Identifying unidirectional and dynamic habitat filters to faunal recolonisation in restored mine-pits

Authors


Correspondence author. E-mail: m.craig@murdoch.edu.au

Summary

1. There is increasing evidence that passive faunal recolonisation of restored areas can take decades or even centuries, reducing benefits to biodiversity from restoration. Thus, there is a need to develop restoration and management strategies that facilitate and accelerate faunal recolonisation. This requires identification of habitat features that act as filters to slow or prevent recolonisation and whether those filters are temporally unidirectional or dynamic.

2. We investigated successional patterns of reptiles and mammals in restored mine-pits in south-western Australia to identify potential filters to faunal recolonisation. We sampled reptiles and small mammals using 30 trapping grids across each of four restoration ages (4, 8, 12 and 17-years old) and unmined forest and assessed vegetation structure to identify animal–habitat relationships.

3. Mammal communities in restored areas converged on unmined forest communities as restoration matured, and all species recolonised rapidly, indicating there were no filters to mammal recolonisation. In contrast, reptile communities did not converge in the same way, indicating there were filters that slowed or prevented reptile recolonisation. We identified three reptile species that were slow, or failed, to recolonise restored areas and a fourth species that rapidly recolonised but disappeared as restoration matured. We identified low coarse woody debris (CWD) volumes and high overstorey stem densities as likely filters; the former is a unidirectional filter that will decrease gradually over long time frames, possibly centuries, while the latter is a dynamic filter that fluctuates in its intensity over short time frames of years.

4.Synthesis and applications. Our study adds to growing evidence that filters to faunal recolonisation may be widespread in restored areas, with important implications for restoration practices. Firstly, examining individual species may more effectively identify filters than examining community successional patterns. Secondly, filters can persist over long time frames, possibly centuries, so management, such as the provision and accelerated development of CWD, may need to occur over similar time frames. Lastly, filters can be dynamic and repeated management interventions, such as thinning, may be required to overcome these filters. The growing evidence for filters suggests that facilitating faunal recolonisation is more complex than simply returning vegetation to restoration sites.

Introduction

Given past and current levels of habitat clearance and degradation, restoration is likely to become increasingly important in conserving global biodiversity (Hobbs & Harris 2001). However, given that fauna contribute most of the biodiversity in ecosystems, restored areas will need to support most of the original fauna if they are to be effective (e.g. Thomas 2009). Despite this, many restoration projects simply revegetate and assume that animals will naturally recolonise; this assumption is called the ‘Field of Dreams’ hypothesis, which states that ‘if you build it they will come’ (Palmer, Ambrose & Poff 1997). However, evidence is increasing that in some systems not all fauna recolonise restored areas even after several decades (e.g. Martin et al. 2004; Kanowski et al. 2006). Species that fail to recolonise are often specialised and are most vulnerable to habitat clearance and fragmentation (Ryan 1999; Vesk et al. 2008). Thus, it is likely these species will be lost from landscapes unless restoration can provide suitable habitat within the first few decades. If restoration is to be effective in helping preserve global biodiversity, it should aim to facilitate, accelerate and maintain faunal recolonisation rather than relying on passive recolonisation.

A number of recolonisation strategies have been proposed (e.g. Thomas 2009; Lindenmayer et al. 2010), but their incorporation into restoration planning is still not widespread (Brennan, Nichols & Majer 2005). This may be due to the lack of a conceptual framework through which to interpret barriers to faunal recolonisation and examine generalities across studies. However, community ecology-based assembly rules, which seek to understand how communities assemble and why some species occur in particular places (e.g. Wallem et al. 2010), provide a fertile framework through which to interpret faunal recolonisation of restored areas (e.g. Summerville, Conoan & Steichen 2006). In particular, trait-filter models, which predict how species’ traits limit species presence within assemblages (e.g. Purcell et al. 2009), could improve our understanding of why some species move from regional species pools into restored areas whereas others do not.

Although trait-filter models have been widely applied to restored plant communities (e.g. Hough-Snee et al. 2011), they have been infrequently used in understanding faunal community assembly (but see Summerville, Conoan & Steichen 2006). However, the mobility of most terrestrial fauna means that they can respond more rapidly to filters than plants, which are sessile and, often, long-lived. This means filters to faunal recolonisation can be both unidirectional (decreasing gradually over time) or dynamic (fluctuating in intensity over time), depending on whether filters develop over long time frames [e.g. coarse woody debris (CWD)] or change rapidly (e.g. canopy cover). Typically, it has been assumed that filters are unidirectional and that once a species has recolonised restored areas they will persist there. However, if filters are dynamic, species may recolonise restored areas but fail to persist. The temporal nature of filters has important implications for management because reducing the effect of unidirectional filters may require only single management inputs, whereas reducing the effects of dynamic filters is likely to require ongoing and repeated management inputs.

