Pollutant is modelled as a swarm of tracer particles released at the stack. Each particle is advected by the wind vector and jittered by a small random walk every frame — a simple Lagrangian form of the advection-diffusion equation. Ground sensors sample the local particle density within a fixed radius each frame; a sensor turns from green to yellow to red as that count crosses alarm thresholds.
x(t+dt) = x(t) + wind·dt + N(0,σ)·√dt
concentration(sensor) = Σ 1[ |particle − sensor| < r ]
source_est = Σ(cᵢ · posᵢ) / Σ cᵢ, over alarmed sensors i
- Wind speed / direction — sets the advection vector; faster wind stretches the plume into a narrower, longer streak.
- Emission rate — how many tracer particles the stack releases per second; higher rates raise concentration readings everywhere downwind.
- Sensor density — spacing of the ground sensor grid; a denser network resolves the plume edge better and lowers localization error.
- Source localization — a weighted centroid of every alarmed sensor's position, weighted by its measured concentration. It is a simplified stand-in for real source-term-estimation algorithms used in environmental monitoring networks.
Real-world relevance: this is the same sense-and-triangulate idea behind city-scale air-quality sensor grids and industrial leak-detection networks — cheap local sensors plus a fusion algorithm can localize an emission source without ever seeing it directly.