Fragmentation of the Southern Brown Bandicoot Isoodon obesulus : Unraveling Past Climate Change from Vegetation Clearing

Distribution modeling and vegetation suitability mapping were undertaken to assess (1) the role that past climate change played in fragmenting a subspecies of the endangered southern brown bandicoot Isoodon obesulus and (2) the impacts of land cover change on the subspecies following European settlement of Australia. Based on a selection of bioclimatic variables, disjunctions in the broad distribution of I. obesulus obesulus were found. Vegetation maps representing the time of European settlement revealed two clear features. First, vegetation that was unsuitable for the subspecies corresponded to climatic disjunctions in its distribution, and, second, substantial additional areas were predicted to have suitable vegetation but not suitable climate. Vegetation mapping showed considerable change over two centuries after European settlement, so thatmost places that formerly had suitable climate and vegetation were cleared. Our analysis demonstrates that clearing of native vegetation has masked naturally occurring disjunctions in the distribution of I. o. obesulus. This finding provides evidence that fragmented, regional-scale populations of I. o. obesulus existed prior to European settlement. Implications for conservation planning are discussed.


Introduction
Fragmentation has been variously described as either naturally occurring or anthropogenic disruption to habitat continuity that can operate across multiple spatial, temporal, and functional domains [1][2][3].Most fragmentation studies focus on the patch scale and biotic effects of relatively recent anthropogenic impacts, for example, loss of habitat through vegetation clearing, increased isolation of remnants, and exposure to edge effects [4][5][6][7].But fragmentation can be considered at much broader spatial and temporal scales including global-scale vicariance events, such as continental drift in the Paleocene-Pliocene, or regional-scale disruption to species distributions during the Pleistocene and Holocene.For example, marsupial evolution and radiation in Australia and South America are largely understood in the context of the continents' long-term isolation following separation from Gondwana [8], and at a different, finer scale, the distribution of oligostenothermic species such as the mountain pygmypossum Burramys parvus in south-eastern Australia has been fragmented by the change to the thermocline after the last glacial maximum [9].Scale is a key issue in the present study in which we examine how natural and anthropogenic factors have interacted to fragment the distribution of an endangered species, the southern brown bandicoot Isoodon obesulus in south-eastern Australia.Resolution of two modes of fragmentation, long-term natural versus recent anthropogenic change, is important for understanding inter-regional genetic variation in this species and has important implications for planning conservation initiatives, particularly regarding selection of stock for captive breeding and translocation of individuals between wild populations.[17] and Pope et al. [21].Originally compiled by Rees and Paull [17] from distribution records published by Ashby et al. [11], Friend [22], Gordon et al. [23], Hocking [24], Kemper [14], Menkhorst and Seebeck [18], and Paull [19].NSW: New South Wales, NT: Northern Territory, SA: South Australia, TAS: Tasmania, QLD: Queensland, VIC: Victoria, WA: Western Australia.
Isoodon obesulus is a medium-small (∼0.5-1.8 kg), omnivorous, ground-dwelling marsupial that has a highly fragmented distribution across the margins of southern and eastern Australia (Figure 1).The total distribution of the species Isoodon obesulus in Australia includes five subspecies [10] with the subject of the present study being Isoodon obesulus obesulus in the States of South Australia, Victoria, and New South Wales.Its closest relatives are located in the far north of Queensland (I.o. peninsulae), in Tasmania (I.o. affinis), on islands of Nuyts Archipelago in the Great Australian Bight (I.o. nauticus), and in south-western Western Australia (I. o. fusciventer).Within the distribution of the subspecies, discrete regional populations exist, and this is the scale of fragmentation we address for I. o. obesulus.In the 1800s, I. o. obesulus was reported to be the most common bandicoot and one of the most abundant mammals in south-eastern Australia [11,12], but by the 1920s it was extremely rare [13].Clearing of native vegetation, modification of remnant habitats, introduction of exotic carnivores, and changes to fire regimes have been cited as causes for its decline [14][15][16].Because its numbers have been so severely reduced [11,13,14,[17][18][19][20], it is now listed as endangered under the Australian Commonwealth Environment Protection and Biodiversity Conservation (EPBC) Act 1999.
With vast areas of native vegetation having been cleared throughout the range of I. o. obesulus [25][26][27][28], it is uncertain whether its distribution was already fragmented at the time of European settlement in 1788 by long-standing, naturally occurring barriers.If past climate change, for example, caused disjunctions within the distribution of I. o. obesulus, then some populations that exist today may have been isolated long enough to be substantially genetically distinct.Conversely, if fragmentation was caused by vegetation clearing following European settlement, these distinctions would be less pronounced and perhaps not of consequence for conservation planning.
The explicit aim of this paper is, therefore, to determine if the distribution of I. o. obesulus was continuous or fragmented throughout its total range at the time of European settlement of Australia.Specifically, was the fragmented distribution pattern that we see today attributable solely to vegetation change since European settlement in 1788 or to a combination of natural and anthropogenic factors?To clarify this issue, we combined a climate modeling approach with vegetation suitability mapping.

