A New Non-linear Technique for Measurement of Splitting Functions of Normal Modes of the Earth

Wednesday, 17 December 2014
Guy Masters1, Surya Pachhai2 and Hrvoje Tkalcic2, (1)University of California San Diego, La Jolla, CA, United States, (2)Australian National University, Canberra, ACT, Australia
Normal modes are the vibrating patterns of the Earth in response to the large earthquakes. Normal mode spectra are split due to Earth’s rotation, ellipticity, and heterogeneity. The normal mode splitting is visualized through splitting functions, which represent the local radial average of Earth’s structure seen by a mode of vibration. The analysis of the splitting of normal modes can provide unique information about the lateral variation of the Earth’s elastic properties that cannot be directly imaged in body wave tomographic images. The non-linear iterative spectral fitting of the observed complex spectra and autoregressive linear inversion have been widely utilized to compute the Earth’s 3-D structure. However, the non-linear inversion requires a model of the earthquake source and the retrieved 3-D structure is sensitive to the initial constraints. In contrast, the autoregressive linear inversion does not require the source model. However, this method requires many events to achieve full convergence. In addition, significant disagreement exists between different studies because of the non-uniqueness of the problem and limitations of different methods.

We thus apply the neighbourhood algorithm (NA) to measure splitting functions. The NA is an efficient model space search technique and works in two steps: In the first step, the algorithm finds all the models compatible with given data while the posterior probability density of the model parameters are obtained in the second step. The NA can address the problem of non-uniqueness by taking advantage of random sampling of the full model space. The parameter trade-offs are conveniently visualized using joint marginal distributions. In addition, structure coefficients uncertainties can be extracted from the posterior probability distribution. After demonstrating the feasibility of NA with synthetic examples, we compute the splitting functions for the mode 13S2 (sensitive to the inner core) from several large earthquakes. We confirm earlier observations of this mode and demonstrate that the newly developed technique is capable of reconstructing normal modes with fewer events.