UVM Theses and Dissertations
Format:
Print
Author:
Wei, Xinyu
Title:
Dept./Program:
Civil and Environmental Engineering
Year:
2006
Degree:
PhD
Abstract:
Long-term monitoring (LTM) is the process of evaluating groundwater state variables, such as water levels, temperature, and chemical data over a period of time. The purpose of this project is to evaluate the concept of optimization of a LTM network design by reducing the sampling locations and frequency while satisfying the constraint that the estimation variances at specified locations will not exceed a certain level. The algorithm of this LTM network design consists of a combination of stochastic simulation using Monte Carlo methods, a static Kalman filter and a genetic algorithm. The Monte Carlo simulation is used to produce the spatio-temporal concentration random field, static Kalman filter is applied to condition the concentration random field with measurements at selected locations and times, and the genetic algorithm is used to select the heuristic sampling scheme that takes a minimum number of samples but satisfies the variance constraints.
This LTM is implemented in an intermediate-scale, fully-characterized subsurface research facility. Multiple aquifer tests on the hydraulic conductivity (K) of the soil were performed, a three-dimensional continuous tracer injection experiment was conducted, and concentration measurements of solute were taken by a dense sampling network to create an exhaustive transport dataset. The implementation demonstrates that the prediction of concentrations based solely upon hydraulic conductivity measurements, uncalibrated concentration simulations and a static Kalman filter can be erroneous if the differences between concentrations simulated by the models and the measurement data are significant. In this case, a Kalman filter concentration update should be replaced by a bias-correction scheme. In addition, a standard deviation (StD) constraint is recommended over the coefficient of variation (CV) constraint in the optimal design algorithm. Distributed computing is recommended for Monte Carlo simulations. A final sampling scheme saved 53.3% of the sampling cost with actual errors less than 5% at 92.4% of the unsampled instances.
This LTM is implemented in an intermediate-scale, fully-characterized subsurface research facility. Multiple aquifer tests on the hydraulic conductivity (K) of the soil were performed, a three-dimensional continuous tracer injection experiment was conducted, and concentration measurements of solute were taken by a dense sampling network to create an exhaustive transport dataset. The implementation demonstrates that the prediction of concentrations based solely upon hydraulic conductivity measurements, uncalibrated concentration simulations and a static Kalman filter can be erroneous if the differences between concentrations simulated by the models and the measurement data are significant. In this case, a Kalman filter concentration update should be replaced by a bias-correction scheme. In addition, a standard deviation (StD) constraint is recommended over the coefficient of variation (CV) constraint in the optimal design algorithm. Distributed computing is recommended for Monte Carlo simulations. A final sampling scheme saved 53.3% of the sampling cost with actual errors less than 5% at 92.4% of the unsampled instances.