The grid is a stand-in for a satellite image tile, one cell per pixel-patch. Deforestation ignites at random frontier cells and spreads to forested neighbours (real clearing grows from existing cleared edges, it rarely appears mid-canopy). A pollution plume diffuses outward from a fixed source the same way a discharge spreads through water or air. Neither process is "detected" as it happens — an orbiting satellite only sees the ground each time its scan line crosses it, exactly like the cyan bar sweeping the grid below.
Δforest = forest(t_prev_pass) − forest(t_scan)
alert ⇔ Δ > sensitivity (else: change happens, but stays unflagged)
- Deforestation rate — how fast ignited clearings eat into neighbouring forest cells, and how often a brand-new illegal clearing ignites.
- Pollution spread rate — diffusion strength carrying the plume outward from its source each pass.
- Detector sensitivity — the change-detection threshold a computer-vision model would use between two image passes; too high and slow creeping loss goes unflagged, too low and sensor noise starts to false-positive.
- Satellite scan speed — how many grid rows the scan line covers per second, standing in for revisit frequency — a slower revisit means more change accumulates, undetected, between passes.
Real-world relevance: this is the same change-detection idea behind Global Forest Watch and Sentinel-2-based deforestation alerts — compare a new satellite pass against the last one, flag pixels whose spectral signature moved more than the noise floor, and route only those to a human analyst.