Patterns of rain forest plant endemism in subtropical Australia relate to stable mesic refugia and species dispersal limitations


  • Lui C. Weber,

    Corresponding author
    1. School of Biological Sciences, The University of Queensland, St Lucia, QLD, Australia
    • Correspondence: Lui C. Weber, School of Biological Sciences, The University of Queensland, St Lucia, QLD 4072, Australia.


    Search for more papers by this author
  • Jeremy VanDerWal,

    1. Centre for Tropical Biodiversity and Climate Change, School of Marine and Tropical Biology, James Cook University of North Queensland, Townsville, QLD, Australia
    Search for more papers by this author
  • Susanne Schmidt,

    1. School of Agriculture and Food Sciences, The University of Queensland, St. Lucia, QLD, Australia
    Search for more papers by this author
  • William J. F. McDonald,

    1. Queensland Herbarium, Queensland Government Department of Environment and Resource Management, Brisbane Botanic Gardens – Mt Coot-tha, Toowong, QLD, Australia
    Search for more papers by this author
  • Luke P. Shoo

    1. School of Biological Sciences, The University of Queensland, St Lucia, QLD, Australia
    2. Centre for Tropical Biodiversity and Climate Change, School of Marine and Tropical Biology, James Cook University of North Queensland, Townsville, QLD, Australia
    Search for more papers by this author



Our aims were to identify centres of endemism and to infer whether these areas have functioned as refugia for subtropical rain forest plants through historical climate fluctuations.


Subtropical eastern Australia (23–33° S; 145–155° E).


We collated 25,000 records of 179 endemic rain forest plants to identify geographical areas with unusually high concentrations of endemic taxa and range-restricted endemics. We then tested whether centres of endemism coincide with other features indicating refugia, including habitat stability over 120,000 years, and we related dispersal patterns to past habitat stability using seed weight as a surrogate for dispersal ability of endemic plant taxa.


We identified five main centres of endemism. Historical stability and other processes affecting diversity, including current rainfall, rain forest area, and topographic complexity, explained 58% of variation in plant-weighted endemism. Taxa with poor dispersal ability were concentrated in the areas that were most stable historically.

Main conclusions

Several lines of evidence suggest that centres of endemism have functioned as important refugia for subtropical rain forest taxa through historical climate fluctuations. The highest concentrations of range-restricted endemic species occur in locations that are predicted to have maintained stable rain forest habitat over at least the past 120,000 years. This association was independent from other factors that were expected to promote diversity (i.e. rain forest area and current environmental suitability). These locations have disproportionately high concentrations of species with poor dispersal ability (large-seeded species).


The concept of climatic refugia is an important idea in biogeography (Haffer, 1982). Traditionally, the term ‘refugium’ or ‘refugia’ was used to describe areas in the Northern Hemisphere into which species retracted and survived during Quaternary ice ages (Keppel et al., 2012). Recently, however, the term has also been used to refer to environments in non-glaciated landscapes in the Americas, Africa, Asia, Australia and the Pacific that have sheltered species from past climate change, or that may function similarly in the future (Keppel et al., 2012), thus allowing survival of species or ecosystems by maintaining climatic conditions similar to those prior to climatic change or by offering more stable habitat than surrounding areas (Hilbert et al., 2007; Bennett & Provan, 2008; Ohlemüller et al., 2008; VanDerWal et al., 2009). This broader definition has necessitated recognition of diverse refugial types varying along spatial, temporal and disturbance dimensions (Couper & Hoskin, 2008; Ashcroft, 2010; Stewart et al., 2010; Keppel et al., 2012). Here, we investigate rain forest refugia in eastern Australia associated with climatic change, focusing on larger areas that have remained suitable for rain forest through a number of glacial cycles (stable, mesic macrorefugia).

It is uncertain how refugia can best be defined and identified (Bennett & Provan, 2008; Ashcroft, 2010), but they are generally considered to have distinctive patterns of biodiversity, including: high species diversity (Médail & Diadema, 2009); high endemism; concentrations of climatic relict, poorly dispersed and vegetatively reproducing species; or other traits that reduce survival elsewhere including drought or fire intolerance (see Keppel et al., 2012, and references therein).

The persistence of palaeoendemic species and the accumulation of neoendemic species in refugia can generate areas with unusually high concentrations of endemic species (Keppel et al., 2012). While some endemism can be expected by chance as a result of differing species range sizes (Laffan & Crisp, 2003; Jetz et al., 2004), areas combining higher levels of endemism, concordant patterns between plant and animal distributions, and more stable climatic conditions are likely to have functioned as refugia, allowing the survival of relicts and evolution of new species (McGuigan et al., 1998; Moritz et al., 2001). Endemism (particularly of range-restricted taxa with low dispersal ability) is positively correlated with past climatic stability globally (Jansson, 2003; Sandel et al., 2011). There are numerous examples of concordance between climatic refugia and centres of endemism in the European Alps (Tribsch & Schönswetter, 2003; Casazza et al., 2008), Africa (Fjeldså & Lovett, 1997), and Australian Wet Tropics (Hilbert et al., 2007; Graham et al., 2010), although this notion is contested in some areas (Knapp & Mallet, 2003). Areas with high concentrations of range-restricted endemic species, termed ‘centres of endemism’ (Jetz et al., 2004) may be particularly indicative of refugia. This is because many range-restricted species also share ecological traits that reduce survival outside or expansion from refugia, such as dependency on relict climates or poor dispersal (Keppel et al., 2012). Although some endemism in centres of endemism may be attributed to recently evolved species from widespread linages, these areas may function as future refugia following species range expansion and contraction. An area can function as a refugium for lineages that were previously more widespread and contracted leading to evolution of neoendemic species in refugia. The majority of neoendemic Australian rain forest plants are likely to represent the latter case due to long-term contraction of rain forest.