Alcoa of Australia has mined bauxite in the jarrah forest of Western Australia since 1963 and restored its mine-pits since 1966 (Koch 2007). Long-term research indicates that most vertebrate species in unmined jarrah forest recolonise restored sites (Nichols & Nichols 2003; Nichols & Grant 2007), although some species are either unrecorded or recorded very rarely in restored areas, indicating there may be filters to recolonisation. Most of these species are reptiles, which are the slowest vertebrate group to recolonise restored areas (Nichols & Nichols 2003). However, the filters that prevent, or slow, their recolonisation have not been identified so restoration practices cannot be modified to facilitate their return. It is also important to identify successional patterns within restored areas so we can determine whether vertebrates that recolonise persist in restored areas or there are temporally dynamic filters. This will, in turn, inform the type, and time frame, of management strategies in restored areas.

Our study examined reptile and small mammal communities in restored mine-pits of varying ages and adjacent unmined forest to investigate vertebrate successional patterns and use them to identify filters to recolonisation. We analysed vertebrate successional patterns at both community and individual species levels to identify potential filters. At the community level, we concluded that filters to recolonisation were present if communities failed to converge on those in unmined forest as restoration matured. At the species level, we concluded that filters to recolonisation were present if species were absent from restoration or were present in young restoration but disappeared as restoration matured and showed relationships with habitat variables that were also absent from the restored area or differed greatly in value between restoration and unmined forest. Our study objectives were to (i) determine whether habitat filters were present, (ii) if so, identify them and their temporal nature (unidirectional or dynamic), (iii) examine the efficacy of community metrics or individual species in identifying filters and (iv) make recommendations for restoration practices and management to facilitate faunal recolonisation.

Materials and methods

Study Area

This study was conducted on Alcoa of Australia’s Jarrahdale and Huntly mines in south-western Australia. Jarrahdale mine (32°17′S, 116°07′E), 10 km north of Karnet, was mined from 1963 to 1998 with mine-pit restoration completed by 2001. Huntly mine (32°36′S, 116°06′E), 10 km north of Dwellingup, has been mined since 1969, with mining still continuing. The climate at both Karnet and Dwellingup is Mediterranean with cool, wet winters and warm, dry summers. Rainfall averages 1164 mm year−1 at Karnet and 1249 mm year−1 at Dwellingup with >75% falling between May and September at both localities.

The original vegetation at both mines was jarrah forest that was logged at varying intensities several times in the past century (Havel 1989). Jarrah forest is a eucalypt forest with a canopy up to 30 m tall consisting almost exclusively of two eucalypts, jarrah Eucalyptus marginata Smith and marri Corymbia calophylla (Lindley) K. D. Hill & L. A. S. Johnson. Following mining, both mines consist of a mosaic of unmined jarrah forest and restored mine-pits of varying ages. Restored mine-pits have similar species compositions to unmined jarrah forest, although some grass and sedge species are less common. For details of restoration procedures, see Koch (2007).

Experimental Design and Terrestrial Vertebrate and Vegetation Sampling

We investigated terrestrial vertebrate successional patterns in restored mine-pits using a chronosequence experiment with five treatments: unmined forest (the reference community) and four ages of restoration – 4, 8, 12 and 17-years old. The oldest restoration age was chosen as it was the first year when current restoration prescriptions were implemented; other ages were chosen because they represented different stages of vegetation succession within restored mine-pits (Norman et al. 2006). For each treatment, we selected six sites, four at Huntly and two at Jarrahdale, and treatments were interspersed, except the 17-year old restored sites at Huntly which were a few kilometres apart from the remaining sites (Craig et al. 2007).

We sampled terrestrial vertebrates using trapping grids consisting of nine pit-traps (four 850-mL plastic take-away containers, three 20-L plastic buckets and two 40-cm-long 15-cm-diameter PVC tubing) located every 3 m along 29 m aluminium fly-wire drift fences; two paired funnel traps placed along drift fences 7 m from each end; four Elliott traps placed 5 m either side of the two end pit-traps; and Sheffield cage-traps placed at one end of drift fences (Craig et al. 2007). These trapping grids have been shown to provide accurate estimates of relative abundance across a range of restored and unmined forests, precluding the need to correct for detectability (Craig et al. 2009). Because of the number of sites, we conducted trapping sessions over 2 weeks and sampled half the sites each week (two sites at Huntly and one at Jarrahdale in each treatment). Vertebrates were sampled for four nights each week in spring (17–21 and 23–28 October 2005), summer (2–6 and 9–13 January 2006), autumn (20–24 and 27–31 March 2006) and winter (15–19 and 22–26 May 2006) totalling 1728 trap nights per treatment or 8640 trap nights overall. All vertebrates captured were individually marked and released at their capture site.