Distribution Model.
The distribution of suitable climate for I. o. obesulus was modeled using a combination of climatic data generated with BIOCLIM in ANUCLIM 5.1 [29,30], generalized additive models (GAMs) [31] in SPLUS (7.0), and ArcView GIS 3.3 (ESRI, Redlands, CA, USA).Species distribution modeling with BIOCLIM has been undertaken for many vertebrate species (see, e.g., [32][33][34][35][36]).Typically, these types of analyses generate statistical profiles of climatic variables at point localities where species have been recorded then apply homocline matching to model potential bioclimatic distributions.For the present study, a modified approach was adopted whereby locality records were used to define sampling areas (not points) within which I. o. obesulus has been recorded.Between these known "presence" areas, we established "pseudo-absence" areas where the species has never been recorded.To assume that a species is absent from an area may not be valid when few occurrence records exist [37], but in the case of I. o. obesulus there are hundreds of records dating back to the Nineteenth Century.The approach of establishing presence/pseudo-absence data was taken to reduce unknown sampling biases often present in wildlife survey data, for example, where many records are co-located due to biased field survey effort, and it allowed probability estimates to be calculated for each discrete sampling area, thus avoiding the issue of arbitrarily defining a "core" bioclimatic distribution as encountered in other studies (see, e.g., [38,39]).
Locality records for I. o. obesulus were compiled from the South Australian Museum, Museum Victoria, the Australian Museum, the Atlas of Victorian Wildlife database, the New South Wales Wildlife Atlas database, and field survey results of Paull [19] and Rees and Paull [17].These combined sources totaled 1561 records, of which one-fifth were considered to be unique, accurate, and reliable based on the following criteria: (1) records duplicated between data sources were used only once; (2) records with an estimated positional error >1500 m were excluded from analysis; (3) records of I. o. obesulus hairs in predator scats were omitted because scats were collected an unknown distance from where prey was captured;  The selected bandicoot locality records were mapped in ArcView and a hexagonal sampling array with a cell size of 0.25 ∘ was overlaid within a 200 km buffer zone from the coast (as guided by the experimental results of VanDerWal et al. [41]) using the DNR sampling tool (V2.1, 22 June 2004).Hexagonal cells in the array were then assigned binary values of "1" for presence (79 cells) or "0" for pseudo-absence (768 cells) (Figure 3).Next, a systematically distributed sample of 19,812 points was established across the hexagonal array, with each point spaced 0.05 ∘ apart.An elevation value (m above sea level) was extracted for each point using a digital elevation model (DEM) with a cell size of 0.025 ∘ and the ArcView extension Get Grid Value.Values for each of the 35 default BIOCLIM climatic variables [30] were calculated for each of the 19,812 points.The final stage of data preparation was to calculate the arithmetic mean values of the 35 BIOCLIM variables for each hexagonal presence/pseudo-absence cell based on the sampling points that fell within them.
Based on these data, GAMs were developed after first testing for correlation between the climatic variables and choosing those with least correlation for the modeling International Journal of Ecology 0 100 (km) process.BIOCLIM variables that were Pearson correlated with any other variables by >0.7 were culled from analysis leaving 11 candidate explanatory climatic variables (annual mean temperature, isothermality, maximum temperature of the warmest period, minimum temperature of the coldest period, annual precipitation, precipitation seasonality, precipitation of coldest quarter, annual mean radiation, radiation seasonality, radiation of wettest quarter, and highest period moisture index).Three GAMs were calculated using backward selection of variables at the significance level of  = 0.05, with each model having progressively fewer explanatory variables to highlight those aspects of climate that had the greatest influence on the distribution of I. o. obesulus.Finally, spatial predictions for I. o. obesulus were made by mapping probability estimates from the GAMs across the hexagonal array.

