Prediction 1: Small mammals will recognise and avoid the odour of V. gouldii
To test our first prediction, we used faecal material from V. gouldii as an odour source to present to small mammals. Varanus gouldii was captured during a long-term live trapping program in the study area (Dickman et al. 1999a, 2010, 2014; Greenville et al. 2016a) or by hand noosing following opportunistic sightings, and faeces produced during handling were placed in vials and frozen at -2 – -4°C within 30 min. We used a giving-up density (GUD) experiment to gauge small mammal responses to the odour. The GUD is the amount of food that remains when an animal has finished foraging in a patch with an enriched food source. We provided food in an inedible matrix and allowed animals to choose between treatments with these food sources. If there are no constraints on foraging, individuals should spend adequate time in patches and consume as much food as needed. However, if an animal experiences a stressor, such as predation risk, in a food patch then GUDs increase and animals spend less time in the patch (Brown 1988; Kotler et al. 1991).
To implement a GUD experiment, we established 54 sites, each >25 m apart, along the bases or mid-sections of sand dunes where V. gouldii and small mammals were active. At each site we set up a food patch. This was a half-buried plastic bowl (15 cm diameter, 4 cm deep) containing food of high value to small mammals: mealworms Tenebrio molitor, which were energetically profitable for S. youngsoni (Fisher and Dickman 1993) or peanut quarters, which we considered would be of value to the study rodents (Murray and Dickman 1994; Murray et al. 1999). Either 10 mealworms or 10 peanut quarters were placed in each bowl, mixed into 200 ml of sifted sand. The underside of the bowls was smeared with a mixture of Vaseline and Coopex insecticide powder (Bayer, Pymble, New South Wales) to repel ants from the bowls. Twenty-seven bowls were set with each type of food, with the order of placement randomized. To assess small mammal responses to V. gouldii odour, we used cotton buds—6 cm-long sticks tipped with absorbent cotton wool (Johnson & Johnson, Sydney, New South Wales)—for odour presentation. The tip of a cotton bud was applied to fresh goanna faeces and then the stick was inserted vertically into the centre of a food bowl such that the odour-bearing cotton tip was ~2 cm above the bowl surface. To quantify the magnitude of response to V. gouldii odour, we also used cotton buds dipped in eucalyptus oil as a pungency control treatment (i.e., representing a strong but familiar odour) and cotton buds dipped in water as a procedural control (Kovacs et al. 2012). The three treatment odours were allocated randomly to bowls with each food type, so that n = 9 for each odour for the bowls containing mealworms and n = 9 for bowls with peanut quarters.
As our target small mammal species are nocturnal, the bowls were set up in the early evening and revisited near dawn when foraging was expected to have concluded. The GUD (number of mealworms or peanuts remaining in each bowl) was then recorded. The bowls were set with food but without the odour treatment for one night to allow animals to accustom to them. The bowls were then recharged each evening with fresh food (mealworms and peanut quarters) and new cotton buds with fresh odours (sand goanna, pungency, procedural control) over seven consecutive nights. Disposable latex gloves were used at all times to minimise cross-contamination of odours. To identify the small mammal species that had visited the food bowls, and thus whose GUDs were being measured, we smoothed the sand in a 10 cm-radius around each bowl and examined the footprints that were left each morning. The prints of the target species are readily distinguishable by differences in size, numbers of toes and imprint of the heel (Moseby et al. 2009; Dickman et al. 2010). GUDs could be recorded reliably if a bowl had been visited by one species or, if two or more species had visited, we could clearly discern the last forager by the overprinting of its tracks on those of earlier foragers. We discarded the results for any bowls if we could not reliably read the prints or if the site had been disturbed by wind or other species such as Australian ravens Corvus coronoides. The experiment testing our first prediction was carried out once, at Main Camp, in November 2014.
V. gouldii will be attracted to the odour of small mammals
To test our second prediction, we created artificial burrows that simulated those of the study mammals. The burrows were made by hammering PVC pipe into the soil and then gently removing it, with soil inside, to create a vertical 'burrow' that was 2.5 cm or 3.5 cm wide and ~25 cm deep. The narrower (2.5 cm) burrows were used on one occasion only (September 2012) and then abandoned in favour of the wider burrows to facilitate ease of creation and subsequent manipulation of odours. Clusters of four burrows were constructed at 32 – 50 sites along the bases or mid-sections of sand dunes, with each cluster spaced >200 m from the next. In sand dune habitat V. gouldii occupies activity areas of 5.9 ha (274 m linear distance if the area is assumed to be circular), but take a week to cover them (Bolton and Moseby 2004). Thus we assumed that our burrow clusters would be accessed by no more than a single V. gouldii provided they were set for a week or less. By contrast, the artificial burrows within a site were spaced 2 – 3 m apart so that an individual V. gouldii encountering one burrow in a cluster would have an approximately equal chance of encountering the others.
