Disentangling chronic regeneration failure in endangered woodland ecosystems

. Ecological restoration of degraded ecosystems requires the facilitation of natural regeneration by plants, often augmented by large-scale active revegetation. The success of such projects is highly variable. Risk factors may be readily identi ﬁ able in a general sense, but it is rarely clear how they play out individually, or in combination. We addressed this problem with a ﬁ eld experiment on the survival of, and browsing damage to, 1275 hand-planted buloke ( Allocasuarina luehmannii ) seedlings in a nationally endangered, semi-arid woodland community. Buloke seedlings were planted in 17 sites representing four landscape contexts and with three levels of protection from kangaroo and lagomorph browsing. We censused seedlings and measured herbivore activity four times during the ﬁ rst 400 d post-planting and ﬁ tted models of mortality and browse hazard to these data using survival analysis. Increasing lagomorph activity was associated with higher mortality risk, while kangaroo activity was not. Seedling survival was lowest for each treatment within extant buloke woodland, and the highest survival rates for guarded seedlings were in locations favored by lagomorphs. Damage from browsing was nearly ubiquitous after one year for surviving unguarded seedlings, despite moderate browser activity. On average, unguarded seedlings showed a decline in height, whereas fully guarded seedlings grew 2.3 cm across the survey period. This study demonstrates buloke seedlings should be protected from browsers, even with browsers maintained at moderate to low density, and the location that maximizes survival, and possibly growth rates, is adjacent to dunes. Further work will test this heuristic in an analysis of cost-effective revegetation strategies for this endangered community.


INTRODUCTION
Ecosystems heavily modified or displaced by agriculture may be at risk of ecological collapse due to the loss of biotic components or ecological functions (Keith et al. 2013, Bland et al. 2018. Ecological restoration of such ecosystems requires facilitation of natural regeneration processes, which is often augmented by large-scale, active revegetation programs (Vesk and MacNally 2006, Molin et al. 2018, Rohr et al. 2018. Ecological interventions are expensive and have high uncertainty, and conservation budgets are typically small (Curtis and Lockwood 2000, McLeod 2004, Cooke et al. 2010, so an understanding of the processes underpinning success and failure is critical. Risk factors that may impede seedling survival are often readily identifiable in a general sense, such as water stress, interspecific competition, and herbivory (Close et al. 2005). However, it is rarely clear how these risk factors might play out individually, or in combination across spatially heterogeneous landscapes. Such understandings are required to plan and manage cost-effective restoration of ecosystems (Dorrough et al. 2008, McBride et al. 2010. For example, hazards of water stress may be independent of, or weakly correlated with, herbivore pressure and the two hazards may differ widely in their consequences for seedlings. Further, mere survival is not enough when exposed to strong grazing or browsing pressure. To attain the population-sustaining characteristics of mature individuals, seedlings and saplings must grow well enough to escape their prospective grazers and browsers (Vesk and Dorrough 2006).
Here, we examine the complex interaction of risk factors that potentially contribute to regeneration failure for an endangered woodland ecosystem in southeast Australia. The role of herbivores in limiting plant regeneration is a primary management concern in the ecosystem, an entity circumscribed as Buloke Woodlands of the Riverina and Murray-Darling Depression Bioregions (Department of the Environment and Energy 2008). This semi-arid woodland ecosystem was extensively cut from the 1850s to promote pasture growth for cattle and sheep, and most was later cleared for cereal cropping (Cheal et al. 2011). For remnant vegetation, this regime resulted in the extirpation of indigenous fauna and flora, and the introduction of alien species including annual weeds and herbivores, most notably the European rabbit (Oryctolagus cuniculus). Concerns about the influence of browsing and grazing herbivores on the species diversity and regeneration of dominant woody species in this area date back at least 50 yr (Cochrane and McDonald 1966). Since then, the largest remnants were incorporated into protected areas (Cheal et al. 2011) and livestock grazing concessions were phased out in the 1970-1990s to facilitate natural regeneration (Cheal 1986, Land Conservation Council 1989, Durham 2001). However, no signal of adequate recruitment or regeneration to replace the aging stock of remaining mature trees has emerged.
Since the removal of livestock from protected areas, there has been increasing emphasis on the threat that introduced rabbits and the native western gray kangaroo (Macropus fuliginosus) pose to restoration. These two herbivores can impede seedling regeneration across a broad swathe of Australian ecosystems (Cheal 1986, Coulson et al. 1989, Bird et al. 2012, Taylor and Pegler 2016, Dillon et al. 2018. Both species preferentially feed on grasses and herbs but they will browse shrubs and seedlings when preferred options become scarce (Coulson and Norbury 1988, Bird et al. 2012, Mutze et al. 2016b.
Previous studies from temperate Australia have demonstrated that rabbits are capable of significant browsing damage and mortality even at low densities <1 ha À1 (Lange and Graham 1983, Bird et al. 2012, Mutze et al. 2014, Forsyth et al. 2015. The degrading impact of kangaroos at high densities is clear (Cheal 1986, Coulson et al. 1989, Sluiter et al. 1997, and their population is subject to annual monitoring and control (Morris et al. 2019), but the impacts kangaroos may have on woody perennial species at lowmoderate densities are unclear. Kangaroos are typically presented as a subordinate browsing threat in studies of both rabbits and kangaroos (Bird et al. 2012, Mutze et al. 2016a with several studies explicitly separating these effects (Cooke 1987, Allcock and Hik 2004, Denham and Auld 2004, Bird et al. 2012. We conducted a survival experiment on handplanted buloke (Allocasuarina luehmannii R. T. Baker (L. A. S. Johnson)) seedlings using exclosures designed to distinguish the browsing impacts of kangaroos and rabbits. We planted seedlings in distinct spatial contexts representing variation in habitat favorability for kangaroos or rabbits. We expected that seedlings in habitats favored by rabbits would suffer high mortality, but the likely impact of kangaroos was uncertain. Buloke was selected as the target species because it presents the most persistent regeneration failure among the structurally dominant species of the endangered buloke woodland ecological community (Gowans et al. 2010). We examined the variation in seedling browsing and mortality risk with exclosure treatment, herbivore abundance, habitat features, and site over time. These data can immediately inform future planting strategies and can also feed into cost-effectiveness analyses with varying levels of protection and herbivore control.

