Enhancing the accuracy of raster-based algorithms for forest fire spread modelling

Abstract : We introduce in this paper a raster-based algorithm for forest fire growth modelling. The computation is performed on a grid of 2-D cells representing the landscape. During the simulation, the different rates of spread are calculated on the edges of a graph with the proviso that local fires are modelled by ellipses. In this graph, the neighbouhood of a given vertex is defined by a set of intra-cell edges whose orientation and direction depend on a local ellipse. The neighbourhood concerned has therefore a basically dynamic character. Next, we compare the results of this algorithm with those of another one involving static neighbourhoods. In both cases, fire spread is determined using the Dijkstra's shortest path algorithm. We conclude that the facts show that the model involving dynamic neighbourhoods provides more accurate results.