A multivariate analysis of biophysical factors and forest fires in Spain, 1991-2005

The main goal of this study was to explain the relationship between forest fires and different climatic, topographic and vegetation factors, establishing explanatory models from multivariate analysis. The study area comprised peninsular Spain. Two dependent variables were considered: probability of burning and fire size class, from a forest-fire map derived from visual analysis of satellite images from 1991 to 2005 (3337 fires greater than 25 ha). Logistic regression, discriminant analysis and regression trees were used to analyse the probability of burning. The models showed a significant relationship with land cover and slope, where the classification achieved an agreement of similar to 66%, and this was very similar for the three statistical methods used. Discriminant analysis and regression trees were used to model fire size class. These models appeared more related to ecozones and climatic variables (winter precipitation and mean summer temperature). In this case, the best classification results were obtained in the category of very large fires (>5000 ha), with an agreement above 80%. Regression trees achieved better results for fire size class models.