Human and biophysical influence on fire ignition likelihood in protected areas as a function of fire size in west-central Spain

Forest fires affect Mediterranean ecosystem, often affecting protected areas. Because these normally harbour vegetation in a better conservation state and more continuous in space, it is important to determine how they burn compared to other areas. In this study we modelled fire ignition likelihood in west-central Spain as a function of biophysical and anthropogenic variables, with a special focus on natural areas that have been recently protected by the EU Natura 2000 Network. During the 2001-2015 period more than 9000 ignitions (≥1ha) were recorded in the Spanish National Forest Fire Statistics (EGIF). We characterized each ignition point with a series of biophysical (topography, radiation and land use-land cover [LULC] types) and anthropogenic (distance to highways and roads, population density, farm density, protected areas, and forest interfaces [WUI, WAI, WGI]) variables. We built and compared statistical models of fire likelihood using the MaxEnt software for three different fire sizes: ≥ 1ha (n=9089), ≥ 10ha (n=1927) and ≥ 100ha (n=292) using a 50% random test percentage in each model. Models for the likelihood of having small and medium fires (≥ 1ha and ≥ 10ha) showed the lowest performance (AUC = 0.65, AUC = 0.73). Biophysical variables barely showed importance in explaining fire activity (except for radiation). Conversely, anthropic variables like distance to roads and settlements, population density, and farm density were important predictors. Models for fires ≥ 100ha showed the best performance (AUC = 0.84). Large fire likelihood was mainly explained by biophysical variables like radiation, elevation and some LULC types (e.g., grasslands, agrarian, shrublands, and oak forests), compared to those of anthropic origin. Protected areas showed the greatest contribution to explain the ignitions of large fires. Our models highlight the different relations of biophysical and anthropogenic variables with the likelihood of fire ignitions according to their final size.