Decline in semi-arid reptile occurrence following habitat loss and fragmentation

Study area

The study was conducted on a mixed-agricultural property 12 km north of Balranald in south-eastern Australia (34°32′13″S, 143°31′11″E). The study area was selected due to the spatial configuration of remnant habitat in two agricultural land use types (cropping and grazing) coupled with a private reserve. Historically, the entire property was grazed by sheep (Ovis aries). In 1975, a large fire burnt through the entire property leaving only a small number of unburnt trees and resulting in relatively even-aged stands of regrowth. In 1998, a 4000-ha reserve was established on the property to offset a 3-year phase of land clearing for agricultural purposes, whereby all livestock grazing in the reserve ceased. Between 1998 and 2001, a network of remnant mallee patches was retained. Grazing occurs across the eastern section of the property and in 2003, crops were established on approximately half of the cleared land and sown biennially. During the study period, the cropping areas remained fallow and stocking levels were 1000 head of sheep per 1500 ha. Pest control, including 1080 baiting and lethal destruction via shooting, was applied throughout the cropping and grazing areas, targeting European red foxes (Vulpes vulpes), feral cats (Felis catus) and European rabbits (Oryctolagus cuniculus). No pest control was conducted in the reserve during the study period.

The main vegetation communities in the study area include sandplain mallee woodland (hereafter referred to as sandplain mallee) and dune mallee woodland (hereafter referred to as dune mallee). Mallee woodlands are widespread throughout the arid zone and have experienced significant habitat loss and extensive clearing due to the expansion of agricultural enterprises (Clarke et al. 2021). Sandplain mallee is dominated by yorrell (Eucalyptus gracilis) and oil mallee (E. oleosa) overstorey, on fertile red-brown soils with a shrubby understorey of various chenopod and Acacia sp. species (OEH 2022). Dune mallee is dominated by white mallee (E. dumosa) and slender-leaved red mallee (E. leptophylla) overstorey, typically containing a low soil clay content with deep red sand dunes supporting T. scariosa hummocks (OEH 2022). The climate is classified as semi-arid, the average annual rainfall is 323 mm and the mean minimum and maximum temperature ranges from 16.6 to 33.1°C in summer and 3.5 to 15.7°C in winter (BOM 2022).

Study design

In total, 20 sites were selected in this study (Fig. 1), classified as either reserve or patch. Eight sites were established within the reserve to serve as reference sites and 12 sites were established within patches of remnant vegetation and ranged in size from three to 21 ha (mean = 14 ha). Each site was further stratified by vegetation type (sandplain mallee or dune mallee) and dominant land use (continuous vegetation, cropland or grazing land) (Fig. 2). The distance between sites ranged from 900 m to 1510 m in the cropping landscape, 500 m to 1700 m in the grazing landscape and 400 m to 2400 m in the reserve, thus ensuring independence between sites and survey periods. Because we were primarily interested in sampling a wide variety of patches, area-controlled survey effort was not employed; instead, a single site was established within each patch. Time and budget constraints also prevented additional sites being established within the larger remnant patches. All sites were constrained to the same property to avoid confounding effects associated with potential differences in land management practices.

Fig. 1.

Location of the study area in New South Wales, south-eastern Australia.

Fig. 2.

Study design showing the spatial arrangement of the 20 survey sites stratified by site type. SM, sandplain mallee; DM, dune mallee.

Survey protocol

At each site, reptiles were surveyed on three repeat occasions between November 2021 and February 2022 using pitfall and funnel traps (Baumgardt et al. 2021), with active searches of natural habitat constrained by time (1 h) and area (1 ha). Repeated surveying was conducted to minimise bias in species detection levels and account for potential issues associated with imperfect detection (MacKenzie et al. 2009). Active searches involved scanning the area for reptile activity (visual encounters), turning over and inspecting logs or debris, sifting through leaf litter and lifting loose bark on trees. Reptiles were captured by hand where possible, identified to species level using Wilson and Swan (2021) and released at their point of capture.