Study system and sites
Our experiment was located in the Pine Plains management area of Wyperfeld National Park in ❖ www.esajournals.org northwest Victoria, Australia (Appendix S1: Fig. S1). The region typically experiences hot summers and mild to cool winters, with highly variable rainfall throughout the year, which typically occurs as sporadic, localized, and often intense rain events. The long-term mean annual rainfall of 332 mm (AE109 standard deviation [SD], Bureau of Meteorology Walpeup Research Station No. 76064) was exceeded in 2016 and 2017 when this study took place, with 394 mm and 355 mm recorded, respectively (Appendix S1: Fig. S2).
Pine Plains contains the largest (~700 ha), albeit highly degraded, remnant of the endangered buloke woodland (Cheal et al. 2011). These woodlands are dominated by buloke and slender cypress-pine (Callitris gracilis). The understorey is highly simplified, with an occasional shrub layer and a ground layer dominated by native and introduced herbs and grasses (Gowans and Gibson 2005).

Reproductive biology and regeneration niche of buloke
Buloke is a long-lived tree in the Casuarinaceae family. Although listed as vulnerable in Victoria, it occurs over a wide latitudinal range of Australia (~16°-37°S) inland of the Great Dividing Range (Atlas of Living Australia 2019). Buloke is dioecious or sub-dioecious (Conomikes et al. 2011). It is wind pollinated; males may produce copious pollen and females prodigious quantities of cones (Raymond 1990). It can reproduce sexually, and suckers readily following root zone disturbance (Murdoch 2005). As a nitrogen fixer, buloke seedlings are presumed to be highly palatable (Mutze et al. 2016b).

Herbivore species
The European rabbit has become a major pest over much of Australia (Kearney et al. 2019). They consume grasses and forbs but will also feed on seedlings, saplings, shrubs, bark, and tubers (Bird et al. 2012, Mutze et al. 2016a and can browse foliage up to 60 cm in height. Rabbits have been monitored and controlled (fumigation and warren ripping) at Wyperfeld NP since the 1970s. Since numbers crashed by an order of magnitude following the introduction of a biological control agent (myxoma virus), they have largely been maintained at or below target levels of <1 rabbit/transect km (Sandell 2002, Parks Victoria 2016. European hare (Lepus europaeus) occur at lower densities and have less irruptive population dynamics and potentially a lower capacity for ecological impact than rabbits. We included them here because we could not reliably distinguish the fecal pellets of each species, which we used as our measure of herbivore activity.
The western gray kangaroo is a large, social macropodid marsupial (17-72 kg; Coulson 2008) with a preference for heterogeneous habitats that provide both food and shelter (Arnold et al. 1989, Coulson 1993, Garnick et al. 2016. The western gray kangaroo is generally considered a grazer, but will also browse on shrubs and tree species, particularly if grass availability is low Norbury 1988, Morgan and. Kangaroo population control has been undertaken by ground-shooting at Wyperfeld NP since the 1980s (Morris et al. 2019).
While feral goats (Capra hircus) were present and were recorded on camera traps, so few goat fecal pellets were recorded that goats were excluded from our analyses. Similarly, red kangaroos (Osphranter rufus) were present in the park but none was observed within 4 km of the study sites, so this species was not considered further.

