Abstract

This dissertation contributes the ASSIST (Adaptive Strategies for Sampling in
Space and Time) framework for improving long-term groundwater monitoring
(LTGM) decisions across space and time while accounting for the influences
of systematic model errors (or predictive bias). The new framework combines
Monte Carlo based contaminant flow-and-transport modeling, bias-aware ensemble
Kalman filtering (EnKF), many-objective evolutionary optimization, and visual
analytics-based decision support. The ASSIST framework allows decision makers
to forecast the value of investments in new observations for many objectives simultaneously.
Information tradeoffs are evaluated using an EnKF to forecast plume
transport in space and time in the presence of uncertain and biased model predictions
that are conditioned on uncertain measurement data. The goal of the
ASSIST framework is to provide decision makers with a fuller understanding of
the information tradeoffs they must confront when performing long-term groundwater
monitoring network design.

Each chapter of this dissertation focuses on and addresses a specific challenge to
LTGM network design. The scaling challenges of LTGM design are first explored
in order to provide a basis for advancing the size and scope of LTGM design problems
that can be effectively solved using multi-objective evolutionary algorithms
(MOEAs). In addition, complex decision variable interdependencies that exist in
large LTGM design problems cause traditional MOEAs to fail as problem sizes
increase (defined in terms of increasing numbers of decisions and objectives). 
To address this, a new more robust MOEA termed the Epsilon-Dominance Hierarchical
Bayesian Optimization Algorithm (ε-hBOA) was developed to learn and
exploit the complex interdependencies that exist for large LTGM design problems.
Building on the scalable many-objective optimization capabilities of ε-hBOA, the
ASSIST framework contributes visual analytical tools, capable of providing decision
makers with an improved understanding of the complex spatial and temporal
tradeoffs that often exist between their LTGM design objectives. Finally, a biasaware
EnKF framework was developed that dramatically enhances the accuracy of
groundwater flow-and-transport forecasts in the presence of systematic modeling
errors (or biases), while making computational innovations that again expand the
size and scope of LTGM problems that can be addressed.

This dissertation demonstrates that the forecasting, search, and visualization
components of the ASSIST framework combine to represent a significant advance
for LTGM network design that has a strong potential to innovate our future 
characterization, prediction, and management of groundwater systems.