A framework for predicting stand-level fire behavior from forest community data in former prairie and savanna

Background/Question/Methods As development pressures continue to expand the extent of the wildland-urban interface (WUI), the ability to predict fire regimes there becomes increasingly important. Such predictions will be particularly valuable to land managers who seek to reduce wildfire risk and to restore imperiled ecosystems within the WUI. Our study focused on remnant and former upland prairie and oak savanna ecosystems in the southern Willamette Valley, Oregon, which were widespread prior to Euro-American settlement but now occupy less than 2% of their historic range. Prairie and savanna grasslands provide habitat for several endangered species, as well as important ecosystem services, such as the regulation of fire regimes. We sampled over 250 plots from seven sites that were grasslands with few to no trees circa 1850 but now have markedly different communities, ranging from prairie to dense forest. We collected data on community composition, topography and fuel loadings. With the BehavePlus fire model, we calculated surface and crown fire parameters: heat per unit area and crown fire transition ratio, respectively. We built two classification and regression trees (CARTs) that used plant community data to group plots on the basis of their surface and crown fire behavior.

Results/Conclusions Fuel loads differed significantly by community type, although trends in fuel loadings were neither monotonic across communities nor intuitive. Fuel characteristics were extremely sensitive to topography, and may result from successional history and the presence of exotic invasive species. Though the CARTs were statistically significant, they generally had poor predictive power, which is indicative of the amount of variability in fuel loads across community types. There was greater variability in fire behavior for communities that support more intense fires, indicating that land managers can improve the precision of their predictions by managing for less intense fire regimes. The CARTs suggested that surface fires differed among nine community types and crown fire behavior differed among five community types. There was poor agreement among definitions of communities as determined by the CARTs based upon fire behavior, and communities based on stand density and species composition. The CART community classifications are significantly better at predicting surface and crown fire behavior than conventional vegetation classification systems. A fire-behavior-based classification system improves land managers' understanding of potential fire behavior in different community types. Our results provide a quantitative basis for determining the relative benefits of different land management alternatives, including oak savanna restoration, in attenuating surface or crown fire behavior.