The fire spread predictions used in the current United States fire prediction system are based upon a semi-empirical formulation first presented by Rothermel in 1972. This system has been implemented operationally in the form of programmable hand-held calculators in the late 1970s (Rothermel, 1983), the BEHAVE minicomputer program in the middle 1980s (Andrews, 1986), and the FARSITE fire spread model in the middle 1990s (Finney, 1998). The Rothermel fire behavior model, and variants thereof, require detailed stand-level vegetation properties.
The thirteen standard fuel models used nationally, abbreviated as the National Forest Fire Laboratory (NFFL) models, may not apply everywhere. Custom fuel models can be accomodated into existing fire spread prediction models such as BEHAVE and FARSITE. Developing custom fuel models is more of an art based on trial and error comparison between fuel model input parameters and fire spread predictions rather than science. In other words, direct estimates of fuel properties from ground-based sampling are used to guide values supplied as inputs to new models, but some "fudging" of these estimates is still required. By examining the performance of each of the models against these variables you will gain some insight as to the fire behavior that can be expected for a particular area.