Uncertainty quantification in viscoelastic multiparameter full waveform inversion with Hamiltonian Monte Carlo

Jinji Li, Kristopher A. Innanen

Accurately quantifying uncertainty in multiparameter viscoelastic full waveform inversion (FWI) remains a major challenge due to the strong coupling among elastic and anelastic parameters and the high dimensionality of the model space. While conventional Markov Chain Monte Carlo (MCMC) methods provide a rigorous Bayesian framework, their high computational cost often limits their practicality for large-scale inversions. In this study, we explore the feasibility of Hamiltonian Monte Carlo (HMC) for viscoelastic multiparameter FWI, leveraging Hamiltonian dynamics to improve sampling efficiency and convergence stability. Through synthetic experiments on both simple anomaly and Marmousi-type models, we demonstrate that HMC-FWI successfully reconstructs the main subsurface structures and provides reliable estimates of model uncertainties. Among the inverted parameters, V_P and V_S show higher reconstruction fidelity, while ρ exhibit slightly higher artifacts and uncertainty near the surface. The attenuation parameters (Q_P and Q_S) display moderate updates, suggesting a potential convergence issue. Overall, the results confirm that HMC enables efficient posterior exploration and meaningful uncertainty characterization in viscoelastic multiparameter inversion, providing a robust probabilistic alter- native to deterministic FWI approaches. Future work should focus on further improving the efficiency and ensuring convergence of the algorithm.