Gondwanan rain forests were isolated when Australia was separated from Antarctica by sea-floor spreading and the continents became separated by ocean c. 49 Ma (Truswell, 1993). Rain forests were still widespread in Australia in the late Eocene (35 Ma) including in inland areas that are now arid (Macphail, 2007). Following collision with Asia during the Miocene numerous rain forest plant lineages migrated into northern Australia (Sniderman & Jordan, 2011). The loss of rain forests in central Australia around 10 Ma was a consequence of increasing aridity (Byrne, 2008). A further contraction of this community has occurred within the past 5 Myr (Crisp et al., 2004; Byrne, 2008). Drying climates resulted in rain forest becoming restricted to areas retaining suitable climate along the Great Dividing Range and east coast (Floyd, 1990; Crisp et al., 2004). Rain forests now cover < 1% of Australia (Webb & Tracey, 1981a) and are limited to areas with low susceptibility to bush fires (Bowman, 2000). Not all rain forest patches are stable over long time-scales (Graham et al., 2010) and many current rain forest areas represent expansions since the Last Glacial Maximum (LGM), as indicated by charcoal deposits, and are not considered refugia (Hopkins et al., 1993). Northern Australian rain forests have been enriched by many Asian plant lineages since the Miocene, yet those in subtropical Australia also retain significant numbers of Gondwanan lineages. Conversely, Asian lineages are comparatively rare in temperate Australian rain forests, including subtropical montane forests (Sniderman & Jordan, 2011). Persistence of these lineages indicates that some areas may be long-term refugia (Heads, 2009).

Phylogeographical, pollen and biodiversity studies focusing on fauna have provided evidence for a link between refugia and centres of endemism in the Wet Tropics (north-eastern Australia) (Williams & Pearson, 1997; Schneider & Moritz, 1999; Yeates et al., 2002; Graham et al., 2006; Hilbert et al., 2007; Moussalli et al., 2009). The link between endemism and refugia is less well understood for Australian subtropical rain forest flora. Past studies have been limited by state borders, relied on qualitative methods, focused on a single species group, or considered all flora without distinguishing processes specifically affecting rain forests (McGuigan et al., 1998; Queensland CRA/RFA Steering Committee, 1998; Crisp et al., 2001; National Land & Water Resources Audit, 2001). This is a notable gap in our knowledge given the high biodiversity values (Adam, 1987; Hunter, 2004) and projected vulnerability to future climate change of Australian subtropical rain forest flora (Australian National University, 2009; Laidlaw et al., 2011). The importance of subtropical and tropical forests of Australia as globally significant biodiversity hotspots has recently been recognized due to high levels of endemism and habitat loss (Williams et al., 2011).

Here, we evaluate spatial patterns of richness and endemism for rain forest flora in subtropical Australia and consider whether centres of endemism have functioned as refugia. Specifically, we test distribution patterns of endemic wet rain forest plants against random models and correlate patterns of endemism with a palaeoclimate model for rain forest habitat stability over 120,000 years (Graham et al., 2010). We hypothesize that centres of rain forest plant endemism should coincide with areas of high environmental stability, sheltering palaeoendemic species and favouring the evolution of new species through divergence of isolated populations (Ponniah & Hughes, 2004; Rossetto & Kooyman, 2005; Mast et al., 2008). We contrast the influence of historical habitat stability against other processes potentially governing patterns of richness, such as rain forest patch size, current environmental suitability and topographic heterogeneity. We also examine associations between dispersal ability of endemic plant taxa and modelled rain forest stability and hypothesize that large-seeded species are most likely to be restricted to historically stable habitat. Poor dispersal ability is likely to have reduced the range expansion of large-seeded species following the amelioration of climate since the LGM.

Materials and methods

Study area

Four broad rain forest formations (subtropical, dry, warm and cool temperate) occur in subtropical Australia. Subtropical rain forests (STRF) have outliers north and south of their core distribution: Eungella (QLD, 21° S), which supports more tropical species; and Illawarra (NSW, 34° S), from which many STRF indicator species are absent (Webb & Tracey, 1981b). We adopt a stricter definition of STRF excluding these outliers, making Kroombit Tops (23.5° S) and Barrington Tops (33° S) the northern and southern limits of STRF, respectively (Fig. 1). Kroombit Tops is also a northern outlier of warm temperate rain forest (WTRF) and Barrington Tops represents the most southerly stand of Nothofagus moorei cool temperate rain forest (CTRF) (Floyd, 1990; Queensland CRA/RFA Steering Committee, 1998). The limits of the study area also correspond to major biogeographical barriers: the dry St. Lawrence Gap to the north and the Cassilis Gap (Hunter Valley) to the south (Moussalli et al., 2009; Chapple et al., 2011; Rix & Harvey, 2012).

Figure 1.

The study region in subtropical eastern Australia showing the pre-1750 distribution of wet rain forest types (black shading) from National Vegetation Information Systems mapping and subregional boundaries (black lines). Subregional boundaries were defined by 800 mm mean annual rainfall isohyets (western extent) and known biogeographical barriers for rain forest taxa reported in literature (see Appendix S2 for details). The QLD–NSW state border is also shown (grey line).

