R script with data for vegetation and macropod scat analysis
Chard, Matthew (2022), R script with data for vegetation and macropod scat analysis, Dryad, Dataset, https://doi.org/10.5061/dryad.dr7sqvb0r
Fire and herbivores alter vegetation structure and function. Future fire activity is predicted to increase, and quantifying changes in vegetation communities arising from post-fire herbivory is needed to better manage natural environments.
We investigated the effects of post-fire herbivory on understory plant communities in a coastal eucalypt forest in south-eastern Australia. We quantified herbivore activity, understory plant diversity, and dominant plant morphology following a wildfire in 2017 using two sizes of exclosures. Statistical analysis incorporated the effect of exclusion treatments, time since fire (TSF), and the effect of a previous prescribed burn.
Exclusion treatments altered herbivore activity, but TSF did not. Herbivory reduced plant species richness, diversity and evenness and promoted the dominance of the most abundant plants within the understory. Increasing TSF reduced community diversity and evenness and influenced morphological changes to the dominant understory plant species, increasing size and dead material while decreasing abundance. We found the legacy effects of a previous prescribed burn had no effect on herbivores or vegetation within our study.
Foraging by large herbivores resulted in a depauperate vegetation community. As post-fire herbivory can alter vegetation communities, we postulate that management burning practices may exacerbate herbivore impacts.
Future fire management strategies to minimise herbivore-mediated alterations to understory vegetation could include aggregating management burns into larger fire sizes or linking fire management with herbivore management. Restricting herbivore access following fire (planned or otherwise) can encourage a more diverse and species-rich understory plant community. Future research should aim to determine how vegetation change from post-fire herbivory contributes to future fire risk.
We conducted this study at Booderee National Park (BNP; 35.1489415° S, 150.6454625° E; Figure 1) on the southeast coast of Australia, approximately 200 km south of Sydney. The Park is ~6,500 ha in area and co-managed by the Wreck Bay Aboriginal Community and Parks Australia. The dominant vegetation class in the park is Sydney Coastal Dry Sclerophyll Forest (45% of the park area) which is characterised by canopy species of Eucalyptus pilularis, Corymbia gummifera, and Eucalyptus botryoides, midstory species of Banksia serrata and Monotoca eliptica, and an understory dominated by Pteridium esculentum, Lomandra longifolia and Lepidosperma concavum (Taws, 1997).
Three species of macropod in BNP meet the “large herbivore” classification (> 2 kg; sensu Danell et al., 2006). They are the eastern grey kangaroo (Macropus giganteus); swamp wallaby (Wallabia bicolor); and red-necked wallaby (M. rufogriseus). No other large terrestrial herbivore species are currently found in BNP. All three macropods have previously demonstrated pyric herbivory responses with most studies identifying a preference for recently burnt patches due to a higher quality of foraging resources (Southwell & Jarman, 1987; Meers & Adams, 2003; Foster et al., 2015; Parkins et al., 2019).
We quantified the interacting effects of post-fire herbivory on vegetation communities using two randomised, blocked experiments. In June 2012, we established three blocks of six 25 x 25 m plots (0.0625 ha, hereafter referred to as ‘small’ plots) within Sydney Coastal Dry Sclerophyll Forest, with plots spaced 150 m apart and blocks 2 km apart. We manipulated grazing pressure by macropods using three methods of fencing: (1) open (i.e no fencing), (2) partial fencing - intermediate access with gates at two corners of the plot which were opened or closed at two-month intervals and (3) closed (completely fenced). We constructed 1.1 m tall fences which prevented access by macropods (Foster et al., 2015). We conducted low-intensity, prescribed burns in August 2012 within half of the plots in each block so that each fencing treatment had one burnt and one unburnt pair. Controlled fires were extinguished after burning a 50 x 50 m area and removing approximately 95% of the understory vegetation. This facilitated examination of two burning treatments across three herbivory measures within all three blocks. Although a wildfire in 2017 burnt all plots, we describe our small plots as ‘burnt’ or ‘unburnt’ as per the initial prescribed burn conducted in 2012.
In September 2017, a wildfire burned 1,600 ha of BNP, including each small experimental block, again removing approximately 95% of the understory vegetation. In our study, we recorded time since fire as time since the 2017 wildfire. In July 2018, we established an additional four blocks of two 200 x 200m plots (4 ha, hereafter referred to as ‘large’ plots) in forest vegetation. Large plots were spaced 300 m apart and blocks at least 2 km apart. Again, we manipulated herbivore grazing pressure using two randomly allocated fencing treatments: (1) open (no fencing); and (2) closed (completely fenced).
We conducted scat surveys every two months from October 2018 to February 2020 in all plots, within two 25 m x 2 m transects (small plots), and four 50 m x 2 m transects (large plots), in which macropod scats were counted and removed from the transect. We used macropod scat counts as an index of herbivore activity (Murphy & Bowman, 2007). We conducted vegetation surveys annually in spring, in all plots. We used five point-intercept transects of 20 m (small plots) and four 50 m transects (large plots) within each plot to record understory plant species (< 3 m in height) at 1 m intervals. We used site-level data to calculate four vegetation community measures: species richness, diversity (Simpson’s reciprocal index – 1/D), evenness (Shannon evenness index), and dominance (Berger-Parker index; Magurran, 2013). Using the same point-intercept transects, when a bracken plant was present, we recorded its physical attributes including width (measured parallel with the transect), height to bottom-most frond, top height, and percentage of dead vegetation. We also recorded the number of bracken plants intercepted at the 20 or 50 points along each transect. Both scat and vegetation surveys encompassed the post-wildfire period from September 2018 to February 2020.
We analysed the influence of exclosure fences, time since fire (TSF), and the 2012 prescribed burn on: (1) scat counts, (2) plant community measures and (3) bracken attributes in R (Core, R Team, 2016). We fit models from a candidate set of nine models (small plots) and two models (large plots) for each response in a Bayesian framework using the ‘brms’ package (Bürkner, 2017). The models we constructed used all possible combinations of exclusion treatment (open/partial/closed), TSF, and prescribed burn (burnt/unburnt) for each response variable (Table S1; S2; S3). We selected appropriate regression distributions for each variable after testing for assumptions of normality and homogeneity of variance (see Tables S1; 2; 3; Hanea et al., 2015).
Our response variables were: (1) number of macropod scats, with scat counts being summed at two-month intervals for small plots to allow for effective analysis of the partial treatments (as every second count was effectively zero); (2) understory plant richness, diversity, evenness, and dominance, with vegetation measures calculated using the ‘diversityresult’ function from the ‘BiodiversityR’ package (Kindt & Kindt, 2019); and (3) bracken width, height to bottom frond, top height, count of individuals, and percentage of dead material. We treated TSF as a continuous variable for scat surveys, standardised using the ‘scale’ function so that the mean was zero with a standard deviation of one. We included season (for scat surveys) and block as a fixed effect in each model as well as the random effect of plot. We selected appropriate priors for each model and the Rhat values were deemed acceptable (all values = 1; Gelman & Rubin, 1992).
The models were fit using Markov chain Monte-Carlo methods. We ran four chains, each with with 3,000 iterations with the first 1,000 iterations discarded as burn-in for the sampler. We based our inference on the importance of the hypothesised interactions by selecting the most parsimonious model using lowest WAIC and simplest model using the ‘loo’ package (Burnham & Anderson, 2002; Vehtari et al., 2017). We present results for most parsimonious models for macropod scats, vegetation community measures and bracken morphology from small and large plots.