Drone swarm coordination for detection and monitoring of dynamic environmental phenomena

Dynamic environmental hazards present a critical challenge for disaster management due to their rapidly evolving intensity and unpredictable boundaries. Wildfires serve as the primary use case for this study, selected for their spatially complex perimeters and because the propagating fire front acts as a dynamic obstacle that imposes real-time geometric constraints on the monitoring area. Conventional monitoring systems, such as satellites or single UAVs, often lack the spatiotemporal resolution required for persistent situational awareness in these complex settings. To address this limitation, this diploma thesis proposes a Hierarchical Drone Swarm Architecture that employs a hybrid control strategy, effectively combining centralized high-level planning with decentralized reactive behaviors. Within a grid-based simulation framework based on Stochastic Cellular Automata for simulating fire propagation, the system utilizes a centralized leader agent to generate the initial paths and optimize the resource allocation for the subordinate drones of the swarm. Simultaneously, the subordinate agents navigate based on the initial paths in the obstacle environment to detect the fire spots and execute a decentralized tracking logic, when they are assigned to a fire perimeter, maintaining close-range surveillance. This architecture is evaluated against two baseline configurations: a stochastic random walk (representing purely reactive search) and a systematic sweep (representing exhaustive search). Experimental results across varying swarm sizes, environmental scales, and initial fire ignition points demonstrate that the proposed hierarchical swarm successfully balances mission speed with resource conservation. In the detection phase, the system delivers superior energy efficiency compared to exhaustive search methods, while in the monitoring phase it achieves high fire-tracking coverage. Furthermore, scalability analysis confirms that by maintaining a constant agent density in the operational area, the proposed architecture effectively mitigates the complexity of larger environments, sustaining rapid detection times and high-quality monitoring coverage even as the region of interest expands.