Islands in the sky – could complex topography help us rewild beyond the fence?

Study site

Mt. Talaterang (−35.2774S, 150.2709E) is a 317-ha mesa (Fig. 1). The mesa has relatively few on-foot access points; with one on its most northerly aspect, at Pallin Pass, and another on the western edge above Dummal creek, both requiring a traverse over complex terrain (boulders, slots, and hauls) to reach the plateau. We also studied the area at Little Forest plateau (−35.2387S, 150.3228E) where a camera trap site was established approximately 3 km east of the closest camera trap site on Mt. Talaterang (Fig. 1). Little Forest plateau is approximately 2954 ha. This site is geographically lower-lying and has more extensive connectivity to the surrounding landscape than Mt. Talaterang, with well-maintained public vehicle access roads (Mt. Bushwalker Road/Tianjara Fire trail, Pointers Gap Road) as well as multiple foot-passes (Egans Gap, Martins Gap, Ingold Gap, Gadara Point, among other unnamed on-foot access points).

Fig. 1.

Study area. Survey sites at Mt. Talaterang (red) and Little Forest Plateau (brown) in Morton National Park, NSW, Australia.

Camera trap deployment

In April 2022, we deployed 16 Swift Enduro camera traps on Mt. Talaterang, spaced at a minimum of 250 m apart, with camera traps placed along remnant walking tracks, animal trails, or in open landscape amongst large flat rocky outcrops that characterise much of the south-western portion of this plateau (Fig. 2). Mean camera elevation for cameras deployed on the plateau was 687 m (range 619.4–761.5 m, Supplementary Table S1). On Little Forest Plateau, we deployed 20 Swift Enduro camera traps spaced a minimum of 350 m apart along a transect that used existing roads and walking trails (Fig. 2). Camera trap deployment was limited to existing roads and foot trails as the site lies within the boundaries of the former Tianjara Artillery Training Area and off-track activities carry an Unexploded Ordinance risk. Mean camera elevation on Little Forest plateau was 529 m (range 491.8–584 m; Table S1).

Fig. 2.

Camera trap sites at (a) Little Forest plateau (lower-lying site) and (b) Mt. Talaterang plateau (mesa). (c) Experimental set-up of a motioned-triggered camera facing towards a tea strainer (circled in red) that contained a universal attractant mix (mixture of oats, peanut butter, and tuna oil) fixed to the ground using a tent peg. Tea strainers were placed between 1.5 m and 5 m in front of each camera.

At both sites, we attached cameras to vegetation at heights of approximately 50 cm and aimed parallel to the ground. At all sites, we fixed a tea strainer containing a universal attractant mix (mixture of oats, peanut butter and tuna oil) to the ground using a tent-peg at between 1.5 m and 5 m in front of each camera. Attractants could be smelt, but not accessed by animals. Cameras were active for 24 h per day and set to take five photographs with no rest period using infra-red flash. Cameras were deployed at Mt. Talaterang for 79–118 days (giving a total survey effort of 1724 trap days), and at Little Forest Plateau for 119 days (giving a total survey effort of 2380 trap days).

We uploaded images from each camera to the Wildlife Insights image management online platform (Wildlife Insights 2022), which integrates machine learning models to assist with the identification of blank images and species. We also manually checked images and classified to the species level all mammals (except rodents, which were classified to the genus level). The catalogued data set was then downloaded from Wildlife Insights for analysis. Of the images analysed, a domestic dog (Canis familiaris) was identified as separate to a dingo (Canis dingo) by either the presence of a collar, a human walking the dog on a lead, or a human appearing in images captured immediately after those of the dog. Furthermore, based on previous genetic analysis of Cairns et al. (2021) demonstrating the rarity of feral dogs within Australia, any non-fox canid captured without the aforementioned domesticated identifiers was classed as a dingo in subsequent analysis.

