Rocky landform specialists of the Mount Isa Inlier: camera trapping reveals seasonal occupancy and habitat associations of a unique faunal assemblage in an ancient landscape

Study area

The study area (Fig. 2) lies within Queensland’s Mount Isa Inlier bioregion and is characterised by hot, monsoonal summers, and dry winters with warm days and cool nights. Mean annual temperatures in the region are between 17.3 and 32°C (range = 8.7–37.3°C) (Bureau of Meteorology 2024). Rainfall is summer-dominant with an annual mean of 459.8 mm (Bureau of Meteorology 2024). Total survey area by minimum convex polygon was 11,523 km².

Fig. 2.

The study area showing camera station locations, major access routes, and local terrain ruggedness (Riley et al. 1999) to show the extent of rocky landforms within the study area.

Survey conditions

Although temperatures across the study area are relatively stable from one location to another, rainfall is often patchily distributed and totals at one location may not reflect totals at another. Weather data was therefore obtained from three Bureau of Meteorology weather stations within the survey area (Mount Isa Airport station ID 029127, Cloncurry Airport station ID 029141, and Trepell Airport station ID 037036), and averages calculated across all three stations. The study commenced following a below-average wet season, with an average total rainfall of ~390 mm between October 2017 and April 2018 (Supplementary data Fig. S1). Average daily temperature during the survey fluctuated similarly to long-term means. Total rainfall exceeded the long-term mean with ~599 mm recorded (Fig. S1). Almost 60% of rain fell in February 2019 alone, and >90% from January to March 2019.

Field surveys

Field surveys investigated the presence of vertebrate fauna on rocky landforms over a 13-month period by repeated targeted surveys using ScoutGuard/BolyGuard SG562-C white-flash PIR camera traps. Cameras were set to 24 h operation, ‘high’ sensitivity, three captures per trigger and a 0 s delay between photographs within each trigger. This camera has a trigger speed of 1.2 s and an in-built delay period of 15 s between trigger events; however, field testing showed that, with the chosen settings, the delay period between triggers was ~30 s. Several cameras were inadvertently set to video mode for at least one sampling period during the survey. These data were omitted from analysis.

We undertook a total of nine sampling periods at 60 camera stations (Fig. 2), all on rocky landforms and each consisting of a single camera. Intercamera station distance was between 772 and 24,462 m (mean ± s.d. = 3265 ± 4026 m). To assess whether camera stations were spatially independent we collated home range data for all identified specialists (or their analogues if no data were available: Supplementary data Table S1). Home range is generally more readily available than ranging distance or dispersal data and still provides some context to the likely maximum area covered by an individual. Home range size cannot account for occasional long-range movements of some taxa (e.g. Dickman et al. 1995; Higgins and Davies 1996; Puckey et al. 2004; Sharp 2009; Burnett et al. 2023), but such movements are typically stochastic (e.g. in relation to rainfall) and difficult to control for. Home range size also provides no information on home range shape, which can take many forms and is difficult to standardise. Noting these limitations, we calculated a hypothetical circular home range with a 386 m radius (equating to a minimum 772 m between camera stations), giving a minimum non-overlapping area of 46.8 ha between adjacent cameras. This calculation suggests that all camera stations could be considered spatially independent for all rocky landform specialists except the euro (Osphranter robustus erubescens) and potentially the purple-necked rock-wallaby (Petrogale purpureicollis). On average, however, camera stations were much further apart than 772 m, with a mean of >3 km, suggesting sufficient spacing to satisfy an assumption of spatial independence for all specialists.

Cameras were initially deployed and baited between 17 and 26 May 2018 and were operated continuously until 10 June 2019 (Supplementary data Table S2). During this period, cameras were serviced (batteries and memory cards changed) and rebaited on eight occasions with a service interval ranging from 27 to 68 days (mean ± s.d. = 50.88 ± 11.72 days). Bait consisted of a simple peanut butter lure (approximately 125 g) smeared directly onto the rock surface in the centre of the camera’s field of view. In each sampling period, between 56 and 60 camera stations were active simultaneously, over a total of 21,965 camera days (Table S2). Cameras were mounted horizontally on either a Joby GorillaPod 3 K tripod (Vitec Imaging Solutions, Italy; n = 46), star picket (n = 7) or tree (n = 7) depending on station-specific mounting opportunities. Distance to bait varied between 90 and 159 cm from bare rock (mean ± s.d. = 115.46 ± 13.56 cm). Regardless of the mounting method, cameras were deployed with the lens ~20 cm above the ground surface. Where necessary, the area within a camera’s field of view was cleared of vegetation to reduce the incidence of false triggers from vegetation movement. In August 2018, cameras were fitted with heat shields made from air-cell insulation (Kingspan Insulation, Australia) to protect them from extreme temperatures; however, these were later removed after interference from corvids (Corvus spp.) which resulted in cameras becoming dislodged.