Rain forest types

We focus on wetter rain forests of subtropical Australia, defined by Webb (1959) as complex notophyll vine forest (CNVF), notophyll vine forest (NVF) and simple notophyll vine forest (SNVF); or subtropical, littoral and warm temperate rain forests sensu Floyd (1990) that form the A1 ecofloristic province (Webb & Tracey, 1981a). We also include wetter forms of Araucarian notophyll vine forests (ANVF), as they grade into CNVF and share endemic species, and higher-elevation, wet, cool temperate and submontane rain forests (notophyll–microphyll vine fern forest, microphyll fern forest, microphyll mossy forest and thickets) of ecofloristic province A2 (Webb, 1959, 1968; Webb & Tracey, 1981a). Plant taxa beyond the scope of this study included species from dry rain forests, Araucarian microphyll vine forests (AMVF) and semi-evergreen vine thickets (SEVT), which extend to low rainfall regions (ecofloristic provinces C1 and C2) (Webb & Tracey, 1981a; McDonald, 2006).

Endemic species and data set

We targeted rain forest vascular plant species with distributions that were completely contained within the study region (23–33° S), while excluding species of rain forest margins, wet sclerophyll, dry rain forest and vine thickets (Harden et al., 2006, 2007). Species with endemic subspecies or forms, that have another subspecies occurring outside the study area, were included as endemic subspecies. Two taxa that had multiple subspecies endemic to the study area were grouped to species level as not all records were identified to subspecies. The complete list of endemic taxa included in the analysis is provided in Appendix S1 in Supporting Information.

Occurrence data for target taxa were compiled from the following data sets: (1) specimen data from the Queensland Herbarium database (HERBRECS) (Johnson, 1991) (accessed 6 May, 2010); (2) the WWF Vine Forest Atlas (Forster et al., 1991) (accessed 14 May, 2010); (3) Queensland Herbarium CORVEG database (McDonald & Dillewaard, 1994) (accessed 26 July, 2011); and (4) Atlas of NSW Wildlife (, accessed 13 August, 2010). The NSW Atlas incorporates specimen-backed records from the NSW Herbarium and field records from environmental scientists. The combined data set was then validated against published species distributions (Andrews, 1990; Harden et al., 2006, 2007; Floyd, 2008) and checked for geocoding errors.

All records with poor accuracy (> 10 km) were removed. To reduce the potential effect of forest clearing on species richness, records from all years were retained. Finally, the data set was supplemented with 1084 field observations of 149 taxa by the first author (L.C.W.). Taxonomy was corrected to conform to the Census of Queensland Flora (Bostock & Holland, 2010) except for new species (Acalypha sp. ‘Big Scrub’ and Endiandra lowiana). The final data set consisted of 25,418 records for 179 endemic taxa.

Patterns of endemism

Endemism was assessed using three measures: local richness, weighted endemism and subregional endemism. Spatial patterns of endemic taxa were analysed using Biodiverse 0.15 (Laffan et al., 2010). Species records were aggregated into 0.05 degree (c. 5.5 km × 5 km) grids, with moving window analyses conducted across two spatial scales (Laffan & Crisp, 2003): (1) each grid cell plus the four adjacent grid cells; and (2) each grid cell and the nearest 28 surrounding grid cells in a circular radius (neighbourhood). The use of moving window neighbourhoods balances resolution against interpolation of data gaps (Crisp et al., 2001). Local richness was calculated as the number of taxa recorded in each grid cell and its set of nearest neighbours at each spatial scale. Weighted endemism (WE) was the sum of inverse range sizes (number of 5.5 × 5 km grid cells) for all taxa in the neighbourhood (Peterson & Watson, 1998; Crisp et al., 2001).

We used spatial Monte Carlo randomizations to test whether observed values of local richness or WE were statistically greater than a random pattern. Specifically, values in all grid cells were randomly swapped to another location with one or more taxa recorded. Tests were run for 1000 iterations for each grid cell neighbourhood against richness and WE. Areas that scored higher than the random model for more than 900 of the 1000 randomizations (P ≥ 0.90) were considered significant centres of endemism for WE, or foci of endemism for species richness (Casazza et al., 2008). Results from Biodiverse were visualised in diva-gis 7.3.0 (Hijmans et al., 2011). Assessment of subregional endemism is detailed below under subregions.

Determinants of endemism

We modelled species richness and weighted endemism as a function of historical rain forest stability, area, precipitation and topographical complexity. Each of these parameters is detailed in turn.


We estimated historical stability of rain forest using climate data from Hadley Centre coupled models, version 3 (HadCM3), run back over 120,000 years, testing ice-sheet forcings and validating results against polar ice cores (Singarayer & Valdes, 2010). This was carried out in several steps. First, the current distribution of rain forest in eastern Australia from Cape York to Tasmania was modelled using Maxent 3.3.3 (Phillips et al., 2006). Training data (presence and background points) were derived from the pre-European (1750) distribution of rain forest using the Australian National Vegetation Information System (NVIS) (Australian Government Department of the Environment & Water Resources, 2006) and a 1 km × 1 km sampling grid overlaid on the continental land surface. Environmental data used for modelling included eight climate variables that could be generated using current and historical climate surfaces. These variables included annual mean temperature, temperature seasonality, mean temperature of the warmest quarter, mean temperature of the coldest quarter, annual precipitation, precipitation seasonality, precipitation of the wettest quarter and precipitation of the driest quarter. Current climate data (4-km resolution) were provided by Robert Hijmans and generated using anusplin (Hijmans et al., 2005) with terrestrial surfaces extended to include areas that would be exposed if sea levels were 125 m lower than at present in order to allow estimation of historical climate.

