Here we will quickly introduce ambient seismic noise and give a short overview of how the noise source inversion works. To get a short overview of SANS, the inversion method, and what is offered on the website you can check out the following video abstract about the latest publication (Igel et al., 2022). For further reading, you can find relevant publications here.
Even though we don’t usually feel it, the ground underneath our feet is in constant motion due to various different natural and man-made seismic noise sources. One of the main natural sources of seismic noise is the open ocean. The source mechanisms of how this background noise is generated by the ocean has been studied intensively and is considered well-understood. Seismic noise between frequencies of around 0.05–0.2 Hz can be classified into primary and secondary microseisms. Primary microseismic sources with a frequency between 0.05 and 0.1 Hz are predominantly excited along coastlines by the interaction of ocean waves with the sloping ocean floor. Secondary microseisms, in contrast, originate from interfering ocean waves and are predominantly excited in the open ocean between frequencies of 0.1 and 0.2 Hz. In our studies, we focus on the noise source distribution of the secondary microseisms. The below animation illustrates how these sources are generated: two ocean waves overlap which in turn create a pressure wave that travels vertically down to the ocean floor. This then creates a tiny displacement everywhere at the ocean bottom. Since the source strength depends on the wave height, the distribution of noise sources changes constantly, particularly if there are storms.
To get information about where the noise sources are located, we look at the cross-correlations of seismic noise data. As illustrated below, if we had a completely homogeneous source distribution, we would have a symmetric cross-correlation. In contrast, if the source distribution becomes more heterogeneous - for example a stronger blob on the right-hand side of this station pair – we get an asymmetry in the cross-correlation. So the asymmetry in the ambient noise cross-correlations contains information about where the dominant noise sources are. To be able to invert for the noise source distribution we need to efficiently forward model cross-correlations. We can do this by using (i) pre-computed wavefields and (ii) spatially variable grids. By taking a measurement in both the synthetically computed cross-correlations and observed cross-correlations we can compute finite-frequency sensitivity kernels. Currently we are using the so-called logarithmic energy ratio as measurement. The sensitivity kernels provide us with information where an increase (or decrease) in noise source strength will improve the noise source model. By summing up the sensitivity kernels for all station pairs we obtain the gradient, which is then used to update the model. This process allows us to iteratively improve our noise source distribution. The whole inversion workflow is visualised in the illustration below. Figure 3: Illustration of the inversion workflow.
The data for the previous day is automatically downloaded at 4am every morning using Obspy. We use a station list with 414 stations (global) or 153 stations (North Atlantic) for which we have already pre-computed the wavefields. Once all the available data has been downloaded we perform simple pre-processing steps:
The inversion requires an initial model to compute the first set of cross-correlations which should be based on prior knowledge. Since secondary microseismic sources can only be created in the open ocean we ignore all grid points on land. A reasonable initial model would then be a homogeneous noise source distribution in the ocean. However, to further improve the workflow and steer the inversion in the right direction we used a different noise-imaging method to create an initial model: Matched-Field Processing (MFP). In contrast to the inversion framework, MFP is a data-driven approach which tests different noise source locations and subsequently correlates and stacks the data to get an estimate of the noise source strength. This process is illustrated in the animation below. Figure 4: Animation of the MFP method.
Tolman, H.L. & Chalikov, D. (1996) Source terms in a third-generation wind wave model. J. Phys. Oceanogr. https://doi.org/10.1175/1520-0485(1996)026<2497:STIATG>2.0.CO;2 Krischer, L., Megies, T., Barsch, R., Beyreuther, M., Lecocq, T., Caudron, C. & Wassermann, J. (2015) ObsPy: A bridge for seismology into the scientific Python ecosystem. Comput. Sci. Discov., 8, 0–17, IOP Publishing. https://doi.org/10.1088/1749-4699/8/1/014003 Ermert, L., Sager, K., Afanasiev, M., Boehm, C. & Fichtner, A. (2017) Ambient Seismic Source Inversion in a Heterogeneous Earth: Theory and Application to the Earth’s Hum. J. Geophys. Res. Solid Earth, 122, 9184–9207. https://doi.org/10.1002/2017JB014738 Ermert, L., Igel, J.K.H., Sager, K., Stutzmann, E., Nissen-Meyer, T. & Fichtner, A. (2020) Introducing noisi: a Python tool for ambient noise cross-correlation modeling and noise source inversion. Solid Earth, 11, 1597–1615. https://doi.org/10.5194/se-11-1597-2020 Bowden, D., Sager, K., Fichtner, A. & Chmiel, M. (2020) Connecting Beamforming and Kernel-based Source Inversion. Geophys. J. Int., 1–14. https://doi.org/10.1093/gji/ggaa539 Igel, J.K.H., Ermert, L.A. & Fichtner, A. (2021) Rapid finite-frequency microseismic noise source inversion at regional to global scales. Geophys. J. Int., 227, 169–183, Oxford University Press. https://doi.org/10.1093/gji/ggab210 If you are using this framework in your research, please cite the following: Igel, J. K. H., Bowden, D. C. & Fichtner, A. (2023) SANS: Publicly Available Daily Multi-Scale Seismic Ambient Noise Source Maps. Journal of Geophysical Research: Solid Earth: e2022JB025114. https://doi.org/10.1029/2022JB025114
