A multidimensional statistical framework to explore seasonal profile, severity and land-use preferences of wildfires in a Mediterranean country/Un cadre statistique multidimensionnel pour explorer le profil saisonnier, la severite et le ciblage des terres par les feux de forets dans un pays mediterraneen/Un marco estadistico multidimensional para explorar el perfil estacional, la gravedad y las preferencias

This study analyses spatio-temporal patterns of wildfires in Greece using a multidimensional statistical framework based on non-parametric correlations, principal component analysis, clustering and stepwise discriminant analysis. Specifically, we assess the frequency, seasonal profile, severity and land-use type of 135 178 wildfires which occurred between 2000-2012 in Greece, one of the countries most affected by fire in Europe. Our results show that both the number of fires and the average size of the area covered by fire show a specific seasonal pattern with a marked increase during the dry season. Principal component analysis identifies three dimensions linked with the main type of land-use affected by the fires: (i) medium and large fires primarily affected landscapes composed of forests, mixed woodlands/shrublands and croplands; (ii) small fires mainly affected fragmented landscapes, i.e. those with mosaics of different crops, market gardens and non-vegetated, abandoned or marginal areas; (iii) fires affecting wetlands and pastures occurred particularly in late summer and showing medium-low severity. Hierarchical clustering highlights similarities in spatio-temporal patterns between fire indicators (ignition date, burnt land cover classes, fire size, fire density). K-means clustering allows us to distinguish between low-severity fires occurring in the wet season from intense and frequent fires occurring in the dry season but with distinct land-use selectivity. The research reported here contributes insight into the complexity of wild fires in the Mediterranean region and supports the design of more effective fire prevention measures including sustainable forest management practices and careful regional planning to minimise risk factors. Keywords: fire frequency, fire severity, land-use types, fire-smart management, Greece Cette etude analyse les traces spatio-temporels des feux de forets en Grece en utilisant un cadre statistique multidimensionnel base sur des correlations non-parametriques, une analyse de l'element principal, et une analyse discriminatoire en bloc et echelonnee. Nous evaluons specifiquement la frequence, les profils saisonniers, la severite et le ciblage des terres de 135 178 feux de forets subis par la Grece, l'un des pays les plus affectes par le feu en Europe, entre 2000 et 2012. Nos resultats montrent que le nombre de feux, ainsi que la taille moyenne de l'etendue couverte par le feu suivent un profil saisonnier specifique, avec