Resistance and resilience: quantifying relative extinction risk in a diverse assemblage of Australian tropical rainforest vertebrates


*Correspondence: Joanne L. Isaac, Centre for Tropical Biodiversity and Climate Change, School of Marine and Tropical Biology, James Cook University, Townsville, Queensland 4811, Australia. E-mail:


Aim  Assessing the relative vulnerability of species within an assemblage to extinction is crucial for conservation planning at the regional scale. Here, we quantify relative vulnerability to extinction, in terms of both resistance and resilience to environmental change, in an assemblage of tropical rainforest vertebrates.

Location  Wet Tropics Bioregion, north Queensland, Australia.

Methods  We collated data on 163 vertebrates that occur in the Australian Wet Tropics, including 24 frogs, 33 reptiles, 19 mammals and 87 birds. We used the ‘seven forms of rarity’ model to assess relative vulnerability or resistance to environmental change. We then develop a new analogous eight-celled model to assess relative resilience, or potential to recover from environmental perturbation, based on reproductive output, potential for dispersal and climatic niche marginality.

Results  In the rarity model, our assemblage had more species very vulnerable and very resistant than expected by chance. There was a more even distribution of species over the categories in the resilience model. The three traits included in each model were not independent of each other; species that were widespread were also habitat generalists, while species with narrow geographical ranges tended to be locally abundant. In the resilience model, species with low reproductive output had a narrow climatic niche and also a low capacity to disperse. Frogs were the most vulnerable taxonomic group overall. The model categories were compared to current IUCN category of listed species, and the product of the two models was best correlated with IUCN listings.

Main conclusions  The models presented here offer an objective way to predict the resistance of a species to environmental change, and its capacity to recover from disturbance. The new resilience model has similar advantages to the rarity model, in that it uses simple information and is therefore useful for examining patterns in assemblages with many poorly known species.


A crucial task in contemporary conservation biology is to determine which species are most at risk of extinction from human-induced threats including climate change and habitat loss and fragmentation, in order to prioritize conservation efforts (Olden et al., 2008). Previous studies have identified a variety life-history and ecological traits that have been correlated with risk of extinction. The most commonly cited predictors of extinction are small geographical range size, low local abundance, ecological specialization, large body size, slow generation times/slow life history and high trophic level (e.g. McKinney, 1997; Purvis et al., 2000; Johnson, 2002; O’Grady et al., 2004).

McKinney (1997) states that evidence from the palaeontological literature indicates two general factors that reduce extinction risk of a species relative to other species within the same clade. These are (1) resistance – or the ability of a species to withstand an environmental perturbation; and (2) resilience – the ability of a species to recover from an environmental change. While McKinney (1997) discussed these traits specifically with regard to risk of extinction in rare species, they can clearly be applied more generally as predictors of extinction risk (e.g. Knapp et al., 2001). Traits that are likely to confer resistance to environmental change include a wide geographical range, high local abundance and low habitat specialization (McKinney, 1997).

However, there is less evidence from the fossil record to indicate which traits promote resilience. Small-bodied species do appear more resilient, likely because small body size is commonly linked to faster generation times and a faster life history in general (Cardillo, 2003). Other traits that promote resilience are likely to include relatively high reproductive rates in order to maintain (or regain) a viable population size (Crawford, 1997), and high dispersal potential in order to reach suitable new habitat (Knapp et al., 2001). As the primary drivers of the current extinction crisis appear to be habitat loss and climate change (Pimm, 2008), another trait that may promote resilience should be a wide environment niche, which should facilitate the colonization of habitats with changed or new environmental conditions (Crawford, 1997).

The three main traits likely to determine resistance to environmental change are essentially encapsulated in the rarity model of Rabinowitz (1981; Rabinowitz et al., 1986), which ranks species according to their combinations to relative geographical range size, local abundance and habitat specificity. Kattan (1992) extended the rarity model to explore vulnerability of Colombian birds to extinction and the model has since been applied by a number of other studies to determine rarity and extinction risk (e.g. Pitman et al., 1999; Yu & Dobson, 2000; Harcourt et al., 2002). However, as outlined above, these three traits probably give little information about the potential of a species to recover following environmental change. A similar model to explore relative resilience among species in an assemblage may be informative in determining conservation priorities. Furthermore, previous studies that have applied the rarity model to determine vulnerability have generally examined patterns within a single taxonomic group (e.g. Kattan, 1992 (birds); Pitman et al. (1999) (trees); Yu & Dobson, 2000 (mammals); Harcourt et al. (2002) (primates)).

