for volcanic earthquakes using source location-dependent amplification corrections: Application to Izu Oshima volcano, Japan

Summary The amplitude source location (ASL) method has been widely applied to volcanic earthquakes that do not show clear wave onsets at active volcanoes. To reduce the effects of shallow velocity structures just beneath seismic stations, observed amplitudes are generally corrected by using seismic coda waves of regional earthquakes. However, even using such amplitude corrections, ASL may not accurately locate the source of volcanic earthquakes occurring in a strongly heterogeneous medium in volcanic regions. To overcome this limitation, we propose a source location-dependent amplitude correction method and introduce seismic velocity structure with a vertical gradient. We analyze 339 volcano-tectonic earthquakes (VTs) recorded by eight seismic stations deployed across the Izu Oshima volcano edifice. Reference hypocenters are determined from manually picked P- and S-wave arrival times and used to evaluate the performance of the proposed amplitude-based method. We divide the VT dataset into two subsets, which are used as reference and test datasets. The amplification factors, which are the ratio of observed to predicted amplitudes of known VTs, are calculated for the reference dataset. The amplification factors are further interpolated onto a two-dimensional grid using the ordinary kriging method with an exponential variogram model and then replicated at each depth level of the three-dimensional grid. Using interpolated source-location dependent amplification factors (SLAF), we correct the observed amplitudes of VTs in the test dataset and estimate the source locations by minimizing the amplitude residual in a grid search scheme. We use the 8-16 Hz frequency band, a velocity gradient k of 0.33 s−1 and v0 1.5 km s−1, and a constant attenuation factor (Q) of 50. The obtained source locations show good agreement with the locations determined from P- and S-wave onsets. Median misfits are approximately 1.1 km and 1.0 km in the vertical and horizontal directions, respectively. Meanwhile, the median misfits from conventional ASL are approximately 2.2 km and 2.3 km in the vertical and horizontal directions, respectively, with the estimated location concentrated close to the summit. These results demonstrate that our proposed method substantially improves relocation consistency and reduces summit-ward bias.