Analysis of Vegetation Change.
Throughout southeastern Australia, I. o. obesulus has been documented in native vegetation that includes heathland, shrubland, woodland, and forest communities (Table 1).Common attributes are dense ericaceous shrubs in the ground stratum (0-1 m), thought to be associated with predator avoidance [19].Isoodon o. obesulus also makes limited use of some exotic vegetation, for example, blackberry Rubus spp.[19], but only where suitable native vegetation exists nearby.Vegetation types that are avoided include wet and closed forests and communities with grassy lower strata.Based on these known habitat relationships, weightings were used to rank the suitability of vegetation types represented in two digitized vegetation maps, namely, Vegetation of Australia 1788 and Vegetation of Australia 1988 [42, 43].A modified version of Bryan's [44] generic GIS method for mapping habitat suitability (Table 2(a)) was used in ArcView GIS to combine the dominant species of the tallest stratum and the growth form of the lowest stratum, to produce spatial predictions of vegetation suitable for I. o. obesulus (Table 2(b)).These two particular vegetation attributes were selected because they directly related to the structure and floristic composition of I. o. obesulus habitats as summarized in Table 1.They covered the entire study area and were the best available representation at this scale of vegetation at the time of European settlement and two centuries later at a nominal scale of 1 : 5,000,000.

Statistical Predictions.
Three GAMs were calculated that incorporated BIOCLIM variables relating to precipitation and temperature.They contained six, four, and three significant variables with adjusted  2 values of 0.518, 0.403, and 0.378, respectively (Figure 4).Temperature annual range (TAR) and precipitation of the driest period (PDP) were inversely related with the probability of I. o. obesulus occurring within a hexagonal cell being greatest when values of TAR were low (approximately <20 ∘ C) and PDP was high (approximately >15 mm).Other explanatory variables (annual precipitation (AP), precipitation seasonality (PS), isothermality (ISO), and annual mean temperature (AMT)) approximated Gaussian functions, indicating that I. o. obesulus was most likely to occur in places where these variables took values intermediate to the ranges provided in Figure 4.
Figure 4 indicates how each variable would respond if the others were held at their mean values.To illustrate the interaction between variables, a three-dimensional representation of the three-variable model (PDP + TAR + AP) is provided in Figure 5.This clearly shows the inverse relationship between TAR and PDP and that the greatest probabilities for I. o. obesulus occur when values of TAR are small and values of PDP are large.Figure 5 also shows that when these two variables interact with AP, the probability of I. o. obesulus occurring increases above  = 0.5 only if AP exceeds ∼500 mm, reaching a maximum of  = 0.94 at AP = 891 mm and PDP = 21.6 mm and TAR 13.6 ∘ C before decreasing to  < 0.5 at AP = 1495 mm.

