The significance of topographic complexity in habitatselection and persistence of a declining marsupial inthe Kimberley region of Western Australia

RosemaryHohnenABE Katherine TuftAB Sarah LeggeBC NaomiWaltersB Lucy JohansonBScott CarverA Ian J RadfordD and Christopher N JohnsonA

ASchool of Biological Sciences University of Tasmania Private Bag 55 Hobart Tas 7001 AustraliaBMornington Wildlife Sanctuary Australian Wildlife Conservancy PMB 925 Derby WA 6728 AustraliaCPresent address National Environmental Science Program Threatened Species Recovery HubCentre for Biodiversity Conservation University of Queensland St Lucia Qld 4072 Australia

DDepartment of Parks and Wildlife PO Box 942 Kununurra WA 6743 AustraliaECorresponding author Email rhohnenutaseduau

Abstract Mammalian species in northernAustralia are declining The resources thatmany species from this region requireto persist in the landscape remain poorly understood We examined habitat selection and diet of the scaly-tailed possum(Wyulda squamicaudata hereafter calledWyulda) in the north-westKimberleyWesternAustralia in relation to variation incomplexity of rocky habitat habitat heterogeneity and recent fire historyWefittedGPS tags to 23Wyulda between January2013 and February 2014 and analysed step selection between GPS fixes to describe habitat choice We assessed diet bymicroscopic analysis of plant fragments from 47 faecal samples Individual Wyulda preferentially foraged in locations withhigh rock complexity and high habitat heterogeneity in a wide variety of habitats but denned exclusively in complex rockpiles They used savannas of a range of post-fire ages including recently burnt (1ndash2 months after fire) and long unburnt(gt24months after fire) They were highly frugivorous with on average 77 of plant fragments per scat sample identified asfruit epidermal layers Overall rock complexity appears to be an important landscape attribute forWyulda as it may provideden sites and protect fire-sensitive landscape features such as fruiting trees and habitat heterogeneity

Additional keywords Bayesian hierarchical logistic regressionmodels diet discrete choicemodels fire regime northernmammal decline small mammal vegetation structure Wyulda squamicaudata

Received 2 March 2016 accepted 22 August 2016 published online 20 September 2016

Introduction

Populations of many mammal species across northern Australiaare currently declining (Corbett et al 2003 Andersen et al 2005Woinarski et al 2011 2014) Several contributing threats havebeen identified including the impacts of introduced herbivores(Legge et al 2011a) feral cats (Felis catus) and contemporaryfirepatterns (Woinarski et al 2011) Despite these declines little isknown regarding the resources many of these species require topersist in the landscape

Loss of habitat heterogeneity and of fire-sensitive vegetationresources resulting from increased fire intensity and frequencyhas been proposed as an important driver of mammal decline(Firth et al 2010 Hohnen et al 2015 Leahy et al 2016) Inrecent years dominant fire patterns in northern Australia haveshifted from those prevalent during indigenous occupation(Vigilante 2001) to dominance by large intense fires that burnprimarily in the mid-to-late dry season and recur every 1ndash3 years(Andersen et al 1998 Vigilante 2001 Fisher et al 2003 Legge

et al 2011b) Such patterns can change vegetation structure andcomposition decreasing vegetation heterogeneity (Russell-Smith et al 2012) causing mortality of fruiting and hollow-bearing trees (Russell-Smith et al 1998 Williams et al 1999Vigilante and Bowman 2004b) and driving declines in fire-sensitive vegetation such as rainforest habitats (Russell-Smithand Bowman 1992) In the short term intense fires can removeground cover vegetation used as a shelter and food resource bysmall mammals

The identification of areas or resources that provide refugein fire-prone landscapes can inform conservation managementstrategies that aim to conserve biodiversity (Banks et al 2011Pereoglou et al 2011) In northernAustralia there is evidence thattopographically complex areas may be especially important Forexample several mammal species such as the monjon (Petrogaleburbidgei) golden-backed tree-rat (Mesembriomys macrurus)brush-tailed rabbit-rat (Conilurus penicillatus) and Kimberleyrock-rat (Zyzomys woodwardi) that have declined substantially

Journal compilation CSIRO 2016 wwwpublishcsiroaujournalsajz

CSIRO PUBLISHING

Australian Journal of Zoologyhttpdxdoiorg101071ZO16015

now persist only in the topographically complex north-westKimberley having retracted from previously wider distributions(Sawle 1988 Start et al 2007 2012Woinarski et al 2014)Morebroadly across Australia mammal species that use rock featuresand habitats as foraging or shelter resources have lower thanexpected rates of extinction (Burbidge and McKenzie 1989Smith and Quin 1996) However few studies have examinedhow such species use these complex landscapes in relation toadjacent open areas particularly after they have been affectedby fire (Driscoll et al 2010)

Wyulda squamicaudata (hereafter calledWyulda) is a possumspecies endemic to the Kimberley region of Western AustraliaThe species is largely restricted to the rugged north-westKimberley region and has disappeared from the central and eastKimberley with the exception of Emma Gorge in the CockburnRanges where it was recently rediscovered (Doody et al 2012Woinarski et al 2014) Limited studies have identified thatWyulda den within rock piles during the day (Runcie 1999) anduse rocky savanna and vine thicket habitats for foraging(Humphreys et al 1984 Bradley et al 1987 Burbidge andWebb2008)

In this paper we examine how Wyulda use rocky and openhabitats at sites partially burnt by a severe and extensive firetypical of contemporary fire patterns We aimed to (1) test forpreferences between topographically complex and open savannahabitats (2) examine theeffect offire historyon theuseof savannahabitats and (3) identify key foraging and denning resourcesFrom this information we suggest habitat characteristics that willsupport the persistence of Wyulda in the landscape

Methods

Study areas

This study was conducted between January 2013 and February2014 in the Artesian Range Wildlife Sanctuary ~180 km north-west of Derby in the Kimberley region of Western Australia(Fig 1) This 172 820-ha sanctuary is managed by the AustralianWildlife Conservancy It is characterised by areas of opensavanna eucalyptspinifex woodlands plus topographicallycomplex areas dissected by sandstone gorges that often supportsmall rainforest pockets Average annual rainfall is 1200mm94 falling in the NovemberndashApril wet season (Bureau ofMeteorology 2015)

Weworked at three sites (Fig 2) Site 1 on the Charnley Rivergorge ~16 km upstream from its entrance to the Walcott InletSite 2 on theCharnleyRiver gorge 6 kmdownstream fromSite 1and Site 3 20 km south-west of Site 2 on the southern edge ofthe Artesian Range All sites consisted of a narrow rainforestgully bordered by steep cliffs and surrounded by various typesof savanna

We identified eight habitat types differing in floristics andrecent fire history boulder scree rainforest river edge and fivesavanna types (described below in Fig 2 Fig 3 andAppendix 1)Rainforest and river edge are all topographically protected fromfire Savannas were divided into long-unburnt (burnt gt24monthspreviously) and burnt within the preceding 12 months Long-unburnt savanna had a ground layer dominated by spinifex(Triodia spp) Recently burnt savanna habitats were morefloristically variable and were classed into four types based on

Isdell River

Charnely Rive

r

Calde

r Rive

r

Walcott Inlet

Area enlarged

N

0 125 25 Kilometres

River

Artesian Range Sanctuary

Site 1

Site 2

Site 3

Fig 1 The location of the three sites in the Artesian Range north-west Kimberley Western Australia

B Australian Journal of Zoology R Hohnen et al

understorey plants and substrate (1) Hibiscus keneallyi savannaon skeletal sandy soils over a sandstone substrate (2) Sorghumstipoideum savanna on shallow sandy soils over sandstone

(3) Chrysopogon fallax savanna on shallow sandy soils oversandstone substrate and (4) Heteropogon contortus savannaon skeletal soils over basalt We assessed floristic differences

0 01 02 04 KmN

Site 1 Site 2

Site 3Habitat type

Boulder scree

Rainforest

River edge

Triodia spp savanna

Hibiscus keneallyi savanna

Sorghum stipoideum savanna

Chrysopogon fallax savanna

Heteropogon contortus savanna

River

Inaccessible

Fig 2 Habitat composition of sites in the Artesian Range north-west Kimberley Western Australia

(a) (b)

Fig 3 Photos of (a) an individual male Wyulda squamicaudata and (b) typical rugged terrain in which this species is found

Habitat use by Wyulda squamicaudata Australian Journal of Zoology C

between habitat types by analysing vegetation survey data byordination as in Hohnen et al (2015) The availability of thesevegetation types varied between sites (Table 1)

Trapping

We captured Wyulda in treadle-operated wire cage traps(L 30 cm H 30 cm W 70 cm Sheffield Wire ProductsWelshpool Western Australia) At each site 30 traps were set10ndash30m apart and baited with a mixture of peanut butteroats honey and apple Wyulda were monitored at Site 1 in fourseparate months twice in the wet season (January 2013December 2013) and twice in the dry season (April 2013 June2013) Sites 2 and 3 were each visited once in the dry season(May and August 2013 respectively) and once in the wet season(February 2013 and January 2014 respectively) for a total of4470 trap-nights

On first capture each animal was weighed sexed and taggedwith a nanotransponder microchip (Trovan United Kingdom)Faecal pellets were collected from traps for diet analysisIndividuals were considered adult if theyweighed gt1000 g or forfemales if their teats showed evidence of current or previouslactation Trapped individuals gt1065 g were fitted with GPSVHF composite collars (Telemetry Solutions Concord CAUSA) Collars weighed 32 g constituting lt3 of the animalrsquosbody weight Each unit consisted of a GPS logger and a VHFtransmitter attached to a strong lightweight material collarDuring January and February 2013 the GPS collars were not yetavailable so VHF collars were used These collars weighed 30 g(SirtrackHavelockNorthNewZealand)lt3of an individualrsquosbody weight and were fastened around the neck with a leatherstrap

The GPS collars logged foraging locations hourly between1900 and 0500 hours for the maximum battery life of the collars(three weeks) Radio-collared Wyulda were located betweenone and four times each night bymeans of a radio receiver (TitleyScientific Brisbane Australia) and a Yagi antenna (SirtrackHavelock North New Zealand) using the homing methodoutlined in White and Garrot (1990) Individuals were located aminimum of 1 h after a previous fix to minimise autocorrelationAs Wyulda were thought to den underground where there is noGPS reception den sites were located daily using the VHF signalfor both collar types

Data analysis

Diet analysis

A reference collection of the epidermal layers of 161 plantspecies (including 161 leaf 6 flower and 18 fruit slides) weremade from plant specimens collected at the three sites Referenceslides were made by soaking fragments of plant specimen in42 g Lndash1 sodium hypochlorite for 24 h or until 90of thematerialwas visibly bleached following themethodsoutlined inTuft et al(2011) The material was then rinsed stained with 02 g Lndash1

gentian violet and mounted onto a slide using corn syrup Of the108 scat samples collected 47 were selected for analysis tomaximise the representation of the sexes (20 male 28 female)sites (25 Site 1 11 Site 2 12 Site 3) and seasons (24 wet season24 dry season) To prepare scat samples for analysis sampleswere broken up loosely and also soaked in 42 g Lndash1 sodiumhypochlorite for 24 h or until 90 of the material was visiblybleached The material was rinsed in water through a 250-mmsieve whole seeds were removed identified and countedRemaining leafmaterialwas stained using02 g Lndash1 gentian violetandmounted on a slide using corn syrup For each scat specimen100 identifiable fragments of plant material were counted intransects across the slides Unidentifiable fragments (where theepidermal layer was not visible) were uncommon and were notincluded in counts

Movement analysis

GPS points were first screened to remove erroneous fixescaused byGPS error These included location points found on theopposite side of the impassable Charnley River (at Sites 1 and 2see Fig 2) or that displayed all of the following characteristicsdistance from the last fix gt300m and turning angle (in relation tothe previous and the followingfix)gt180 (Bjoslashrneraas et al 2010)The combination of these characteristics indicated implausiblespikes in which an individual appeared to deviate at high speedfrom his or her trajectory to a distant location and then return tothe original trajectory Horizontal dilution of precision (HDOP)values were not used to filter fixes as data from two stationary testGPS collars indicated that there was no relationship betweendistance to known location and HDOP value

Home ranges were estimated by minimum convex polygons(MCP) and fixed kernel density estimation methods in theprogram Geospatial Modelling Environment (Beyer 2012) and

Table 1 Proportions of habitat types available at three sites in the Artesian Range north-west Kimberley Western AustraliaAn asterisk indicates that last fire is unknown but gt24 months preceding commencement of study period

Habitat type Last fire Age Site 1 Site 2 Site 3(months) Area

(ha)Proportion

()Area(ha)

Proportion()

Area(ha)

Proportion()

Boulder scree 60+ 669 383 359 203 000 000Rainforest 60+ 960 549 361 204 938 505River edge 60+ 667 382 558 315 000 000Triodia spp savanna 24+ 617 353 000 000 615 331Hibiscus keneallyi savanna Nov 2012 1ndash12 14555 8333 16448 9078 1081 582Sorghum stipoideum savanna Oct 2013 3 000 000 000 000 8923 4804Chrysopogon fallax savanna Oct 2013 3 000 000 000 000 1137 612Heteropogon contortus savanna Oct 2013 3 000 000 000 000 5882 3166

D Australian Journal of Zoology R Hohnen et al

interpreted in ArcMap 101 (ESRI) Kernel density estimateswere calculated with a plug-in bandwidth The 95 kernel (K95)utilisation contour estimated total home range and the 50(K50) kernel utilisation contour estimated the core home rangeAll points (both foraging and denning) were included in theMCPand kernel estimates and all areas that the animals could not use(including the river and land on the opposite side of the river)were removed from further consideration Home-range estimateswere tested for normality using the ShapirondashWilk W test andnormalised by log-transformation Differences in home-rangesize between sex and season were examined using generalisedlinear models with a model-averaging approach using thepackage lme4 and MuMIn in the statistical program R ver 303(R Development Core Team 2005 Barton 2013 Bates et al2014)

We calculated the total percentage home-range overlap ofK95 and core K50 home-range estimates for all neighbouringindividuals Home-range overlap was also compared using PHR(Percentile home range) and UDOI (Utilisation distributionoverlap index) indices which consider heterogeneity of spaceuse within home ranges (Fieberg and Kochanny 2005) (seeAppendix 2 for more detail) Differences in overlap extent werecompared between dyads using the asymptotic KruskallndashWallistest as datasets did not meet assumptions of normality

We used a discrete choice modelling approach to describestep selection of Wyulda during foraging periods Coefficientswere estimated from case-control logistic regression modelsusing the lsquoclogitrsquo command in the package survival (Therneau2014) within the statistical program R ver 303 This regressionmethod accounts for differing availabilities of habitats betweenobservations by matching every known animal location to achoice set of randomly generated locations To generate randompoints we first calculated the average distance travelled betweenfixes taken at 1-h 2-h or 3-h (and so on) intervals Then forexample if between fixA andB there was a gap of 2 h five pointswere randomly generated within a circle (with A at its centroid)that had a radius consisting of the average distance travelled byWyulda in 2 h clipped by the individualrsquos 95 kernel home-range estimate Those five points represent locations where theanimal could have gone and were compared in the modellingprocess to point B the location where the animal did go

For all points where Wyulda were found we recorded habitattype and rock complexity (Appendix 3) The variable lsquorockcomplexityrsquo comprised four categories cliffs high complexity(boulder size gt2m) medium complexity (boulder size lt2m) andlow complexity (no exposed rock) Habitat and rockiness mapsfor each site were created in ArcMap 101 using geo-referencedsatellite images from Google Earth (Google Inc 2009) Othervariables investigatedwere distance topermanentwater (referredto as lsquodistance to waterrsquo) number of habitat types within a20-m radius (lsquohabitat heterogeneityrsquo) distance to unburnt habitat(lsquodistance to unburnt patchrsquo) fire (lsquotime since firersquo) andthe distance from the point to the centre of the home range(lsquodistance to home range centroidrsquo) We chose 20m as the radiusof the circle within which to calculate habitat heterogeneity aswe aimed to take into accountGPS error aswell as an individualrsquosperception of where it is relative to the habitats around itWe tested for correlation and redundancy between explanatoryvariables by identifying correlation coefficients of gt05

(Fox 2002) The variables lsquofirersquo lsquodistance towaterrsquo and lsquodistanceto unburnt patchrsquo were strongly correlated with one another solsquodistance to waterrsquo and lsquodistance to unburnt patchrsquo were omittedfrom further analysis Numerical variables including habitatheterogeneity time sincefire anddistance tohome rangecentroidwere standardised (mean = 0 standard deviation = 1) beforeanalysis

We created 31 candidatemodels based on combinations of thevariables lsquohabitat typersquo lsquotime since firersquo lsquohabitat heterogeneityrsquolsquorock complexityrsquo and lsquodistance to home range centroidrsquo withlsquoindividualrsquo considered as a random effect Candidate modelswere ranked on the basis of the Akaikersquos information criterionadjusted for small sample sizes (AICc) The selection forlandscape characteristics by Wyulda was measured by thecoefficient estimates and the odds ratio which representsthe relative change in the odds of selection for each unit of thepredictor variableWe usedmodel averaging to estimate variablecoefficients and their weights according to Anderson (2008)

Den site selection

To explore den site selection by Wyulda we used Bayesianhierarchical logistic regression models For each individual wecompared known den sites with an equal number of randomlygenerated den sites within that individualrsquos home range Modelscomprised a binomial response variable (that described if a pointwas real (1) or generated (0)) and all combinations of theaforementioned predictor variables including habitat type rockcomplexity time since fire habitat heterogeneity and distance tohome range centroid To account for variation in individual useof habitats lsquoindividualrsquo was included as a random effect in allmodels (See Appendix 4 for more details) A Markov ChainMonte Carlo (MCMC) procedure was used to estimate posteriordistributions using the package MCMCglmm (Hadfield 2010) inR ver 303 MCMC chains were run with randomised startingvalues aburn-inof 50 000 iterations a thinning interval of 50 anda total of 200 000 iterations Convergence of MCMC chains wasevaluated following Gelman and Hill (2006) The DevianceInformation Criterion (DIC) was used to compare all possiblemodels and calculate model weightsWe compared all subsets ofthe most complex model We used model averaging to estimatevariable coefficients posterior credible intervals and variableweights according to Anderson (2008)

Results

Between January 2013 and January 2014 collars were deployedfor one month each on 34 occasions on 23Wyulda (15 males and19 females) Of those deployments 29 were GPS collars and5 were VHF giving a total of 3314 location points Nineindividuals were tracked on more than one occasion and theirhome ranges were calculated separately for each tracking period(Appendix 5) The number of individuals known to be alive ateach site varied between 5 and 14 (Appendix 6)

Diet

From the scat samples 4700 fragments of plant material wereidentified Fragments of leaf came from 42 species fruit from 14species and flower from one species (Appendix 7) The mostcommon items were Ficus platypoda fruit (on average 47 of

Habitat use by Wyulda squamicaudata Australian Journal of Zoology E

fragments per sample) Owenia vernicosa fruit (10) Vitexacuminata fruit (10) Grewia breviflora fruit (8)V acuminata leaf (3) Strychnos lucida fruit (2)O vernicosaleaf (2) Cryptocarya cunninghamii leaf (2) and Acaciadimorpha leaf (1) Whole seeds were also frequently found inscats the most common being F platypoda (on average 55 ofseeds per sample) followed by V acuminata (15) UnknownSpecies 3 (6)UnknownSpecies 2 (4) andG breviflora (4)F brachypoda Unknown Species 1 Trachymene dendrothrixand Cryptocarya cunninghamii seeds were also present in lownumbers Plant species that took up on average a largerpercentage composition per sample were also likely to be foundin a greater number of samples (Appendix 7)

Home range

Estimated home-range sizewas not correlatedwith the number offixes per individual (Appendix 8) Mean home-range size was817 134 ha (MCP) or 802 129 ha (K95) and mean corehome-range size (K50) was 214 038 ha (Table 2) Meanhome-range size formaleswas 932 210 (MCP) or 981 233(K95) and for females was 725 176 (MCP) or 661 137(K95) In the sets ofmodels describing both variation inMCPandK95 home-range estimates with sex and season multiple modelswere competitive with one another (Appendix 9) For the MCPmodels the confidence intervals of the model-averagedcoefficient estimates overlapped zero for the variable lsquosexrsquo butdid not overlap zero for the variable lsquoseasonrsquo The coefficientestimate of the variable lsquodry seasonrsquowas positive For the K95models both thevariables lsquosexrsquo and lsquoseasonrsquohadmodel-averagedcoefficient estimates and confidence intervals that overlappedzero (Appendix 10)

Home-range overlap

Home-range overlap occurred frequently between neighbouringWyulda (Table 3) Extent of total home-range overlap differedbetween dyad types for the static K95 contour PHR andUDOIindices of total home-range overlap (Table 4) Total home-rangeoverlapoccurredmost frequentlybetween individuals of differentsexes (Appendix 11) and least frequently between neighbouringmales (Table 3) The extent of core home-range overlap alsodiffered between sexdyads andwasmost commonbetweenmalesand females (Fig 4 Appendix 11)

Foraging site selection

Two models (model weights of 30 and 16) were withintwo DAIC of the top model (model weight of 41 Table 5)The variables lsquohabitat typersquo lsquorock complexityrsquo lsquohabitatheterogeneityrsquo and lsquodistance to home range centroidrsquo all had highvariable importance weights of gt99 and the variable timesince fire had a low importance weight of 27 Model-averagedcoefficient estimates suggested that Wyulda selected againstboth lowandmedium levels of rockycomplexity (Fig 5Table 6)Weak but positive selection was detected for locations at greaterdistances from the centre of an individualrsquos home range Wyuldaselected areas of high habitat heterogeneity while foraging andused long unburnt habitats such as rainforest and savannadominated by Triodia spp as well as recently burnt savanna(1ndash12 months after fire) dominated by Hibiscus kenealleyi andSorghum stipodeum in preference to recently burntHeteropogoncontortusndashdominated savanna There was no selection for oragainst river-edge habitat types or in relation to the variable lsquotimesince firersquo

Den site selection

Individuals were tracked to a total of 736 den sites all of whichwere in rock features 548 in piles of rocky scree 108 in cracks incliffs 57 in vertical clefts in large rocky slabs and 23 in massiverocky outcrops Of the models built on the den site data only fivewere competitive (within 2 DDIC) with the top model (Table 7)Importance weights were highest for the variables lsquohabitat typersquoand lsquorock complexityrsquo (92 and 83 respectively) followed bylsquodistance to home range centroidrsquo (68) lsquohabitat heterogeneityrsquo(59) and lsquotime since firersquo (36) Model-averaged coefficientestimates suggest that Wyulda selected against denning intopographically simple habitats and at long distances from theirhome range centroids (Fig 6)Wyulda selected to den inHibiscuskeneallyi and Sorghum stipodeumndashdominated savanna habitatsboth 1ndash12 months after fire and against denning in long-unburnt

Table 2 The mean home-range size ofW squamicaudata across sexessites and seasons

MCPplusmn se(ha)

K95plusmn se(ha)

K50plusmn se(ha)

Female (n= 19) 725 plusmn 176 661 plusmn 137 194 plusmn 046Male (n= 15) 932 plusmn 210 981 plusmn 233 239 plusmn 065Dry season (n= 21) 1018 plusmn 199 955 plusmn 196 213 plusmn 050Wet season (n= 13) 492 plusmn 092 555 plusmn 090 214 plusmn 061Site 1 (n= 13) 725 plusmn 107 639 plusmn 059 125 plusmn 012Site 2 (n= 8) 1433 plusmn 491 139 plusmn 483 461 plusmn 123Site 3 (n= 13) 528 plusmn 068 603 plusmn 094 150 plusmn 026GPS (n= 29) 874 plusmn 155 857 plusmn 148 191 plusmn 038VHF (n= 5) 486 plusmn 095 486 plusmn 100 347 plusmn 145Total (n= 34) 817 plusmn 134 802 plusmn 129 214 plusmn 038

Table3 Frequencyandaveragehome-rangeoverlapbetweenadjacentWsquamicaudatamonitoredconcurrently at three sites in theArtesianRangenorth-west Kimberley Western Australia (mean se) using two-dimensional (K95and K50) and three-dimensional (PRH and UDOI) indices

Tnd=Total no of neighbouring dyads Tndo =Total no of neighbouring dyads that overlapped F female M male

Sex Tnd Tndo Tndo Mean total overlap Mean core overlapdyad (K95) (K50) K95 PRH UDOI K50 PRH UDOI

FF 24 22 6 2019 plusmn 456 027 plusmn 006 011 plusmn 004 437 plusmn 239 006 plusmn 003 005 plusmn 002FM 37 34 16 3700 plusmn 604 048 plusmn 007 033 plusmn 007 2068 plusmn 545 021 plusmn 005 013 plusmn 004MF 37 34 16 2424 plusmn 405 037 plusmn 006 033 plusmn 007 1199 plusmn 319 011 plusmn 002 007 plusmn 002MM 26 22 6 1210 plusmn 302 020 plusmn 004 004 plusmn 001 071 plusmn 032 002 plusmn 001 001 plusmn 000

F Australian Journal of Zoology R Hohnen et al

rainforest and river-edge habitats Credible intervals of thevariable lsquoTriodia spp savannarsquo included zero suggesting neutralselection Wyulda also appeared to select for older post-fire agesof all habitat types but this variable had a low importanceweight

Discussion

This study sought to test utilisation of complex landscapes andsurrounding areas by a declining mammal in the presence ofintensified fire regimes Rock complexity appears to be a keyhabitat feature selected for by Wyulda Individuals avoidedforaging in habitats with low rock complexity and consistently

denned within rock structures Wyulda preferred to forage inareas of high habitat heterogeneity using both habitats that wererecently burnt (1ndash2 months after fire) habitats that were longunburnt (gt24months after fire) and habitats that very rarely burn(boulder scree river edge and rainforest) Within these habitatsWyulda were highly frugivorous with on average 77 of plantmaterial fragments per scat sample identified as fruit epidermallayers Rock structure may contribute to protecting habitatheterogeneity and the availability of fruiting tree species fromextensive hot fires that can homogenise habitats and kill trees(Williams et al 1999 Vigilante and Bowman 2004a) Overallour results demonstrate diverse uses of heterogeneous habitat byWyulda and emphasise the critical importance of these habitats inthe conservation of small mammals in northern Australia

Previous research has suggested that rocky complexity is a keyhabitat feature forWyulda as the specieswas detected at siteswithrock and rainforest elements (Humphreys et al 1984 Bradleyet al 1987 Doody et al 2012) both for foraging and as den sites(Runcie 1999)Wyuldarsquos preference for rocky habitats could be aby-product of their denning requirements however it is likelythat other factors are involvedRockystructuremayprovide coverfrom predators such as cats which appear most adept at huntingin more open habitats (McGregor et al 2015) particularly after

Table 4 Asymptotic KruskalndashWallis tests of home-range overlapbetween sex dyads using two-dimensional (K95 and K50) and

three-dimensional (PRH and UDOI) indicesValues shown in bold are significant at Plt 005

Total overlap (95) Core overlap (50)c2 P c2 P

K 8272 0041 7863 0049PRH 9099 0028 6921 0074UDOI 8485 0037 4125 0248

04Kilometres

(a) (b) (c)

(d) (e) (f )

(g) (h)

N

Fig 4 Core home-range overlap (K50) between all male (dashed lines) and female (solid lines)Wsquamicaudata tracked in theArtesianRange north-westKimberleyWesternAustralia (a) Site1 January2013wet season (b) Site 1 April 2013 dry season (c) Site 1 July 2013 dry season (d) Site 1 December 2013wet season (e) Site 2 February 2013 wet season (f) Site 2 August 2013 dry season (g) Site 3 May 2013 dryseason (h) Site 3 January 2014 wet season Note that different individuals during each month-long trackingperiod can be distinguished by differing line weight or dash interval

Habitat use by Wyulda squamicaudata Australian Journal of Zoology G

understorey vegetation has been simplified by intense fire orgrazing (McGregor et al 2014) Reduced hunting success inrocky habitats may explain why cat density in this environment isvery low (Hohnen et al 2016) Also at a slightly larger spatialscale rocky landscapes tend to burn more patchily than opensavannas thus retaining more heterogeneity in vegetation age(Price et al 2003) In this study Wyulda selected for high levelsof habitat heterogeneity at both den and foraging sites and usedall ages of post-fire savannas as long as rocky features werepresent Habitat heterogeneity appears to be important for severalvertebrate species in northern Australia (Bolton and Latz 1978Ingleby and Westoby 1992 Pardon et al 2003 Southgate et al2007) and elsewhere (Doumas and Koprowski 2013)

While Wyulda are capable of consuming a wide range ofplant species fruit is clearly a key component of their diet Thisadds to information from a previous radio-tracking study that

identified Wyulda foraging within Xanthostemon paradoxusXanthostemon eucalyptoides and Planchonia careya (Runcie1999) all fruit-bearing species Fruit also makes an importantcontribution to the diet for other arboreal marsupials native tonorth-western Australia (Kerle 1985) Intense and frequent firescan kill or delay the fruiting of tree species (Williams et al 1999Vigilante and Bowman 2004a Atchison et al 2005 Atchison2009) Rocky structures (such as gorges or rock outcrops) canprovide conditions such as shade and water retention that favoursome fruiting species (such as Ficus platypoda) and also protectthem from fire Thus rock structure may contribute to theprotection of fire-sensitive foraging resources used by Wyulda

While the results of this study suggest that Wyulda favourrocky landscapes it is unclear whether the observed relationshipreflects a true habitat preference or is a result of recent reductionsof habitat quality in the open savannas caused by current fireregimes grazing by introduced herbivores and the impacts offeral cats Wyulda have been detected in fauna surveys on rockyescarpments of Charnley River Station separated from sites usedin this studybymore than15 kmof openbasalt plains (KTuft andS Legge unpubl data) On a larger scale Emma Gorge (wherethe species was recently rediscovered in the east Kimberley) isseparated from other historical records in the north Kimberley(eg Mitchell Plateau) by more than 200 km and large stretchesof open plains (Doody et al 2012) The highly disjunct natureof this distribution suggests that there may have been a timewhen the species moved more freely over open landscapesA recent phylogenetic study found that unlike several saxicolousrock wallaby species in the region Wyulda showed no deepstructure between eastern and western populations suggestingrecent colonisation or connectivity across the Kimberley (Potter

Table 5 Top five candidate discrete choice models derived fromforaging data for Wyulda squamicaudata

n1 no of real points included in the model n0 no of generated pointsincluded in the model Abbreviations are as follows hab_type habitat typehet habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K AIC DAIC w

Foraging model (n1 = 2615 n0 = 13071)hab_type+ het+ rock 3 914189 000 041hab_type+ het + rock + distc 4 914254 065 030hab_type+ fire + het + rock 4 914373 184 016hab_type+ fire + het + rock + distc 5 914425 236 013hab_type+ het 2 915147 958 000

