FW-ANSI2D: A Python package for seismic ambient noise cross-correlation modelling and source inversion

Abstract

We present FW-ANSI2D, an open-source package for full waveform forward and inverse modelling of seismic ambient noise cross-correlations in two dimensions. It belongs to the so-called ‘interferometry without Green’s function retrieval’ class of methods, wherein noise cross-correlations are modelled numerically, for arbitrary spatio-spectral distributions of noise source power spectral density (PSD). A multifrequency inversion for the source PSD is achieved via a nonlinear finite-frequency waveform inversion technique, implemented under the assumption of a fixed Earth structure model. FW-ANSI2D seamlessly integrates with other open-source Python packages for seismic wave propagation modelling, most notably a C-based numerical solver for acoustic modelling in 2-D media. Whilst the package is currently limited to the acoustic modelling regime and flat geometries (Earth’s sphericity is unaccounted for), it is a powerful tool for ambient noise analysis at local scales. It exhibits certain advantages over other waveform inversion techniques for ambient noise sources, such as an enhanced ability to resolve sources outside the receiver network, and a relatively high tolerance for velocity model inaccuracies. Moreover, it is amenable to Hessian-based optimization, which ensures speedy convergence ( iterations) of the nonlinear inverse problem, compared to purely gradient-based methods. We introduce the package in detail, describing both the serial and parallel versions of the code, and present synthetic tests designed to assess the efficacy of the inversion technique in realistic field scenarios. These tests show that reasonably good inversions are achievable even with relatively sparse receiver networks, and with sources lying outside the modelling domain.