Spatial Predictions.
When the probability estimates from the three GAMs were mapped to the hexagonal sampling units, the spatial predictions matched all known major disjunctions in the distribution of I. o. obesulus (Figures 1-3, cf.Figures 6(a)-6(c)).The largest of these gaps occurred at the Coorong lagoon in South Australia (SA), and at Port Phillip Bay and the La Trobe Valley in Victoria (VIC).In terms of areas where I. o. obesulus currently exists, all three models predicted the region with the most suitable climate to be from East Gippsland (VIC) to Eden in New South Wales (NSW).The six-variable model (AP + PDP + PS + TAR + ISO + AMT) predicted the Mount Lofty Ranges (SA) to also have highly suitable climate for I. o. obesulus, with other noteworthy areas being the western part of Kangaroo Island (SA), the Mount Gambier-Portland region (SA-VIC border), the Otway Ranges-Geelong region (VIC), the Melbourne-Wilsons Promontory region (VIC), and the Batemans Bay-Durras region (NSW).The bioclimatic models A key feature of the spatial predictions was that having fewer terms incorporated into the GAMs resulted in progressively less fragmentation because having fewer terms reduced the fit of the models.The six-variable model incorporated the greatest number of explanatory terms, most closely reflected the distribution of the input records, and concurred best with the known distribution of I. o. obesulus (Figure 1).The four-variable and three-variable models indicated less habitat fragmentation; for example, the Mount Gambier-Portland and Otway Ranges-Geelong regions coalesced in these instances.The six-variable model was the only one that showed the Grampian Ranges (VIC) to be fragmented from its neighbouring regions due to climatic factors.

Vegetation Suitability in 1788 and 1988.
The 1788 vegetation suitability map indicated that the distribution of I. o. obesulus was likely fragmented at the time of European settlement with unsuitable plant cover existing immediately to the south-east of the Mount Lofty Ranges, between the Mount Gambier-Portland and Otway-Geelong regions, and to the north of Port Phillip Bay (Figure 7(a)).Based on what is known of habitat use by the subspecies (Table 1), parts of the Sydney region, including the Cumberland Plain, would formerly have had unsuitable lower stratum vegetation dominated by tussocky or tufted grasses on shale substrate, yet optimal areas of heath did, and still do, exist to the north and south of Sydney on sandstone [59,60].Large areas of suboptimal vegetation occurred in 1788 in places between the Mount Gambier-Portland and Otway Ranges-Geelong regions, as well as extending north-eastwards from the Melbourne-Wilsons Promontory region towards Sydney.Substantial areas of optimal vegetation were mapped in places where the climate models predicted low probabilities of I. o. obesulus being present, for example, inland from the Coorong lagoon, in Victoria's Central Highlands, and in the Blue Mountains.On Kangaroo Island, vegetation was predicted to be more suitable for the subspecies in the eastern half of the island compared to the western half, which was the reverse direction indicated for suitable climate (Figure 6 c.f. Figure 7).
By 1988, vegetation cover for I. o. obesulus had become broadly unsuitable in most of the areas where it was formerly optimal (Figure 7(b)), and this occurred particularly where the species had the greatest probabilities of occurring according to the climate models (Figures 6(a)-6(c)).Of all the regions where I. o. obesulus is currently known to exist, only the Grampian Ranges in Victoria appear to have avoided major anthropogenic transformation of formerly optimal vegetation cover.