Site establishment
A total of 17 locations were randomly selected within areas identified as the former distribution of the buloke woodland community (Gowans and Gibson 2005). We accepted points as suitable buloke habitat if a live or dead buloke tree was located within 200 m. We discarded points if they were located where buloke trees were deemed unlikely to have been present, such as on a dune crest or former lakebed (Cheal et al. 2011), or where there was evidence of a potentially confounding factor, for example, recent fire.
Six sites were located within extant open woodland structure still dominated by mature buloke trees, hereafter "buloke woodland." Five sites were located in open grassy areas 198-331 m from continuous canopy cover, and six were located in open grassy areas adjacent to cover (3-33 m). Three of these latter sites sat adjacent to low Eucalyptus mallee woodland and ❖ www.esajournals.org three adjacent to dune ridges dominated by Acacia shrubs. These different contexts were selected to capture variation in herbivore activity, as informed by park rangers: Kangaroos often use mallee vegetation for shelter and shade and feed in adjacent grasslands, and rabbits favor dune habitats for the formation of warrens, while acacias provide good cover from both aerial and terrestrial predators. Although open grasslands may provide good forage, the lack of nearby shelter suggests a lower herbivory risk for planted seedlings.
We established the 17 (50 9 50 m) sites over 6 weeks in spring (22 October-1 December) 2016. In each site, we randomly selected 75, 2-m 2 squares from a 25 9 25 grid (Appendix S1). We randomly assigned one of three herbivore access treatments (n = 25; Appendix S1: Fig. S3) to each of the selected cells. Treatments were Open, providing access to all herbivores; Partial, excluding large herbivores (goats and kangaroos) but allowing access to lagomorphs; and Total, excluding all herbivores. Seedlings were planted using a Hamilton forestry tube tree planter (Noble 1993) into loosened soil, to a depth of 2 cm below the surface, and then, 1 L of water was applied. Pre-treatment of the buloke seedlings is described in Appendix S1.

Site variation
Four north-south orientated transects were set up at 10-m intervals across each site for quantifying site variation. To provide an index of herbivore activity at each site, we established 16 fecal pellet accumulation plots (Putman 1984) of 15.75 m 2 (r = 2.24 m) at 10-m intervals along each transect (accounting for 10% of site area; Appendix S1). Along the same transects, we estimated cover abundance of vegetation strata and ground cover attributes using point intercept method; obtained distance to tree cover from GIS; and measured soil textural characteristics (Appendix S1). We then used principal component analysis on scaled data (Appendix S1: Fig. S4) to evaluate the influential component axes as alternative predictors of mortality and browsing. We tested these component axes against site context (as a categorical variable) and the individual site characteristic variables, and the results are presented in Appendix S1.

Seedling survey
Seedlings were surveyed on four occasions: December 2016 (10-47 d post-planting), February 2017, April 2017, and December 2017 (364-406 d post-planting). Whenever a dead seedling was encountered (no green plant tissue visible; Bird et al. 2012), we assigned a cause of death (Appendix S1: Table S1) and recorded the final height. During the last survey, we measured the height and stem diameter for all seedlings, recorded the status of all seedlings as live or dead, and categorized the level of damage consistent with browsing (Appendix S1: Table S1). We adopted a conservative approach to herbivore damage, modeling hazard based only on the moderate to extreme cases with damage to the apical (main) stem.

