Does livestock fencing progress woodlands towards desired states?
(Part Three)
Welcome to part three of my research project. If you got through part one and two, and you're here, then well done, you must be a sucker for punishment.
Introduction
Livestock exclusion is a major tool to restore vegetation in agricultural landscapes of Australia (Prober et al. 2011b), China (Su et al. 2015), northern Sinai (El-Bana et al. 2003), north America (Yeo 2005), Saudi Arabia (Al-Rowaily et al. 2015) and Argentina (KrÖPfl et al. 2013), and is often the first step in restoring ecosystems degraded by farming. Livestock grazing in Australian temperate grassy woodlands has converted understories formerly dominated by tall perennial native grasses and geophytes to assemblages of shorter native grasses and exotic annual grasses and herbs (Dorrough et al. 2004a; Dorrough et al. 2004b; Lunt et al. 2007a; Pettit et al. 1995; Prober and Thiele 1995). Compositional shifts, induced by heavy ungulate grazing has dramatically reduced the diversity of native woodland flora. In 2010 it was estimated that 92% of grassy box woodlands have been extensively cleared for agriculture and modified for livestock grazing (Department of Environment Climate Change and Water NSW 2010.). A number of Australian temperate grassy ecosystems are listed as threatened under national legislation.
Fencing to exclude livestock grazing is a common first-management-step for restoration of woodland biota in agricultural landscapes. However, results vary and are hard to predict (Dorrough 2012). Some studies show positive results, with fenced sites having increased canopy cover and recruitment, greater species richness and cover of native understorey plants, lower soil surface compaction and convergence of species composition with reference sites over time (Prober et al. 2011b; Spooner et al. 2002). The effectiveness of fencing to reduce exotic cover is also uncertain. Lunt et al. (2007b), Briggs et al. (2008) and Pettit et al. (1995) all found that grazing exclusion had no effect on exotic species richness and exotic cover. Conversely, Prober et al. (2011b) found grazing exclusion reduced exotic cover in temperate woodlands of Western Australia. Western Australian York Gum woodlands are considered low productivity and occur in a region that receives 320–469mm mean annual rainfall (Prober et al. 2011b) compared to Jarrah woodlands which receive 650 - 700 mean annual rainfall (Pettit et al. 1995). Resource-poor ecosystems have been shown to respond to livestock removal through increased native species dominance and inter-tussock plants (Lunt et al. 2007a; Lunt et al. 2007b). This may be due to resource constraints that result in slower rates of competitive exclusion, assuming that prior overgrazing has not depleted native seed banks.
The variability in responses to fencing may be broadly related to four explanatory factors: site productivity (availability of water and/or nutrients), level of degradation, plant palatability and environmental conditions.
The intermediate disturbance model posits that ecological communities attain their highest diversity under moderate levels of disturbance because such regimes strike a trade-off between opposing forces of species elimination (Connell 1975). On one hand, low levels of disturbance allow elimination of weak competitors by strong competitors. On the other, high levels of disturbance eliminate species that are unable to respond by regenerating. Optimal coexistence occurs when levels of disturbance are sufficient to disrupt competitive exclusion but insufficiently severe to directly eliminate many species. In grazed ecosystems, maximum diversity may be expected at intermediate levels of grazing because this reduces biomass of dominant plants and allows recruitment of less competitive small plants. Lunt et al. (2007b) suggested that this relationship is most likely to hold in high-productivity systems, such as basalt grasslands (Tremont 1994). Support for the intermediate disturbance model is equivocal in grazed ecosystems. It was supported by a study of temperate grasslands in Victoria by Schultz et al. (2011), who found species richness declined in the absence of livestock grazing, due to unmanaged phytomass accumulation and consequent suppression of non-dominant species. In contrast, Lunt et al. (2007b) found grazing exclusion had no effect on native species richness. They attributed this result to low grazing levels (0.3 – 0.6 DSE) prior to exclusion, initial site degradation and site productivity.
Understanding the circumstances in which the intermediate disturbance model applies and those in which it fails will improve the ability of managers to predict outcomes of restoration through adjustments to grazing intensity. For example, if the dominant species are unpalatable to livestock, grazing cannot promote species diversity. In fact, the opposite may be expected. This has been well illustrated in grasslands of temperate NSW dominated by unpalatable grasses such as Nassella trichotoma (serrated tussock) (Lunt et al. 2007a). In these cases, livestock preferentially graze associated species that are more palatable and nutritious than Nassella trichotoma, reducing their abundance and gradually promoting a further dominance of unpalatable species (Campbell 1998). Reduced competition from other species further promotes seedling recruitment of Nassella, leading to increasing dominance by an unpalatable species in an increasingly species-poor pasture and also to a decline in carrying capacity for livestock (Campbell 1998).
Alternative stable state theory (Scheffer et al. 2001; Suding et al. 2004) proposes that ecological thresholds define the limits to the recovery of degraded ecosystems. Several studies support the existence of alternative stable states and thresholds delimiting healthy and degraded woodlands (Pettit and Froend 2001; Price et al. 2010; Spooner and Briggs 2008). In these cases ecological thresholds may be defined as boundaries of species composition or function between alternative ecosystem states (Spooner and Allcock 2006). Once an ecosystem passes a threshold to a more degraded state, it is posited that the transition cannot be reversed by simply removing the disturbing pressure (Yates and Hobbs 1997b) and usually requires intervention to promote recovery (Standish et al. 2014).
