Epigaeic invertebrate diversity of post‐mining revegetation evaluated by comparison with adjacent remnant woodlands

This study evaluated the validity of invertebrate diversity for evaluating the condition of minesite rehabilitation. Because the aim of rehabilitation of mined lands in this region is to establish a self‐sustaining ecosystem that resembles “bushland,” a variety of remnant woodlands in the vicinity of the minesite were selected as reference sites. Reference sites encompassed a range of soil types (alluvial, coarse‐textured, or clay) and understorey structural types (grassy, grass, and shrub or shrubby). Grassland sites (cleared woodlands) were also sampled to provide a negative control. Epigaeic invertebrates of three rehabilitation sites were compared with five reference sites and three grassland sites using pitfall traps over 6 years. Sites were sampled annually and the annual data within each of two time periods (T1: 2008–2010 and T2: 2011–2013) were pooled. Ants, beetles, and springtails were identified as morphospecies, providing site species richness and composition data. Springtail metrics of all three habitats, reference, rehabilitation, and grassland, resembled each other in T1 as well as T2, indicating that rehabilitation rapidly provided suitable conditions for this group. In contrast, the ant and beetle composition of rehabilitation sites differed significantly from that of reference sites in T1 but not T2, although only ants matched the expected negative control with grassland sites remaining distinct from reference sites in both time periods. Thus, ant assemblages of rehabilitation sites show a positive trajectory toward native woodlands and away from grasslands. These findings demonstrate that rehabilitation sites are developing diverse invertebrate assemblages comparable to local remnant woodlands in relatively short time frames.

• Invertebrates identified to order level were found to provide useful data to evaluate trophic structure.

Introduction
Open-cut coal mines in the Bowen Basin in central Queensland create large areas of mined lands requiring revegetation. Most coal mines in this region aim to restore native vegetation and associated fauna but not necessarily the original landscape. Typically, these landscapes differ from the original, pre-mining conditions in having steeper slopes comprised of homogenized overburden, usually overlain with a thin layer of preserved topsoil of relatively poor quality (Erskine & Fletcher 2013). Notwithstanding this, it is still possible to provide conditions that allow the development of native vegetation and fauna. This is termed rehabilitation, defined as the process of reinstating a level of ecosystem functionality on degraded sites where ecological restoration is not the aspiration, as a means of enabling the ongoing provision of ecosystem goods and services (McDonald et al. 2016).
In the Bowen Basin, the aim of most mine rehabilitation is to establish a self-sustaining ecosystem, resembling natural habitats that were present before mining; native woodlands in this instance. Typically, reference sites comprising natural vegetation in the vicinity of the minesite are used to provide benchmarks against which the progress of the rehabilitation can be evaluated (Brady & Noske 2010;Gould 2012;Oliver et al. 2022). The use of local reference sites in remnant vegetation also provides a benchmark for calibrating impacts of regional influences outside the mining footprint such as climatic drivers of ecosystem change.
While rehabilitation monitoring typically focuses on vegetation, it is now appreciated that all trophic levels, including invertebrates, need to be evaluated to determine the progress of rehabilitated vegetation (Nichols & Grant 2007;Majer 2009;McDonald et al. 2016). Invertebrates play a vital functional role in natural landscapes, providing important ecosystem services such as decomposition, soil turnover and aeration, the release of nutrients, pollination, and herbivory, as well as providing food resources for higher levels, such as vertebrates, in the food web (Stork & Eggleton 1992;Lavelle et al. 2006;Hallmann et al. 2017). Without invertebrates, it has been acknowledged that ecosystems would cease to function (Wilson 1987). Invertebrates are the most diverse of all animal groups and biodiversity is an important criterion upon which the success of ecosystem rehabilitation attempts is judged (Andersen et al. 2004;McDonald et al. 2016). Higher biodiversity implies more niches have been filled, as well as contributing to redundancy in trophic networks and ecosystem processes. Conserving biodiversity is regarded as essential for maintaining ecosystem functioning (Cross et al. 2019).
Previous studies have shown that patterns observed with one taxon do not necessarily reflect those seen in others when comparing restored and reference communities (Longcore 2003;Majer et al. 2007;. Consequently, rehabilitation guidelines recommend that more than one taxon are targeted (Cristescu et al. 2012;Cross et al. 2019). Three taxa were evaluated in this study. Ants were chosen because of their importance in soil health, high relative abundance, and involvement in various functional processes such as predation and seed removal; beetles because of their role in predation and soil and dung turnover, high diversity, and the wide range of niches occupied; and springtails because of their importance in decomposition processes and the microprocessing of organic matter (Stork & Eggleton 1992). All three taxa have been used to evaluate land restoration following mining (Cristescu et al. 2012;Bowie et al. 2019;Casimiro et al. 2019). In addition, the trophic structure was evaluated using the order level of classification (Majer & Brown 1998;. Many of the taxa equate to important feeding guilds such as detritivores, herbivores, and predators (Stork & Eggleton 1992).
This study sought to improve understanding of the development of biodiversity of local fauna in rehabilitation sites on novel coal mine landforms by comparing them to nearby stands of remnant woodlands (reference sites) that were present in the pre-mining landscape. Nearby cleared woodlands (grasslands) were also sampled to provide a further point of contrast for the developing vegetation associated with rehabilitation. As the woody vegetation develops, it is expected that bioindicators associated with successful rehabilitation should show a trajectory toward remnant woodlands but differentiate from grassland sites. Epigaeic invertebrate assemblages of three rehabilitation sites were compared with three grassland and five reference sites, based on annual sampling over two time periods, 2008-2010 and 2011-2013. The metrics used were alpha diversity (i.e., site species richness) and similarity in species composition based on three taxa (ants, beetles, and springtails). In addition, invertebrate order-level abundance data were used to determine if the trophic structure of rehabilitation sites resembled that of reference sites.

