Study area and time
We conducted our surveys on two neighboring cattle stations, Dungowan (16°42′S, 132°16′E) and Camfield (17°2′S, 131°17′E), located in the northern margin of the Tanami Desert in the Northern Territory, Australia. Bore-fed reservoirs at AWP on both stations consist of a mix of earthen dams and tanks made of plastic or steel (Fig. 2). At both reservoir types, livestock are supplied with water through troughs located within 50 m of the reservoir (Fig. 1d). The troughs are fed by gravity and fitted with a float-valve to prevent them from over-flowing. Permanent fences prevent livestock from accessing the water stored in dams.
The study area has a mean annual rainfall of 580 mm, of which 96% falls in the wet season (November to April) and 4% in the dry season (May–October; Australian Bureau of Meteorology). The vegetation of the study area consists of open semi-arid savannah woodland with the dominant woody species of lancewood (Acacia shirleyi) and eucalypts (Eucalyptus leucophloia) and an understory dominated by grasses (Eriachne spp. and Sorghum spp.). We surveyed the foraging activity of sand goannas and the abundance of cane toads, skinks and dragons in April and November 2012, April and November 2013 and September 2014.
Cane toad abundance at AWP and along road transects
We estimated the abundance of toads in the direct vicinity of AWP by conducting nocturnal 4 m × 150 m strip transects radiating away from the AWP (n = 4 per AWP) using handheld 12 V spotlights with 25 W halogen bulbs. Cane toad abundance was calculated as the sum of individuals encountered along the four transects. We conducted a total of 42 cane toad counts at 31 AWP (ten dams and 21 tanks) (Online Resource 1); four dams and seven tanks were sampled twice during the study period with a minimum of 12 months between surveys.
Cane toads frequently travel along roads during dispersal periods (Brown et al. 2006). To document the distribution of toads with respect to distance from AWP, we conducted nocturnal surveys along low-use single lane dirt roads in a 4WD vehicle during a period when many toads had dispersed away from their dry season refuges at the end of the wet season in April 2012. Because of logistical constraints during field work, it was not possible to conduct road transects from all AWP. Hence, distance mediated effects were evaluated at a subset of dams (n = 4) and tanks (n = 5). We surveyed toad abundance over a total of 110 km at distances of up to 12 km from both reservoir types. The surveys were undertaken at a speed of 20 km/h and an observer noted with a GPS the location of all toads sighted. Toad activity was documented as number of toads per 500 m transect section.
Goanna activity indices
Sand goannas are difficult to survey using mark-recapture methods because they rarely enter traps (Letnic et al. 2004) and, in habitats with dense understory vegetation such as our study area, are difficult to sight and approach for the purposes of noosing, hand capture or visual surveys. Previous studies have used the occurrence of fresh goanna tracks and pits that goannas create whilst foraging to index goanna abundance (Paltridge 2002; Bird et al. 2014; Read and Scoleri 2014). Both indices have been validated against known abundances in other varanid species (Anson et al. 2014). Following these previous studies, we used two methods to index goanna activity, the occurrence of tracks (i.e. footprints and tail drag marks) crossing single-lane dirt roads and the occurrence of recent goanna foraging pits. Our track count index provided a measurement of goanna activity over a 24 h period, while the foraging pit index provided a cumulative measure of goanna activity for a period of approximately 1 month prior to our surveys. We conducted all monitoring under environmental conditions that favored lizard activity and ensured equal and high detection probabilities among track plots and surveys (Jessop et al. 2013a).
The track-based index of goanna activity was derived by scoring the occurrence (presence/absence) of tracks crossing 50 m track plots located along road transects radiating from dams and tanks. The transects were situated on low-use single lane dirt roads (Paltridge 2002; Read and Scoleri 2014). We surveyed 403 track plots over a total of 201.5 km (Online Resource 2). Each track plot consisted of a 50 m road section that was cleared of tracks on the day before the survey. Track plots were spaced 500 m apart and located between 0 and 12 km from the nearest AWP. We walked along each track plot and recorded the presence or absence of fresh goanna tracks (i.e., distinctive tail drags and claw imprints). As daily activity areas of sand goannas are unlikely to exceed an area of 200 m by 200 m (Green and King 1978), we are confident that each recorded track originated from a different individual.