Assuming threshold behaviour of grazed woodlands, several authors suggest that fencing to exclude livestock alone is not enough to promote the recovery native woodland flora (Spooner et al. 2002; Yates and Hobbs 1997b). Farming appears to leave legacies that form ecological barriers inhibiting the regeneration of native species. These legacies include changes in soil nutrient concentrations (caused by fertiliser application to increase productivity), degradation of surface soil structure, reduced soil water infiltration rates, changes in microclimate at ground level and in the soil and depleted native seed banks (Cramer et al. 2007; Prober et al. 2002b; Spooner et al. 2002; Standish et al. 2006; Standish et al. 2007; Yates and Hobbs 1997a; Yates and Hobbs 1997b; Yates et al. 1995). It is unknown how long these legacies persist once land is retired from agricultural production (Standish et al. 2006). Prober et al. (2011b) found no evidence to suggest fencing led to declines in topsoil nutrients concentrations over 22 years of livestock exclusion. When native grassy systems that have been transformed by agriculture cross restoration thresholds, grazing tolerant species well adapted to disturbance persist while more sensitive species expire. Prober et al. (2002b) found exotic annuals were more abundant and native species richness lower in places with unnaturally elevated soil nutrients or soil moisture levels. Even long after retirement, plant-soil feedbacks can maintain a state of exotic annual dominance (Prober et al. 2002a; Prober et al. 2002b; Standish et al. 2009). In these situations, recovery of native species is unlikely without intervention and fraught with failure if managed symptomatically.
Grassy ecosystems may undergo fluctuations related to seasonal and interannual variation in weather (Burrows 2004; Schultz et al. 2014). This variation among seasons may have contributed to findings of Briggs et al. (2008) and may sometimes obscure or confound directional changes related to grazing. For example, rainfall had more influence on annual species richness and cover than grazing exclusion (Austin et al. 1981; Lunt et al. 2007b). Jackson and Bartolome (2002) proposed a model of Californian grasslands in which seasonal weather patterns were the primary determinant of community change rather than grazing. Similarly, climate and rainfall are considered the primary factors determining the distribution and function of tropical savannahs (Sankaran et al. 2005; Scholes and Walker 2004).
With so much variability and uncertainty on the effectiveness of livestock fencing we sought answers to two important questions that remain ambiguous in the literature. Firstly, does fencing initiate the divergence of species composition and soil conditions from continually grazed vegetation over time and, more specifically, does it initiate convergence in these properties with reference site conditions over time? Secondly, how does the response to fencing depend on legacies from prior livestock grazing and/or environmental conditions?
This study uses cross-fenceline contrasts of grazed and ungrazed woodlands, and explicit comparisons with nearby reference woodlands to explore the following three hypotheses:
Species composition and soil conditions of fenced plots diverge from unfenced plots over time.
Species composition and soil conditions of fenced plots converge with reference site conditions over time
Responses to fencing depend on time since exclusion of livestock grazing, degree of degradation of the initial state and/or environmental conditions.
To test these hypotheses, compositional trajectories based on chronosequences of fenced and unfenced plots were assessed relative to ungrazed reference sites using ordinations of floristic and topsoil chemistry data. Convergence and divergence were modelled as a function of time-since-livestock-exclusions, initial state and environmental variables.
While a number of previous studies have used fences to compare the effects of livestock exclusion on woodland restoration, we believe this is one of the first to evaluate the rate of compositional convergence of fenced vegetation to reference communities. Furthermore, very few studies have directly evaluated recovery rates of topsoil conditions as a result of livestock exclusion (Prober et al. 2011b).
Methods
Experimental design
The study used cross-fence line contrasts of paired sites matched to nearby reference sites with no history of farming. Each pair consists of a fenced plot excluding livestock grazing (treatment plot) and an unfenced plot being actively grazed by livestock (control plot). Paired plots were located within 100 metres of each other, more than 5 metres from fence lines, and matched as closely as possible in dominant tree species, structure, aspect, slope, topographic position and land management history to their treatment site. Remnant vegetation in little-used country cemeteries were targeted as reference sites. These small remnants were fenced and protected from farming disturbances during European settlement and commonly support grazing-sensitive species (Prober and Thiele 1995; Stuwe and Parsons 1977).. Site selection criteria were developed and strictly applied to minimise the constraints that are typical of space-for-time (SFT) substitution comparative studies (Pickett 1989) and limit any non-treatment differences between plots within each pair. The criteria ensured that control plots (grazed sites) and treatment plots (exclusion sites) shared a common initial state. That is, the plots within each pair shared similar grazing history; the control plot maintained a similar grazing regime and intensity to that of the treatment plot before it was destocked.
Landholder Interviews and Site Selection
Informal landholder interviews (described in (Briggs et al. 2008; Prober et al. 2011a) were the first means of further prioritising candidate sites. The aim of the interview was to gain an appreciation of the level and consistency of grazing pressure imposed on the vegetation prior to retirement and vegetation in actively grazed paddocks. Landholders were asked about the number, type of animal and grazing frequency. This allowed a standardised stocking rate of dry sheep equivalent (DSE) per unit area (hectares) to be estimated for each paddock. Estimating stocking rates of paddocks gave insight into whether grazing histories matched across the fence-line.
Quantitative stocking rates were classified into four broad qualitative grazing intensity classes: negligible/nil (0-0.1 DSE), light (1.5 DSE), moderate (3.5 DSE) and heavy (5 DSE). Grazing intensity classes were based on carrying capacity estimation of the land, as well as landholder opinion. For example, a DSE of 5 on highly productive soils would impose a lower grazing pressure than if the same DSE of 5 was applied to low productive soils.
Below is the list of landholder interview questions.
What are the dominant trees and understorey cover?
What type of stock do you run?
How long has the paddock been retired for?
What was the stocking rate (dry sheep equivalents (DSE)) and grazing technique (set stocked, rotational) of the retired paddock?
Is there another paddock that is still being actively grazed nearby with a similar grazing history to that described in question 4?
Has grazing pressure been consistent of the grazed paddock since retirement of the destocked paddock?
A site reconnaissance was scheduled if a property had the following attributes:
A canopy (or evidence of) dominated by White Box, Yellow Box or Blakely’s Red Gum, with a grassy understorey.
A retired paddock between 1 and 20 years.
A nearby actively grazed paddock sharing similar habitat and landscape position to the retired paddock.
Similar stocking rates on both the grazed and retired paddock.
