Assessing uncertainties in high-resolution, multifrequency receiver-function inversion: A comparison with borehole data

We use teleseismic P-to-S converted waves from a permanent station to estimate the uncertainties in a 1D elastic model of the shallow crust (0-7 km depth) obtained from the inversion of receiver function (RF) data. Our earth model consists of layers with a constant S-wave velocity V-S and P-to S-wave velocity ratio (V-P/V-S). We apply a Bayesian formulation and transdimensional Monte Carlo sampling to compute the posterior uncertainties of the earth model. The model uncertainties rely on a realistic representation of the data uncertainties, and we estimate directly from the stacking of the teleseismic data, a full-error covariance matrix. To explore the effect of the number of teleseismic events and the RF frequency content, we compare the results of inverting a single RF computed for a cut-off filter frequency of 4 Hz with the joint inversion of four RFs computed from independent ensembles in a larger pool of events for cutoff frequencies of 0.5, 1, 2, and 4 Hz. The inversion results are compared with the lithostratigraphy and sonic-log measurements from a 7 km deep borehole drilled near the seismic station. The inversion of a single RF results in larger uncertainties in the recovered V-S profile and in the depth to seismic discontinuities compared with the multifrequency inversion. Moreover, the multifrequency inversion predicts more accurately the depth to a velocity inversion at approximately 6 km below the surface and matches more closely the borehole sonic-log data. Our results indicate that RF data can be used to map shallow (3-5 km depth) crustal interfaces with uncertainties in the order of 300-500 m, whereas uncertainties are consistently smaller (< 300 m) for interfaces in the top kilometer.