Trap arrays consisted of 3 × 20-L buckets spaced 15 m apart, connected by a 30-m drift fence (damp coarse), with two funnel traps (dimensions: 75 cm long × 18 cm wide × 18 cm high) placed either side at the 10-m and 20-m points along the fence. Funnel traps are a complementary method to pitfall traps because they often capture more snake species and larger-sized lizards (Thompson and Thompson 2007). In total, 25 trap nights were applied to each site during each of the three survey periods, amounting to 1500 trap nights (i.e. three buckets + two funnels × five nights × three surveys × 20 sites). Traps were checked each morning and before sundown. In each bucket, a paper plate, sand and leaf litter were placed in the bottom to provide shelter to captured animals. Branches and leaf litter were placed on top of funnel traps to provide shade. A single trapping array was established at each site and positioned at least 50 m from the edge of the patches to avoid potential confounding associated with edge effects. Each individual reptile was weighed to the nearest g using a handheld spring scale (Persola) and calico bag, measured (total length and snout–vent–length) to the nearest mm using a ruler and sexed from external features when possible. Individuals were marked using a permanent marker pen to identify recaptures and released 5 m from their point of detection. Richgro© ant sand was applied around the pits and funnels, and all funnel traps were closed when temperatures were forecasted to exceed 33°C to minimise animal welfare issues.

Statistical analysis

Prior to modelling species occurrence, we investigated which species could be reliably detected at a site by using single season occupancy-detection models (MacKenzie et al. 2002), fit using the ‘unmarked’ package (Fiske and Chandler 2011). Species were considered at this stage if they occurred at >25% of sites and had a minimum of 10 detections. Occupancy–detectability models were fit to the repeated surveys (n = 3) with a single predictor variable for detectability indicating the survey method type: either pitfall/funnel trapping or active search. This generated a nightly detection probability for each of the two survey methods. Next, we calculated the cumulative probability of detecting each species, if present, across the 15 nights of pitfall/funnel trapping and the three active searches following Kéry (2002). To reduce the probability of false absences, we used a threshold adapted from Nimmo et al. (2014) where only species with >80% cumulative probability of being detected (if present) were included in subsequent modelling of presence/absence or abundance. Eight of the 10 species meeting the selection criterion met the cumulative detection probability threshold. All species were classified as either habitat generalists or specialists, specifically within the Murray–Darling Depression bioregion. Justification for the designation of species as habitat specialists or habitat generalists was based on the primary literature cited in Greer (2022), and macro and microhabitat accounts reported in Wilson and Swan (2021) and the Atlas of Living Australia (ALA) (2023) (see Supplementary Table S1). Thus, habitat specialists were species restricted to a specific type of vegetation community or microhabitat type within the specified bioregion.

Because we were interested in the influence of habitat loss and fragmentation on the presence/absence of reptile species, we fit a series of Generalised Linear Models (GLM) using logistic regression. The robust slider (Lerista punctatovittata) was present at 95% of sites, so models of presence/absence for this species would not be informative. Instead, this species was modelled using abundance (count data) as the response variable, thus specifying a Poisson distribution. Patch size can influence species occurrence patterns; therefore, we first explored the relationship between patch area (ha) (n = 12) and species presence/absence. We then considered six additional GLMs for each species. The first model concatenated all categorical variables considered during the experimental design (site type × vegetation type × matrix type) to create a categorical variable with six levels. This model was supported if a species was sensitive to all three predictor variables. The second model included a four-level categorical variable that concatenated site type and vegetation type, the third model included a three-level categorical variable that captured the landscape matrix surrounding the site (i.e. continuous vegetation, cropland or grazing land), the fourth model included only vegetation type and the fifth model included only site type. The sixth model was a ‘null’ model and included only an intercept term and would be supported if the species did not respond significantly to any of the other predictor variables. Some models failed to converge due to complete separation and were therefore refit within a Bayesian framework using the brglm package (Kosmidis 2021).

