A Framework of Bayesian Joint Inversion Using Multiple Surface Wave Dispersion Data and Receiver Functions for Shallow Subsurface Structures

The application of multiple surface wave dispersion (SWD) data is emerging as an essential factor for inferring the subsurface shear wave velocity (Vs) structure. Well-defined subsurface profiles covering shallow (surface) to deep (>1km) structure are important due to the necessity for investigation of the reference rock condition to improve the accuracy of seismic hazard assessment in terms of site-specific parameters (e.g., Vs30, bedrock depth, and site-classification). We incorporated high-frequency surface wave phase velocities derived from multichannel analysis of surface wave (5-20 Hz) and frequency-wavenumber analysis (2-5 Hz), lower frequency group velocities from ambient noise cross-correlations (<1 Hz), and jointly high-frequency teleseismic receiver functions (RF) (Gaussian alpha=5.0). Since SWD and RF are particularly sensitive to absolute Vs and impedance contrast, respectively, the joint inversion can be highly effective. For the inversion scheme, we adopt a transdimensional and hierarchical Bayesian joint inversion, which automatically controls the balance between the dimension of model (i.e., the number of layers) and the level of fitting to data, and it provides a statistically rigorous ensemble of model solutions as a multi-dimensional posterior probability distribution (PPD). We verified this methodology using multiple SWD data from a field experiment and RF data from a local accelerograph station (KG.HKU) in the southern Korean Peninsula. The resulting model was probabilistically selected and interpreted based on PPD that is often overlooked in linear inversions due to the selection of an optimum (e.g., mean or best-fitting) model. The obtained model represented a well-constrained subsurface structure with accurate Vs and interface contrast depth over a large depth range (from surface to ~1km) with realistic uncertainties. Based on the result, we confirmed the framework of Bayesian joint inversion using multiple SWDs and RF data is reliable to be applied to entire accelerograph networks in South Korea to provide accurate subsurface Vs structure models and corresponding uncertainties encompassing site parameters and reference rock conditions.