Here, we determine relative risk of extinction in 163 species in four broad taxonomic groups (frogs, reptiles, birds, mammals) that occur in the Wet Tropics Bioregion (WTB), Australia. Approximately 50% of the WTB was listed as a World Heritage Area in 1988, and the region stretches along the north-east coast of Queensland, from Cooktown in the north to Townsville in the south (Williams et al., 2008), encompassing an area of approximately 18,500 km2. The vegetation in the region is primarily tropical rainforest, including montane vine and fern cloud forest, but there are also large areas of lowland wet sclerophyll and mangrove forest (Williams et al., 2008). This region has high biodiversity and conservation value, primarily due to the high number of endemic montane tropical rainforest species (Williams et al., 2003). While much of the region is now protected, the landscape is severely fragmented in places due to past clearing for agriculture (Laurance, 1997), and the WTB has been identified as being particularly susceptible to climate change (Williams et al., 2003). However, many of the endemic species in the assemblage are poorly known and data on population trends are lacking; as a result many species have not been evaluated by the IUCN. Thus, our aim is to determine relative extinction risk, in terms of both resistance and resilience, using the information available for the species in our assemblage. First we use Rabinowitz's rarity model, as implemented by Kattan (1992), to determine which species are likely to be most vulnerable to an initial environmental perturbation. We then develop a new model to determine the relative potential of a species to recover from an environmental perturbation, a resilience index, based on measures of reproductive output, climatic niche marginality and potential for dispersal. Finally, we tested how well these two indices were correlated with IUCN categories of risk for the subset of our sample of species that has been evaluated by the IUCN.


Traits and models

The 163 species included in the analysis included 87 birds, 25 frogs, 33 reptiles and 18 mammals; these numbers represent an estimated 23%, 41%, 20% and 16% of the total number of species in each vertebrate group, respectively, occurring in the WTB (Williams, 2006). Of the included species, 12 birds, 21 frogs, 16 reptiles and nine mammals are regionally endemic. The majority of data were collected during intensive field surveys across the Wet Tropics, including bird, reptile and frog surveys and spotlighting and live-trapping surveys for mammals (Williams, 2006). Every effort was been made to minimize taxonomic bias in survey effort. The body size of species included ranged widely, from 15 mm to 1.8 m, suggesting that any survey bias resulting from sampling species of a particular body size range is also minimal. These 163 species were chosen as a subset of the wider fauna in the region on the basis that they had complete data available for each of the traits used in the models; many other species in our assemblage currently do not have sufficiently robust estimates available for traits including habitat specificity and local abundance to be included. The 163 species, and the raw data for each of the six traits, can be found in Appendix S1 in Supporting Information.

To rank species in our assemblage by relative resistance to environmental perturbation, we used the rarity model of Rabinowitz (1981; Rabinowitz et al., 1986). This model has been discussed in detail elsewhere (e.g. Kattan, 1992; Yu & Dobson, 2000). Briefly, the model uses geographical range size, local abundance and habitat specificity. Species are dichotomized for each of these variables to form an eight-celled model that reflects different types of rarity and commonness (Fig. 1a). Kattan (1992) extended the model by assigning a number between 1 and 8 to each cell to indicate vulnerability. Species assigned the value 1 are rare in all three traits; species assigned the value 8 are common in all three traits. Of the remaining cells in the model, three are rare in two traits and three are rare in only one. We use the method of Kattan (1992) to rank intermediate cells based on the reasoning that species with a narrow geographical range are more vulnerable at a global level, and species with restricted habitat specificity are more vulnerable regardless of their abundance (Table 1a). This model is referred to as the vulnerability index (VI; Kattan, 1992).

Figure 1.

The number of species in each category of (a) the vulnerability model, and (b) the resilience model. Black bars show the observed number of species while open bars show the number of species expected to occur in each category based on chance.