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus05 00 05 10

Coefficent estimate

Var

iabl

e

Fig 5 Model-averaged coefficient estimates with 95 credible intervals of discrete choice models describing foraging-siteselection byW squamicaudata in the Artesian Range north-west Kimberley Western Australia The reference habitat typewas Heteropogon contortus savanna and level of rock cover was cliff

H Australian Journal of Zoology R Hohnen et al

et al 2014) Historical records from the east Kimberley suggestthat the species did once persist inmuch drier habitats (Alexander1919)

Wyulda in this study continued to den and forage in habitatsthat had been recently burnt (1ndash12 months after fire) a patternalso observed in several other northern Australian mammalstudies (Hohnen et al 2015) For pale field rats (Rattus tunneyi)in the central Kimberley Leahy et al (2016) found that whilepopulation size decreased after an intense fire individual rats

stayed within their original home ranges and did not move toadjacent unburnt areas Low-intensity fire did not displacenorthern quolls (Dasyurus hallucatus) from their home ranges inthe north Kimberley (Cook 2010) or significantly affect home-range size or location of northern bettongs (Bettongia tropica) innorth-east Queensland (Vernes and Pope 2001) The continueduse of savannahabitats afterfire byWyulda suggests that the costsof shifting home range are greater than any resource bottleneckcaused by recent or frequent fires The minimal home-rangeoverlap between adjacentmales and adjacent females in this studysuggests adegreeof territoriality thatmayalso inhibit home-rangemovement

The average home-range size estimated in this study(817 134 ha (MCP) or 802 129 ha (K95)) was larger thanprevious estimates of 10 ha (Runcie 1999) and a maximum of

Table 6 Odds ratios derived from the model-averaged coefficientestimates from discrete choice models of foraging data for Wyulda

squamicaudata

Variable Foraging dataOdds ratio Robust se

Recently burnt habitatsHibiscus keneallyii savanna 194 029Sarga stipodeum savanna 180 031

Long-unburnt habitatsRainforest 226 028River edge 106 030Long-unburnt savanna 157 029

Rock complexityRock complex 097 015Rock medium 083 012Rock simple 071 019

Other variablesFire 098 005Habitat heterogeneity 112 003Home-range distance 106 008

Table 7 Top five candidate models derived from den data forWyuldasquamicaudata using Bayesian hierarchical logistic regression modelsn1 number of real points included in the model n0 number of generatedpoints included in the model Abbreviations are as follows hab_type habitattype het habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K DIC DDIC w

Den model (n1 = 736 n0 = 736)hab_type + rock + het + distc 4 3008 000 017hab_type + rock + distc 3 3058 050 014hab_type + rock + het 3 3066 058 013hab_type + rock + het +fire + distc 5 3123 115 010hab_type + rock +fire + distc 4 3147 139 009hab_type + het 2 3178 170 007

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus400 minus200 0 200

Coefficent estimate

Var

iabl

e

Fig 6 Model-averaged coefficient estimates with 95 credible intervals of Bayesian hierarchical logistic regressionmodels describing den site selection byW squamicaudata in the Artesian Range north-west KimberleyWestern AustraliaThe reference habitat type was Heteropogon contortus savanna and level of rock cover was cliff

Habitat use by Wyulda squamicaudata Australian Journal of Zoology I

272m long (Humphreys et al 1984) Potentially this reflectsdifferences between the two studies in habitat quality andor thenumber of location points recorded per individual Home-rangeestimates from this study were larger than those of other northKimberley arboreal marsupials including the northern brushtailpossum (Trichosurus vulpecula arhnemensis) (087ndash112 haKerle 1998) and the cohabiting saxicolous rock ringtail possum(Petropseudes dahli) (05ndash12 ha Runcie 2000) While home-range size did not vary between sexes overlap occurred muchmore frequently and extensively between males and femalesthan between individuals of the same sex Several other possumspecies native to northern Australia move and den in familygroups (Kerle 1998 Runcie 2000) Potentially Wyulda alsopractice some degree of sociality and pair bonding

Currently the distribution of Wyulda is restricted to thetopographically complex north Kimberley with a small isolatedpopulation persisting in the east Kimberley While this speciesmay have once been more widely distributed in the centraland east Kimberley including in a wider range of habitatshabitat heterogeneity the availability of fruiting tree species andparticularly the availability of topographic and rock complexitywere key habitat features selected for by Wyulda Rockcomplexity may provide some buffer to Wyulda from threatssuch as intense fires and feral cats offer shelter from predationduring the post-fire period with cliffs and rock outcrops formingnatural barriers to such fires protecting sensitive vegetationand promoting habitat heterogeneity Mammal declines haveoccurred in topographically complex areas of the Arnhem Landplateau (Woinarski 2000 Ziembicki et al 2015) KakaduNational Park (Woinarski et al 2001) and the south-westKimberley (McKenzie 1981 Sawle 1988) Thus substratecomplexity alone does not guarantee mammal species securityManagement strategies that aim to maintain availability ofhabitat heterogeneity and fruiting trees may support thepersistence of Wyulda in rocky habitats Such strategies couldinclude lighting small-scale low-intensity fires early in the dryseason to prevent the incidence and extent of large-scale fires thatoccur late in the dry season

Acknowledgements

This studywas funded by supporters of the AustralianWildlife Conservancyan Australian Research Council Discovery Grant (100100033) and theDepartment of Parks and Wildlife (Western Australia) The study methodswere approved by the University of Tasmania Animal Ethics Committee(Permit no A12516) Fieldwork would not have been possible without thesupport of staff at theAustralianWildlifeConservancyrsquosMorningtonWildlifeSanctuary Thanks go in particular to Alex Hartshorne Hannah Cliff JillianSmith Joel Murray David James Peter Richardson Kaely Kreger DavinaBright Tom Crawford and Iris Bleach

References

Alexander W B (1919) A new species of marsupial of the subfamilyPhalangerinae Journal of the Royal Society of Western Australia 431ndash36

Andersen A N Braithwaite R W Cook G D Corbett L K WilliamsR J Douglas M M Gill A M Setterfield S A and Muller W J(1998) Fire research for conservation management in tropical savannasintroducing the Kapalga fire experiment Australian Journal of Ecology23 95ndash110 doi101111j1442-99931998tb00708x

Andersen A N Cook G D Corbett L K Douglas M M Eager R WRussell-Smith J SetterfieldSAWilliamsR J andWoinarski JCZ(2005) Fire frequency and biodiversity conservation in Australiantropical savannas implications from the Kapalga fire experiment AustralEcology 30 155ndash167 doi101111j1442-9993200501441x

Anderson D R (2008) lsquoModel Based Inferences in the Life Sciencesrsquo(Springer Science and Business Media New York)

Atchison J (2009) Human impacts on Persoonia falcata Perspectives onpost-contact vegetation change in the Keep River region Australia fromcontemporary vegetation surveysVegetationHistoryandArchaeobotany18 147ndash157 doi101007s00334-008-0198-y

Atchison J Head L and Fullagar R (2005) Archaeobotany of fruit seedprocessing in a monsoon savanna environment evidence from the KeepRiver region Northern Territory Australia Journal of ArchaeologicalScience 32 167ndash181 doi101016jjas200403022

Banks S C Dujardin M McBurney L Blair D Barker M andLindenmayer D B (2011) Starting points for small mammal populationrecovery after wildfire recolonisation or residual populationsOikos 12026ndash37 doi101111j1600-0706201018765x

Barton K (2013)MuMInmulti-model inference R package version 1913Bates D Maechler M Bolker B and Walker S (2014) lme4 linear

mixed-effects models using Eigen and S4 R package version 11-7Beyer H L (2012) Geospatial modelling environment Spatial Ecology

LLCBjoslashrneraas K Van Moorter B Rolandsen C M and Herfindal I (2010)

Screening global positioning system location data for errors usinganimal movement characteristics Journal of Wildlife Management 741361ndash1366 doi1021932009-405

Bolton B and Latz P (1978) The western hare-wallaby Lagorchesteshirsutus (Gould) (Macropodidae) in the Tanami Desert WildlifeResearch 5 285ndash293 doi101071WR9780285

BradleyAKemper CKitchenerDHumphreysW andHowR (1987)Small mammals of the Mitchell Plateau region Kimberley WesternAustralia Wildlife Research 14 397ndash413 doi101071WR9870397

Burbidge A A andMcKenzie N L (1989) Patterns in the modern declineof Western Australian vertebrate fauna ndash causes and conservationimplications Biological Conservation 50 143ndash198 doi1010160006-3207(89)90009-8

Burbidge A A and Webb M J (2008) Scaly-tailed possum In lsquoTheMammals of Australiarsquo 3rd edn (Eds S Van Dyck and R Strahan)pp 277ndash278 (New Holland Publishers Sydney)

Bureau of Meteorology (2015) Climate data online Commonwealth ofAustralia Bureau of Meteorology

Cook A C (2010) Habitat use and home range of the northern quollDasyurus hallucatus effects of fire MSc thesis The University ofWestern Australia Perth

Corbett L C Andersen A N and Muumlller W J (2003) Terrestrialvertebrates In lsquoFire inTropical SavannasTheKapalgaExperimentrsquo (EdsA N Andersen G D Cook and R J Williams) pp 126ndash52 (SpringerVerlag New York)

Doody J S Rhind D Castellano CM and BassM (2012) Rediscoveryof the scaly-tailed possum (Wyulda squamicaudata) in the easternKimberleyAustralianMammalogy 34 260ndash262 doi101071AM11039

Doumas S L and Koprowski J L (2013) Effect of heterogeneity inburn severity on Mexican fox squirrels following the return of fireInternational Journal of Wildland Fire 22 405ndash413 doi101071WF12046

Driscoll D A Lindenmayer D B Bennett A F Bode M BradstockR A CaryG J ClarkeM F Dexter N FenshamR Friend G GillM James S Kay G Keith D A MacGregor C Russell-Smith JSalt D Watson J E M Williams R J and York A (2010) Firemanagement for biodiversity conservation key research questions andour capacity to answer them Biological Conservation 143 1928ndash1939doi101016jbiocon201005026

J Australian Journal of Zoology R Hohnen et al

understorey plants and substrate (1) Hibiscus keneallyi savannaon skeletal sandy soils over a sandstone substrate (2) Sorghumstipoideum savanna on shallow sandy soils over sandstone

(3) Chrysopogon fallax savanna on shallow sandy soils oversandstone substrate and (4) Heteropogon contortus savannaon skeletal soils over basalt We assessed floristic differences

0 01 02 04 KmN

Site 1 Site 2

Site 3Habitat type

Boulder scree

Rainforest

River edge

Triodia spp savanna

Hibiscus keneallyi savanna

Sorghum stipoideum savanna

Chrysopogon fallax savanna

Heteropogon contortus savanna

River

Inaccessible

Fig 2 Habitat composition of sites in the Artesian Range north-west Kimberley Western Australia

(a) (b)

Fig 3 Photos of (a) an individual male Wyulda squamicaudata and (b) typical rugged terrain in which this species is found

Habitat use by Wyulda squamicaudata Australian Journal of Zoology C

between habitat types by analysing vegetation survey data byordination as in Hohnen et al (2015) The availability of thesevegetation types varied between sites (Table 1)

Trapping

We captured Wyulda in treadle-operated wire cage traps(L 30 cm H 30 cm W 70 cm Sheffield Wire ProductsWelshpool Western Australia) At each site 30 traps were set10ndash30m apart and baited with a mixture of peanut butteroats honey and apple Wyulda were monitored at Site 1 in fourseparate months twice in the wet season (January 2013December 2013) and twice in the dry season (April 2013 June2013) Sites 2 and 3 were each visited once in the dry season(May and August 2013 respectively) and once in the wet season(February 2013 and January 2014 respectively) for a total of4470 trap-nights

On first capture each animal was weighed sexed and taggedwith a nanotransponder microchip (Trovan United Kingdom)Faecal pellets were collected from traps for diet analysisIndividuals were considered adult if theyweighed gt1000 g or forfemales if their teats showed evidence of current or previouslactation Trapped individuals gt1065 g were fitted with GPSVHF composite collars (Telemetry Solutions Concord CAUSA) Collars weighed 32 g constituting lt3 of the animalrsquosbody weight Each unit consisted of a GPS logger and a VHFtransmitter attached to a strong lightweight material collarDuring January and February 2013 the GPS collars were not yetavailable so VHF collars were used These collars weighed 30 g(SirtrackHavelockNorthNewZealand)lt3of an individualrsquosbody weight and were fastened around the neck with a leatherstrap

The GPS collars logged foraging locations hourly between1900 and 0500 hours for the maximum battery life of the collars(three weeks) Radio-collared Wyulda were located betweenone and four times each night bymeans of a radio receiver (TitleyScientific Brisbane Australia) and a Yagi antenna (SirtrackHavelock North New Zealand) using the homing methodoutlined in White and Garrot (1990) Individuals were located aminimum of 1 h after a previous fix to minimise autocorrelationAs Wyulda were thought to den underground where there is noGPS reception den sites were located daily using the VHF signalfor both collar types

Data analysis

Diet analysis

A reference collection of the epidermal layers of 161 plantspecies (including 161 leaf 6 flower and 18 fruit slides) weremade from plant specimens collected at the three sites Referenceslides were made by soaking fragments of plant specimen in42 g Lndash1 sodium hypochlorite for 24 h or until 90of thematerialwas visibly bleached following themethodsoutlined inTuft et al(2011) The material was then rinsed stained with 02 g Lndash1

gentian violet and mounted onto a slide using corn syrup Of the108 scat samples collected 47 were selected for analysis tomaximise the representation of the sexes (20 male 28 female)sites (25 Site 1 11 Site 2 12 Site 3) and seasons (24 wet season24 dry season) To prepare scat samples for analysis sampleswere broken up loosely and also soaked in 42 g Lndash1 sodiumhypochlorite for 24 h or until 90 of the material was visiblybleached The material was rinsed in water through a 250-mmsieve whole seeds were removed identified and countedRemaining leafmaterialwas stained using02 g Lndash1 gentian violetandmounted on a slide using corn syrup For each scat specimen100 identifiable fragments of plant material were counted intransects across the slides Unidentifiable fragments (where theepidermal layer was not visible) were uncommon and were notincluded in counts

Movement analysis

GPS points were first screened to remove erroneous fixescaused byGPS error These included location points found on theopposite side of the impassable Charnley River (at Sites 1 and 2see Fig 2) or that displayed all of the following characteristicsdistance from the last fix gt300m and turning angle (in relation tothe previous and the followingfix)gt180 (Bjoslashrneraas et al 2010)The combination of these characteristics indicated implausiblespikes in which an individual appeared to deviate at high speedfrom his or her trajectory to a distant location and then return tothe original trajectory Horizontal dilution of precision (HDOP)values were not used to filter fixes as data from two stationary testGPS collars indicated that there was no relationship betweendistance to known location and HDOP value

Home ranges were estimated by minimum convex polygons(MCP) and fixed kernel density estimation methods in theprogram Geospatial Modelling Environment (Beyer 2012) and

Table 1 Proportions of habitat types available at three sites in the Artesian Range north-west Kimberley Western AustraliaAn asterisk indicates that last fire is unknown but gt24 months preceding commencement of study period

Habitat type Last fire Age Site 1 Site 2 Site 3(months) Area

(ha)Proportion

()Area(ha)

Proportion()

Area(ha)

Proportion()

Boulder scree 60+ 669 383 359 203 000 000Rainforest 60+ 960 549 361 204 938 505River edge 60+ 667 382 558 315 000 000Triodia spp savanna 24+ 617 353 000 000 615 331Hibiscus keneallyi savanna Nov 2012 1ndash12 14555 8333 16448 9078 1081 582Sorghum stipoideum savanna Oct 2013 3 000 000 000 000 8923 4804Chrysopogon fallax savanna Oct 2013 3 000 000 000 000 1137 612Heteropogon contortus savanna Oct 2013 3 000 000 000 000 5882 3166

D Australian Journal of Zoology R Hohnen et al

interpreted in ArcMap 101 (ESRI) Kernel density estimateswere calculated with a plug-in bandwidth The 95 kernel (K95)utilisation contour estimated total home range and the 50(K50) kernel utilisation contour estimated the core home rangeAll points (both foraging and denning) were included in theMCPand kernel estimates and all areas that the animals could not use(including the river and land on the opposite side of the river)were removed from further consideration Home-range estimateswere tested for normality using the ShapirondashWilk W test andnormalised by log-transformation Differences in home-rangesize between sex and season were examined using generalisedlinear models with a model-averaging approach using thepackage lme4 and MuMIn in the statistical program R ver 303(R Development Core Team 2005 Barton 2013 Bates et al2014)

We calculated the total percentage home-range overlap ofK95 and core K50 home-range estimates for all neighbouringindividuals Home-range overlap was also compared using PHR(Percentile home range) and UDOI (Utilisation distributionoverlap index) indices which consider heterogeneity of spaceuse within home ranges (Fieberg and Kochanny 2005) (seeAppendix 2 for more detail) Differences in overlap extent werecompared between dyads using the asymptotic KruskallndashWallistest as datasets did not meet assumptions of normality

We used a discrete choice modelling approach to describestep selection of Wyulda during foraging periods Coefficientswere estimated from case-control logistic regression modelsusing the lsquoclogitrsquo command in the package survival (Therneau2014) within the statistical program R ver 303 This regressionmethod accounts for differing availabilities of habitats betweenobservations by matching every known animal location to achoice set of randomly generated locations To generate randompoints we first calculated the average distance travelled betweenfixes taken at 1-h 2-h or 3-h (and so on) intervals Then forexample if between fixA andB there was a gap of 2 h five pointswere randomly generated within a circle (with A at its centroid)that had a radius consisting of the average distance travelled byWyulda in 2 h clipped by the individualrsquos 95 kernel home-range estimate Those five points represent locations where theanimal could have gone and were compared in the modellingprocess to point B the location where the animal did go

For all points where Wyulda were found we recorded habitattype and rock complexity (Appendix 3) The variable lsquorockcomplexityrsquo comprised four categories cliffs high complexity(boulder size gt2m) medium complexity (boulder size lt2m) andlow complexity (no exposed rock) Habitat and rockiness mapsfor each site were created in ArcMap 101 using geo-referencedsatellite images from Google Earth (Google Inc 2009) Othervariables investigatedwere distance topermanentwater (referredto as lsquodistance to waterrsquo) number of habitat types within a20-m radius (lsquohabitat heterogeneityrsquo) distance to unburnt habitat(lsquodistance to unburnt patchrsquo) fire (lsquotime since firersquo) andthe distance from the point to the centre of the home range(lsquodistance to home range centroidrsquo) We chose 20m as the radiusof the circle within which to calculate habitat heterogeneity aswe aimed to take into accountGPS error aswell as an individualrsquosperception of where it is relative to the habitats around itWe tested for correlation and redundancy between explanatoryvariables by identifying correlation coefficients of gt05

(Fox 2002) The variables lsquofirersquo lsquodistance towaterrsquo and lsquodistanceto unburnt patchrsquo were strongly correlated with one another solsquodistance to waterrsquo and lsquodistance to unburnt patchrsquo were omittedfrom further analysis Numerical variables including habitatheterogeneity time sincefire anddistance tohome rangecentroidwere standardised (mean = 0 standard deviation = 1) beforeanalysis

We created 31 candidatemodels based on combinations of thevariables lsquohabitat typersquo lsquotime since firersquo lsquohabitat heterogeneityrsquolsquorock complexityrsquo and lsquodistance to home range centroidrsquo withlsquoindividualrsquo considered as a random effect Candidate modelswere ranked on the basis of the Akaikersquos information criterionadjusted for small sample sizes (AICc) The selection forlandscape characteristics by Wyulda was measured by thecoefficient estimates and the odds ratio which representsthe relative change in the odds of selection for each unit of thepredictor variableWe usedmodel averaging to estimate variablecoefficients and their weights according to Anderson (2008)

Den site selection

To explore den site selection by Wyulda we used Bayesianhierarchical logistic regression models For each individual wecompared known den sites with an equal number of randomlygenerated den sites within that individualrsquos home range Modelscomprised a binomial response variable (that described if a pointwas real (1) or generated (0)) and all combinations of theaforementioned predictor variables including habitat type rockcomplexity time since fire habitat heterogeneity and distance tohome range centroid To account for variation in individual useof habitats lsquoindividualrsquo was included as a random effect in allmodels (See Appendix 4 for more details) A Markov ChainMonte Carlo (MCMC) procedure was used to estimate posteriordistributions using the package MCMCglmm (Hadfield 2010) inR ver 303 MCMC chains were run with randomised startingvalues aburn-inof 50 000 iterations a thinning interval of 50 anda total of 200 000 iterations Convergence of MCMC chains wasevaluated following Gelman and Hill (2006) The DevianceInformation Criterion (DIC) was used to compare all possiblemodels and calculate model weightsWe compared all subsets ofthe most complex model We used model averaging to estimatevariable coefficients posterior credible intervals and variableweights according to Anderson (2008)

Results

Between January 2013 and January 2014 collars were deployedfor one month each on 34 occasions on 23Wyulda (15 males and19 females) Of those deployments 29 were GPS collars and5 were VHF giving a total of 3314 location points Nineindividuals were tracked on more than one occasion and theirhome ranges were calculated separately for each tracking period(Appendix 5) The number of individuals known to be alive ateach site varied between 5 and 14 (Appendix 6)

Diet

From the scat samples 4700 fragments of plant material wereidentified Fragments of leaf came from 42 species fruit from 14species and flower from one species (Appendix 7) The mostcommon items were Ficus platypoda fruit (on average 47 of

Habitat use by Wyulda squamicaudata Australian Journal of Zoology E

fragments per sample) Owenia vernicosa fruit (10) Vitexacuminata fruit (10) Grewia breviflora fruit (8)V acuminata leaf (3) Strychnos lucida fruit (2)O vernicosaleaf (2) Cryptocarya cunninghamii leaf (2) and Acaciadimorpha leaf (1) Whole seeds were also frequently found inscats the most common being F platypoda (on average 55 ofseeds per sample) followed by V acuminata (15) UnknownSpecies 3 (6)UnknownSpecies 2 (4) andG breviflora (4)F brachypoda Unknown Species 1 Trachymene dendrothrixand Cryptocarya cunninghamii seeds were also present in lownumbers Plant species that took up on average a largerpercentage composition per sample were also likely to be foundin a greater number of samples (Appendix 7)

Home range

Estimated home-range sizewas not correlatedwith the number offixes per individual (Appendix 8) Mean home-range size was817 134 ha (MCP) or 802 129 ha (K95) and mean corehome-range size (K50) was 214 038 ha (Table 2) Meanhome-range size formaleswas 932 210 (MCP) or 981 233(K95) and for females was 725 176 (MCP) or 661 137(K95) In the sets ofmodels describing both variation inMCPandK95 home-range estimates with sex and season multiple modelswere competitive with one another (Appendix 9) For the MCPmodels the confidence intervals of the model-averagedcoefficient estimates overlapped zero for the variable lsquosexrsquo butdid not overlap zero for the variable lsquoseasonrsquo The coefficientestimate of the variable lsquodry seasonrsquowas positive For the K95models both thevariables lsquosexrsquo and lsquoseasonrsquohadmodel-averagedcoefficient estimates and confidence intervals that overlappedzero (Appendix 10)

Home-range overlap

Home-range overlap occurred frequently between neighbouringWyulda (Table 3) Extent of total home-range overlap differedbetween dyad types for the static K95 contour PHR andUDOIindices of total home-range overlap (Table 4) Total home-rangeoverlapoccurredmost frequentlybetween individuals of differentsexes (Appendix 11) and least frequently between neighbouringmales (Table 3) The extent of core home-range overlap alsodiffered between sexdyads andwasmost commonbetweenmalesand females (Fig 4 Appendix 11)

Foraging site selection

Two models (model weights of 30 and 16) were withintwo DAIC of the top model (model weight of 41 Table 5)The variables lsquohabitat typersquo lsquorock complexityrsquo lsquohabitatheterogeneityrsquo and lsquodistance to home range centroidrsquo all had highvariable importance weights of gt99 and the variable timesince fire had a low importance weight of 27 Model-averagedcoefficient estimates suggested that Wyulda selected againstboth lowandmedium levels of rockycomplexity (Fig 5Table 6)Weak but positive selection was detected for locations at greaterdistances from the centre of an individualrsquos home range Wyuldaselected areas of high habitat heterogeneity while foraging andused long unburnt habitats such as rainforest and savannadominated by Triodia spp as well as recently burnt savanna(1ndash12 months after fire) dominated by Hibiscus kenealleyi andSorghum stipodeum in preference to recently burntHeteropogoncontortusndashdominated savanna There was no selection for oragainst river-edge habitat types or in relation to the variable lsquotimesince firersquo

Den site selection

Individuals were tracked to a total of 736 den sites all of whichwere in rock features 548 in piles of rocky scree 108 in cracks incliffs 57 in vertical clefts in large rocky slabs and 23 in massiverocky outcrops Of the models built on the den site data only fivewere competitive (within 2 DDIC) with the top model (Table 7)Importance weights were highest for the variables lsquohabitat typersquoand lsquorock complexityrsquo (92 and 83 respectively) followed bylsquodistance to home range centroidrsquo (68) lsquohabitat heterogeneityrsquo(59) and lsquotime since firersquo (36) Model-averaged coefficientestimates suggest that Wyulda selected against denning intopographically simple habitats and at long distances from theirhome range centroids (Fig 6)Wyulda selected to den inHibiscuskeneallyi and Sorghum stipodeumndashdominated savanna habitatsboth 1ndash12 months after fire and against denning in long-unburnt

Table 2 The mean home-range size ofW squamicaudata across sexessites and seasons

MCPplusmn se(ha)

K95plusmn se(ha)

K50plusmn se(ha)

Female (n= 19) 725 plusmn 176 661 plusmn 137 194 plusmn 046Male (n= 15) 932 plusmn 210 981 plusmn 233 239 plusmn 065Dry season (n= 21) 1018 plusmn 199 955 plusmn 196 213 plusmn 050Wet season (n= 13) 492 plusmn 092 555 plusmn 090 214 plusmn 061Site 1 (n= 13) 725 plusmn 107 639 plusmn 059 125 plusmn 012Site 2 (n= 8) 1433 plusmn 491 139 plusmn 483 461 plusmn 123Site 3 (n= 13) 528 plusmn 068 603 plusmn 094 150 plusmn 026GPS (n= 29) 874 plusmn 155 857 plusmn 148 191 plusmn 038VHF (n= 5) 486 plusmn 095 486 plusmn 100 347 plusmn 145Total (n= 34) 817 plusmn 134 802 plusmn 129 214 plusmn 038

Table3 Frequencyandaveragehome-rangeoverlapbetweenadjacentWsquamicaudatamonitoredconcurrently at three sites in theArtesianRangenorth-west Kimberley Western Australia (mean se) using two-dimensional (K95and K50) and three-dimensional (PRH and UDOI) indices

Tnd=Total no of neighbouring dyads Tndo =Total no of neighbouring dyads that overlapped F female M male

Sex Tnd Tndo Tndo Mean total overlap Mean core overlapdyad (K95) (K50) K95 PRH UDOI K50 PRH UDOI

FF 24 22 6 2019 plusmn 456 027 plusmn 006 011 plusmn 004 437 plusmn 239 006 plusmn 003 005 plusmn 002FM 37 34 16 3700 plusmn 604 048 plusmn 007 033 plusmn 007 2068 plusmn 545 021 plusmn 005 013 plusmn 004MF 37 34 16 2424 plusmn 405 037 plusmn 006 033 plusmn 007 1199 plusmn 319 011 plusmn 002 007 plusmn 002MM 26 22 6 1210 plusmn 302 020 plusmn 004 004 plusmn 001 071 plusmn 032 002 plusmn 001 001 plusmn 000

F Australian Journal of Zoology R Hohnen et al

rainforest and river-edge habitats Credible intervals of thevariable lsquoTriodia spp savannarsquo included zero suggesting neutralselection Wyulda also appeared to select for older post-fire agesof all habitat types but this variable had a low importanceweight

Discussion

This study sought to test utilisation of complex landscapes andsurrounding areas by a declining mammal in the presence ofintensified fire regimes Rock complexity appears to be a keyhabitat feature selected for by Wyulda Individuals avoidedforaging in habitats with low rock complexity and consistently

denned within rock structures Wyulda preferred to forage inareas of high habitat heterogeneity using both habitats that wererecently burnt (1ndash2 months after fire) habitats that were longunburnt (gt24months after fire) and habitats that very rarely burn(boulder scree river edge and rainforest) Within these habitatsWyulda were highly frugivorous with on average 77 of plantmaterial fragments per scat sample identified as fruit epidermallayers Rock structure may contribute to protecting habitatheterogeneity and the availability of fruiting tree species fromextensive hot fires that can homogenise habitats and kill trees(Williams et al 1999 Vigilante and Bowman 2004a) Overallour results demonstrate diverse uses of heterogeneous habitat byWyulda and emphasise the critical importance of these habitats inthe conservation of small mammals in northern Australia

Previous research has suggested that rocky complexity is a keyhabitat feature forWyulda as the specieswas detected at siteswithrock and rainforest elements (Humphreys et al 1984 Bradleyet al 1987 Doody et al 2012) both for foraging and as den sites(Runcie 1999)Wyuldarsquos preference for rocky habitats could be aby-product of their denning requirements however it is likelythat other factors are involvedRockystructuremayprovide coverfrom predators such as cats which appear most adept at huntingin more open habitats (McGregor et al 2015) particularly after

Table 4 Asymptotic KruskalndashWallis tests of home-range overlapbetween sex dyads using two-dimensional (K95 and K50) and

three-dimensional (PRH and UDOI) indicesValues shown in bold are significant at Plt 005

Total overlap (95) Core overlap (50)c2 P c2 P

K 8272 0041 7863 0049PRH 9099 0028 6921 0074UDOI 8485 0037 4125 0248

04Kilometres

(a) (b) (c)

(d) (e) (f )

(g) (h)

N

Fig 4 Core home-range overlap (K50) between all male (dashed lines) and female (solid lines)Wsquamicaudata tracked in theArtesianRange north-westKimberleyWesternAustralia (a) Site1 January2013wet season (b) Site 1 April 2013 dry season (c) Site 1 July 2013 dry season (d) Site 1 December 2013wet season (e) Site 2 February 2013 wet season (f) Site 2 August 2013 dry season (g) Site 3 May 2013 dryseason (h) Site 3 January 2014 wet season Note that different individuals during each month-long trackingperiod can be distinguished by differing line weight or dash interval

Habitat use by Wyulda squamicaudata Australian Journal of Zoology G

understorey vegetation has been simplified by intense fire orgrazing (McGregor et al 2014) Reduced hunting success inrocky habitats may explain why cat density in this environment isvery low (Hohnen et al 2016) Also at a slightly larger spatialscale rocky landscapes tend to burn more patchily than opensavannas thus retaining more heterogeneity in vegetation age(Price et al 2003) In this study Wyulda selected for high levelsof habitat heterogeneity at both den and foraging sites and usedall ages of post-fire savannas as long as rocky features werepresent Habitat heterogeneity appears to be important for severalvertebrate species in northern Australia (Bolton and Latz 1978Ingleby and Westoby 1992 Pardon et al 2003 Southgate et al2007) and elsewhere (Doumas and Koprowski 2013)

