Peter J. MCDONALDand Gary W. LUCK

1Flora and Fauna Division， Department of Land Resource Management， Northern Territory Government， Alice Springs，NT， Australia

2School of Environmental Sciences， Charles Sturt University， Albury， NSW， Australia

3Institude for Land， Water and Society， Charles Sturt University， Albury， NSW， Australia

Preliminary Insights into the Habitat Preferences of the Centralian Bandy Bandy (Vermicella vermiformis) (Squamata: Elapidae) in Central Australia

Peter J. MCDONALD1，2*and Gary W. LUCK3

1Flora and Fauna Division， Department of Land Resource Management， Northern Territory Government， Alice Springs，NT， Australia

2School of Environmental Sciences， Charles Sturt University， Albury， NSW， Australia

3Institude for Land， Water and Society， Charles Sturt University， Albury， NSW， Australia

Bandy Bandy's （Vermicella spp.） are a striking， black-and-white ringed genus of small elapid snakes endemic to Australia. All taxa are burrowers and little is known of their biology and ecology. We investigated the habitat preferences of the only arid-dwelling species， the centralian bandy bandy （Vermicella vermiformis）， in the MacDonnell Ranges west of Alice Springs in the Northern Territory. Using systematic road-cruising， we encountered 16 V. vermiformis over a 12 months period between 2009 and 2010. We used logistic regression to model the occurrence of the species against a range of different habitat variables collected at multiple scales. Despite the small sample size， V. vermiformis exhibited a clear preference for acacia shrubland habitats， with acacia variables present in all AICcranked models in the 95% conf i dence set. The factors driving this association， together with the preference for habitat not burnt in the most recent wildf i res， may be related to the abundance of their only known prey， blind snakes （Ramphotyphlops spp.）.

Snake， Arid， Acacia， Fire， Ramphotyphlops

1. Introduction

Bandy Bandy's （Vermicella spp.） are a black-and-white ringed genus of small （total length 362-760 mm） elapid（Elapidae） snakes endemic to the eastern， central and northern regions of the Australian mainland and nearby islands （Wilson and Swan， 2013）. All taxa are burrowers and are generally only observed above ground at night and often after rainfall （Wilson and Swan， 2013）. Little is known of their biology and ecology， though females are generally larger than males， clutch sizes range from 1-13 and are smaller in smaller species， and there is evidence of dietary preference for blind snakes （Ramphotyphlops spp.） （Keogh and Smith， 1996）. Of the f i ve known taxa， the centralian bandy bandy （Vermicella vermiformis） is the only truly arid-dwelling species and occurs in two apparently disjunct populations in the Northern Territory，including a population in the MacDonnell Ranges bioregion surrounding the town of Alice Springs （Wilson and Swan， 2013）. Here， we report on a field study of the habitat preferences of the MacDonnell Ranges population of V. vermiformis， carried out west of Alice Springs （Figure 1）. We sampled snakes using systematic road-cruising and， using logistic regression， modelled the occurrence of the species against a range of habitat variables recorded around each presence and absence location.

2. Materials and Methods

Bandy Bandy's were sampled by systematic roadcruising as part of a broader study into the ecology ofarid-zone snakes in the MacDonnell Ranges bioregion（for more detail and a map of the study area， see McDonald）. We drove along a 77 km sealed road transect which included part of Namatjira Drive （23°48′33″ S，133°13′16″ E to 23°40′1″ S， 132°38′8″ E） and all of the Ormiston Gorge access road （23°41′5″ S， 132°42′34″ E to 23°37′57″ S， 132°43′39″ E） west of the town Alice Springs. This transect runs through a range of habitat types typical of the bioregion and was driven on 77 nights over a 12 months period between August 2009 and July 2010. Each night the transect was driven end-to-end twice at 40-60 km/hr， with the start point alternated between the east and west ends. The location of all V. vermiformis individuals encountered was marked with a hand held GPS. All live animals were caught， individually marked by scale clipping and released on the road verge adjacent to the point of capture. Road-kill animals were removed from the road surface. All snake handling and processing procedures were approved by the Charles Sturt University Animal Care and Ethics Committee （approval number 09/064） and the Department of Natural Resources，Environment， the Arts and Sport/NT Parks and Wildlife Service （research permit number 35656）.

Figure 1 An adult female centralian bandy bandy （Vermicella vermiformis）.

