Uncertainty-quantified 3D ambient noise tomography using transdimensional Monte Carlo inversion

Traditional two-step surface-wave tomography often yields discontinuous models and compound uncertainty. We present the first fully 3-D transdimensional Bayesian inversion with adaptive Voronoi parameterization and reversible-jump MCMC for near-surface engineering-scale arrays, providing voxel-level uncertainty estimates. From 1 week of ambient-noise records acquired by a 101-station linear array (120 m spacing) across the F1 fault zone, we extracted phase velocities via frequency–wavenumber analysis of Rayleigh waves (0.5–3 s). The resulting 3-D Vs. model reveals (i) 300–800 m s -1 in the upper 50 m, (ii) 2.1 ± 0.05 km s -1 at 0–1 km, (iii) 2.6–2.9 ± 0.08 km s -1 at 1–3 km, and (iv) 2.8–3.1 ± 0.12 km s -1 at 3–5 km beneath the fault trace. Voxel-wise 1σ uncertainties range from <5% in the shallowest 2 km to 12% at 5 km depth. These Vs. values and their uncertainties can be directly converted to engineering mechanical parameters: shear modulus G = ρVs 2 , Young’s modulus E = 2G (1+ν), and Poisson’s ratio ν, enabling quantitative assessment of excavation stability, tunnel lining design, and slope stability across the F1 fault zone. The 3-D Bayesian framework mitigates over-fitting biases inherent in sequential inversions and offers critical, uncertainty-aware constraints for multi-stage tectonic reconstruction of the North China Craton destruction belt.