Imaging Noise Sources: Comparing a Data-Driven, Matched-Field Processing Technique and Noise Source Inversion

Abstract:

A number of different methods have been developed to locate noise sources on a spatial domain. One class of methods, with tools like Matched Field Processing (MFP), offer an efficient data-driven approach to locate noise sources. Such methods test different possible noise sources by time-shifting observations and subsequently correlating or stacking. A slightly different approach treats the problem as a rigorous gradient-based full-waveform inverse problem. This requires iterative simulations of the correlation wavefield, but ultimately allows for the incorporation of prior information and iterative updates. These two classes of methods are inevitably similar, as discussed by Bowden et al., 2020 (in review), though that work focused solely on theory and simulated examples.

Here we use real data to compare and contrast the methods. We use stations in Europe and North America to image microseism noise sources in the northern Atlantic, related most likely to oceanic swell and storms (Igel et al., 2020, in prep). This allows us to demonstrate that the application of MFP yields similar results for a lower computational cost, but ultimately can not offer the same resolution or validation offered by the noise source inversion. Other applications with real data will be explored in this presentation, particularly in the case of smaller aperture arrays for local scale source imaging. We also specifically discuss situations where the extra computation in an inversion framework is warranted or not.