|Thesis abstract: |
The presence of multiple, institutionally independent but physically interconnected actors is a distinctive feature of the complexity characterizing most of the decision-making problems in environmental contexts. As dealing with many conflicting stakeholders requires to replace the concept of optimality by that of Pareto efficiency, the presence of many decision makers requires some kind of coordination and a cooperative attitude of the involved parties, who agree on adopting a fully coordinated strategy to maximize the system-level performance. These assumptions are often unpracticable in real world contexts, where the decision makers generally belong to different institutions or countries. In these situations, they independently pursue local interests and produce negative externalities leading to a low system-wide efficiency. Game theory and simulation-based approaches are generally used to analyze these issues from a descriptive standpoint, while their prescriptive use in decision support systems to design coordination mechanisms between the originally self-interested decision-making actors remains a challenge. This thesis contributes a novel decision analytic framework based on multi-agent system (MAS) to study water resources planning and management problems in complex decision-making contexts. The aim of the proposed decision analytic framework is to combine descriptive and prescriptive methods, which provide informative tools to represent the actual decision-making context as well as decision support procedures to recommend proper coordination mechanisms. The adoption of an agent-based framework naturally allows the representation of a set of decision makers or stakeholders (agents), which act in the same environment and need to coordinate to maximize the system-wide efficiency in the use of the available water. This agent-based perspective aims to move beyond the traditional centralized approach to water resources management and to explore different levels of cooperation, from fully coordinated strategies and full information exchange to completely uncoordinated practices. Moreover, coupling the agent-based modeling scheme with state-of-the-art Control Theory techniques allows a better understanding of the feedbacks between agents objectives, agents decisions, and the environment they share, as well as the description of their co-evolution and co-adaptation under change. The proposed framework is demonstrated in different problems characterized by distinctive decision-making contexts, which require to adopt different tools and methodologies. The framework is first applied in a hypothetical water allocation planning problem to discuss the issue of balancing system-wide efficiency and solutions practicability. MAS methods based on distributed constraint optimization problems are used to support a watershed authority in evaluating different levels of coordination. Then, the advantages of coordination mechanisms based on the exchange of information are estimated in two real world case studies. The first one assesses the value of cooperation and information exchange in transboundary river basins. The second one shows the potential of co-adapting water demand and supply in agricultural water systems, under current and projected hydroclimatic conditions. Finally, in the last application, the framework combines tools to identify and refine the current operation of the Conowingo reservoir in the Lower Susquehanna system, for balancing evolving demands and system uncertainties. It also introduces a novel method based on input variable selection techniques to support the identification of effective policy mechanisms for environmental protection.