Camera trap analyses

For the purposes of analyses, we collated all camera trap records for each camera trap site. We used RStudio Desktop (Posit Team 2023) to filter the dataset for 5-min intervals for each species’ records obtained from the same camera location. We used these filtered records for further analysis as independent events. We transformed all data using a 4th root transformation to avoid very common species dominating the analyses. We compared the relative abundance and diversity of species between sites using the Bray-Curtis similarity matrix, which does not use joint absences to compare sites. We visualised separation between sites using non-metric multidimensional scaling (nMDS). We compared Mt. Talaterang and Little Forest plateau sites using a Permutational Multivariate Analysis of Variance (PERMANOVA). We used a SIMPER analysis to determine which species contributed most to the separation between groups. All analyses were performed using PRIMER-E (Clarke and Gorley 2016) with PERMANOVA (Anderson et al. 2008) extension.

Identifying other mesas in NSW

We used topographic maps of NSW to identify other mesas across the state with similar topology to Mt. Talaterang, with the goal of identifying other sky-island safe havens. We carried out all spatial data modification and analysis in QGIS ver. 3.8 (QGIS.org 2023). We used a Topographic Positioning Index based landform classification (Weiss 2001; Jenness 2008) of a 3-s Digital Elevation Model of Australia (Gallant et al. 2009) clipped to NSW and reprojected to EPSG: 32356 – WGS 72/UTM Zone 56S. We used a small neighbour (400 m) and long neighbour (2000 m) combination that distinguished Mt. Talaterang from the surrounding landscape as a collection of points classified as mesa surrounded on all sides by points classified as high ridges. We kept all other parameters as default. We used this parameter combination as it allowed Mt. Talaterang to be distinguished from the surrounding topology and hence allowed us to identify other mesas with the same topological characteristics. We counted all mesas in NSW with the same landform classification as Mt. Talaterang and measured the ellipsoidal area in hectares of all points classified as a flat mesa top for each mesa. We recorded all mesas of 10 ha or greater as potential sites for further investigation as sky-island safe havens.

Camera trap data

We found mammal abundances at Mt. Talaterang and Little Forest Plateau differed significantly (Pseudo-F = 14.621; d.f. = 1, 30; P < 0.001) and there was clear separation between the two sites in the nMDS (Fig. 3, Table 1). Importantly, red foxes were absent from Mt. Talaterang and feral cats were detected significantly less on the mesa-top than the surrounding landscape (Table 1). Two other invasive species identified at the lower-lying Little Forest Plateau site were absent on Mt. Talaterang: (1) the European rabbit (Oryctolagus cuniculus); and (2) the domestic dog, (Table 1).

Fig. 3.

nMDS results on mammal abundances at Mt. Talaterang (mesa) and Little Forest Plateau (lower-lying site).

The species that contributed most to the separation between sites were red foxes (percent contribution 13.83%), Bare-nosed Wombats (Vombatus ursinus) (percent contribution 11.5%), feral cats (percent contribution 8.92%), Swamp Wallabies (Wallabia bicolor) (percent contribution 8.65%), and Eastern Grey Kangaroos (Macropus giganteaus) (percent contribution 8.18%) (Table 1). As with red foxes, rabbits, and dogs, Eastern Grey Kangaroos and dingoes were absent from Mt. Talaterang. Antechinus (likely Antechinus stuartii) were not detected on Little Forest Plateau (Table 1).

Mesas across NSW

We identified 91 mesas of ≥10 ha or greater with similar landform identity as Mt. Talaterang (Table S2). Of the other 91 NSW mesas, 14 were 50 ha or larger, and 5 were 100 ha or larger, with the 2 largest, greater than 300 ha in area (Fig. 4).

Fig. 4.

Map of New South Wales, Australia showing locations of mesas for potential use as sky-island safe havens and location of Mt. Talaterang, with insert map of Australia.

