Patterns of wildfire occurrence at the landscape level were characterised during the period 1990–94 in Portugal. Based on land-cover information within 5591 burned patches (larger than 5 ha) and in the surrounding landscape, selection ratio functions were used to measure fire preference or avoidance for different land-cover types in 12 regions of the country. Shrublands were the most fire-prone land cover, whereas annual crops, permanent crops and agro-forestry systems were the most avoided by fire. In terms of forest types, conifer plantations were more susceptible to fire than eucalyptus, and broadleaved forests were the least fire-prone. There were regional variations in land-cover susceptibility to fire, which may be explained by differences in climate, management, ignition patterns, firefighting strategies, and regional availability. A cluster analysis of regional variations in selection ratios for all land covers allowed the identification of three main geographical areas with similar fire selection patterns. These results can be used for planning landscape-scale fuel management in order to create landscapes with a lower fire hazard.
We have assessed the fire proneness of the main forest types in Portugal classified according to the main species, using three different approaches: the use of resource selection ratios applied to burned patches, the proportion of randomly located plots that were burned and the proportion of burned National Forest Inventory plots. The results allowed ranking fire proneness according to the following decreasing order: maritime pine forests, eucalyptus forests, unspecified broadleaf forests, unspecified conifer forests, cork oak forests, chestnut forests, holm oak forests and stone pine forests. In order to understand the obtained results we have assessed the structure of the different forest types using the percent cover of seven vegetation layers (C1–C7), a Tree Dominance Index, a Height Index and a Cover Index (IC). Structural variables and stand composition were used to predict fire probability according to binary logistic modelling. Only four structural variables and stand composition provided significant results, the latter being the most important variable for explaining fire probability. These models were used to predict fire probability for different stand types as a function of IC.