Probabilidade bayesiana e regress？o logística na avalia？？o da susceptibilidade de ocorrência de incêndios de grande magnitude

the aim of this paper is to apply a susceptibility model to high magnitude wildfires - with the latter defined as the small fraction of the total number of occurrences that causes most of the annual damage. this type of frequency/ magnitude relation is characteristic of wildfire regimes in southern european countries. drawing on the analysis of burnt-area maps for the period 1990-2007 in the castelo branco district, a wildfire classification method is put forth and a model is tested using two alternative data integration techniques: one based on logistic regression, the other on bayesian probability. the results indicate that the bayesian technique has slightly greater predictive capability and confirm that the proposed model is adjusted to the behaviour of wildfires considered to be of high magnitude. although the suggested model may usefully undergo future improvements in order to increase its predictive capability, it can already be used to complement other forms of susceptibility/hazard analysis, by highlighting the areas that are most likely to be affected by the most destructive wildfire events.