Parallel fuzzy cellular automata for data-driven simulation of wildfire spreading

Cellular Automata (CA) have been introduced many decades ago as one of the most efficient parallel computational models able to simulate various physical processes and systems where the interactions are local. In this paper, we are trying to advance the application of CA in modeling wildfires by accounting for the fuzziness intrinsic to the numerous environmental variables and mechanisms engaged with the emergence of the phenomenon itself. The proposed Fuzzy CA (FCA) model adopts a data-driven approach, based on evolutionary optimization, which allows incorporating knowledge from real wildfires in order to enhance its accuracy. The main difficulty for doing so arrives from the computational complexity of the proposed framework and the burden of computational resources needed for its application, which would prevent the real-time prediction of fire spread scenarios. In order to tackle the aforementioned difficulties, we propose model's fully parallel implementations in Graphical Processing Units (GPUs) and Field Programmable Gate Arrays (FPGAs) hardware. In the article, we first investigate the speedup achieved by the developed parallel implementations. Then, we present and discuss two applications to heterogeneous landscapes through comparisons with observed wildfires. Moreover, we compare the proposed framework with two different modelling approaches and results found are really promising.