Akaike’s Information Criterion (AIC) was used to identify the best-performing model(s) for each species using the package ‘AICmcodavg’ (Mazerolle 2020). The AICc value was used to determine models with a considerable level of support. The AICc of candidate models were compared against the best model (lowest AICc) (Burnham and Anderson 2002). Models with AICc values <2 were considered to have the greatest support in explaining species presence (and abundance for L. punctatovittata) (Burnham and Anderson 2002). The coefficient of determination (R²) was calculated for each best fit model(s) to determine ‘goodness of fit’. Model diagnostics were examined (QQ plots, residual vs predicted values) using DHARMAa package (Hartig 2022). Models with significant effects (P-value ≤ 0.05) were plotted using packages ‘ggeffects’ (Lüdecke 2018) and ‘ggplot2’ (Wickham 2016). All analysis was performed in R ver. 4.2.2 (R Core Team 2022).

Summary statistics

In total, 480 individuals from 31 species and nine families were detected (Table 1). Acknowledging the uneven sampling effort between reserve and patch sites, we recorded a total of 332 individuals (69%) from 24 species in the eight reserve sites and 148 individuals from 12 sites in the remnant patches (17 species accounting for 73 individuals detected in the cropping matrix and 16 species with 75 individuals in the grazing matrix). The most abundant species was the eastern mallee dragon (Ctenophorus spinodomus), accounting for 114 observations (24% of total detections), followed by the eastern robust slider (L. punctatovittata; 70 observations) and dwarf three-toed slider (Lerista timida; 46 observations).

Nine species (29%) were recorded only in the reserve and accounted for 152 detections, whereas seven species (23%) were only found in patches and accounted for 18 detections. Species restricted to the reserve included C. spinodomus, nobbi dragon (Diporiphora nobbi), eastern stone gecko (Diplodactylus vittatus), eastern beaked gecko (Rhynchoedura ormsbyi), red-tailed worm-lizard (Aprasia inaurita), Mitchell’s short-tailed snake (Suta nigriceps), royal ctenotus (Ctenotus regius), Bougainville’s slider (Lerista bougainvillii) and the desert skink (Liopholis inornata). Species found exclusively in patches included the eastern thick-tailed gecko (Underwoodisaurus milii), eastern brown snake (Pseudonaja textilis), bandy bandy (Vermicella annulata), Bynoe’s gecko (Heteronotia binoei), shingleback (Tiliqua rugosa), dark-spined blind snake (Anilios bicolor) and prong-snouted blind snake (Anilios bituberculatus). L. punctatovittata and L. timida were the only species detected across all treatments (see Fig. S1).

Detection probability by survey method

In total, 10 species met the criteria to be modelled, occurring at >25% of sites with a minimum of 10 detections (Table 2). There was no significant difference in the probability of detection across survey methods for six species: barred wedge-snouted skink (Ctenotus schomburgkii); common dwarf skink (Menetia greyii); L. punctatovittata; sand goanna (Varanus gouldii); shrubland pale-flecked morethia (Morethia obscura); and the southern spinifex ctenotus (C. atlas). Four species, D. nobbi, M. boulengeri, L. timida and C. spinodomus, were significantly less likely to be detected using pitfall/funnel traps compared with active searches (see Table S2). The cumulative detection probability for the pitfall/funnel trap nights and three active searches revealed that eight of the 10 species satisfied the requirements of having a >80% chance of being detected given their presence at a site (Table 2).

