On the Identification of Soil Hydraulic Parameters

Drainage experiments are commonly conducted to determine soil hydraulic parameters. Experiments generally measure the temporal evolution of the cumulative outflow as well as pressure heads and water contents at several levels in the column. Then the analysts try to calibrate the model by fitting the model response to the data series. The aim of the present work is to investigate the quality of inverting Richards-Van Genuchten Mualem (R-VGM) equations prior to experiment. For this purpose, we invert R-VGM in a Bayesian framework from artificial data series. Prior to the inverse modeling, global sensitivity analyses of R-VGM are performed with polynomial chaos expansion method. Then, we use Monte Carlo Markov Chain (MCMC) to estimate the marginal posterior distribution of the parameters. Such an approach allows to address the following issues: Is the model overparametrized? What are the parameters that can reasonably be estimated? And from which measurements?