Bayesian inference of preferential seepage path by gradient-based Markov Chain Monte Carlo

Abstract

The region or the path of preferential seepage flow is inversely identified by a gradient-based Markov Chain Monte Carlo method called Hamiltonian Monte Carlo (HMC). Observing hydraulic head and discharge rate of seepage water, HMC method estimates the domain or the path of preferential seepage flow by changing the shapes of finite elements over which the seepage flow is numerically solved by the finite element method. One simple synthetic example is solved in this article, and the numerical result shows that HMC method with a moving mesh performs well for this geometric inverse problem.