Mitigating artifacts in back-projection source imaging with implications for frequency-dependent properties of the Tohoku-Oki earthquake

Abstract

Comparing teleseismic array back-projection source images of the 2011 Tohoku-Oki earthquake with results from static and kinematic finite source inversions has revealed little overlap between the regions of high-and low-frequency slip. Motivated by this interesting observation, back-projection studies extended to intermediate frequencies, down to about 0.1 Hz, have suggested that a progressive transition of rupture properties as a function of frequency is observable. Here, by adapting the concept of array response function to non-stationary signals, we demonstrate that the "swimming artifact", a systematic drift resulting from signal non-stationarity, induces significant bias on beamforming back-projection at low frequencies. We introduce a "reference window strategy" into the multitaper-MUSIC back-projection technique and significantly mitigate the "swimming artifact" at high frequencies (1 s to 4 s). At lower frequencies, this modification yields notable, but significantly smaller, artifacts than time-domain stacking. We perform extensive synthetic tests that include a 3D regional velocity model for Japan. We analyze the recordings of the Tohoku-Oki earthquake at the USArray and at the European array at periods from 1 s to 16 s. The migration of the source location as a function of period, regardless of the backprojection methods, has characteristics that are consistent with the expected effect of the "swimming artifact". In particular, the apparent up-dip migration as a function of frequency obtained with the USArray can be explained by the "swimming artifact". This indicates that the most substantial frequency-dependence of the Tohoku-Oki earthquake source occurs at periods longer than 16 s. Thus, low-frequency back-projection needs to be further tested and validated in order to contribute to the characterization of frequency-dependent rupture properties.

Lower edge of locked Main Himalayan Thrust unzipped by the 2015 Gorkha earthquake

“…We adopted the Multitaper-MUSIC array processing technique 32 , which resolves more closely spaced sources and is less sensitive to aliasing, yielding a sharper image of the rupture process than the standard beamforming approach 33 . We also applied a 'reference window' strategy 34 , which eliminates the 'swimming' artefacts, a systematic apparent drift of the high-frequency energy towards the station arrays.…”

Multiarray rupture imaging of the devastating 2015 Gorkha, Nepal, earthquake sequence

“…Rupture processes of earthquakes are commonly imaged by the backprojection method), first proposed by Ishii et al [2005] and Krüger and Ohrnberger [2005]. However, phantom projecting ("swimming") due to trade-offs between path length and rupture duration and along-path smearing of backprojected P energy [Meng et al, 2012] can lead to remaining uncertainties and/or artifacts in the rupture model. This issue is addressed here by disallowing P energy at a reference seismogram of a central station to contribute to more than one subevent [Zhang and Ge, 2010;Zhang et al, 2011] or using a reference window strategy [Meng et al, 2012].…”

The 2013 Okhotsk deep-focus earthquake: Rupture beyond the metastable olivine wedge and thermally controlled rise time near the edge of a slab

“…Backprojection (BP) aims at imaging the spatiotemporal distribution of seismic energy release of an earthquake rupture. We applied the Multitaper‐MUSIC array processing technique [ Meng et al ., ] with the “reference window” strategy [ Meng et al ., ], which enables higher‐resolution and weaker “swimming” artifacts than with standard beamforming methods. The relative simplicity and high coherency of the wavefield produced by very deep sources enables source imaging at higher frequencies than previously achieved in BP studies of shallow megathrust earthquakes [ Ishii et al ., ; Meng et al ., ; Koper et al ., ].…”