Managers at Eglin Air Force Base (EAFB) in the Florida Panhandle use frequent fire to sustain healthy large stands of fire-dependent longleaf pine (Pinus palustris Mill.). Overstory stand structure, understory plant diversity, surface fuel structure, and fire regime are all linked through mutual feedbacks. However, only overstory stand structure can be characterized with sensitivity using airborne lidar. We used canopy height and density metrics generated from the lidar to predict maps of tree density, basal area, and dominant tree species; these 3 responses were summarized from tree measures tallied in 0.65 ha rectangular monitoring plots (n = 195) distributed randomly across EAFB. Our question: Could we use these maps to “indirectly” predict surface fuels and understory plant diversity, by virtue of their links to overstory structure? We employed imputation as our modeling strategy because it allows for prediction of multivariate responses, and the Random Forests machine learning algorithm to assign the predictive model weights. We also applied a Ripley’s L statistic to the tree stem map data, to see if the frequent, low intensity fire regime was causing naturally regenerating longleaf pine to be either more dispersed or more clustered than random.
Results/Conclusions
We found mapped predictions of most surface fuel components and understory plant diversity, which had no weight in the models, were nonetheless significantly correlated with the overstory structure predictions upon aggregation to EAFB land management blocks. Across 425 management blocks, the absolute values of Spearman rank correlations between 16 measures of surface fuels or understory plant diversity and 2 measures of overstory structure (tree density and basal area) ranged from 0.11 to 0.64; only 2 of 32 were not significant. Aggregation to the management block level makes sense because decisions about whether or not to burn are implemented for entire management blocks. Indeed, surface fuel and understory plant diversity measures were also significantly correlated to fire history variables summarized from an independent geodatabase of fire history records. Across 425 management blocks, the absolute values of Spearman rank correlations between 16 measures of surface fuels or understory plant diversity and 2 measures of fire history (number of fires and years since last fire) ranged from 0.09 to 0.69; again, only 2 of 32 were not significant. We also determined that naturally regenerating longleaf pine trees have a clearly clustered distribution. We conclude that frequent fires promote patchy stands and plant diversity.