Estimations of ambient seismic noise source distribution with waveform inversions of multi-component Rayleigh-wave crosscorrelations

The ambient seismic noise source distribution is important in ambient noise imaging and monitoring studies. People commonly assume that seismic sources are distributed isotropically around receivers and use cross-correlation functions as approximations to the surface-wave Green's function between the two sensors. However, in practice this assumption is not valid most of the time. Thus the cross-correlations contain both source and structure information, and one must unravel the source information if one wants to estimate the structure information. Furthermore, estimates of the source location(s) are also useful to study and monitor source processes. Based on these two motivations, we implement an elastic waveform inversion of cross-correlations to estimate ambient seismic noise source distributions. In this presentation we discuss the use of different misfit functions (e.g. full waveforms or travel-time) in the source inversion and related sensitivities to inversion parameters (e.g. the starting model). We explain the physics behind the different types of source kernels derived from the misfit functions. With the ultimate goal to improve the accuracy of the source estimation via the cross-correlations, we also discuss the benefits of using the multicomponent cross-correlations during noise estimation as opposed to only the vertical component correlations. Finally, we show that our inversion scheme provides an accurate estimation on a real dataset and compare with other source imaging methods (e.g. matched-field processing).