Discussion
4.1.Fragmentation Factors.Our analysis highlights that the distribution of I. o. obesulus has been fragmented by a combination of underlying climatic factors, natural vegetation patterns, and subsequent "superimposed" anthropogenic clearing of native vegetation.Major geographic barriers, such as rivers or highlands, are not obvious at the points of disjunction, but unrecognized physical discontinuities may exist.Broad-scale soil variations, for example, may have contributed to the fragmented distribution of I. o. obesulus at the time of European settlement, which would be generally reflected in the vegetation mapping.The models we developed were based on the assumption that the climatic niche of I. o. obesulus has not altered since 1788 and that no major changes to regional climate have taken place.This may not, in fact, be realistic as some analysts suggest that rainfall and minimum temperatures in south-eastern Australia have trended slightly upwards over the past century [61].Regardless, the distribution of optimal vegetation has certainly changed in a radical way throughout the study area due to land clearing for agriculture, establishment of urban areas, and other human land uses.It is difficult to separate the influence of climate from the influence of vegetation in defining the distribution of I. o. obesulus: climate clearly underpins its distribution but the vegetation patterns are strongly related to climate [62].Comparing the c. 1988 map to the c. 1788 map of vegetation suitability reveals considerable change over 200 years so any "natural" fragmentation in the distribution of I. o. obesulus that formerly existed has been masked by humaninduced transformations to land cover.The models indicate where former, natural, and long-standing disjunctions most likely existed and, importantly, highlight that very few places remain where optimal vegetation and suitable climate for I. o. obesulus coincide.
The relationship between I. o. obesulus and vegetation is reasonably well understood particularly regarding the requirement for dense cover for nesting, foraging, and   predator avoidance [10]  dry (desiccated), soil-dwelling invertebrate food may not be abundant [48,55].Prolonged periods of very dry or very wet conditions may, therefore, detrimentally affect the soil microhabitat of invertebrate food sought by I. o. obesulus.The temperature and moisture variables selected in our models would theoretically interact to produce optimal soil moisture regimes for I. o. obesulus and its food sources.by humans has contributed toward a substantial decline in its distribution and abundance.Many other ground-dwelling mammal species are similarly affected, and it is important to recognize this when planning their conservation.We have identified long-standing patterns of fragmentation in the distribution of I. o. obesulus that have likely caused genetic isolation and contributed to the speciation process despite these patterns having become masked by anthropogenic influences.We conclude that conservation scenarios that do not consider the "past" as well as more recent environmental history may be unsound.

Figure 3 :
Figure 3: Hexagonal sampling scheme used to characterize climate throughout the distribution of Isoodon obesulus obesulus in southeastern Australia.Gray hexagons have locality records of I. o.obesulus associated with them whereas hollow hexagons have no records and were, therefore, treated as regions where the subspecies is absent.BIOCLIM values were calculated based on overlaid points illustrated in the inset.

Figure 5 :
Figure 5: Three-dimensional representation of the three-term model at five values of annual precipitation (AP) in mm.Values on the x, y and z-axes of each graph range between 0 and 21.6 mm for precipitation of the driest period (PDP), 13.6-29.4∘ C for temperature annual range (TAR), and 0-1 for probability (PROB).

Figure 6 :
Figure 6: Spatial predictions of the distribution of Isoodon obesulus obesulus based on the probability estimates from generalized additive models, using (a) six variables, (b) four variables, and (c) three variables.

Table 1 :
Vegetation types used by Isoodon obesulus obesulus throughout its range.Eucalyptus sieberi, Eucalyptus agglomerata, Eucalyptus globoidea, Eucalyptus cypellocarpa, and Eucalyptus gummifera (Andrew Claridge pers.obs.)Sydney region Open Eucalyptus woodlands and heathlands on sandstone ridges.Associations of Eucalyptus sieberi, E. capitellata, E. gummifera, E. sparsifolia, E. racemosa, E. piperita, E. pilularis, and Angophora costata (Andrew Claridge pers.obs.) predicted low probabilities of I. o. obesulus occurring near Maldon in central Victoria, in the Blue Mountains (NSW), at Khancoban (NSW), and north of Hawkesbury River near Sydney (NSW), all of which are associated with unverified records of the subspecies.Interestingly, two hexagons at the base of Yorke Peninsula (SA) were predicted by each model to have relatively high probabilities of I. o. obesulus occurring ( = 0.41-0.88),although no input records came from there.