The Maxent model was then projected back in time at 5000-year intervals. Historical climate data were derived from HadCM3 anomaly data (Singarayer & Valdes, 2010) downscaled from 1° (c. 110 km at the equator) to 4-km resolution using a cubic spline method and applied to current climate. Anomalies were calculated as the difference (absolute for temperature and proportion for precipitation) between modelled past climate and modelled pre-industrial climate. The potential distribution of rain forest was clipped based on digital elevation data and estimated sea levels for each timeframe.

Finally, we allowed for a spatially dynamic representation of habitat stability. That is, suitable habitat may have persisted by shifting contiguously in response to changing climate through geological time. Dispersal of rain forest was incorporated using the Viterbi least cost path (Viterbi, 1967) and a migration rate of 10 m per year. We acknowledge that migration rates are likely to vary between individual taxa (Vittoz & Engler, 2007). Our estimated rate is a compromise between the extremes of no dispersal and full dispersal. It is probably a conservative estimate for the rain forest community as a whole, and a closer reflection of the dispersal rates of taxa that have ecological traits which are expected to constrain expansion from refugia, such as large seeds or asexual reproduction (e.g. clonal suckering).

Stability was calculated by summing the negative log probabilities of the Viterbi migration paths for all time steps, following the approach outlined in Graham et al. (2010) (Fig. 2a). Note that stability, as modelled, does not distinguish between rain forest types, so temperate rain forest may have replaced subtropical rain forest during glacial maxima; however, our model provides an overall estimate of the spatial continuity of rain forest habitats over time.

Figure 2.

Spatial layers used to model species richness and weighted endemism for subtropical rain forest plants in eastern Australia. (a) Modelled rain forest stability over the past 120,000 years (120 kyr) incorporating 10 m year−1 migration rate (Graham et al., 2010). High stability areas (red > 0.8) are numbered: 1, Sunshine Coast; 2, Border Ranges; 3, Dorrigo. (b) Elevation (m) and (c) mean annual rainfall (mm) in the eastern Australian study area; data from Geoscience Australia 250 m DEM-9S data set and anuclim rainfall interpolation. (d) The standard deviation of elevation in each 5.5 km × 5 km grid cell within the study area; note that SD of elevation was calculated per forest patch not per grid cell for the regression analysis.


We used pre-European (1750) NVIS rain forest and the Region Group tool in ESRI ArcGIS (ESRI Inc., to identify clusters of cells that formed unique, unconnected forest patches. The Zonal Statistics tool in ArcGIS 10.1 was used to estimate geographical area of patches.


Maximum precipitation for each forest patch was extracted from a spatial layer of mean annual precipitation (Fig. 2b).

Topographic complexity

The standard deviation of elevation was extracted for each forest patch from a 250-m resolution digital elevation model [created from the GEODATA 9 Second Digital Elevation Model (DEM-9S) Version 2 data set; Geoscience Australia, Canberra] derived using anuclim 5.1 software (Houlder et al., 2000) in combination with the digital elevation model (Fig. 2c,d).

Clearly, observed richness and endemism in a grid cell is not independent of neighbouring cells. Two measures were employed to reduce the problem of spatial autocorrelation in the analyses. First, all data were aggregated to the level of unique rain forest patches prior to analysis using the Zonal Statistics tool in ArcGIS. Second, we converted spatial data for species richness and weighted endemism to a 5 km × 5 km point grid and excluded patches that did not have any intersections with grid centres containing information on richness and endemism. This approach reduced the number of patches in the analysis from 18,800 (mostly small clustered patches) to 90 (a representative sample of patches of varying in size). Highly skewed variables were log-transformed prior to analysis. We tested for pairwise correlations between independent variables and then used ordinary least squares regression to model the relationship between endemic richness or weighted endemism and predictor variables.

Ecological traits of species

We used seed weight and dispersal mode as a surrogate for dispersal ability (Bolmgren & Eriksson, 2010). Dispersal mode was inferred from fruit descriptions or published studies in the region (Green, 1993; Butler et al., 2007). Seed weights for each taxon were derived from Floyd (2008). In instances where seed data were not available for a species, seed weights were estimated from closely related species with a similar fruit size listed in the Kew Seed Information Database ( or Floyd (2008) (Appendix S1). We used the Quantum GIS 1.7 intersection tool (QGIS Development Team, 2011) to extract information on historical stability for all location records for target taxa. We then examined the relationship between mean historical stability of rain forest (across all locations for each species) and seed weight using ordinary least squares regression and linear quantile regression. Quantile regression was included to examine the relationship between stability and the upper limits (95th percentile) of seed weight (Cade et al., 1999). All analyses were performed using R 2.13.1 (R Development Core Team, 2011) and the quantreg 4.10 library (


Subregions were derived by partitioning the study area based on known biogeographical barriers for rain forest taxa reported in the literature, which also coincided with high concentrations of latitudinal range limits for endemic species (see Appendix S2). The western margin of the study area was defined by the 800 mm rainfall isohyet that encompassed all but three endemic flora records. Endemic richness was compared at the subregion rather than grid level, by counting subregional endemic taxa (restricted to a single subregion) and regional endemics, occurring in two or more subregions. Total (subtropical) endemic richness was counted as the sum of these two classes.


Our endemic flora data set included 179 endemic vascular plant taxa (166 species and 13 subspecies), including one lycophyte, six fern species, 26 basal angiosperms, 10 monocotyledons and 136 eudicots (Appendix S1).

