CAN MORE MEASUREMENTS MAKE A MODEL WORSE: A TRANSPORT EXPERIMENT IN AN EXHAUSTIVELY SAMPLED SANDSTONE SLAB

An advantage of numerical models over analytic methods is their ability to include field measurements, such as hydraulic conductivity (K) at measured locations. Poor model fit with observed transport is typically attributed to lack of field data. Previous research has shown that despite including the finest currently possible K measurements (0.15 cm) of a slab of Massillon sandstone, traditional numerical models fail to capture significant details of observed solute transport. Unwanted spatial averaging of the K field occurs due to sub-centimeter scale heterogeneities within the sandstone. The resulting inaccurate K field causes the classic advection-dispersion equation (ADE) used within typical numerical models to produce poor results. We present a new method which uses statistically similar ensemble realizations within typical numerical models that can reproduce the early breakthrough and late tails seen in natural aquifers. Using MODFLOW and particle tracking code, ensemble results capture the actual tracer transport better than models using the 8,649 K measurements directly. Results suggest that the K statistics (i.e. mean, variance, and correlation structure) of an aquifer are more important in obtaining accurate solute transport predictions than the measurements themselves.