Study Area and Study System
The study area covers the Central West of NSW, Australia (Figure 1). The study area is characterised by steep hills in the east, to gently undulating landscapes and alluvial river plains. The geology is based on a series of faults and folds of sedimentary parent material, interspersed with granite intrusions, basalts from old volcanoes and overland flows. Soils range from highly fertile alluvials and basalts to deep sands of the Pilliga sandstones. Rainfall is relatively evenly distributed throughout the year and ranges from 550 to 900 mm per year.
The study system closely follows the legal definition of box gum woodland and derived native grassland under the National Environment Protection and Biodiversity Conservation Act 1999 and NSW Bioidveristy Conservation Act 2016 (then Threatened Species Conservation Act 1995). Prior to European settlement grassy box woodlands formed an almost continuous band covering several million hectares on the slopes and tablelands of Victoria, NSW and southern Queensland (Beadle 1981). The community’s preference for productive soils, has led to extensive clearing for agriculture and modification for livestock grazing. It has been estimated that 93% of its original extent has been cleared in NSW (Department of Environment Climate Change and Water NSW 2010), and is poorly conserved in the national conservation reserve system (Specht 1981; Yates and Hobbs 1999).
The predominant canopy comprises White Box (Eucalyptus albens), Yellow Box (Eucalyptus melliodora) and/ or Blakely’s Red Gum (Eucalyptus blakelyi), and the ecological community that they dominate is thus coined ‘Box Gum Woodland’. The latter two species become locally dominant along non-permanent watercourses or on deeper soils of valleys (Moore 1953). Shrubs are generally sparse or absent, though they may be locally common. The trees form an open canopy above a rich diversity of graminoids and herbs (Prober and Thiele 1993). The community becomes shrubbier on poorer, shallower soils (Prober 1996), owing to greater reservation in national parks and conservation reserves and exclusion from legal descriptions.
Figure 1: Location of study sites, major towns, roads and National Parks (green shading). Pale green shading is private land, mostly with a >180 year history of livestock grazing. Stars represent reference site locations. Circles represent the location of paired sites (fenced and unfenced).
Sampling
Sampling was undertaken over six field survey campaigns between 28 January to 30 April 2015. Paired plots were located as close together as possible and stratified both temporally and spatially. Sample sites were grouped into four age classes. Age classes were based on livestock exclusion time (in years) of fenced plot. The four age classes were 0-4, 5-9, 10-14 and 15-19 years of livestock exclusion. Stratification ensured a minimum of 3 sample sites per age class. To mitigate any effect of sampling effort across age classes and time taken to complete sampling, a maximum of two sample sites in any one age class in one region were sampled during any one survey campaign.
Floristic Surveys
Floristic sampling was undertaken from 28th of January to the 19th March 2015. To identify changes in vegetation composition, floristic surveys were undertaken at each of the 14 pairs of fenced and unfenced plots and five reference sites (33 in total). Each site consisted of two 20 metre x 20 metre (400m2) plots. To control site variability whilst avoiding selection bias, fenced and unfenced pairs were randomly located on either side of a fence within a common domain so that they matched as closely as possible in dominant tree species, structure, aspect, slope, topographic position and land management. Paired sites were located within 150 metres from each other. A handheld geographical positioning system (GPS) was used to record the coordinates of the plot corner.
Twenty, 1 m2 quadrats were randomly located within each plot. All vascular plant species observed rooting within the bounds of the quadrat were identified to species-level, or genus-level for nonflowering monocots. Presence records of each taxon was tallied across the 20 quadrats to give a frequency score (i.e. if a plant was observed in 5 quadrats, irrespective of density, it was awarded 5). Plant species not captured 1 m2 quadrats, but observed growing in the 400 m2 plot were given a nominal score of 0.1 out of 20.
Topsoil Properties
To quantify soil nutrient levels at each site, soil samples were collected during 19th – 28th of April 2015, before peak plant nutrient uptake during the spring growing season (Tisdale et al. 1985). Soil collection for laboratory analysis involved composite soil cores (Brown 1999) A plastic pipe was used to sample soil cores (0-7 cm deep and 2.5 cm in diameter). One subsample was collected within 10 floristic quadrats.
Subsamples from each plot were collected on the same day, homogenised and stored in sealed clip-lock plastic bags for transport to the laboratory for analysis of cation exchange capacity (CEC), extractable K, Na, Mg, Ca, Al, Mn, pH, electrical conductivity (EC), loss on ignition (LOI), nitrate (NO3-N), available phosphorus (Av. P), total kjeldahl nitrogen (TKN), total organic carbon (TOC) and bulk density (DENS). These soil chemical properties were chosen because at high-levels they have be shown to act as barriers against system recovery or influence vegetation response trajectories (Prober et al. 2002b; Standish et al. 2006).
Data Analysis
Species Composition
Presence-absence records from quadrats within each plot were summed to give frequency scores for all vascular plant species. Three site by species frequency matrices were constructed for all 33 sample plots (14 paired sites and 5 reference plots). The first matrix contained all vascular plant species (native and exotic combined), totalling 294 species. The second matrix included only native species, totalling 199 species. The third, excluded native species and totalled 95 exotic species.
Changes in vegetation composition were assessed by unconstrained ordination based on a latent variable model (LVM) (Hui et al. 2015) using the “boral” package (Hui 2016) in R v.3.2.3 (R Core Team 2016). A model based ordination was chosen because it correctly accounts for the underlying properties of the abundance data (most notably its mean-variance relationship) on a species by species level, but more importantly because it allows direct estimation of the position of each site along an unobserved ecological gradient (Hui et al. 2015). These unobserved gradients (latent variables) are analogous to ordinations axes, so for consistency, models were fit with two latent variables. A random effect was assigned to each site to account for site randomisation and variability among sites. Three models were run for each of the three datasets. Each model was assigned a different seed value for Markov Chain Monte Carlo (MCMC) sampling (Hui 2016). To achieve a stable model, the number of iterations was increased until ordinations appeared similar between models with different seed values and MCMC chains converged. Once stability was reached, the model with the lowest deviance information criterion (DIC) was selected for each dataset.
