Introducing probability and uncertainty in groundwater modeling with FEMWATER-LHS

Abstract

A three-dimensional finite element groundwater model for density-dependent flow and transport through saturated–unsaturated porous media was written, taking into account uncertainty modeling by means of Latin Hypercube Sampling (LHS) using a restricted pairing algorithm. The original FEMWATER code was rewritten and a LHS shell combined with rank order correlation was added. The new code was named FEMWATER-LHS and comes with an Argus Open Numerical Environments Graphical User Interfaces (Argus ONE™ GUI). The hybrid Lagrangian–Eulerian finite element method was incorporated in the transport module, thus the combined flow and transport can handle a wide range of real-world problems. To demonstrate the applicability for uncertainty and sensitivity analysis using Latin Hypercube Sampling, the FEMWATER-LHS code was applied to two benchmark test and good estimates of the cumulative distribution function in comparison with results from large random or LHS-samples were obtained. Latin Hypercube Samples estimates seem to improve slightly between m = 20 and 80. The statistical sensitivity was calculated by standardized rank regression coefficients and partial rank correlation coefficients to clarify the relationship between output variable and critical input parameter values. The saturated water content θ s or porosity in material one was the most sensitive parameter followed by the saturated hydraulic conductivity contributing to the uncertainty in concentration.