Table 1.  (a) Classes of vulnerability to environmental perturbation (VI), based on Rabinowitz (1981); ranks of vulnerability in each cell are based on those found in Kattan (1992). (b) Ranks for the resilience index (RI), or the potential for a species to recover following perturbation. As with the VI, species that have low potential to recover on all three traits (see text) are assigned a rank of 1 – indicating the lowest potential for population recovery.
Vegetation specializationDistribution
Reproductive outputDispersal potential
Climatic nicheBroad8674

To rank species by potential to recover from environmental disturbance – their resilience – we developed an analogous eight-celled model using the following traits: reproductive output (indicated by clutch or litter size), climatic niche marginality and potential for long distance dispersal. As in the vulnerability index, species were dichotomized for each variable and each cell was designated a number ranging from 1 to 8. To align with the VI, species that demonstrated a low capacity to recover based on all three traits were assigned a value of 1, and those with high potential to recover in all three traits were assigned a value of 8 (Table 1b). Intermediate values were assigned according to the following rationale. The species of the Wet Tropics have been identified as being particularly at risk from climate change (e.g. Williams et al., 2003), and the most important factor determining recovery under climate change is likely to be the ability to recolonize habitats with changed climatic conditions. The ability to reproduce fast enough to maintain a viable population size will then be more important than ability to disperse. If a species is able to disperse to a new environment/habitat but does not have a sufficiently broad environmental tolerance to survive there, or the ability to reproduce fast enough, it will still go extinct. Thus, we ranked traits in order of the following importance: climatic niche marginality > reproductive output > potential for dispersal. Hereafter, this model is referred to as the resilience index (RI).

The measurement and determination of the traits included in both models are explained below. Separating species into two groups for each trait was a necessarily arbitrary exercise. As our main aim was to rank species according to their relative extinction risk within the assemblage, we primarily split species above and below the median value found in our assemblage, except with respect to range size where we considered that for this assemblage, it was more useful (in terms of regional conservation effort) to classify species as endemic or non-endemic.

Geographical range size

Species were categorized as either endemic to the WTB or not, using the list provided in Williams (2006).

Local abundance

Abundance (average abundance n/ha) estimates were calculated from surveys of all species across the region as described in Williams et al. (2003). Species were categorized according to their local abundance based on the median value for the entire data set. Species with values ≥ the median value (1.27/ha) were classed as abundant, species with values < than the median value were classed as scarce.

Habitat specificity

The criteria used to rank species according to their habitat specificity are based on a measure of vegetation specialization. Vegetation groups for each occurrence point for each species were extracted from a spatial vegetation layer created by Stanton & Stanton (2005). We used index level 3 vegetation classification; this is a fairly broad classification level (e.g. rainforest, eucalypt forest, cleared land) but was deemed the most useful index level given the wide variation in occurrence data among species. Vegetation specialization represents the greatest proportion of occurrence records for a species in a single vegetation group. For example, if a species had 30 occurrence records and 15 of them were within habitat classified as rainforest, a vegetation specialization value of 0.5 was assigned to that species. Species with proportions ≥ the median (0.7 = 70% of observations within one habitat type) were classed as having high habitat specialization; those with values < 0.7 had ‘low’ habitat specialization.

Reproductive output

Reproductive output was estimated based on the average number of offspring produced in a single reproductive event, using clutch/litter sizes reported in the primary literature. For a few poorly known species, clutch size had to be inferred from closely related species. In order for clutch sizes to be comparable across taxa, they were standardized within groups. For example it was not meaningful to compare a frog with a clutch size of 3000 to a mammal with a litter size of 1. Clutch size for each species was therefore standardized against the largest clutch size in that group (birds, frogs, mammals and reptiles), giving values of between 0 and 1 for each group. Species with values ≥ the median standardized value (0.1111) were classed as having ‘high’ reproductive output; those with values lower than the median were classed as having ‘low’ reproductive output.

Climatic niche marginality

Climatic niche marginality was calculated using Ecological Niche Factor Analysis (ENFA; Hirzel et al., 2002). Inputs of ENFA are locations for which the species occurs and the continuous environmental data representing the study area. Here, occurrences of species were from multiple sources as described in Williams (2006) and the environmental data represented annual mean temperature, temperature seasonality, maximum temperature of the warmest week, coldest temperature of the coldest week, annual precipitation, precipitation seasonality, precipitation of the driest quarter and precipitation of the wettest quarter, all of which were created using the Anuclim 5.1 software (McMahon et al., 1995). ENFA uses modified principal components analysis to return values for ‘marginality’, which is a measure of niche position quantifying the difference of the input data mean from the background mean. Given identical background environmental space, the values of marginality different species are directly comparable; species with higher ENFA values are more marginal. Species with values above the median (1.154) were classed as having a ‘narrow’ climatic niche, those below the median value were classed as having a ‘broad’ climatic niche.

