Assessing a Stochastic Fire Spread Simulator

Several wildfire simulators have been proposed in recent years. Often, such simulators have been judged to be reasonable if they are capable of predicting mean fire growth accurately. However, when the management objective is a burn probability map, this kind of assessment will be insufficient, since it does not address the uncertainties intrinsic to fire behaviour. The problem of comparing output from a stochastic fire growth simulator with real fire behaviour is difficult in general. Focussing on a single (important) aspect of the fire behaviour provides a way through this difficulty In this paper, we propose a method to judge the appropriateness of a particular stochastic fire spread simulator by comparing the variability in the rates of spread in simulated output with the variability in the rates of spread seen in some experimental micro-scale fires. The methodology can be applied to other stochastic fire spread models and to large-scale fires.