We tested the ability of V. gouldii to find and discriminate prey odours by placing small balls of cotton wool bearing the body odour of the study species at the bottom of each artificial burrow. Body odours were used in preference to urinary or faecal odours as integumentary chemicals are most likely to be deposited in burrows as small mammals enter and exit them. Within each cluster, we presented the odour of three species of small mammals—one odour per artificial burrow—and placed fresh cotton wool in the fourth burrow to serve as an odourless control. Field visits in September and November 2012 used the odours of N. alexis, P. hermannsburgensis and R. villosissimus; the latter species disappeared from the study area in 2012 (Greenville et al. 2013), and its odours were replaced by those of S. youngsoni for experiments carried out in September 2013, April and November 2014. We term these two periods phase 1 and phase 2, respectively. Cotton wool was imbued with the odour of these species by providing captive or wild-caught animals with balls of this material as bedding for periods of >12 h. The material was then frozen at -2 – -4°C to reduce degradation of volatile components (Fardell et al. 2021), and inserted into the artificial burrows by operators wearing latex gloves and using clean forceps. As the cotton wool could be seen at the bottom of the burrows when the sun was overhead, a small (4 × 4 cm) patch of clean brown cloth was placed above the cotton wool to obscure it and ensure that odour was the only cue provided to foraging V. gouldii. Sand in a 30 cm radius around each artificial burrow was smoothed to capture the tracks of visiting V. gouldii.
The burrows and their experimental odours were set up in the mornings before V. gouldii became active, and were checked in the evening. We recorded the tracks of V. gouldii in the 30 cm radius around burrows and whether burrows had been dug into. This goanna characteristically excavates burrows by tearing at the soil from one side of a burrow with their forelimbs and following the burrow to the bottom where prey is usually located. Here, we recorded digging if a burrow had been partially or fully excavated. If a burrow in a cluster had been dug, we relocated the burrow cluster to a new position about 10 m away. All burrow sites were moved to new positions after 2 – 7 days; varanids quickly learn to distinguish profitable from non-profitable feeding sites and modes of prey capture (Manrod et al. 2008), and we assumed that individual V. gouldii would lose interest in small mammal odours with successive unsuccessful digs. Cotton wool balls bearing the experimental odours were replenished every 2 days to maintain their freshness, and inspections of goanna activity were made each evening.
Prediction 3: Small mammals will be less mobile and will show higher burrow fidelity in the absence than the presence of V. gouldii
Test of our third prediction required a small scale experimental removal and relocation of V. gouldii so that we could compare small mammal mobility and burrow fidelity in the presence and absence of this apex reptilian predator. We selected an area of ~50 ha at Main Camp with access tracks and three trapping grids on which we could capture V. gouldii, and commenced a removal program in September 2009. Between this time and November 2011, 39 V. gouldii were captured and relocated in similar habitat >5 km away. This goanna is not territorial and occupies areas that shift over time as animals track prey (Bolton and Moseby 2004), and our tracking of a subset (n = 11) of relocated individuals indicated that they rapidly established new burrows and maintained their mass and condition following translocation (CD, NH, unpub. data).
To gauge the degree to which the removal area was free of V. gouldii, we set up three 30 m × 1 m transects on each of the three removal grids and a further nine transects on unsealed vehicle tracks through the removal area. The transects were raked and then smoothed by dragging a half-filled hessian sack along their length to create a suitable surface to record animal tracks, and were checked daily for three days on each of six field trips between September 2009 and November 2011. The tracks of V. gouldii were recorded on 12 of the 18 transects (67%) in September 2009 but fell to 0 on three field trips in 2011 as the removal protocol progressed. It is likely that some V. gouldii activity remained in the removal area in 2011 as we found tracks away from the transects on two occasions in that year. However, the results indicate that activity of V. gouldii on the removal area had been reduced to nearly zero. By contrast, similar monitoring at the same time of a control area 7 km away, where no removals were undertaken, showed that V. gouldii activity on transects remained relatively unchanged. There, 10 of 18 transects (56%) yielded sand goanna tracks in September 2009, compared with 8 – 14 of the 18 transects (44% – 78%) on the field trips in 2011.
In view of the above results, we used radio-tracking to quantify small mammal mobility and burrow fidelity in the control and removal areas on three occasions in May, August, and October – November 2011. Small mammals were live-trapped on long-term trapping grids in each area (Dickman et al. 1999b, 2014) and equipped with single-stage radio-tags with 10-cm trailing whip antennae that were attached either using a plastic cable tie collar (N. alexis and P. hermannsburgensis) or by cyanoacrylate glue to fur between the scapulae (S. youngsoni); no R. villosissimus were radio-tracked. Tags weighed 0.4 – 1.0 g and were attached to animals that were randomly selected from those captured provided that their weight gain with the tag was <5%. Tags were supplied by Biotrack (Wareham, United Kingdom), Holohil (Ontario, Canada), or Titley Electronics (Ballina, New South Wales, Australia) and used frequencies between 150 and 151 MHz. Detailed protocols for tag fitting and animal release are provided by Dickman et al. (2010) and Haythornthwaite and Dickman (2006).