Patterns of endemism

Analysis at small versus larger scale (i.e. moving window of 4 versus 28 grid cells) yielded similar results, except for Barrington Tops, which was not a significant centre of endemism at the larger scale. For simplicity, we report results from the finer-scale analysis. Omitting the seven taxa from the data set that are currently recognized as endemic subspecies (with other subspecies outside the study area) did not notably alter the size or location of centres of endemism. Given the trend to revise disjunct subspecies as endemic species, for example Mischocarpus ailae and Actephila grandifolia (Forster, 2005; Guymer, 2009), we report results based on the full data set including endemic subspecies (see below). Increasing the number of Monte Carlo randomizations from 500 to 1000 did not appreciably change delineation of significant areas of endemism suggesting that the number of iterations was appropriate for the analysis.

Broadening the latitudinal range of the study area to also encompass rain forests between Mackay and Wollongong (21−34° S) would have added approximately 35 endemic species and minimally increased richness and endemism for the Bulburin and Barrington Tops areas. The majority of the species with distributional limits near Mackay or Wollongong are not subtropical endemics, but rather tropical species extending to the Wet Tropics or temperate species extending to Victoria and Tasmania.

Foci of endemism

One major and two minor foci of endemism were identified. The major focus comprised a portion of the Border Ranges centred on the Mount Warning Caldera–Big Scrub area (Mount Tamborine to Ballina) (Fig. 3). This area had the highest richness of endemic taxa with grid neighbourhoods containing in excess of 50 taxa. Nine neighbourhoods contained > 100 taxa, including the richest single grid containing 117 taxa. Minor foci (only 1–2 cells significant for richness) were located in adjacent parts of the Border Ranges as well as Dorrigo to the south and the Sunshine Coast to the north.

Figure 3.

Patterns of local endemic richness and Monte Carlo test results for subtropical rain forest plant taxa in eastern Australia. (a) Richness (number of endemic taxa) per 5.5 km × 5 km grid cell including the four nearest grid cells. (b) P-values for richness per grid cell including the four nearest grid cells, compared to 1000 Monte Carlo randomizations; red areas have significantly more endemism than a random model (P ≥ 0.9) (1, Sunshine Coast; 2, Border Ranges-Big Scrub; 3, Dorrigo).

Centres of endemism

Five centres of endemism were identified (Fig. 4). Of these areas, the Border Ranges including the former Big Scrub harboured the highest WE scores (18–27), endemic richness, with 142 taxa (78% of all subtropical endemics) and most subregional endemics with 45 species found only in this centre, encompassing 64% of all the subregional endemics within the five centres (Table 1). Two other major centres (> five subregional endemics) lie to the north and south of the Border Ranges in the Sunshine Coast and Dorrigo–Ebor areas, respectively, with WE scores from four to ten. Two minor centres of endemism (two to three subregional endemics) were identified at Bulburin and Barrington Tops with WE scores of two to six (Table 1, Appendix S2: Table S2.2).

Figure 4.

Patterns of weighted endemism and Monte Carlo test results for subtropical rain forest plant taxa in eastern Australia. (a) Weighted endemism (WE; the sum of inverse range sizes) per 5.5 km × 5 km grid cell including the four nearest grid cells. (b) P-values for WE per grid cell including four nearest cells, compared to 1000 Monte Carlo randomizations; red areas have significantly more endemism than a random model (P ≥ 0.9) (1, Bulburin; 2, Fraser Island–Sunshine Coast; 3, Border Ranges–Big Scrub; 4, Dorrigo–Ebor; 5, Barrington Tops).

Table 1. Summary of total subtropical (regional + subregional) endemic and subregional endemic wet rainforest plant taxa richness, in subregions containing the five identified centres of endemism in rain forest of subtropical eastern Australia. Centres of endemism are listed according to their geographical position from north to south. Bold indicates major centres of endemism (> five narrowly endemic taxa)
Centre of endemismTotal (regional + subregional) endemic richness (no. of taxa)Fraction of total endemism (%)Subregional endemic richness (no. of taxa)Fraction of subregional endemism (%)
1. Bulburin2916.245.7
2. Sunshine Coast–Fraser Island 70 39 13 18.5
3. Border Ranges–Big Scrub 142 79.3 45 64.2
4. Dorrigo–Ebor 88 49.1 6 8.5
5. Barrington Tops3016.722.8

Determinants of endemism

Pairwise comparisons generally indicated low correlation between independent variables (Spearman's ρ < 0.59 across all comparisons, Appendix S3: Table S3.1). Maximum richness tended to occur in forest patches with both high rainfall and high historical stability, with the linear model predicting richness using log patch area, elevation SD, log max rainfall and max stability (log richness = log area + elevation SD + log max rainfall + max stability), explaining 56% of the variance in maximum richness (Fig. 5a–d, Table 2). The same variables were important for explaining weighted endemism but there was an additional positive effect of area (Fig. 5e–h, Table 2). The linear model (log WE = log area + elevation SD + log max rainfall + max stability) explained 58% of the variance in maximum weighted endemism among forest patches. Elevational standard deviation within forest patches was not found to be important in explaining richness or weighted endemism.

Figure 5.

Partial regression plots showing relationships between (a–d) maximum endemic plant taxa richness, or (e–h) weighted endemism (sum of inverse range sizes for all taxa recorded in a 5.5 km × 5 km grid cell and its four nearest neighbours), and (a,e) area (ha); (b,f) elevational standard deviation (m), (c,g) maximum mean annual rainfall (mm); and (d,h) maximum stability. Each dot represents a unique rain forest patch located in the study areas in eastern Australia. Regression lines (least squares) are shown for statistically significant relationships (see Table 2).

Table 2. Summaries of linear models testing the effects of historical stability, rainfall, topographic variability and patch area, on endemic plant taxa richness and weighted endemism in subtropical Australian rain forests
FactorEstimateSEt-value P
  1. SD, standard deviation; SE, standard error.*P < 0.05, **P < 0.01,***P < 0.001.

