Influence of Conceptual Model Selection on Three-Dimensional Tracer Transport: Comparison of Predictions and Observations in a Nonuniform Sand Aquifer

This paper presents the results of a field study involving bromide transport through an unconfined sand aquifer with a nonuniform permeability distribution. The tracer test was conducted in preparation for a pilot-scale surfactant enhanced aquifer remediation demonstration in Oscoda, Michigan, USA. A bromide pulse was injected through an array of three fully screened wells and extracted through a single fully screened well approximately 8m away. Concentrations were measured at the extraction well and at 26 individual multilevel sample ports located between the injection and extraction wells. Four alternative modeling approaches were used to explore the influence of conceptual model selection on predictions of three-dimensional transport. In these approaches, alternative representations of the distribution of hydraulic conductivity, K, were employed: 1) a uniform, effective K; 2) perfect stratification with uniform effective K values assigned to each 0.3m layer; 3) conditional sequential Gaussian simulation; and 4) conditional sequential indicator simulation. Grain size distribution information from 167 core samples and hydraulic conductivity measurements from 10 repacked samples provided the basis for geostatistical modeling of the spatial variability of model parameters (variance ln(K) = 0.29). Geostatistical realizations were simulated on a 0.3 x 0.3 x 0.3m grid increment, commensurate with the grain size distribution measurement scale of support. Standard simulation software packages MODFLOW and MT3DMS were used to model three-dimensional groundwater flow and bromide transport. Conceptual models incorporating formation variability are better able to capture observed breakthrough behavior at the fully screened extraction well. Comparisons of simulated breakthrough curves with individual multilevel observation ports demonstrate generally good agreement at points located near the injection wells and poor agreement at points located farther away. This behavior could be explained by the presence of smaller scale preferential flow pathways that were not incorporated in the chosen conceptual models.