Building Rothermel fire behaviour fuel models by Genetic Algorithm optimization

A method to build and calibrate custom fuel models was developed by linking Genetic Algorithms (GA) to the Rothermel spread model. GA randomly generates solutions of fuel model parameters to form an initial population. Solutions are validated against observations of fire rate of spread via a goodness-of-fit metric. The population is selected for its best members, crossed over, and mutated within a range of model parameters values, until a satisfactory fitness is reached. We showed that GA improved the performance of Rothermel model in three published custom fuel models for litter, grass, and shrub fuels (RMSE decreased by 39%, 19% and 26%). We applied GA to calibrate a mixed grass-shrub fuel model, using fuel and fire behaviour data from fire experiments in dry heathlands of Southern Europe. The new model had significantly lower prediction error against a validation dataset than either standard or custom fuel models built using average values of inventoried fuels, and predictions of the Fuel Characteristics Classification System. GA proved a useful tool to calibrate fuel models and improve Rothermel model predictions. GA allows exploration of a continuous space of fuel parameters, making fuel model calibration computational effective and easily reproducible, and does not require fuel sampling. We suggest GA as a viable method to calibrate custom fuel models in fire modelling systems based on the Rothermel model.