Log taxa richness (adj. R2 = 0.56) P < 0.001***
Intercept−23.0752.803−8.232 < 0.001***
Log area0.0620.0501.227 0.224
Elevation SD−0.0020.002−1.3670.176
Log max rainfall3.0720.4656.603 < 0.001***
Max stability3.7791.9061.983 0.051
Log weighted endemism + 1 (adj. R2 = 0.58) P < 0.001***
Intercept−14.3031.667−8.581 < 0.001***
Log area0.08770.030 2.924 0.005**
Elevation SD−0.0020.001−1.550 0.125
Log max rainfall1.7140.277 6.194 < 0.001***
Max stability2.6661.133  0.020*

Ecological traits of species

Seed weights varied across seven orders of magnitude from 0.005 mg in orchids and ferns to 55 g in large-fruited Lauraceae (Endiandra virens). Non-animal dispersed propagules were significantly lighter than animal dispersed ones (Kruskal–Wallis P < 0.00001). There was a positive relationship between seed weight and stability for animal dispersed taxa (d.f. = 131, F = 6.024, P = 0.015) but not non-animal dispersed taxa (d.f. = 45, F < 0.001, = 0.994) (Fig. 6). Of 39 taxa with seed weights > 2 g, 37 occurred at sites with a mean stability > 0.7. Nevertheless, stability did not explain much variance in seed weight, even for animal-dispersed taxa (adj. R2 = 0.037). Quantile regression indicates the potential form of the limiting relationship between stability and seed weight (Fig. 6a).

Figure 6.

Relationships between seed weight and mean stability of rain forest habitat in eastern Australia for (a) taxa with animal-dispersed seeds and (b) taxa with non-animal-dispersed seeds. Solid line is a quantile regression fitted to the top 5% of data, the dotted line is least squares regression fitted to all data.


We identified five distinct centres of endemism in subtropical Australian rain forests. Several lines of evidence suggest that these centres have functioned as refugia through historical climate fluctuations. The highest concentrations of range-restricted endemic species occur in locations predicted to have maintained more stable rain forest habitat over at least the last 120,000 years and, by extension, supported larger or more numerous patches of suitable habitat during the LGM. The association between endemism and stability remains even after accounting for rain forest area, topographical complexity and current environmental suitability (i.e. rainfall). These same areas also have disproportionately high representation of species with ecological traits suggesting poor capacity for dispersal (large animal-dispersed seeds).

Patterns of richness and endemism

Our analyses, based on a comprehensive qualitative study of plant endemism in rain forests of subtropical Australia, provides important refinements to patterns of endemism detected in previous studies, showing major centres treated as equivalent to have differing levels of endemism, while some smaller centres are less important than previously recognized (see also Appendix S2). Specifically, our data reaffirm the importance of the Border Ranges and New England–Dorrigo centres identified in previous assessments (Boden & Given, 1995; Crisp et al., 2001; Laffan & Crisp, 2003). We also identified the Sunshine Coast area as a major centre of endemism for wet rain forest flora. Importantly, we show that the Border Ranges region (Big Scrub, McPherson Ranges and Mount Warning Caldera) has the highest richness and endemism and contains at least 78% of all endemic wet rain forest plant taxa, with 65% occurring in close proximity near Springbrook and Lamington National Parks. Notably, the area with the richest endemic flora also has the highest rainfall in subtropical Australia: 3109 mm mean annual rainfall at Springbrook Forestry Station (1914–2003; Bureau of Meteorology, Climate Data Online, Mount Warning volcano (including Springbrook) was suspected by Floyd (1990) to be the most important refugium after the Wet Tropics and we find that in addition to outstanding endemism and richness, this area has the highest modelled rain forest stability on mainland Australia outside the Wet Tropics (J. VanDerWal, unpublished data). We are confident that the patterns of richness and endemism observed here result from differences in diversity between areas rather than sampling bias, because sampling coverage is high across the broader study area (Appendix S2: Fig. S2.3), and plot data recording all species in a 1-ha area shows similar patterns (Laidlaw et al., 2000), as does species diversity pooled by subregion (Kooyman et al., 2011).

Determinants of endemism

High levels of plant richness in the Border Ranges have previously been attributed to coincidence with the McPherson Macleay Overlap (MMO), where temperate and tropical species co-occur (Burbidge, 1960; Laidlaw et al., 2000). However, this fails to explain why some areas within the MMO are richer than others. Our results indicate that the MMO is not just an overlap of northern and southern species. The Border Ranges area (the eastern portion in particular) is predicted to have retained a greater area of stable habitat for rain forest during climatic variability of the past 120,000 years than areas to the north, west and south including other parts of the MMO. It also hosts a large number of narrow-range endemics. Our results are supported by other studies using different methods. Kooyman et al. (2011) found that the Nightcap Range (part of the Border Ranges) has lower taxonomic evenness and more phylobeta diversity than the Washpool area, indicating that Washpool has experienced extinctions of many lineages that persist in the border ranges refugia. Multiple distinct centres of endemism are likely to be driven by the contraction and persistence of palaeoendemic lineages such as Eidothea and the more recent speciation of lineages such as Actephila and Fontainea in disjunct rain forest refugia separated by dry and fire prone climatic barriers (Shapcott, 1998; Rossetto et al., 2000).

An alternative explanation is that the higher richness of the Border Ranges may be an artefact of random arrangement of species distributions within a bounded domain, i.e. the mid-domain effect (MDE) (Colwell & Lees, 2000). We believe this is unlikely for at least two reasons. First, range-restricted endemic species targeted by this study are expected to be less affected by the MDE than species with larger ranges and should be more likely to be randomly distributed within the domain (Colwell & Lees, 2000). Second, at the subregional level, the pattern of disproportionately high richness persists even when the bounded domain (and hence selection of ‘endemic’ species) is reconfigured to place Border Ranges at the northern or southern extreme of the domain (Appendix S3: Table S3.2).

