A randomized approach to improve the accuracy of wildfire simulations using cellular automata

The accuracy of most Cellular Automata (CA) models for simulating wildfires is limited by the fact that spread directions are constrained to the few angles imposed by the regular lattice of cells. To mitigate such problem, this paper proposes a new CA in which a local randomization of the spread directions is introduced. The suggested technique, inspired by a method already used for simulating lava flows, is empirically investigated under homogeneous conditions and by comparison with the wildfire simulator FARSITE. According to the presented results, the adopted randomization can significantly improve the accuracy of fire spread simulations based on a standard center-to-center ignition scheme.