The number of large, high-severity fires has increased in California montane forests during the past thirty years due to fuel accumulation related to a century of fire exclusion, and recent warming. Fires generate a mosaic of vegetation types and structures which can alter future vegetation patterns through self-reinforcing fire-vegetation feedbacks. Severity of subsequent fires, or reburns, has often matched severity patterns of prior burns, especially in the case of high severity fire begetting high severity fire. Consistency in reburn severity behavior may partly reflect the effects of topography on local climate and fire behavior, but it is also likely to arise due to different vegetation types and structures that create spatial variability in different levels of fire hazard based on fuel characteristics. Our objectives were to determine the critical controls on fire severity for initial fires in fire excluded forests, and reburns. We were particularly interested in whether severity patterns in reburns were more consistent with strong topographic controls or with fire-vegetation interactions. We used Random Forest, a machine learning technique, to identify the influence of terrain, fire weather, and fuels on patterns of fire severity for >100 wildfires with daily fire progression maps between 2001 and 2017. Separate models were developed for initial fires and reburns. Input variables for the analysis included remote sensing derived variables for fire severity, and pre and post fire vegetation and fuels, terrain based variables derived from digital elevation models, and daily fire weather derived from gridded data sets.
Results/Conclusions
Variable importance for the initial fire model indicate that terrain, weather and fuels weather played strong roles in determining fire severity. For the reburn model, previous fire severity was by far the most important variable with moderate influence by other vegetation, fuel, and terrain variables. Weather variables were notably less influential than in the initial burn model. Our results suggest potential for persistent vegetation type conversions that are in part driven by the stochastic influences from weather conditions in the initial burn that are not readily explainable using topographic setting alone. This study highlights the potential for fire-vegetation interactions to generate new and persistent mosaics of fire severity and vegetation structure that are both desirable and undesirable for land managers.