We conducted vegetation assessments between 19 January and 30 June 2006 to quantify treatment differences and explore relationships between terrestrial vertebrates and habitat structure. To collect data on CWD and overstorey and understorey composition, we set up 100-m2 plots with bottom edges centred 2 m from the end of trapping grid drift–fences. CWD (defined as >5 cm diameter at largest end) volume was calculated from lengths and end diameters of all CWD on plots. Where CWD extended outside plots, we only measured sections within plots. Overstorey and understorey stem densities were calculated by recording the height of all overstorey plants (>3 m in height) present in plots and, because of greater densities, all understorey plants (0·6–3 m in height) in the bottom right-hand quarters of these plots. Canopy height was the average of the five tallest overstorey plants on each plot. We also collected structural data from 0·25-m2 plots located 5 and 10 m on one side and 5, 10 and 15 m on the other side of each 20-L bucket pit-trap on grids, with plots falling along lines running perpendicular to drift fences. In each plot, we estimated overall vegetation cover in three strata (0–1, 1–2 and 2–5 m), canopy cover using densiometers while facing away from drift fences and percentage cover of bare ground and litter.

Statistical Analysis

To investigate community-level patterns, we first created resemblance matrices between sites using a Bray-Curtis similarity measure on untransformed data for reptile and mammal communities separately. For each community, we then used the resemblance matrix to (i) visually represent site differences using principal coordinates analysis (PCO) and overlaid vectors representing species relationships with PCO axes; (ii) analyse treatment (unmined or restoration) and location (Huntly or Jarrahdale) differences using two-way permutational anova (permanova), with 9999 permutations, and conducted post-hoc pairwise tests between all ages of restoration and unmined forest; (iii) analyse differences in similarities with unmined sites for each treatment (all ages of restoration and unmined forest) using a one-way anova; and (iv) examine the relationship between community composition and habitat structure using distance-based linear models (DISTLM), after removing highly correlated structural variables, namely bare ground (r28 = −0·96 with litter cover) and cover from 1 to 2 m (r28 = 0·75 and −0·81 with cover from 2 to 5 m and canopy height, respectively), and normalising structural variables. For mammal analyses, we excluded one site where no mammals were trapped. To investigate changes in vegetation structure, we first created a resemblance matrix between sites using a Euclidean similarity measure on normalised variables then conducted analyses (i) and (ii) described above.

To identify differences in vertebrate abundances between restoration and unmined forest, we analysed individual species using two-way anovas, with location (Jarrahdale or Huntly) and treatment (unmined or restoration) as factors and also analysed community metrics (number and species richness of reptiles and mammals) to supplement community analyses. To investigate successional patterns, we conducted univariate regressions between restoration age (i.e. excluding unmined sites) and abundance with the same variables. To examine structural differences between unmined forest and restoration and understand structural changes with restoration succession, we conducted the same analyses on the ten structural variables measured.

As ecological knowledge of jarrah forest vertebrates is practically non-existent, we posed general a priori hypotheses describing three candidate response types (linear, threshold, quadratic) to examine relationships between vertebrate abundance and structural variables. After removing both correlated structural variables (see above), we then confronted vertebrate abundances (species captured ≥4 times) with three forms of the remaining eight variables plus an intercept-only model, resulting in 25 models per species. For threshold models, we applied a pseudo-threshold form to save parameter estimates (Franklin et al. 2000). We did not employ more complex, multi-predictor models owing to limited data and the lack of a priori predictions of species responses. We ranked models using AICc, or chose the form with the lowest AICc where two forms of a predictor were supported and retained models with weights >0·1 (Burnham & Anderson 2002). To determine which variables were well supported, we summed the model weights of all models containing each variable and considered summed weights >0·4 to be well supported (Converse, White & Block 2006).

All analyses were conducted using permanova+ for Primer (Primer-E, 2008), Statistica 6·0 (StatSoft, 2001), JMP 3·2·5 (SAS, 1999) and R 2·13 (R Development Core Team, 2011).