The minor centres of endemism (Bulburin and Barrington) are predicted to have been less stable than the major centres of endemism and exhibit reduced subregional endemism. Palaeovegetation studies support this pattern: the pollen record from a 900 m elevation swamp at Barrington Tops shows alpine vegetation during the LGM, with Nothofagus CTRF returning 9000 yr bp and persisting until the present (Sweller & Martin, 2001). It is likely that despite the lack of pollen at 900 m, CTRF persisted in nearby small lower-elevation refugia during the LGM, as two narrow-endemic species occur only in CTRF at Barrington, and the related Nothofagus cunninghamii persisted in glacial microrefugia in Tasmania (Worth et al., 2009). Also, genetics of the CTRF tree Atherosperma moschatum indicate survival in refugia near Barrington and/or the Blue Mountains (Worth et al., 2011).

Climatic, geological and topographical factors have been important in the formation of refugia on other continents and in the Australian Wet Tropics (Kreft & Jetz, 2007; Dobrowski, 2011). The maintenance of high rainfall over yearly and geological time-scales appears to be an important factor governing environmental suitability for rain forest and persistence of dependent plants, through factors such as reducing the frequency and severity of fires. In subtropical Australia, the dominance of high pressure systems causes areas with coastlines aligned perpendicular to the dominant south-east trade winds and high coastal mountain ranges to experience higher rainfall (Webb & Tracey, 1981a; Floyd, 1990). These factors may be especially important in maintaining rainfall during El Niño events and periods of lowered glacial sea levels when coastlines are further east (Hilbert & Ostendorf, 2001). In our study the combination of rain forest area, rainfall, stability and topographical variability explained 56% of the variation in endemic flora richness and 58% for weighted endemism. Topographical variation was not a significant predictor of weighted endemism possibly because the local contribution of complexity is highly mediated by geographical position (e.g. inland mountains receive lower rainfall and are more prone to fires).

All centres of endemism, with the exception of the granitic Bulburin and sandy Fraser Island areas, are focused on extinct volcanoes and erosional landscapes derived from them (Geoscience Australia, 2010). A series of volcanoes formed along the east coast during the Tertiary (20–30 Ma), creating high mountains and fertile soils, both rare environments on an overall flat and oligotrophic continent (Orians & Milewski, 2007; Cohen, 2012). This pattern is contrary to recognized trends for refugia elsewhere in the world, as volcanic eruptions and glaciations are considered to constitute clean slate or tabula rasa disturbances (Nordal, 1987). Areas subject to such disturbances are generally not considered to be refugia (Keppel et al., 2012). We assert that areas that have experienced tabula rasa disturbances can become refugia given enough time: 20.5 Myr in the case of the Mount Warning volcano, which supports 45 narrowly endemic wet rain forest flora taxa and a similar number of narrowly endemic non-rain forest flora (Boden & Given, 1995).

Dry river valleys or areas with lower rainfall appear to function as dispersal barriers promoting isolation between stable rain forest refugia and high levels of narrow-range endemism (Floyd, 1990; Ponniah & Hughes, 2004). For endemic plant taxa, turnover and disjunctions in distributions are concentrated on the margins of these climatic barriers (see Appendix S2: Fig. S2.1). Distributions of endemic rain forest animals, correspond closely with centres of plant endemism, reinforcing other evidence for refugia near these locations (see Appendix S2: Table S2.1). Furthermore, molecular phylogeny of rain forest flora and fauna shows divergence between refugia across dry climatic barriers within subtropical Australia. These patterns are particularly strong in the frog genera Philoria and Taudactylus, leaf tailed geckos (Phyllurus and Saltuarius) and freshwater crayfish (Euastacus) (Couper et al., 2000; Knowles et al., 2004; Shull et al., 2005).

Ecological traits of species

Areas that have served as refugia are expected to contain higher concentrations of species with poor dispersal and asexual reproduction as well as other traits that would reduce survival outside refugia (Richards et al., 2003; Keppel et al., 2012). Endemic species in our data set with larger, heavy seeds of 10–60 g are essentially confined to stable refugia. Butler et al. (2007) also detected a positive correlation between large-seeded fleshy fruits and rainfall in rain forests of the northern half of our study area. They proposed that slow replacement of trees in wet rain forest reduces selection for smaller seed size, allowing evolution of large seeds. While this theory could explain the evolution of large seeds in ancient wet rain forest, current distributions of large-seeded species may not reflect conditions under which they evolved. The present concentration of large-seeded species in higher-rainfall areas may instead reflect contraction to refugia during the LGM (Longmore, 1997; Hilbert et al., 2007) and the extinction of coevolved seed-dispersing megafauna, including cassowaries, from subtropical Australia 40 ka (Miller, 1962; Low, 2011). Loss of dispersers would have reduced the ability of larger-seeded species to expand from glacial refugia as climate became more favourable (Rossetto & Kooyman, 2005; Rossetto et al., 2008).