Study Area
German Creek mine is situated in the Bowen Basin in Central Queensland, Australia, 205 km west-northwest of Rockhampton (22 58 0 ; 148 33 0 ). About 50% of the German Creek mining lease is uncleared and has a complex mosaic of vegetation types including remnant woodlands and cleared woodlands dominated by pasture grasses. The study region spans the Tropic of Capricorn and has a warm, seasonally wet/dry climate that is typified by long, hot summers and mild winters (Hutchinson et al. 2005). Mean daily maximum temperatures peak in December (34 C) and are lowest in July (22 C) while mean daily minima are highest in January at 23 C and lowest in July at 11 C. Based on the nearest rainfall station, the annual average rainfall is 608 mm (Bureau of Meteorology 2018). Summers are relatively wet with almost 60% of the rainfall falling between December and March. The 2008-2013 study period encompassed a relatively wet climatic cycle with wet season rainfall average to above average, with the wet season from December 2010 to March 2011 exceptional with more than twice the average rainfall, coincident with a La Niña weather pattern from late 2010 to 2012.

Site Selection
The mine at German Creek has been operating since 1981. The mining process involves removing overburden to a 50-100 m depth, typically resulting in the creation of an elevated landform requiring rehabilitation. Rehabilitation is undertaken progressively as old pits close and new pits open. Due to the variable nature of soil pre-treatment, differences in traits of the underlying spoil material, hydrology, proximity to remnant vegetation, climatic conditions when planted, and differences in plant species composition of seed mixes, the result of rehabilitation cannot be predicted with certainty and may be regarded as novel or hybrid ecosystems (Vickers et al. 2012;Doley & Audet 2013;Erskine & Fletcher 2013).
At the time of the study, the operators of the mine were required to rehabilitate mined areas to "bushland," defined as an ecosystem capable of sustaining native flora and fauna species. As such, the target ecosystem for rehabilitation was unspecified remnant woodlands in the vicinity of the minesite. Due to the novel or hybrid nature of rehabilitated ecosystems in the Bowen Basin, it was recommended that the evaluation of Restoration Ecology September 2023 rehabilitation should be based on properties found naturally in a broad range of local vegetation types rather than a specific vegetation type (Doley & Audet 2013). Reference sites were used to define the range of variation in the invertebrate diversity levels of the target ecosystem (Oliver et al. 2022), remnant woodlands, as a measure of rehabilitation success.
There were some pragmatic constraints on reference site selection: for access purposes, sites had to be on the mining lease but away from active mining operations; in addition, sites had to have secure tenure such that they were located in areas of the lease that were likely to be present in the post-mining landscape (i.e., unmined at mine closure). Five reference sites were randomly located within patches of vegetation meeting these criteria; providing a variety of microhabitat conditions relevant to epigaeic invertebrates, such as soils and understorey structural types (e.g. grassy understorey or a mix of grasses and shrubs) (Fig. 1). These consisted of two reference sites with a different understorey structure in each of the two most widespread vegetation/soil types: Poplar Box Eucalyptus populnea woodlands on fertile alluvial soils associated with river and creek flats (W2 and W3) and E. populnea woodlands with a low tree layer of Buloke Allocasuarina luehmannii on low hills with coarse-textured soils of low fertility (W4 and W5). A single representative of a third vegetation type associated with clay plains, E. populnea and Brigalow Acacia harpophylla (W1) provided an example of a shrubby woodland. Typical of landforms in this region, the five reference sites were all on gentle slopes of less than 4 slope. As a negative control (Oliver et al. 2022), representing the starting vegetation type (i.e., a grassland) and as a counterpoint to the developing wooded vegetation of rehabilitation (Bowie et al. 2019), three grassland sites (cleared woodlands) on a similar range of soil types as the woodlands were also selected (P1 on fertile soils, P2 on clay plains and P3 on coarse-textured soils). Typical of pastures in the Central Queensland region, grassland sites were dominated by the introduced pasture grass, Buffel Grass Cenchrus ciliaris (Eyre et al. 2009).
To evaluate rehabilitation progress, three rehabilitated sites that had been established for at least 6 years at the outset of the study in 2008 were selected (Table 1; Fig. 1). All three sites had been landscaped, topsoiled using soils stockpiled before mining, ripped and seeded with a mixture of grasses and native woodland species (Eucalyptus and Acacia). R1 had a relatively deep layer of topsoil on a steep landform with a dense cover of introduced grasses, mainly Rhodes Grass Chloris virgatus and Buffel Grass. R2 was on a gently sloping landform with a moderate cover of mainly Buffel Grass. R3 was on a steep landform with little topsoil and grass cover but with a dense mix of shrubs and trees. All three rehabilitation sites had a developing low tree layer and understorey. However, the basal area of stems was lower than reference sites (1.7-4.3 m 2 /ha compared with 6-9 m 2 /ha). While rehabilitation sites had a lower cover of grasses and other ground-layer vegetation than reference and grassland sites (20-69% compared with 60-84% and 70-89%, respectively); all sites had more than 60% groundcover when leaf and woody litter cover were included (62-93% for rehabilitation sites compared with 95-99% for reference sites and 93-99% for grassland sites). All sites were at least 1 km apart.

