A Random Characterization of Deterministic Fire Spread

Francis M. Fujioka

USDA Forest Service

Riverside, California

Figure 1. Simulation of the growth of the Williams Fire, Angeles National Forest, California, on 23 September 2002, using the FARSITE fire modeling system. Arrows depict afternoon surface winds simulated by the MM5 mesoscale weather model.

Although recent research has spawned increasingly sophisticated models of fire behavior, accurate simulation of real fire events for operations remains extraordinarily difficult. Real fires deviate from predicted behavior randomly, in the sense that the spatial and temporal variability of the errors are unpredictable. This suggests that a deterministic fire behavior model should have a random component, which describes the error characteristics of the model in a probabilistic context.

This study outlines a framework for describing the random aspects of real-world fires, using an underlying deterministic fire behavior model. Such a framework is useful for quantifying the variability of model errors and statistically testing their significance with respect to stated hypotheses. Tactical operations will have information about uncertainties of the spread model to perform quantitative risk assessments. A fire behavior model with a plausible random component would also be useful as a training tool, to help fire practitioners deal with unexpected situations.