S017-07
Near Real-Time Global Ambient Noise Source Inversions
Near Real-Time Global Ambient Noise Source Inversions
Wednesday, 9 December 2020: 04:26
Virtual
Abstract:
With the rise of full-waveform ambient noise tomography, knowledge of the heterogeneous noise source distribution could greatly advance our near real-time monitoring capabilities and possibly contribute to constraining the current and past ocean state utilising oceanographic theories (e.g. Ardhuin et al., 2011).
We use a finite-frequency sensitivity kernel approach to invert for the time- and space-dependent noise source distribution of the secondary microseisms in the North Atlantic on a daily basis. By implementing (i) pre-computed high-frequency wavefields from wave propagation solvers like AxiSEM (Nissen-Meyer et al., 2014), and (ii) spatially variable grids, we are able to rapidly forward model cross-correlations for any given global noise source distribution up to a frequency of 0.2 Hz. Adjoint techniques allow us to subsequently compute the finite-frequency sensitivity kernels, which enable gradient-based iterative inversions for the power-spectral density of the noise source distribution.
In combination with various seismological python libraries we create a framework that allows us to download, process, and invert observed continuous noise data within a few hours (depending on the array size) requiring only a list of stations. Several synthetic inversions on a regional and global scale show promising results, with noise sources on a scale of several hundred kilometres resolved well. Inverting for observed cross-correlations for several consecutive days with stations in North America and Europe demonstrates the spatio-temporal nature of the noise distribution in the North Atlantic. Automating the workflow could lead to publicly available daily ambient noise source maps on a regional and global scale.
We use a finite-frequency sensitivity kernel approach to invert for the time- and space-dependent noise source distribution of the secondary microseisms in the North Atlantic on a daily basis. By implementing (i) pre-computed high-frequency wavefields from wave propagation solvers like AxiSEM (Nissen-Meyer et al., 2014), and (ii) spatially variable grids, we are able to rapidly forward model cross-correlations for any given global noise source distribution up to a frequency of 0.2 Hz. Adjoint techniques allow us to subsequently compute the finite-frequency sensitivity kernels, which enable gradient-based iterative inversions for the power-spectral density of the noise source distribution.
In combination with various seismological python libraries we create a framework that allows us to download, process, and invert observed continuous noise data within a few hours (depending on the array size) requiring only a list of stations. Several synthetic inversions on a regional and global scale show promising results, with noise sources on a scale of several hundred kilometres resolved well. Inverting for observed cross-correlations for several consecutive days with stations in North America and Europe demonstrates the spatio-temporal nature of the noise distribution in the North Atlantic. Automating the workflow could lead to publicly available daily ambient noise source maps on a regional and global scale.