Statistical model
We modeled two aspects of the fate of buloke seedlings using survival analysis (Cox and Oakes 1984, Muenchow 1986, Mills 2011, Austin 2017: the hazard of being browsed by vertebrate herbivores, and the hazard of seedling mortality. The hazard of seedling mortality includes any other factors such as physiological stress from water deficit, pathogen attack, and physical damage during planting or trampling by wildlife postestablishment. The response variable in survival analysis is the instantaneous rate of occurrence of the event (baseline hazard), in our case seedling mortality or seedling browsing. The baseline hazard function was derived from the binary response variable (dead = 1/alive = 0; browsed = 1/not browsed = 0) and time (number of elapsed days since planting), which was supplied to the model as a log-transformed offset to represent degree of exposure to browsing.
Mortality was modeled as a pseudo-Poisson process using a complementary log link (cloglog). The predicted response was a linear function of site context, treatment type, and browser activity (site mean deposition rate of lagomorph and kangaroo pellets) plus the interaction of context with treatment type. Continuous covariates were centered and rescaled by two SDs following Gelman (2008). Site was coded as a random effect.
The model described above imposes a constant baseline hazard. For example, in the case of mortality, it assumes that a seedling has a constant ❖ www.esajournals.org instantaneous risk of mortality throughout the experiment. In principle, and with trends in the data, that assumption seemed too simple. We included a quadratic polynomial term on the log of elapsed days, to allow for the cumulative mortality probability to increase more slowly as seedlings became established. No additional smooth term was required to fit the model of browse hazard.
We fit the models in a Bayesian framework using the package greta (Golding 2018) for R (R Core Team 2018). The model parameter estimates were derived from posterior probability distributions generated by Markov Chain Monte-Carlo (MCMC) simulations. The MCMC algorithm allows a computer to efficiently explore the value space for all model parameters simultaneously (see van Ravenzwaaij et al. 2018). We ran four MCMC chains, sampling 10,000 iterations after discarding 2000 samples as burn-in. Initial parameter values except for the intercept were drawn from a random normal distribution centered on 0 (AE0.4 SD). For the intercept, we drew initial values randomly from a normal distribution with mean of À5, informed by preliminary modeling. The model was evaluated on Gelman-Rubin statistics, complete mixing of posterior chains, and by inspecting prediction plots.
The list of variables included in exploratory models is presented in Appendix S1: Table S2, data, and code are available via https://github.c om/dhduncan/buloke_survival.

RESULTS
The herbivore activity index confirmed our assumptions regarding the different contexts (Fig. 1); wattle dunes and mallee vegetation were most favored by lagomorphs and kangaroos (respectively), and open grassland sites are least favored by both species.
We converted the herbivore activity indices to densities following Mutze et al. (2014) for lagomorphs, and Coulson and Raines (1985) for kangaroos. These conversions suggest that 0-2 lagomorphs and <0.1 kangaroos were present per ha, respectively (Appendix S2).

Seedling mortality
Overall, 60% of the 1275 planted seedlings survived the experimental period of just over 400 d. Survival averaged 30% in the open treatment cohort, 75% in the partial exclusion treatment, and 77% in the total exclusion cohort. Of seedlings that survived the year in the open treatment, only 2.5% had escaped browsing damage, compared with 45% of the partial exclusion cohort and virtually all the total exclusion cohort (Fig. 2).
Across all treatments, up to 30% of seedlings died without browsing. Most of those occurred in the first few months following planting. The most common cause of mortality for seedlings was browsing only in the open treatment (73%), while in the partial treatment only 6.5% of dead seedlings had been browsed (Fig. 2). So few seedlings in the total exclusion treatment appeared browsed that we excluded them from the statistical model of browse hazard.
Our observation period of 406 d coincided with favorable environmental conditions compared with long-term averages. Twice the average monthly rainfall fell in the second month post-planting, and monthly rainfall was around or above the long-term average for 10 of the 12 months (Appendix S1: Fig. S2).
The modeled baseline daily mortality hazard for a seedling in a buloke woodland context, with no protection from browsing (open treatment), was 0.005 (À5.33) on the complementary log-log scale, which translates to an expected cumulative probability of survival over a 365-d period of around 0.2 under median herbivore activity (Fig. 3).
Seedlings planted in the open treatment in buloke woodland (the base case) proved to have the highest mortality risk (Fig. 3). Guards excluding all herbivores (Total) or those that would allow access by small herbivores (Partial) resulted in a 50% reduction of mortality after one year. Guarded treatments located adjacent to wattle dunes were particularly effective, where mean mortality after one year was predicted to be only around 20%, compared to around 65% without guards (Fig. 3).
Kangaroo activity did not predict mortality, with a mean effect centered near zero with high uncertainty (Fig. 4). By contrast, higher lagomorph activity tended to increase the mortality risk, though the 95% credible interval still included 0 (Fig. 4). For a seedling planted in Buloke woodland (base case), if lagomorph pellets were set to the 90% quartile (0.05 pellets per plot per day, estimated to be equivalent to 1 rabbit/ha; Mutze et al. 2014), the expected probability of survival decreased to 0.07.