Model selection

We found no significant relationships between patch size (n = 12) and occurrence patterns for any species. Therefore, patch size was not included in subsequent model building. The model with site type (reserve or patch) was the best-fitting model for C. spinodomus, D. nobbi, C. schomburgkii and M. obscura (Table 3). All four species were more likely to occur in the reserve (probability of occurrence 75%+) compared with patches (Table 4, Fig. 3a–d). The landscape matrix model (continuous, cropping or grazing) was supported by C. schomburgkii and M. obscura (Table 3). Both species were more likely to be detected in continuous vegetation followed by patches in the grazing matrix (Table 4, Fig. 4a, b). The only model with substantial support for C. atlas and L. timida was the vegetation-type model (sandplain mallee or dune mallee) (Table 3). Ctenotus atlas was significantly more likely to occur in dune mallee (Table 4, Fig. 4c), whereas L. timida was more likely to occur in sandplain mallee, although the difference was not significant (Table 4, Fig. 4d). The site and vegetation-type model had substantial support from M. obscura and V. gouldii (Table 3). M. obscura was most likely to occur in sandplain mallee reserve sites and was least likely to occur in sandplain mallee patches (Table 4, Fig. 4e). Varanus gouldii was most likely to occur in dune mallee vegetation patches (Table 4, Fig. 4f). The best-performing model that explained L. punctatovittata abundance included site type × vegetation type × matrix type (Table 3). Lerista punctatovittata was significantly more abundant in dune mallee patches surrounded by cropland and sandplain mallee in the reserve (Table 4, Fig. 4g). Confidence intervals around these predictions are wide in many instances.

Fig. 3.

The probability of occurrence of (a) Ctenophorus spinodomus, (b) Diporiphora nobbi, (c) Ctenotus schomburgkii and (d) Morethia obscura between reserve and patch sites. Error bars represent 95% confidence intervals.

Fig. 4.

The probability of occurrence of: (a) C. schomburgkii and (b) M. obscura on sites surrounded by continuous, cropping or grazing land use types; (c) C. atlas and (d) L. timida in sandplain mallee or dune mallee vegetation type; (e) M. obscura and (f) V. gouldii in site and vegetation type combinations; and (g) L. punctatovittata estimated abundance across treatments (RC, reserve continuous; PC, patch cropping; PG, patch grazing; S, sandplain; D, dune). Error bars represent 95% confidence intervals.

Which species are sensitive to habitat fragmentation?

Reptiles are often highly sensitive to habitat loss and fragmentation (Keinath et al. 2017). Of the eight species we modelled, five species were classified as habitat specialists (C. spinodomus, D. nobbi, C. schomburgkii, C. atlas and M. obscura). Of these species, C. spinodomus, D. nobbi, C. schomburgkii and M. obscura were significantly associated with the reserve, with between 73% and 83% probability of occurring in the reserve and only 3–10% probability of occurring in patches. These results are consistent with previous studies on mallee reptiles in south-central NSW, where D. nobbi (syn. Amphibolurus nobbi), M. obscura and C. schomburgkii were found to be biased towards a mallee woodland reserve and C. spinodomus (syn. Ctenophorus fordi) was detected exclusively within a reserve (Driscoll 2004). Driscoll (2004) concluded that D. nobbi and C. spinodomus are extremely sensitive to fragmentation, occurring in high abundances in the reserve while being absent from patches. By contrast, Driscoll and Hardy (2005) found genetic evidence indicating high dispersal and migration in D. nobbi from a reserve within a fragmented agricultural landscape to ungrazed linear remnants. It is possible that the patches in our study were too widely spaced (averaging 1000 m) between neighbouring patches, or the reserve, to enable sufficient dispersal and maintain viable populations within remnant patches, or that disturbance by livestock grazing (and trampling) influenced shelter and breeding site suitability. The habitat specialist, C. atlas, was the only species that was not significantly associated with the reserve. This may be due to C. atlas having the ability to disperse through fragmented landscapes or because its requirements of mid to late post-fire successional stage Triodia cover were met within patches dominated by dune mallee (Smith et al. 2011; Verdon et al. 2020).

Three of the eight species modelled were classified as habitat generalists (V. gouldii, L. timida and L. punctatovittata). These three species were widespread across the study area. L. punctatovittata occurred at 100% of patches and L. timida and V. gouldi occurred at 83% and 50% respectively, suggesting these species are less vulnerable to processes associated with habitat loss and fragmentation. Furthermore, we found L. punctatovittata was significantly more abundant in dune mallee patches in the cropping landscape compared with any other site type. L. punctatovittata is a nocturnal, fossorial species associated with extensive leaf litter (Henle 1989). Driscoll (2004) found this species attained extremely high numbers along linear roadsides, conceivably due to increased nutrients and food resources. Cultivation around patches may act as a barrier to dispersal and reduced competition for resources with specialists could also explain the higher numbers observed. Lerista punctatovitta also responds positively to destocking (Neilly et al. 2021), so another plausible explanation is that this species has increased in response to the reduced level of grazing pressure across the property.

