•  
  •  
 

Abstract

Prescribed burning is used in West Virginia, USA to return the important disturbance process of fire to oak and oak-pine forests. Species composition and structure are often the main goals for re-establishing fire with less emphasis on fuel reduction or reducing catastrophic wildfire. In planning prescribed fires land managers could benefit from the ability to predict mortality to overstory trees. In this study, wildfires and prescribed fires in West Virginia were examined to determine if specific landscape and terrain characteristics were associated with patches of high/moderate post-fire change. Using the ensemble machine learning approach of Random Forest, we determined that linear aspect was the most important variable associated with high/moderate post-fire change patches, followed by hillshade, aspect as class, heat load index, slope/aspect ratio (sine transformed), average roughness, and slope in degrees. These findings were then applied to a statewide spatial model for predicting post-fire change. Our results will help land managers contemplating the use of prescribed fire to spatially target landscape planning and restoration sites and better estimate potential post-fire effects.