Model abstraction techniques for predicting subsurface contaminant transport

Guber, A.K., Y.A.Pachepsky, A.M.Yakirevich, T.J.Nicholson, R.E.Cady

Presented at the All Scientist Meeting (2012-03-15 to 2012-03-16 )

Model abstraction is a methodology for reducing complexity of a simulated domain while maintaining the validity of the simulation results with respect to the questions that the simulation is aimed at addressing. The need for model abstraction stems from the needs to improve reliability and to reduce uncertainty of simulations, to make modeling and its results more understandable and transparent to the end users, and to enable more efficient use of available resources in data collection and computations. This study focuses on applications of model abstraction to contingency planning and management of potential and actual contaminant release sites within the scope of the US NRC operations. The OPE3 experimental field site near Beltsville, MD, has been extensively studied for more than 10 years using geophysical, biophysical, remote sensing, soil and groundwater monitoring methods. A major focus of the data analysis was on the existence of subsurface structural units and features that may drastically change the fate and transport of contaminants in the vadose zone, as well as of projected trajectories of the contaminant plume in groundwater. We found that solute transport in soils and shallow groundwater at the site was affected by (i) the presence of a restrictive fine-textured layer that was not fully continuous laterally; (ii) the complex topography of the restrictive layer favoring preferential flow and transport along preferred pathways along its surface relief; (iii) the presence of natural capillary barriers; and (iv) possible funnel flow in a coarse-textural layer located between fine-textured layers. New field experimental study was designed to observe the lateral transport of a surface-applied conservative tracer pulse when transport was controlled by regular irrigation pulses and natural precipitation. The results of the experiment were used to test applicability of three of model abstraction techniques, specifically: (A-1) aggregation, i.e. replacement heterogeneous soil profile with a homogeneous soil, (A-2) parameter determination, i.e. replacement of calibrated parameters with values estimated using pedofransfer functions; (A-3) model conceptualization, i.e. ignorance of the unsaturated zone by locating source of release within the saturation zone. Different initial and boundary conditions were generated to run calibrated and abstracted HYDRUS-3D model. Applicability of model abstractions was evaluated using performance indicators (PI): peak concentrations, time of peak concentrations, and total fluxes computed in 100 locations across the simulation domain. Abstractions A-1 and A-2 resulted in inadequate predictions of all PI, indicating high model sensitivity to the transport parameters and complexity of the simulated system, while results of abstraction A-3 did not deviate significantly from the calibrated model. Overall, this work demonstrated the usefulness of model abstraction in simulations of flow and transport in variably-saturated subsurface.