While Wyulda are capable of consuming a wide range ofplant species fruit is clearly a key component of their diet Thisadds to information from a previous radio-tracking study that

identified Wyulda foraging within Xanthostemon paradoxusXanthostemon eucalyptoides and Planchonia careya (Runcie1999) all fruit-bearing species Fruit also makes an importantcontribution to the diet for other arboreal marsupials native tonorth-western Australia (Kerle 1985) Intense and frequent firescan kill or delay the fruiting of tree species (Williams et al 1999Vigilante and Bowman 2004a Atchison et al 2005 Atchison2009) Rocky structures (such as gorges or rock outcrops) canprovide conditions such as shade and water retention that favoursome fruiting species (such as Ficus platypoda) and also protectthem from fire Thus rock structure may contribute to theprotection of fire-sensitive foraging resources used by Wyulda

While the results of this study suggest that Wyulda favourrocky landscapes it is unclear whether the observed relationshipreflects a true habitat preference or is a result of recent reductionsof habitat quality in the open savannas caused by current fireregimes grazing by introduced herbivores and the impacts offeral cats Wyulda have been detected in fauna surveys on rockyescarpments of Charnley River Station separated from sites usedin this studybymore than15 kmof openbasalt plains (KTuft andS Legge unpubl data) On a larger scale Emma Gorge (wherethe species was recently rediscovered in the east Kimberley) isseparated from other historical records in the north Kimberley(eg Mitchell Plateau) by more than 200 km and large stretchesof open plains (Doody et al 2012) The highly disjunct natureof this distribution suggests that there may have been a timewhen the species moved more freely over open landscapesA recent phylogenetic study found that unlike several saxicolousrock wallaby species in the region Wyulda showed no deepstructure between eastern and western populations suggestingrecent colonisation or connectivity across the Kimberley (Potter

Table 5 Top five candidate discrete choice models derived fromforaging data for Wyulda squamicaudata

n1 no of real points included in the model n0 no of generated pointsincluded in the model Abbreviations are as follows hab_type habitat typehet habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K AIC DAIC w

Foraging model (n1 = 2615 n0 = 13071)hab_type+ het+ rock 3 914189 000 041hab_type+ het + rock + distc 4 914254 065 030hab_type+ fire + het + rock 4 914373 184 016hab_type+ fire + het + rock + distc 5 914425 236 013hab_type+ het 2 915147 958 000

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus05 00 05 10

Coefficent estimate

Var

iabl

e

Fig 5 Model-averaged coefficient estimates with 95 credible intervals of discrete choice models describing foraging-siteselection byW squamicaudata in the Artesian Range north-west Kimberley Western Australia The reference habitat typewas Heteropogon contortus savanna and level of rock cover was cliff

H Australian Journal of Zoology R Hohnen et al

et al 2014) Historical records from the east Kimberley suggestthat the species did once persist inmuch drier habitats (Alexander1919)

Wyulda in this study continued to den and forage in habitatsthat had been recently burnt (1ndash12 months after fire) a patternalso observed in several other northern Australian mammalstudies (Hohnen et al 2015) For pale field rats (Rattus tunneyi)in the central Kimberley Leahy et al (2016) found that whilepopulation size decreased after an intense fire individual rats

stayed within their original home ranges and did not move toadjacent unburnt areas Low-intensity fire did not displacenorthern quolls (Dasyurus hallucatus) from their home ranges inthe north Kimberley (Cook 2010) or significantly affect home-range size or location of northern bettongs (Bettongia tropica) innorth-east Queensland (Vernes and Pope 2001) The continueduse of savannahabitats afterfire byWyulda suggests that the costsof shifting home range are greater than any resource bottleneckcaused by recent or frequent fires The minimal home-rangeoverlap between adjacentmales and adjacent females in this studysuggests adegreeof territoriality thatmayalso inhibit home-rangemovement

The average home-range size estimated in this study(817 134 ha (MCP) or 802 129 ha (K95)) was larger thanprevious estimates of 10 ha (Runcie 1999) and a maximum of

Table 6 Odds ratios derived from the model-averaged coefficientestimates from discrete choice models of foraging data for Wyulda

squamicaudata

Variable Foraging dataOdds ratio Robust se

Recently burnt habitatsHibiscus keneallyii savanna 194 029Sarga stipodeum savanna 180 031

Long-unburnt habitatsRainforest 226 028River edge 106 030Long-unburnt savanna 157 029

Rock complexityRock complex 097 015Rock medium 083 012Rock simple 071 019

Other variablesFire 098 005Habitat heterogeneity 112 003Home-range distance 106 008

Table 7 Top five candidate models derived from den data forWyuldasquamicaudata using Bayesian hierarchical logistic regression modelsn1 number of real points included in the model n0 number of generatedpoints included in the model Abbreviations are as follows hab_type habitattype het habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K DIC DDIC w

Den model (n1 = 736 n0 = 736)hab_type + rock + het + distc 4 3008 000 017hab_type + rock + distc 3 3058 050 014hab_type + rock + het 3 3066 058 013hab_type + rock + het +fire + distc 5 3123 115 010hab_type + rock +fire + distc 4 3147 139 009hab_type + het 2 3178 170 007

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus400 minus200 0 200

Coefficent estimate

Var

iabl

e

Fig 6 Model-averaged coefficient estimates with 95 credible intervals of Bayesian hierarchical logistic regressionmodels describing den site selection byW squamicaudata in the Artesian Range north-west KimberleyWestern AustraliaThe reference habitat type was Heteropogon contortus savanna and level of rock cover was cliff

Habitat use by Wyulda squamicaudata Australian Journal of Zoology I

272m long (Humphreys et al 1984) Potentially this reflectsdifferences between the two studies in habitat quality andor thenumber of location points recorded per individual Home-rangeestimates from this study were larger than those of other northKimberley arboreal marsupials including the northern brushtailpossum (Trichosurus vulpecula arhnemensis) (087ndash112 haKerle 1998) and the cohabiting saxicolous rock ringtail possum(Petropseudes dahli) (05ndash12 ha Runcie 2000) While home-range size did not vary between sexes overlap occurred muchmore frequently and extensively between males and femalesthan between individuals of the same sex Several other possumspecies native to northern Australia move and den in familygroups (Kerle 1998 Runcie 2000) Potentially Wyulda alsopractice some degree of sociality and pair bonding

Currently the distribution of Wyulda is restricted to thetopographically complex north Kimberley with a small isolatedpopulation persisting in the east Kimberley While this speciesmay have once been more widely distributed in the centraland east Kimberley including in a wider range of habitatshabitat heterogeneity the availability of fruiting tree species andparticularly the availability of topographic and rock complexitywere key habitat features selected for by Wyulda Rockcomplexity may provide some buffer to Wyulda from threatssuch as intense fires and feral cats offer shelter from predationduring the post-fire period with cliffs and rock outcrops formingnatural barriers to such fires protecting sensitive vegetationand promoting habitat heterogeneity Mammal declines haveoccurred in topographically complex areas of the Arnhem Landplateau (Woinarski 2000 Ziembicki et al 2015) KakaduNational Park (Woinarski et al 2001) and the south-westKimberley (McKenzie 1981 Sawle 1988) Thus substratecomplexity alone does not guarantee mammal species securityManagement strategies that aim to maintain availability ofhabitat heterogeneity and fruiting trees may support thepersistence of Wyulda in rocky habitats Such strategies couldinclude lighting small-scale low-intensity fires early in the dryseason to prevent the incidence and extent of large-scale fires thatoccur late in the dry season

Acknowledgements

This studywas funded by supporters of the AustralianWildlife Conservancyan Australian Research Council Discovery Grant (100100033) and theDepartment of Parks and Wildlife (Western Australia) The study methodswere approved by the University of Tasmania Animal Ethics Committee(Permit no A12516) Fieldwork would not have been possible without thesupport of staff at theAustralianWildlifeConservancyrsquosMorningtonWildlifeSanctuary Thanks go in particular to Alex Hartshorne Hannah Cliff JillianSmith Joel Murray David James Peter Richardson Kaely Kreger DavinaBright Tom Crawford and Iris Bleach

References

Alexander W B (1919) A new species of marsupial of the subfamilyPhalangerinae Journal of the Royal Society of Western Australia 431ndash36

Andersen A N Braithwaite R W Cook G D Corbett L K WilliamsR J Douglas M M Gill A M Setterfield S A and Muller W J(1998) Fire research for conservation management in tropical savannasintroducing the Kapalga fire experiment Australian Journal of Ecology23 95ndash110 doi101111j1442-99931998tb00708x

Andersen A N Cook G D Corbett L K Douglas M M Eager R WRussell-Smith J SetterfieldSAWilliamsR J andWoinarski JCZ(2005) Fire frequency and biodiversity conservation in Australiantropical savannas implications from the Kapalga fire experiment AustralEcology 30 155ndash167 doi101111j1442-9993200501441x

Anderson D R (2008) lsquoModel Based Inferences in the Life Sciencesrsquo(Springer Science and Business Media New York)

Atchison J (2009) Human impacts on Persoonia falcata Perspectives onpost-contact vegetation change in the Keep River region Australia fromcontemporary vegetation surveysVegetationHistoryandArchaeobotany18 147ndash157 doi101007s00334-008-0198-y

Atchison J Head L and Fullagar R (2005) Archaeobotany of fruit seedprocessing in a monsoon savanna environment evidence from the KeepRiver region Northern Territory Australia Journal of ArchaeologicalScience 32 167ndash181 doi101016jjas200403022

Banks S C Dujardin M McBurney L Blair D Barker M andLindenmayer D B (2011) Starting points for small mammal populationrecovery after wildfire recolonisation or residual populationsOikos 12026ndash37 doi101111j1600-0706201018765x

Barton K (2013)MuMInmulti-model inference R package version 1913Bates D Maechler M Bolker B and Walker S (2014) lme4 linear

mixed-effects models using Eigen and S4 R package version 11-7Beyer H L (2012) Geospatial modelling environment Spatial Ecology

LLCBjoslashrneraas K Van Moorter B Rolandsen C M and Herfindal I (2010)

Screening global positioning system location data for errors usinganimal movement characteristics Journal of Wildlife Management 741361ndash1366 doi1021932009-405

Bolton B and Latz P (1978) The western hare-wallaby Lagorchesteshirsutus (Gould) (Macropodidae) in the Tanami Desert WildlifeResearch 5 285ndash293 doi101071WR9780285

BradleyAKemper CKitchenerDHumphreysW andHowR (1987)Small mammals of the Mitchell Plateau region Kimberley WesternAustralia Wildlife Research 14 397ndash413 doi101071WR9870397

Burbidge A A andMcKenzie N L (1989) Patterns in the modern declineof Western Australian vertebrate fauna ndash causes and conservationimplications Biological Conservation 50 143ndash198 doi1010160006-3207(89)90009-8

Burbidge A A and Webb M J (2008) Scaly-tailed possum In lsquoTheMammals of Australiarsquo 3rd edn (Eds S Van Dyck and R Strahan)pp 277ndash278 (New Holland Publishers Sydney)

Bureau of Meteorology (2015) Climate data online Commonwealth ofAustralia Bureau of Meteorology

Cook A C (2010) Habitat use and home range of the northern quollDasyurus hallucatus effects of fire MSc thesis The University ofWestern Australia Perth

Corbett L C Andersen A N and Muumlller W J (2003) Terrestrialvertebrates In lsquoFire inTropical SavannasTheKapalgaExperimentrsquo (EdsA N Andersen G D Cook and R J Williams) pp 126ndash52 (SpringerVerlag New York)

Doody J S Rhind D Castellano CM and BassM (2012) Rediscoveryof the scaly-tailed possum (Wyulda squamicaudata) in the easternKimberleyAustralianMammalogy 34 260ndash262 doi101071AM11039

Doumas S L and Koprowski J L (2013) Effect of heterogeneity inburn severity on Mexican fox squirrels following the return of fireInternational Journal of Wildland Fire 22 405ndash413 doi101071WF12046

Driscoll D A Lindenmayer D B Bennett A F Bode M BradstockR A CaryG J ClarkeM F Dexter N FenshamR Friend G GillM James S Kay G Keith D A MacGregor C Russell-Smith JSalt D Watson J E M Williams R J and York A (2010) Firemanagement for biodiversity conservation key research questions andour capacity to answer them Biological Conservation 143 1928ndash1939doi101016jbiocon201005026

J Australian Journal of Zoology R Hohnen et al

between habitat types by analysing vegetation survey data byordination as in Hohnen et al (2015) The availability of thesevegetation types varied between sites (Table 1)

Trapping

We captured Wyulda in treadle-operated wire cage traps(L 30 cm H 30 cm W 70 cm Sheffield Wire ProductsWelshpool Western Australia) At each site 30 traps were set10ndash30m apart and baited with a mixture of peanut butteroats honey and apple Wyulda were monitored at Site 1 in fourseparate months twice in the wet season (January 2013December 2013) and twice in the dry season (April 2013 June2013) Sites 2 and 3 were each visited once in the dry season(May and August 2013 respectively) and once in the wet season(February 2013 and January 2014 respectively) for a total of4470 trap-nights

On first capture each animal was weighed sexed and taggedwith a nanotransponder microchip (Trovan United Kingdom)Faecal pellets were collected from traps for diet analysisIndividuals were considered adult if theyweighed gt1000 g or forfemales if their teats showed evidence of current or previouslactation Trapped individuals gt1065 g were fitted with GPSVHF composite collars (Telemetry Solutions Concord CAUSA) Collars weighed 32 g constituting lt3 of the animalrsquosbody weight Each unit consisted of a GPS logger and a VHFtransmitter attached to a strong lightweight material collarDuring January and February 2013 the GPS collars were not yetavailable so VHF collars were used These collars weighed 30 g(SirtrackHavelockNorthNewZealand)lt3of an individualrsquosbody weight and were fastened around the neck with a leatherstrap

The GPS collars logged foraging locations hourly between1900 and 0500 hours for the maximum battery life of the collars(three weeks) Radio-collared Wyulda were located betweenone and four times each night bymeans of a radio receiver (TitleyScientific Brisbane Australia) and a Yagi antenna (SirtrackHavelock North New Zealand) using the homing methodoutlined in White and Garrot (1990) Individuals were located aminimum of 1 h after a previous fix to minimise autocorrelationAs Wyulda were thought to den underground where there is noGPS reception den sites were located daily using the VHF signalfor both collar types

Data analysis

Diet analysis

A reference collection of the epidermal layers of 161 plantspecies (including 161 leaf 6 flower and 18 fruit slides) weremade from plant specimens collected at the three sites Referenceslides were made by soaking fragments of plant specimen in42 g Lndash1 sodium hypochlorite for 24 h or until 90of thematerialwas visibly bleached following themethodsoutlined inTuft et al(2011) The material was then rinsed stained with 02 g Lndash1

gentian violet and mounted onto a slide using corn syrup Of the108 scat samples collected 47 were selected for analysis tomaximise the representation of the sexes (20 male 28 female)sites (25 Site 1 11 Site 2 12 Site 3) and seasons (24 wet season24 dry season) To prepare scat samples for analysis sampleswere broken up loosely and also soaked in 42 g Lndash1 sodiumhypochlorite for 24 h or until 90 of the material was visiblybleached The material was rinsed in water through a 250-mmsieve whole seeds were removed identified and countedRemaining leafmaterialwas stained using02 g Lndash1 gentian violetandmounted on a slide using corn syrup For each scat specimen100 identifiable fragments of plant material were counted intransects across the slides Unidentifiable fragments (where theepidermal layer was not visible) were uncommon and were notincluded in counts

Movement analysis

GPS points were first screened to remove erroneous fixescaused byGPS error These included location points found on theopposite side of the impassable Charnley River (at Sites 1 and 2see Fig 2) or that displayed all of the following characteristicsdistance from the last fix gt300m and turning angle (in relation tothe previous and the followingfix)gt180 (Bjoslashrneraas et al 2010)The combination of these characteristics indicated implausiblespikes in which an individual appeared to deviate at high speedfrom his or her trajectory to a distant location and then return tothe original trajectory Horizontal dilution of precision (HDOP)values were not used to filter fixes as data from two stationary testGPS collars indicated that there was no relationship betweendistance to known location and HDOP value

Home ranges were estimated by minimum convex polygons(MCP) and fixed kernel density estimation methods in theprogram Geospatial Modelling Environment (Beyer 2012) and

Table 1 Proportions of habitat types available at three sites in the Artesian Range north-west Kimberley Western AustraliaAn asterisk indicates that last fire is unknown but gt24 months preceding commencement of study period

Habitat type Last fire Age Site 1 Site 2 Site 3(months) Area

(ha)Proportion

()Area(ha)

Proportion()

Area(ha)

Proportion()

Boulder scree 60+ 669 383 359 203 000 000Rainforest 60+ 960 549 361 204 938 505River edge 60+ 667 382 558 315 000 000Triodia spp savanna 24+ 617 353 000 000 615 331Hibiscus keneallyi savanna Nov 2012 1ndash12 14555 8333 16448 9078 1081 582Sorghum stipoideum savanna Oct 2013 3 000 000 000 000 8923 4804Chrysopogon fallax savanna Oct 2013 3 000 000 000 000 1137 612Heteropogon contortus savanna Oct 2013 3 000 000 000 000 5882 3166

D Australian Journal of Zoology R Hohnen et al

interpreted in ArcMap 101 (ESRI) Kernel density estimateswere calculated with a plug-in bandwidth The 95 kernel (K95)utilisation contour estimated total home range and the 50(K50) kernel utilisation contour estimated the core home rangeAll points (both foraging and denning) were included in theMCPand kernel estimates and all areas that the animals could not use(including the river and land on the opposite side of the river)were removed from further consideration Home-range estimateswere tested for normality using the ShapirondashWilk W test andnormalised by log-transformation Differences in home-rangesize between sex and season were examined using generalisedlinear models with a model-averaging approach using thepackage lme4 and MuMIn in the statistical program R ver 303(R Development Core Team 2005 Barton 2013 Bates et al2014)

We calculated the total percentage home-range overlap ofK95 and core K50 home-range estimates for all neighbouringindividuals Home-range overlap was also compared using PHR(Percentile home range) and UDOI (Utilisation distributionoverlap index) indices which consider heterogeneity of spaceuse within home ranges (Fieberg and Kochanny 2005) (seeAppendix 2 for more detail) Differences in overlap extent werecompared between dyads using the asymptotic KruskallndashWallistest as datasets did not meet assumptions of normality

We used a discrete choice modelling approach to describestep selection of Wyulda during foraging periods Coefficientswere estimated from case-control logistic regression modelsusing the lsquoclogitrsquo command in the package survival (Therneau2014) within the statistical program R ver 303 This regressionmethod accounts for differing availabilities of habitats betweenobservations by matching every known animal location to achoice set of randomly generated locations To generate randompoints we first calculated the average distance travelled betweenfixes taken at 1-h 2-h or 3-h (and so on) intervals Then forexample if between fixA andB there was a gap of 2 h five pointswere randomly generated within a circle (with A at its centroid)that had a radius consisting of the average distance travelled byWyulda in 2 h clipped by the individualrsquos 95 kernel home-range estimate Those five points represent locations where theanimal could have gone and were compared in the modellingprocess to point B the location where the animal did go

For all points where Wyulda were found we recorded habitattype and rock complexity (Appendix 3) The variable lsquorockcomplexityrsquo comprised four categories cliffs high complexity(boulder size gt2m) medium complexity (boulder size lt2m) andlow complexity (no exposed rock) Habitat and rockiness mapsfor each site were created in ArcMap 101 using geo-referencedsatellite images from Google Earth (Google Inc 2009) Othervariables investigatedwere distance topermanentwater (referredto as lsquodistance to waterrsquo) number of habitat types within a20-m radius (lsquohabitat heterogeneityrsquo) distance to unburnt habitat(lsquodistance to unburnt patchrsquo) fire (lsquotime since firersquo) andthe distance from the point to the centre of the home range(lsquodistance to home range centroidrsquo) We chose 20m as the radiusof the circle within which to calculate habitat heterogeneity aswe aimed to take into accountGPS error aswell as an individualrsquosperception of where it is relative to the habitats around itWe tested for correlation and redundancy between explanatoryvariables by identifying correlation coefficients of gt05

(Fox 2002) The variables lsquofirersquo lsquodistance towaterrsquo and lsquodistanceto unburnt patchrsquo were strongly correlated with one another solsquodistance to waterrsquo and lsquodistance to unburnt patchrsquo were omittedfrom further analysis Numerical variables including habitatheterogeneity time sincefire anddistance tohome rangecentroidwere standardised (mean = 0 standard deviation = 1) beforeanalysis

We created 31 candidatemodels based on combinations of thevariables lsquohabitat typersquo lsquotime since firersquo lsquohabitat heterogeneityrsquolsquorock complexityrsquo and lsquodistance to home range centroidrsquo withlsquoindividualrsquo considered as a random effect Candidate modelswere ranked on the basis of the Akaikersquos information criterionadjusted for small sample sizes (AICc) The selection forlandscape characteristics by Wyulda was measured by thecoefficient estimates and the odds ratio which representsthe relative change in the odds of selection for each unit of thepredictor variableWe usedmodel averaging to estimate variablecoefficients and their weights according to Anderson (2008)

Den site selection

To explore den site selection by Wyulda we used Bayesianhierarchical logistic regression models For each individual wecompared known den sites with an equal number of randomlygenerated den sites within that individualrsquos home range Modelscomprised a binomial response variable (that described if a pointwas real (1) or generated (0)) and all combinations of theaforementioned predictor variables including habitat type rockcomplexity time since fire habitat heterogeneity and distance tohome range centroid To account for variation in individual useof habitats lsquoindividualrsquo was included as a random effect in allmodels (See Appendix 4 for more details) A Markov ChainMonte Carlo (MCMC) procedure was used to estimate posteriordistributions using the package MCMCglmm (Hadfield 2010) inR ver 303 MCMC chains were run with randomised startingvalues aburn-inof 50 000 iterations a thinning interval of 50 anda total of 200 000 iterations Convergence of MCMC chains wasevaluated following Gelman and Hill (2006) The DevianceInformation Criterion (DIC) was used to compare all possiblemodels and calculate model weightsWe compared all subsets ofthe most complex model We used model averaging to estimatevariable coefficients posterior credible intervals and variableweights according to Anderson (2008)

Results

Between January 2013 and January 2014 collars were deployedfor one month each on 34 occasions on 23Wyulda (15 males and19 females) Of those deployments 29 were GPS collars and5 were VHF giving a total of 3314 location points Nineindividuals were tracked on more than one occasion and theirhome ranges were calculated separately for each tracking period(Appendix 5) The number of individuals known to be alive ateach site varied between 5 and 14 (Appendix 6)

Diet

From the scat samples 4700 fragments of plant material wereidentified Fragments of leaf came from 42 species fruit from 14species and flower from one species (Appendix 7) The mostcommon items were Ficus platypoda fruit (on average 47 of

Habitat use by Wyulda squamicaudata Australian Journal of Zoology E

fragments per sample) Owenia vernicosa fruit (10) Vitexacuminata fruit (10) Grewia breviflora fruit (8)V acuminata leaf (3) Strychnos lucida fruit (2)O vernicosaleaf (2) Cryptocarya cunninghamii leaf (2) and Acaciadimorpha leaf (1) Whole seeds were also frequently found inscats the most common being F platypoda (on average 55 ofseeds per sample) followed by V acuminata (15) UnknownSpecies 3 (6)UnknownSpecies 2 (4) andG breviflora (4)F brachypoda Unknown Species 1 Trachymene dendrothrixand Cryptocarya cunninghamii seeds were also present in lownumbers Plant species that took up on average a largerpercentage composition per sample were also likely to be foundin a greater number of samples (Appendix 7)

Home range

Estimated home-range sizewas not correlatedwith the number offixes per individual (Appendix 8) Mean home-range size was817 134 ha (MCP) or 802 129 ha (K95) and mean corehome-range size (K50) was 214 038 ha (Table 2) Meanhome-range size formaleswas 932 210 (MCP) or 981 233(K95) and for females was 725 176 (MCP) or 661 137(K95) In the sets ofmodels describing both variation inMCPandK95 home-range estimates with sex and season multiple modelswere competitive with one another (Appendix 9) For the MCPmodels the confidence intervals of the model-averagedcoefficient estimates overlapped zero for the variable lsquosexrsquo butdid not overlap zero for the variable lsquoseasonrsquo The coefficientestimate of the variable lsquodry seasonrsquowas positive For the K95models both thevariables lsquosexrsquo and lsquoseasonrsquohadmodel-averagedcoefficient estimates and confidence intervals that overlappedzero (Appendix 10)

Home-range overlap

Home-range overlap occurred frequently between neighbouringWyulda (Table 3) Extent of total home-range overlap differedbetween dyad types for the static K95 contour PHR andUDOIindices of total home-range overlap (Table 4) Total home-rangeoverlapoccurredmost frequentlybetween individuals of differentsexes (Appendix 11) and least frequently between neighbouringmales (Table 3) The extent of core home-range overlap alsodiffered between sexdyads andwasmost commonbetweenmalesand females (Fig 4 Appendix 11)

Foraging site selection

Two models (model weights of 30 and 16) were withintwo DAIC of the top model (model weight of 41 Table 5)The variables lsquohabitat typersquo lsquorock complexityrsquo lsquohabitatheterogeneityrsquo and lsquodistance to home range centroidrsquo all had highvariable importance weights of gt99 and the variable timesince fire had a low importance weight of 27 Model-averagedcoefficient estimates suggested that Wyulda selected againstboth lowandmedium levels of rockycomplexity (Fig 5Table 6)Weak but positive selection was detected for locations at greaterdistances from the centre of an individualrsquos home range Wyuldaselected areas of high habitat heterogeneity while foraging andused long unburnt habitats such as rainforest and savannadominated by Triodia spp as well as recently burnt savanna(1ndash12 months after fire) dominated by Hibiscus kenealleyi andSorghum stipodeum in preference to recently burntHeteropogoncontortusndashdominated savanna There was no selection for oragainst river-edge habitat types or in relation to the variable lsquotimesince firersquo

Den site selection

Individuals were tracked to a total of 736 den sites all of whichwere in rock features 548 in piles of rocky scree 108 in cracks incliffs 57 in vertical clefts in large rocky slabs and 23 in massiverocky outcrops Of the models built on the den site data only fivewere competitive (within 2 DDIC) with the top model (Table 7)Importance weights were highest for the variables lsquohabitat typersquoand lsquorock complexityrsquo (92 and 83 respectively) followed bylsquodistance to home range centroidrsquo (68) lsquohabitat heterogeneityrsquo(59) and lsquotime since firersquo (36) Model-averaged coefficientestimates suggest that Wyulda selected against denning intopographically simple habitats and at long distances from theirhome range centroids (Fig 6)Wyulda selected to den inHibiscuskeneallyi and Sorghum stipodeumndashdominated savanna habitatsboth 1ndash12 months after fire and against denning in long-unburnt

Table 2 The mean home-range size ofW squamicaudata across sexessites and seasons

MCPplusmn se(ha)

K95plusmn se(ha)

K50plusmn se(ha)

Female (n= 19) 725 plusmn 176 661 plusmn 137 194 plusmn 046Male (n= 15) 932 plusmn 210 981 plusmn 233 239 plusmn 065Dry season (n= 21) 1018 plusmn 199 955 plusmn 196 213 plusmn 050Wet season (n= 13) 492 plusmn 092 555 plusmn 090 214 plusmn 061Site 1 (n= 13) 725 plusmn 107 639 plusmn 059 125 plusmn 012Site 2 (n= 8) 1433 plusmn 491 139 plusmn 483 461 plusmn 123Site 3 (n= 13) 528 plusmn 068 603 plusmn 094 150 plusmn 026GPS (n= 29) 874 plusmn 155 857 plusmn 148 191 plusmn 038VHF (n= 5) 486 plusmn 095 486 plusmn 100 347 plusmn 145Total (n= 34) 817 plusmn 134 802 plusmn 129 214 plusmn 038

Table3 Frequencyandaveragehome-rangeoverlapbetweenadjacentWsquamicaudatamonitoredconcurrently at three sites in theArtesianRangenorth-west Kimberley Western Australia (mean se) using two-dimensional (K95and K50) and three-dimensional (PRH and UDOI) indices

Tnd=Total no of neighbouring dyads Tndo =Total no of neighbouring dyads that overlapped F female M male

Sex Tnd Tndo Tndo Mean total overlap Mean core overlapdyad (K95) (K50) K95 PRH UDOI K50 PRH UDOI

FF 24 22 6 2019 plusmn 456 027 plusmn 006 011 plusmn 004 437 plusmn 239 006 plusmn 003 005 plusmn 002FM 37 34 16 3700 plusmn 604 048 plusmn 007 033 plusmn 007 2068 plusmn 545 021 plusmn 005 013 plusmn 004MF 37 34 16 2424 plusmn 405 037 plusmn 006 033 plusmn 007 1199 plusmn 319 011 plusmn 002 007 plusmn 002MM 26 22 6 1210 plusmn 302 020 plusmn 004 004 plusmn 001 071 plusmn 032 002 plusmn 001 001 plusmn 000

F Australian Journal of Zoology R Hohnen et al

rainforest and river-edge habitats Credible intervals of thevariable lsquoTriodia spp savannarsquo included zero suggesting neutralselection Wyulda also appeared to select for older post-fire agesof all habitat types but this variable had a low importanceweight

Discussion

This study sought to test utilisation of complex landscapes andsurrounding areas by a declining mammal in the presence ofintensified fire regimes Rock complexity appears to be a keyhabitat feature selected for by Wyulda Individuals avoidedforaging in habitats with low rock complexity and consistently

denned within rock structures Wyulda preferred to forage inareas of high habitat heterogeneity using both habitats that wererecently burnt (1ndash2 months after fire) habitats that were longunburnt (gt24months after fire) and habitats that very rarely burn(boulder scree river edge and rainforest) Within these habitatsWyulda were highly frugivorous with on average 77 of plantmaterial fragments per scat sample identified as fruit epidermallayers Rock structure may contribute to protecting habitatheterogeneity and the availability of fruiting tree species fromextensive hot fires that can homogenise habitats and kill trees(Williams et al 1999 Vigilante and Bowman 2004a) Overallour results demonstrate diverse uses of heterogeneous habitat byWyulda and emphasise the critical importance of these habitats inthe conservation of small mammals in northern Australia

Previous research has suggested that rocky complexity is a keyhabitat feature forWyulda as the specieswas detected at siteswithrock and rainforest elements (Humphreys et al 1984 Bradleyet al 1987 Doody et al 2012) both for foraging and as den sites(Runcie 1999)Wyuldarsquos preference for rocky habitats could be aby-product of their denning requirements however it is likelythat other factors are involvedRockystructuremayprovide coverfrom predators such as cats which appear most adept at huntingin more open habitats (McGregor et al 2015) particularly after