Browsing risk
At the end of one year, most unguarded planted seedlings were damaged consistent with browsing on the apical stem, from 70% of seedlings in open grassy contexts to 93% of those adjacent to mallee woodlands (Fig. 2), where the hazard in open treatments was significantly higher (Fig. 5). Our model predicts that 100% of seedlings would be damaged inside the first 6 months adjacent to mallee woodlands (Fig. 6), though for other contexts it seems a matter of when, rather than if, unguarded seedlings will suffer browsing damage.
Seedling guards greatly reduced browse hazard. Negligible browse damage occurred in total exclusion guards, and, overall, partial exclusion guards reduced the browse damage hazard to around 30-60% (Figs. 5, 6). However where lagomorph activity was higher (Fig. 5), the browse hazard doubled as lagomorph activity moved from 5 th to 95th quantiles of the observed range (Fig. 7).
Our model identified a negative interaction between kangaroo activity and the browsing hazard for that same cohort of partially guarded seedlings (Fig. 5). Although lagomorph activity in open grassy sites was negligible according to our activity index, around one in four seedlings in partial exclusion treatments was damaged by browsing (Fig. 2).

Seedling growth
Growth, measured by change in height (cm) and stem diameter (mm), was consistent with the pattern of mortality and browsing damage. Seedling diameters more than doubled on average (125% increase) across all treatments over the course of the experiment, but seedlings in the open treatment increased less than those in partial and total treatments (Table 1). Mean height change was negative overall (À20%) and only positive (+6.4%, or 2.3 cm) in the total exclusion cohort. Seedlings in the open treatment lost on average two-thirds of their height by the end of the experiment, or at death (whichever came first).

