Acoustic full waveform inversion for 2-D ambient noise source imaging

Abstract

We present a method for estimating seismic ambient noise sources, by acoustic full waveform inversion of interstation cross-correlations. The method is valid at local scales for laterally heterogeneous media, and ambient noise sources confined to the Earth's surface. Synthetic tests performed using an actual field array geometry, are used to illustrate three unique aspects of our work. First: the method is able to recover noise sources of arbitrary spatial distribution, both within and outside the receiver array, with high fidelity. This holds true for complex velocity models and does not require a good initial guess for inversion, thereby bridging a clear gap in the existing research literature. Second: we analyse the extent of biases in source inversion, that arise due to inaccurate velocity models. Our findings indicate that source inversion using simplified (e.g. homogeneous) velocity models may work reliably when lateral variations in velocity structure are limited to 5 or 10\% in magnitude, but is vitiated by strong variations of 20\% or higher. Finally, our technique is implemented without the adjoint method, which is usually inextricably linked to full waveform inversion. Inversions are performed using source kernels computed for each receiver pair, and this approach is computationally tractable for real-world problems with small aperture seismic arrays.