Regional mapping of ultra-low velocity zones beneath the Coral Sea using Bayesian inference

Abstract:

Forward waveform modeling of seismic wave conversions at the core-mantle boundary (CMB) points to a strong decrease in P- and S-wave velocity, and an increase in density in a thin zone above the CMB known as the ultra-low velocity zone (ULVZ). However, physical interpretation of ULVZs is challenging due to non-uniqueness of model parameters and lack of rigorous uncertainty estimates. Here, we present results from Bayesian waveform inversions of ScP waves (shear waves converted to, and reflected as compressional waves at the CMB) that sample a wide region beneath the Coral Sea off northeast Australia. The waveforms are obtained from short-period transportable arrays in southeast Australia and the Warramunga array in Northern Territory. The inversion does not require explicit noise and ULVZ parameterization (i.e. number of ULVZ layers and noise parameters treated as unknowns). Model uncertainties are quantified and distinguished from parameter variability as a function of depth and at various locations. The study reveals complex ULVZs, some well and some weakly constrained, with multiple layers as likely solutions. A common feature in all well-constrained results is that the S-wave velocity decreases as a function of depth with narrow uncertainties while P-wave velocity and density have wider uncertainties. Furthermore, ULVZ height varies as a function of location, at the CMB which implies lateral variability of these structures. S and P velocities are decreased by up to 50% and 30%, respectively, whereas density increases up to 30% with respect to the 1-D reference model. These strong perturbations indicate the presence of melt-rich iron material in the lowermost mantle beneath the Coral Sea. In contrast, weakly constrained ULVZs can be a result of: (a) an insufficient number of waveforms to reduce the incoherent noise and/or (b) incoherent pre-/post-cursors due to the 3D shape of ULVZs which cannot be accounted for by a 1-D forward model.