Transcript
ORIGINALARTICLE The interdependence of fire, grass,kangaroos and Australian Absrcines: acase study from central Arnhem Land,northern Australia Brett P. Murphy* and David M. J. S. Bowman School for Environmental Research, CharlesDarwin University, Darwin, NT 0909, Australia *Correspondence: Brett Murphy, School forEnvironmental Research, Charles DarwinUniversity, Darwin, NT 0909, Australia.E-mail:
[email protected] ABSTRACT Aim To describe the nexus between Absrcinal landscape burning and patterns of habitat use by kangaroos in a tropical savanna habitat mosaic, and to provideevidence to evaluate the claim that Absrcinal landscape burning is a gamemanagement tool. Location Central Arnhem Land, a stronghold of traditional Absrcinal culture,in the monsoon tropics of northern Australia. Methods The abundance of kangaroo scats was recorded throughout alandscape burnt by Absrcinal people, and used as a proxy for the intensity of habitat use by kangaroos. Scat abundance was assessed along field traversestotalling 112 km, at three time periods: (1) 1–4 weeks following mid-dry seasonburning (July 2003); (2) in the late dry season (November 2003); and (3) in thefollowing mid-dry season (July 2004). We compared the intensity with whichkangaroos used burnt vs. unburnt areas in various habitat types, with time sincemid-dry season burning. Scats were collected from areas that had been burnt to avarying extent and the abundance of carbon and nitrogen stable isotopes ( d 13 Cand d 15 N) and carbon to nitrogen ratios (C : N) determined. Results There was clear evidence of an interaction between burning and habitattype on the abundance of kangaroo scats. Scats were much more abundant inburnt vs. unburnt areas in the moist habitats, but the opposite effect was observedin the dry rocky habitats, with higher scat abundance in unburnt areas. Thisinteractive effect of burning and habitat type on scat abundance was observedimmediately (< 4 weeks) following fire, and was still present one year later. Highconcentrations of nitrogen in resprouting grasses indicate that burnt areas may provide kangaroos with greater access to nutrients. The isotopic composition of scats indicates that kangaroos feeding in extensively burnt areas were consumingmore grasses, and possibly sedges, than kangaroos feeding in unburnt areas. Main conclusions The fine-scale mosaic of burnt and unburnt areas created by mid-dry season Absrcinal landscape burning has clear effects on the distributionof kangaroos. Kangaroos move into burnt moist habitats and away from burntdry, rocky habitats. Isotopic analysis of scats suggests that the mechanism drivingthis effect is the increased abundance of nitrogen rich grasses in burnt moisthabitats. Keywords Australia, indigenous ecological knowledge, landscape fire, landscape ecology,macropod, Macropus , stable isotopes, tropical savanna, wildfire management. Journal of Biogeography ( J. Biogeogr .) (2007) 34 , 237–250 ª 2006 The Authors www.blackwellpublishing.com/jbi 237 Journal compilation ª 2006 Blackwell Publishing Ltd doi:10.1111/j.1365-2699.2006.01591.x INTRODUCTION In 1848, the European explorer Thomas Mitchell suggestedthat if it were not for Absrcinal landscape burning, southernAustralia would be covered in ‘thick jungle’ rather than openforests. He explained the logic of Absrcinal burning with hissuccinct and oft-quoted passage: ‘fire, grass, kangaroos, andhuman inhabitants seem all dependent upon each other forexistence in Australia’ (Mitchell, 1848). It is remarkable thatdespite considerable controversies about the positive andnegative short- and longer term effects of Absrcinal burning,Mitchell’s conjecture has received little scientific attention.Indeed, despite the importance of wildland fire management,little is known about the motives, geographic scale andecological importance of indigenous fire managementthroughout the world. The absence of firsthand data make itdifficult to evaluate the hypothesis that Absrcinal burning hadan initial catastrophic impact on the Australian environmentsuch as that postulated by Flannery (1994) and Miller et al. (2005) or reconcile these views with the countervailingarguments based upon more recent historical data (e.g.Bowman, 1998). The Australian debate has parallels in NorthAmerica where there is also a spectrum of