Title: Integrating model abstraction into monitoring strategies Authors
|Guber, Andrey -|
|Yakirevich, Alex -|
|Feng, Pan -|
|Kuznetsov, Mikhail -|
|Van Genuchten, Martinus -|
|Cady, Ralph -|
|Nicholson, Thomas -|
Submitted to: Nureg Series
Publication Type: Government Publication
Publication Acceptance Date: February 20, 2013
Publication Date: N/A
Interpretive Summary: Modeling water flow and contaminant transport in variably saturated soils and ground water is the cornerstone of environmental and water resources management. Management-oriented modeling strives to strike a balance between the necessary level of details required and the feasibility of having good mathematical description of those details. The methodology of finding such models, via simplification of more complex base models, is called model abstraction. We developed a systematic approach to model abstraction in our previous research. In this work, we used this systematic approach to develop a model abstraction method for a groundwater monitoring network design. Using state-of-the-art modeling tools in conjunction with the results of unique experiments on tracer transport in groundwater conducted at the Beltsville Agricultural Research Center OPE3 site, we documented the validity of this model abstraction method for use in groundwater monitoring design. This model abstraction method improved the accuracy of model predictions, and allowed for discrimination between conceptually different models. Results of this work are intended to be used in reviews of licensing applications to the U. S. Nuclear Regulatory commission, and are expected to be useful in a wide range of environmental and water resources management applications, as they provide methods and real-world examples of tailoring monitoring networks to different site-specific levels of information about subsurface material locations and properties.
Technical Abstract: This study was designed and performed to investigate the opportunities and benefits of integrating model abstraction techniques into monitoring strategies. The study focused on future applications of modeling to contingency planning and management of potential and actual contaminant release sites within the scope of US NRC operations. The main objective was to develop methods for incorporating model abstraction techniques into the design of subsurface hydrologic monitoring and performance assessment programs. A comprehensive review of groundwater monitoring network (GMN) design techniques demonstrated the increased use recently of physically based pollutant fate and transport models in the design of monitoring networks. Since groundwater monitoring network designs are based on a conceptual model of the presumed subsurface flow and transport conditions, and since abstraction of the model structure leads to a range of conceptual models and their mathematical equivalents, an opportunity arises to design improved GMNs that decrease uncertainty in predictions of individual abstracted models, or in weighted predictions from several models combined via model averaging. Another opportunity is to use model abstraction via parameter determination to augment existing monitoring networks for discriminating between different conceptual and mathematical abstractions. These opportunities are pursued in this work and applied to models that are obtained via model abstraction for fully three-dimensional (3D) flow and transport in variably saturated flow domains. The methodology was applied to data from the USDA-ARS OPE3 experimental field study involving lateral transport of a surface-applied conservative tracer pulse, with transport controlled by regular irrigation pulses and natural precipitation. The vadose zone was monitored for soil water contents and pressure heads, while groundwater levels and tracer concentrations in groundwater at three depths were also recorded. Fine-scale ground-penetration radar survey and time-lapsed electrical resistivity surveys, along with borehole logs, contributed to the development of the original conceptual model in which subsurface structures exerted strong controls on flow and transport. The HYDRUS-3D and TOUGH2 codes were both calibrated manually by first adjusting saturated hydraulic conductivities, and then the longitudinal dispersivity while keeping a constant ratio of the longitudinal and transversal dispersivities. The FULL-3D code was employed to evaluate the new code QUASI-3D proposed in this study to perform limited input domain abstraction assuming one-dimensional flow in the unsaturated zone above the capillary fringe, and three-dimensional flow in the saturated zone and the capillary fringe. Complex fields of the flow and transport variables were obtained, with peak tracer concentration and the time to peak concentration at observation locations serving as performance indicators. A comprehensive sensitivity analysis was performed to identify promising directions of model simplification in the model abstraction process. Four examples of model abstraction were developed using the manually calibrated HYDRUS-3D software as the base model. The examples included using pedotransfer functions for the hydraulic conductivity, profile aggregation, ignoring the unsaturated zone, and combining the use of pedotransfer functions with scaling. Abstraction of parameter estimation using pedotransfer functions was found not to be efficient due to the high sensitivity of the HYDRUS-3D modeling results on the saturated hydraulic conductivity. The scale mismatch between measurements reflected in the pedotransfer database and the computational grid discretization was a plausible explanation of the poor efficiency of the pedotransfer abstraction process. The pedotransfer abstraction was complemented for this reason with scaling abstraction. The increase in the saturated hydraulic conductivity with the spatial scale was simulated assuming a power-law dependence on the ratio of characteristic lengths. The accuracy of simulations based on the concurrent use of pedotransfer and scaling abstraction was very close to the calibrated model, thus confirming the efficiency of systematic model abstraction. Two approaches were developed and implemented in this work to integrate model abstraction in the monitoring network augmentation. The first approach was based on the statistics of decreasing model uncertainties, improvements in the parameter estimation process, and plume detection. Locations selected for the pedotransfer abstraction were close to those of the base model if the peak concentration was used as the performance indicator. Soil profile homogenization was the only abstraction technique that generated a monitoring network dissimilar to the network obtained with the calibrated HYDRUS-3D model. The second approach aimed at using model abstraction in the monitoring augmentation for model discrimination purposes. Ensembles of pedotransfer functions were used to generate the model prediction uncertainty measures, while information-theories were used to find a location where the metric of evidence of difference was maximized. Overall, incorporating model abstraction in monitoring strategies appeared to be a logical step given that both the mathematical models and the groundwater monitoring network designs are based on conceptual models of subsurface structural and geochemical conditions. Developments in the fields of data assimilation and geophysical data fusion will help further to interface the two technologies. Since the use mathematical fate and transport models in groundwater monitoring designs is expanding, one should expect further intertwining of modeling and monitoring based upon a common conceptual modeling basis.