Extracting reliable empirical Green's functions using weighted cross-correlation functions of ambient seismic noise in west-central and southern Brazil

Ambient seismic noise is now routinely used to study the Earth's interior. For an isotropic homogeneous medium, the basic assumption to extract seismic phases from a station pair is that the sources of seismic noise are distributed in such a way that there is a uniform energy flux around the station pair. In general, however, some particular azimuthal directions may dominate the energy flux, which directly affects the extracted interstation empirical Green's function (EGF). To solve this problem, we analysed synthetic cross-correlation functions (CCFs) from seismic pulses propagated in isotropic and anisotropic heterogeneous half-space media towards a station pair under the assumptions of uniformly and non-uniformly distributed sources of noise. A reliable EGF signal can be extracted by applying three processing steps: (1) normalizing the number of repeated stationary sources, (2) normalizing the energy of each excited source and (3) selecting coherent CCFs in the final stacking. In this way, three different classes of station pairs were identified based on the number of CCFs used in the stacking procedure. We introduced and applied a new method based on weighted root-mean-square stacking (WRMS) to the CCFs of more than 33 months of ambient noise recorded from January 2016 to September 2018 at 75 broad-band stations in West-Central Brazil. In the case of non-uniform distribution of source of noise, simple classical linear stacking of CCFs produces distorted EGFs. However, the waveform extracted by the WRMS method is very similar to the Rayleigh waves excited by an earthquake (on 2017 January 3) near one receiver observed at the other receiver. Moreover, synthetic tests and a comparison between extracted and earthquake signals show that although the WRMS method can extract the main part of the signal that is propagated on the shortest path, it cannot recover the energy parts propagated on multipath. Despite the N-W directionality in the geometry of the array, the rose-diagram results indicate no significant spatial variations in the energy level of EGFs extracted by the WRMS stacking, whereas the EGFs extracted by the classical linear stacking indicate the extreme directionality of energy flow in different period ranges. Rayleigh wave group and phase velocity tomographic maps resolved by the EGFs derived from the WRMS method indicate a clear boundary along the Asuncion and Rio Grande Arches between the Chaco-Parana' and the Parana' basins at the shorter period, while the tomographic maps in the same periods which were calculated by other stacking methods cannot clearly separate basins and arcs. Our tomographic maps at longer periods indicate variations of Moho depth and lithospheric velocities.