The foraging-pit based index of goanna activity was derived by scoring the presence or absence of recent goanna foraging signs during 2 min active searches in the vicinity of each track plot (Jessop et al. 2013a). Whilst digging for fossorial prey, sand goannas leave characteristic ellipsoid foraging pits that often show deep scratch marks left by their strong forelimbs during excavation of the soil (Read and Scoleri 2014). We estimated the approximate age of foraging pits based on two criteria: the amount of leaf litter and other debris in the excavation and the coloration and texture of the excavated soil (initially darker and softer than the topsoil, gradually fading and hardening over the course of several weeks). To provide an indication of the age of foraging pits that we encountered, we excavated pits similar to goanna foraging pits and monitored them over a 2 month period. This allowed us to classify foraging pits into the two age classes of recent (i.e. younger than approximately 1 month) and old (i.e. older than 1 month). Only the presence of foraging pits younger than approximately 1 month was used for further analyses.
Small lizard abundance
To monitor the abundance of small lizards, we conducted 198 active diurnal searches (total search duration of 1980 min) following the methods of Lunney and Barker (1986) (Online Resource 3). During each of the surveys in April and November 2012 and 2013, we conducted 40 active diurnal searches (20 sites located near dams, 20 near tanks), during the survey in September 2014 we conducted 38 searches (20 sites located near dams, 18 near tanks). Active search sites comprised 1 ha (100 m × 100 m) plots and were spaced a minimum of 2 km apart and located along the same transects used to survey goanna activity. At each site, active searches were conducted simultaneously by two observers who portioned their search effort so that each observer restricted their search to a 50 × 100 m quadrat within each site. We recorded reptiles encountered on the ground, under logs, in litter, in grass and on stems and branches of trees. An observers’ experience bias was avoided by using random observer combinations for each survey. Each site was actively searched for 10 min (5 min per observer) and was conducted between 9:00 and 10:30 am. To prevent double counting, observers avoided walking the same paths twice. Sighted reptiles were identified to family level by their pattern and size and the microhabitat they were encountered in. All encountered skink species belonged to four genera (Carlia, Ctenotus, Lialis and Menetia), all encountered dragon species belonged to two genera (Amphibolurus and Diporiphora). The total number of individuals recorded during 10 min of active search was used as an index of the abundance of skinks and dragons at each active search site.
To investigate the alternative hypotheses that habitat disturbance by cattle or predation by dingoes or feral cats were factors influencing the abundance of goannas and/or smaller lizards, we recorded the presence of tracks of cattle, dingoes, and feral cats at each tracking plot. An index of cattle activity for each track plot and each active search site was expressed as the percentage of track plots with fresh tracks within a 1.5 km radius. To account for the wide-ranging habitat of dingoes, dingo activity was expressed as mean values obtained for each sub-site. Cat activity was omitted from the analyses owing to the low activity of cats (only 3.8% of the track plots contained cat tracks).
The reduction of vegetation coverage by fire is known to influence the abundance of sand goannas and smaller lizard species (Letnic et al. 2004; Bird et al. 2014). To investigate whether differences in the fire history could explain differences in the abundance of goannas, skinks and dragons we obtained data on the fire history of each active search site from the North Australian Fire Information.
Cane toad abundance at AWP and along road transects
We analyzed differences in cane toad density in the direct vicinity of AWP using a generalized linear mixed model (GLMM) with a Poisson distribution and a log link function. To account for multiple sampling between years, AWP identity was included as random factor. We analyzed differences in cane toad density along 12 km road transects radiating from the two different reservoir types using a generalized additive model (GAM) with a Poisson distribution and a log link function. To account for the nested structure of the data, we included the identity of each transect and the year in which the survey was conducted in as random factors. All GLMM and GAM analyses were performed in R Version 3.0.3 using the ‘glmm’ and ‘mgcv’ libraries.
Goanna activity and small lizard abundance along transects
We combined the track and foraging pit indices of goanna activity in our analysis and defined a plot as indicating recent goanna activity if goanna tracks and/or recent foraging pits were present. Because our road surveys revealed a decline of cane toad abundance at distances of up to 3 km from dams, followed by a steady count to distances up to 12 km (see “Results” section), we divided our 12 km transects into two sections (i.e. < 3 and > 3 km). For the initial analysis of goanna activity and small lizard abundance along transects we compared the averaged goanna activity and lizard abundance of each transect section between individual transects using GLMM with a normal distribution and log link function. Type of nearest AWP, distance to AWP and the interaction of type and distance were included as fixed factors in the models. To account for the nested structure of the data, we included the identity of each transect and the year in which the survey was conducted in as random factors.
