Limitations on Accuracy of Wildland Fire Behavior Predictions[1]

 

Many mathematical models exist for calculating various features of wildland fire behavior.  Some are easy to use, some very complicated, but all will be found to produce results which do not always agree with observed fire behavior.  In some instances, the disagreement can be quite significant.

 

There are three principal reasons for such disagreement, no matter which models are used:

 

1. The model may not be applicable to the situation.
2. The model’s inherent accuracy may be at fault.
3. The data used in the model may be inaccurate.

 

Model Applicability

 

If one applies a model in a situation for which the model was not intended to be used, the “error” in the model’s prediction can be very large.  Most models have the following kinds of limitations and should not be expected to predict what they do not pretend to represent:

 

1. The fuel bed modeled is continuous, uniform, and homogeneous.  The more the real fuel situation departs from this ideal, the more erratic the prediction will be when compared to real fire behavior.
2. Some models assume that the fuel bed is a single layer and is contiguous to the ground (e.g., Andrews et al. 2003); in other words, there is no distinct gap between fuel layers (e.g., a forest stand with ground/surface fuels and crown or aerial fuels).
3. Fire spread by spotting (flying embers or firebrands) is not modeled by some models (e.g., Andrews et al. 2003), so fire spread in those situations where spotting is important will likely be poorly estimated.
4. Fire whirlwinds and similar extreme, fire-induced atmospheric disturbances are not modeled.  Guidelines exist as to when such phenomena are expected, but actual predictions are not yet within the state-of-the-modeler’s art.

 

Accuracy of Model Relationships

 

Wildfires, being infrequent, unpredicted, and often occurring in inconvenient locations, are not ideal candidates for instrumentation and measurement.  As a consequence, data to test theoretical or empirical formulae for wildfire behavior accumulate slowly.  Model testing probably will continue to rely mostly on the occasional “windfall” wildfire observation and to a lesser extent on experimental fires.

 

The relationships between variables in most models must be viewed as weakly tested, semi-empirical  in nature, and subject to exception.  Where tests have been possible with sufficient rigor to test model relationship accuracy, they have usually shown prediction errors to be within a few tens of percent on the average.  Fire behavior varies over many orders of magnitude, and model builders consider models successful if the relationships predict fire behavior within a factor of two or three over a range of two or three decades[2].  This can be taken as roughly representative of the current state-of-the-art in fire behavior model accuracy, including both the effects of applicability and internal accuracy.  So until the limitations of model applicability outlined above are relaxed by further research, improvements in model relationship accuracy beyond the current level are unlikely to increase the overall accuracy of predictions.

 

The most important source of error in any particular prediction may be difficult to pin down to model applicability, model accuracy, or data accuracy.  But the internal consistency of a well-disciplined mathematical model allows one to use it to assess the impact of changes in important variables for specific situations, even if the model overpredicts or underpredicts systematically, whether due to model inapplicability, model inaccuracy, or data errors.

 

Accuracy of Data

 

Fire behavior models should be sensitive to those parameters known to affect fire behavior, such as variations in fuel moisture, wind speed, slope, fuel bed depth, and others.  If these data are not known accurately enough, model output may be significantly in error.  It is easy to recognize the nature of and the effects of errors in data such as the wind speed, the slope, or the fuel moisture content.  More subtle, yet equally important descriptors, such as fuel particle surface/volume ratios, the loading of fuel components in each size class, and the proportions of live and dead components, must also be specified  accurately in order to predict fire behavior realistically.

 

Because models of phenomena as complex as wildfire are, generally, quite nonlinear, the output may be highly sensitive to the same parameter in a different value range.  For this reason, it is difficult to make a valid quantitative statement about the relationship between input data accuracy and output accuracy.  The model in question must be used to establish its requirements for data accuracy, considering the range of values of the variables used for input.

 

If any general rule is valid, however, it is that most likely data accuracy will not be the factor which limits the validity of fire behavior model predictions.  The usually dominating error source is that the fuel complex is not uniform, continuous, homogeneous, and consolidated into a single layer.  Nor is the wind speed constant, the slope everywhere the same, nor the fuel moisture content the same from place to place.  After model applicability, probably the next most important error source is inherent model accuracy.  If standard techniques are followed, it is unlikely that data accuracy would be the dominant error source.  If no measurements are made, however, but estimates are used, the accuracy of the estimates may cause errors as large as the first two sources, or even larger.

 

References

 

Albini, F.A. 1976. Estimating wildfire behavior and effects. USDA Forest Service, Intermountain Forest and Range Experiment Station, Ogden, Utah. General Technical Report INT-30. 92 pp.

 

Andrews, P.L., Bevins, C.D., and Seli, R.C. 2003. BehavePlus fire modeling system, version 2.0: user’s guide. USDA Forest Service, Rocky Mountain Research Station, Fort Collins, Colo. General Technical Report RMRS-GTR-106WWW. 132 pp.

 

McArthur, A.G. 1977. Fire danger rating systems. Special paper prepared for FAO/UNESCO Consultation on Forest Fires in the Mediterranean Region (May 9-18, 1977, Marseille, France). Food and Agriculture Organization of the United Nations, Rome, Italy. FAO Document FO:FFM/77/3-01. 15 pp.