Application condition of linearized model uncertainty estimate

This paper studies the application condition of linearized model uncertainty estimate numerically for the Magnetotelluric (MT) problem using Bayesian inference theory, based on a three layer case with different parameter combinations. The nonlinear uncertainty estimate is calculated with relatively intensive computation, via Markov chain Monte Carlo method of Metropolis Hastings, giving an unbiased sampling from the posterior probability density (PPD), while the linearized uncertainty estimate which can be achieved cheaply, is based on the Gaussian distribution centered at the optimal model calculated based on a nonlinear optimization method, adaptive simplex simulated annealing (ASSA). The results show that linear uncertainty estimate can be evaluated as an alternative to the nonlinear uncertainty estimate for certain parameter combinations (model structure).