Abstract:
Society faces many complex management problems, particularly in the area of shared public resources such as ecosystems. Existing decision making processes are often guided by personal experience and political ideology rather than state-of-the-art scientific understanding. This dissertation envisions a future in which multiple stakeholders are provided with computational tools for formalizing their management preferences and computing optimal solutions based on state-of-the-art computational simulations. To make this vision a reality, advances are required in optimization and visualization, and this dissertation presents research on both topics within the formalism of the Markov decision process (MDP). First, it describes an interactive visualization system for understanding the MDP under user-defined management policies, reward functions, and transition dynamics. Second, it presents a method for optimizing management policies for the user-parameterized MDPs. The research is illustrated and validated using a combination of benchmark MDPs and an application to the management of wildfire in ponderosa pine forests. For the wildfire problem, an excellent high-fidelity model of forest growth and wildfire behavior is employed. However, this model is extremely slow, which prevents interactive visualization and optimization. To address simulation computational expense, the dissertation also presents a method for creating a fast surrogate model and shows that this model is sufficiently accurate to support policy optimization and visualization.
