Applying GPU parallel technology to accelerate FARSITE forest fire simulator

Abstract

Forest fires are one of the most common natural hazards in the countries of the Mediterranean region. The forest fires spread simulators have proven to be very effective tools in the fight against these disasters. But to be able to use these simulators in an effective way it is necessary to be able to obtain realistic predictions of the behavior of the fire in a relatively short time. In the last decades, the advances in the field of computing have allowed different strategies to improve the effectiveness of simulators. With this objective, it has been tried to take advantage of the computing power of the GPUs (Graphic Processors Units) to accelerate the simulation of the propagation of fires. The problem with the use of GPUs is that the data transfer between the memory of the system and the GPU memory increases the execution time. For this reason, the computation operations have to compensate the transfer time of the data from the CPU (Central Processor Unit) to the memory of the GPU; which means, that the data volume must be sufficient to compensate this effect. The main objective of this work is to analyze the execution time of the forest fire spread simulation in a CPU and in a GPU, and to see from what data volume the execution in GPU is more efficient than in the CPU. For this study, a fire simulator has been designed based on the basic model for one point evolution in the simulator FARSITE. As study case, a synthetic fire was used where the fire progresses in a linear front and in a single direction, maintaining constant wind, terrain and vegetation conditions for all points of the fire front and throughout the simulation.