Characterizing the Spatial Patterns and Spatially Explicit Probabilities of Post-Fire Vegetation residual patches in Boreal Wildfire Scars

Wildfire is one of the main natural disturbances that consume a substantial amount of forest cover, influencing and reshaping the landscape mosaic of boreal forests. Wildfires do not burn the entire landscape; they rather create a complex mosaic of post-fire landscape structure with different degrees of burn severity. The resulting spatial mosaic includes fully burned, partially burned, and unburned areas. Even though the most visible components of a fire disturbed landscape are the completely burned areas, a considerable number of residual patches of various size, shape, and composition are retained following a fire. The residual patches refer to remnants of the pre-fire forest ecosystem that left completely unaltered within the fire footprint. Improved understanding of the patterns and characteristics of wildfire residuals provides insights for investigating the effects of fire disturbances, emulating forest disturbances in harvesting operations, and improving forest management planning. Knowledge about the post-fire residuals relies on how well we measure the patterns and characteristics of post-fire residuals, determine the factors that explain their occurrence and patterns, and what consistent measurement framework we use to understand the patterns and predict their likely occurrence. In this study, the patterns and characteristics of post-fire residuals was initially examined based on eleven boreal wildfire events within northwestern Ontario; each ignited by lightning and never suppressed. The wildfire events were occurred in ecoregion 2W during the fire seasons of 2002 and 2003. In order to design a consistent and repeatable method for measuring the patterns of residuals, an integrate approach has been designed. This involves assessing the spatial patterns where the composition, configuration, and fragmentation of residual patches were assessed based on selected spatial metrics; examining the importance of predictor variables that explain residuals and their marginal effects on residual patch occurrence using Random Forest (RF) ensemble method; and developing a spatially explicit predictive model using the RF method where the combined effects of the variables were examined. Finally, the three approaches are applied and evaluated using a recent and independent data from the extensive RED084 wildfire event that occurred in 2011 within the adjacent ecoregion (3S). The effects of analytical scale (i.e., spatial resolution) on characterizing the spatial patterns, determining the relative variable importance, and predicted probabilities of residual patches are assessed. The results show that the composition and configuration of wildfire residuals vary as a function of measurement, spatial resolutions, and fire event sizes, suggesting the variation in fire intensity and severity across the fire events. The patterns of wildfire residuals are also sensitive to changing scale, but the responses of the spatial metrics to changing spatial resolutions are grouped into three categories: monotonic change and predictable response in which three shape related metrics (LSI, MSI, and FRAC) show a predictable responsible; monotonic change with no simple scaling rule; and non-monotonic change with erratic response. The results also reveal that the factors that are incorporated in this study interactively affect the occurrence and distribution of residual patches, but natural firebreak features (e.g., wetlands and surface water) were among the most important predictors to explain wildfire residuals. Furthermore, the model implemented to predict residual patches has a reasonable or high predictive performance (‘marginal’ to ‘strong’ model performance) when it was applied in wildfire events that occurred in the same ecoregion. However, the predictive power of the model is low for the independent fire event (RED084). The overall findings of this dissertation reveal that the 1) predictive model based on RF is robust enough to determine the relative importance of the predictors and their marginal effect; 2) the model was flexible enough to identify areas where wildfire residuals are likely to occur; and 3) there is a repeatable, robust measurement framework for characterizing residual patches and understanding their variability across different wildfire events.