Latent variable coordinates of each sample plot were extracted for all three datasets to elucidate trends of convergence towards reference sites with fencing age and site condition.
For each dataset, Euclidean vector distances were measured between plots based on their latent variable co-ordinates to calculate the following (Figure 2):
vegetation divergence after fencing (Di), represented by the compositional distance between fenced and unfenced plots within each pair;
post treatment state (Fi), represented by the compositional distance between each fenced plot and its closest reference plot;
pre-treatment or initial state (Ui), represented by the compositional distance between each unfenced plot to the same reference plots used to calculate Fi from its fenced pair; and
convergence after fencing (Ci), represented by the difference in distance between Fi and Ui.
To test overall convergence of fenced plots towards reference plots a one sample t-test with a 95% confidence interval was calculated for Ci. Convergence was inferred if the mean value was positive and confidence intervals did not overlap with zero.
To examine causes of vegetation response, Linear Mixed Models (LMMs) (Hadfield 2010) were fitted to predict divergence and convergence responses using the “MCMCglmm” package in R v.3.2.2 (R Core Team 2016). MCMC estimation was utilised for its relative ease in fitting observation-level random effects, which was desirable as previously mentioned. DIC was calculated and residuals checked for each LMM to ascertain the best fit and appropriate transformation. Response variable Di was square root transformed and Ci was log transformed with a constant added to overcome negative values. Di and Ci were modelled separately as a function of fencing age (Ai), initial state (Ui), initial state and fencing age (Ui + Ai) and initial state and fencing age with an interaction (Ui + Ai + Ui.Ai). Another series of LMMs were fitted as above substituting available Phosphorus levels in unfenced plots for distance-based values of Ui (initial state). Available Phosphorus top soil level was selected as a representation of initial state because the majority of sites showed signs of application of super phosphate fertiliser, consistent with information elicited from landholders – a general farming practice undertaken to increase productivity (Donald 1970). In addition, available P was correlated with many other soil constituents (loss of ignition, total kjeldahl nitrogen, total organic carbon and bulk density), suggesting that it was a useful measure of overall soil fertility.
The change in mean frequency (calculated as the percentage increase/decrease) of all taxa in fenced plots and unfenced plots was calculated to identify which species were most affected by livestock grazing exclusion. The highest mean frequency scores of reference plots were calculated to describe the most abundant species in reference plots.
Soil Condition
A Principal Component Analysis (PCA) was applied to topsoil properties using “prcomp” package in R v.3.2.3 (R Core Team 2016). The PCA was based on the correlation matrix (data were scaled and standardised). The number of principal components retained was based on those required to account for 90% of the variance.
Euclidean vector distances were measured between plots based on principal component scores to calculate the following:
soil condition divergence after fencing (Di), represented by the distance between fenced and unfenced plots within each pair;
post treatment state (Fi), represented by the distance between each fenced plot and its closest reference plot;
pre-treatment or initial state (Ui), represented by the distance between each unfenced plot to the same reference plots used to calculate Fi from its fenced pair; and
convergence after fencing (Ci), represented by the difference in distance between Fi and Ui.
To test overall convergence of fenced plots towards reference plots a one sample t-test with a 95% confidence interval was calculated for Ci. Convergence was inferred if the mean value was positive and confidence intervals did not overlap with zero.
To examine changes in soil conditions Linear Mixed Models (LMMs) were fitted to predict divergence and convergence responses using “MCMCglmm” package (Hadfield 2010) in R v.3.2.2 (R Core Team 2016). DIC was calculated and residuals inspected for each LMM to ascertain the best fit and appropriate transformation. The response variable, Di, was square root transformed, and Ci was log transformed with a constant added to overcome negative values. Di and Ci were modelled separately as a function of fencing age (Ai), initial state (Ui), initial state and fencing age (Ui + Ai) and initial state and fencing age with an interaction (Ui + Ai + Ui.Ai).
Figure 2: Hypothetical ordination diagram showing Euclidean distances measured between the coordinates of each plot to calculate divergence (Di), post-treatment state (Fi) and pre-treatment state (Ui). The left example shows a fenced plot that is both diverging from its initial state (paired unfenced plot) and converging with its most similar reference plot. The right example shows a fenced plot diverging from its initial state, but not converging with its most similar reference plot.
Results
Species Composition
The ordination revealed that fenced plots diverged from unfenced plots in species composition, but there was no consistency in vector lengths between fenced and unfenced plots (i.e. degree of compositional divergence was not always greater at older sites). Compositional divergence of fenced from unfenced plots (Di) was therefore unrelated to time-since-livestock-exclusion (Ai) for all three datasets (combined species composition 95% CI = 0.0384 ±0.0124, P= 0.6156; Table 2).
The ordination showed reference plots clustered at high positive values of each latent variable (Figure 3). Fenced plots were generally positioned closer to reference plots than their paired unfenced plot. One sample t-tests confirmed an overall convergence of fenced plots with reference plots for combined species composition (95% CI 0.1690:0.0689, P= 0.0034), but not for native or exotic components individually (Table 2). A slightly higher number of fenced plots converged with reference plots (i.e. had positive values of Ci) in combined species composition compared to native species only and exotic species only (n=12, n=10, n=9, respectively).
The ordination showed no clear organisation of sites by fencing age (i.e. older fenced sites were not positioned closer to reference plots relative to younger fenced plots). Consequently, compositional convergence of fenced plots towards reference states (Ci) was unrelated to time-since-livestock-exclusion for all three datasets (combined species composition (95% CI = -0.0101:0.1956, P = 0.4600; Table 2).
The ordination revealed vegetation change after fencing varied in both magnitude and direction in relation to reference states. Some fenced plots remained close to their unfenced pair; while others moved a greater distance towards reference plots, irrespective of fencing age. Others transition away from both their initial state and reference state. The ordination showed no clear evidence indicating a relationship between degree of compositional divergence and degree of compositional convergence.