In order to test for any habitat preferences， we returned to each V. vermiformis encounter location during the day to record a range of habitat variables. We recorded the percentage of each of the three dominant vegetation types（% hummock grassland， HUMCK_50； acacia shrubland，ACAC_50， % riparian woodland/grassland， RIP_50），as well as several groundcover variables （% acacia litter cover， ACAC_LIT； eucalyptus litter， EUC_LIT； % hummock grass cover， HUMCK_COVER， % tussock grass cover， TUSSK_COVER； % rock cover ROCK_ COVER； % rock outcrop cover， OUTCRP_COVER； no. logs ＞ 20 cm diameter， LOGS）， within a 50 m radius on both side sides of the road verge surrounding each snake encounter locality， using visual estimation （Table 1）. In addition， we recorded the percentage of each of the three dominant vegetation types within a 500 m radius in order to test for larger-scale effects of the surrounding vegetation types on V. vermiformis occurrence （Table 1）. The 500 m scale vegetation data were sourced from a 1:25 000 GIS layer of vegetation communities produced by the Northern Territory Government （McDonald et al.，2012）.

Because wildf i re is considered an important landscape process potentially influencing snake occurrence in arid Australia （McDonald et al.， 2012）， we also recorded the percentage of land area burnt in the most recent wildf i res （2002） at both the 50 m and 500 m scales （BURNT_50 and BURNT_500， respectively） （Table 1）. These data were collected by visual estimation at the 50 m scale andfrom fire-scar mapping derived from Landsat imagery at the 500 m scale （Table 1； McDonald et al.， 2012）. Because binary logistic regression requires both presence and absence data， all variables were also recorded at 50 randomly selected ‘pseudo-absence' locations along the road transect， each a minimum distance of 500 m from the nearest presence location.

Table 1 Comparison of habitat variables at sites where Vermicella vermiformis were present and absent.

The difference in environmental values between the presence and absence sites was initially examined with the mean values and tested with the non-parametric Mann Whitney U-test （Table 1）. We tested for collinearity among the independent （predictor） variables using the Spearman correlation coeff i cient. Where pairs of variables exhibited a correlation coeff i cient of greater than 0.6， one of the pair was removed from further analysis. This was carried out with consideration of biological meaning （e.g. MULGA_50 and MULGLIT_COVER were correlated and we retained the former as that variable encompasses more habitat characteristics）. All retained variables were modelled using binary logistic regression， with presence（1） or absence （0） as the dependent variable. We also modelled combinations of independent variables where these combinations made biological sense （e.g.，HUMCK_50 + HUMCK_COVER）. Models were ranked using Akaike's Information Criterion corrected for small sample size （AICc）. This is appropriate when the number of data points/maximum number of fitted parameters is less than 40 （Symonds and Moussalli， 2011）. Models with smaller values of AICchave greater support as explanations for variation in the response variable relative to other models in the set considered. We assessed the relative strength of models subsequent to the best fi tting model by comparing the difference in criterion values of the best ranked model （smallestvalue；） with model i （i） （Symonds and Moussalli， 2011）. The best ranked modelhas aivalue of 0 and subsequent models are scored asi=-， where AICciis the AICcvalue of the modelcompared with the best ranked model. Models withivalues less than 2 are considered to be essentially as good asthe best model in explaining variation in the response，ivaluesup to 6 should be considered plausible explanations，ivalues between 7-10 may be rejected， andivalues greater than 10 should be considered implausible and rejected （Symonds and Moussalli， 2011）.

We also calculated Akaike weights （wi） for each model，which can be interpreted as the probability of a model being the best model of those considered. We present the 95% confidence set of models， where the summed wiequals a minimum of 0.95. To test the overall goodness offi t of each model， we applied the Hosmer and Lemeshow statistic. Signi fi cant P-values （P ＜ 0.05） from this test are evidence of lack of fi t. We also included the classi fi cation table outputs for % accuracy of each model in predicting presenceand absence. AICcvalues， difference in criterion values （i）， Akaike weights and diagnostic measures were calculated manually. All other analyses were run in SPSS（PASW Statistics 17.0）.

3. Results

Over the 12 months of sampling， we located 16 V. vermiformis on the road transect. Of these， 15 were live and 1 was road-kill， 7 were female （mean SVL = 520 ± 36 mm）， 5 were male （mean SVL = 486 ± 24 mm） and the sex of a further 4 was not recorded （due to them being too agitated to safely probe）. The best AICcranked model included the area of acacia shrubland within a 50 m radius （ACACA_50） and the area of recently burnt vegetation （BURNT_50）， also within a 50 m radius（Table 2）. The probability of occurrence of V. vermiformis increased with increasing area of acacia shrubland and decreased with increasing area of burnt vegetation. The acacia variable was present in two other models and the fi re variable in the next best ranked model within the 95% conf i dence set. The combined wiof models that included the ACAC_50 variable was 0.97 which can be interpreted as there being a 97% probability of this variable being included in the best model （Symonds and Moussalli，2011）. The only other predictor variable within the 95% conf i dence set was area of acacia shrubland within a 500 m radius （ACACA_500） and this variable was present in the second and third ranked models for a combined wiof 0.47 （Table 2）. There was no evidence of a lack of f i t in the three models， though the ability of the models in predicting the presence of V. vermiformis was relatively low （Table 2）.

