Can local ADE-based models predict contaminant breakthrough?

Abstract:

A solute breakthrough curve measured during a two-well tracer test was successfully predicted by multiple modelers using six 3D contaminant transport models. Water was injected into an alluvial sand aquifer and pumped out 38.3 m away at the steady rate of 250 gallons/minute. The injected water was spiked with bromide for 76.6 hours; the outflow concentration was measured for 32.5 days. Based on previous testing, the horizontal hydraulic conductivity of the 21.6 m thick aquifer varied by a factor of 7 among 12 layers. Each modeler used the same limited data set consisting of K_h(z), bromide concentration in the injection well over time, the above measured well and aquifer parameters and aquifer porosity and dispersivities. Groups using four MODFLOW-based models with brick-shaped grid blocks and two models with curvilinear elements following the arc-shaped flowlines predicted bromide breakthrough curves that matched the measured concentrations reasonably well. The fit between calculated and measured breakthrough curves degenerated as the number of model layers and/ or grid blocks decreased, reflecting a loss of model predictive power as the level of characterization lessened.

How can these results for a moderately-heterogeneous sand aquifer during a forced-gradient test be applied to modeling transport under natural gradient conditions in the highly-heterogeneous aquifer of the MADE site experiments or similar ones? These results constitute a best-case scenario for the use of local ADE-based models. Such models lose predictive power as (1) the spatial and temporal scale of transport increase, (2) the spatial flow field and concentration heterogeneity compared to the aquifer characterization data density increase, (3) the contaminant of concern is conserved less well due to various processes, (4) the contaminant source strength over time and space becomes less well known, (5) characterization of boundary and initial conditions diminishes, (6) the model grid refinement in time and space lessens; and (7) aquifer dispersivities increase. The recipe for predictive power in local ADE-based models of highly heterogeneous aquifers is omniscience. Under realistic levels of characterization such models are best used qualitatively rather than for quantitative predictions.