Prediction Uncertainty in Simulated Groundwater Quality Trends
Date of Completion
Although about 2 billion people worldwide rely on groundwater for their drinking water, our knowledge of the subsurface is sparse and imperfect. One of the primary tools to investigate groundwater is the simulation model. Sophisticated models are used that depict detailed subsurface processes, but model uncertainty is often not understood. A better understanding of model uncertainty is needed to make rational management decisions about this important yet imprecisely understood resource. ^ This work is motivated by the need to understand basin-scale changes in groundwater quality in transient groundwater systems. Basic information needs, some of which are addressed here, must be understood better before proceeding to the broader question. This work is divided into three sections. The first section describes a new algorithm to estimate pumping-well recharge-area uncertainty and its use to assess atmospheric tracer data in reducing model uncertainty. The second section describes new methods for calculating solute breakthrough curves in pumping wells. Simulation modeling involves compromise between time and accuracy, and this section provides guidelines for choosing among several popular solution methods. It also provides new algorithms to improve particle-based transport simulation near a pumping well. The final section combines previous work and applies uncertainty in particle-based simulations of breakthrough curves at a pumping well in a synthetic groundwater flow system. Future work will apply these methods to real basin-scale problems. ^ Model uncertainty is estimated using calibration-constrained Monte Carlo simulation with Latin Hypercube Sampling. Parameter covariance estimated using modified Gauss-Newton nonlinear regression is used to generate Monte Carlo realizations that are conditioned on model performance. Parameter correlations are important in groundwater simulation models, and Latin Hypercube Sampling increases simulation efficiency and preserves correlation. Convolution-based particle tracking is used to calculate breakthrough curves at pumping wells. To overcome the discrete nature of particle simulations, truncated Gaussian Kernel Density estimation is used to construct response functions. Increased hydraulic gradients near the pumping well are calculated by embedding an analytical solution in a commonly used semi-analytic particle tracking algorithm. Particle tracking is shown to be efficient and accurate in estimating model parameters. ^
Starn, J. Jeffrey, "Prediction Uncertainty in Simulated Groundwater Quality Trends" (2012). Doctoral Dissertations. Paper AAI3529384.