Potential for long distance dispersal

An index based on the sum of the following characteristics, which are known to influence capability for dispersal (Bowman et al., 2002; Bowman, 2003; Charrette et al., 2006), was calculated for each species: flight (yes = 1, no = 0); known movements from the literature (1 = sedentary (< 5 km); 2 = local dispersion (5–10 km); 3 = altitudinal migrations/nomadic (10–50); 4 = long-distance migrations (> 50); and body size (1 = 0–100; 2 = 101–250; 3 = 251–500; 4 = 500–1000; 5 ≥ 1000). For frogs, the further category of whether or not they had a freshwater tadpole dispersal stage (1, 0) was included. Species with values above the median ranked value (4) were classed as having ‘high’ potential for dispersal; those below the median had ‘low’ potential for dispersal.

Data on the life-history traits used in the analyses were sourced primarily from Strahan (1995: mammals), Higgins et al. (2006: birds); Greer (2008: reptiles) and Cogger (2000: reptiles and frogs).

There is some degree of taxonomic bias in our data set, primarily due to the larger total number of birds, but relatively fewer endemic birds, compared to the other groups. We thus examined the effect of removing a random sample of 47 non-endemic birds from our data set – reducing the total number of birds to 40, including 12 endemics. This resulted in minor changes to the median values of local abundance (+ 0.01), habitat specificity (+ 0.04) and climatic marginality (–0.05), changing the final ranking on the VI for eight species of reptile and rankings on the RI for four birds, two frogs and two reptiles. Given that removing these 40 widespread birds influenced the ranking of only 4% of species in each model, we conclude that the effect of taxonomic bias on our results is minimal and conducted our final analysis on the full set of data.


We initially tested whether the degree of collinearity between the traits used in the models could have influenced our results, particularly since the ‘dispersal’ category was determined partly by body size, which is known to be related to many other life-history and ecological traits. However, the tolerance values for all of the predictor traits were > 0.1, indicating that the degree of collinearity was likely to be insufficient to affect results (Quinn & Keough, 2002).

We then examined patterns in the VI and RI separately. We used χ2 tests for homogeneity to determine whether the number of species occurring in the eight categories in each model differed from that expected by chance. The logm () function in the MASS package, implemented in the program R, was used to produce loglinear models to test for independence between the three categorical variables in each of the two models. We tested for mutual independence (variables are pairwise independent), conditional independence (A is independent of B, given C), and a three-way interaction in all cases. We then examined differences among the four main taxonomic groups (birds, mammals, reptiles and frogs). Wilcoxon rank sum tests were used to determine if the distribution of the summed rank values differed across the four groups. Post-hoc pairwise (Tukey–Kramer HSD) tests were used to assess which groups were significantly different.

We examined how well the models correlated to current IUCN category for listed species using correlation coefficients. Current IUCN category was converted to a numerical ranked value as follows: Critical = 1; Endangered = 2; Vulnerable = 3; Near Threatened = 4; and Low Risk/Least Concern = 5. We applied these ranks, and the eight ranks in each model, as ordinal variables, to examine correlations between values of the VI and RI, and between various combinations of the VI and RI against IUCN category. The Spearman's Rho (rs) non-parametric correlation coefficient was used to assess the strength of relationships, as data could not strictly be classed as continuous and were not normally distributed.



The proportion of species that fell into the different categories for the vulnerability index varied widely. The largest proportion of species fell into the two most vulnerable and two least vulnerable categories (Fig. 1a). The highest proportion fell into category 7; consisting of locally scarce, widespread species with low habitat specificity. A χ2 test for homogeneity showed that the distribution of species within the groups differed from that expected by chance (χ2 = 10.97; P = 0.012; d.f. = 3). There were less species in categories 1 and 8 than expected; those which were either rare in all three traits or common in all three traits. Conversely, there were more species than expected in categories 2 and 7; these being endemic species with high local abundance and a high degree of habitat specialization, and non-endemic species with low local abundances and low habitat specialization, respectively.

