A Hierarchical Agent-Based Model for Decision Making Under Uncertainty in Tactical Firefighting

In recent years, wildfires have become increasingly more frequent and severe causing significant strain on limited firefighting resources. The need for effective tactical decision-making during suppression operations has therefore never been more crucial. This research develops an integrated approach that combines agent-based modeling (ABM) and a Monte Carlo Tree Search (MCTS) to create a hierarchical framework that can optimize, under uncertain conditions, the deployment of ground and aerial firefighting assets in a tactical firefighting scenario. Current predictive fire spread models and Agent-Based formulations lack integration with optimization techniques that can dynamically hedge against the inherent uncertainties present in tactical firefighting. Furthermore, existing tactical decision-making frameworks fail to address the sequential, path-dependent nature of tactical wildfire suppression decisions. This work addresses these limitations by introducing a Monte Carlo Tree Search into the decision-making hierarchy which enables a systematic exploration and evaluation of decision sequences under uncertainty. To test the performance of the proposed method, the combined ABM-MCTS framework was compared to simpler baseline algorithms that use deterministic heuristics to inform decision making regarding asset allocation. The benchmarks conducted in this work clearly show that the ABM-MCTS framework outperforms all other heuristic algorithms by consistently leading to outcomes that yield considerably less damage and losses.