views concerning theecological effects of native American landscape burning,ranging from negligible (e.g. Grissino-Mayer et al. , 2004), tovery specialized ecological outcomes, particularly increasingthe abundance of economically valuable plant species (Ander-son, 2002).The recent ethnographic work of Bowman et al. (2001)emphasized that among the many uses of fire in centralArnhem Land, northern Australia, the most important relate tothe hunting of kangaroos, and a similar conclusion wasreached by Vigilante (2003) working in the Kimberley regionof north-western Australia. The use of fire as a hunting tool hasbeen widely reported, being used to flush kangaroos or drivethem toward waiting hunters (Finlayson, 1946; Hallam, 1975;Kimber, 1983; Russell-Smith et al. , 1997; Bowman et al. , 2001;Vigilante, 2003). The less direct uses of fire in kangaroohunting incorporate aspects of habitat manipulation. Fire isreported to have been used to produce young grass shoots inburnt areas, which attract kangaroos, making them easier tolocate and hunt (Kimber, 1983; Bright, 1994; Bowman & Vigilante, 2001). Regenerating areas are also thought toprovide improved grazing, increasing the carrying capacity for kangaroos (Gould, 1971; Haynes, 1991; Latz, 1995; Russell-Smith et al. , 1997; Bowman & Vigilante, 2001). Such use of fireas a land management tool led Jones (1969) to coin the phrase‘fire-stick farming’, although Bowman et al. (2001) suggestthat ‘fire-stick ranching’ might be more apt given the centrality of game management.Bowman & Prior (2004) point out that while ‘fire-stick farming’ is now a generally accepted concept, little effort hasever been made to test it. Indeed, despite abundant anecdotalevidence of Absrcinal use of fire to attract kangaroos andmaintain kangaroo populations, little empirical evidence exists.There are numerous ecological studies from other parts of Australia that show that some kangaroo species are attracted toburnt areas (Noble et al. , 1986; Southwell & Jarman, 1987;Lundie-Jenkins, 1993; McCullough & McCullough, 2000;Meers & Adams, 2003). To the best of our knowledge, theonly quantitative evidence of kangaroos being attracted toburnt areas in monsoonal northern Australia is the brief account provided by Yibarbuk et al. (2001). However, therehas been little investigation of the mechanisms that cause thisresponse by kangaroos, although it is widely assumed thatkangaroos are attracted initially to exposed green remnants,and later to newly emerging shoots (Southwell & Jarman,1987). Evidence from North America suggests that grazers areable to locate nitrogen rich, easily digestible food sources innewly burnt areas (Hobbs & Spowart, 1984).This study focuses on an area in central Arnhem Land wherethe land management practices of local Absrcinal people havechanged little since European settlement. The area haspreviously been used to explore various effects of Absrcinalland management on the biological landscape (Yibarbuk et al. ,2001; Bowman & Prior, 2004; Bowman et al. , 2004). Weexplore the ecological rationale of Aboriginal landscapeburning by examining its effects on patterns of habitat useby kangaroos and the stable isotope composition of kangarooscats. METHODSStudy area Arnhem Land is a large area ( c. 97 000 km 2 ) in northernAustralia, proclaimed an Absrcinal reserve in the 1930s andowned by an Absrcinal land trust since the 1970s (Fig. 1).Today it is largely managed to support a traditional subsistenceeconomy. Korlorrbirrahda is a small isolated community incentral Arnhem Land, based around one extended family andusually home to fewer than 20 people. Such a community isknown in northern Australia as an ‘outstation’. The clan estatesurrounding Korlorrbirrahda is relatively unusual in havinghad a continuous history of Aboriginal occupation sinceEuropean colonization, interrupted for only several years inthe 1950s (Yibarbuk et al. , 2001). Detailed descriptions of thestudy area are provided by Yibarbuk et al. (2001) and Bowman& Prior (2004). Field traverses In early July 2003, we witnessed local Absrcinal men settingfire to vegetation around two localities near Korlorrbirrahdaoutstation: Dukaladjarranj, a site frequently used by localAbsrcinal people for camping; and Yaiminyi, an abandonedoutstation (Fig. 1). We were accompanying the men onkangaroo