Results

Identifying Filters Through Community Succession

Reptile communities did not differ between locations (Pseudo-F1,20 = 1·25, = 0·272) but did differ between treatments (Pseudo-F4,20 = 1·67, = 0·020) although there was no interaction (Pseudo-F4,20 = 1·12, = 0·314). Post-hoc tests revealed unmined sites differed from all restoration ages (all < 0·016) but no restoration age differed from another (0·131 < 0·868). Resemblances to unmined forest differed significantly between treatments (F4,25 = 4·80, = 0·005), with unmined sites being more similar to each other than to any restoration age (< 0·05; Fig. 1). PCOs also indicated restored sites were not converging on unmined sites as they matured, with Hemiergis initialis and Cryptoblepharus buchananii the species most associated with unmined sites (Fig. 2). Neither overall reptile abundance nor species richness differed between locations or restored and unmined sites or were related to restoration age (Table 1).

Figure 1.

 Mean (±SE) similarity of mammal (top) and reptile communities (bottom) between each treatment and unmined forest (i.e. UM represents similarities between unmined forest plots). Numbers on x-axes represent restoration age.

Figure 2.

 PCOs of reptile communities (top), mammal communities (middle) and vegetation structure (bottom) in 4-year (□), 8-year (Δ), 12-year (◊) and 17-year-old restoration (○) and unmined forest (•), showing all sites (left) and centroids of each treatment (right). For reptile and mammal communities, vector overlays of species’ correlations with PCO axes are shown. The first three letters of each species’ generic and specific name identify vectors.

Table 1.   Results of analyses on vertebrate species and community metrics
VariablenLocation (L)Treatment (T)L × TUnivariate regression
F1,26F1,26F1,26r22
  1. Location (Jarrahdale vs. Huntly) and treatment (unmined vs. restoration) report the results of two-way anovas on species abundances and community metrics. Univariate regression reports these abundances regressed against restoration age. n is the number of individuals (or species) captured.

  2. *< 0·05, **< 0·01, ***< 0·001.

  3. Introduced.

Community metrics
 No. of reptiles2701·432·730·55−0·12
 Reptile species richness200·362·180·01−0·29
 No. of mammals1460·042·511·160·08
 Mammal species richness63·995·44*0·440·16
Reptiles
 Acritoscincus trilineatus (Gray, 1838)721·593·122·930·08
 Cryptoblepharus buchananii Gray, 1838320·80***57·78***20·80*** 
 Ctenotus delli (Storr, 1974)20·060·540·06 
 Ctenotus labilladieri (Duméril & Bibron, 1839)80·230·230·080·14
 Egernia napoleonis (J.E. Gray, 1838)40·0410·00**0·040·29
 Hemiergis initialis (Werner, 1910)740·023·670·090·23
 Lerista distinguenda (Werner, 1910)70·330·330·33−0·09
 Menetia greyii (J.E. Gray, 1845)101·351·351·89−0·12
 Morethia obscura (Storr, 1972)380·0521·38***0·05−0·46*
 Tiliqua rugosa (J.E. Gray, 1825)100·292·620·29−0·02
 Christinus marmoratus (Gray, 1845)21·3912·48**1·39 
 Diplodactylus polyophthalmus (Günther, 1867)163·910·030·54−0·46*
 Aprasia pulchella (Gray, 1839)40·390·393·47 
 Lialis burtonis (Gray, 1835)20·060·540·06 
 Pogona minor (Sternfeld, 1919)80·561·820·56−0·09
 Varanus rosenbergi (Mertens, 1957)20·250·250·25 
 Ramphotyphlops australis (Gray, 1845)42·280·820·82 
 Notechis scutatus (Peters, 1861)1    
 Parasuta gouldi (Gray, 1841)1    
 Parasuta nigriceps (Günther, 1863)20·250·250·25 
Mammals
 Antechinus flavipes (Waterhouse, 1838)494·37*0·140·140·27
 Sminthopsis griseoventer (Kitchener, Stoddart & Henry, 1984)1    
 Cercatetus concinnus (Gould, 1845)201·141·520·170·14
 Trichosurus vulpecula (Kerr, 1792)2912·46**0·443·110·65***
 Mus musculus (Linnaeus, 1758)450·513·260·51−0·66***
 Rattus rattus (Linnaeus, 1758)21·161·161·16 

