Sites
To elucidate relationships between fire intensity, severity, and effects, we collected data from five different prescribed burns, conducted on five prairies owned by county, state, and non-profit agencies in the South Puget Sound, Washington, USA. The five sites were Glacial Heritage Preserve (46 ° 53′ 37.32″ N, 123 ° 3′ 8.64″ W), Tenalquot Preserve (46 ° 53′ 50.6394″ N, 122 ° 43″ 53.0394″ W), Mima Mounds Natural Area Preserve (46 ° 52′ 13.7994″ N, 123 ° 2′ 25.44″ W), Scatter Creek Wildlife Area (46 ° 50′ 20.76″ N, 122 ° 59′ 37.32″ W), and West Rocky Wildlife Area (46 ° 53′ 33.36″ N, 122 ° 52′ 11.2794″ W). All five prairies have similar glacial outwash soils (Spanaway, or Nisqually, or both soil types) and similar management goals, but are at different stages in the restoration process. All prairies have been actively managed to reduce the cover of invasive species, recover native forbs and bunchgrasses, and act as reintroduction sites for rare species such as the Taylor’s checkerspot butterfly (Euphydryas editha taylori [W.H. Edwards 1888]; state endangered, federally endangered) and golden paintbrush (Castilleja levisecta Greenm; state endangered, federally threatened). Prescribed burns have occurred annually across different sections of all prairies since 2001, creating slightly different fire histories in each unit. The goal for these sites is to eventually achieve a fire return interval of 2 yr to 5 yr for any given patch of prairie. The burns evaluated for this study took place between 15 August and 15 September over two years (2011 and 2012) and were characterized as having either predominantly low or moderate severity.
Two of the sites, Glacial Heritage Preserve and Tenalquot Preserve, were studied more closely for ignition pattern treatment effects on fire intensity, fire severity, and vegetation communities. The 12.2 ha unit at Glacial Heritage was burned in August 2011, marking the third fire in 10 years at that location. The 8.1 ha, August 2012 fire at Tenalquot was the first known fire to occur since European settlement of the area. Approximately half of each burn unit experienced a head fire and half experienced a backing fire. Prior to both burns, sites were boom-sprayed with a grass-specific herbicide (sethoxydim) in the spring to kill non-native grasses and then sprayed with a broad-spectrum herbicide (glyphosate) during the fall post-fire flush to kill fast-growing non-native forbs. This integrated treatment of fire and herbicide has been the most successful management strategy to control invasive species and prepare sites for seed augmentation (Stanley et al. 2008). Areas were then seeded with Roemer’s fescue (Festuca roemeri [Pavlick] E.B. Alexeev) two months post fire. Despite findings of deleterious effects of certain broadleaf herbicides on seed germination of both monocots and dicots in dry grasslands (Wagner and Nelson 2014), we have not seen any evidence of pre-emergent, off-target effects of grass-specific (A. Lincoln, Center for Natural Lands Management, Olympia, Washington, USA, unpublished data) or broad-spectrum (Stanley et al. 2008) herbicides on prairie species when they are applied at appropriate times. For this study, herbicide and seeding treatments were applied evenly across the burn units, while fire ignition pattern was not, allowing evaluation of head fires versus backing fires within this restoration context.
Sampling Design
At all five sites, we established five to ten transects that were either 5 m or 10 m apart, depending upon size and shape of the burn unit. We collected plant community and fire severity data in 1 m2 quadrats placed every 10 m along each transect. At Glacial Heritage and Tenalquot, there were 50 quadrats per burn unit; at Mima Mounds, Scatter Creek, and West Rocky, there were 100 quadrats per burn unit. At Glacial Heritage and Tenalquot, where ignition pattern treatments were applied, there were 21 to 29 quadrats per ignition pattern treatment (sample size was dependent upon where ignition lines met within the burn unit), so sample size between ignition patterns and sites varied slightly. Due to the fact that we could not replicate ignition pattern treatments within each burn, ignition pattern data are pseudo-replicated within a site (Hurlbert 1984, but see Oksanen 2003 for a rebuttal). Despite this design challenge, we still believe that these data provide a unique opportunity to evaluate fire behavior, severity, and effects associated with fire ignition patterns.
Fire Intensity
Fire intensity is typically characterized by measures of fire temperature and rate of spread (RoS; Keeley 2009). In all five burn units, we measured maximum temperature reached at the soil surface using 10 sets of copper tags painted with 24 temperature sensitive paints (OmegaLaq, Stamford, Connecticut, USA), ranging from 79 ° C to 760 ° C, in increments of approximately 28° C. We placed tags in 10 randomly chosen quadrats within each burn. At Glacial Heritage and Tenalquot, half of these quadrats experienced a head fire and half experienced a backing fire (n = 5 per burn). At these two sites, we measured RoS by timing the spread of the flaming front over a 1 m distance at two to four different locations within each burn treatment (head fire, backing fire). Additionally, we visually estimated flame lengths at the same points at which RoS measurements were taken to help characterize the fire intensity. We collected these observational data solely for the purpose of characterizing the burns and therefore did not subject them to statistical analysis.