Initial state (Ui) did not explain the degree of compositional divergence after fencing for combined species (95% CI -3931:0.1956, P = 0.4660) and native species (95% CI -0.2499:0.0412, P =0.6467). Conversely, compositional divergence of exotic species after fencing was weakly related to initial state (Ui) (95% CI 0.0128:0.5428, P = 0.0489). All other models including both initial state and time-since-livestock-exclusion, with or without interactions terms, failed to explain divergence. Initial state did not explain degree of compositional convergence, either individually or in combination with time-since-livestock-exclusion, with or without an interaction term.
A model including time-since-livestock-exclusion, available Phosphorus in unfenced plots (as an indicator of initial state) and their interaction, proved to be a good predictor of degree of compositional divergence based on exotic species (95% CI = 0.0035:0.1311, P = 0.0467). Otherwise, unfenced Phosphorus level was not a good predictor of convergence or divergence. All other LMMs were not significant.
The frequencies of the top 10 increaser species after fencing increased markedly (6-fold to 36-fold) and 80% were native, but only one was among the most abundant native species in reference plots. An even mix of native and exotic species decreased in abundance after fencing and the top 10 declined in frequency by at least 90%.
Figure 3: Boral unconstrained ordination of latent variables of combined species composition
Soil Condition
Five principal components (PCs) accounted for 90% of the variation in soil condition variables. PC1 accounted for 61% and PC2 accounted for 12% of the variation. Soil properties CEC, Extractable Ca, LOI, TKN, TOC and bulk density were correlated with PC1. Extractable Mn and pH correlated with PC2.
The PCA showed no clear organisation of reference plots relative to fenced/unfenced plots (Figure 4). There was no consistency in vector lengths between fenced and unfenced plots with time of livestock exclusion (i.e. older fenced plots had not moved further away than younger fenced plots). Divergence of fenced plot soil conditions from unfenced plots was unrelated to time-since-livestock-exclusion (95% CI = -0.0279:0.0447, P= 0.680).
There was no consistent shift of fenced plots from their unfenced pair towards reference plots. The ordination also showed notable overlap between these groups. A one sample t-test therefore showed that fencing produced no convergence in soil condition relative to reference sites (95% CI -0.0071:0.2903, P= 0.060). Consistent with this, convergence of fenced plots towards reference states was unrelated to time-since-livestock-exclusion (95% Ci = 0.0169:0.0415, P= 0.436).
Initial state did not explain the degree of divergence of fenced from unfenced plots (95% CI = -0.1275:0.2244, P = 0.224) and degree of convergence of fenced plots towards reference state (95% CI = -0.1035:0.2770, P= 0.353).
Initial state did not explain degree of soil divergence or convergence when time-since-livestock-exclusion was added to the model, with or without interactions terms.
Figure 4: PCA of soil properties
Discussion
While the species composition of fenced plots diverged from that of their paired unfenced plot at a number of sites, divergence varied greatly among sites and was unrelated to time-since-livestock-exclusion. Hypothesis one was therefore not supported for species composition, consistent with earlier findings by Lunt et al. (2007b). The results showed that the compositional divergence of fenced vegetation from unfenced vegetation was due to an increased abundance of native species, and decreased abundance of exotic and other native species.
Although, the unconstrained ordination showed evidence of compositional convergence of fenced plots with reference plots, the degree of convergence was unrelated to time-since-livestock-exclusion. Hypothesis two was therefore not supported for species composition. This finding is supported by other fencing studies of temperate woodlands in Western Australia (Pettit et al. 1995; Prober et al. 2011b) and may be explained partly by poor recovery of soil condition over time (see below).
Initial state weakly predicted the degree of compositional divergence of fenced from unfenced plots in exotic species, partially supporting hypothesis three. This suggests that livestock altered the composition of weeds present in the vegetation. Fencing may release some weeds suppressed by grazing, but this was not clearly supported by the data, which showed that only two exotic species were among the top 10 increasers and four were among the top ten decreasers.
A poor relationship between degree of compositional divergence and degree of compositional convergence suggest that fenced plots that become more different from their initial state over time do not necessarily become more similar to the reference state. The composition is changing, but not in the direction intended by management, suggesting the system does not follow equilibrium theory after disturbance removal and highlighting the importance of site selection. Nonetheless, the observed changes involve increases in abundance of some native species and declines in some exotic species, suggesting the development of a novel community that is more suitable for some elements of the native woodland flora than sites subjected to ongoing livestock grazing.
There was no evidence that soil properties diverged from unfenced controls or converged with reference states, supporting the null model. Hypothesis one was not supported for soil conditions, because there was no evidence that soil properties of fenced sites diverged from those of unfenced sites over time. Similarly, hypothesis two was not supported for soil properties due to the lack of evidence for convergence towards the reference state over time. Few studies have evaluated recovery of topsoil chemistry properties as a result of livestock exclusion. Poor soil recovery is consistent with other temperate woodlands studies in Victoria (Duncan et al. (2007) and Western Australia (Standish et al. (2006); Prober et al. (2011b). Initial state did not explain the degree to which soil properties of fenced plots converged towards those of reference states and the degree to which they diverged from unfenced plots.
In summary, exclusion of livestock by fencing resulted in vegetation change, promoting increased abundance of mostly native species, and decreased abundance of exotic and some other native species, but these changes did not increase compositional resemblance with reference states that had never been grazed by livestock. No decreaser species, and only one increaser species (Themeda triandra) after fencing were found to be most abundant in reference sites. This finding suggests that fencing of degraded woodland vegetation creates a novel plant community dissimilar to species composition of reference states, characterised by land-use legacies from farming practices (such as residual fertilisers), and which, relative to unfenced vegetation that livestock continue to graze, has greater abundance of some native species and reduced abundance of some native and exotic species. Such an interpretation is consistent with Alternative Stable State models (Prober et al. 2002b; Schröder et al. 2005).