Mammal communities differed significantly between locations (Pseudo-F1,20 = 13·80, < 0·001), with Jarrahdale having more Trichosurus vulpecula and less Antechinus flavipes than Huntly, and treatments (Pseudo-F4,20 = 4·47, < 0·001), although there was no interaction (Pseudo-F4,20 = 1·18, = 0·377). Treatment differences were complex, but were primarily between young restoration and older restoration and unmined forest combined. Resemblances to unmined forest did not differ between treatments (F4,25 = 2·20, = 0·098) (Fig. 1) and PCOs indicated restored sites were converging on unmined sites as they matured, with Mus musculus strongly associated with young restoration (Fig. 2). Overall mammal abundance did not differ between unmined forest and restoration or between locations and was not related to restoration age (Table 1). Mammal species richness did not differ between locations and was not related to restoration age, but was significantly greater on restored than unmined sites (2·4 ± 0·2 vs. 1·5 ± 0·4 per site) (Table 1). When analyses were restricted to native mammals, the only differences were that overall native mammal species richness did not differ between restored and unmined sites and both native mammal abundance and species richness were positively correlated with restoration age (r28 = 0·64 and 0·46, = 0·001 and 0·025, respectively).

Identifying Filters Through Individual Species Successional Patterns

We recorded 20 reptile species across all sites (Table 1). Three species (Christinus marmoratus, C. buchananii and Egernia napoleonis) were significantly more abundant in unmined forest (Table 1) and appeared to be late successional, with the first two species only recorded in unmined forest (Fig. 3). Two species (Diplodactylus polyophthalmus and Morethia obscura) appeared to prefer open habitats with low canopy cover (see Table 2), being most abundant in 4-year-old restoration and unmined forest (Fig. 3). Mobscura was more abundant in unmined forest than restoration (3·3 ± 0·5 vs. 0·8 ± 0·3 individuals per site) but Dpolyophthalmus was not; abundances of both species were negatively related to restoration age (Table 1). Hemiergis initialis appeared to be a successional stage increaser, recolonising rapidly and becoming more abundant as restoration matured (Fig. 3). Five species (Ctenotus labilladieri, Lerista distinguenda, Menetia greyii, Tiliqua rugosa and Pogona minor) appeared to be generalists, being found in low abundance in most treatments (Table 1; Fig. 3). One species (Acritoscincus trilineatus) appeared most abundant in 8-, 12- and 17-year-old restoration, but there was little variation between treatments at Huntly with the overall pattern being driven by Jarrahdale (Fig. 3), so we consider it a generalist. The remaining species were captured too infrequently (Table 1) to determine their successional response. Cbuchananii was the only species that differed in abundance between locations and showed a significant interaction (Table 1). This resulted from two individuals being captured at Jarrahdale and one at Huntly which, with fewer Jarrahdale sites, led to a result with statistical, but no biological, significance.

Figure 3.

 Mean (±SE) abundance (individuals per grid) of Egernia napoleonis, Cryptoblepharus buchananii and Christinus marmoratus (late successional), Hemiergis initialis (successional stage increaser), Morethia obscura and Diplodactylus polyophthalmus (open habitat specialists) and Acritoscincus trilineatus, Ctenotus labilladieri and Lerista distinguenda (generalists) in each of four ages of restoration and unmined forest. Key is as for Fig. 1. The inset for Atrilineatus shows variation in treatment responses between Jarrahdale (JA) and Huntly (HU).

Table 2.   Results of modelling effects of structural variables on community metrics and individual species
SpeciesVariableFormDirection†Adjusted r2Model weightVariable weight‡
  1. CWD, coarse woody debris.

  2. Variable, response form and direction, along with model adjusted r2 and model weights are presented.

  3. †Peaked = maxima in middle portion of x-axis, trough = minima in middle portion of x-axis.

  4. ‡Derived from summing weights of all models containing that variable.

  5. *P < 0.05, **P < 0.01, ***P < 0.001.

Community metrics
 Reptile abundanceCWDThresholdIncreased0·11*0·160·36
 Reptile species richnessCanopy coverQuadraticPeaked0·21*0·280·35
Canopy heightQuadraticTrough0·20*0·220·25
 Mammal abundanceCanopy coverQuadraticTrough0·17*0·190·21
Cover from 0 to 1 mLinearIncreased0·11*0·140·27
 Mammal species richnessCWDQuadraticTrough0·22**0·200·35
Canopy heightQuadraticPeaked0·190·110·17
Reptiles
 Acritoscincus trilineatusUnderstorey DensityThresholdDecreased0·060·120·25
 Ctenotus labilladieriNone – intercept   0·13 
 Egernia napoleonisCWDLinearIncreased0·43***0·480·84
 Hemiergis initialisCanopy heightLinearIncreased0·19**0·230·51
CWDThresholdIncreased0·15*0·110·23
 Lerista distinguendaNone – intercept   0·11 
 Menetia greyiiCWDQuadraticTrough0·160·200·44
 Morethia obscuraOverstorey densityThresholdDecreased0·40***0·600·75
Canopy heightQuadraticTrough0·36*0·120·13
 Tiliqua rugosaOverstorey densityLinearIncreased0·090·160·30
 Diplodactylus polyophthalmusLeaf litterLinearDecreased0·11*0·170·35
 Pogona minorUnderstorey densityLinearIncreased0·38***0·630·98
Mammals
 Antechinus flavipesUnderstorey densityLinearIncreased0·12*0·190·40
 Mus musculusCanopy coverQuadraticTrough0·44***0·350·38
Leaf litterThresholdDecreased0·40***0·290·49
 Trichosurus vulpeculaUnderstorey densityThresholdDecreased0·39***0·760·95
 Cercatetus concinnusUnderstorey densityThresholdDecreased0·21**0·430·81