Animals were located at night every 1–4 h using a 3-element hand-held Yagi antenna and a Regal 1000 (Titley Electronics, Ballina, New South Wales, Australia) or TR-2 (Telonics Inc., Mesa, Arizona) receiver. If animals were in open areas, we approached them from downwind, walking on sand to reduce noise and using red light to establish visual contact to determine their location precisely. If animals were in covered areas or could not be approached closely, we estimated locations by triangulation of 2–3 bearings taken in the direction of the peak signal using a prismatic compass. Bearings were taken from known triangulation points ensuring that angles were >20° and <120° from each other (Kenward 2001). Pilot trials using tags placed at known locations indicated that bearings were accurate to within ± 5° up to distances of ~110 m (Dickman et al. 2010). By day, we located animals in burrows by walking along dune crests to pick up their signals and then walking along the line of peak signal strength. Burrows were pinpointed by removing the Yagi antenna and homing in on the signal, to within 1–2 m of the animal, using the coaxial cable and receiver. Animal locations and bearings were flagged using marker tape and later placed on fine-scale maps of the study area.
Data Processing and Statistical Analyses
To test our first prediction concerning recognition and avoidance by small mammals of V. gouldii odour, we compared GUDs between the three odour treatments (no odour – control; pungency control – eucalyptus; goanna faecal material – goanna) at all sites that were visited, using a linear mixed-effects model. Treatment and small mammal species were set as fixed effects and survey day as a random effect in a lmer model using the lme4 package (Bates et al. 2015) in R. Stepwise comparisons indicated that food type and GUD station had little effect on model results, and thus were not retained in the most parsimonious model. Model fit was assessed using residual diagnostics (Hartig 2021). Post hoc comparisons between the treatment groups and for each species between treatments were made on the final model using estimated marginal means via emmeans (Lenth 2021), which allows for the error variances that have been specified in the model, and Tukey P-value adjustments for comparing the families of three and twelve estimates.
To test our second prediction concerning the attraction of V. gouldii to small mammal odour, we counted the numbers of artificial burrows that elicited a response from V. gouldii. Responses were recorded as digging activity or as tracks within 30 cm of the artificial burrows that indicated investigative activity, and were analysed separately. Analyses were carried out separately for the species complements present in 2012 (i.e., with the inclusion of R. villosissimus – phase 1) and 2013–2014 (with S. youngsoni – phase 2). Survey session was considered a nominal factor that differed for each new location of the artificial burrows within one of the three main study locations. Survey day within each survey session was also considered a nominal factor. Generalised least squares (GLS) models were used to account for heterogeneity of variances, which were observed across the treatments, survey days, survey sessions, and locations. Following Zuur et al. (2009) we incorporated the nominal variables using the nlme package (Pinheiro et al. 2021) in R to implement GLS with a variance structure that allowed for differences in variance across survey days per survey session per location. We also incorporated an exponential variance structure for the covariance to account for the variance spread across different survey sessions alone or by survey day per survey session, depending on the data and best fit model. Log likelihood and Akaike information criterion comparisons were used to determine the most parsimonious models. The fixed explanatory variables retained in all models were treatment odour, survey day and survey session; location was not retained as it had no significant effect. The only other difference between the models was that survey day was not retained in the exponential variance structure for models of the track data for both species complements. Graphical validation of the optimal model was obtained by comparing box plots of the raw data against the model estimates, by plotting the residuals against the fitted values for the nominal explanatory variables, and via histograms and Q-Q plots of the model residuals (Zuur et al. 2009). Post hoc comparisons between the treatment groups were made on the final model using estimated marginal means via emmeans (Lenth 2021), which allows for the error variances that have been specified in the GLS model, and Tukey P-value adjustments for comparing a family of four estimates. All analyses were carried out in R (R core team 2021).
To analyse small mammal movements and burrow fidelity (prediction 3), we mapped the locations of animals that we had sighted and estimated the positions of triangulated animals using an iterative maximum likelihood estimator (LOAS, Ecological Software Solutions, Sacramento, California). Movements were calculated as the distance covered between successive signal locations, per unit time (m/h), with the distance (di) moved by an individual between its first signal location i(xi, yi) and the next (xi+1, yi+1) calculated using White and Garrott’s (1990) equation:
d i = √ (xi+1 – xi)2 + (yi+1 – yi)2
We assessed small mammal fidelity to burrows, (f), using the index: f = Nmax/N
where Nmax is the maximum number of visits by an individual to the same burrow over the period of radio-tracking, and N is the total number of its visits to all burrows. Index values thus range from f = 1 for an individual that uses a single burrow to f = 1/N for an individual that uses all burrows once. The index formula is the same as the Berger-Parker index of species dominance (Magurran 2004), and its use here follows Dickman et al. (2010). Only small numbers of the study species were radio-tracked on each of the three sampling occasions, so results were pooled and rates of movement and burrow fidelity compared for each species between the control and V. gouldii removal areas using analysis of variance. The burrow fidelity data for S. youngsoni were log-transformed to improve the variance structure prior to analysis; no other transformations were required.