4. Discussion

Although relatively few V. vermiformis were encountered during the study， it is unlikely that any other sampling method would have resulted in more snake encounters. Funnel trapping has been used with some success targeting snakes in Australia and elsewhere （Thompson and Thompson， 2007）， but extensive funnel trapping throughout our study region has not yet resulted in a V. vermiformis capture （P. McDonald， unpubl. data）. V. vermiformis were frequently observed moving on the road surface， rather than lying motionless， and it is likely that individuals encountered were simply moving from one location to another， rather than actively seeking out the warm road surface for thermoregulation. Despite the small sample size， the best models predicting the occurrence of this species appeared to be reasonably robust. Although V. vermiformis were located in other habitats （e.g. riparian woodland/grassland）， they appear to exhibit a preference for acacia shrubland in the study area， particularly at the finer 50 m scale. A preference for acacia shrubland has also been demonstrated for the sympatric small elapid Parasuta monachus and it was suggested that this preference may be related to the availability of litter-dwelling skinks， an important food resource for that species （McDonald et al.， 2012）.

Understanding the factors driving the acacia association for V. vermiformis was beyond the scope of our study；however， it is possible that the species occurs more frequently in acacia habitats because they support higher densities of its known food， blindsnakes （Ramphotyphlops spp.）. Blindsnakes are generally encountered even lessfrequently above ground than V. vermiformis （McDonald，2012）， making this a difficult hypothesis to test. The apparent preference for areas not recently burnt by wildfire was more puzzling as it could reasonably be expected that a species which， along with its prey， spends most of its time below ground would be resistant to the effects of temporary groundcover loss associated with wildf i re. One possibility is that ants， the larvae and pupae of which are the primary food source of blindsnakes（Webb and Shine， 1993）， are negatively affected by fire and that this may result in a negative flow-on effect for bandy bandy populations. Although no studies have investigated the fire ecology of ants in Australia's arid zone， a study from tropical Australia showed that fire regime had a profound effect on terrestrial ant community composition （Anderson， 1991）. Because blindsnakes have been shown to specialise on particular ant groups （Webb and Shine， 1993）， it is plausible that blindsnakes reach greater abundance in areas with increased time-since-f i re due to these areas supporting more abundant or diverse ant food sources. Further research will provide more insight into the robustness of the relationship with fire history and the factors driving the habitat association we observed.

Table 2 AICcranked models explaining the occurrence of Vermicella vermiformis in the study area.

Acknowledgements This project was partially funded by an operating grant through Charles Sturt University. We thank the people who volunteered their time to assist with the snake sampling， including Kelly KNIGHTS，Katherine WILLIAMS， Gareth CATT， Greg FYFE， Paul GARDNER， Simon RATHBONE and Peter NUNN.

Anderson A. N. 1991. Responses of ground-foraging ant communities to three experimental fire regimes in a savannah forest of tropical Australia. Biotropica， 23: 575-585

Keogh J. S., Smith S. A. 1996. Taxonomy and natural history of the Australian bandy-bandy snakes （Elapidae: Vermicella） with a description of two new species. J Zool， 240: 677-701

McDonald P. J. 2012. Snakes on roads: an arid Australian perspective. J Arid Environ， 79: 116-119

McDonald P. J., Luck G. W., Pavey C. R., Wassens S. 2012. Importance of f i re in inf l uencing the occurrence of snakes in and upland region of arid Australia. Austral Ecol， 37: 855-864

Symonds M. R. E., Moussalli A. 2011. A brief guide to model selection， multimodel inference and model averaging in behavioural ecology using Akaike's information criterion. Behav Ecol Sociobiol， 65: 13-21

Thompson G. G., Thompson S. A. 2007. Usefulness of funnel traps in catching small reptiles and mammals， with comments on the effectiveness of the alternatives. Wildlife Res， 34: 491-497

Webb J. K., and Shine, R. 1993. Dietary habitats of Australian Blindsnakes （Typhlopidae）. Copeia， 1993: 762-770

Wilson, S., Swan, G. 2013. A complete guide to reptiles of Australia. Sydney: New Holland Publishers

10.3724/SP.J.1245.2014.00049

*Corresponding author: Peter McDonald， from Department of Land Resource Management， Northern Territory Government， Alice Springs，Australia， with his research focusing on landscape ecology and conservation biology in Australia.

E-mail: peterj.mcdonald@nt.gov.au

25 October 2013 Accepted: 20 February 2014

Asian Herpetological Research 2014， 5（1）: 49-53