We recorded six mammal species across all sites (Table 1). Mus musculus was early successional (Fig. 4). Its abundance was negatively related to restoration age, but did not differ between restored and unmined sites because of its rarity in older restoration (Table 1). Trichosurus vulpecula, which was more abundant at Jarrahdale than Huntly (1·9 ± 0·4 vs. 0·5 ± 0·3 per site), appeared to be a seral stage increaser (Fig. 4) whose abundance was positively related to restoration age but did not differ between restored sites and unmined forest (Table 1). Two species (Antechinus flavipes and Cercatetus concinnus) appeared to be generalists, and their abundances did not differ between restoration and unmined forest (Table 1). Antechinus flavipes was least common in young restoration (Fig. 4) and was more abundant at Huntly than Jarrahdale (2·2 ± 0·5 vs. 0·5 ± 0·2 individuals per site), while C. concinnus was recorded in all restoration ages with no obvious pattern (Fig. 4). Sminthopsis griseoventer and Rattus rattus were captured too infrequently to determine their successional pattern.

Figure 4.

 Mean (± SE) abundance (individuals per grid) of Mus musculus, Trichosurus vulpecula, Antechinus flavipes and Cercatetus concinnus in the four restoration ages and unmined forest. Key is as for Fig. 1.

Identifying Filters Through Animal–Habitat Relationships

Vegetation structure differed between locations (Pseudo-F1,20 = 3·20, = 0·004) although, univariately, only understorey stem densities differed (F1,26 = 4·30, = 0·048), being greater at Huntly than Jarrahdale (10 800 ± 1316 vs. 5200 ± 950 stems ha−1). Vegetation structure also differed between treatments (Pseudo-F1,20 = 8·67, < 0·001) and changed as restoration matured. All ages of restoration were significantly different from one another and unmined forest (all < 0·028), except 8- and 12-year (t= 1·48, = 0·086) and 12- and 17-year-old restoration (t= 1·02, = 0·419). Unmined forest had less cover from 1 to 2 and 2 to 5 m, lower overstorey stem densities, taller canopies and more CWD than restored areas, but other structural variables did not differ between forest types (Table S1, Supporting Information). As restoration matured, it developed more canopy and litter cover, taller canopies and less bare ground and cover from 1 to 2 m, but the remaining structural variables did not change (Table S1, Supporting Information). The last four variables gradually changed as restoration matured, but canopy cover increased from 4- to 8-year-old restoration then remained relatively constant.

Structural variables explained little variation in reptile communities (adjusted r2 of best model = 0·13) and marginal tests from the DISTLM revealed only canopy height explained significant amounts of this variation (Pseudo-F1,28 = 3·03, P = 0·001) and then only 9·8%. Structural variables also explained little variation in reptile community metrics with no variables well supported (Table 2). At the individual species level, there were well-supported variables for five species with Pogona minor showing a linear increase with understorey density, E. napoleonis a linear increase with CWD volume, M. obscura a decreasing threshold response to overstorey stem density, Hemiergis initialis a linear increase with canopy height and Menetia greyii a trough-shaped quadratic relationship with CWD volume (Table 2).

Structural variables explained much of the variation in mammal communities (adjusted r2 of best model = 0·42) and marginal tests from the DISTLM revealed that understorey density (Pseudo-F1,28 = 6·74, P < 0·001), litter cover (Pseudo-F1,28 = 5·02, P = 0·002), canopy height (Pseudo-F1,28 = 4·28, P = 0·006) and canopy cover (Pseudo-F1,28 = 3·09, P = 0·032) all univariately explained significant amounts of this variation: 20·0, 15·7, 13·7 and 10·3%, respectively. Conversely, structural variables explained little variation in mammal community metrics with no variable well supported (Table 2). At the individual species level, there were well-supported variables for three species with both Trichosurus vulpecula and C. concinnus showing a decreasing threshold relationship with understorey density and Mus musculus a trough-shaped quadratic relationship with canopy cover (Table 2).