Fire Severity
We classified fire severity using the Grassland Burn Severity classification scheme used by the National Park Service (USDI National Park Service 2003). Briefly, the scheme includes five measures of burn severity: un-burned, scorched, lightly burned, moderately burned, and heavily burned. These categories are based on the amount of organic matter consumed and how deep the soil charring occurred, with complete consumption and soil charring occurring in moderately and heavily burned categories. We recorded the percent cover of each fire severity category in the same 1 m2 plots monitored for vegetation in each burn unit one week post burn.
Plant Community
We collected species richness data in all quadrats across each burn unit. We collected pre-burn and post-burn data at the height of the growing season, between 1 June and 15 June each year. For analysis, we collapsed the data from each plot from species level to origin (native or exotic) and life form (annual or perennial; forb, grass, or shrub) to control for the species differences between sites and to facilitate comparisons with other ecosystems.
Data Analysis
The relationship between fire intensity, fire severity, and vegetation response to fire. The data used for this analysis consisted of quadrat-level fire severity and maximum temperature data from all five prairies. We conducted analyses in either PCORD v6 (MjM Software Design, Glenneden Beach, Oregon, USA) or R (version 3.1.1; R Core Team 2014). Since the fire severity data were multivariate in nature, and cover values of the five severity classes summed to 100 %, we created a Fire Severity Index (FSI). We used a Bray-Curtis distance measure and a 1-axis polar ordination with subjective poles (ranging from least to most severe) to create a score. This score had negative values, so we added 1 to ease interpretation: lower values correlated with unburned and light severity, and higher values correlated with moderate and high severity. We ran linear regressions on the 50 quadrats across the five prairies that had fire intensity measures taken (10 quadrats from each prairie) to test the relationship between fire intensity and severity.
Next, we used the TITAN function (Taxa Indicator Threshold ANalysis) in R (Baker and King 2010) to assess how vegetation changed with increasing maximum temperature and fire severity using data from the same 50 quadrats mentioned above. The TITAN function uses a novel technique combining indicator species analysis, change point analysis (Qian et al. 2003), and randomization procedures to detect how taxa change in distribution (cover, abundance, or frequency) along a gradient, as well as to detect possible community thresholds across that gradient. A community threshold represents synchronous response of multiple taxa to the same gradient value; in this case, it would be synchronous response of life form groups to fire severity. Briefly, TITAN converts indicator values to Z scores, and then sums all the z scores for each point along a gradient in which a given taxa is present or absent. The highest sum (Z) score of present taxa is evidence of a positive threshold, and the highest sum (Z) score for absent taxa is evidence for a negative threshold (Baker and King 2010). The TITAN function also gives two values, purity and reliability, that evaluate the confidence surrounding the indicator value estimates. Purity is the proportion of 500 bootstrap replicates with the same response direction (positive or negative). Reliability is the proportion of 500 bootstrap replicates with P values ≤ 0.05. Life form groups that have high purity (>0.90) have high fidelity to a given response direction, and those with high reliability (>0.90) have strong statistical confidence.
Influence of ignition pattern on fire intensity, fire severity, and plant community. To evaluate effects of ignition pattern on fire intensity (measured by surface temperatures), we ran a two-way ANOVA in SAS 9.3 (SAS Institute, Cary, North Carolina, USA), with ignition pattern and site as fixed effects and an alpha of 0.05. We also displayed descriptive statistics for rate of spread and flame length for both sites to help characterize the burn.
To analyze vegetation and fire severity response to ignition patterns, we used multi-response permutation procedures (MRPP) to compare treatment differences (head or backing fire) on the pre- to post-burn change in the plant community using a Bray-Curtis distance measure. The MRPP produces two test statistics, T and A. The T statistic describes the amount of separation between multivariate groups (similar to a t-value in a Students t-test); higher negative values mean stronger separation. The A value is an effect size statistic, describing the degree of similarity among multivariate samples within a group. A values closer to 1 signify that more samples within a group are identical, while values closer to 0 signify that more samples are different. We used non-metric multidimensional scaling (NMDS) to ordinate results using PC-ORD v6. If significant community differences were present, we used indicator species analysis (ISA) on origin-life form group to assess how the number of species within each group was associated with a given ignition pattern using the Indicator Species Function in R (Bakker 2008). We followed the same approach on the percent cover of each fire severity class per 1 m2, with the severity class as the “species.”