Environmental covariates

Seven environmental covariates were recorded for each camera station (Table 1). Lithology, landform and plant species richness were recorded in the field. In some instances, landform could not be determined from visual assessment in the field and was later confirmed by assessment of aerial imagery. Elevation, aspect and ruggedness were recorded for each camera station in a geographic information system (GIS) environment (QGIS Development Team 2020).

Rocky landform specialist species selection

Not all species observed on rocky landforms are necessarily rocky landform specialists (i.e. saxicolous species). Additionally, the likelihood of recording non-target species on camera during this study was increased by baiting cameras with peanut butter, which is likely to attract a range of species not confined to rocky landforms. All taxa identified in camera trap images were therefore identified as rocky landform specialists or non-specialists. We adhere to the definition used by Freeland et al. (1988) in defining a rocky landform specialist as a species with a consistent association with rocky habitats across its entire distribution, and we support our classification of each rocky landform specialist through available evidence in the scientific literature (Table 2). Non-specialists were all species recorded for which this consistent association with rocky habitats was either not reported or not supported by the scientific literature. Further analysis was conducted only on taxa classified as rocky landform specialists. We also determined the dietary niche of rocky landform specialists, either from direct reports of a species’ diet, or inferred from the reported diet of congeners (Table 2).

Multispecies multiseason occupancy model

Camera trap data for each sampling period were truncated to the first 21 days to match the duration of the shortest sampling period (May 2019). Total effort of this truncated dataset was 10,966 camera days. We produced a detection database of rocky landform specialists in the R package camtrapR (Niedballa et al. 2016) using an independence interval of 15 min and retained all species with ≥10 detections across the entire study for further analysis. An independence interval of 15 min was chosen based on analysis of a histogram of detections of all analysed species at increasing time from the previous detection, which showed a clear peak for all species at <5 min (Supplementary data Fig. S2). We assumed that any detection interval beyond 5 min was therefore a suitable independence level and selected 15 min as a conservative interval. Species detection data were used to produce detection histories with an occasion length of 7 days to reduce computing time. Detection histories were subsequently used to develop a multispecies community occupancy model using the communityModel function in camtrapR (Niedballa et al. 2016). This function is generally utilised for developing single-season occupancy models and does not allow for dynamic multiseason models; however, by partitioning detection data into discrete sampling periods we were able to produce a multispecies multiseason occupancy model by including survey as a random effect (J. Niedballa pers. comm., Leibniz Institute for Zoo and Wildlife Research, April 2024). We also incorporated site-level covariates of aspect, basic lithology, landform, ruggedness and plant species richness. We omitted elevation from our analysis as it showed a significant moderate correlation with ruggedness (Kendall’s tau, τ_b = 0.21, P < 0.0001), and we deemed ruggedness to be more ecologically relevant. All covariates were included as random effects in the model to calculate species-specific occupancy estimates for each covariate alongside community occupancy estimates. Random effects were included for occupancy and detection probability intercepts to give species-specific responses for these parameters. The final model was implemented in the JAGS framework within camtrapR (Niedballa et al. 2016) using 15,000 iterations with a burn-in of 3000 across three chains. The model was tested for convergence with the Gelman–Rubin statistic (values <1.1 indicate convergence: Gelman et al. 2004; Supplementary data Table S3) and visual assessment of trace plots.