hunts as part of fieldwork for an unrelated projectconcerning the isotopic composition of kangaroo bones. Theirsetting of fires was unsolicited by us.We returned to the areas where the fires were lit between1 week and 4 weeks later, and undertook a series of field B. P. Murphy and D. M. J. S. Bowman 238 Journal of Biogeography 34 , 237–250 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd traverses on foot. Five traverses, totalling 60.6 km, wereundertaken from Dukaladjarranj and five traverses, totalling51.2 km, were undertaken from Yaiminyi (Fig. 1). Theroutes taken were recorded with a handheld GlobalPositioning System (GPS). GPS waypoints were used tosplit the routes into sections that were each classified as oneof five habitat types common in the study area (Table 1).The sections were also classified as either burnt orunburnt in the current dry season. Although most burntsections had been burnt 1 week previously, some sectionshad been burnt earlier. The ten traverses were repeated inNovember 2003 and July 2004 with the same informationrecorded. Scat abundance Our study was restricted to members of the kangaroo genus Macropus , represented in the study area by M. agilis Gould(agile wallaby), M. antilopinus Gould (antilopine wallaroo), M. bernardus Rothschild (black wallaroo) and M robustus Gould (common wallaroo). Throughout this paper, we use theterm ‘kangaroo’ to refer to any Macropus species.The abundance of kangaroo scats is a good proxy for theintensity of utilization of an area as a feeding site by kangaroos,because kangaroos mainly defecate while feeding (Johnson et al. , 1987). This approach has been used successfully innumerous studies of kangaroos and other macropods (Floyd,1980; Johnson & Jarman, 1987; Lundie-Jenkins, 1993). Whilekangaroo scats are easily distinguishable from other animals inthe study area (Telfer et al. , in press), no attempt was made todistinguish the scats of the four kangaroo species. Telfer et al. (in press) found that while the scats of the kangaroo speciespresent in the study area have minor differences in size andshape, their attempts to distinguish them met with only limited success (> 30% misclassification rate).During each traverse, we recorded the number of ‘scatencounters’ within each section delineated by GPS way-points. Scat encounters were defined as the number of linearmetres along the route where one or more scats were presentwithin 50 cm of the centreline of the route. We created an Figure 1 Location of the study area withincentral Arnhem Land (a, b). The locations of the ten traverses made from Dukaladjarranjand Yaiminyi are shown within the study area (c). Habitat types are shown in relationto the field traverses (d, e), however the twohabitat types ‘rocky savanna’ and ‘rocky washouts’ are not differentiated. Habitatdescriptions are provided in Table 1. Grass, kangaroos and Absrcinal fire management Journal of Biogeography 34 , 237–250 239 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd index of scat abundance, based on probability of encoun-tering scats:Scat abundance index ¼ Scate encountersDistance walked (m) : To minimize the confounding effect of older scats accumu-lating in unburnt sites, we only recorded fresh scats thatretained their shiny black coating without any obviouscracking or deterioration. Scats such as this have an approxi-mate age of up to three months (Telfer et al. , in press). Assessment of scat visibility In other areas, the accuracy of scat counts has been found to beinfluenced by vegetation characteristics such as height andcover (Bulinski & McArthur, 2000). We therefore set out toconfirm there were not major differences in the visibility of scats in different habitats and in burnt vs. unburnt areas. To dothis, we conducted a simple exercise in bushland c. 10 kmsouth of the city of Darwin, where at seven different locations,ten strings 200 m in length were laid out in random directions.Twenty groups of small painted pebbles, roughly the same size,shape and colour as kangaroo scats, were randomly placedalong the length of the string, within 50 cm of it. Group sizesranged randomly from one to seven. Later in the day, one of uswalked quickly along the length of the strings, counting thegroups of artificial scats seen. This allowed us to estimate scatvisibility as a proportion of