Fig. 4: Analysis of the M_L3.5 local event (2020-04-14 03:27 UTC, 0.6 km S of the station, 7.5 km below the surface).

a Infrasound waveform (from PARK.14). Subsequent panels show CLEAN results as an image and traditional beamforming results as dots; traditional beamforming results necessarily agree with CLEAN results but only give a single solution and fail to show the wavefield’s complexity. Inset shows three traces (subarray: PARK.09, PARK.14, PARK.17) with amplitude magnified by 30x, showing that coherent signal continues to arrive after amplitude drops. b CLEAN beamforming of 20 sensors shows that the wavefield is dominated by low-slowness primary infrasound for the first ~5 s and by secondary infrasound subsequently, although low-power primary infrasound is still detected for tens of seconds afterward as different seismic waves continue to arrive. The low-slowness feature around time 67–70 s is weak but coherent primary infrasound, perhaps from a different small aftershock. The dashed line at 2 s km^-1, corresponding to an infrasound incidence angle of 42°, indicates the boundary between inferred seismic-coupled primary infrasound and other infrasound. c Coherent ambient noise before the primary infrasound is visible with traditional beamforming (dots with consistent backazimuth of -90 to -100); it is also present in the CLEAN results, but the image colors in this plot do not show it due to its low power. Secondary earthquake infrasound arrives from many directions initially, whereas the latest-arriving infrasound comes strictly from the northwest. Repeating the analysis using only three stations yields consistent but less-precise results for both (d) horizontal slowness and (e) backazimuth. A comparison of single-source beamforming between the three-element sub-array and the full array in horizontal slowness (f) and backazimuth (g) shows that the three-element sub-array tends to have slownesses that differ more from the expected 3 s km^-1, as well as greater scatter in both slowness and backazimuth.