How important is the matrix?

The matrix can have a substantial influence on ecological communities and species distribution patterns (Michael et al. 2008; Munguia-Vega et al. 2013). Cropland in particular can act as a barrier to reptiles and restrict movement among remnant vegetation (Kay et al. 2016; Hansen et al. 2020). We predicted that patches in the cropping matrix would have lower occurrence rates than continuous vegetation (reserve) or patches in the grazing matrix. However, we found little difference in species occurrence patterns in patches among matrix types. Matrix type only featured in the models for C. schomburgkii and M. obscura, where both species were more likely to occur in sites surrounded by continuous vegetation and therefore were more likely to occur in the reserve. The limited support for our prediction could be attributed to the relatively small sample size, lack of replication, relatively recent land clearing and potentially confounding influence of unmeasured disturbances. Structural similarities and the same time since clearing between cropping and grazing matrices may indicate similar ecological impacts on reptiles. Pulsford et al. (2017) found that the intensity of matrix use for agricultural purposes was more important than matrix type itself in predicting reptile distribution patterns in temperate woodland landscapes. Therefore, improving matrix quality and appropriately designed matrix improvements can help increase conservation outcomes in modified landscapes (Franklin and Lindenmayer 2009; Driscoll et al. 2013).

How important is vegetation type in explaining species distribution patterns?

At large spatial scales, reptile communities are strongly influenced by vegetation communities. Sass (2006) found that mallee sites supporting Triodia maintained higher abundance and diversity of reptile species than areas without Triodia. We predicted that habitat specialists would be restricted to dune mallee, due to the presence of Triodia, whereas habitat generalists would be distributed across vegetation types. We found support for this prediction for C. atlas, which was associated with dune mallee. This pattern is consistent with previous studies that found Triodia cover to be an important predictor of C. atlas occurrence (Verdon et al. 2020; Bell et al. 2021a). We found limited support for a vegetation type effect for C. spinodomus, a species reported to have affiliations with early–mid stage Triodia cover (Nimmo et al. 2012; Sadlier et al. 2019; Verdon et al. 2020). However, the lack of a vegetation type relationship in our study may be due to the species’ complete absence from patches, including those with Triodia, hence the overriding effect of fragmentation. The only other species that exhibited a vegetation type effect was M. obscura, which was most likely to occur in sandplain mallee within the reserve. Preference for chenopod-dominated vegetation communities with open canopies and abundant leaf litter are consistent with documented habitat relationships for this species (Triska et al. 2016; Dundas et al. 2021).

Management implications

There are several broader management implications that stem from this study. First, offsetting habitat loss by establishing in-perpetuity conservation areas representative of the vegetation communities affected by agricultural intensification is an important factor in preserving reptile diversity on private land. Second, remnant patches within cleared areas could be better managed to improve habitat suitability for reptile specialists. Although we were unable to explain the mechanisms behind the decline in habitat specialists in this study, strong habitat affiliations with foundation species such as Triodia suggest that declining habitat condition and/or extent could be responsible. Triodia cover can be substantially altered due to agricultural activities, especially soil compaction and increased soil nutrients (Bell et al. 2021b). Fencing remnants to control livestock grazing pressure immediately following habitat fragmentation may help to preserve ground cover condition (Pulsford et al. 2017). Small-scale ecological burns could also be trialled to promote Triodia growth and post-fire seral stages. Third, improved matrix management through managing stocking levels and strategically orientating crops (Kay et al. 2016) could reduce barrier effects between remnants. Lastly, future research could trial small-scale translocations (Watson and Watson 2015) to investigate the feasibility of recovering locally extinct species.