Table 4 Asymptotic KruskalndashWallis tests of home-range overlapbetween sex dyads using two-dimensional (K95 and K50) and

three-dimensional (PRH and UDOI) indicesValues shown in bold are significant at Plt 005

Total overlap (95) Core overlap (50)c2 P c2 P

K 8272 0041 7863 0049PRH 9099 0028 6921 0074UDOI 8485 0037 4125 0248

04Kilometres

(a) (b) (c)

(d) (e) (f )

(g) (h)

N

Fig 4 Core home-range overlap (K50) between all male (dashed lines) and female (solid lines)Wsquamicaudata tracked in theArtesianRange north-westKimberleyWesternAustralia (a) Site1 January2013wet season (b) Site 1 April 2013 dry season (c) Site 1 July 2013 dry season (d) Site 1 December 2013wet season (e) Site 2 February 2013 wet season (f) Site 2 August 2013 dry season (g) Site 3 May 2013 dryseason (h) Site 3 January 2014 wet season Note that different individuals during each month-long trackingperiod can be distinguished by differing line weight or dash interval

Habitat use by Wyulda squamicaudata Australian Journal of Zoology G

understorey vegetation has been simplified by intense fire orgrazing (McGregor et al 2014) Reduced hunting success inrocky habitats may explain why cat density in this environment isvery low (Hohnen et al 2016) Also at a slightly larger spatialscale rocky landscapes tend to burn more patchily than opensavannas thus retaining more heterogeneity in vegetation age(Price et al 2003) In this study Wyulda selected for high levelsof habitat heterogeneity at both den and foraging sites and usedall ages of post-fire savannas as long as rocky features werepresent Habitat heterogeneity appears to be important for severalvertebrate species in northern Australia (Bolton and Latz 1978Ingleby and Westoby 1992 Pardon et al 2003 Southgate et al2007) and elsewhere (Doumas and Koprowski 2013)

While Wyulda are capable of consuming a wide range ofplant species fruit is clearly a key component of their diet Thisadds to information from a previous radio-tracking study that

identified Wyulda foraging within Xanthostemon paradoxusXanthostemon eucalyptoides and Planchonia careya (Runcie1999) all fruit-bearing species Fruit also makes an importantcontribution to the diet for other arboreal marsupials native tonorth-western Australia (Kerle 1985) Intense and frequent firescan kill or delay the fruiting of tree species (Williams et al 1999Vigilante and Bowman 2004a Atchison et al 2005 Atchison2009) Rocky structures (such as gorges or rock outcrops) canprovide conditions such as shade and water retention that favoursome fruiting species (such as Ficus platypoda) and also protectthem from fire Thus rock structure may contribute to theprotection of fire-sensitive foraging resources used by Wyulda

While the results of this study suggest that Wyulda favourrocky landscapes it is unclear whether the observed relationshipreflects a true habitat preference or is a result of recent reductionsof habitat quality in the open savannas caused by current fireregimes grazing by introduced herbivores and the impacts offeral cats Wyulda have been detected in fauna surveys on rockyescarpments of Charnley River Station separated from sites usedin this studybymore than15 kmof openbasalt plains (KTuft andS Legge unpubl data) On a larger scale Emma Gorge (wherethe species was recently rediscovered in the east Kimberley) isseparated from other historical records in the north Kimberley(eg Mitchell Plateau) by more than 200 km and large stretchesof open plains (Doody et al 2012) The highly disjunct natureof this distribution suggests that there may have been a timewhen the species moved more freely over open landscapesA recent phylogenetic study found that unlike several saxicolousrock wallaby species in the region Wyulda showed no deepstructure between eastern and western populations suggestingrecent colonisation or connectivity across the Kimberley (Potter

Table 5 Top five candidate discrete choice models derived fromforaging data for Wyulda squamicaudata

n1 no of real points included in the model n0 no of generated pointsincluded in the model Abbreviations are as follows hab_type habitat typehet habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K AIC DAIC w

Foraging model (n1 = 2615 n0 = 13071)hab_type+ het+ rock 3 914189 000 041hab_type+ het + rock + distc 4 914254 065 030hab_type+ fire + het + rock 4 914373 184 016hab_type+ fire + het + rock + distc 5 914425 236 013hab_type+ het 2 915147 958 000

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus05 00 05 10

Coefficent estimate

Var

iabl

e

Fig 5 Model-averaged coefficient estimates with 95 credible intervals of discrete choice models describing foraging-siteselection byW squamicaudata in the Artesian Range north-west Kimberley Western Australia The reference habitat typewas Heteropogon contortus savanna and level of rock cover was cliff

H Australian Journal of Zoology R Hohnen et al

et al 2014) Historical records from the east Kimberley suggestthat the species did once persist inmuch drier habitats (Alexander1919)

Wyulda in this study continued to den and forage in habitatsthat had been recently burnt (1ndash12 months after fire) a patternalso observed in several other northern Australian mammalstudies (Hohnen et al 2015) For pale field rats (Rattus tunneyi)in the central Kimberley Leahy et al (2016) found that whilepopulation size decreased after an intense fire individual rats

stayed within their original home ranges and did not move toadjacent unburnt areas Low-intensity fire did not displacenorthern quolls (Dasyurus hallucatus) from their home ranges inthe north Kimberley (Cook 2010) or significantly affect home-range size or location of northern bettongs (Bettongia tropica) innorth-east Queensland (Vernes and Pope 2001) The continueduse of savannahabitats afterfire byWyulda suggests that the costsof shifting home range are greater than any resource bottleneckcaused by recent or frequent fires The minimal home-rangeoverlap between adjacentmales and adjacent females in this studysuggests adegreeof territoriality thatmayalso inhibit home-rangemovement

The average home-range size estimated in this study(817 134 ha (MCP) or 802 129 ha (K95)) was larger thanprevious estimates of 10 ha (Runcie 1999) and a maximum of

Table 6 Odds ratios derived from the model-averaged coefficientestimates from discrete choice models of foraging data for Wyulda

squamicaudata

Variable Foraging dataOdds ratio Robust se

Recently burnt habitatsHibiscus keneallyii savanna 194 029Sarga stipodeum savanna 180 031

Long-unburnt habitatsRainforest 226 028River edge 106 030Long-unburnt savanna 157 029

Rock complexityRock complex 097 015Rock medium 083 012Rock simple 071 019

Other variablesFire 098 005Habitat heterogeneity 112 003Home-range distance 106 008

Table 7 Top five candidate models derived from den data forWyuldasquamicaudata using Bayesian hierarchical logistic regression modelsn1 number of real points included in the model n0 number of generatedpoints included in the model Abbreviations are as follows hab_type habitattype het habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K DIC DDIC w

Den model (n1 = 736 n0 = 736)hab_type + rock + het + distc 4 3008 000 017hab_type + rock + distc 3 3058 050 014hab_type + rock + het 3 3066 058 013hab_type + rock + het +fire + distc 5 3123 115 010hab_type + rock +fire + distc 4 3147 139 009hab_type + het 2 3178 170 007

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus400 minus200 0 200

Coefficent estimate

Var

iabl

e

Fig 6 Model-averaged coefficient estimates with 95 credible intervals of Bayesian hierarchical logistic regressionmodels describing den site selection byW squamicaudata in the Artesian Range north-west KimberleyWestern AustraliaThe reference habitat type was Heteropogon contortus savanna and level of rock cover was cliff

Habitat use by Wyulda squamicaudata Australian Journal of Zoology I

272m long (Humphreys et al 1984) Potentially this reflectsdifferences between the two studies in habitat quality andor thenumber of location points recorded per individual Home-rangeestimates from this study were larger than those of other northKimberley arboreal marsupials including the northern brushtailpossum (Trichosurus vulpecula arhnemensis) (087ndash112 haKerle 1998) and the cohabiting saxicolous rock ringtail possum(Petropseudes dahli) (05ndash12 ha Runcie 2000) While home-range size did not vary between sexes overlap occurred muchmore frequently and extensively between males and femalesthan between individuals of the same sex Several other possumspecies native to northern Australia move and den in familygroups (Kerle 1998 Runcie 2000) Potentially Wyulda alsopractice some degree of sociality and pair bonding

Currently the distribution of Wyulda is restricted to thetopographically complex north Kimberley with a small isolatedpopulation persisting in the east Kimberley While this speciesmay have once been more widely distributed in the centraland east Kimberley including in a wider range of habitatshabitat heterogeneity the availability of fruiting tree species andparticularly the availability of topographic and rock complexitywere key habitat features selected for by Wyulda Rockcomplexity may provide some buffer to Wyulda from threatssuch as intense fires and feral cats offer shelter from predationduring the post-fire period with cliffs and rock outcrops formingnatural barriers to such fires protecting sensitive vegetationand promoting habitat heterogeneity Mammal declines haveoccurred in topographically complex areas of the Arnhem Landplateau (Woinarski 2000 Ziembicki et al 2015) KakaduNational Park (Woinarski et al 2001) and the south-westKimberley (McKenzie 1981 Sawle 1988) Thus substratecomplexity alone does not guarantee mammal species securityManagement strategies that aim to maintain availability ofhabitat heterogeneity and fruiting trees may support thepersistence of Wyulda in rocky habitats Such strategies couldinclude lighting small-scale low-intensity fires early in the dryseason to prevent the incidence and extent of large-scale fires thatoccur late in the dry season

Acknowledgements

This studywas funded by supporters of the AustralianWildlife Conservancyan Australian Research Council Discovery Grant (100100033) and theDepartment of Parks and Wildlife (Western Australia) The study methodswere approved by the University of Tasmania Animal Ethics Committee(Permit no A12516) Fieldwork would not have been possible without thesupport of staff at theAustralianWildlifeConservancyrsquosMorningtonWildlifeSanctuary Thanks go in particular to Alex Hartshorne Hannah Cliff JillianSmith Joel Murray David James Peter Richardson Kaely Kreger DavinaBright Tom Crawford and Iris Bleach

References

Alexander W B (1919) A new species of marsupial of the subfamilyPhalangerinae Journal of the Royal Society of Western Australia 431ndash36

Andersen A N Braithwaite R W Cook G D Corbett L K WilliamsR J Douglas M M Gill A M Setterfield S A and Muller W J(1998) Fire research for conservation management in tropical savannasintroducing the Kapalga fire experiment Australian Journal of Ecology23 95ndash110 doi101111j1442-99931998tb00708x

Andersen A N Cook G D Corbett L K Douglas M M Eager R WRussell-Smith J SetterfieldSAWilliamsR J andWoinarski JCZ(2005) Fire frequency and biodiversity conservation in Australiantropical savannas implications from the Kapalga fire experiment AustralEcology 30 155ndash167 doi101111j1442-9993200501441x

Anderson D R (2008) lsquoModel Based Inferences in the Life Sciencesrsquo(Springer Science and Business Media New York)

Atchison J (2009) Human impacts on Persoonia falcata Perspectives onpost-contact vegetation change in the Keep River region Australia fromcontemporary vegetation surveysVegetationHistoryandArchaeobotany18 147ndash157 doi101007s00334-008-0198-y

Atchison J Head L and Fullagar R (2005) Archaeobotany of fruit seedprocessing in a monsoon savanna environment evidence from the KeepRiver region Northern Territory Australia Journal of ArchaeologicalScience 32 167ndash181 doi101016jjas200403022

Banks S C Dujardin M McBurney L Blair D Barker M andLindenmayer D B (2011) Starting points for small mammal populationrecovery after wildfire recolonisation or residual populationsOikos 12026ndash37 doi101111j1600-0706201018765x

Barton K (2013)MuMInmulti-model inference R package version 1913Bates D Maechler M Bolker B and Walker S (2014) lme4 linear

mixed-effects models using Eigen and S4 R package version 11-7Beyer H L (2012) Geospatial modelling environment Spatial Ecology

LLCBjoslashrneraas K Van Moorter B Rolandsen C M and Herfindal I (2010)

Screening global positioning system location data for errors usinganimal movement characteristics Journal of Wildlife Management 741361ndash1366 doi1021932009-405

Bolton B and Latz P (1978) The western hare-wallaby Lagorchesteshirsutus (Gould) (Macropodidae) in the Tanami Desert WildlifeResearch 5 285ndash293 doi101071WR9780285

BradleyAKemper CKitchenerDHumphreysW andHowR (1987)Small mammals of the Mitchell Plateau region Kimberley WesternAustralia Wildlife Research 14 397ndash413 doi101071WR9870397

Burbidge A A andMcKenzie N L (1989) Patterns in the modern declineof Western Australian vertebrate fauna ndash causes and conservationimplications Biological Conservation 50 143ndash198 doi1010160006-3207(89)90009-8

Burbidge A A and Webb M J (2008) Scaly-tailed possum In lsquoTheMammals of Australiarsquo 3rd edn (Eds S Van Dyck and R Strahan)pp 277ndash278 (New Holland Publishers Sydney)

Bureau of Meteorology (2015) Climate data online Commonwealth ofAustralia Bureau of Meteorology

Cook A C (2010) Habitat use and home range of the northern quollDasyurus hallucatus effects of fire MSc thesis The University ofWestern Australia Perth

Corbett L C Andersen A N and Muumlller W J (2003) Terrestrialvertebrates In lsquoFire inTropical SavannasTheKapalgaExperimentrsquo (EdsA N Andersen G D Cook and R J Williams) pp 126ndash52 (SpringerVerlag New York)

Doody J S Rhind D Castellano CM and BassM (2012) Rediscoveryof the scaly-tailed possum (Wyulda squamicaudata) in the easternKimberleyAustralianMammalogy 34 260ndash262 doi101071AM11039

Doumas S L and Koprowski J L (2013) Effect of heterogeneity inburn severity on Mexican fox squirrels following the return of fireInternational Journal of Wildland Fire 22 405ndash413 doi101071WF12046

Driscoll D A Lindenmayer D B Bennett A F Bode M BradstockR A CaryG J ClarkeM F Dexter N FenshamR Friend G GillM James S Kay G Keith D A MacGregor C Russell-Smith JSalt D Watson J E M Williams R J and York A (2010) Firemanagement for biodiversity conservation key research questions andour capacity to answer them Biological Conservation 143 1928ndash1939doi101016jbiocon201005026

J Australian Journal of Zoology R Hohnen et al

interpreted in ArcMap 101 (ESRI) Kernel density estimateswere calculated with a plug-in bandwidth The 95 kernel (K95)utilisation contour estimated total home range and the 50(K50) kernel utilisation contour estimated the core home rangeAll points (both foraging and denning) were included in theMCPand kernel estimates and all areas that the animals could not use(including the river and land on the opposite side of the river)were removed from further consideration Home-range estimateswere tested for normality using the ShapirondashWilk W test andnormalised by log-transformation Differences in home-rangesize between sex and season were examined using generalisedlinear models with a model-averaging approach using thepackage lme4 and MuMIn in the statistical program R ver 303(R Development Core Team 2005 Barton 2013 Bates et al2014)

We calculated the total percentage home-range overlap ofK95 and core K50 home-range estimates for all neighbouringindividuals Home-range overlap was also compared using PHR(Percentile home range) and UDOI (Utilisation distributionoverlap index) indices which consider heterogeneity of spaceuse within home ranges (Fieberg and Kochanny 2005) (seeAppendix 2 for more detail) Differences in overlap extent werecompared between dyads using the asymptotic KruskallndashWallistest as datasets did not meet assumptions of normality

We used a discrete choice modelling approach to describestep selection of Wyulda during foraging periods Coefficientswere estimated from case-control logistic regression modelsusing the lsquoclogitrsquo command in the package survival (Therneau2014) within the statistical program R ver 303 This regressionmethod accounts for differing availabilities of habitats betweenobservations by matching every known animal location to achoice set of randomly generated locations To generate randompoints we first calculated the average distance travelled betweenfixes taken at 1-h 2-h or 3-h (and so on) intervals Then forexample if between fixA andB there was a gap of 2 h five pointswere randomly generated within a circle (with A at its centroid)that had a radius consisting of the average distance travelled byWyulda in 2 h clipped by the individualrsquos 95 kernel home-range estimate Those five points represent locations where theanimal could have gone and were compared in the modellingprocess to point B the location where the animal did go

For all points where Wyulda were found we recorded habitattype and rock complexity (Appendix 3) The variable lsquorockcomplexityrsquo comprised four categories cliffs high complexity(boulder size gt2m) medium complexity (boulder size lt2m) andlow complexity (no exposed rock) Habitat and rockiness mapsfor each site were created in ArcMap 101 using geo-referencedsatellite images from Google Earth (Google Inc 2009) Othervariables investigatedwere distance topermanentwater (referredto as lsquodistance to waterrsquo) number of habitat types within a20-m radius (lsquohabitat heterogeneityrsquo) distance to unburnt habitat(lsquodistance to unburnt patchrsquo) fire (lsquotime since firersquo) andthe distance from the point to the centre of the home range(lsquodistance to home range centroidrsquo) We chose 20m as the radiusof the circle within which to calculate habitat heterogeneity aswe aimed to take into accountGPS error aswell as an individualrsquosperception of where it is relative to the habitats around itWe tested for correlation and redundancy between explanatoryvariables by identifying correlation coefficients of gt05

(Fox 2002) The variables lsquofirersquo lsquodistance towaterrsquo and lsquodistanceto unburnt patchrsquo were strongly correlated with one another solsquodistance to waterrsquo and lsquodistance to unburnt patchrsquo were omittedfrom further analysis Numerical variables including habitatheterogeneity time sincefire anddistance tohome rangecentroidwere standardised (mean = 0 standard deviation = 1) beforeanalysis

We created 31 candidatemodels based on combinations of thevariables lsquohabitat typersquo lsquotime since firersquo lsquohabitat heterogeneityrsquolsquorock complexityrsquo and lsquodistance to home range centroidrsquo withlsquoindividualrsquo considered as a random effect Candidate modelswere ranked on the basis of the Akaikersquos information criterionadjusted for small sample sizes (AICc) The selection forlandscape characteristics by Wyulda was measured by thecoefficient estimates and the odds ratio which representsthe relative change in the odds of selection for each unit of thepredictor variableWe usedmodel averaging to estimate variablecoefficients and their weights according to Anderson (2008)

Den site selection

To explore den site selection by Wyulda we used Bayesianhierarchical logistic regression models For each individual wecompared known den sites with an equal number of randomlygenerated den sites within that individualrsquos home range Modelscomprised a binomial response variable (that described if a pointwas real (1) or generated (0)) and all combinations of theaforementioned predictor variables including habitat type rockcomplexity time since fire habitat heterogeneity and distance tohome range centroid To account for variation in individual useof habitats lsquoindividualrsquo was included as a random effect in allmodels (See Appendix 4 for more details) A Markov ChainMonte Carlo (MCMC) procedure was used to estimate posteriordistributions using the package MCMCglmm (Hadfield 2010) inR ver 303 MCMC chains were run with randomised startingvalues aburn-inof 50 000 iterations a thinning interval of 50 anda total of 200 000 iterations Convergence of MCMC chains wasevaluated following Gelman and Hill (2006) The DevianceInformation Criterion (DIC) was used to compare all possiblemodels and calculate model weightsWe compared all subsets ofthe most complex model We used model averaging to estimatevariable coefficients posterior credible intervals and variableweights according to Anderson (2008)

Results

Between January 2013 and January 2014 collars were deployedfor one month each on 34 occasions on 23Wyulda (15 males and19 females) Of those deployments 29 were GPS collars and5 were VHF giving a total of 3314 location points Nineindividuals were tracked on more than one occasion and theirhome ranges were calculated separately for each tracking period(Appendix 5) The number of individuals known to be alive ateach site varied between 5 and 14 (Appendix 6)

Diet

From the scat samples 4700 fragments of plant material wereidentified Fragments of leaf came from 42 species fruit from 14species and flower from one species (Appendix 7) The mostcommon items were Ficus platypoda fruit (on average 47 of

Habitat use by Wyulda squamicaudata Australian Journal of Zoology E

fragments per sample) Owenia vernicosa fruit (10) Vitexacuminata fruit (10) Grewia breviflora fruit (8)V acuminata leaf (3) Strychnos lucida fruit (2)O vernicosaleaf (2) Cryptocarya cunninghamii leaf (2) and Acaciadimorpha leaf (1) Whole seeds were also frequently found inscats the most common being F platypoda (on average 55 ofseeds per sample) followed by V acuminata (15) UnknownSpecies 3 (6)UnknownSpecies 2 (4) andG breviflora (4)F brachypoda Unknown Species 1 Trachymene dendrothrixand Cryptocarya cunninghamii seeds were also present in lownumbers Plant species that took up on average a largerpercentage composition per sample were also likely to be foundin a greater number of samples (Appendix 7)

Home range

Estimated home-range sizewas not correlatedwith the number offixes per individual (Appendix 8) Mean home-range size was817 134 ha (MCP) or 802 129 ha (K95) and mean corehome-range size (K50) was 214 038 ha (Table 2) Meanhome-range size formaleswas 932 210 (MCP) or 981 233(K95) and for females was 725 176 (MCP) or 661 137(K95) In the sets ofmodels describing both variation inMCPandK95 home-range estimates with sex and season multiple modelswere competitive with one another (Appendix 9) For the MCPmodels the confidence intervals of the model-averagedcoefficient estimates overlapped zero for the variable lsquosexrsquo butdid not overlap zero for the variable lsquoseasonrsquo The coefficientestimate of the variable lsquodry seasonrsquowas positive For the K95models both thevariables lsquosexrsquo and lsquoseasonrsquohadmodel-averagedcoefficient estimates and confidence intervals that overlappedzero (Appendix 10)

Home-range overlap

Home-range overlap occurred frequently between neighbouringWyulda (Table 3) Extent of total home-range overlap differedbetween dyad types for the static K95 contour PHR andUDOIindices of total home-range overlap (Table 4) Total home-rangeoverlapoccurredmost frequentlybetween individuals of differentsexes (Appendix 11) and least frequently between neighbouringmales (Table 3) The extent of core home-range overlap alsodiffered between sexdyads andwasmost commonbetweenmalesand females (Fig 4 Appendix 11)

Foraging site selection

Two models (model weights of 30 and 16) were withintwo DAIC of the top model (model weight of 41 Table 5)The variables lsquohabitat typersquo lsquorock complexityrsquo lsquohabitatheterogeneityrsquo and lsquodistance to home range centroidrsquo all had highvariable importance weights of gt99 and the variable timesince fire had a low importance weight of 27 Model-averagedcoefficient estimates suggested that Wyulda selected againstboth lowandmedium levels of rockycomplexity (Fig 5Table 6)Weak but positive selection was detected for locations at greaterdistances from the centre of an individualrsquos home range Wyuldaselected areas of high habitat heterogeneity while foraging andused long unburnt habitats such as rainforest and savannadominated by Triodia spp as well as recently burnt savanna(1ndash12 months after fire) dominated by Hibiscus kenealleyi andSorghum stipodeum in preference to recently burntHeteropogoncontortusndashdominated savanna There was no selection for oragainst river-edge habitat types or in relation to the variable lsquotimesince firersquo

Den site selection

Individuals were tracked to a total of 736 den sites all of whichwere in rock features 548 in piles of rocky scree 108 in cracks incliffs 57 in vertical clefts in large rocky slabs and 23 in massiverocky outcrops Of the models built on the den site data only fivewere competitive (within 2 DDIC) with the top model (Table 7)Importance weights were highest for the variables lsquohabitat typersquoand lsquorock complexityrsquo (92 and 83 respectively) followed bylsquodistance to home range centroidrsquo (68) lsquohabitat heterogeneityrsquo(59) and lsquotime since firersquo (36) Model-averaged coefficientestimates suggest that Wyulda selected against denning intopographically simple habitats and at long distances from theirhome range centroids (Fig 6)Wyulda selected to den inHibiscuskeneallyi and Sorghum stipodeumndashdominated savanna habitatsboth 1ndash12 months after fire and against denning in long-unburnt

Table 2 The mean home-range size ofW squamicaudata across sexessites and seasons

MCPplusmn se(ha)

K95plusmn se(ha)

K50plusmn se(ha)

Female (n= 19) 725 plusmn 176 661 plusmn 137 194 plusmn 046Male (n= 15) 932 plusmn 210 981 plusmn 233 239 plusmn 065Dry season (n= 21) 1018 plusmn 199 955 plusmn 196 213 plusmn 050Wet season (n= 13) 492 plusmn 092 555 plusmn 090 214 plusmn 061Site 1 (n= 13) 725 plusmn 107 639 plusmn 059 125 plusmn 012Site 2 (n= 8) 1433 plusmn 491 139 plusmn 483 461 plusmn 123Site 3 (n= 13) 528 plusmn 068 603 plusmn 094 150 plusmn 026GPS (n= 29) 874 plusmn 155 857 plusmn 148 191 plusmn 038VHF (n= 5) 486 plusmn 095 486 plusmn 100 347 plusmn 145Total (n= 34) 817 plusmn 134 802 plusmn 129 214 plusmn 038

Table3 Frequencyandaveragehome-rangeoverlapbetweenadjacentWsquamicaudatamonitoredconcurrently at three sites in theArtesianRangenorth-west Kimberley Western Australia (mean se) using two-dimensional (K95and K50) and three-dimensional (PRH and UDOI) indices

Tnd=Total no of neighbouring dyads Tndo =Total no of neighbouring dyads that overlapped F female M male

Sex Tnd Tndo Tndo Mean total overlap Mean core overlapdyad (K95) (K50) K95 PRH UDOI K50 PRH UDOI

FF 24 22 6 2019 plusmn 456 027 plusmn 006 011 plusmn 004 437 plusmn 239 006 plusmn 003 005 plusmn 002FM 37 34 16 3700 plusmn 604 048 plusmn 007 033 plusmn 007 2068 plusmn 545 021 plusmn 005 013 plusmn 004MF 37 34 16 2424 plusmn 405 037 plusmn 006 033 plusmn 007 1199 plusmn 319 011 plusmn 002 007 plusmn 002MM 26 22 6 1210 plusmn 302 020 plusmn 004 004 plusmn 001 071 plusmn 032 002 plusmn 001 001 plusmn 000

F Australian Journal of Zoology R Hohnen et al

rainforest and river-edge habitats Credible intervals of thevariable lsquoTriodia spp savannarsquo included zero suggesting neutralselection Wyulda also appeared to select for older post-fire agesof all habitat types but this variable had a low importanceweight

Discussion

This study sought to test utilisation of complex landscapes andsurrounding areas by a declining mammal in the presence ofintensified fire regimes Rock complexity appears to be a keyhabitat feature selected for by Wyulda Individuals avoidedforaging in habitats with low rock complexity and consistently

denned within rock structures Wyulda preferred to forage inareas of high habitat heterogeneity using both habitats that wererecently burnt (1ndash2 months after fire) habitats that were longunburnt (gt24months after fire) and habitats that very rarely burn(boulder scree river edge and rainforest) Within these habitatsWyulda were highly frugivorous with on average 77 of plantmaterial fragments per scat sample identified as fruit epidermallayers Rock structure may contribute to protecting habitatheterogeneity and the availability of fruiting tree species fromextensive hot fires that can homogenise habitats and kill trees(Williams et al 1999 Vigilante and Bowman 2004a) Overallour results demonstrate diverse uses of heterogeneous habitat byWyulda and emphasise the critical importance of these habitats inthe conservation of small mammals in northern Australia

Previous research has suggested that rocky complexity is a keyhabitat feature forWyulda as the specieswas detected at siteswithrock and rainforest elements (Humphreys et al 1984 Bradleyet al 1987 Doody et al 2012) both for foraging and as den sites(Runcie 1999)Wyuldarsquos preference for rocky habitats could be aby-product of their denning requirements however it is likelythat other factors are involvedRockystructuremayprovide coverfrom predators such as cats which appear most adept at huntingin more open habitats (McGregor et al 2015) particularly after

Table 4 Asymptotic KruskalndashWallis tests of home-range overlapbetween sex dyads using two-dimensional (K95 and K50) and

three-dimensional (PRH and UDOI) indicesValues shown in bold are significant at Plt 005

Total overlap (95) Core overlap (50)c2 P c2 P

K 8272 0041 7863 0049PRH 9099 0028 6921 0074UDOI 8485 0037 4125 0248

04Kilometres

(a) (b) (c)

(d) (e) (f )

(g) (h)

N

Fig 4 Core home-range overlap (K50) between all male (dashed lines) and female (solid lines)Wsquamicaudata tracked in theArtesianRange north-westKimberleyWesternAustralia (a) Site1 January2013wet season (b) Site 1 April 2013 dry season (c) Site 1 July 2013 dry season (d) Site 1 December 2013wet season (e) Site 2 February 2013 wet season (f) Site 2 August 2013 dry season (g) Site 3 May 2013 dryseason (h) Site 3 January 2014 wet season Note that different individuals during each month-long trackingperiod can be distinguished by differing line weight or dash interval

Habitat use by Wyulda squamicaudata Australian Journal of Zoology G

understorey vegetation has been simplified by intense fire orgrazing (McGregor et al 2014) Reduced hunting success inrocky habitats may explain why cat density in this environment isvery low (Hohnen et al 2016) Also at a slightly larger spatialscale rocky landscapes tend to burn more patchily than opensavannas thus retaining more heterogeneity in vegetation age(Price et al 2003) In this study Wyulda selected for high levelsof habitat heterogeneity at both den and foraging sites and usedall ages of post-fire savannas as long as rocky features werepresent Habitat heterogeneity appears to be important for severalvertebrate species in northern Australia (Bolton and Latz 1978Ingleby and Westoby 1992 Pardon et al 2003 Southgate et al2007) and elsewhere (Doumas and Koprowski 2013)

While Wyulda are capable of consuming a wide range ofplant species fruit is clearly a key component of their diet Thisadds to information from a previous radio-tracking study that

identified Wyulda foraging within Xanthostemon paradoxusXanthostemon eucalyptoides and Planchonia careya (Runcie1999) all fruit-bearing species Fruit also makes an importantcontribution to the diet for other arboreal marsupials native tonorth-western Australia (Kerle 1985) Intense and frequent firescan kill or delay the fruiting of tree species (Williams et al 1999Vigilante and Bowman 2004a Atchison et al 2005 Atchison2009) Rocky structures (such as gorges or rock outcrops) canprovide conditions such as shade and water retention that favoursome fruiting species (such as Ficus platypoda) and also protectthem from fire Thus rock structure may contribute to theprotection of fire-sensitive foraging resources used by Wyulda

While the results of this study suggest that Wyulda favourrocky landscapes it is unclear whether the observed relationshipreflects a true habitat preference or is a result of recent reductionsof habitat quality in the open savannas caused by current fireregimes grazing by introduced herbivores and the impacts offeral cats Wyulda have been detected in fauna surveys on rockyescarpments of Charnley River Station separated from sites usedin this studybymore than15 kmof openbasalt plains (KTuft andS Legge unpubl data) On a larger scale Emma Gorge (wherethe species was recently rediscovered in the east Kimberley) isseparated from other historical records in the north Kimberley(eg Mitchell Plateau) by more than 200 km and large stretchesof open plains (Doody et al 2012) The highly disjunct natureof this distribution suggests that there may have been a timewhen the species moved more freely over open landscapesA recent phylogenetic study found that unlike several saxicolousrock wallaby species in the region Wyulda showed no deepstructure between eastern and western populations suggestingrecent colonisation or connectivity across the Kimberley (Potter

