Spatially modeling wildland fire severity in pine forests of Galicia, Spain

Abstract

Fire is a major disturbance in forests and one of the most important carbon emissions sources, which contributes to climate change. Carbon emissions are directly correlated with the degree of organic matter consumption or fire severity. Gaining knowledge about the relative strength of the various explanatory variables is essential to mitigate its environmental impact. We tested an approach that combines wind modeling, light detection and ranging (LiDAR), remotely sensed vegetation indices and topography data for assessing the occurrence of high-severity fire using the random forests ensemble learning method. Data from four wildfires that occurred in Galicia (northwestern Spain) were used to exemplify the application of this approach. The models predicted high-severity occurrence with a classification accuracy ranging from 77 to 94%. High-severity fire occurred more frequently in areas of high simulated wind speed, and more pronouncedly, for cases reported as wind-driven fires. High severity also occurred more frequently in areas of high terrain roughness, on sunny slopes and in low canopy base height stands. This approach allowed predicting spatially explicit fire severity at a mean scale level (resolution of 25 m) with accuracy rates from 80 to 95%. This approach may be helpful for fire managers when delimiting and planning fuel treatments for severity mitigation or during fire suppression, and for post hoc case studies.