Significance of occupancy modelling was determined by analysis of 95% credible intervals of effect sizes (95% CIs not including zero are considered significant). Because baited camera traps are effectively a point survey with an indeterminate area of influence, occupancy estimates derived from studies utilising them should more correctly be considered estimates of site utilisation unless the methodology explicitly considers home range size, population density and dispersal ability of the target species (Mackenzie 2005; Efford and Dawson 2012; Rovero et al. 2013). Therefore, we consider effect sizes (±95% CIs) in this study as determining whether a given covariate significantly affects changes in the proportion of the study area utilised (hereafter ‘rates of use’), rather than affecting true occupancy rates.

Rocky landform specialists

Camera trap surveys recorded 34,861 independent detections of 70 taxa that could be identified to at least genus level (Fig. 3; Supplementary data Table S4; Table S5). We identified 15 taxa as rocky landform specialists and recorded 31,672 detections of these taxa across the entire survey (~91% of all detections). Over the complete survey period, rocky landform specialists had a detection rate (number of camera days detected per 100 camera days) of 3.06 (±6.26) compared to 0.18 (±0.89) for non-specialists (Table S4).

Fig. 3.

Number of detections (square root-transformed) of rocky landform specialists and non-specialists recorded on camera traps during the complete study period. Only binomial nomenclature is provided in the interests of space. Common names for all taxa are given in Tables S4 and S5. Values at the end of each bar are actual total detections based on a 15 min independence interval. (*) Based on field observations, Corvus sp. are almost certainly C. coronoides, but C. orru and C. bennetti also occur in the bioregion and were considered too difficult to differentiate in camera trap imagery. (^) Cryptoblepharus sp. includes the rocky landform specialist C. zoticus, which occurs syntopically with the non-specialists C. australis and C. pannosus and was too difficult to differentiate in camera trap imagery in this study. (#) Denotes invasive species.

We recorded 28,546 detections of 59 taxa within the 21-day sampling window, of which 26,190 were of 14 rocky landform specialists (~92% of all detections within this window). Of the 14 specialists, 12 had a sufficient number of detections (i.e. ≥10) to support further investigation (Table 2). The final rocky landform specialist community analysed consisted of two birds, four mammals, and six reptiles (Table 2).

Sampling period effects

There was a significant effect of sampling period on community rates of use (indicated by effect size 95% CIs), with a positive effect of the November 2018 sampling period and a negative effect of the April 2019 sampling period compared to the initial sampling period in May 2018 (Fig. 4). For individual rocky landform specialists, sampling period had a significant effect on rates of use for nine out of 12 species (indicated by effect size 95% CIs) (Fig. 5).

Fig. 4.

Effect size (correlation coefficient) of sampling period and environmental covariates on estimates of utilisation rates (±95% credible interval) by the community. Credible intervals that do not include zero are depicted by filled circles and indicate a significant effect of a covariate (for continuous variables) or a significant effect of a covariate level compared to the reference covariate level (for categorical variables) on estimates of utilisation rates by the community.

Fig. 5.

Effect size (correlation coefficient) of sampling period and environmental covariates on estimates of utilisation rates (±95% credible interval) by individual rocky landform specialists. Credible intervals that do not include zero are depicted by filled circles and indicate a significant effect of a covariate (for continuous variables) or a significant effect of a covariate level compared with the reference covariate level (for categorical variables) on estimates of each species’ utilisation rates.

Environmental covariate effects

There was a significant effect (indicated by effect size 95% CIs) of aspect on estimates of community rates of use (Fig. 4) and for six individual species (Fig. 5). At the community level, north-facing slopes were utilised less frequently than south- and west-facing slopes. At the species level, if aspect had a significant effect on a species, there was always at least a significant positive effect of west-facing slopes compared with south-facing slopes. If there was additionally a significant effect of south- and/or east-facing slopes the effect was always positive compared with north-facing slopes.

There was a significant effect (indicated by effect size 95% CIs) of lithology on estimates of community rates of use (Fig. 4) and for eight individual species (Fig. 5). At the community level there was a significant positive effect of metamorphic and igneous geologies compared with sedimentary geologies. At the species level, if lithology had a significant effect on a species, there was always at least a positive effect of igneous lithologies compared with sedimentary ones. If there was additionally a significant effect of metamorphic lithologies the effect was always positive compared with sedimentary ones.

There was a significant positive effect (indicated by effect size 95% CIs) on estimates of community rates of use on hills compared with jumpups (Fig. 4). Landform had a significant effect on nine individual species (Fig. 5), with each showing a significant positive response to one or more landform types compared with jumpups. No landform type appeared to be consistently important between species.