Discussion

Mammal communities in restored areas converged towards those in unmined forest as restoration matured, but reptile communities did not. This strongly indicates that restoration contains filters that slow or prevent recolonisation by reptiles; however, the community-level approach was poor at identifying those filters. Our analyses did not identify any structural variables that were well supported as being related to reptile community metrics, while analyses on overall reptile communities identified canopy height as the only variable related to community composition. However, we suspect the relationship with canopy height is correlative rather than causal. Many studies have found no evidence that canopy height is important in structuring reptile communities (e.g. Taylor & Fox 2001a; Craig et al. 2010) and canopy height becomes more similar to unmined forest as restoration matured, but reptile communities did not. There were significant relationships between mammal community composition and four structural variables, but we know that mammal communities converged towards unmined communities as restoration matured; therefore, these variables could not be acting as filters. We propose that community-level analyses were not effective at detecting filters because individual species displayed a range of successional patterns that counteracted one another, thereby obscuring relationships with vegetation structure at a community level. While studies do use community-level analyses to identify filters to recolonisation (e.g. Longcore 2003; Lindenmayer et al. 2010), our study suggests this approach will not always effectively identify filters, while acknowledging there is often no alternative in invertebrate communities studies.

In contrast, individual species analyses appeared more successful at identifying filters for reptiles. Five species showed significant relationships with structural variables, although only three identified potential filters. Pogona minor increased linearly with understorey density, possibly because it frequently utilises understorey plants (Craig et al. 2007), but this species is more common in restoration than unmined forest so understorey density cannot be acting as a filter. Menetia greyii was most abundant at high and low CWD volumes, implying that CWD cannot act as a filter for this species. Hemiergis initialis was positively related to canopy height, which is a unidirectional filter and slowly decreasing its effect as the restoration matures. Given jarrah growth rates (Koch & Ward 2005), it will probably cease to be a filter after several decades. However, Craig et al. (2010) found no relationship between H. initialis and canopy height, and it is difficult to propose an ecological mechanism whereby canopy height directly influences a litter-inhabiting skink. M. obscura showed a decreasing threshold response with overstorey stem density. While the exact mechanism behind the relationship is unclear, it is probably due to vegetation cover and solar insolation influencing the thermal suitability of the habitat (Craig et al. 2010). This species increased when overstorey stem densities in restoration were reduced by thinning and burning (Craig et al. 2010) but, 5 years later, numbers in thinned and burned restoration had decreased (Smith 2011). This suggests that this filter is dynamic with both Mobscura abundance and vegetation cover increasing and decreasing over short time frames as succession proceeds within restored areas. E. napoleonis linearly increased with CWD volume, probably because it utilises CWD extensively (Craig et al. 2011; Christie et al. in press), suggesting low CWD volumes in restoration are a unidirectional filter preventing recolonisation. We would expect the effect of this filter to decrease and Enapoleonis abundance to increase gradually over time. The time frame over which CWD will develop naturally in restoration is unclear but, given average ages of Eucalyptus marginata and Corymbia calophylla and average rates of trees fall (Whitford 2002), it will probably be centuries before CWD accumulates sufficiently to reduce this filter to Enapoleonis recolonisation, particularly as it prefers large CWD (Craig et al. 2011).

Studies across a range of ecosystems have shown that species fail to recolonise restored areas, even after several decades (e.g. Martin et al. 2004; Kanowski et al. 2006; Desrochers, Keagy & Cristol 2008), suggesting that filters to recolonisation may be widespread. The extent of the filters identified in this study is unclear. Canopy height is often related to structural complexity (McElhinny et al. 2005), and several studies have shown that lack of structural complexity can be a filter to recolonisation (Lomov, Keith & Hochuli 2009), suggesting that canopy height could potentially be an indirect filter in other forest and woodland ecosystems. Vegetation cover affects solar insolation and microclimate (e.g. Martens, Breshears & Meyer 2000) both of which are critical in influencing habitat use by many fauna species (e.g. Hill et al. 2001; Pike, Webb & Shine 2011). This suggests that structural variables that relate to vegetation cover, such as canopy, mid-storey or understorey cover, or understorey or overstorey stem density, have the potential to act as filters to faunal recolonisation in many forest and woodland ecosystems. Indeed, Taylor & Fox (2001b) found that canopy and understorey cover were major determinants of lizard succession in restored minesites, probably due to differing thermal requirements between species. CWD is a critical component of forested ecosystems, and many fauna species are CWD-dependent (e.g. Lindenmayer et al. 2002). A lack of CWD is often cited as a factor preventing faunal recolonisation of restored sites across a range of faunal groups (e.g. Martin et al. 2004; Grimbacher & Catterall 2007), suggesting this filter is likely to be widespread in ecosystems naturally containing CWD (Vesk et al. 2008). While the filters identified in this study are potentially widespread in other forest and woodland systems, the complexity of fauna–habitat relationships means that they almost certainly represent only a small subset of the range of potential filters that could slow or prevent faunal recolonisation. It is likely that filters relate to species traits as well, as found by Summerville, Conoan & Steichen (2006). Unfortunately, the lack of information on the ecology of jarrah forest reptiles prevents us from examining those relationships in our study, but it is an important area for future research as it would further improve our ability to predict which species might be susceptible to filters.

