Sajad A. Sarab¹ , Jahangir Feghhi¹ , Hamid  Goshtasb²

1 Department of Forestry and Forest Economics, Faculty of Natural Resources, University of Tehran, I.R., Iran
2 University of Environment, Karaj, Iran
Author    Correspondence author
International Journal of Molecular Evolution and Biodiversity, 2013, Vol. 3, No. 4   doi: 10.5376/ijmeb.2013.03.0004
Received: 26 Apr., 2013    Accepted: 02 May, 2013    Published: 20 Feb., 2014

This is an open access article published under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Sajad ei al., 2013, Determining the Main Parameters Affecting on Forest Fire Using MLP Neural Network (Forests of Western Iran: Izeh), International Journal of Molecular Evolution and Biodiversity, Vol.3, No.4, 15~23 (doi: 10.5376/ijmeb.2013.03.0004)

Wildfires destroy the vegetation and pose numerous environmental and social hazards. Climatic factors are among those which affect the occurrence and expansion of wildfire. determining all the climatic factors that control the happening and expansion of fire in nature requires enormous funds, time and energy which should be devoted to collection and analysis of collected data, so we decided to focus on the most important climatic factors that influence wildfires. Accordingly, this study was carried out in Izeh; a city in Khuzestan Province where is prone to wildfire due to its rich plant cover and characteristic climatic conditions. To achieve the above-mentioned objective, a an Multi-Layer Perceptron (MLP) neural network was employed. In this study, 11 parameters extracted by the synoptic were used on a monthly basis in three consecutive years. In order to develop a proper model which is able to forecast the probability of fire occurrence, several parameters were selected as inputs of the network. Based on the evaluations the best possible network was developed on the basis of maximum temperature, minimum relative humidity, mean relative humidity and daylight hours. The best model was composed of 4 inputs in the first layer, two hidden layers with 4 neurons in each one and an output layer with one variable (4-4-4-1). This study reveals that the proposed network is capable of predicting wildfire occurrence with the highest accuracy and minimum error in forests and pastures of the studied area and other areas with similar climatic conditions.

Forest fire; Climate; Hazard forecast; Multilayer perceptron; Zagros