Bayesian soil parameter estimation: Results of percolation-drainage vs infiltration laboratory experiments

• Published in: Journal of hydrology (Amsterdam) 2018-10, Vol.565, p.770-778
• Main Authors: Younes, A, Zaouali, J., Fahs, M., Slama, F., Grunberger, O., Mara, T.A.
• Format: Article
• Language: English
• Subjects: Applications; Drainage experiment; Infiltration experiment; Markov Chain Monte Carlo; Richards’ equation; Statistics; Uncertainty ranges
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2018.08.082

•Unsaturated soil parameters are investigated for an undisturbed soil.•Percolation-drainage and infiltration laboratory experiments are performed.•Hydraulic parameters are estimated in a statistical framework using MCMC.•Both experiments yield similar mean estimated parameters.•Strong discrepancies occur between the obtained confidence intervals. In this work, we conducted two laboratory column experiments on an undisturbed sandy soil. The first deals with a percolation-drainage experiment whereas the second deals with an infiltration of a constant water flux at the surface of the unsaturated soil. A Bayesian assessment of the soil parameters is performed for both experiments with the Markov Chain Monte Carlo (MCMC) method using measurements of pressure head inside the column and cumulative outflow collected during the experiments. The results show that both experiments can be well reproduced with the mathematical model based on the Richards equation and the van-Genchten/Mualem models. Furthermore, the inversion of the two laboratory experiments yields similar results in terms of mean estimated parameter values but strong discrepancies occur between confidence intervals used to quantify uncertainty on the estimated parameters. Compared to the percolation-drainage experiment, the infiltration experiment yields more accurate parameters with narrower uncertainty regions.