Contextually-based framework for improved data reduction in regional scale analytic element groundwater models

This dissertation introduced a conceptual framework for improved data reduction for numerical environmental models. Its key conceptual idea was a simulation-driven feature simplification, which based a simplification decision on the factors affecting accuracy and runtime of the simulation model. The framework led to the development of a new simplification algorithm, inclusion of contextually based constraints into simplification decisions, a contextually based approach for evaluating data simplification, and methods for identifying the appropriate simplification parameters. The conceptual framework was fully developed within the context of Analytic Element Method (AEM) of two-dimensional steady-state groundwater modeling applied to large groundwater domains.;As part of the development and implementation of the framework, a new Hybrid Three-Dimensional simplification algorithm was designed to improve accuracy and performance of the AEM simulation models. A contextually based simplification operator was developed using constraints derived from physical characteristics that have a direct influence on the results of the AEM groundwater model. Results of the simplification algorithms were evaluated based on AEM groundwater model objectives using a proposed multi-objective function, which measured multiple types of model errors and computational resource consumption. Optimization of the simplification parameters was conducted to identify appropriate simplification level and constraints based on specific modeling objectives. Influence of changes in model parameters on prediction accuracy and optimal simplification parameters were also conducted as part of the experiments.;The multi-objective function provided a contextually based approach to compare and evaluate simulation results from different sets of simplification parameters with regard to the specified modeling objectives. Evaluation of simplification results within the context of the environmental model confirmed that the simplification algorithms operated within the modeling context were superior to traditional cartographic simplification algorithms. The Hybrid Three-Dimensional simplification was very effective in the study sites dominated by a large number of rivers. In the study site with a large number of lakes with a relatively small number of rivers, the Hybrid algorithm was not as effective but it still outperformed the cartographic simplification algorithms. The introduction of contextually based constraint simplification helped improve both the Hybrid and cartographic algorithms. However, the constraints were effective only in the models with lower level of details. Optimization of simplification parameters showed that different simplification levels and constraints were required for different modeling objectives and study site characteristics. Changes in groundwater model parameters or element specifications, such as the aquifer hydraulic conductivity or well pumping rates, resulted in different optimal simplification parameters, and so model parameters should be calibrated prior to optimizing simplification parameters. With an initial set of optimized simplification parameters, prediction scenarios with respect to changes in individual analytic element specification, such as well pumping rates, yielded the same level of accuracy. However, prediction scenarios involving changes in the underlying property of the aquifer, such as hydraulic conductivity, yielded a different amount of model error.