Study Area
The High Park Fire was located in Larimer County at the northern end of the Front Range, Colorado (Figure 1). The fire was ignited by lightning on 9 June 2012 and burned more than 35 000 hectares over the course of 22 days (NWCG 2012). The burned area included approximately 5000 hectares of mature lodgepole pine. Soil burn severity was high in much of the area that had been dominated by lodgepole pine, leading the Burned Area Emergency Response (BAER) team to recommend extensive aerial mulching treatments within severely burned areas, particularly those with steep slopes or where excessive erosion would threaten public health or safety (NWCG 2012). The initial treatments were completed during April 2013, less than one year after the fire event but prior to the two-year seedling germination pulse that has been observed in lodgepole pine ecosystems (Turner et al. 1999). Most of the treatments utilized straw mulch. While some of the fire area was mulched with wood, small sample sizes prevented us from considering wood mulch.
Data Collection
Twelve field sites were sampled between June and August 2014 as part of a larger study on the combined effects of mountain pine beetle (MPB; Dendroctonus ponderosae Hopkins) mortality and fire on lodgepole pine regeneration. We located the study sites randomly, stratifying for MPB mortality and crown consumption based on 25 m resolution raster layers created by Stone (2015). During the stratification, we elected to combine the low and moderate burn severity classifications because moderate severity areas occupied narrow transition zones surrounding high severity patches, and both low (minimal needle scorch) and moderate (needles mostly scorched) severity represented the effects of surface fire (Stone 2015). We did not stratify field sites based on the presence of mulch; instead, we recorded mulch cover as we encountered it within the field sites. At each field site, we systematically placed a 25 cm × 25 cm quadrat every 10 m within a 70 m × 70 m (approximately 0.5 hectare) grid, for a total of 64 quadrat locations per site (with the exception of the first site, which was 100 m × 100 m). In each quadrat, we recorded lodgepole pine seedling abundance, and visually estimated cover percentage of the following variables: mulch, litter, total vegetation, and growable area. Litter cover consisted largely of post-fire needlecast, as there was little unburned litter in the study area. Growable area is the inverse of rock + coarse wood cover (i.e., the portion of the quadrat in which a seedling could conceivably establish).
We included other covariates known to explain the distribution of post-fire lodgepole pine seedlings (Wright 2016). To estimate the potential seed source surrounding each quadrat, we used a wedge prism to record the basal area of lodgepole pine that bore cones. We also field-verified our crown consumption variable using the wedge prism to create a binary variable classifying whether the tree crowns at each quadrat location had been consumed. If fewer than 50 % of the trees tallied with the basal area prism retained needles at the time of sampling, that quadrat site was classified as high crown consumption. Otherwise, quadrat sites were classified as low crown consumption. Typically, tree crowns were either fully consumed or intact due to the stratified sampling design, which located sites in areas of more or less uniformly high or low crown consumption. We relied on the remotely sensed pre-fire MPB layer created by Stone (2015) to classify MPB mortality as “high” in a site if greater than 50 % of the pixels within that site were classified as MPB mortality; otherwise it was classified as “low.” Most sites had close to 0 % or 100 % MPB mortality—again because our original site stratification selected areas with a uniform MPB mortality classification. We recorded the quadrat slope by placing an inclinometer along the quadrat frame. We calculated site-level elevation by averaging elevation data extracted from a 25 m digital elevation model (Stone 2015).
Analysis
We restricted our analysis to field sites in which at least one quadrat had been treated with straw mulch. Additionally, quadrats with growable area of less than 10 % were excluded from the analysis, leaving a total of 783 quadrats distributed over 12 sites. We used unbiased conditional inference trees with random effects (unbiased RE-EM trees; Fu and Simonoff 2015) to examine the drivers of lodge-pole pine seedling abundance. Unbiased RE-EM tree algorithms improve upon previously available models by incorporating conditional inference for recursive partitioning, which is less prone to bias than other recursive partitioning algorithms such as CART (classification and regression trees; Breiman et al. 1984), assumes no underlying distribution in the data, and does not require pruning to avoid over-fitting (Hothorn et al. 2006).. Additionally, unbiased RE-EM trees also incorporate random effects, which allows for the analysis of grouped data (Sela and Simonoff 2012, Fu and Simonoff 2015). We specified site as the random effect; all other variables were treated as fixed effects. All statistical analyses were performed in R (R Core Team 2015) using a function for unbiased RE-EM trees developed by Fu and Simonoff (2015).