Management Implications

Where filters to faunal recolonisation are identified, management actions should focus on reducing the effects of these filters to facilitate faunal recolonisation. In our system, reducing the impact of low CWD volumes requires more CWD, preferably up to 50 logs ha-1 (Christie et al. in press), to be added as part of the restoration process. However, CWD in mine-pits may only last 70 years before being consumed by fire (Grigg & Steele 2011), long before large CWD will develop naturally, so management strategies that aim to kill trees and accelerate CWD formation also need to be considered. Adding CWD has been shown to accelerate faunal recolonisation in other studies (Madden & Fox 1997; Márquez-Ferrando et al. 2009; Barton et al. 2011), suggesting it is likely to be an effective management strategy across a range of systems. Management strategies to reduce the filters of low canopy heights and high overstorey stem densities will be similar, as reducing overstorey stem densities increases growth rates in jarrah trees (Stoneman et al. 1997). Thinning and burning minesite restoration was an effective strategy for reducing both these filters in our system (Craig et al. 2010), and thinning has also been shown to influence faunal community composition in a range of ecosystems (Hagar, Howlin & Ganio 2004; Converse, White & Block 2006). Individual species respond variably to thinning, however (e.g. Converse, White & Block 2006), so use of this management strategy will depend on the species of interest. Furthermore, vegetation cover, which was probably the proximal filter in our system, is a dynamic filter, so any benefit of thinning and burning may be short-lived (Smith 2011). This implies that management strategies that reduce vegetation cover may need to be repeatedly applied if filters are to be reduced to levels that enable populations of species of interest to be maintained. Although our study has focused on habitat filters that are relatively easy to restore compared to some others (e.g. soil water-logging), the effect of all filters should potentially be reducible given unlimited resources, so which filters can be reduced through management in restored sites will be primarily constrained by finances and logistics.

Conclusions

In our study, both community-level and species-level analyses indicated there were filters to passive reptile recolonisation, but only species-level analyses appeared successful at identifying these filters. Low CWD volumes were identified as a unidirectional filter that could delay passive recolonisation of some species for centuries. Overstorey stem density fluctuates over relatively short time frames (years), so its filter effect is dynamic. These findings have important implications for restoration practices. Firstly, many species are known to be unable to passively recolonise restored areas (e.g. Kanowski et al. 2006), implying that filters to recolonisation are likely to be widespread in many restored areas and that community-level analyses might be poor at identifying these filters. Secondly, filters can persist over long time frames, possibly centuries, so management may also need to occur over similar time frames. This management will need to ensure that slow-developing habitat resources, such as CWD or tree hollows (Vesk et al. 2008), are provided in restored areas, and their natural development accelerated to facilitate recolonisation by fauna dependent on these resources. Lastly, filters can be dynamic and repeated management interventions, such as thinning and burning, may be required to overcome these filters and maintain populations of susceptible fauna in restored areas (Craig et al. 2010). The growing evidence for the impact of filters suggests that facilitating faunal recolonisation is more complex than simply returning vegetation to the pre-impacted state at restoration sites.

Acknowledgements

We thank Anna Whitfield, Dion Trevithick-Harney, Alicia Sparnon, Adam Peck, Angela Mercier, Chantelle Jackson, Kaitlyn Height, Jason Fraser, Finlay Bender and Jane Adcroft who helped install trapping grids and Jonathan Anderson, Angela Mercier and Rod Armistead for help with vegetation surveys. Claire Stevenson from the Western Australian Museum identified the Sminthopsis species. This project was conducted with Department of Environment and Conservation permit SF005179 and Murdoch University Animal Ethics committee approval W1152/05. Financial support was provided by Alcoa of Australia and ARC Linkage Grant LP0455309.

Ancillary