Table 5 Top five candidate discrete choice models derived fromforaging data for Wyulda squamicaudata

n1 no of real points included in the model n0 no of generated pointsincluded in the model Abbreviations are as follows hab_type habitat typehet habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K AIC DAIC w

Foraging model (n1 = 2615 n0 = 13071)hab_type+ het+ rock 3 914189 000 041hab_type+ het + rock + distc 4 914254 065 030hab_type+ fire + het + rock 4 914373 184 016hab_type+ fire + het + rock + distc 5 914425 236 013hab_type+ het 2 915147 958 000

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus05 00 05 10

Coefficent estimate

Var

iabl

e

Fig 5 Model-averaged coefficient estimates with 95 credible intervals of discrete choice models describing foraging-siteselection byW squamicaudata in the Artesian Range north-west Kimberley Western Australia The reference habitat typewas Heteropogon contortus savanna and level of rock cover was cliff

H Australian Journal of Zoology R Hohnen et al

et al 2014) Historical records from the east Kimberley suggestthat the species did once persist inmuch drier habitats (Alexander1919)

Wyulda in this study continued to den and forage in habitatsthat had been recently burnt (1ndash12 months after fire) a patternalso observed in several other northern Australian mammalstudies (Hohnen et al 2015) For pale field rats (Rattus tunneyi)in the central Kimberley Leahy et al (2016) found that whilepopulation size decreased after an intense fire individual rats

stayed within their original home ranges and did not move toadjacent unburnt areas Low-intensity fire did not displacenorthern quolls (Dasyurus hallucatus) from their home ranges inthe north Kimberley (Cook 2010) or significantly affect home-range size or location of northern bettongs (Bettongia tropica) innorth-east Queensland (Vernes and Pope 2001) The continueduse of savannahabitats afterfire byWyulda suggests that the costsof shifting home range are greater than any resource bottleneckcaused by recent or frequent fires The minimal home-rangeoverlap between adjacentmales and adjacent females in this studysuggests adegreeof territoriality thatmayalso inhibit home-rangemovement

The average home-range size estimated in this study(817 134 ha (MCP) or 802 129 ha (K95)) was larger thanprevious estimates of 10 ha (Runcie 1999) and a maximum of

Table 6 Odds ratios derived from the model-averaged coefficientestimates from discrete choice models of foraging data for Wyulda

squamicaudata

Variable Foraging dataOdds ratio Robust se

Recently burnt habitatsHibiscus keneallyii savanna 194 029Sarga stipodeum savanna 180 031

Long-unburnt habitatsRainforest 226 028River edge 106 030Long-unburnt savanna 157 029

Rock complexityRock complex 097 015Rock medium 083 012Rock simple 071 019

Other variablesFire 098 005Habitat heterogeneity 112 003Home-range distance 106 008

Table 7 Top five candidate models derived from den data forWyuldasquamicaudata using Bayesian hierarchical logistic regression modelsn1 number of real points included in the model n0 number of generatedpoints included in the model Abbreviations are as follows hab_type habitattype het habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K DIC DDIC w

Den model (n1 = 736 n0 = 736)hab_type + rock + het + distc 4 3008 000 017hab_type + rock + distc 3 3058 050 014hab_type + rock + het 3 3066 058 013hab_type + rock + het +fire + distc 5 3123 115 010hab_type + rock +fire + distc 4 3147 139 009hab_type + het 2 3178 170 007

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus400 minus200 0 200

Coefficent estimate

Var

iabl

e

Fig 6 Model-averaged coefficient estimates with 95 credible intervals of Bayesian hierarchical logistic regressionmodels describing den site selection byW squamicaudata in the Artesian Range north-west KimberleyWestern AustraliaThe reference habitat type was Heteropogon contortus savanna and level of rock cover was cliff

Habitat use by Wyulda squamicaudata Australian Journal of Zoology I

272m long (Humphreys et al 1984) Potentially this reflectsdifferences between the two studies in habitat quality andor thenumber of location points recorded per individual Home-rangeestimates from this study were larger than those of other northKimberley arboreal marsupials including the northern brushtailpossum (Trichosurus vulpecula arhnemensis) (087ndash112 haKerle 1998) and the cohabiting saxicolous rock ringtail possum(Petropseudes dahli) (05ndash12 ha Runcie 2000) While home-range size did not vary between sexes overlap occurred muchmore frequently and extensively between males and femalesthan between individuals of the same sex Several other possumspecies native to northern Australia move and den in familygroups (Kerle 1998 Runcie 2000) Potentially Wyulda alsopractice some degree of sociality and pair bonding

Currently the distribution of Wyulda is restricted to thetopographically complex north Kimberley with a small isolatedpopulation persisting in the east Kimberley While this speciesmay have once been more widely distributed in the centraland east Kimberley including in a wider range of habitatshabitat heterogeneity the availability of fruiting tree species andparticularly the availability of topographic and rock complexitywere key habitat features selected for by Wyulda Rockcomplexity may provide some buffer to Wyulda from threatssuch as intense fires and feral cats offer shelter from predationduring the post-fire period with cliffs and rock outcrops formingnatural barriers to such fires protecting sensitive vegetationand promoting habitat heterogeneity Mammal declines haveoccurred in topographically complex areas of the Arnhem Landplateau (Woinarski 2000 Ziembicki et al 2015) KakaduNational Park (Woinarski et al 2001) and the south-westKimberley (McKenzie 1981 Sawle 1988) Thus substratecomplexity alone does not guarantee mammal species securityManagement strategies that aim to maintain availability ofhabitat heterogeneity and fruiting trees may support thepersistence of Wyulda in rocky habitats Such strategies couldinclude lighting small-scale low-intensity fires early in the dryseason to prevent the incidence and extent of large-scale fires thatoccur late in the dry season

Acknowledgements

This studywas funded by supporters of the AustralianWildlife Conservancyan Australian Research Council Discovery Grant (100100033) and theDepartment of Parks and Wildlife (Western Australia) The study methodswere approved by the University of Tasmania Animal Ethics Committee(Permit no A12516) Fieldwork would not have been possible without thesupport of staff at theAustralianWildlifeConservancyrsquosMorningtonWildlifeSanctuary Thanks go in particular to Alex Hartshorne Hannah Cliff JillianSmith Joel Murray David James Peter Richardson Kaely Kreger DavinaBright Tom Crawford and Iris Bleach

References

Alexander W B (1919) A new species of marsupial of the subfamilyPhalangerinae Journal of the Royal Society of Western Australia 431ndash36

Andersen A N Braithwaite R W Cook G D Corbett L K WilliamsR J Douglas M M Gill A M Setterfield S A and Muller W J(1998) Fire research for conservation management in tropical savannasintroducing the Kapalga fire experiment Australian Journal of Ecology23 95ndash110 doi101111j1442-99931998tb00708x

Andersen A N Cook G D Corbett L K Douglas M M Eager R WRussell-Smith J SetterfieldSAWilliamsR J andWoinarski JCZ(2005) Fire frequency and biodiversity conservation in Australiantropical savannas implications from the Kapalga fire experiment AustralEcology 30 155ndash167 doi101111j1442-9993200501441x

Anderson D R (2008) lsquoModel Based Inferences in the Life Sciencesrsquo(Springer Science and Business Media New York)

Atchison J (2009) Human impacts on Persoonia falcata Perspectives onpost-contact vegetation change in the Keep River region Australia fromcontemporary vegetation surveysVegetationHistoryandArchaeobotany18 147ndash157 doi101007s00334-008-0198-y

Atchison J Head L and Fullagar R (2005) Archaeobotany of fruit seedprocessing in a monsoon savanna environment evidence from the KeepRiver region Northern Territory Australia Journal of ArchaeologicalScience 32 167ndash181 doi101016jjas200403022

Banks S C Dujardin M McBurney L Blair D Barker M andLindenmayer D B (2011) Starting points for small mammal populationrecovery after wildfire recolonisation or residual populationsOikos 12026ndash37 doi101111j1600-0706201018765x

Barton K (2013)MuMInmulti-model inference R package version 1913Bates D Maechler M Bolker B and Walker S (2014) lme4 linear

mixed-effects models using Eigen and S4 R package version 11-7Beyer H L (2012) Geospatial modelling environment Spatial Ecology

LLCBjoslashrneraas K Van Moorter B Rolandsen C M and Herfindal I (2010)

Screening global positioning system location data for errors usinganimal movement characteristics Journal of Wildlife Management 741361ndash1366 doi1021932009-405

Bolton B and Latz P (1978) The western hare-wallaby Lagorchesteshirsutus (Gould) (Macropodidae) in the Tanami Desert WildlifeResearch 5 285ndash293 doi101071WR9780285

BradleyAKemper CKitchenerDHumphreysW andHowR (1987)Small mammals of the Mitchell Plateau region Kimberley WesternAustralia Wildlife Research 14 397ndash413 doi101071WR9870397

Burbidge A A andMcKenzie N L (1989) Patterns in the modern declineof Western Australian vertebrate fauna ndash causes and conservationimplications Biological Conservation 50 143ndash198 doi1010160006-3207(89)90009-8

Burbidge A A and Webb M J (2008) Scaly-tailed possum In lsquoTheMammals of Australiarsquo 3rd edn (Eds S Van Dyck and R Strahan)pp 277ndash278 (New Holland Publishers Sydney)

Bureau of Meteorology (2015) Climate data online Commonwealth ofAustralia Bureau of Meteorology

Cook A C (2010) Habitat use and home range of the northern quollDasyurus hallucatus effects of fire MSc thesis The University ofWestern Australia Perth

Corbett L C Andersen A N and Muumlller W J (2003) Terrestrialvertebrates In lsquoFire inTropical SavannasTheKapalgaExperimentrsquo (EdsA N Andersen G D Cook and R J Williams) pp 126ndash52 (SpringerVerlag New York)

Doody J S Rhind D Castellano CM and BassM (2012) Rediscoveryof the scaly-tailed possum (Wyulda squamicaudata) in the easternKimberleyAustralianMammalogy 34 260ndash262 doi101071AM11039

Doumas S L and Koprowski J L (2013) Effect of heterogeneity inburn severity on Mexican fox squirrels following the return of fireInternational Journal of Wildland Fire 22 405ndash413 doi101071WF12046

Driscoll D A Lindenmayer D B Bennett A F Bode M BradstockR A CaryG J ClarkeM F Dexter N FenshamR Friend G GillM James S Kay G Keith D A MacGregor C Russell-Smith JSalt D Watson J E M Williams R J and York A (2010) Firemanagement for biodiversity conservation key research questions andour capacity to answer them Biological Conservation 143 1928ndash1939doi101016jbiocon201005026

J Australian Journal of Zoology R Hohnen et al

fragments per sample) Owenia vernicosa fruit (10) Vitexacuminata fruit (10) Grewia breviflora fruit (8)V acuminata leaf (3) Strychnos lucida fruit (2)O vernicosaleaf (2) Cryptocarya cunninghamii leaf (2) and Acaciadimorpha leaf (1) Whole seeds were also frequently found inscats the most common being F platypoda (on average 55 ofseeds per sample) followed by V acuminata (15) UnknownSpecies 3 (6)UnknownSpecies 2 (4) andG breviflora (4)F brachypoda Unknown Species 1 Trachymene dendrothrixand Cryptocarya cunninghamii seeds were also present in lownumbers Plant species that took up on average a largerpercentage composition per sample were also likely to be foundin a greater number of samples (Appendix 7)

Home range

Estimated home-range sizewas not correlatedwith the number offixes per individual (Appendix 8) Mean home-range size was817 134 ha (MCP) or 802 129 ha (K95) and mean corehome-range size (K50) was 214 038 ha (Table 2) Meanhome-range size formaleswas 932 210 (MCP) or 981 233(K95) and for females was 725 176 (MCP) or 661 137(K95) In the sets ofmodels describing both variation inMCPandK95 home-range estimates with sex and season multiple modelswere competitive with one another (Appendix 9) For the MCPmodels the confidence intervals of the model-averagedcoefficient estimates overlapped zero for the variable lsquosexrsquo butdid not overlap zero for the variable lsquoseasonrsquo The coefficientestimate of the variable lsquodry seasonrsquowas positive For the K95models both thevariables lsquosexrsquo and lsquoseasonrsquohadmodel-averagedcoefficient estimates and confidence intervals that overlappedzero (Appendix 10)

Home-range overlap

Home-range overlap occurred frequently between neighbouringWyulda (Table 3) Extent of total home-range overlap differedbetween dyad types for the static K95 contour PHR andUDOIindices of total home-range overlap (Table 4) Total home-rangeoverlapoccurredmost frequentlybetween individuals of differentsexes (Appendix 11) and least frequently between neighbouringmales (Table 3) The extent of core home-range overlap alsodiffered between sexdyads andwasmost commonbetweenmalesand females (Fig 4 Appendix 11)

Foraging site selection

Two models (model weights of 30 and 16) were withintwo DAIC of the top model (model weight of 41 Table 5)The variables lsquohabitat typersquo lsquorock complexityrsquo lsquohabitatheterogeneityrsquo and lsquodistance to home range centroidrsquo all had highvariable importance weights of gt99 and the variable timesince fire had a low importance weight of 27 Model-averagedcoefficient estimates suggested that Wyulda selected againstboth lowandmedium levels of rockycomplexity (Fig 5Table 6)Weak but positive selection was detected for locations at greaterdistances from the centre of an individualrsquos home range Wyuldaselected areas of high habitat heterogeneity while foraging andused long unburnt habitats such as rainforest and savannadominated by Triodia spp as well as recently burnt savanna(1ndash12 months after fire) dominated by Hibiscus kenealleyi andSorghum stipodeum in preference to recently burntHeteropogoncontortusndashdominated savanna There was no selection for oragainst river-edge habitat types or in relation to the variable lsquotimesince firersquo

Den site selection

Individuals were tracked to a total of 736 den sites all of whichwere in rock features 548 in piles of rocky scree 108 in cracks incliffs 57 in vertical clefts in large rocky slabs and 23 in massiverocky outcrops Of the models built on the den site data only fivewere competitive (within 2 DDIC) with the top model (Table 7)Importance weights were highest for the variables lsquohabitat typersquoand lsquorock complexityrsquo (92 and 83 respectively) followed bylsquodistance to home range centroidrsquo (68) lsquohabitat heterogeneityrsquo(59) and lsquotime since firersquo (36) Model-averaged coefficientestimates suggest that Wyulda selected against denning intopographically simple habitats and at long distances from theirhome range centroids (Fig 6)Wyulda selected to den inHibiscuskeneallyi and Sorghum stipodeumndashdominated savanna habitatsboth 1ndash12 months after fire and against denning in long-unburnt

Table 2 The mean home-range size ofW squamicaudata across sexessites and seasons

MCPplusmn se(ha)

K95plusmn se(ha)

K50plusmn se(ha)

Female (n= 19) 725 plusmn 176 661 plusmn 137 194 plusmn 046Male (n= 15) 932 plusmn 210 981 plusmn 233 239 plusmn 065Dry season (n= 21) 1018 plusmn 199 955 plusmn 196 213 plusmn 050Wet season (n= 13) 492 plusmn 092 555 plusmn 090 214 plusmn 061Site 1 (n= 13) 725 plusmn 107 639 plusmn 059 125 plusmn 012Site 2 (n= 8) 1433 plusmn 491 139 plusmn 483 461 plusmn 123Site 3 (n= 13) 528 plusmn 068 603 plusmn 094 150 plusmn 026GPS (n= 29) 874 plusmn 155 857 plusmn 148 191 plusmn 038VHF (n= 5) 486 plusmn 095 486 plusmn 100 347 plusmn 145Total (n= 34) 817 plusmn 134 802 plusmn 129 214 plusmn 038

Table3 Frequencyandaveragehome-rangeoverlapbetweenadjacentWsquamicaudatamonitoredconcurrently at three sites in theArtesianRangenorth-west Kimberley Western Australia (mean se) using two-dimensional (K95and K50) and three-dimensional (PRH and UDOI) indices

Tnd=Total no of neighbouring dyads Tndo =Total no of neighbouring dyads that overlapped F female M male

Sex Tnd Tndo Tndo Mean total overlap Mean core overlapdyad (K95) (K50) K95 PRH UDOI K50 PRH UDOI

FF 24 22 6 2019 plusmn 456 027 plusmn 006 011 plusmn 004 437 plusmn 239 006 plusmn 003 005 plusmn 002FM 37 34 16 3700 plusmn 604 048 plusmn 007 033 plusmn 007 2068 plusmn 545 021 plusmn 005 013 plusmn 004MF 37 34 16 2424 plusmn 405 037 plusmn 006 033 plusmn 007 1199 plusmn 319 011 plusmn 002 007 plusmn 002MM 26 22 6 1210 plusmn 302 020 plusmn 004 004 plusmn 001 071 plusmn 032 002 plusmn 001 001 plusmn 000

F Australian Journal of Zoology R Hohnen et al

rainforest and river-edge habitats Credible intervals of thevariable lsquoTriodia spp savannarsquo included zero suggesting neutralselection Wyulda also appeared to select for older post-fire agesof all habitat types but this variable had a low importanceweight

Discussion

This study sought to test utilisation of complex landscapes andsurrounding areas by a declining mammal in the presence ofintensified fire regimes Rock complexity appears to be a keyhabitat feature selected for by Wyulda Individuals avoidedforaging in habitats with low rock complexity and consistently

denned within rock structures Wyulda preferred to forage inareas of high habitat heterogeneity using both habitats that wererecently burnt (1ndash2 months after fire) habitats that were longunburnt (gt24months after fire) and habitats that very rarely burn(boulder scree river edge and rainforest) Within these habitatsWyulda were highly frugivorous with on average 77 of plantmaterial fragments per scat sample identified as fruit epidermallayers Rock structure may contribute to protecting habitatheterogeneity and the availability of fruiting tree species fromextensive hot fires that can homogenise habitats and kill trees(Williams et al 1999 Vigilante and Bowman 2004a) Overallour results demonstrate diverse uses of heterogeneous habitat byWyulda and emphasise the critical importance of these habitats inthe conservation of small mammals in northern Australia

Previous research has suggested that rocky complexity is a keyhabitat feature forWyulda as the specieswas detected at siteswithrock and rainforest elements (Humphreys et al 1984 Bradleyet al 1987 Doody et al 2012) both for foraging and as den sites(Runcie 1999)Wyuldarsquos preference for rocky habitats could be aby-product of their denning requirements however it is likelythat other factors are involvedRockystructuremayprovide coverfrom predators such as cats which appear most adept at huntingin more open habitats (McGregor et al 2015) particularly after

Table 4 Asymptotic KruskalndashWallis tests of home-range overlapbetween sex dyads using two-dimensional (K95 and K50) and

three-dimensional (PRH and UDOI) indicesValues shown in bold are significant at Plt 005

Total overlap (95) Core overlap (50)c2 P c2 P

K 8272 0041 7863 0049PRH 9099 0028 6921 0074UDOI 8485 0037 4125 0248

04Kilometres

(a) (b) (c)

(d) (e) (f )

(g) (h)

N

Fig 4 Core home-range overlap (K50) between all male (dashed lines) and female (solid lines)Wsquamicaudata tracked in theArtesianRange north-westKimberleyWesternAustralia (a) Site1 January2013wet season (b) Site 1 April 2013 dry season (c) Site 1 July 2013 dry season (d) Site 1 December 2013wet season (e) Site 2 February 2013 wet season (f) Site 2 August 2013 dry season (g) Site 3 May 2013 dryseason (h) Site 3 January 2014 wet season Note that different individuals during each month-long trackingperiod can be distinguished by differing line weight or dash interval

Habitat use by Wyulda squamicaudata Australian Journal of Zoology G

understorey vegetation has been simplified by intense fire orgrazing (McGregor et al 2014) Reduced hunting success inrocky habitats may explain why cat density in this environment isvery low (Hohnen et al 2016) Also at a slightly larger spatialscale rocky landscapes tend to burn more patchily than opensavannas thus retaining more heterogeneity in vegetation age(Price et al 2003) In this study Wyulda selected for high levelsof habitat heterogeneity at both den and foraging sites and usedall ages of post-fire savannas as long as rocky features werepresent Habitat heterogeneity appears to be important for severalvertebrate species in northern Australia (Bolton and Latz 1978Ingleby and Westoby 1992 Pardon et al 2003 Southgate et al2007) and elsewhere (Doumas and Koprowski 2013)

While Wyulda are capable of consuming a wide range ofplant species fruit is clearly a key component of their diet Thisadds to information from a previous radio-tracking study that

identified Wyulda foraging within Xanthostemon paradoxusXanthostemon eucalyptoides and Planchonia careya (Runcie1999) all fruit-bearing species Fruit also makes an importantcontribution to the diet for other arboreal marsupials native tonorth-western Australia (Kerle 1985) Intense and frequent firescan kill or delay the fruiting of tree species (Williams et al 1999Vigilante and Bowman 2004a Atchison et al 2005 Atchison2009) Rocky structures (such as gorges or rock outcrops) canprovide conditions such as shade and water retention that favoursome fruiting species (such as Ficus platypoda) and also protectthem from fire Thus rock structure may contribute to theprotection of fire-sensitive foraging resources used by Wyulda

While the results of this study suggest that Wyulda favourrocky landscapes it is unclear whether the observed relationshipreflects a true habitat preference or is a result of recent reductionsof habitat quality in the open savannas caused by current fireregimes grazing by introduced herbivores and the impacts offeral cats Wyulda have been detected in fauna surveys on rockyescarpments of Charnley River Station separated from sites usedin this studybymore than15 kmof openbasalt plains (KTuft andS Legge unpubl data) On a larger scale Emma Gorge (wherethe species was recently rediscovered in the east Kimberley) isseparated from other historical records in the north Kimberley(eg Mitchell Plateau) by more than 200 km and large stretchesof open plains (Doody et al 2012) The highly disjunct natureof this distribution suggests that there may have been a timewhen the species moved more freely over open landscapesA recent phylogenetic study found that unlike several saxicolousrock wallaby species in the region Wyulda showed no deepstructure between eastern and western populations suggestingrecent colonisation or connectivity across the Kimberley (Potter

Table 5 Top five candidate discrete choice models derived fromforaging data for Wyulda squamicaudata

n1 no of real points included in the model n0 no of generated pointsincluded in the model Abbreviations are as follows hab_type habitat typehet habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K AIC DAIC w

Foraging model (n1 = 2615 n0 = 13071)hab_type+ het+ rock 3 914189 000 041hab_type+ het + rock + distc 4 914254 065 030hab_type+ fire + het + rock 4 914373 184 016hab_type+ fire + het + rock + distc 5 914425 236 013hab_type+ het 2 915147 958 000

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus05 00 05 10

Coefficent estimate

Var

iabl

e

Fig 5 Model-averaged coefficient estimates with 95 credible intervals of discrete choice models describing foraging-siteselection byW squamicaudata in the Artesian Range north-west Kimberley Western Australia The reference habitat typewas Heteropogon contortus savanna and level of rock cover was cliff

H Australian Journal of Zoology R Hohnen et al

et al 2014) Historical records from the east Kimberley suggestthat the species did once persist inmuch drier habitats (Alexander1919)

Wyulda in this study continued to den and forage in habitatsthat had been recently burnt (1ndash12 months after fire) a patternalso observed in several other northern Australian mammalstudies (Hohnen et al 2015) For pale field rats (Rattus tunneyi)in the central Kimberley Leahy et al (2016) found that whilepopulation size decreased after an intense fire individual rats

stayed within their original home ranges and did not move toadjacent unburnt areas Low-intensity fire did not displacenorthern quolls (Dasyurus hallucatus) from their home ranges inthe north Kimberley (Cook 2010) or significantly affect home-range size or location of northern bettongs (Bettongia tropica) innorth-east Queensland (Vernes and Pope 2001) The continueduse of savannahabitats afterfire byWyulda suggests that the costsof shifting home range are greater than any resource bottleneckcaused by recent or frequent fires The minimal home-rangeoverlap between adjacentmales and adjacent females in this studysuggests adegreeof territoriality thatmayalso inhibit home-rangemovement

The average home-range size estimated in this study(817 134 ha (MCP) or 802 129 ha (K95)) was larger thanprevious estimates of 10 ha (Runcie 1999) and a maximum of

Table 6 Odds ratios derived from the model-averaged coefficientestimates from discrete choice models of foraging data for Wyulda

squamicaudata

Variable Foraging dataOdds ratio Robust se

Recently burnt habitatsHibiscus keneallyii savanna 194 029Sarga stipodeum savanna 180 031

Long-unburnt habitatsRainforest 226 028River edge 106 030Long-unburnt savanna 157 029

Rock complexityRock complex 097 015Rock medium 083 012Rock simple 071 019

Other variablesFire 098 005Habitat heterogeneity 112 003Home-range distance 106 008

Table 7 Top five candidate models derived from den data forWyuldasquamicaudata using Bayesian hierarchical logistic regression modelsn1 number of real points included in the model n0 number of generatedpoints included in the model Abbreviations are as follows hab_type habitattype het habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K DIC DDIC w

Den model (n1 = 736 n0 = 736)hab_type + rock + het + distc 4 3008 000 017hab_type + rock + distc 3 3058 050 014hab_type + rock + het 3 3066 058 013hab_type + rock + het +fire + distc 5 3123 115 010hab_type + rock +fire + distc 4 3147 139 009hab_type + het 2 3178 170 007

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus400 minus200 0 200

Coefficent estimate

Var

iabl

e

Fig 6 Model-averaged coefficient estimates with 95 credible intervals of Bayesian hierarchical logistic regressionmodels describing den site selection byW squamicaudata in the Artesian Range north-west KimberleyWestern AustraliaThe reference habitat type was Heteropogon contortus savanna and level of rock cover was cliff

Habitat use by Wyulda squamicaudata Australian Journal of Zoology I

272m long (Humphreys et al 1984) Potentially this reflectsdifferences between the two studies in habitat quality andor thenumber of location points recorded per individual Home-rangeestimates from this study were larger than those of other northKimberley arboreal marsupials including the northern brushtailpossum (Trichosurus vulpecula arhnemensis) (087ndash112 haKerle 1998) and the cohabiting saxicolous rock ringtail possum(Petropseudes dahli) (05ndash12 ha Runcie 2000) While home-range size did not vary between sexes overlap occurred muchmore frequently and extensively between males and femalesthan between individuals of the same sex Several other possumspecies native to northern Australia move and den in familygroups (Kerle 1998 Runcie 2000) Potentially Wyulda alsopractice some degree of sociality and pair bonding

Currently the distribution of Wyulda is restricted to thetopographically complex north Kimberley with a small isolatedpopulation persisting in the east Kimberley While this speciesmay have once been more widely distributed in the centraland east Kimberley including in a wider range of habitatshabitat heterogeneity the availability of fruiting tree species andparticularly the availability of topographic and rock complexitywere key habitat features selected for by Wyulda Rockcomplexity may provide some buffer to Wyulda from threatssuch as intense fires and feral cats offer shelter from predationduring the post-fire period with cliffs and rock outcrops formingnatural barriers to such fires protecting sensitive vegetationand promoting habitat heterogeneity Mammal declines haveoccurred in topographically complex areas of the Arnhem Landplateau (Woinarski 2000 Ziembicki et al 2015) KakaduNational Park (Woinarski et al 2001) and the south-westKimberley (McKenzie 1981 Sawle 1988) Thus substratecomplexity alone does not guarantee mammal species securityManagement strategies that aim to maintain availability ofhabitat heterogeneity and fruiting trees may support thepersistence of Wyulda in rocky habitats Such strategies couldinclude lighting small-scale low-intensity fires early in the dryseason to prevent the incidence and extent of large-scale fires thatoccur late in the dry season

Acknowledgements

This studywas funded by supporters of the AustralianWildlife Conservancyan Australian Research Council Discovery Grant (100100033) and theDepartment of Parks and Wildlife (Western Australia) The study methodswere approved by the University of Tasmania Animal Ethics Committee(Permit no A12516) Fieldwork would not have been possible without thesupport of staff at theAustralianWildlifeConservancyrsquosMorningtonWildlifeSanctuary Thanks go in particular to Alex Hartshorne Hannah Cliff JillianSmith Joel Murray David James Peter Richardson Kaely Kreger DavinaBright Tom Crawford and Iris Bleach

References

Alexander W B (1919) A new species of marsupial of the subfamilyPhalangerinae Journal of the Royal Society of Western Australia 431ndash36

Andersen A N Braithwaite R W Cook G D Corbett L K WilliamsR J Douglas M M Gill A M Setterfield S A and Muller W J(1998) Fire research for conservation management in tropical savannasintroducing the Kapalga fire experiment Australian Journal of Ecology23 95ndash110 doi101111j1442-99931998tb00708x

Andersen A N Cook G D Corbett L K Douglas M M Eager R WRussell-Smith J SetterfieldSAWilliamsR J andWoinarski JCZ(2005) Fire frequency and biodiversity conservation in Australiantropical savannas implications from the Kapalga fire experiment AustralEcology 30 155ndash167 doi101111j1442-9993200501441x

Anderson D R (2008) lsquoModel Based Inferences in the Life Sciencesrsquo(Springer Science and Business Media New York)

Atchison J (2009) Human impacts on Persoonia falcata Perspectives onpost-contact vegetation change in the Keep River region Australia fromcontemporary vegetation surveysVegetationHistoryandArchaeobotany18 147ndash157 doi101007s00334-008-0198-y

Atchison J Head L and Fullagar R (2005) Archaeobotany of fruit seedprocessing in a monsoon savanna environment evidence from the KeepRiver region Northern Territory Australia Journal of ArchaeologicalScience 32 167ndash181 doi101016jjas200403022

Banks S C Dujardin M McBurney L Blair D Barker M andLindenmayer D B (2011) Starting points for small mammal populationrecovery after wildfire recolonisation or residual populationsOikos 12026ndash37 doi101111j1600-0706201018765x

Barton K (2013)MuMInmulti-model inference R package version 1913Bates D Maechler M Bolker B and Walker S (2014) lme4 linear

mixed-effects models using Eigen and S4 R package version 11-7Beyer H L (2012) Geospatial modelling environment Spatial Ecology

LLCBjoslashrneraas K Van Moorter B Rolandsen C M and Herfindal I (2010)

Screening global positioning system location data for errors usinganimal movement characteristics Journal of Wildlife Management 741361ndash1366 doi1021932009-405

Bolton B and Latz P (1978) The western hare-wallaby Lagorchesteshirsutus (Gould) (Macropodidae) in the Tanami Desert WildlifeResearch 5 285ndash293 doi101071WR9780285

BradleyAKemper CKitchenerDHumphreysW andHowR (1987)Small mammals of the Mitchell Plateau region Kimberley WesternAustralia Wildlife Research 14 397ndash413 doi101071WR9870397