Loglinear models determined that the three variables were not independent of one another. Range size and habitat specialization interacted in a pairwise manner, and the interaction between range size and abundance approached significance (Table 2a). Species that were widespread (non-endemic) were habitat generalists while species with small geographical ranges (endemics) tended to be locally abundant.

Table 2.  (a) Results of loglinear models to test for independence between the three variables in the vulnerability model, and (b) the three traits in the resilience model.
ModelLikelihood ratio χ2d.f.P
Complete independence
 Range + Abundance + Habitat47.094< 0.0001
Conditional independence
 Range × Abundance5.2020.074
 Range × Habitat32.862< 0.0001
 Abundance × Habitat2.8520.240
Three-way interaction
 Range × Abundance × habitat1.2010.274
ModelLikelihood ratio χ2d.f.P
Complete independence
 Marginality + Dispersal + Reproduction17.9540.0012
Conditional independence
 Marginality × Dispersal6.3020.042
 Marginality × Reproduction6.7920.033
 Dispersal × Reproduction1.0320.598
Three-way interaction
 Marginality × Dispersal × Reproduction0.1810.674


The proportion of species that fell into each category of the RI was somewhat different to those found in the VI. The category with the lowest potential to recover from a perturbation contained 9.82% of species, while the largest proportion (21.29%) of species had high capacity to recover (Fig. 1b). Other categories had intermediate numbers of species, although category 4 had the lowest percentage of species assigned to it; these were species with low dispersal potential, narrow climatic niche but high reproductive output. A chi-square test found that the distribution of species among the eight categories did not differ from that expected by chance, although test approached significance (χ2 = 7, d.f. = 4, P = 0.072). The biggest deviation appeared to be in group 2, which had more species than expected; these were species with high potential for dispersal, but low reproductive output and a narrow climatic niche. There were somewhat fewer species than expected in groups 4 and 6; these were species with a narrow climatic niche and low dispersal potential but high reproductive output, and those with wide climatic niche, high dispersal potential but low reproductive output, respectively.

The loglinear models determined that the three variables in the resilience model were not independent of one another. Some variables interacted in a pairwise manner, but there was no significant three-way interaction (Table 2b). Species with low reproductive output tended to have a more marginal climatic niche and also a low capacity to disperse.

Patterns of resistance and resilience among species

The distribution of species among the categories for each model differed greatly among the four taxonomic groups (Fig. 2a,b). Birds were most represented in VI group 7, with 37% of species non-endemic with low habitat specificity but low local abundances. On resilience, the majority of birds (44%) fell into the group with the highest potential for recovery, having wide climatic niches, high reproductive output and high potential for dispersal. For mammals, the highest proportion (27%) of mammals also fell into VI category 7; however, 20% were also classified into the highest risk group. Among frogs, 53% of species were categorized into VI groups 1 or 2, the majority (48%) being endemic species with high habitat specificity but also high local abundances. Frogs also had low potential to recover from environmental perturbation, with 60% assigned to RI group 1. The majority of reptiles (33%) were also endemic species with high habitat specificity and high local abundances. However, in comparison to frogs, more reptiles (33%) had a slightly greater capacity for recovery and were classified into RI group 3, having low reproductive output and low capacity for dispersal, but a wide climatic niche.

Figure 2.

The proportion of each taxa that falls into each category for (a) the vulnerability model, and (b) the resilience model. Birds are shown by dark grey bars, mammals by black bars, frogs by light grey bars and reptiles by white bars.

Non-parametric tests showed that the distribution of the sum of the ranked values differed between the four taxonomic groups (VI: χ2 = 20.66, d.f. = 3, P = 0.0001; RI: χ2 = 36.93, d.f. = 3, P < 0.0001). Post-hoc tests show that for the VI, frogs and mammals had significantly lower rank values than birds and for the RI, frogs had significantly lower rank values than all other groups. No other pairwise differences were found.

There was a significant correlation between the ranks of the two indices (rs = 0.57, P < 0.0001), indicating that species that were vulnerable to an initial environmental perturbation were also likely to have low resilience and thus a reduced capacity to recover. The correlation was strongest in mammals (rs = 0.53, P = 0.025), but was also high and statistically significant in all the other groups (reptiles: rs = 0.51, P = 0.003; birds: rs = 0.50, P < 0.0001; frogs: rs = 0.44, P = 0.029).

