Swarm Intelligence for Dynamic 3D Pollution Peak Detection Using Drones

Timely localisation of airborne pollution (AP) sources and concentration peaks is critical for public health protection, and rapid response to hazardous releases. This paper studies dynamic pollution peak (PP) detection in a 3D environment using a drone swarm controlled by swarm intelligence (SI) and swarm-inspired drone-based (DB) methods. We use a dynamic 3D CO2 plume benchmark with wind transitions and time-bounded peaks to model nonstationary concentration fields, where multiple peaks appear and decay during a fixed search horizon. We compare baseline SI algorithms, Particle Swarm Optimisation (PSO), Firefly Algorithm (FA), and Artificial Bee Colony (ABC), with DB methods for multi-peak detection (MPD), namely Drone Swarm Intelligence (DSI), Drone Firefly Algorithm (DFA), Drone Bee Colony (DBC), and Drone Hill Climber (DHC). Across four scenarios and swarm sizes N ϵ {4, 10, 25, 50} , performance is assessed using peak concentration captured (Peak PPM), time-to-detection in iterations sτ(u) , and runtime T(s). Results on the most challenging four-peak scenario show that ABC is consistently robust among SI baselines, while DB methods reach near-maximal Peak PPM at moderate and large swarms but exhibit distinct efficiency trade-offs, with DBC achieving very low iteration-to-detection at the cost of higher runtime. A static-to-dynamic transfer comparison further shows that static plume evaluation can invert method rankings under nonstationary conditions, supporting the need for dynamic benchmarks when assessing swarm-based pollution localisation.