Burbidge A A andMcKenzie N L (1989) Patterns in the modern declineof Western Australian vertebrate fauna ndash causes and conservationimplications Biological Conservation 50 143ndash198 doi1010160006-3207(89)90009-8

Burbidge A A and Webb M J (2008) Scaly-tailed possum In lsquoTheMammals of Australiarsquo 3rd edn (Eds S Van Dyck and R Strahan)pp 277ndash278 (New Holland Publishers Sydney)

Bureau of Meteorology (2015) Climate data online Commonwealth ofAustralia Bureau of Meteorology

Cook A C (2010) Habitat use and home range of the northern quollDasyurus hallucatus effects of fire MSc thesis The University ofWestern Australia Perth

Corbett L C Andersen A N and Muumlller W J (2003) Terrestrialvertebrates In lsquoFire inTropical SavannasTheKapalgaExperimentrsquo (EdsA N Andersen G D Cook and R J Williams) pp 126ndash52 (SpringerVerlag New York)

Doody J S Rhind D Castellano CM and BassM (2012) Rediscoveryof the scaly-tailed possum (Wyulda squamicaudata) in the easternKimberleyAustralianMammalogy 34 260ndash262 doi101071AM11039

Doumas S L and Koprowski J L (2013) Effect of heterogeneity inburn severity on Mexican fox squirrels following the return of fireInternational Journal of Wildland Fire 22 405ndash413 doi101071WF12046

Driscoll D A Lindenmayer D B Bennett A F Bode M BradstockR A CaryG J ClarkeM F Dexter N FenshamR Friend G GillM James S Kay G Keith D A MacGregor C Russell-Smith JSalt D Watson J E M Williams R J and York A (2010) Firemanagement for biodiversity conservation key research questions andour capacity to answer them Biological Conservation 143 1928ndash1939doi101016jbiocon201005026

J Australian Journal of Zoology R Hohnen et al

rainforest and river-edge habitats Credible intervals of thevariable lsquoTriodia spp savannarsquo included zero suggesting neutralselection Wyulda also appeared to select for older post-fire agesof all habitat types but this variable had a low importanceweight

Discussion

This study sought to test utilisation of complex landscapes andsurrounding areas by a declining mammal in the presence ofintensified fire regimes Rock complexity appears to be a keyhabitat feature selected for by Wyulda Individuals avoidedforaging in habitats with low rock complexity and consistently

denned within rock structures Wyulda preferred to forage inareas of high habitat heterogeneity using both habitats that wererecently burnt (1ndash2 months after fire) habitats that were longunburnt (gt24months after fire) and habitats that very rarely burn(boulder scree river edge and rainforest) Within these habitatsWyulda were highly frugivorous with on average 77 of plantmaterial fragments per scat sample identified as fruit epidermallayers Rock structure may contribute to protecting habitatheterogeneity and the availability of fruiting tree species fromextensive hot fires that can homogenise habitats and kill trees(Williams et al 1999 Vigilante and Bowman 2004a) Overallour results demonstrate diverse uses of heterogeneous habitat byWyulda and emphasise the critical importance of these habitats inthe conservation of small mammals in northern Australia

Previous research has suggested that rocky complexity is a keyhabitat feature forWyulda as the specieswas detected at siteswithrock and rainforest elements (Humphreys et al 1984 Bradleyet al 1987 Doody et al 2012) both for foraging and as den sites(Runcie 1999)Wyuldarsquos preference for rocky habitats could be aby-product of their denning requirements however it is likelythat other factors are involvedRockystructuremayprovide coverfrom predators such as cats which appear most adept at huntingin more open habitats (McGregor et al 2015) particularly after

Table 4 Asymptotic KruskalndashWallis tests of home-range overlapbetween sex dyads using two-dimensional (K95 and K50) and

three-dimensional (PRH and UDOI) indicesValues shown in bold are significant at Plt 005

Total overlap (95) Core overlap (50)c2 P c2 P

K 8272 0041 7863 0049PRH 9099 0028 6921 0074UDOI 8485 0037 4125 0248

04Kilometres

(a) (b) (c)

(d) (e) (f )

(g) (h)

N

Fig 4 Core home-range overlap (K50) between all male (dashed lines) and female (solid lines)Wsquamicaudata tracked in theArtesianRange north-westKimberleyWesternAustralia (a) Site1 January2013wet season (b) Site 1 April 2013 dry season (c) Site 1 July 2013 dry season (d) Site 1 December 2013wet season (e) Site 2 February 2013 wet season (f) Site 2 August 2013 dry season (g) Site 3 May 2013 dryseason (h) Site 3 January 2014 wet season Note that different individuals during each month-long trackingperiod can be distinguished by differing line weight or dash interval

Habitat use by Wyulda squamicaudata Australian Journal of Zoology G

understorey vegetation has been simplified by intense fire orgrazing (McGregor et al 2014) Reduced hunting success inrocky habitats may explain why cat density in this environment isvery low (Hohnen et al 2016) Also at a slightly larger spatialscale rocky landscapes tend to burn more patchily than opensavannas thus retaining more heterogeneity in vegetation age(Price et al 2003) In this study Wyulda selected for high levelsof habitat heterogeneity at both den and foraging sites and usedall ages of post-fire savannas as long as rocky features werepresent Habitat heterogeneity appears to be important for severalvertebrate species in northern Australia (Bolton and Latz 1978Ingleby and Westoby 1992 Pardon et al 2003 Southgate et al2007) and elsewhere (Doumas and Koprowski 2013)

While Wyulda are capable of consuming a wide range ofplant species fruit is clearly a key component of their diet Thisadds to information from a previous radio-tracking study that

identified Wyulda foraging within Xanthostemon paradoxusXanthostemon eucalyptoides and Planchonia careya (Runcie1999) all fruit-bearing species Fruit also makes an importantcontribution to the diet for other arboreal marsupials native tonorth-western Australia (Kerle 1985) Intense and frequent firescan kill or delay the fruiting of tree species (Williams et al 1999Vigilante and Bowman 2004a Atchison et al 2005 Atchison2009) Rocky structures (such as gorges or rock outcrops) canprovide conditions such as shade and water retention that favoursome fruiting species (such as Ficus platypoda) and also protectthem from fire Thus rock structure may contribute to theprotection of fire-sensitive foraging resources used by Wyulda

While the results of this study suggest that Wyulda favourrocky landscapes it is unclear whether the observed relationshipreflects a true habitat preference or is a result of recent reductionsof habitat quality in the open savannas caused by current fireregimes grazing by introduced herbivores and the impacts offeral cats Wyulda have been detected in fauna surveys on rockyescarpments of Charnley River Station separated from sites usedin this studybymore than15 kmof openbasalt plains (KTuft andS Legge unpubl data) On a larger scale Emma Gorge (wherethe species was recently rediscovered in the east Kimberley) isseparated from other historical records in the north Kimberley(eg Mitchell Plateau) by more than 200 km and large stretchesof open plains (Doody et al 2012) The highly disjunct natureof this distribution suggests that there may have been a timewhen the species moved more freely over open landscapesA recent phylogenetic study found that unlike several saxicolousrock wallaby species in the region Wyulda showed no deepstructure between eastern and western populations suggestingrecent colonisation or connectivity across the Kimberley (Potter

Table 5 Top five candidate discrete choice models derived fromforaging data for Wyulda squamicaudata

n1 no of real points included in the model n0 no of generated pointsincluded in the model Abbreviations are as follows hab_type habitat typehet habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K AIC DAIC w

Foraging model (n1 = 2615 n0 = 13071)hab_type+ het+ rock 3 914189 000 041hab_type+ het + rock + distc 4 914254 065 030hab_type+ fire + het + rock 4 914373 184 016hab_type+ fire + het + rock + distc 5 914425 236 013hab_type+ het 2 915147 958 000

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus05 00 05 10

Coefficent estimate

Var

iabl

e

Fig 5 Model-averaged coefficient estimates with 95 credible intervals of discrete choice models describing foraging-siteselection byW squamicaudata in the Artesian Range north-west Kimberley Western Australia The reference habitat typewas Heteropogon contortus savanna and level of rock cover was cliff

H Australian Journal of Zoology R Hohnen et al

et al 2014) Historical records from the east Kimberley suggestthat the species did once persist inmuch drier habitats (Alexander1919)

Wyulda in this study continued to den and forage in habitatsthat had been recently burnt (1ndash12 months after fire) a patternalso observed in several other northern Australian mammalstudies (Hohnen et al 2015) For pale field rats (Rattus tunneyi)in the central Kimberley Leahy et al (2016) found that whilepopulation size decreased after an intense fire individual rats

stayed within their original home ranges and did not move toadjacent unburnt areas Low-intensity fire did not displacenorthern quolls (Dasyurus hallucatus) from their home ranges inthe north Kimberley (Cook 2010) or significantly affect home-range size or location of northern bettongs (Bettongia tropica) innorth-east Queensland (Vernes and Pope 2001) The continueduse of savannahabitats afterfire byWyulda suggests that the costsof shifting home range are greater than any resource bottleneckcaused by recent or frequent fires The minimal home-rangeoverlap between adjacentmales and adjacent females in this studysuggests adegreeof territoriality thatmayalso inhibit home-rangemovement

The average home-range size estimated in this study(817 134 ha (MCP) or 802 129 ha (K95)) was larger thanprevious estimates of 10 ha (Runcie 1999) and a maximum of

Table 6 Odds ratios derived from the model-averaged coefficientestimates from discrete choice models of foraging data for Wyulda

squamicaudata

Variable Foraging dataOdds ratio Robust se

Recently burnt habitatsHibiscus keneallyii savanna 194 029Sarga stipodeum savanna 180 031

Long-unburnt habitatsRainforest 226 028River edge 106 030Long-unburnt savanna 157 029

Rock complexityRock complex 097 015Rock medium 083 012Rock simple 071 019

Other variablesFire 098 005Habitat heterogeneity 112 003Home-range distance 106 008

Table 7 Top five candidate models derived from den data forWyuldasquamicaudata using Bayesian hierarchical logistic regression modelsn1 number of real points included in the model n0 number of generatedpoints included in the model Abbreviations are as follows hab_type habitattype het habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K DIC DDIC w

Den model (n1 = 736 n0 = 736)hab_type + rock + het + distc 4 3008 000 017hab_type + rock + distc 3 3058 050 014hab_type + rock + het 3 3066 058 013hab_type + rock + het +fire + distc 5 3123 115 010hab_type + rock +fire + distc 4 3147 139 009hab_type + het 2 3178 170 007

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus400 minus200 0 200

Coefficent estimate

Var

iabl

e

Fig 6 Model-averaged coefficient estimates with 95 credible intervals of Bayesian hierarchical logistic regressionmodels describing den site selection byW squamicaudata in the Artesian Range north-west KimberleyWestern AustraliaThe reference habitat type was Heteropogon contortus savanna and level of rock cover was cliff

Habitat use by Wyulda squamicaudata Australian Journal of Zoology I

272m long (Humphreys et al 1984) Potentially this reflectsdifferences between the two studies in habitat quality andor thenumber of location points recorded per individual Home-rangeestimates from this study were larger than those of other northKimberley arboreal marsupials including the northern brushtailpossum (Trichosurus vulpecula arhnemensis) (087ndash112 haKerle 1998) and the cohabiting saxicolous rock ringtail possum(Petropseudes dahli) (05ndash12 ha Runcie 2000) While home-range size did not vary between sexes overlap occurred muchmore frequently and extensively between males and femalesthan between individuals of the same sex Several other possumspecies native to northern Australia move and den in familygroups (Kerle 1998 Runcie 2000) Potentially Wyulda alsopractice some degree of sociality and pair bonding

Currently the distribution of Wyulda is restricted to thetopographically complex north Kimberley with a small isolatedpopulation persisting in the east Kimberley While this speciesmay have once been more widely distributed in the centraland east Kimberley including in a wider range of habitatshabitat heterogeneity the availability of fruiting tree species andparticularly the availability of topographic and rock complexitywere key habitat features selected for by Wyulda Rockcomplexity may provide some buffer to Wyulda from threatssuch as intense fires and feral cats offer shelter from predationduring the post-fire period with cliffs and rock outcrops formingnatural barriers to such fires protecting sensitive vegetationand promoting habitat heterogeneity Mammal declines haveoccurred in topographically complex areas of the Arnhem Landplateau (Woinarski 2000 Ziembicki et al 2015) KakaduNational Park (Woinarski et al 2001) and the south-westKimberley (McKenzie 1981 Sawle 1988) Thus substratecomplexity alone does not guarantee mammal species securityManagement strategies that aim to maintain availability ofhabitat heterogeneity and fruiting trees may support thepersistence of Wyulda in rocky habitats Such strategies couldinclude lighting small-scale low-intensity fires early in the dryseason to prevent the incidence and extent of large-scale fires thatoccur late in the dry season

Acknowledgements

This studywas funded by supporters of the AustralianWildlife Conservancyan Australian Research Council Discovery Grant (100100033) and theDepartment of Parks and Wildlife (Western Australia) The study methodswere approved by the University of Tasmania Animal Ethics Committee(Permit no A12516) Fieldwork would not have been possible without thesupport of staff at theAustralianWildlifeConservancyrsquosMorningtonWildlifeSanctuary Thanks go in particular to Alex Hartshorne Hannah Cliff JillianSmith Joel Murray David James Peter Richardson Kaely Kreger DavinaBright Tom Crawford and Iris Bleach

References

Alexander W B (1919) A new species of marsupial of the subfamilyPhalangerinae Journal of the Royal Society of Western Australia 431ndash36

Andersen A N Braithwaite R W Cook G D Corbett L K WilliamsR J Douglas M M Gill A M Setterfield S A and Muller W J(1998) Fire research for conservation management in tropical savannasintroducing the Kapalga fire experiment Australian Journal of Ecology23 95ndash110 doi101111j1442-99931998tb00708x

Andersen A N Cook G D Corbett L K Douglas M M Eager R WRussell-Smith J SetterfieldSAWilliamsR J andWoinarski JCZ(2005) Fire frequency and biodiversity conservation in Australiantropical savannas implications from the Kapalga fire experiment AustralEcology 30 155ndash167 doi101111j1442-9993200501441x

Anderson D R (2008) lsquoModel Based Inferences in the Life Sciencesrsquo(Springer Science and Business Media New York)

Atchison J (2009) Human impacts on Persoonia falcata Perspectives onpost-contact vegetation change in the Keep River region Australia fromcontemporary vegetation surveysVegetationHistoryandArchaeobotany18 147ndash157 doi101007s00334-008-0198-y

Atchison J Head L and Fullagar R (2005) Archaeobotany of fruit seedprocessing in a monsoon savanna environment evidence from the KeepRiver region Northern Territory Australia Journal of ArchaeologicalScience 32 167ndash181 doi101016jjas200403022

Banks S C Dujardin M McBurney L Blair D Barker M andLindenmayer D B (2011) Starting points for small mammal populationrecovery after wildfire recolonisation or residual populationsOikos 12026ndash37 doi101111j1600-0706201018765x

Barton K (2013)MuMInmulti-model inference R package version 1913Bates D Maechler M Bolker B and Walker S (2014) lme4 linear

mixed-effects models using Eigen and S4 R package version 11-7Beyer H L (2012) Geospatial modelling environment Spatial Ecology

LLCBjoslashrneraas K Van Moorter B Rolandsen C M and Herfindal I (2010)

Screening global positioning system location data for errors usinganimal movement characteristics Journal of Wildlife Management 741361ndash1366 doi1021932009-405

Bolton B and Latz P (1978) The western hare-wallaby Lagorchesteshirsutus (Gould) (Macropodidae) in the Tanami Desert WildlifeResearch 5 285ndash293 doi101071WR9780285

BradleyAKemper CKitchenerDHumphreysW andHowR (1987)Small mammals of the Mitchell Plateau region Kimberley WesternAustralia Wildlife Research 14 397ndash413 doi101071WR9870397

Burbidge A A andMcKenzie N L (1989) Patterns in the modern declineof Western Australian vertebrate fauna ndash causes and conservationimplications Biological Conservation 50 143ndash198 doi1010160006-3207(89)90009-8

Burbidge A A and Webb M J (2008) Scaly-tailed possum In lsquoTheMammals of Australiarsquo 3rd edn (Eds S Van Dyck and R Strahan)pp 277ndash278 (New Holland Publishers Sydney)

Bureau of Meteorology (2015) Climate data online Commonwealth ofAustralia Bureau of Meteorology

Cook A C (2010) Habitat use and home range of the northern quollDasyurus hallucatus effects of fire MSc thesis The University ofWestern Australia Perth

Corbett L C Andersen A N and Muumlller W J (2003) Terrestrialvertebrates In lsquoFire inTropical SavannasTheKapalgaExperimentrsquo (EdsA N Andersen G D Cook and R J Williams) pp 126ndash52 (SpringerVerlag New York)

Doody J S Rhind D Castellano CM and BassM (2012) Rediscoveryof the scaly-tailed possum (Wyulda squamicaudata) in the easternKimberleyAustralianMammalogy 34 260ndash262 doi101071AM11039

Doumas S L and Koprowski J L (2013) Effect of heterogeneity inburn severity on Mexican fox squirrels following the return of fireInternational Journal of Wildland Fire 22 405ndash413 doi101071WF12046

Driscoll D A Lindenmayer D B Bennett A F Bode M BradstockR A CaryG J ClarkeM F Dexter N FenshamR Friend G GillM James S Kay G Keith D A MacGregor C Russell-Smith JSalt D Watson J E M Williams R J and York A (2010) Firemanagement for biodiversity conservation key research questions andour capacity to answer them Biological Conservation 143 1928ndash1939doi101016jbiocon201005026

J Australian Journal of Zoology R Hohnen et al

understorey vegetation has been simplified by intense fire orgrazing (McGregor et al 2014) Reduced hunting success inrocky habitats may explain why cat density in this environment isvery low (Hohnen et al 2016) Also at a slightly larger spatialscale rocky landscapes tend to burn more patchily than opensavannas thus retaining more heterogeneity in vegetation age(Price et al 2003) In this study Wyulda selected for high levelsof habitat heterogeneity at both den and foraging sites and usedall ages of post-fire savannas as long as rocky features werepresent Habitat heterogeneity appears to be important for severalvertebrate species in northern Australia (Bolton and Latz 1978Ingleby and Westoby 1992 Pardon et al 2003 Southgate et al2007) and elsewhere (Doumas and Koprowski 2013)

While Wyulda are capable of consuming a wide range ofplant species fruit is clearly a key component of their diet Thisadds to information from a previous radio-tracking study that

identified Wyulda foraging within Xanthostemon paradoxusXanthostemon eucalyptoides and Planchonia careya (Runcie1999) all fruit-bearing species Fruit also makes an importantcontribution to the diet for other arboreal marsupials native tonorth-western Australia (Kerle 1985) Intense and frequent firescan kill or delay the fruiting of tree species (Williams et al 1999Vigilante and Bowman 2004a Atchison et al 2005 Atchison2009) Rocky structures (such as gorges or rock outcrops) canprovide conditions such as shade and water retention that favoursome fruiting species (such as Ficus platypoda) and also protectthem from fire Thus rock structure may contribute to theprotection of fire-sensitive foraging resources used by Wyulda

While the results of this study suggest that Wyulda favourrocky landscapes it is unclear whether the observed relationshipreflects a true habitat preference or is a result of recent reductionsof habitat quality in the open savannas caused by current fireregimes grazing by introduced herbivores and the impacts offeral cats Wyulda have been detected in fauna surveys on rockyescarpments of Charnley River Station separated from sites usedin this studybymore than15 kmof openbasalt plains (KTuft andS Legge unpubl data) On a larger scale Emma Gorge (wherethe species was recently rediscovered in the east Kimberley) isseparated from other historical records in the north Kimberley(eg Mitchell Plateau) by more than 200 km and large stretchesof open plains (Doody et al 2012) The highly disjunct natureof this distribution suggests that there may have been a timewhen the species moved more freely over open landscapesA recent phylogenetic study found that unlike several saxicolousrock wallaby species in the region Wyulda showed no deepstructure between eastern and western populations suggestingrecent colonisation or connectivity across the Kimberley (Potter

Table 5 Top five candidate discrete choice models derived fromforaging data for Wyulda squamicaudata

n1 no of real points included in the model n0 no of generated pointsincluded in the model Abbreviations are as follows hab_type habitat typehet habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K AIC DAIC w

Foraging model (n1 = 2615 n0 = 13071)hab_type+ het+ rock 3 914189 000 041hab_type+ het + rock + distc 4 914254 065 030hab_type+ fire + het + rock 4 914373 184 016hab_type+ fire + het + rock + distc 5 914425 236 013hab_type+ het 2 915147 958 000

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus05 00 05 10

Coefficent estimate

Var

iabl

e

Fig 5 Model-averaged coefficient estimates with 95 credible intervals of discrete choice models describing foraging-siteselection byW squamicaudata in the Artesian Range north-west Kimberley Western Australia The reference habitat typewas Heteropogon contortus savanna and level of rock cover was cliff

H Australian Journal of Zoology R Hohnen et al

et al 2014) Historical records from the east Kimberley suggestthat the species did once persist inmuch drier habitats (Alexander1919)

Wyulda in this study continued to den and forage in habitatsthat had been recently burnt (1ndash12 months after fire) a patternalso observed in several other northern Australian mammalstudies (Hohnen et al 2015) For pale field rats (Rattus tunneyi)in the central Kimberley Leahy et al (2016) found that whilepopulation size decreased after an intense fire individual rats

stayed within their original home ranges and did not move toadjacent unburnt areas Low-intensity fire did not displacenorthern quolls (Dasyurus hallucatus) from their home ranges inthe north Kimberley (Cook 2010) or significantly affect home-range size or location of northern bettongs (Bettongia tropica) innorth-east Queensland (Vernes and Pope 2001) The continueduse of savannahabitats afterfire byWyulda suggests that the costsof shifting home range are greater than any resource bottleneckcaused by recent or frequent fires The minimal home-rangeoverlap between adjacentmales and adjacent females in this studysuggests adegreeof territoriality thatmayalso inhibit home-rangemovement

The average home-range size estimated in this study(817 134 ha (MCP) or 802 129 ha (K95)) was larger thanprevious estimates of 10 ha (Runcie 1999) and a maximum of

Table 6 Odds ratios derived from the model-averaged coefficientestimates from discrete choice models of foraging data for Wyulda

squamicaudata

Variable Foraging dataOdds ratio Robust se

Recently burnt habitatsHibiscus keneallyii savanna 194 029Sarga stipodeum savanna 180 031

Long-unburnt habitatsRainforest 226 028River edge 106 030Long-unburnt savanna 157 029

Rock complexityRock complex 097 015Rock medium 083 012Rock simple 071 019

Other variablesFire 098 005Habitat heterogeneity 112 003Home-range distance 106 008

Table 7 Top five candidate models derived from den data forWyuldasquamicaudata using Bayesian hierarchical logistic regression modelsn1 number of real points included in the model n0 number of generatedpoints included in the model Abbreviations are as follows hab_type habitattype het habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K DIC DDIC w

Den model (n1 = 736 n0 = 736)hab_type + rock + het + distc 4 3008 000 017hab_type + rock + distc 3 3058 050 014hab_type + rock + het 3 3066 058 013hab_type + rock + het +fire + distc 5 3123 115 010hab_type + rock +fire + distc 4 3147 139 009hab_type + het 2 3178 170 007

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus400 minus200 0 200

Coefficent estimate

Var

iabl

e

Fig 6 Model-averaged coefficient estimates with 95 credible intervals of Bayesian hierarchical logistic regressionmodels describing den site selection byW squamicaudata in the Artesian Range north-west KimberleyWestern AustraliaThe reference habitat type was Heteropogon contortus savanna and level of rock cover was cliff

Habitat use by Wyulda squamicaudata Australian Journal of Zoology I

272m long (Humphreys et al 1984) Potentially this reflectsdifferences between the two studies in habitat quality andor thenumber of location points recorded per individual Home-rangeestimates from this study were larger than those of other northKimberley arboreal marsupials including the northern brushtailpossum (Trichosurus vulpecula arhnemensis) (087ndash112 haKerle 1998) and the cohabiting saxicolous rock ringtail possum(Petropseudes dahli) (05ndash12 ha Runcie 2000) While home-range size did not vary between sexes overlap occurred muchmore frequently and extensively between males and femalesthan between individuals of the same sex Several other possumspecies native to northern Australia move and den in familygroups (Kerle 1998 Runcie 2000) Potentially Wyulda alsopractice some degree of sociality and pair bonding

Currently the distribution of Wyulda is restricted to thetopographically complex north Kimberley with a small isolatedpopulation persisting in the east Kimberley While this speciesmay have once been more widely distributed in the centraland east Kimberley including in a wider range of habitatshabitat heterogeneity the availability of fruiting tree species andparticularly the availability of topographic and rock complexitywere key habitat features selected for by Wyulda Rockcomplexity may provide some buffer to Wyulda from threatssuch as intense fires and feral cats offer shelter from predationduring the post-fire period with cliffs and rock outcrops formingnatural barriers to such fires protecting sensitive vegetationand promoting habitat heterogeneity Mammal declines haveoccurred in topographically complex areas of the Arnhem Landplateau (Woinarski 2000 Ziembicki et al 2015) KakaduNational Park (Woinarski et al 2001) and the south-westKimberley (McKenzie 1981 Sawle 1988) Thus substratecomplexity alone does not guarantee mammal species securityManagement strategies that aim to maintain availability ofhabitat heterogeneity and fruiting trees may support thepersistence of Wyulda in rocky habitats Such strategies couldinclude lighting small-scale low-intensity fires early in the dryseason to prevent the incidence and extent of large-scale fires thatoccur late in the dry season

Acknowledgements

This studywas funded by supporters of the AustralianWildlife Conservancyan Australian Research Council Discovery Grant (100100033) and theDepartment of Parks and Wildlife (Western Australia) The study methodswere approved by the University of Tasmania Animal Ethics Committee(Permit no A12516) Fieldwork would not have been possible without thesupport of staff at theAustralianWildlifeConservancyrsquosMorningtonWildlifeSanctuary Thanks go in particular to Alex Hartshorne Hannah Cliff JillianSmith Joel Murray David James Peter Richardson Kaely Kreger DavinaBright Tom Crawford and Iris Bleach

References

Alexander W B (1919) A new species of marsupial of the subfamilyPhalangerinae Journal of the Royal Society of Western Australia 431ndash36

Andersen A N Braithwaite R W Cook G D Corbett L K WilliamsR J Douglas M M Gill A M Setterfield S A and Muller W J(1998) Fire research for conservation management in tropical savannasintroducing the Kapalga fire experiment Australian Journal of Ecology23 95ndash110 doi101111j1442-99931998tb00708x

Andersen A N Cook G D Corbett L K Douglas M M Eager R WRussell-Smith J SetterfieldSAWilliamsR J andWoinarski JCZ(2005) Fire frequency and biodiversity conservation in Australiantropical savannas implications from the Kapalga fire experiment AustralEcology 30 155ndash167 doi101111j1442-9993200501441x

Anderson D R (2008) lsquoModel Based Inferences in the Life Sciencesrsquo(Springer Science and Business Media New York)

Atchison J (2009) Human impacts on Persoonia falcata Perspectives onpost-contact vegetation change in the Keep River region Australia fromcontemporary vegetation surveysVegetationHistoryandArchaeobotany18 147ndash157 doi101007s00334-008-0198-y

Atchison J Head L and Fullagar R (2005) Archaeobotany of fruit seedprocessing in a monsoon savanna environment evidence from the KeepRiver region Northern Territory Australia Journal of ArchaeologicalScience 32 167ndash181 doi101016jjas200403022

Banks S C Dujardin M McBurney L Blair D Barker M andLindenmayer D B (2011) Starting points for small mammal populationrecovery after wildfire recolonisation or residual populationsOikos 12026ndash37 doi101111j1600-0706201018765x

Barton K (2013)MuMInmulti-model inference R package version 1913Bates D Maechler M Bolker B and Walker S (2014) lme4 linear

mixed-effects models using Eigen and S4 R package version 11-7Beyer H L (2012) Geospatial modelling environment Spatial Ecology

LLCBjoslashrneraas K Van Moorter B Rolandsen C M and Herfindal I (2010)

Screening global positioning system location data for errors usinganimal movement characteristics Journal of Wildlife Management 741361ndash1366 doi1021932009-405

Bolton B and Latz P (1978) The western hare-wallaby Lagorchesteshirsutus (Gould) (Macropodidae) in the Tanami Desert WildlifeResearch 5 285ndash293 doi101071WR9780285

BradleyAKemper CKitchenerDHumphreysW andHowR (1987)Small mammals of the Mitchell Plateau region Kimberley WesternAustralia Wildlife Research 14 397ndash413 doi101071WR9870397

Burbidge A A andMcKenzie N L (1989) Patterns in the modern declineof Western Australian vertebrate fauna ndash causes and conservationimplications Biological Conservation 50 143ndash198 doi1010160006-3207(89)90009-8

Burbidge A A and Webb M J (2008) Scaly-tailed possum In lsquoTheMammals of Australiarsquo 3rd edn (Eds S Van Dyck and R Strahan)pp 277ndash278 (New Holland Publishers Sydney)

Bureau of Meteorology (2015) Climate data online Commonwealth ofAustralia Bureau of Meteorology

Cook A C (2010) Habitat use and home range of the northern quollDasyurus hallucatus effects of fire MSc thesis The University ofWestern Australia Perth

Corbett L C Andersen A N and Muumlller W J (2003) Terrestrialvertebrates In lsquoFire inTropical SavannasTheKapalgaExperimentrsquo (EdsA N Andersen G D Cook and R J Williams) pp 126ndash52 (SpringerVerlag New York)

Doody J S Rhind D Castellano CM and BassM (2012) Rediscoveryof the scaly-tailed possum (Wyulda squamicaudata) in the easternKimberleyAustralianMammalogy 34 260ndash262 doi101071AM11039

Doumas S L and Koprowski J L (2013) Effect of heterogeneity inburn severity on Mexican fox squirrels following the return of fireInternational Journal of Wildland Fire 22 405ndash413 doi101071WF12046

Driscoll D A Lindenmayer D B Bennett A F Bode M BradstockR A CaryG J ClarkeM F Dexter N FenshamR Friend G GillM James S Kay G Keith D A MacGregor C Russell-Smith JSalt D Watson J E M Williams R J and York A (2010) Firemanagement for biodiversity conservation key research questions andour capacity to answer them Biological Conservation 143 1928ndash1939doi101016jbiocon201005026

J Australian Journal of Zoology R Hohnen et al

et al 2014) Historical records from the east Kimberley suggestthat the species did once persist inmuch drier habitats (Alexander1919)