Ruggedness had no effect (indicated by effect size 95% CIs) on estimates of community rates of use (Fig. 4) but had a significant positive effect (i.e. rates of use increased with increasing ruggedness values) for three species and a significant negative effect (i.e. rates of use increased with decreasing ruggedness values) for five species (Fig. 5).

Plant species richness had no effect (indicated by effect size 95% CIs) on estimates of community rates of use (Fig. 4) but had a significant positive effect (indicated by effect size 95% CIs) on one species (Fig. 5).

Environmental covariate effects

Slope aspect affects abiotic factors such as light, heat, moisture and soil composition (Bale et al. 1998; Burnett et al. 2008; Nyman et al. 2015), and the underutilisation of northerly slopes may therefore reflect suboptimal conditions for the community. In the Mount Isa Inlier, the sun tracks north over most of the year, resulting in relatively higher levels of solar radiation hitting northern slopes. Although rocky landforms typically provide sheltering opportunities such as cracks and crevices that may help maintain stable thermal regimes (Shi et al. 2015; Millar and Hickman 2021), shelter sites that are too deep, shallow, thick or thin can be unsuitable as daily refuges and can even lead to mortality (Huey et al. 1989). The reduction in rates of use of northern slopes may therefore be because shelter sites on these slopes have less suitable morphology and are unable to effectively buffer incoming solar radiation. Further research on shelter sites of rocky landform specialists and the thermal regimes of these shelter sites are necessary to shed any further light on this possibility. Other studies from the southern hemisphere have shown that south-facing slopes are cooler than north-facing slopes in spring and summer (Speziale and Ezcurra 2015), and steep landforms with a higher concentration of south-facing slopes have stronger features of ecological refugia (i.e. to which a species can retreat, persist in and expand from: Buschke et al. 2020). In the Yarra Ranges of Victoria, northern slopes are significantly hotter than southern ones, with western and eastern slopes in between (Nyman et al. 2015). A similar pattern likely occurs in north-west Qld, so eastern and western slopes may receive enough of a reduction in solar radiation within the Mount Isa Inlier to be ecologically equivalent to south-facing aspects. Alternatively, east- and south-facing slopes may also provide basking sites for reptiles, as well as some mammals (Geiser et al. 2002; Creese et al. 2019), during the cooler parts of the day towards sunrise and sunset respectively, leading to increased rates of use of these areas compared with north-facing slopes.

The importance of metamorphic and igneous lithologies compared with sedimentary ones, at both the community and species level, highlights the importance of these broad lithology groupings. This trend may reflect the proportional availability of finer-scale categories of lithology and the ages and varying propensity of these geologies to weather and break down to form suitable sheltering and denning sites within the study area (Twidale and Campbell 2005). Sedimentary landforms surveyed in this study are predominantly laterite jumpups, which are extremely old and weathered and provide few sheltering opportunities. By contrast, igneous and metamorphic landforms generally consist of complex, often deeply fissured habitat with an abundance of sheltering and denning opportunities (e.g. granite boulder piles and quartzite scree).

Our data suggest that all landforms across the bioregion are utilised at similar rates by the community, except for a small positive increase in rates of use of hills compared with jumpups. Additionally, there were significant effects of landform on nine specialists, but these effects were not consistent among specialists. Unfortunately, detailed analysis of the true importance of some landform types (notably ‘outcrop’ and ‘ravine’) was hampered by small sample sizes, resulting in large credible intervals overlapping zero, especially at the community level. As discussed above, jumpups in the bioregion are predominantly composed of laterite, which is extremely weathered and provides relatively few sheltering opportunities. Beyond this, it is difficult to comment on why certain landforms may increase or depress rates of use, but it is likely that it is a result of a combination of site characteristics missing from our broadscale assessment, such as the abundance of preferred dietary items, water availability, vegetative cover, and presence of shelter sites (e.g. Soderquist and Nally 2000; Blevins and With 2011; Hohnen et al. 2016a). Our study therefore highlights landforms that may be potentially important for a range of rocky specialists, but further research is required to determine exactly why they are important.

