A dataset of 42 experimental fires in maritime pine (Pinus pinaster Ait.) stands was used to develop fuel models to describe pine litter and understorey surface fuel complexes. A backtracking calibration procedure quantified the surface fuel bed characteristics that best explained the observed rate of fire spread. The study suggested the need for two distinct fuel models to adequately characterise the variability in fire behaviour in this fuel type. In these heterogeneous fuel beds the fuel models do not necessarily represent the inventoried average fuel conditions.

Evaluation against the modelling data produced mean absolute errors of 0.8 and 0.6 m min^–1 in rate of spread, respectively, for the litter and understorey fuel models, with little evidence of bias. The fuel models predicted the rate of spread of a validation dataset with comparable error. Comparison of the behaviour and evaluation statistics produced by the study fuel models with fuel models developed from inventoried fuel data alone revealed an improvement on model performance for the current study approach for the litter fuel model and comparable behaviour for the understorey one.

We examined model behaviour through comparative analysis with models used operationally to predict fire spread in pine stands. Large departures from model behaviour essentially occur when the models are exercised outside the range of the model development dataset. The discrepancies in predicted fire behaviour were hypothesised to arise not from differences in fuel complex structure but from the selected functional relationships that determine the effect of wind and fuel moisture on rate of spread.