Correlations between models and IUCN category

Current IUCN category of listed species (n = 126) correlated most strongly with the product of VI*RI rank (Table 3, Fig. 3). However, the strength of correlation was not much different to that for RI rank only, or RI + VI. The VI index alone had the lowest correlation to current IUCN category. Only three species were in the highest risk category of both the RI and VI, these were two frogs (Taudactylus rheophilus and T. acutirostris) and one reptile (Carphodactylus laevis). Both frogs are currently listed as Critically Endangered by the IUCN, and are thought to be extinct. However, C. laevis has not yet been assessed by the IUCN. A total of 12 species fell into the highest risk category on one index and the second highest risk category on the other (see Appendix S1). A number of these species are currently listed as Lower Risk by the IUCN, including four endemic microhylid frogs (Cophixalus infacetus, C. ornatus, Austrochaperina fryi and A. robusta: VI = 2; RI = 1), and one endemic bird (Priondura newtoniana golden bowerbird; VI = 1; RI = 2).

Table 3.  Correlations between a species category on the vulnerability index (VI), the resilience index (RI) and combinations of the two indices, with the current IUCN category of listed species.
ModelSpearman's Rho (rs)P
VI0.43< 0.0001
RI0.52< 0.0001
VI + RI0.51< 0.0001
VI*RI0.52< 0.0001
Figure 3.

The product of a species vulnerability index category * its resilience index category plotted against its current IUCN category. Frogs are shown as black squares, mammals as grey triangles and birds as open circles (no reptiles in our data set are currently assessed by the IUCN). IUCN categories are: CR – Critically Endangered; EN – Endangered; VU – Vulnerable; NT – Near Threatened; LR – Lower Risk/Least Concern.


This study demonstrates that a relatively high proportion of Wet Tropics vertebrates have both low resistance and resilience and are potentially at high risk of extinction under environmental change.

Our results from the VI generally confirm the niche-based hypothesis (Brown, 1984, 1995), that within an assemblage there should be many rare and many common species, resulting in a bimodal pattern across categories. However, unlike the patterns shown in the data of Kattan (1992) and Yu & Dobson (2000), the highest proportion of species in our assemblage was in groups 2 and 7, not groups 1 and 8. This is because, contrary to the predictions of the niche-based hypothesis, our assemblage had many endemic species that were locally abundant habitat specialists, and also many widespread, generalist species that were locally sparse. The high local abundance of specialist endemics may be a result of adaptation to local environmental conditions (e.g. Dobzhansky, 1950). High local abundance is likely to be advantageous in increasing the resistance of endemic species and buffering populations against the initial impacts of environmental change (Manne & Pimm, 2001) and may explain the persistence of some restricted endemic specialists through geological time (Johnson, 1998; Williams et al., 2006). Very specific local environmental conditions, common in tropical montane systems including our study region, may also mean that widespread generalists are at a competitive disadvantage compared to endemics, explaining their relatively low abundance (Reif et al., 2006).

The bimodal pattern evident in the distribution of species among VI categories was not evident among RI categories, where species were generally more evenly distribution and the majority fell into the most resilient category. This assemblage is predicted to be at high risk from climate change (Williams et al., 2003), and resilience is likely to be a particularly important trait influencing recovery from climate change-induced events such as severe cyclones and fire (IPCC, 2007). However, patterns differed considerably among taxonomic groups and many frogs in particular appear to have a very low capacity to recover following environmental perturbation. Indeed, overall frogs were both most vulnerable to an environmental perturbation and had the least capacity to recover, while birds had the lowest vulnerability and highest resilience. Reptiles and mammals fell into intermediate categories, although reptiles appeared to have lower resistance, while mammals had a lower capacity to recover.

The two indices correlated well with IUCN listings, indicating that a combination of these two simple models could be a useful indicator of relative extinction risk in assemblages characterized by many poorly known species, for which data on population trends are lacking. Primarily, IUCN assessments are based on evidence of temporal changes in global populations, while our classifications are essentially based around static measures of regional species status and life-history traits – providing a somewhat different basis to classification (see Dobson & Smith, 1997). Nonetheless, secondary assessment criteria used by the IUCN can include extent of occurrence and measures of abundance – traits that are also included in the rarity model. However, we found that scores from the rarity model alone actually correlated least well with current IUCN category, which suggests that the likelihood for circularity in this analysis is low. We found that the best correlation with current IUCN category was with the product of two indices, suggesting that resistance and resilience may work in a multiplicative way to determine extinction risk. These results, combined with evidence from the log linear models (see below) and other studies (e.g. Davies et al., 2004; Olden et al., 2008), lend further support to the idea that many traits act synergistically to increase extinction risk.