Wyulda in this study continued to den and forage in habitatsthat had been recently burnt (1ndash12 months after fire) a patternalso observed in several other northern Australian mammalstudies (Hohnen et al 2015) For pale field rats (Rattus tunneyi)in the central Kimberley Leahy et al (2016) found that whilepopulation size decreased after an intense fire individual rats

stayed within their original home ranges and did not move toadjacent unburnt areas Low-intensity fire did not displacenorthern quolls (Dasyurus hallucatus) from their home ranges inthe north Kimberley (Cook 2010) or significantly affect home-range size or location of northern bettongs (Bettongia tropica) innorth-east Queensland (Vernes and Pope 2001) The continueduse of savannahabitats afterfire byWyulda suggests that the costsof shifting home range are greater than any resource bottleneckcaused by recent or frequent fires The minimal home-rangeoverlap between adjacentmales and adjacent females in this studysuggests adegreeof territoriality thatmayalso inhibit home-rangemovement

The average home-range size estimated in this study(817 134 ha (MCP) or 802 129 ha (K95)) was larger thanprevious estimates of 10 ha (Runcie 1999) and a maximum of

Table 6 Odds ratios derived from the model-averaged coefficientestimates from discrete choice models of foraging data for Wyulda

squamicaudata

Variable Foraging dataOdds ratio Robust se

Recently burnt habitatsHibiscus keneallyii savanna 194 029Sarga stipodeum savanna 180 031

Long-unburnt habitatsRainforest 226 028River edge 106 030Long-unburnt savanna 157 029

Rock complexityRock complex 097 015Rock medium 083 012Rock simple 071 019

Other variablesFire 098 005Habitat heterogeneity 112 003Home-range distance 106 008

Table 7 Top five candidate models derived from den data forWyuldasquamicaudata using Bayesian hierarchical logistic regression modelsn1 number of real points included in the model n0 number of generatedpoints included in the model Abbreviations are as follows hab_type habitattype het habitat heterogeneity rock rock complexity distc distance to home

range centroid fire time since fire

Model K DIC DDIC w

Den model (n1 = 736 n0 = 736)hab_type + rock + het + distc 4 3008 000 017hab_type + rock + distc 3 3058 050 014hab_type + rock + het 3 3066 058 013hab_type + rock + het +fire + distc 5 3123 115 010hab_type + rock +fire + distc 4 3147 139 009hab_type + het 2 3178 170 007

Rainforest

River edge

Hibiscus keneallyi savanna

Sorghum stipodeum savanna

Triodia spp savanna

High rock complexity

Medium rock complexity

Low rock complexity

Fire

Habitat hetrogeniety

Distance to centroid

minus400 minus200 0 200

Coefficent estimate

Var

iabl

e

Fig 6 Model-averaged coefficient estimates with 95 credible intervals of Bayesian hierarchical logistic regressionmodels describing den site selection byW squamicaudata in the Artesian Range north-west KimberleyWestern AustraliaThe reference habitat type was Heteropogon contortus savanna and level of rock cover was cliff

Habitat use by Wyulda squamicaudata Australian Journal of Zoology I

272m long (Humphreys et al 1984) Potentially this reflectsdifferences between the two studies in habitat quality andor thenumber of location points recorded per individual Home-rangeestimates from this study were larger than those of other northKimberley arboreal marsupials including the northern brushtailpossum (Trichosurus vulpecula arhnemensis) (087ndash112 haKerle 1998) and the cohabiting saxicolous rock ringtail possum(Petropseudes dahli) (05ndash12 ha Runcie 2000) While home-range size did not vary between sexes overlap occurred muchmore frequently and extensively between males and femalesthan between individuals of the same sex Several other possumspecies native to northern Australia move and den in familygroups (Kerle 1998 Runcie 2000) Potentially Wyulda alsopractice some degree of sociality and pair bonding

Currently the distribution of Wyulda is restricted to thetopographically complex north Kimberley with a small isolatedpopulation persisting in the east Kimberley While this speciesmay have once been more widely distributed in the centraland east Kimberley including in a wider range of habitatshabitat heterogeneity the availability of fruiting tree species andparticularly the availability of topographic and rock complexitywere key habitat features selected for by Wyulda Rockcomplexity may provide some buffer to Wyulda from threatssuch as intense fires and feral cats offer shelter from predationduring the post-fire period with cliffs and rock outcrops formingnatural barriers to such fires protecting sensitive vegetationand promoting habitat heterogeneity Mammal declines haveoccurred in topographically complex areas of the Arnhem Landplateau (Woinarski 2000 Ziembicki et al 2015) KakaduNational Park (Woinarski et al 2001) and the south-westKimberley (McKenzie 1981 Sawle 1988) Thus substratecomplexity alone does not guarantee mammal species securityManagement strategies that aim to maintain availability ofhabitat heterogeneity and fruiting trees may support thepersistence of Wyulda in rocky habitats Such strategies couldinclude lighting small-scale low-intensity fires early in the dryseason to prevent the incidence and extent of large-scale fires thatoccur late in the dry season

Acknowledgements

This studywas funded by supporters of the AustralianWildlife Conservancyan Australian Research Council Discovery Grant (100100033) and theDepartment of Parks and Wildlife (Western Australia) The study methodswere approved by the University of Tasmania Animal Ethics Committee(Permit no A12516) Fieldwork would not have been possible without thesupport of staff at theAustralianWildlifeConservancyrsquosMorningtonWildlifeSanctuary Thanks go in particular to Alex Hartshorne Hannah Cliff JillianSmith Joel Murray David James Peter Richardson Kaely Kreger DavinaBright Tom Crawford and Iris Bleach

References

Alexander W B (1919) A new species of marsupial of the subfamilyPhalangerinae Journal of the Royal Society of Western Australia 431ndash36

Andersen A N Braithwaite R W Cook G D Corbett L K WilliamsR J Douglas M M Gill A M Setterfield S A and Muller W J(1998) Fire research for conservation management in tropical savannasintroducing the Kapalga fire experiment Australian Journal of Ecology23 95ndash110 doi101111j1442-99931998tb00708x

Andersen A N Cook G D Corbett L K Douglas M M Eager R WRussell-Smith J SetterfieldSAWilliamsR J andWoinarski JCZ(2005) Fire frequency and biodiversity conservation in Australiantropical savannas implications from the Kapalga fire experiment AustralEcology 30 155ndash167 doi101111j1442-9993200501441x

Anderson D R (2008) lsquoModel Based Inferences in the Life Sciencesrsquo(Springer Science and Business Media New York)

Atchison J (2009) Human impacts on Persoonia falcata Perspectives onpost-contact vegetation change in the Keep River region Australia fromcontemporary vegetation surveysVegetationHistoryandArchaeobotany18 147ndash157 doi101007s00334-008-0198-y

Atchison J Head L and Fullagar R (2005) Archaeobotany of fruit seedprocessing in a monsoon savanna environment evidence from the KeepRiver region Northern Territory Australia Journal of ArchaeologicalScience 32 167ndash181 doi101016jjas200403022

Banks S C Dujardin M McBurney L Blair D Barker M andLindenmayer D B (2011) Starting points for small mammal populationrecovery after wildfire recolonisation or residual populationsOikos 12026ndash37 doi101111j1600-0706201018765x

Barton K (2013)MuMInmulti-model inference R package version 1913Bates D Maechler M Bolker B and Walker S (2014) lme4 linear

mixed-effects models using Eigen and S4 R package version 11-7Beyer H L (2012) Geospatial modelling environment Spatial Ecology

LLCBjoslashrneraas K Van Moorter B Rolandsen C M and Herfindal I (2010)

Screening global positioning system location data for errors usinganimal movement characteristics Journal of Wildlife Management 741361ndash1366 doi1021932009-405

Bolton B and Latz P (1978) The western hare-wallaby Lagorchesteshirsutus (Gould) (Macropodidae) in the Tanami Desert WildlifeResearch 5 285ndash293 doi101071WR9780285

BradleyAKemper CKitchenerDHumphreysW andHowR (1987)Small mammals of the Mitchell Plateau region Kimberley WesternAustralia Wildlife Research 14 397ndash413 doi101071WR9870397

Burbidge A A andMcKenzie N L (1989) Patterns in the modern declineof Western Australian vertebrate fauna ndash causes and conservationimplications Biological Conservation 50 143ndash198 doi1010160006-3207(89)90009-8

Burbidge A A and Webb M J (2008) Scaly-tailed possum In lsquoTheMammals of Australiarsquo 3rd edn (Eds S Van Dyck and R Strahan)pp 277ndash278 (New Holland Publishers Sydney)

Bureau of Meteorology (2015) Climate data online Commonwealth ofAustralia Bureau of Meteorology

Cook A C (2010) Habitat use and home range of the northern quollDasyurus hallucatus effects of fire MSc thesis The University ofWestern Australia Perth

Corbett L C Andersen A N and Muumlller W J (2003) Terrestrialvertebrates In lsquoFire inTropical SavannasTheKapalgaExperimentrsquo (EdsA N Andersen G D Cook and R J Williams) pp 126ndash52 (SpringerVerlag New York)

Doody J S Rhind D Castellano CM and BassM (2012) Rediscoveryof the scaly-tailed possum (Wyulda squamicaudata) in the easternKimberleyAustralianMammalogy 34 260ndash262 doi101071AM11039

Doumas S L and Koprowski J L (2013) Effect of heterogeneity inburn severity on Mexican fox squirrels following the return of fireInternational Journal of Wildland Fire 22 405ndash413 doi101071WF12046

Driscoll D A Lindenmayer D B Bennett A F Bode M BradstockR A CaryG J ClarkeM F Dexter N FenshamR Friend G GillM James S Kay G Keith D A MacGregor C Russell-Smith JSalt D Watson J E M Williams R J and York A (2010) Firemanagement for biodiversity conservation key research questions andour capacity to answer them Biological Conservation 143 1928ndash1939doi101016jbiocon201005026

J Australian Journal of Zoology R Hohnen et al

272m long (Humphreys et al 1984) Potentially this reflectsdifferences between the two studies in habitat quality andor thenumber of location points recorded per individual Home-rangeestimates from this study were larger than those of other northKimberley arboreal marsupials including the northern brushtailpossum (Trichosurus vulpecula arhnemensis) (087ndash112 haKerle 1998) and the cohabiting saxicolous rock ringtail possum(Petropseudes dahli) (05ndash12 ha Runcie 2000) While home-range size did not vary between sexes overlap occurred muchmore frequently and extensively between males and femalesthan between individuals of the same sex Several other possumspecies native to northern Australia move and den in familygroups (Kerle 1998 Runcie 2000) Potentially Wyulda alsopractice some degree of sociality and pair bonding

Currently the distribution of Wyulda is restricted to thetopographically complex north Kimberley with a small isolatedpopulation persisting in the east Kimberley While this speciesmay have once been more widely distributed in the centraland east Kimberley including in a wider range of habitatshabitat heterogeneity the availability of fruiting tree species andparticularly the availability of topographic and rock complexitywere key habitat features selected for by Wyulda Rockcomplexity may provide some buffer to Wyulda from threatssuch as intense fires and feral cats offer shelter from predationduring the post-fire period with cliffs and rock outcrops formingnatural barriers to such fires protecting sensitive vegetationand promoting habitat heterogeneity Mammal declines haveoccurred in topographically complex areas of the Arnhem Landplateau (Woinarski 2000 Ziembicki et al 2015) KakaduNational Park (Woinarski et al 2001) and the south-westKimberley (McKenzie 1981 Sawle 1988) Thus substratecomplexity alone does not guarantee mammal species securityManagement strategies that aim to maintain availability ofhabitat heterogeneity and fruiting trees may support thepersistence of Wyulda in rocky habitats Such strategies couldinclude lighting small-scale low-intensity fires early in the dryseason to prevent the incidence and extent of large-scale fires thatoccur late in the dry season

Acknowledgements

This studywas funded by supporters of the AustralianWildlife Conservancyan Australian Research Council Discovery Grant (100100033) and theDepartment of Parks and Wildlife (Western Australia) The study methodswere approved by the University of Tasmania Animal Ethics Committee(Permit no A12516) Fieldwork would not have been possible without thesupport of staff at theAustralianWildlifeConservancyrsquosMorningtonWildlifeSanctuary Thanks go in particular to Alex Hartshorne Hannah Cliff JillianSmith Joel Murray David James Peter Richardson Kaely Kreger DavinaBright Tom Crawford and Iris Bleach

References

Alexander W B (1919) A new species of marsupial of the subfamilyPhalangerinae Journal of the Royal Society of Western Australia 431ndash36

Andersen A N Braithwaite R W Cook G D Corbett L K WilliamsR J Douglas M M Gill A M Setterfield S A and Muller W J(1998) Fire research for conservation management in tropical savannasintroducing the Kapalga fire experiment Australian Journal of Ecology23 95ndash110 doi101111j1442-99931998tb00708x

Andersen A N Cook G D Corbett L K Douglas M M Eager R WRussell-Smith J SetterfieldSAWilliamsR J andWoinarski JCZ(2005) Fire frequency and biodiversity conservation in Australiantropical savannas implications from the Kapalga fire experiment AustralEcology 30 155ndash167 doi101111j1442-9993200501441x

Anderson D R (2008) lsquoModel Based Inferences in the Life Sciencesrsquo(Springer Science and Business Media New York)

Atchison J (2009) Human impacts on Persoonia falcata Perspectives onpost-contact vegetation change in the Keep River region Australia fromcontemporary vegetation surveysVegetationHistoryandArchaeobotany18 147ndash157 doi101007s00334-008-0198-y

Atchison J Head L and Fullagar R (2005) Archaeobotany of fruit seedprocessing in a monsoon savanna environment evidence from the KeepRiver region Northern Territory Australia Journal of ArchaeologicalScience 32 167ndash181 doi101016jjas200403022

Banks S C Dujardin M McBurney L Blair D Barker M andLindenmayer D B (2011) Starting points for small mammal populationrecovery after wildfire recolonisation or residual populationsOikos 12026ndash37 doi101111j1600-0706201018765x

Barton K (2013)MuMInmulti-model inference R package version 1913Bates D Maechler M Bolker B and Walker S (2014) lme4 linear

mixed-effects models using Eigen and S4 R package version 11-7Beyer H L (2012) Geospatial modelling environment Spatial Ecology

LLCBjoslashrneraas K Van Moorter B Rolandsen C M and Herfindal I (2010)

Screening global positioning system location data for errors usinganimal movement characteristics Journal of Wildlife Management 741361ndash1366 doi1021932009-405

Bolton B and Latz P (1978) The western hare-wallaby Lagorchesteshirsutus (Gould) (Macropodidae) in the Tanami Desert WildlifeResearch 5 285ndash293 doi101071WR9780285

BradleyAKemper CKitchenerDHumphreysW andHowR (1987)Small mammals of the Mitchell Plateau region Kimberley WesternAustralia Wildlife Research 14 397ndash413 doi101071WR9870397

Burbidge A A andMcKenzie N L (1989) Patterns in the modern declineof Western Australian vertebrate fauna ndash causes and conservationimplications Biological Conservation 50 143ndash198 doi1010160006-3207(89)90009-8

Burbidge A A and Webb M J (2008) Scaly-tailed possum In lsquoTheMammals of Australiarsquo 3rd edn (Eds S Van Dyck and R Strahan)pp 277ndash278 (New Holland Publishers Sydney)

Bureau of Meteorology (2015) Climate data online Commonwealth ofAustralia Bureau of Meteorology

Cook A C (2010) Habitat use and home range of the northern quollDasyurus hallucatus effects of fire MSc thesis The University ofWestern Australia Perth

Corbett L C Andersen A N and Muumlller W J (2003) Terrestrialvertebrates In lsquoFire inTropical SavannasTheKapalgaExperimentrsquo (EdsA N Andersen G D Cook and R J Williams) pp 126ndash52 (SpringerVerlag New York)

Doody J S Rhind D Castellano CM and BassM (2012) Rediscoveryof the scaly-tailed possum (Wyulda squamicaudata) in the easternKimberleyAustralianMammalogy 34 260ndash262 doi101071AM11039

Doumas S L and Koprowski J L (2013) Effect of heterogeneity inburn severity on Mexican fox squirrels following the return of fireInternational Journal of Wildland Fire 22 405ndash413 doi101071WF12046

Driscoll D A Lindenmayer D B Bennett A F Bode M BradstockR A CaryG J ClarkeM F Dexter N FenshamR Friend G GillM James S Kay G Keith D A MacGregor C Russell-Smith JSalt D Watson J E M Williams R J and York A (2010) Firemanagement for biodiversity conservation key research questions andour capacity to answer them Biological Conservation 143 1928ndash1939doi101016jbiocon201005026

J Australian Journal of Zoology R Hohnen et al

Fieberg J andKochannyCO (2005)Quantifyinghome-rangeoverlap theimportanceof theutilizationdistribution JournalofWildlifeManagement69 1346ndash1359 doi1021930022-541X(2005)69[1346QHOTIO]20CO2

Firth R S C Brook BWWoinarski J C Z and Fordham D A (2010)Decline and likely extinction of a northern Australian native rodent thebrush-tailed rabbit-rat Conilurus penicillatus Biological Conservation143 1193ndash1201 doi101016jbiocon201002027

Fisher R Vigilante T Yates C and Russell-Smith J (2003) Patterns oflandscape fire and predicted vegetation response in the north Kimberleyregion of Western Australia International Journal of Wildland Fire 12369ndash379 doi101071WF03021

Fox J (2002) lsquoAn R and S-plus Companion to Applied Regressionrsquo (SagePublications Thousand Oaks CA)

Gelman A and Hill J (2006) lsquoData Analysis using Regression andMultilevelHierarchicalModelsrsquo (CambridgeUniversity Press London)

Google Inc (2009) Google Earth Pro (Version 7028415) (Mountain ViewCA)

Hadfield J D (2010)MCMCmethods for multi-response generalized linearmixed models the MCMCglmm R Package Journal of StatisticalSoftware 33(2) 1ndash22 doi1018637jssv033i02

Hohnen R Tuft K D Legge S Radford I Carver S and Johnson C N(2015) Post-fire habitat use of the golden-backed tree-rat (Mesembriomysmacrurus) in the north-west Kimberley Western Australia AustralEcology 40 941ndash952 doi101111aec12278

HohnenRTuftKMcGregorHLeggeSRadford I and JohnsonCN(2016)Occupancyof the invasive feral cat varieswith habitat complexityPLoS One in press

Humphreys W F How R A Bradley J Kemper C M and KitchenerD J (1984) The biology of Wyulda squamicaudata Alexander 1919In lsquoPossums and Glidersrsquo (EdS A P Smith and I D Hume)pp 162ndash169 (Surrey Beatty Sydney)

Ingleby S and Westoby M (1992) Habitat requirements of the spectacledhare-wallaby (Lagorchestes conspicillatus) in the Northern Territoryand Western Australia Wildlife Research 19 721ndash741 doi101071WR9920721

Kerle J (1985) Habitat preference and diet of the northern brushtail possumTrichosurus vulpecula in the Alligator Rivers region NT Proceedingsof the Ecological Society of Australia 13 161ndash176

Kerle JA (1998)Thepopulationdynamicsof a tropical possumTrichosurusvulpecula arnhemensis Wildlife Research 25 171ndash181 doi101071WR96113

Leahy L Legge S M Tuft K McGregor H Barmuta L Jones M andJohnson C N (2016) Amplified predation after fire suppresses rodentpopulations in Australiarsquos tropical savannas Wildlife Research 42705ndash716 doi101071WR15011

Legge S Kennedy M S Lloyd R Murphy S A and Fisher A (2011a)Rapid recovery of mammal fauna in the central Kimberley northernAustralia following the removal of introduced herbivores AustralEcology 36 791ndash799 doi101111j1442-9993201002218x

Legge S Murphy S A Kingswood R Maher B and Swan D (2011b)EcoFire restoring the biodiversity values of the Kimberley regionby managing fire Ecological Management amp Restoration 12 84ndash92doi101111j1442-8903201100595x

McGregor H W Legge S Jones M E and Johnson C N (2014)Landscape management of fire and grazing regimes alters the fine-scalehabitat utilisation by feral cats PLoS One 9(10) e109097 doi101371journalpone0109097

McGregor H W Legge S Jones M E and Johnson C N (2015) Feralcats are better killers in open habitats revealed by animal-borne videoPLoS One doi101371journalpone0133915in press

McKenzie N L (1981) lsquoWildlife of the Edgar Ranges Area South-WestKimberley Western Australiarsquo (Western Australia Wildlife ResearchCenter Perth)

Pardon L G Brook B W Griffiths A D and Braithwaite R W (2003)Determinants of survival for the northern brown bandicoot under alandscape-scale fire experiment Journal of Animal Ecology 72 106ndash115doi101046j1365-2656200300686x

PereoglouFMacgregorCBanksS FordFWood J andLindenmayerD (2011) Refuge site selection by the eastern chestnut mouse inrecently burnt heath Wildlife Research 38 290ndash298 doi101071WR11007

Potter S Rosauer D Doody J SWebbM J andEldridgeMD (2014)Persistence of a potentially rare mammalian genus (Wyulda) providesevidence for areasof evolutionary refugiawithin theKimberleyAustraliaConservation Genetics 15 1085ndash1094 doi101007s10592-014-0601-4

Price O Russel-Smith J and Edwards A (2003) Fine-scale patchinessof different fire intensities in sandstone heath vegetation in northernAustralia International Journal of Wildland Fire 12 227ndash236doi101071WF03040

R Development Core Team (2005) R (version 303) In lsquoR A Languageand Environmentrsquo (R Foundation for Statistical Computing Vienna)

Runcie M J (1999) Movements dens and feeding behaviour of the tropicalscaly-tailed possum (Wyulda squamicaudata) Wildlife Research 26367ndash373 doi101071WR98015

Runcie M J (2000) Biparental care and obligate monogamy in the rock-haunting possum Petropseudes dahli from tropical Australia AnimalBehaviour 59 1001ndash1008 doi101006anbe19991392

Russell-Smith J andBowmanDM J S (1992) Conservation ofmonsoonrainforest isolates in the Northern Territory Australia BiologicalConservation 59 51ndash63 doi1010160006-3207(92)90713-W

Russell-Smith J Ryan P G Klessa D Waight G and Harwood R(1998) Fire regimes fire sensitive vegetation and fire management ofthe sandstone Arnhem Plateau monsoonal northern Australia Journal ofApplied Ecology 35 829ndash846 doi101111j1365-26641998tb00002x

Russell-Smith J Edwards A C and Price O F (2012) Simplifying thesavanna the trajectory of fire-sensitive vegetation mosaics in northernAustralia Journal of Biogeography 39 1303ndash1317 doi101111j1365-2699201202679x

Sawle M (1988) A survey of rare and uncommon mammals in theKimberley region of Western Australia MayndashJuly 1988 WesternAustralian Heritage Committee

Smith A P and Quin D G (1996) Patterns and causes of extinctionand decline in Australian conilurine rodentsBiological Conservation 77243ndash267 doi1010160006-3207(96)00002-X

Southgate R Paltridge R Masters P and Carthew S (2007) Bilbydistribution and fire a test of alternative models of habitat suitability inthe Tanami Desert Australia Ecography 30 759ndash776 doi101111j20070906-759004956x

Start A N Burbidge A A McKenzie N L and Palmer C (2007) Thestatus of mammals in the north Kimberley Western Australia AustralianMammalogy 29 1ndash16 doi101071AM07001

Start A N Burbidge A AMcDowell M C andMcKenzie N L (2012)The status of non-volant mammals along a rainfall gradient in the south-west Kimberley Western Australia Australian Mammalogy 34 36ndash48doi101071AM10026

Sawle M (1988) lsquoA Survey of Rare and Uncommon Mammals in theKimberley Region of Western Australia MayndashJuly 1988rsquo (WesternAustralian Heritage Committee Perth)

Therneau T (2014) A package for survival analysis in S R package version237-7

Tuft K D Crowther M S and McArthur C (2011) Multiple scales ofdiet selection by brush-tailed rock-wallabies (Petrogale penicillata)Australian Mammalogy 33 169ndash180 doi101071AM10041

Vernes K and Pope L C (2001) Stability of nest range home range andmovement of the northern bettong (Bettongia tropica) followingmoderate-intensity fire in a tropical woodland north-eastern QueenslandWildlife Research 28 141ndash150 doi101071WR00054

Habitat use by Wyulda squamicaudata Australian Journal of Zoology K

Vigilante T (2001) Analysis of explorersrsquo records of aboriginal landscapeburning in the Kimberley region of Western Australia AustralianGeographical Studies 39 135ndash155 doi1011111467-847000136

Vigilante T and Bowman D (2004a) Effects of fire history on the structureand floristic composition of woody vegetation around Kalumburu northKimberley Australia a landscape-scale natural experiment AustralianJournal of Botany 52 381ndash404 doi101071BT03156

Vigilante T and Bowman D (2004b) Effects of individual fire events onthe flower production of fruit-bearing tree species with reference toAboriginal peoplersquos management and use at Kalumburu northKimberley Australia Australian Journal of Botany 52 405ndash415doi101071BT03157

White G C and Garrot R A (1990) lsquoAnalysis of Wildlife RadiotrackingDatarsquo (Academic Press London)

Williams R J Cook G D Gill A M and Moore P H R (1999) Fireregime fire intensity and tree survival in a tropical savanna in northernAustralia Australian Journal of Ecology 24 50ndash59 doi101046j1442-9993199900946x

Woinarski J C Z (2000) The conservation status of rodents in themonsoonal tropics of the Northern Territory Wildlife Research 27421ndash435 doi101071WR97047

Woinarski J C Z Milne D J and Wanganeen G (2001) Changesin mammal populations in relatively intact landscapes of Kakadu

National Park Northern Territory Australia Austral Ecology 26360ndash370 doi101046j1442-9993200101121x

Woinarski J C Z Legge S Fitzsimons J A Traill B J Burbidge A AFisher A Firth R S C Gordon I J Griffiths A D Johnson C NMcKenzie N L Palmer C Radford I Rankmore B Ritchie E GWard S and Ziembicki M (2011) The disappearing mammal faunaof northern Australia context cause and responseConservation Letters4 192ndash201 doi101111j1755-263X201100164x

Woinarski J Burbidge A and Harrison P (2014) lsquoThe Action Plan forAustralian Mammals 2012rsquo (CSIRO Publishing Melbourne)

Ziembicki M R Woinarski J C Webb J K Vanderduys E Tuft KSmith JRitchieEGReardonTBRadford I J PreeceN Perry JMurphy B P McGregor H Legge S Leahy L Lawes M JKanowski J Johnson C N James A Griffiths A D Gillespie GFrankASK FisherA andBurbidgeAA (2015)Stemming the tideprogress towards resolving the causes of decline and implementingmanagement responses for the disappearing mammal fauna of northernAustralia Therya 6 169ndash226 doi1012933therya-15-236

Handling Editor Paul Cooper

L Australian Journal of Zoology R Hohnen et al

Appendix 1 Dominant plant species available in eight habitat types at three sites in the Artesian Range northwestKimberley Western Australia

Habitat type Overstorey Understorey

Boulder scree Gyrocarpus americanus Santalum albumCeltus philippinensis Cucurbitacea spFicus platypoda Triodia bitexturaAcacia debrilla Bridelia tomentosaEucalyptus brachyandra Dodonaea lanceolata

Rainforest Syzigium angopheroides Hypoestes floribundaCryptocarya cunninghamii Strychnos lucidaXanthostemon paradoxus Ampelocissus acetosaTerminalia hadleyana Atalaya hemiglaucaCeltus philippinensis Wrightia pubescens

River edge Melaleuca leucadendra naMelaleuca argenteaPandanus aquaticusSyzigium eucalyptoidesCeltus philippinensis

Triodia spp savanna Owenia vernicosa Triodia bitexturaCorymbia dendromerinx Grevillea wickhamiiTerminalia canescens Calytrix exstipulataAdansonia gregorii Stemodia lythrifoliaBrachychiton fitzgeraldianus Acacia debrilla

Hibiscus keneallyi savanna Owenia vernicosa Triodia bitexturaCorymbia dendromerinx Trachymene dendrothrixTerminalia canesens Hibiscus keneallyiAdansonia gregorii Grevillea wickhamiiBrachychiton fitzgeraldianus Triumfetta triandra

Sorghum stipoideum savanna Owenia vernicosa Sorghum stipoideumTerminalia hadleyana Triodia sp 2Buchanania obovata Heteropogon contortusAcacia debrilla Chrysopogon fallaxBrachychiton viscidulus Phyllanthus aridus

Chrysopogon fallax savanna Eucalyptus miniata Chrysopogon fallaxEucalyptus polycarpa Cymbopogon procerusPlanchonia careya Sorghum stipoideumPandanus spiralus Stemodia lythrifoliaBuchanania obovata Triodia bitextura

Heteropogon contortus savanna Adansonia gregorii Sorghum stipoideumHakea arborescens Heteropogon contortusEucalyptus tectifica Chrysopogon fallaxBauhinia cunninghamii Triodia sp 2Terminalia canescens Cajanus hirtopilosus

Appendix 2 Home range overlap values

PHR calculates the probability that an animalrsquos location jwill be found within the utilisation distribution of animal i (Fieberg and Kochanny 2005) PRH valuesdiffer between individuals of a dyad as they represent the proportion of home range overlapped by another individual relative to an individualrsquos own utilisationdistribution

PHRi j frac14eth eth

Ai

dUDj ethx yTHORNdxdy

where Ai is the home-range area of animal i anddUDj is the estimated UD of animal j UDOI measures the amount of overlap relative to two individuals using thesame space uniformly (Fieberg andKochanny 2005) This index does not differ between individuals of a dyad and values range from0 (no overlap) to 1 (uniformuse and complete overlap)

UDOI frac14 Ai j

ethyen

yen

ethyen

yen

dUDiethx yTHORN dUDjethx yTHORNdxdy

where Aij is the overlapping area of the two UDs

Habitat use by Wyulda squamicaudata Australian Journal of Zoology M

Appendix 3 Rock complexity composition of sites in the Artesian Range north-west Kimberley Western Australia

0 01 02 04 KmN

Inaccessible

Simple

Medium

Complex

Cliff

Site 1 Site 2

Site 3Rock complexity

Appendix 4 Detail on den use models

The use by Wyulda of a den site was assumed to have a Bernoulli distribution with parameter pij

Yijjpij BernoulliethpijTHORNwhere pij is the probability that observation (i) of an individual at a site (j) being at a used pointWemodelled the probability that an observationwas a used pointpij based on predictor variables The probability of a point being used was modelled as

logitethpijTHORN frac14 aji thorn bixi

wherea andb are themodel intercept and slope for observation i varying by individual at site j and xwas the predictor variable for individual i Prior distributionsfor all model parameters in the hierarchy were given with the goal of providing conjugate priors that contain little to no influence on the posterior distributions ofall the model parameters We assumed normal prior distributions on slopes a and intercept b with mean m and variance s2

aj Normalethmas2aTHORN for j frac14 1 k

For the variance parameters s2 we determined and utilised non-informative uniform prior hyperparameter distributions specified as s2 ~Uniform (0 100)which was used across all models

N Australian Journal of Zoology R Hohnen et al

Appendix 5 Home range estimates and tracking effort for Wyulda squamicaudata at three sites in the Artesian Range north-westKimberley Western Australia