Because our expectation in this study system was that even though the biomass of lizards (i.e., as small ectotherms) would be relatively high, the distribution of individuals is expected to be very patchy. This reflects well known observations, that semi-arid lizards, as consequences of sensitivity to heterogeneity in structural habitat resources (e.g. ground vegetation cover, course woody debris) and relatively small home ranges, can be extremely variable in spatial occurrence (Letnic et al. 2004). Our survey design thus considered that an increased number of plots sampled once, rather than fewer plots sampled repeatedly, would permit better encounter rates and less variation in lizard detection in an otherwise very large study area. We however acknowledge, that as a potential trade-off of this approach, our measurements of naive count data could not account for imperfect detection (Guillera-Arroita et al. 2014). Ideally, if time had permitted, we would have performed repeated count surveys on a large number of sites to allow use of potentially more robust count estimation methods (Royle and Nichols 2003).
Structural equation modelling
Because our initial analysis indicated significant differences in cane toad abundance, recent goanna activity and small lizard abundance between transects in the vicinity of tanks and dams (see “Results” section), we used piecewise SEM to further test hypotheses based on a priori knowledge of interactions hypothesized to occur between cane toads, goannas and smaller lizard species (Grace 2006). We constructed our a priori SEM model based on trophic cascade theory and prior knowledge of factors impacting the abundance of small terrestrial lizards. As opposed to classical SEM, where covariance matrices are used, piecewise SEM uses localized estimates to deduce direct and indirect effect pathways (Grace 2006; Colman et al. 2014). This approach allows the modelling of data that do not meet the assumptions of classic SEM and the incorporation of exogenous factors such as spatial dependence (Pasanen-Mortensen et al. 2013; Colman et al. 2014). Localized estimates within the SEM were fitted using a GLMM (Poisson log-link function; skink and dragon models) or LMM (goanna model). To account for the nested structure of the data we included the identity of each transect as a random factor. For the GLMM, an observation level random effect was included to account for overdispersion (Harrison 2014).
Our initial models were parameterized with values obtained for each active search site for the variables goanna activity (percentage of plots with recent goanna tracks and/or foraging pits within 1.5 km of the active search site), cattle activity (percentage of plots with cattle tracks within 1.5 km of the active search site) and the number of months since the last fire as well as with mean values obtained for each sub-site for dingo activity (percentage of plots with dingo tracks) to account for the wide-ranging habitat of dingoes. Because the impact of cane toads on goannas was negatively correlated with increasing distance from dams whereas increasing distance from tanks was not correlated with goanna abundance (see “Results” section), we used the distance from the nearest dam to each of the active search sites as a proxy for the impact of cane toads on goannas. We used a backwards step-wise elimination process for model simplification whereby the most non-significant predictor variables were sequentially deleted until all interaction were significant. The most parsimonious model was then selected using Akaike’s Information Criterion for small sample sizes (AICc) as that with the lowest AICc value (Burnham and Anderson 2002). Standardized path coefficients were calculated by normalizing data to fall within one standard deviation of a mean centered on zero and the amount of variance explained by each ‘piece’ of the SEM (i.e., the goanna, skink and dragon models) was assessed using marginal R2 values. The overall fit of the SEM was assessed using a Fisher C test and associated p value. The model is a good representation of the data if the Fisher C p-value is > 0.05. All SEM analyses were performed in R Version 3.0.3 using the ‘piecewiseSEM’ library.
Interaction pathways between variables were determined by applying a priori knowledge, which resulted in the following set of hypothesized pathways (Fig. 5a): (1) Cane toad activity should negatively affect goannas owing to lethal ingestion; (2) the foraging activity of goannas should negatively affect the abundance of skinks and dragons (Olsson et al. 2005); (3) because of selective predation pressure, goanna activity should have a stronger impact on skinks than dragons (Sutherland 2011); (4) we used distance to the nearest dam as a proxy for the impact of cane toads because the impact of cane toads on goannas was negatively correlated with distance from dams (see “Results” section); (5) dingo activity should negatively affect goanna activity owing to predation (Paltridge 2002); (6) habitat modification resulting from grazing by livestock can have detrimental effects on both the abundance of goannas and of smaller lizard species such as skinks and dragons (James 2003); (7) time since fire was hypothesized to positively affect populations of goannas (Bird et al. 2014) and to negatively affect skinks and dragons (Letnic et al. 2004).