Getting beyond the fence

There is a growing need to identify additional strategies that can be integrated with fenced safe haven initiatives to restore mammal populations in Australia. Land procurement for fenced sites in suitable habitats is a continual barrier to safe haven development, along with logistical challenges inherent with fence construction and maintenance costs. For example, a Population Viability Analysis for the Eastern Quoll indicated an additional occupancy area of 437 km² is required to remove the species from the IUCN Red List within 10 years (Wilson et. al. 2022). This required land area is equivalent to 90 fenced reserves similar to the 485 ha Eastern Quoll safe haven, Mulligans Flat Woodland Sanctuary, developed in the ACT (not accounting for edge effects) (Wilson et. al. 2022). Due to the inherent difficulty in securing large parcels of land with suitable habitat for species recovery, geographical features such as mesas, functioning as sky-island refuges that may naturally restrict the incursion of red foxes and feral cats, could serve as viable supplementary translocation sites to aid conservation initiatives. These sites have the potential to bolster the extensive network of protected areas and aid in the management of species such as the Eastern Quoll.

We recognise that the geography of all mesas may not provide total protection from introduced predators. Variation in topography, such as fissures in rocks, waterfalls, and boulder fields, may provide the ability for many species, not only red foxes and feral cats, to move across mesa boundaries, albeit in a restricted manner. Indeed, even very low densities of feral cats can exert significant predation pressure on native species (Moseby et al. 2015, 2021). Similarly, even under intense management regimes, foxes have been known to extirpate reintroduced populations of critical weight range mammals including the Eastern Bettong (Bettongia gaimardi) (Evans et. al. 2021). Therefore, sky-island safe havens may not be suitable for species that are particularly vulnerable to cat and fox predation, or additional management may be required to address potential predator incursion points via the strategic placement of gates and fences. However, some level of ‘leakiness’ may be an advantage of mesas over entirely fenced safe havens that can disrupt long-term movement and migratory routes of non-target species.

As mesas may also be naturally semi-permeable to target taxa, sky-island safe havens could function under a ‘leaky fence model’ facilitating a halo or spill-over effect. In doing so, mesas may reduce the potential negative effects of overpopulation and genetic isolation associated with fenced or island safe havens (Agarwal and Bode 2019; Moseby et al. 2020a). Many mesas in NSW occur in relatively close clusters. For example, there are eight mesas of 10 ha or greater within a 20-km radius in Wollemi National Park. A ‘sky archipelago’ approach could therefore be developed to potentially form networks of protected core areas. However, to ensure the best chances of success individual mesas will need to be assessed on a case-by-case basis to evaluate for habitat suitability, access to appropriate foraging resources, and assess ‘leakiness’ to target species. Furthermore, the surrounding landscape will need to be managed to limit populations of introduced predators to low enough levels for persistence of target species beyond safe haven boundaries (Agarwal and Bode 2019; Moseby et al. 2020a).

The potential remoteness and limited accessibility of mesas may present management hurdles and lead to delays in emergency response times. For example, swift responses required to address red fox and feral cat incursion, or when a bushfire threatens the site may be delayed due to the topographic complexities limiting vehicle or even on-foot access. These limitations may be addressed in part by technological advances; including the deployment of remote ‘Felixer’ grooming traps that have been successfully used to selectively control invasive predators (Moseby et al. 2020b). However, the exposure to invasive predators in low numbers may potentially address another negative consequence of fenced safe havens – that of predator naivety. Exposing naïve mammals to real predation risk may provide individuals within target populations a degree of predator awareness (Moseby et al. 2015), which is key in increasing the likelihood or survival if interactions occur in future. This ‘Goldilocks Zone’, which is the ‘just right’ levels of predation needed to drive selection for predator-resistant native species, while ensuring population viability has been proposed recently (Evans et. al. 2021), and some of the sites identified within this study may provide suitable locations for developing this concept further.

Although sky-island safe havens may hold several advantages over fenced havens or islands, we recognise that mesa-tops may not be suitable for the release and survival of all species. Species that occur at lower elevations, are highly vulnerable to stochastic or extreme weather events, require extensive ranges, or require permanent water sources may not be candidates for release atop particular mesas. However, mesas may provide particularly valuable safe havens for montane species, including some critical weight range mammals within the Great Dividing Range of eastern Australia.