LAUSR

Abstract The increasing occurrence of large and severe fires in Mediterranean forest ecosystems produces major ecological and socio-economic damage. In this study, we aim to identify the main environmental factors driving fire severity in extreme fire events in Pinus fire prone ecosystems, providing management recommendations for reducing fire effects. The study case was a megafire (11,891 ha) that occurred in a Mediterranean ecosystem dominated by Pinus pinaster Aiton in NW Spain. Fire severity was estimated on the basis of the differenced Normalized Burn Ratio from Landsat 7 ETM +, validated by the field Composite Burn Index. Model predictors included pre-fire vegetation greenness (normalized difference vegetation index and normalized difference water index), pre-fire vegetation structure (canopy cover and vertical complexity estimated from LiDAR), weather conditions (spring cumulative rainfall and mean temperature in August), fire history (fire-free interval) and physical variables (topographic complexity, actual evapotranspiration and water deficit). We applied the Random Forest machine learning algorithm to assess the influence of these environmental factors on fire severity. Models explained 42% of the variance using a parsimonious set of five predictors: NDWI, NDVI, time since the last fire, spring cumulative rainfall, and pre-fire vegetation vertical complexity. The results indicated that fire severity was mostly influenced by pre-fire vegetation greenness. Nevertheless, the effect of pre-fire vegetation greenness was strongly dependent on interactions with the pre-fire vertical structural arrangement of vegetation, fire history and weather conditions (i.e. cumulative rainfall over spring season). Models using only physical variables exhibited a notable association with fire severity. However, results suggested that the control exerted by the physical properties may be partially overcome by the availability and structural characteristics of fuel biomass. Furthermore, our findings highlighted the potential of low-density LiDAR for evaluating fuel structure throughout the coefficient of variation of heights. This study provides relevant keys for decision-making on pre-fire management such as fuel treatment, which help to reduce fire severity.