Symbols indicate the same individual tracked on separate month-long occasions

Number Sex Site Season Monthtracked

Recorder No offixes

MCP K95 K50

1 F 1 Wet Jan 13 VHF 47 459 650 1632 F 1 Wet Jan 13 VHF 43 799 267 0673 M 1 Wet Jan 13 VHF 49 580 790 1784 F 2 Wet Feb 13 VHF 48 231 314 0515 F 2 Wet Feb 13 VHF 47 313 377 5556 F 1 Dry Apr 13 GPS 210 1288 985 1607 F 1 Dry Apr 13 GPS 132 772 536 0898$ M 1 Dry Apr 13 GPS 101 797 578 1469 M 1 Dry Apr 13 VHF 130 499 466 08510 F 3 Dry May 13 GPS 144 612 711 16511 M 3 Dry May 13 GPS 68 288 251 04712curren F 3 Dry May 13 GPS 78 414 500 12013 F 3 Dry May 13 GPS 131 343 357 09314 M 3 Dry May 13 GPS 138 664 459 10115curren M 3 Dry May 13 GPS 133 735 610 14116 M 3 Dry May 13 GPS 95 709 920 22117 M 3 Dry May 13 GPS 106 789 1010 30018 F 1 Dry Jan 13 GPS 124 1447 979 20219$ M 1 Dry Jul 13 GPS 75 443 425 06720 F 1 Dry Jul 13 GPS 118 555 670 12621 M 1 Dry Jul 13 GPS 68 379 515 10722 F 2 Dry Aug 13 GPS 114 3441 2895 54823 F 2 Dry Aug 13 GPS 149 1402 856 13124 F 2 Dry Aug 13 GPS 53 281 346 77225 M 2 Dry Aug 13 GPS 97 1164 994 24026 M 2 Dry Aug 13 GPS 96 3699 4052 107927 M 2 Dry Aug 13 GPS 69 934 1290 31328 F 1 Wet Dec 13 GPS 75 180 833 15229 M 1 Wet Dec 13 GPS 108 1228 616 08330curren F 3 Wet Jan 14 GPS 146 254 270 06231 F 3 Wet Jan 14 GPS 92 248 227 04032 F 3 Wet Jan 14 GPS 88 458 533 14733curren M 3 Wet Jan 14 GPS 40 332 602 17634 M 3 Wet Jan 14 GPS 102 1028 1389 342

Habitat use by Wyulda squamicaudata Australian Journal of Zoology O

Appendix 6 The minimum number of Wyulda squamicaudata known to be alive at each site during each trapping period

Month Site Season Males Females Total

January 2013 1 Wet 3 3 6February 2013 2 Wet 2 3 5April 2013 1 Dry 3 2 5May 2013 3 Dry 8 6 14July 2013 1 Dry 3 3 6August 2013 2 Dry 5 4 9December 2013 1 Wet 4 5 9January 2013 3 Wet 3 4 7

Appendix 7 Dietary items identified in faecal samples of Wyulda squamicaudata at three sites in the Artesian Range north-west Kimberleyc average percentage composition per sample s percentage of samples within which that plant was found

Dietary item Total c s Dietary item Total c s

Seed cuticle (fragment) Leaf (fragment) continuedFicus platypoda 2205 4691 8723 Grevillea velutinella 15 032 426Owenia vernicosa 491 1045 3191 Syzygium eucalyptoides 15 032 638Vitex acuminata 477 1015 4043 Trachymene dendrothrix 15 032 426Grewia breviflora 377 802 2979 Acacia dunnii 14 030 426Strychnos lucida 113 240 1915 Distichostemon hispidulus 12 026 213Ficus brachypoda 25 053 638 Eucalyptus tectifica 9 019 213Myristica insipida 19 040 213 Ampelocissus acetosa 8 017 213Syzygium eucalyptoides 18 038 213 Ingofera sp A kimberleyflora 5 011 213Planchonella pohlmaniana 14 030 213 Ficus brachypoda 4 009 213Xanthostemon paradoxus 14 030 213 Fimbristylis sp 4 009 213Unidentified 1 6 013 Smilax australis 4 009 213Terminalia hadleyana 5 011 213 Eucalyptus grandifolia 3 006 213Cryptocarya cunninghamii 4 009 213 Timonius timon 3 006 213Trachymene dendrothrix 3 006 213 Cajanus hirtopilosus 2 004 213

Flower (fragment) Eucalyptus polycarpa 2 004 213Vitex acuminata 86 183 1277 Gonocarpus leptothecus 2 004 213

Leaf (fragment) Grewia breviflora 2 004 213Vitex acuminata 133 283 1915 Hibiscus kenneallyi 2 004 213Owenia vernicosa 81 172 851 Melaleuca minutifolia 2 004 213Cryptocarya cunninghamii 74 157 426 Melaleuca nervosa 2 004 213Acacia dimorpha 49 104 426 Terminalia canescens 2 004 213Chrysopogon fallax 46 098 851 Acacia delibrata 1 002 213Alphitonia excelsa 39 083 426 Grevillea refracta 1 002 213Melaleuca argentea 39 083 426 Gyrocarpus americanus 1 002 213Strychnos lucida 32 068 1277 Seed (whole)Ichnocarpus frutescens 30 064 213 Ficus platypoda 2808 5509 6809Wrightia pubescens 30 064 213 Vitex accuminata 102 1466 2553Eucalyptus rupestris 28 060 638 Seed 3 15 592 851Canarium australianum 23 049 426 Seed 2 512 417 851Syzygium angophoroides 21 045 213 Grewia breviflora 20 413 1064Eucalyptus camaldulensis 20 043 213 Ficus bracypoda 37 123 213Planchonia rupestris 18 038 213 Seed 1 14 117 426Homalanthus novo-guineensis 17 036 426 Trachmene dendrothrix 4 048 426Triodia bitextura 17 036 426 Cryptocraya cunninghamii 1 003 213Macarthuria vertex 16 034 426

P Australian Journal of Zoology R Hohnen et al

Appendix 8 Correlations between (a) the no of fixes and K95 home range estimates and (b) the no of fixes andMCP homerange estimates for Wyulda squamicaudata in the Artesian Range north-west Kimberley Western Australia

(a)

(b)

r2 = 0119

r2 = 0262

Habitat use by Wyulda squamicaudata Australian Journal of Zoology Q

Appendix 9 Generalised linear models describing variation in minimum convex polygon (MCP) and kernel95 contour (K95) home range estimates for Wyulda squamicaudata between sexes and seasons

Home-range estimate Variables AIC DAIC w

MCP Season 7190 000 058Sex + Season 7300 112 033Sex 7700 506 005Null 7720 530 004

K95 Sex + Season 6750 000 032Season 6760 010 030Sex 6790 040 026Null 6930 178 013

Appendix 10 Model-averaged coefficient estimates of explanatory variables describing variation in home rangesize for Wyulda squamicaudata between sexes and seasons

Plt 001

Home-range estimate Variables Coefficient se

MCP Wet season ndash063 023Male 028 023

K95 Wet season ndash040 022Male 037 021

R Australian Journal of Zoology R Hohnen et al

(a)

(b)

(c)

Appendix11 Core (dark grey) and total (light grey) home-rangeoverlapofWyulda squamicaudatadyads for (a) percentagekernel overlap (b) PHR score and (c) UDOI score

Habitat use by Wyulda squamicaudata Australian Journal of Zoology S

wwwpublishcsiroaujournalsajz

Appendix 3 Rock complexity composition of sites in the Artesian Range north-west Kimberley Western Australia

0 01 02 04 KmN

Inaccessible

Simple

Medium

Complex

Cliff

Site 1 Site 2

Site 3Rock complexity

Appendix 4 Detail on den use models

The use by Wyulda of a den site was assumed to have a Bernoulli distribution with parameter pij

Yijjpij BernoulliethpijTHORNwhere pij is the probability that observation (i) of an individual at a site (j) being at a used pointWemodelled the probability that an observationwas a used pointpij based on predictor variables The probability of a point being used was modelled as

logitethpijTHORN frac14 aji thorn bixi

wherea andb are themodel intercept and slope for observation i varying by individual at site j and xwas the predictor variable for individual i Prior distributionsfor all model parameters in the hierarchy were given with the goal of providing conjugate priors that contain little to no influence on the posterior distributions ofall the model parameters We assumed normal prior distributions on slopes a and intercept b with mean m and variance s2

aj Normalethmas2aTHORN for j frac14 1 k

For the variance parameters s2 we determined and utilised non-informative uniform prior hyperparameter distributions specified as s2 ~Uniform (0 100)which was used across all models

N Australian Journal of Zoology R Hohnen et al

Appendix 5 Home range estimates and tracking effort for Wyulda squamicaudata at three sites in the Artesian Range north-westKimberley Western Australia

Symbols indicate the same individual tracked on separate month-long occasions

Number Sex Site Season Monthtracked

Recorder No offixes

MCP K95 K50

1 F 1 Wet Jan 13 VHF 47 459 650 1632 F 1 Wet Jan 13 VHF 43 799 267 0673 M 1 Wet Jan 13 VHF 49 580 790 1784 F 2 Wet Feb 13 VHF 48 231 314 0515 F 2 Wet Feb 13 VHF 47 313 377 5556 F 1 Dry Apr 13 GPS 210 1288 985 1607 F 1 Dry Apr 13 GPS 132 772 536 0898$ M 1 Dry Apr 13 GPS 101 797 578 1469 M 1 Dry Apr 13 VHF 130 499 466 08510 F 3 Dry May 13 GPS 144 612 711 16511 M 3 Dry May 13 GPS 68 288 251 04712curren F 3 Dry May 13 GPS 78 414 500 12013 F 3 Dry May 13 GPS 131 343 357 09314 M 3 Dry May 13 GPS 138 664 459 10115curren M 3 Dry May 13 GPS 133 735 610 14116 M 3 Dry May 13 GPS 95 709 920 22117 M 3 Dry May 13 GPS 106 789 1010 30018 F 1 Dry Jan 13 GPS 124 1447 979 20219$ M 1 Dry Jul 13 GPS 75 443 425 06720 F 1 Dry Jul 13 GPS 118 555 670 12621 M 1 Dry Jul 13 GPS 68 379 515 10722 F 2 Dry Aug 13 GPS 114 3441 2895 54823 F 2 Dry Aug 13 GPS 149 1402 856 13124 F 2 Dry Aug 13 GPS 53 281 346 77225 M 2 Dry Aug 13 GPS 97 1164 994 24026 M 2 Dry Aug 13 GPS 96 3699 4052 107927 M 2 Dry Aug 13 GPS 69 934 1290 31328 F 1 Wet Dec 13 GPS 75 180 833 15229 M 1 Wet Dec 13 GPS 108 1228 616 08330curren F 3 Wet Jan 14 GPS 146 254 270 06231 F 3 Wet Jan 14 GPS 92 248 227 04032 F 3 Wet Jan 14 GPS 88 458 533 14733curren M 3 Wet Jan 14 GPS 40 332 602 17634 M 3 Wet Jan 14 GPS 102 1028 1389 342

Habitat use by Wyulda squamicaudata Australian Journal of Zoology O

Appendix 6 The minimum number of Wyulda squamicaudata known to be alive at each site during each trapping period

Month Site Season Males Females Total

January 2013 1 Wet 3 3 6February 2013 2 Wet 2 3 5April 2013 1 Dry 3 2 5May 2013 3 Dry 8 6 14July 2013 1 Dry 3 3 6August 2013 2 Dry 5 4 9December 2013 1 Wet 4 5 9January 2013 3 Wet 3 4 7

Appendix 7 Dietary items identified in faecal samples of Wyulda squamicaudata at three sites in the Artesian Range north-west Kimberleyc average percentage composition per sample s percentage of samples within which that plant was found

Dietary item Total c s Dietary item Total c s

Seed cuticle (fragment) Leaf (fragment) continuedFicus platypoda 2205 4691 8723 Grevillea velutinella 15 032 426Owenia vernicosa 491 1045 3191 Syzygium eucalyptoides 15 032 638Vitex acuminata 477 1015 4043 Trachymene dendrothrix 15 032 426Grewia breviflora 377 802 2979 Acacia dunnii 14 030 426Strychnos lucida 113 240 1915 Distichostemon hispidulus 12 026 213Ficus brachypoda 25 053 638 Eucalyptus tectifica 9 019 213Myristica insipida 19 040 213 Ampelocissus acetosa 8 017 213Syzygium eucalyptoides 18 038 213 Ingofera sp A kimberleyflora 5 011 213Planchonella pohlmaniana 14 030 213 Ficus brachypoda 4 009 213Xanthostemon paradoxus 14 030 213 Fimbristylis sp 4 009 213Unidentified 1 6 013 Smilax australis 4 009 213Terminalia hadleyana 5 011 213 Eucalyptus grandifolia 3 006 213Cryptocarya cunninghamii 4 009 213 Timonius timon 3 006 213Trachymene dendrothrix 3 006 213 Cajanus hirtopilosus 2 004 213

Flower (fragment) Eucalyptus polycarpa 2 004 213Vitex acuminata 86 183 1277 Gonocarpus leptothecus 2 004 213

Leaf (fragment) Grewia breviflora 2 004 213Vitex acuminata 133 283 1915 Hibiscus kenneallyi 2 004 213Owenia vernicosa 81 172 851 Melaleuca minutifolia 2 004 213Cryptocarya cunninghamii 74 157 426 Melaleuca nervosa 2 004 213Acacia dimorpha 49 104 426 Terminalia canescens 2 004 213Chrysopogon fallax 46 098 851 Acacia delibrata 1 002 213Alphitonia excelsa 39 083 426 Grevillea refracta 1 002 213Melaleuca argentea 39 083 426 Gyrocarpus americanus 1 002 213Strychnos lucida 32 068 1277 Seed (whole)Ichnocarpus frutescens 30 064 213 Ficus platypoda 2808 5509 6809Wrightia pubescens 30 064 213 Vitex accuminata 102 1466 2553Eucalyptus rupestris 28 060 638 Seed 3 15 592 851Canarium australianum 23 049 426 Seed 2 512 417 851Syzygium angophoroides 21 045 213 Grewia breviflora 20 413 1064Eucalyptus camaldulensis 20 043 213 Ficus bracypoda 37 123 213Planchonia rupestris 18 038 213 Seed 1 14 117 426Homalanthus novo-guineensis 17 036 426 Trachmene dendrothrix 4 048 426Triodia bitextura 17 036 426 Cryptocraya cunninghamii 1 003 213Macarthuria vertex 16 034 426

P Australian Journal of Zoology R Hohnen et al

Appendix 8 Correlations between (a) the no of fixes and K95 home range estimates and (b) the no of fixes andMCP homerange estimates for Wyulda squamicaudata in the Artesian Range north-west Kimberley Western Australia

(a)

(b)

r2 = 0119

r2 = 0262

Habitat use by Wyulda squamicaudata Australian Journal of Zoology Q

Appendix 9 Generalised linear models describing variation in minimum convex polygon (MCP) and kernel95 contour (K95) home range estimates for Wyulda squamicaudata between sexes and seasons

Home-range estimate Variables AIC DAIC w

MCP Season 7190 000 058Sex + Season 7300 112 033Sex 7700 506 005Null 7720 530 004

K95 Sex + Season 6750 000 032Season 6760 010 030Sex 6790 040 026Null 6930 178 013

Appendix 10 Model-averaged coefficient estimates of explanatory variables describing variation in home rangesize for Wyulda squamicaudata between sexes and seasons

Plt 001

Home-range estimate Variables Coefficient se

MCP Wet season ndash063 023Male 028 023

K95 Wet season ndash040 022Male 037 021

R Australian Journal of Zoology R Hohnen et al

(a)

(b)

(c)

Appendix11 Core (dark grey) and total (light grey) home-rangeoverlapofWyulda squamicaudatadyads for (a) percentagekernel overlap (b) PHR score and (c) UDOI score

Habitat use by Wyulda squamicaudata Australian Journal of Zoology S

wwwpublishcsiroaujournalsajz

Appendix 5 Home range estimates and tracking effort for Wyulda squamicaudata at three sites in the Artesian Range north-westKimberley Western Australia

Symbols indicate the same individual tracked on separate month-long occasions

Number Sex Site Season Monthtracked

Recorder No offixes

MCP K95 K50

1 F 1 Wet Jan 13 VHF 47 459 650 1632 F 1 Wet Jan 13 VHF 43 799 267 0673 M 1 Wet Jan 13 VHF 49 580 790 1784 F 2 Wet Feb 13 VHF 48 231 314 0515 F 2 Wet Feb 13 VHF 47 313 377 5556 F 1 Dry Apr 13 GPS 210 1288 985 1607 F 1 Dry Apr 13 GPS 132 772 536 0898$ M 1 Dry Apr 13 GPS 101 797 578 1469 M 1 Dry Apr 13 VHF 130 499 466 08510 F 3 Dry May 13 GPS 144 612 711 16511 M 3 Dry May 13 GPS 68 288 251 04712curren F 3 Dry May 13 GPS 78 414 500 12013 F 3 Dry May 13 GPS 131 343 357 09314 M 3 Dry May 13 GPS 138 664 459 10115curren M 3 Dry May 13 GPS 133 735 610 14116 M 3 Dry May 13 GPS 95 709 920 22117 M 3 Dry May 13 GPS 106 789 1010 30018 F 1 Dry Jan 13 GPS 124 1447 979 20219$ M 1 Dry Jul 13 GPS 75 443 425 06720 F 1 Dry Jul 13 GPS 118 555 670 12621 M 1 Dry Jul 13 GPS 68 379 515 10722 F 2 Dry Aug 13 GPS 114 3441 2895 54823 F 2 Dry Aug 13 GPS 149 1402 856 13124 F 2 Dry Aug 13 GPS 53 281 346 77225 M 2 Dry Aug 13 GPS 97 1164 994 24026 M 2 Dry Aug 13 GPS 96 3699 4052 107927 M 2 Dry Aug 13 GPS 69 934 1290 31328 F 1 Wet Dec 13 GPS 75 180 833 15229 M 1 Wet Dec 13 GPS 108 1228 616 08330curren F 3 Wet Jan 14 GPS 146 254 270 06231 F 3 Wet Jan 14 GPS 92 248 227 04032 F 3 Wet Jan 14 GPS 88 458 533 14733curren M 3 Wet Jan 14 GPS 40 332 602 17634 M 3 Wet Jan 14 GPS 102 1028 1389 342

Habitat use by Wyulda squamicaudata Australian Journal of Zoology O

Appendix 6 The minimum number of Wyulda squamicaudata known to be alive at each site during each trapping period

Month Site Season Males Females Total

January 2013 1 Wet 3 3 6February 2013 2 Wet 2 3 5April 2013 1 Dry 3 2 5May 2013 3 Dry 8 6 14July 2013 1 Dry 3 3 6August 2013 2 Dry 5 4 9December 2013 1 Wet 4 5 9January 2013 3 Wet 3 4 7

Appendix 7 Dietary items identified in faecal samples of Wyulda squamicaudata at three sites in the Artesian Range north-west Kimberleyc average percentage composition per sample s percentage of samples within which that plant was found

Dietary item Total c s Dietary item Total c s

Seed cuticle (fragment) Leaf (fragment) continuedFicus platypoda 2205 4691 8723 Grevillea velutinella 15 032 426Owenia vernicosa 491 1045 3191 Syzygium eucalyptoides 15 032 638Vitex acuminata 477 1015 4043 Trachymene dendrothrix 15 032 426Grewia breviflora 377 802 2979 Acacia dunnii 14 030 426Strychnos lucida 113 240 1915 Distichostemon hispidulus 12 026 213Ficus brachypoda 25 053 638 Eucalyptus tectifica 9 019 213Myristica insipida 19 040 213 Ampelocissus acetosa 8 017 213Syzygium eucalyptoides 18 038 213 Ingofera sp A kimberleyflora 5 011 213Planchonella pohlmaniana 14 030 213 Ficus brachypoda 4 009 213Xanthostemon paradoxus 14 030 213 Fimbristylis sp 4 009 213Unidentified 1 6 013 Smilax australis 4 009 213Terminalia hadleyana 5 011 213 Eucalyptus grandifolia 3 006 213Cryptocarya cunninghamii 4 009 213 Timonius timon 3 006 213Trachymene dendrothrix 3 006 213 Cajanus hirtopilosus 2 004 213

Flower (fragment) Eucalyptus polycarpa 2 004 213Vitex acuminata 86 183 1277 Gonocarpus leptothecus 2 004 213

Leaf (fragment) Grewia breviflora 2 004 213Vitex acuminata 133 283 1915 Hibiscus kenneallyi 2 004 213Owenia vernicosa 81 172 851 Melaleuca minutifolia 2 004 213Cryptocarya cunninghamii 74 157 426 Melaleuca nervosa 2 004 213Acacia dimorpha 49 104 426 Terminalia canescens 2 004 213Chrysopogon fallax 46 098 851 Acacia delibrata 1 002 213Alphitonia excelsa 39 083 426 Grevillea refracta 1 002 213Melaleuca argentea 39 083 426 Gyrocarpus americanus 1 002 213Strychnos lucida 32 068 1277 Seed (whole)Ichnocarpus frutescens 30 064 213 Ficus platypoda 2808 5509 6809Wrightia pubescens 30 064 213 Vitex accuminata 102 1466 2553Eucalyptus rupestris 28 060 638 Seed 3 15 592 851Canarium australianum 23 049 426 Seed 2 512 417 851Syzygium angophoroides 21 045 213 Grewia breviflora 20 413 1064Eucalyptus camaldulensis 20 043 213 Ficus bracypoda 37 123 213Planchonia rupestris 18 038 213 Seed 1 14 117 426Homalanthus novo-guineensis 17 036 426 Trachmene dendrothrix 4 048 426Triodia bitextura 17 036 426 Cryptocraya cunninghamii 1 003 213Macarthuria vertex 16 034 426

P Australian Journal of Zoology R Hohnen et al

Appendix 8 Correlations between (a) the no of fixes and K95 home range estimates and (b) the no of fixes andMCP homerange estimates for Wyulda squamicaudata in the Artesian Range north-west Kimberley Western Australia

(a)

(b)

r2 = 0119

r2 = 0262

Habitat use by Wyulda squamicaudata Australian Journal of Zoology Q

Appendix 9 Generalised linear models describing variation in minimum convex polygon (MCP) and kernel95 contour (K95) home range estimates for Wyulda squamicaudata between sexes and seasons

Home-range estimate Variables AIC DAIC w

MCP Season 7190 000 058Sex + Season 7300 112 033Sex 7700 506 005Null 7720 530 004

K95 Sex + Season 6750 000 032Season 6760 010 030Sex 6790 040 026Null 6930 178 013

Appendix 10 Model-averaged coefficient estimates of explanatory variables describing variation in home rangesize for Wyulda squamicaudata between sexes and seasons

Plt 001

Home-range estimate Variables Coefficient se

MCP Wet season ndash063 023Male 028 023

K95 Wet season ndash040 022Male 037 021

R Australian Journal of Zoology R Hohnen et al

(a)

(b)

(c)

Appendix11 Core (dark grey) and total (light grey) home-rangeoverlapofWyulda squamicaudatadyads for (a) percentagekernel overlap (b) PHR score and (c) UDOI score

Habitat use by Wyulda squamicaudata Australian Journal of Zoology S

wwwpublishcsiroaujournalsajz

Appendix 6 The minimum number of Wyulda squamicaudata known to be alive at each site during each trapping period

Month Site Season Males Females Total

January 2013 1 Wet 3 3 6February 2013 2 Wet 2 3 5April 2013 1 Dry 3 2 5May 2013 3 Dry 8 6 14July 2013 1 Dry 3 3 6August 2013 2 Dry 5 4 9December 2013 1 Wet 4 5 9January 2013 3 Wet 3 4 7

Appendix 7 Dietary items identified in faecal samples of Wyulda squamicaudata at three sites in the Artesian Range north-west Kimberleyc average percentage composition per sample s percentage of samples within which that plant was found

Dietary item Total c s Dietary item Total c s

Seed cuticle (fragment) Leaf (fragment) continuedFicus platypoda 2205 4691 8723 Grevillea velutinella 15 032 426Owenia vernicosa 491 1045 3191 Syzygium eucalyptoides 15 032 638Vitex acuminata 477 1015 4043 Trachymene dendrothrix 15 032 426Grewia breviflora 377 802 2979 Acacia dunnii 14 030 426Strychnos lucida 113 240 1915 Distichostemon hispidulus 12 026 213Ficus brachypoda 25 053 638 Eucalyptus tectifica 9 019 213Myristica insipida 19 040 213 Ampelocissus acetosa 8 017 213Syzygium eucalyptoides 18 038 213 Ingofera sp A kimberleyflora 5 011 213Planchonella pohlmaniana 14 030 213 Ficus brachypoda 4 009 213Xanthostemon paradoxus 14 030 213 Fimbristylis sp 4 009 213Unidentified 1 6 013 Smilax australis 4 009 213Terminalia hadleyana 5 011 213 Eucalyptus grandifolia 3 006 213Cryptocarya cunninghamii 4 009 213 Timonius timon 3 006 213Trachymene dendrothrix 3 006 213 Cajanus hirtopilosus 2 004 213

Flower (fragment) Eucalyptus polycarpa 2 004 213Vitex acuminata 86 183 1277 Gonocarpus leptothecus 2 004 213

Leaf (fragment) Grewia breviflora 2 004 213Vitex acuminata 133 283 1915 Hibiscus kenneallyi 2 004 213Owenia vernicosa 81 172 851 Melaleuca minutifolia 2 004 213Cryptocarya cunninghamii 74 157 426 Melaleuca nervosa 2 004 213Acacia dimorpha 49 104 426 Terminalia canescens 2 004 213Chrysopogon fallax 46 098 851 Acacia delibrata 1 002 213Alphitonia excelsa 39 083 426 Grevillea refracta 1 002 213Melaleuca argentea 39 083 426 Gyrocarpus americanus 1 002 213Strychnos lucida 32 068 1277 Seed (whole)Ichnocarpus frutescens 30 064 213 Ficus platypoda 2808 5509 6809Wrightia pubescens 30 064 213 Vitex accuminata 102 1466 2553Eucalyptus rupestris 28 060 638 Seed 3 15 592 851Canarium australianum 23 049 426 Seed 2 512 417 851Syzygium angophoroides 21 045 213 Grewia breviflora 20 413 1064Eucalyptus camaldulensis 20 043 213 Ficus bracypoda 37 123 213Planchonia rupestris 18 038 213 Seed 1 14 117 426Homalanthus novo-guineensis 17 036 426 Trachmene dendrothrix 4 048 426Triodia bitextura 17 036 426 Cryptocraya cunninghamii 1 003 213Macarthuria vertex 16 034 426

P Australian Journal of Zoology R Hohnen et al

Appendix 8 Correlations between (a) the no of fixes and K95 home range estimates and (b) the no of fixes andMCP homerange estimates for Wyulda squamicaudata in the Artesian Range north-west Kimberley Western Australia

(a)

(b)

r2 = 0119

r2 = 0262

Habitat use by Wyulda squamicaudata Australian Journal of Zoology Q

Appendix 9 Generalised linear models describing variation in minimum convex polygon (MCP) and kernel95 contour (K95) home range estimates for Wyulda squamicaudata between sexes and seasons

Home-range estimate Variables AIC DAIC w

MCP Season 7190 000 058Sex + Season 7300 112 033Sex 7700 506 005Null 7720 530 004

K95 Sex + Season 6750 000 032Season 6760 010 030Sex 6790 040 026Null 6930 178 013

Appendix 10 Model-averaged coefficient estimates of explanatory variables describing variation in home rangesize for Wyulda squamicaudata between sexes and seasons

Plt 001

Home-range estimate Variables Coefficient se

MCP Wet season ndash063 023Male 028 023

K95 Wet season ndash040 022Male 037 021

R Australian Journal of Zoology R Hohnen et al

(a)

(b)

(c)

Appendix11 Core (dark grey) and total (light grey) home-rangeoverlapofWyulda squamicaudatadyads for (a) percentagekernel overlap (b) PHR score and (c) UDOI score

Habitat use by Wyulda squamicaudata Australian Journal of Zoology S

wwwpublishcsiroaujournalsajz

Appendix 8 Correlations between (a) the no of fixes and K95 home range estimates and (b) the no of fixes andMCP homerange estimates for Wyulda squamicaudata in the Artesian Range north-west Kimberley Western Australia

(a)

(b)

r2 = 0119

r2 = 0262

Habitat use by Wyulda squamicaudata Australian Journal of Zoology Q

Appendix 9 Generalised linear models describing variation in minimum convex polygon (MCP) and kernel95 contour (K95) home range estimates for Wyulda squamicaudata between sexes and seasons

Home-range estimate Variables AIC DAIC w

MCP Season 7190 000 058Sex + Season 7300 112 033Sex 7700 506 005Null 7720 530 004

K95 Sex + Season 6750 000 032Season 6760 010 030Sex 6790 040 026Null 6930 178 013

Appendix 10 Model-averaged coefficient estimates of explanatory variables describing variation in home rangesize for Wyulda squamicaudata between sexes and seasons

Plt 001

Home-range estimate Variables Coefficient se

MCP Wet season ndash063 023Male 028 023

K95 Wet season ndash040 022Male 037 021

R Australian Journal of Zoology R Hohnen et al

(a)

(b)

(c)

Appendix11 Core (dark grey) and total (light grey) home-rangeoverlapofWyulda squamicaudatadyads for (a) percentagekernel overlap (b) PHR score and (c) UDOI score

Habitat use by Wyulda squamicaudata Australian Journal of Zoology S

wwwpublishcsiroaujournalsajz

Appendix 9 Generalised linear models describing variation in minimum convex polygon (MCP) and kernel95 contour (K95) home range estimates for Wyulda squamicaudata between sexes and seasons

Home-range estimate Variables AIC DAIC w

MCP Season 7190 000 058Sex + Season 7300 112 033Sex 7700 506 005Null 7720 530 004

K95 Sex + Season 6750 000 032Season 6760 010 030Sex 6790 040 026Null 6930 178 013

Appendix 10 Model-averaged coefficient estimates of explanatory variables describing variation in home rangesize for Wyulda squamicaudata between sexes and seasons

Plt 001

Home-range estimate Variables Coefficient se

MCP Wet season ndash063 023Male 028 023

K95 Wet season ndash040 022Male 037 021

R Australian Journal of Zoology R Hohnen et al

(a)

(b)

(c)

Appendix11 Core (dark grey) and total (light grey) home-rangeoverlapofWyulda squamicaudatadyads for (a) percentagekernel overlap (b) PHR score and (c) UDOI score

Habitat use by Wyulda squamicaudata Australian Journal of Zoology S

wwwpublishcsiroaujournalsajz

(a)

(b)

(c)

Appendix11 Core (dark grey) and total (light grey) home-rangeoverlapofWyulda squamicaudatadyads for (a) percentagekernel overlap (b) PHR score and (c) UDOI score

Habitat use by Wyulda squamicaudata Australian Journal of Zoology S

wwwpublishcsiroaujournalsajz