A further indication of lack of circularity in our analysis is that we found a number of species that were shown to be at risk using a combination of the two models but are currently listed as Lower Risk by the IUCN; a number of other species have yet to be evaluated. This finding indicates that the population status of these species should be subject to further investigation and review. We generally found that species that were likely to have low resistance to an initial perturbation were also likely to have low resilience. Our study also demonstrates that the traits included in our models interacted, suggesting that they interact to increase (or decrease) resistance and resilience to environmental perturbations (see also Davies et al., 2004; Olden et al., 2008). Species that were endemic (narrow geographical range) had high habitat specialization, but also tended to be locally abundant. For those traits used to define resilience, we found that species that had a marginal climatic niche had low dispersal potential and also low reproductive output. Species that were marginal and had low dispersal and low reproductive output were also all endemics. These results agree with those of a number of other studies. For example, Cofre et al. (2007) found that rarity, in terms of geographical distribution, was associated with low clutch size, non-migratory status and high habitat specialization in Chilean birds, while Davies et al. (2004) show that abundance and habitat specialization can act synergistically to increase extinction risk, such that species that are both locally rare and specialized are at greater risk of extinction. In an analysis of primates, Harcourt et al. (2002) also found that restricted geographical range was associated with habitat specialization, but found no relationship between range size and local abundance.

However, our results are also atypical. Pitman et al. (1999) also used the rarity model to determine the distribution of tree species in the Amazon rainforest. In contrast to our study, the majority of tree species were common in all three traits. These findings led Ricklefs (2000) to suggest that the study of Pitman et al. (1999) casts doubt over the generalization of Rappoport's Rule which states that high diversity in the tropics is associated with narrow habitat specificity and restricted geographical range (e.g. Stevens, 1989). However, our results are consistent with Rappoport's Rule, as many species, more than expected by chance, had a narrow geographical range and high habitat specialization. In a similar analysis, Kattan (1992) found that Amazon bird species with wide geographical range had broad habitat specificity but, unlike our assemblage, they also high local abundance. Instead, we found that endemic species (with narrow geographical range) were more likely to have high local abundance, a pattern that contradicts the positive relationship generally found between range size and abundance (Gaston & Blackburn, 2000). This pattern does, however, agree with a number of recent studies in montane bird communities (e.g. Manne & Pimm, 2001; Reif et al., 2006). This may suggest that this combination of traits (narrow geographical range but high local abundance) is common in endemic species that have evolved in relatively stable high elevation montane regions. Reif et al. (2006) propose that the high abundances of endemic montane species could be a result of adaptation to local environmental conditions, enabled by climatic stability and isolation of montane forests. They also suggest that species restricted to high elevation montane areas may previously have had larger rangers but become restricted after retreat of montane forest (Reif et al., 2006). Previous studies suggest similar processes may have led to observed patterns in the Wet Tropics; the rainforests in the region underwent extensive contractions during the Quaternary period (Williams & Pearson, 1997) and the evolutionary stability of high elevation rainforest refugia is proposed to be the primary factor in the area being a centre of evolution for low vagility endemic taxa (Graham et al., 2006). However, the pattern of covariation between traits identified here clearly warrants further in-depth exploration in this assemblage of species.

In conclusion, the two models presented here offer an objective way to predict both the resistance of a species to environmental change, and its capacity to recover. The resilience model we have developed has similar advantages to the widely used rarity model of Rabinowitz (1981), in that it uses minimal information and is therefore useful for examining patterns in assemblages such as ours with many poorly known species. Furthermore, in our assemblage the RI appeared to correlate better with current IUCN status than did the VI, suggesting that capacity to recover from a perturbation may be more important than initial effect in determining extinction risk under environmental change.


This research was funded by a James Cook University MTSRF grant, and a James Cook University Research Advancement Grant, awarded to SEW. The data analysed in this study were collected with the support of the Australian Research Council and the Rainforest Cooperative Research Centre. The helpful comments of two anonymous reviewers significantly improved earlier versions of the manuscript.