These conclusions are based on trends over 17 years of carefully matched space-for-time comparative observations. While there was no evidence that convergence increased with time-since-livestock-exclusion over 17 years, a longer time series of data may reveal evidence of convergence with reference states, highlighting the need for ongoing monitoring of fencing projects. These results highlight the importance of site choice for restoration projects, more so than the length of time after fencing, at least for the first two decades.
The results suggest that history of management practices is a critical consideration when selecting restoration sites. Furthermore, fencing alone appears to be insufficient to restore target woodland communities floristically and environmentally. Other management strategies (e.g. active weed control, planting, scalping and seeding) need to be explored as well as, or instead of fencing. In restoration trials on degraded Victorian grasslands, native species were most successfully established on areas scalped prior to seeding (Gibson-Roy et al. 2010). Scalping removes weed seed banks in the topsoil and lowers elevated soil nutrients (phosphorus and nitrogen) that persist for decades thereby reducing competition from weeds (Gibson-Roy et al. 2010). Other trials in New South Wales box gum woodlands using a combination of spring burns and carbon supplements significantly enhanced the establishment of Themeda triandra from sown seed (Prober et al. 2005). Re-establishing native grasses in central west NSW showed that pre-sowing herbicide was effective in controlling a broad range of weed species that compete with natives for resources during establishment phases (Cole et al. 2004; Semple and Koen 1994).
More than 70 sites were excluded from the study because of unknown or inconsistent management histories. This aimed to reduce noise and limitations of space-for-time substitution. Space-for-time substitution sampling designs assume spatial and temporal variation are equivalent and, while commonly used in ecology, are sensitive to short term environmental fluctuations (Pickett 1989). Increasing replication by accepting a lower standard for matching between fenced and unfenced plots, reduces rather than increases the ability to detect trends, a problem identified by Spooner et al. (2002). An expanded survey effort covering multiple seasons would provide a better representation of true species richness (Schultz et al. 2014). Some plant life-cycles’, including annual species and terrestrial orchids are dormant and undetectable in Autumn. This is particularly important when comparing degraded woodlands invaded by annuals to reference woodlands with cryptic species.
Understanding the attributes of the initial state, such as seed availability, propensity for propagule dispersal from off-site sources and level of soil compaction that drive divergence of species composition and soil conditions after fencing and those that drive compositional convergence of fenced sites towards reference states would benefit from further research.
Acknowledgements
We thank the farmers and land managers that allowed this research to be conducted on their property. This research was partly funded by the Ecological Consultants Association of NSW (ECA Conservation Grant) and the Linnean Society of New South Wales (Joyce W. Vickery Research Fund).
Thank you Frank Hemmings for your help with plant identification. Suzanne Prober, thank you for your initial advice on study design and other advice when needed. Thank you, Geoffrey Kay, for promoting the study among landholders involved in the Environmental Stewardship Project.
Literature Cited
Al-Rowaily, S.L., El-Bana, M.I., Al-Bakre, D.A., Assaeed, A.M., Hegazy, A.K., Ali, M.B., 2015. Effects of open grazing and livestock exclusion on floristic composition and diversity in natural ecosystem of Western Saudi Arabia. Saudi Journal of Biological Sciences 22 (4), 430-437.
Austin, M.P., Williams, O.B., Belbin, L., 1981. Grassland dynamics under sheep grazing in an Australian Mediterranean type climate. Vegetatio 46-47 (1), 201-211.
Beadle, N.C.W., 1981. The vegetation of Australia. Fischer, Stuttgart ; New York.
Briggs, S.V., Taws, N.M., Seddon, J.A., Vanzella, B., 2008. Condition of fenced and unfenced remnant vegetation in inland catchments in south-eastern Australia. Australian Journal of Botany 56 (7), 590-599.
Brown, A.J., 1999. Soil sampling and sample handling for chemical analysis, In Soil analysis: an interpretation manual. (eds K.I. Peverill, L.A. Sparrow, D.J. Reuter), pp. 35-54. CSIRO publishing, Melbourne.
Burrows, G.E., 2004. The importance of seasonality in the timing of flora surveys in the South and Central Western Slopes of New South Wales. Cunninghamia 8 (4), 514-520.
Campbell, M., 1998. Biological and ecological impact of serrated tussock (Nassella trichotoma (Nees) Arech.) on pastures in Australia. Plant Protection Quarterly 13 (2), 80-86.
Cole, I., Lunt, I.D., Koen, T., 2004. Effects of soil disturbance, weed control and mulch treatments on establishment of Themeda triandra (Poaceae) in a degraded white box (Eucalyptus albens) woodland in central western New South Wales. Australian Journal of Botany 52 (5), 629-637.
Connell, J.H., 1975. Some mechanisms producing structure in natural communities: a model and evidence from field experiments In Ecology and evolution of communities. (ed. M.C.a.J.D. eds), pp. 460-490. Belknap Press of Harvard University, Cambridge, MA.
Cramer, V.A., Standish, R.J., Hobbs, R.J., 2007. Western Australian Old Fields: Prospects for the Recovery of Native Vegetation in an Ancient and Highly Modified Landscape, In In Old Fields: Dynamics and Restoration of Abandoned Farmland. (eds V.A. Cramer, R.J. Hobbs), pp. 286-306. Island Press, Washington DC.
Department of Environment Climate Change and Water NSW, 2010. National Recovery Plan for White Box - Yellow Box - Blakely’s Red Gum Grassy Woodland and Derived Native Grassland. Department of Environment, Climate Change and Water NSW, Sydney.
Department of Environment Climate Change and Water NSW, 2010. National Recovery Plan for White Box - Yellow Box - Blakely’s Red Gum Grassy Woodland and Derived Native Grassland. Department of Environment, Climate Change and Water NSW,, Sydney.
Donald, C.M., 1970. Temperate pasture species, In Australian grasslands. (ed. R.M. Moore.). Australian National University Press, Canberra.
