Inferring Hydraulic Parameters In a Layered Streambed Based on TDR Measurements Made during Active Flow and Recession: Impacts of High Dimensionality and Model Structural Error

We use the DiffeRential Evolution Adaptive Monte Carlo (DREAM) algorithm to infer four hydraulic parameters in each of seven soil layers (28 parameters in total). Inversions are based on time domain reflectometry measurements made at six depths, ranging from 25 to 225 cm below the streambed surface, during and immediately following an ephemeral stream flow event. Water flow is modeled using HYDRUS-1D. We examine the ability of DREAM to quantify the nonlinear confidence intervals of the parameter estimates during inversion. In addition, we consider both Gaussian and first-order autocorrelated model errors to explore the contribution of model structural error to the quality of the parameter estimations. The results of this study demonstrate that DREAM is a robust inverse algorithm that can be used to assess parameter identifiability and model structural error even under highly heterogeneous conditions.