artificial scat groups seen. Theexercise was carried out in unburnt and recently burnt areas inthree habitat types: rocky savanna, sandsheet savanna andswamp. In riverine habitat, no recently burnt areas could belocated, so the validation exercise was carried out in unburntareas only. These habitat types were similar to those foundnear Korlorrbirrahda outstation, with the exception of rocky washouts which could not be located in this area.The magnitude of the differences in scat visibility betweenhabitats, and unburnt vs. burnt areas was small ( 6 6.5%,Table 2). We concluded that these differences in scat visibility would not greatly affect the results of this study. Scat and grass collection and analysis To examine the effect of burning on scat composition, 282scats samples were collected from different locations duringthe field traverses and the collection locations recorded. Thescats were air-dried, the shiny outer coatings were removedand a 1-cm 3 sample of each scat was ground to a fine powder.Each powdered scat sample was analysed for carbon andnitrogen isotope abundances ( d 13 C and d 15 N, respectively) andcarbon to nitrogen ratios (C : N) with an EA 1110 elementalanalyser (CE Instruments, Rodano, Italy) coupled with an Table 1 Habitat types common in the study area near Korlorrbirrahda outstation, central Arnhem Land. Detailed descriptions of mosthabitat types are provided by Bowman et al. (2004). The nomenclature used by Bowman et al. is also indicated. The habitat types ‘swamp’and ‘rocky washouts’ were not defined by Bowman et al. Habitat type Topography VegetationAreal extent instudy area (%)Bowman et al. (2004)nomenclatureRiverine Fringing larger creeklines Dense overstorey dominated by Melaleuca leucadendra (L.) L., andunderstorey dominated by perennial grasses2 RiverineRocky savanna Areas of rugged, exposedsandstoneLimited to low shrubs and highly flammable hummock grasses( Triodia and Plectrachne spp.).Emergent Eucalyptus spp. present36 Sandstone savannaRocky washouts Adjacent to rocky creeklines,where thin layers of sand havebeen deposited betweensandstone rocksSparse vegetation, dominated by thelow trees Acacia torulosa Benth. ex.F. Muell. and Grevillea pteridifolia Knight, with very sparse annualgrasses2 n.a.Sandsheet savanna Extensive flat areas of deeper sandsOpen forest of Eucalyptus tetrodonta F. Muell. with a sparse understorey 54 Plateau savannaSwamp Seasonally wet depressions andareas adjacent to creeksFew trees, with a ground layerdominated by sedges and perennialgrasses7 n.a. Table 2 Scat groups counted (%) in the scat visibility exerciseconducted in unburnt and burnt areas in different habitats.Standard errors are shown Habitat Unburnt BurntRiverine 93.5 ± 1.7 –Rocky savanna 94.0 ± 1.7 93.5 ± 1.7Sandsheet savanna 94.0 ± 1.7 100.0 ± 0.0Swamp 92.0 ± 1.9 98.5 ± 0.9 B. P. Murphy and D. M. J. S. Bowman 240 Journal of Biogeography 34 , 237–250 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd IsoChrom mass spectrometer (Micromass, Manchester, UK). d 13 C and d 15 N are expressed in per mil ( & ) units, relative tothe Vienna Pee Dee Belemnite and atmospheric N 2 standards,respectively.Scat d 13 C was used as a proxy for the proportion of the dietcontributed by plants using the C 4 photosynthetic pathway (Jones et al. , 1979). This is possible because C 3 plants have amean d 13 C of around ) 26.5 & and C 4 plants have a mean d 13 Cof around ) 12.5 & (van der Merwe, 1982; Cerling et al. , 1997).While the faeces of herbivores tends to have slightly lower d 13 Cthan their diet (0.4–2.0 & ), the relationship between d 13 C of faeces and diet is nearly linear, with a slope close to unity (Jones et al. , 1979). This means that a change of 1.4 & in the d 13 C of herbivore faeces can be interpreted as a change of approximately 0.1 in the proportion of the diet contributed by C 4 plants.In northern Australia, C 4 plants are mainly grasses, andnearly all grasses are C 4 (Hattersley, 1983; Bowman & Cook,2002). This allows us to assume that the contribution of C 4 plants to the diet is equivalent to the contribution of grasses tothe diet. However, this approach did not allow us to assess thedietary contribution of sedges that dominate the swamphabitats because these graminoids often possess the C 3 photosynthetic pathway (Teeri et al. , 1980; Ueno & Takeda,1992; Stock et al. , 2004).In herbivores, scat d 15 N tends to reflect d 15 N of the plantmaterial consumed (Codron et al. , 2005), and fire events tendto increase soil, and hence plant d 15 N (Grogan et al. , 2000).For this reason, we analysed scat d 15 N to determine the extentto which herbivores had been grazing on regrowth following afire. During July 2005, we collected 30 samples of grass foliagefrom unburnt areas and 30 samples from areas burntapproximately 1 month previously. In the unburnt areas, only old, senescent grasses were available, and in the recently burntareas, only fresh resprouting grasses were available. Thesamples were dried, ground to a fine powder, and analysedfor nitrogen isotope abundances ( d 15 N) and total nitrogencontent with a 20/20 Automated Nitrogen Carbon Analysismass spectrometer (Europa Scientific, Crewe, UK). Data analysis To analyse the data from this study, we used an information-theoretic approach, sensu Burnham & Anderson (2001, 2002).This approach utilizes a set of multiple hypotheses, or models,derived a priori to any data analysis. The models are ranked forsuitability using Akaike’s Information Criterion (AIC), thatfavours both model fit and model simplicity. Lower values of AIC indicate greater support for a model, relative to othermodels in the same candidate set. Burnham & Anderson(2002) suggest that a model with an AIC value within 2 of thelowest AIC value in the candidate set can be considered wellsupported, and a model with an AIC value within 10 of thelowest can be considered only weakly supported. For eachmodel in the candidate set, AIC can be used to calculate anAkaike weight (w i ), which represents the probability of thatmodel being the best in the candidate set (Burnham & Anderson, 2002).A set of 13 ecologically plausible candidate models wasdeveloped a priori to explain the observed variation in scatabundance (Table 3). Three predictor variables were includedin the models: 1. habitat type (Table 1); 2. burnt: burnt in July 2003, or unburnt since early 2003; and 3. season: July 2003 (1–4 weeks after fire), November 2003(4 months after fire), or July 2004 (1 year after fire).The candidate models were constructed as generalized linearmodels with a binomial error structure (i.e. logistic regressionmodels), in the computer programme R (Ihaka & Gentleman,1996). The suitability of each model was assessed using QAIC c ,a second order form of AIC, corrected to account for smallsample size and overdispersion in the data (Burnham & Anderson, 2002). Where more than one model had a high levelof empirical support (QAIC c within 2 of the best model), thepredictions of all the highly supported models were weightedaccording to w i and combined (Burnham & Anderson, 2002).To provide additional confirmation that any differences inscat abundance between unburnt and burnt areas were not theresult of differences in scat visibility or destruction of accumulated scats during burning, we repeated the modellingprocedure discussed previously, but only used observationsfrom unburnt areas. In this case, the variable ‘burnt’ wasreplaced with the variable ‘distance’ which was equivalent tothe distance in metres to the nearest area burnt in July 2003.The means of nitrogen content and d 15 N of grass samplescollected in burnt and unburnt areas were compared usinganalysis of variance ( anova ) in the computer programme R.The anova model was compared to a null model (i.e.assuming no difference between the two means) on the basisof AIC c , a second order form of AIC, corrected to account forsmall sample size (Burnham & Anderson, 2002).The same candidate model set and modelling procedure asused on the scat abundance data were used on the scat d 13 C, d 15 N and C : N data. In this case, however, linear mixed effectsmodels were used, with locality (i.e. either Dukaladjarranj or Table 3 Candidate models to explain scat abundance, d 13 C, d 15 Nand C : N ratio Models1 Burnt2 Habitat3 Season4 Burnt + habitat5 Burnt * habitat6 Burnt + season7 Burnt * season8 Habitat + season9 Burnt + habitat + season10 Burnt * habitat + season11 Burnt * season + habitat12 Burnt * habitat + burnt * season13 Null model Grass, kangaroos and Absrcinal fire management Journal of Biogeography 34 , 237–250 241 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd
