Dispersion of contaminants in saturated porous media

The main objective of this paper is to outline the experimental and theoretical investigations performed in an attempt to validate the applicability of finite element based numerical models for the prediction of the behaviour of a conservative tracer at the Twin Lake aquifer, Chalk River Nuclear Laboratories, Chalk River, Ontario. The essential point is that the 3/4 of a million data points obtained at the Twin Lake site from a 40 m natural gradient tracer test provide a unique opportunity for quantifying the system variability and for testing finite element models of the dispersion process. The subject of this discussion is the advection-dispersion model of contaminant transport - its equation and solution by the Galerkin finite element method. The report gives a brief description of the experimental data and the methods for the estimation of transport parameters. Scales of averaging associated with the conceptual formulation of the dispersion process, measurement of process variables, parameter estimation and the numerical models are discussed. The compatibility between the scales is emphasized as a major requirement for predictive modelling. The developed finite element model of the radioiodine transport describes the overall behaviour of the tracer plume but lacks the capability to simulate the fingerlike spreading of the plume due to the fact that the grid does not have an adequately fine space discretization. Unfortunately, a refinement of the grid spacing is limited by the size of the site computer memory. For the advection-dominated transport, as that encountered at the Twin Lake aquifer, the failure to satisfy fine mesh requirement causes numerical dispersion. In general, it was concluded that the conventional finite element model may produce accurate simulation of the tracer cloud provided that the adequately fine space discretization of the grid compatible with the support scale of measurements and the adequately fine time discretization are made. This demands large computing resources. The development of a more finely discretized model for execution on a vector processor is underway. 21 refs