Title: Predicting the Occurrence of Lighting/Human-Caused Wildfires Using Advanced Techniques of Data Mining
Citation: Proceedings of the 6th International Workshop of the EARSeL Special Interest Group on Forest Fires p. 119-122
Publisher: OPOCE
Publication Year: 2007
JRC N°: JRC43121
URI: http://publications.jrc.ec.europa.eu/repository/handle/JRC43121
Type: Articles in periodicals and books
Abstract: The fire scientific community and even more the fire managers are trying to assess and describe the fire occurrence parameters that are involved in the fire phenomena. The heterogeneous nature of the variables in-volved in the process lead to the use of complex predictive models which have to deal with their non-normal dis-tribution in the spatial and temporal domains. Besides, human and natural causes of fire ignition need to be inves-tigated by means of predictor variables that can be spatially related under non-linear and non-additive relationships. Due to these actual limitations, the fire modelling community is seeking innovative and flexible models able to deal with these complex data relationships. This paper aims to overcome the limitations of the common regression techniques by testing alternative method-ologies (Regression Tree) based on powerful models part of the family of the data mining or/and non-parametric techniques. In the framework of long-term wildfire risk assessment, lightning/human-caused occurrence has been used in this study as response variable against several predictors. The wildfire occurrence variable has been calcu-lated by kernel density estimation technique using ignition points collected in the Aragón¿s autonomy (Spain) in a 19 years period (1983-2001). The predictor variables included physical and human layers considered as relevant in the fire distribution processes (road and population density, climate condition, topographic aspects, etc.). Model performance was then evaluated looking the explained variance and the RMSE. The results enlightened the potential of the new considered model that resulted to be able to properly identify the relationships between the considered variables and explain the fire phenomena in the study area.
JRC Directorate:Sustainable Resources

Files in This Item:
There are no files associated with this item.

Items in repository are protected by copyright, with all rights reserved, unless otherwise indicated.