DISCUSSION
Factors that may impede the success of efforts to restore degraded ecosystems are usually identifiable in the generic sense, but the influence of these factors individually and in combination varies between and within heterogeneous sites. This study demonstrates the complex interplay of biotic and abiotic factors that contributed to seedling browsing and survival, and ultimately the success of revegetation efforts to improve ecosystem function.
In a period of above-average rainfall and moderate herbivore activity, 70% of seedlings planted without guards died, the average net change in height was a reduction of more than 50% of starting size, and only 2.5% of remaining seedlings were alive and without browsing damage to the apical stem. These figures help explain why regeneration has been so difficult to achieve in this highly modified ecological community. Average survival varied spatially from near 0% after a year for unguarded seedlings in buloke woodland context to better than 80% for fully protected seedlings adjacent to dunes.
Early mortality was similar among treatments, and given most of those individuals were largely intact at the first census, failure of these seedlings was likely due to moisture stress as has been reported elsewhere (Denham andAuld 2004, Bird et al. 2012). While the annual rainfall was above average, rain events are sporadic and the availability of soil moisture appears to have been insufficient to sustain these individuals. Failure to establish was common to all contexts but was particularly severe in buloke stands, which could indicate greater competition for soil moisture with established adult trees in that environment. Buloke seedlings have been shown to suffer when planted in close proximity to adult trees (Morgan et al. 2013), and for that reason, we did not place seedlings within 13 m of a live adult. Nonetheless, differences in resource availability due to root zone competition might make it more difficult for seedlings to grow roots and survive moisture deficit in woodland contexts.
Survival patterns diverged as expected, and as abundantly demonstrated in the literature (Dillon et al. 2018), survival was far better when seedlings were guarded. However, we found no difference in survival between the types of protective guard-seedlings protected from both lagomorphs and kangaroos, or only kangaroos. While it could be inferred that kangaroos are a more damaging browser of buloke seedlings than rabbits, such a finding would strongly contradict the considerable body of work demonstrating that in semi-arid ecosystems rabbits are the more destructive browsers (Bird et al. 2012, Mutze et al. 2016a, although see Kirkpatrick and Bridle 2013, for a different ecosystem). Converting pellet accumulation data to a density estimate suggested that kangaroos were below target densities, which is in accordance with data collected by the park management agency (see Morris et al. 2019). Like Bird et al. (2012, our model suggested that kangaroo activity across the range observed here did not explain mortality hazard. However, kangaroos may have impacts at higher densities, as demonstrated elsewhere (Cheal 1986, Sluiter et al. 1997. Mortality tended to be higher in sites with greater lagomorph activity, which accords with numerous past studies. Partial treatments were also less effective in reducing browse hazard where lagomorph activity was relatively high. These observations suggest lagomorphs were the more damaging of the two browser groups. We suspect that lagomorphs may not have been strongly motivated to access the seedlings in partial guards, given the availability of alternative forage outside guards. Indeed, Cooke (1987) encountered a similar pattern and with additional trials was able to show that rabbits were not accessing all tree guards designed to exclude only kangaroos.
Our study location Pine Plains is considered a relatively homogenous semi-arid woodland landscape for management purposes, but our results show how variation in habitat type and planting site context influence browsing and mortality hazard in important ways, mostly as a function of soil characteristics and availability of shelter from adult trees and shrubs. We exploited these differences to observe variation in survival under different levels of exposure to herbivores. It follows that the most cost-effective restoration options may similarly exploit that heterogeneity to guard less or differently where those hazards are reduced. We also recognize the need for caution in interpreting these results. While we gained insight about herbivore impacts from exploiting habitat variation, experimental manipulation of herbivore density would provide stronger inference.
Survival of a buloke seedling through a year in a natural setting is a remarkable event, as evidenced by the chronic lack of regeneration in this landscape. Even so, survival for one year does not equate to restoration success. Seedlings need to attain an escape height or bulk such that they are no longer vulnerable to browsing damage under all but the most extreme scarcity of forage. Previous work with buloke suggested that severely browsed seedlings are extremely slow to recover, even if protected from further browsing (Murdoch 2007). Seedlings remain susceptible to browsing damage by herbivores until they are at least seven years of age and are not considered safe from browsers until over nine years of age (>60 mm basal stem diameter; Murdoch 2007) due to the low presentation of foliage. The poor height growth increment in our study reflected a high frequency of damage to apical stems. Damage to apical stems consistent with vertebrate herbivore browsing (due to the bite pattern or ancillary evidence of lateral browsing on branchlets) was evident in around two-thirds of all seedlings in open treatments, and up to half of those in the partial treatments. Even our total protected seedlings only showed average net increase of around 2.3 cm height and 0.27 cm basal diameter over one year, so the recovery period from browsing damage could be substantial. What constitutes escape size is also a function of available resources, as in times of extreme forage scarcity, vertebrate herbivores may damage or kill mature trees and shrubs: Rabbits can ring bark trees and shrubs (Tiver and Andrew 1997), and kangaroos can consume woody plant material including root tissue (Morgan and Pegler 2010).
In practical terms, hand-planted seedlings and saplings will require protection from browsers for a decade and guards will need to be maintained and potentially replaced as plants become larger. Guarding groups of plants rather than individuals may be a viable option. These possibilities could all be examined in an analysis of cost-effective restoration options using growth and survival parameters from our study.
Soil moisture relations may play an important role in the observed growth pattern, and further analysis emphasizing growth response data might benefit from substitution of our categorical site context variable for sand (or clay) percentage. In model testing, sand (or clay) percentage proved a viable alternative continuous predictor in place of site context, with survival higher in sandier sites. A spatial model of cost-effective planting for optimal canopy replacement may also benefit from the use of soil predictors in place of site context.

CONCLUSIONS
Our case study demonstrates how empirical studies can untangle the influence of a range of factors that limit ecosystem recovery. To achieve cost-effective restoration of degraded woodland ecosystems under an adaptive management framework, management agencies need quantitative links between herbivore densities, their impacts and interactions, and the effectiveness of management interventions in the system. Our study cannot satisfy all those requirements, but it does yield an immediately actionable heuristic model of where and how to revegetate endangered buloke woodlands. Seedlings must be protected from vertebrate browsing in all contexts, particularly given that they may need a decade or more of growth to reach escape height. Although habitats near dunes are favored by lagomorphs, and particularly rabbits, if seedlings are given robust protection, the best results may be achieved planting near dunes. These insights could be incorporated into a spatially explicit restoration strategy via cost-effectiveness analyses including plant growth and survival data under varying rainfall and soil moisture conditions together with fluctuations in herbivore density and control scenarios.