Dorrough, J., 2012. How do different levels of grazing and fertilisation affect vegetation composition in temperate Australian grassy ecosystems? Systematic review and meta-analysis. Report to Victorian Department of Sustainability and Environment, Natural Regeneration Australia.
Dorrough, J., Ash, J., McIntyre, S., 2004a. Plant responses to livestock grazing frequency in an Australian temperate grassland. Ecography 27 (6), 798-810.
Dorrough, J., Yen, A., Turner, V., Clark, S.G., Crosthwaite, J., Hirth, J.R., 2004b. Livestock grazing management and biodiversity conservation in Australian temperate grassy landscapes. Australian Journal of Agricultural Research 55 (3), 279-295.
Duncan, D., Moxham, C., Read, C., 2007. Effect of stock removal on woodlands in the Murray Mallee and Wimmera Bioregions of Victoria, Melbourne.
El-Bana, M.I., Nijs, I., Khedr, A.-H.A., 2003. The Importance of Phytogenic Mounds (Nebkhas) for Restoration of Arid Degraded Rangelands in Northern Sinai. Restoration Ecology 11 (3), 317-324.
Gibson-Roy, P., Moore, G., Delpratt, J., Gardner, J., 2010. Expanding horizons for herbaceous ecosystem restoration: the Grassy Groundcover Restoration Project. Ecological Management & Restoration 11 (3), 176-186.
Hadfield, J., 2010. MCMC methods for Multiresponse Generalised Linear Mixed Models: The MCMCglmm R Package.". Journal of Statistical Software 33 (2), 1-22.
Hui, F.K., 2016. Package ‘boral’. Version 1.0. R package version 3.8.4.
Hui, F.K.C., Taskinen, S., Pledger, S., Foster, S.D., Warton, D.I., 2015. Model-based approaches to unconstrained ordination. Methods in Ecology and Evolution 6 (4), 399-411.
Jackson, R.D., Bartolome, J.W., 2002. A state-transition approach to understanding nonequilibrium plant community dynamics in Californian grasslands. Plant Ecology 162 (1), 49-65.
KrÖPfl, A.I., Cecchi, G.A., Villasuso, N.M., Distel, R.A., 2013. Degradation and recvoery processes in semi-arid patchy rangelands of northern Patagonia, Argentina. Land Degradation & Development 24 (4), 393-399.
Lunt, I.D., Eldridge, D.J., Morgan, J.W., Witt, G.B., 2007a. TURNER REVIEW No. 13 A framework to predict the effects of livestock grazing and grazing exclusion on conservation values in natural ecosystems in Australia. Australian Journal of Botany 55 (4), 401-415.
Lunt, I.D., Jansen, A., Binns, D.L., Kenny, S.A., 2007b. Long-term effects of exclusion of grazing stock on degraded herbaceous plant communities in a riparian Eucalyptus camaldulensis forest in south-eastern Australia. Austral Ecology 32 (8), 937-949.
Moore, C.W.E., 1953. The vegetation of the south-eastern Riverina, New South Wales. I. The climax communities. Australian Journal of Botany 1 (3), 485-547.
Pettit, N.E., Froend, R.H., 2001. Long-term changes in the vegetation after the cessation of livestock grazing in Eucalyptus marginata (jarrah) woodland remnants. Austral Ecology 26 (1), 22-31.
Pettit, N.E., Froend, R.H., Ladd, P.G., 1995. Grazing in remnant woodland vegetation: changes in species composition and life form groups. Journal of Vegetation Science 6 (1), 121-130.
Pickett, S.A., 1989. Space-for-Time Substitution as an Alternative to Long-Term Studies, In Long-Term Studies in Ecology. (ed. G. Likens), pp. 110-135. Springer New York.
Price, J.N., Wong, N.K., Morgan, J.W., 2010. Recovery of understorey vegetation after release from a long history of sheep grazing in a herb-rich woodland. Austral Ecology 35 (5), 505-514.
Prober, S., 1996. Conservation of the Grassy White Box Woodlands: Rangewide Floristic Variation and Implications for Reserve Design. Australian Journal of Botany 44 (1), 57-77.
Prober, S., Thiele, K., 1995. Conservation of the Grassy White Box Woodlands: Relative Contributions of Size and Disturbance to Floristic Composition and Diversity of Remnants. Australian Journal of Botany 43 (4), 349-366.
Prober, S.M., Lunt, I.D., Thiele, K.R., 2002a. Determining reference conditions for management and restoration of temperate grassy woodlands: relationships among trees, topsoils and understorey flora in little-grazed remnants. Australian Journal of Botany 50 (6), 687-697.
Prober, S.M., Standish, R.J., Wiehl, G., 2011a. After the fence: Vegetation and topsoil condition in grazed, fenced and benchmark eucalypt woodlands of fragmented agricultural landscapes. Australian Journal of Botany 59 (4), 369-381.
Prober, S.M., Standish, R.J., Wiehl, G., 2011b. After the fence: vegetation and topsoil condition in grazed, fenced and benchmark eucalypt woodlands of fragmented agricultural landscapes. Australian Journal of Botany 59 369-381.
Prober, S.M., Thiele, K.R., 1993. The ecology and genetics of remnant grassy white box woodlands in relation to their conservation. Victorian Naturalist 110 (1), 30-36.
Prober, S.M., Thiele, K.R., Lunt, I.D., 2002b. Identifying ecological barriers to restoration in temperate grassy woodlands: soil changes associated with different degradation states. Australian Journal of Botany 50 (6), 699-712.
Prober, S.M., Thiele, K.R., Lunt, I.D., Koen, T.B., 2005. Restoring ecological function in temperate grassy woodlands: manipulating soil nutrients, exotic annuals and native perennial grasses through carbon supplements and spring burns. Journal of Applied Ecology 42 (6), 1073-1085.
R Core Team, 2016. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.