Ruggedness did not have a significant effect on rates of use by the community, indicating that areas of low, moderate and high ruggedness were utilised at similar rates across the study area. This was in part driven by the importance of ruggedness at the species level, with a positive effect on three species and a negative effect on a further five. Rugged areas provide higher habitat value than less rugged areas due to the increased availability of den sites (e.g. Hernandez-Santin et al. 2022) and protection from feral predators (McDonald et al. 2017; McDonald et al. 2020). It was therefore surprising to find a negative effect on five specialists, especially given four of these (spinifex pigeon, Geophaps plumifera leucogaster; Carpentarian pseudantechinus, Pseudantechinus mimulus; Slater’s ring-tailed dragon, Ctenophorus slateri; and ridge-tailed monitor, Varanus acanthurus) may at least occasionally form part of feral cats’ (Felis catus) diet (Doherty et al. 2015). This relationship may be explained by the fact that, as this study was focused on rocky landforms, we did not survey any areas that could truly be considered ‘not rugged’ (e.g. sandplains) for a true comparison. Feral cat occupancy decreases with increasing ruggedness and habitat complexity (Hohnen et al. 2016b; McDonald et al. 2020), and we recorded a generally low number of detections of cats throughout the study, possibly indicating avoidance of our camera stations by cats. Therefore, although five species may have higher rates of use within the less rugged parts of our study area, this may not be the case in truly non-rugged locations. Further comparative analysis of rates of use of non-rugged locations by rocky landform specialists and feral cats would provide a clearer picture of this hypothesis.

Plant species richness had a significant effect on only a single species. Rates of use by Hosmer’s spiny-tailed skink (Egernia hosmeri) were positively affected by increasing plant richness, and this may be due to the species’ diet. While there are other herbivorous species within the community, E. hosmeri appears to have a more varied diet including leaves, stems, berries and seeds (Shea 1995) compared with the grass- and forb-dominated diets of the euro (O. r. erubescens) and P. purpureicollis (Telfer and Bowman 2006; Creese et al. 2019; Papp 2023) and the seed-dominated diets of G. p. leucogaster and Z. argurus (Frith and Barker 1975; Watts 1977; Begg and Dunlop 1985). Our limited data suggest that sites with increased plant richness may therefore be necessary for E. hosmeri to meet its dietary requirements, but further research is required in a range of areas, including further detailed dietary analysis, comprehensive floristic surveys, and phenological studies.

Our study is the first known attempt to quantify and explore the associations of the rocky landform specialist fauna community of the Mount Isa Inlier. It is difficult to assess how many studies have been conducted in the bioregion, as many are restricted to grey literature in relation to resource extraction and exploration and are therefore not publicly available. Of those studies that are accessible, they have typically been limited to single species and/or short study windows. Single season surveys may not provide a complete picture of a fauna community, especially if members of that community are rare or have low mobility (Thompson et al. 2010). Our study demonstrates this, with the Eromanga Basin knob-tailed gecko (Nephrurus eromanga) not being recorded until the sixth sampling period in January 2019. Even relatively long surveys such as ours may miss species. We know of several rocky landform specialists within the Mount Isa Inlier that were not recorded during our survey despite the significant amount of survey effort. In some instances, this may be random chance, especially if a species is rare or has low mobility, but in other instances our methodologies may have been unsuitable for recording all specialists. Camera traps are generally more effective at detecting a wide range of taxa compared with other census techniques (Wearn and Glover-Kapfer 2019), but our use of this single methodology almost certainly led to an underestimate of certain taxa such as bats and small, cryptic reptiles. Additionally, cameras were baited with peanut butter as it is a cheap and readily available product and is commonly utilised as a universal bait in camera trap studies. However, it is likely that the exclusive use of peanut butter biased our estimates towards those species that are known to be attracted by this type of bait, namely small- to medium-size mammals (Paull et al. 2011; Diete et al. 2016; Goldingay 2024). Other taxa may respond more favourably to other bait types, as shown by others (e.g. Davis et al. 2008; Sebastián-González et al. 2020). It is also possible that some taxa responded negatively to the peanut butter bait. Future studies of the community could explicitly test this hypothesis by comparing site utilisation estimates by using multiple bait types (including unbaited controls) and modelling this as a detection probability covariate.