High-resolution S-velocity structure of the shallow crust via Trans-dimensional Multi-Frequency Receiver Function inversion

Abstract:

The receiver function (RF) technique is a widely used tool for investigating the crust and upper mantle structure beneath a seismic station. Constrains on the local 1D S-wave velocity profile can be obtained from the analysis of the arrival times and the amplitudes of P-to-s waves. In recent years, stochastic approaches have been successfully applied to solve the RF inverse problem in a Bayesian framework, which reconstructs the posterior probability distribution of the investigated parameters. To this broad class of algorithms belongs trans-dimensional (Trans-D) Markov chain Monte Carlo sampling, where the number of unknowns is an unknown itself. The main advantages of such approach are that (1) the resolution in the final results does not depend on subjective choices; and (2) the uncertainties of the inverted parameters can be robustly estimated. Trans-D algorithms have been applied to the RF inverse problem, giving excellent results in reconstructing the entire crustal structure. However, the relative low frequency used (1Hz) prevented resolving small-scale details (<1km). Here, we present a new implementation of a Trans-D algorithm for the solution of the RF inverse problem, which allows for an improved reconstruction of the shallow crust (0-5km depth), with a resolution as high as 0.25km. Two key features have been introduced. First, we adopt a Hierarchical Bayes approach to determine the level of the noise in the observed data. This update addresses our poor knowledge of the data uncertainties, which can bias the final solution of the inverse problem. Second, we jointly invert different RFs computed with different corner frequencies, from about 0.5 to 4 Hz. This approach allows to better constrain the absolute S-wave velocity within the shallow crust. We tested our algorithm using teleseismic data recorded near a 7km-deep borehole, and, thus, our results are directly comparable with lithostratigraphy and sonic log data. Preliminary results show that our algorithm is able to retrieve the depths of the main positive and negative seismic velocity discontinuities. The 1D S-wave velocity model obtained from the posterior probability distribution displays a good match with the P-wave velocity profile measured by the sonic log.