Sankaran, M., Hanan, N.P., Scholes, R.J., Ratnam, J., Augustine, D.J., Cade, B.S., Gignoux, J., Higgins, S.I., Le Roux, X., Ludwig, F., Ardo, J., Banyikwa, F., Bronn, A., Bucini, G., Caylor, K.K., Coughenour, M.B., Diouf, A., Ekaya, W., Feral, C.J., February, E.C., Frost, P.G.H., Hiernaux, P., Hrabar, H., Metzger, K.L., Prins, H.H.T., Ringrose, S., Sea, W., Tews, J., Worden, J., Zambatis, N., 2005. Determinants of woody cover in African savannas. Nature 438 (7069), 846-849.
Scheffer, M., Carpenter, S., Foley, J.A., Folke, C., Walker, B., 2001. Catastrophic shifts in ecosystems. Nature 413 (6856), 591-596.
Scholes, R.J., Walker, B.H., 2004. An African Savanna: Synthesis of the Nylsvley Study. Cambridge University Press.
Schröder, A., Persson, L., De Roos, A.M., 2005. Direct experimental evidence for alternative stable states: a review. Oikos 110 (1), 3-19.
Schultz, N.L., Morgan, J.W., Lunt, I.D., 2011. Effects of grazing exclusion on plant species richness and phytomass accumulation vary across a regional productivity gradient. Journal of Vegetation Science 22 (1), 130-142.
Schultz, N.L., Reid, N., Lodge, G., Hunter, J.T., 2014. Seasonal and interannual variation in vegetation composition: Implications for survey design and data interpretation. Austral Ecology 39 (7), 755-766.
Semple, W.S., Koen, T.B., 1994. Effect of pasture type on regeneration of eucalypts in the woodland zone of south-eastern Australia. Cunninghamiana 8 (1).
Specht, R.L., 1981. Conservation of vegetation types, In Australian Vegetation. (ed. H. Groves), pp. 394-410. Cambridge University Press, Cambridge.
Spooner, P., Lunt, I., Robinson, W., 2002. Is fencing enough? The short-term effects of stock exclusion in remnant grassy woodlands in southern NSW. Ecological Management & Restoration 3 (2), 117-126.
Spooner, P.G., Allcock, K.G., 2006. Using a State-and-Transition Approach to Manage Endangered Eucalyptus albens (White Box) Woodlands. Environmental Management 38 (5), 771-783.
Spooner, P.G., Briggs, S.V., 2008. Woodlands on farms in southern New South Wales: A longer-term assessment of vegetation changes after fencing. Ecological Management & Restoration 9 (1), 33-41.
Standish, R., Cramer, M., Yates, C., 2009. A Revised State-and-Transition Model for the Restoration of Woodlands in Western Australia, In New models for ecosystem dynamics and restoration. (ed. K.N. Suding), pp. 169-187. Island Press, Washington.
Standish, R.J., Cramer, V.A., Hobbs, R.J., Kobryn, H.T., 2006. Legacy of land-use evident in soils of Western Australia's wheatbelt. Plant & Soil 280 (1-2), 189-207.
Standish, R.J., Cramer, V.A., Wild, S.L., Hobbs, R.J., 2007. Seed dispersal and recruitment limitation are barriers to native recolonization of old-fields in western Australia. Journal of Applied Ecology 44 (2), 435-445.
Standish, R.J., Hobbs, R.J., Mayfield, M.M., Bestelmeyer, B.T., Suding, K.N., Battaglia, L.L., Eviner, V., Hawkes, C.V., Temperton, V.M., Cramer, V.A., Harris, J.A., Funk, J.L., Thomas, P.A., 2014. Resilience in ecology: Abstraction, distraction, or where the action is? Biological Conservation 177 (0), 43-51.
Stuwe, J., Parsons, R.F., 1977. Themeda australis grasslands on the Basalt Plains, Victoria: floristics and management effects. Australian Journal of Ecology 2 (4), 467-476.
Su, H., Liu, W., Xu, H., Wang, Z., Zhang, H., Hu, H., Li, Y., 2015. Long-term livestock exclusion facilitates native woody plant encroachment in a sandy semiarid rangeland. Ecology and Evolution 5 (12), 2445-2456.
Suding, K.N., Gross, K.L., Houseman, G.R., 2004. Alternative states and positive feedbacks in restoration ecology. Trends in Ecology & Evolution 19 (1), 46-53.
Tisdale, S.L., Nelson, W.L., Beaton, J.D., 1985. Soil and fertilizer potassium, In Soil fertility and fertilizers. (eds S.I. Tisdale, W.I. Nelson, J.D. Beaton), pp. 249-291. MacMillan Publishing, Co, New York.
Tremont, R., 1994. Life-History Attributes of Plants in Grazed and Ungrazed Grasslands on the Northern Tablelands of New South Wales. Australian Journal of Botany 42 (5), 511-530.
Yates, C.J., Hobbs, R.J., 1997a. Temperate Eucalypt Woodlands: a Review of Their Status, Processes Threatening Their Persistence and Techniques for Restoration. Australian Journal of Botany 45 (6), 949-973.
Yates, C.J., Hobbs, R.J., 1997b. Woodland restoration in the Western Australian wheatbelt: a conceptual framework using a state and transition model. Restoration Ecology 5 (1), 28-35.
Yates, C.J., Hobbs, R.J., 1999. Temperate eucalypt woodlands in Australia- an overview, In Temperate Eucalypt Woodlands in Australia. (ed. Hobbs RJ and Yates CJ), pp. 1-5. Surrey Beattie and Sons, Chipping Norton.
Yates, C.J., Taplin, R., Hobbs, R.J., Bell, R.W., 1995. Factors limiting the recruitment of Eucalyptus salmonophloia in remnant woodlands: II. Post-dispersal seed predation and soil seed reserves. Australian Journal of Botany 43 (2), 145-155.
Yeo, J.J., 2005. Effects of grazing exclusion on rangeland vegetation and soils, East Central Idaho. Western North American Naturalist91-102.