Fire and Smoke: Cellular Automaton Combustion

Seven states, no combustion chemistry

Real combustion is a coupled system of heat transfer, chemical reaction rates and fluid dynamics — genuinely difficult to solve in real time. A cellular automaton sidesteps all of that by never solving the underlying physics directly. Instead, every cell on a grid holds one of a small number of discrete states — wood, fire, ember, ash, smoke, water, wall — and a short set of local rules decides, cell by cell, how those states change from one step to the next, based only on each cell's own state and its immediate neighbours. The visual result looks like fire because the rules were designed to reproduce fire's qualitative behaviour, not because any temperature or reaction rate was ever computed.

The core spread rule

Each simulation step, every wood cell adjacent to at least one fire cell has a chance to ignite; a burning cell has a chance to burn out into an ember and eventually ash; and ember cells radiate heat, so they can still ignite neighbours briefly before cooling. The whole thing is a probabilistic version of a simple neighbour-counting rule, similar in spirit to Conway's Game of Life:

for each cell c:
  if c is WOOD and has a burning neighbour:
      p_ignite = base_rate * fuel_dryness(c) * wind_factor(c, wind_dir)
      if random() < p_ignite:  c → FIRE
  if c is FIRE:
      if random() < burnout_rate:  c → EMBER
  if c is EMBER:
      if random() < cool_rate:     c → ASH
  if c is WATER:
      c stays WATER forever  // ignition probability is always 0

Smoke is produced separately: fire and ember cells emit smoke particles that drift, following a simplified advection rule biased by the wind direction and fading out over time — visually convincing without running an actual fluid solver.

Why wind matters so much

In real fires, wind does two things: it tilts the flame toward unburned fuel, preheating it by radiation and direct contact well before the flame front physically arrives, and it feeds fresh oxygen into the combustion zone, sustaining a higher burn rate. The cellular model captures the net effect with a directional multiplier on the ignition probability — a wood cell downwind of a fire ignites far more readily than an identical cell crosswind or upwind, which reproduces the elongated, wind-stretched fire fronts seen in real wildfires without simulating airflow explicitly.

Firebreaks and where the model bends realism

Water and wall cells have zero ignition probability no matter how many burning neighbours surround them, giving a clean way to test how a river, road or cleared fire line stops a spreading blaze — the same logic real firefighters use when cutting containment lines. The model's honest limitation is that it has no true temperature field or radiative heat transport at a distance; every interaction is nearest-neighbour, so effects like long-range spotting (embers carried by wind to ignite fuel well ahead of the main fire) have to be added as an explicit extra rule rather than emerging naturally from the physics, the way it does in full computational fluid dynamics fire models.

Frequently asked questions

Try it live

Everything above runs in your browser — open Fire and Smoke and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.