Predicting the flame characteristics and rate of spread in fires propagating in a bed of Pinus pinaster using Artificial Neural Networks

Abstract

Physical and geometrical characteristics of flame propagation are very important to better understand the forest fire spread behaviour and to improve risk management tools. Having a tool to predict these characteristics is of practical and theoretical interest for a better understanding of the complex chemical and physical mechanisms which occur during forest fire phenomena. A metamodel is presented based on Artificial Neural Networks (ANNs) for estimating physical and geometrical parameters of the forest fire front, namely the rate of spread (ROS), flame height (Hf) and flame tilt angle (αf). The ANN was developed using literature data obtained from experiments of fire propagation in beds of Pinus pinaster needles. The optimal feedforward ANN architecture with error backpropagation (BPNN) was determined by the cross validation method. The ANN architecture having 5 hidden neurons proved to be the best choice. Comparing the modelled values by the ANN with the experimental data indicates that neural network model provide accurate results. The performance of the ANN model was compared with a metamodelling method using a multilinear regression approximation.