Each cell of the grid is unburned fuel, bare/rock ground, burning, or ash. Every simulation tick, every burning cell tests each of its 8 neighbours for ignition with a probability built from four independent factors (the same structure used in published forest-fire cellular-automaton models such as Alexandridis et al. 2008):
p = p0 · fuel(neighbour) · (1 − humidity) · wind_factor / distance
wind_factor = exp( k · windSpeed · cos(θ) )
θ is the angle between the wind direction and the direction toward that neighbour, so cells downwind of the fire ignite far more readily than cells upwind — a lit front stretches into a wind-driven spear shape rather than a uniform circle. Denser fuel patches both ignite more easily and burn longer; humidity scales every probability down uniformly, so a wet forecast can stall a fire even with fuel and wind in its favour.
- New terrain & ignite — regenerates the vegetation map at the chosen fuel density (clustered, not uniform — gaps act as natural firebreaks) and drops a spark at the western edge.
- Click the grid — drop an additional spot fire anywhere there is still fuel, e.g. ahead of the main front to see spotting.
- Fuel remaining — the share of originally-fuelled cells that have not yet ignited; it never recovers once burned, matching how a real fire consumes its fuel bed.