The pigeons Ptilinopus magnificus and Lopholaimus antarcticus are highly mobile frugivorous birds in subtropical Australia with large gapes capable of swallowing larger seeds than similar birds. Elaeocarpus grandis (4.6 g seed) is eaten by these species and may represent the upper limit of seed weight for highly dispersed species (Floyd, 1990; Green, 1993). Most animal-dispersed species with seeds heavier than five grams are likely to have relied on cassowaries for dispersal, as they still do in tropical Australia and New Guinea (Willson et al., 1989). Now only water, gravity and scatter-hoarding rodents disperse these largest subtropical seeds (Rossetto & Kooyman, 2005). Fourteen subtropical endemic species in our study fit the ‘megafauna fruits’ (Guimarães et al., 2008) or ‘cassowary fruits’ (Willson et al., 1989) syndrome and occur at significantly more stable sites than other endemic species (P < 0.05 Kruskal–Wallis test). These species include Endiandra floydii, Niemeyera whitei and Syzygium hodgkinsoniae. Differing dispersal rates of large- and small-seeded species in the region are supported by biochemical and genetic studies of endemic species (Banfield et al., 1982; Shapcott, 2002).

Our data set contains two species that currently reproduce exclusively through vegetative suckering (Elaeocarpus williamsianus and Davidsonia johnsonii), although genetic evidence suggests these species once had reproduction via seeds. Both clonal species are narrowly endemic to the stable lowlands of the Border Ranges (Rossetto et al., 2004). Asexual reproduction can confer increased site persistence and tolerance of smaller disturbances, and is prominent at species distribution limits and in species with infrequent sexual reproduction (Johnston & Lacey, 1983; Rossetto & Kooyman, 2005; Zobel, 2008). However, loss of sexual reproduction can disadvantage species by reducing dispersal to lateral growth of ramets and increasing risks of population extinction from catastrophic disturbances (Rossetto et al., 2004; van der Merwe et al., 2010). In Australia, catastrophic fires occur in most landscapes (Bowman, 2000) and these events would probably eliminate long-lived fire sensitive clonal species outside fire refugia. Nothofagus moorei, distributed from the Border Ranges to Barrington, exhibits higher levels of suckering in the Border Ranges than further south (Taylor et al., 2005). Two species, Eidothea hardeniana and Eucryphia jinksii, both restricted to single locations on the Mount Warning caldera, also frequently sucker (Forster & Hyland, 1997; Rossetto & Kooyman, 2005). Rossetto & Kooyman (2005) found resprouting ability, seed size and dispersal mode best predicted the proportion of potential habitats occupied by rare rain forest plants (including Eidothea hardeniana) on the southern caldera. Long-lived clonal rain forest plants, while able to tolerate smaller disturbances, may be restricted to refugia where catastrophic disturbances occur infrequently. The age of clones in the study area is not known; however, Lomatia tasmanica from Tasmania is likely to have persisted for 43,600 years in a glacial refugium (Lynch & Balmer, 2004).

Although the refugia modelled here are Quaternary, and the evidence for the persistence of plants and rain forest over much longer time-scales is circumstantial, there is evidence for much older Tertiary refugia in eastern Australia from rain forest invertebrates, for example the Border Ranges clade of assassin spiders (Austrarchaea spp.) diverged from those to the north of the Brisbane Valley c. 22 Ma, coincidental with the eruptions of the Mount Warning volcano (Rix & Harvey, 2012).

Heads (2009) describes the McPherson Macleay Overlap (MMO) as a globally basal centre of endemism due to the presence of endemic species that are basal to more diverse widespread groups. Areas supporting concentrations of ancient lineages that have gone extinct elsewhere are consistent with refugia (Hampe & Jump, 2011). There are more than 30 examples of phylogenetically and geographically isolated lineages with endemic species in the study area (16% of endemic taxa) many of which are of Gondwanan origin (Appendix S3: Table S3.3). All but two of these groups have fossils outside subtropical Australia, indicating more extensive distributions in the past, which is a signature of refugia (Keppel et al., 2012).


Our analysis suggests that major current centres of plant endemism have probably functioned as mesic refugia in subtropical Australia, sheltering during increasing aridity both ancient Gondwanan rain forest taxa and more recently evolved species derived from Asian lineages. Reduced frugivore dispersal, including the extinction of cassowaries, is likely to have limited the expansion of large-seeded plants from refugia after the LGM. We suggest that multiple refugia separated by dry barriers can explain patterns of observed species distributions that are not predicted by the overlap of temperate and tropical species in the MMO bioregion or the mid-domain effect, and that Tertiary volcanoes played a role in the creation of rain forest refugia in an infertile and drying continent.

By identifying probable refugia for subtropical Australian rain forests we have addressed a key knowledge gap for conservation planning and climate change adaptation (Steffen et al., 2009). Focused protection and restoration of refugia and centres of endemism is likely to result in the conservation of more potentially vulnerable species in areas with lower climate change velocity than efforts directed in other areas (Sandel et al., 2011; Shoo et al., 2011; Keppel et al., 2012).


This research was supported by a National Climate Change Adaptation Research Facility grant to L.C.W. and the South East Queensland Climate Adaptation Research Initiative – a partnership between the Queensland and Australian governments, the CSIRO Climate Adaptation National Research Flagship, Griffith University, University of the Sunshine Coast and University of Queensland. The Initiative aims to provide research knowledge to enable the region to adapt and prepare for the impacts of climate change. We thank Robert Hijmans for providing climate data and anonymous referees for providing further comments.


Lui C. Weber is interested in how past and present environmental factors determine the survival of plants, interactions between plants and fauna and how understanding these can improve the conservation and restoration of ecosystems and ecological interactions.

Author contributions: L.C.W. led the research including data compilation, analysis and writing; J.V. provided the palaeoclimate rain forest stability model; W.J.F.M. provided the CORVEG and Vine Forest Atlas datasets; and S.S. and L.P.S. provided ongoing advice and contributed to analysis and preparation of the manuscript.