Ph.D. Proposal
Omar Abyad
(Advisor: Prof. Dimitri Mavris)
"A Simulation-Based Methodology for Optimizing Tactical Aerial Firefighting Operations"
Friday, January 31
9:00 -10:00 a.m.
Weber Space Science and Technology Building (SST II)
Collaborative Design Environment (CoVE)
or
Abstract
Wildfire have become increasingly frequent and intense, imposing a significant burden on limited firefighting resources and underscoring the need for effective decision-making during active wildfire suppression efforts. This dissertation focuses on modeling tactical decisions related to the deployment of ground and aerial firefighting assets, recognizing that swift, well-informed strategies can reduce both the immediate costs of suppression and the lagging indirect costs that persist long after the flames are extinguished. The core premise is that simulation and optimization techniques, when integrated into real-time command and control structures, can augment existing experience-based decision making. However, a review of relevant literature reveals a critical gap; though predictive models and geographic information systems (GIS) are in use, they are not usually coupled with optimization algorithms that ensure optimal resource allocation under the unpredictable dynamics of wildfire behavior.
To address this gap, this research explores different classes of optimization algorithms suitable for sequential decision making in high-risk scenarios. Ultimately, this research concludes that a tree search, and specifically a Monte Carlo Tree Search (MCTS), is best aligned with the requirements of tactical firefighting environments. From this position, three hypotheses are proposed: (1) implementing MCTS in tactical firefighting operations yields improved results compared to existing methods; (2) MCTS is flexible enough to adapt to wildfires of varying scale and complexity, and there exists an optimal search depth that can further boost real-time performance; and (3) surrogate models of fire spread can be developed using Convolutional Neural Networks (CNNs) to replace computationally intensive models such as FARSITE or FlamMap.
This research includes designing a series of experiments to evaluate these hypotheses. The first experiment proposes combing an agent-based modeling (ABM) environment with MCTS. This essentially allows the tree search algorithm to handle resource allocations at the overall operational level while the ABM simulates detailed firefighting activities of aircraft and fire crews within smaller sectors. The second experiment tests the ABM-MCTS framework across various wildfire scenarios to determine adaptability, scalability, and optimal tree-search depth. The third experiment integrates a surrogate model into the framework to assess its performance against a baseline that uses traditional fire-spread models. Collectively, these experiments present a novel methodology that combines agent-based simulation, MCTS-driven optimization, and surrogate modeling to enhance tactical decision making, thereby improving operational efficiency and reducing the direct and indirect costs of wildfires.
Committee
- Prof. Dimitri Mavris – School of Aerospace Engineering (advisor)
- Prof. Daniel Schrage – School of Aerospace Engineering
- Prof. Jenna Jordan – Sam Nunn School of International Affairs
- Dr. Burak Bagdatli – School of Aerospace Engineering