Invertebrate Sampling
Annual sampling was undertaken in the post-wet period (i.e., March to May) over 6 years (2008-2013) using pitfall traps. While not recommended for abundance estimates, pitfall traps are useful in biodiversity surveys aimed at documenting species' presence or absence in specific habitats (King & Hutchinson 2007). At each site, eight pitfall traps (55 mm diameter by 110 mm deep containing approximately 200 mL of 70% ethanol/2% glycerol as a preservative) were set at 25 m intervals along a 200 m transect. Obvious disturbances such as tracks, or meat ant nests were avoided. To avoid "digging-in" effects, where soil disturbance may attract some species of ants, a soil auger of the same diameter as the pitfall traps were used to provide suitably sized holes with minimal soil disturbance (Williams et al. 2012). To ensure weather conditions were not biasing trap catches, the pitfall traps were opened for 3 days and nights (approximately 72 hours) on the same day at all sites. Sites were sampled in either March or April of each year to coincide with favorable conditions for pasture growth and associated invertebrates in the post-wet period.
The eight subsamples from each site were combined and the invertebrates were sorted to "order level" (Naumann 1991), the abundance of each taxon was recorded, and the specimens were preserved in a mixture of 70% ethanol/2% glycerol for longterm storage. Exceptions to the order level of classification were Formicidae (ants) which were enumerated separately from Other Hymenoptera (wasps and bees). Collembola (springtails) were categorized as elongate-bodied (orders Poduromorpha and Entomobryomorpha) or globular springtails (order Symphypleona).
Following substantial rainfall, one of the poduromorph elongate-bodied springtails, Brachystomellidae sp. S626, was super-abundant at one site in 2013, with approximately 72,000 individuals captured in two pitfalls at R3. Data from these two pitfalls were excluded from abundance-based analyses and data were adjusted accordingly.
Ants were identified into genus using taxonomic keys (Shattuck 1999;Andersen 2000) and a voucher collection at Central Queensland University, and then to morphospecies (Oliver & Beattie 1996). Springtails and beetles were identified into family using keys (Naumann 1991) and then into morphospecies. Morphospecies were assigned a unique identifier and vouchered.

Evaluation of Rehabilitation Progress and Data Analysis
Site species richness of each taxon identified to morphospecies level (ants, beetles, and springtails) was used to evaluate change in alpha diversity. To enable the temporal trend (trajectory) to be evaluated, species richness data of sites were based on two time periods, 2008-2010 (T1) and 2011-2013 (T2). To ensure that sampling intensity was sufficient (Oberprieler & Andersen 2020), the 3 years of sample data in each time period were pooled to provide a cumulative measure of species richness for each site, with the sites providing replication within each time period. For each taxon, a two-way ANOVA was performed using time period (T1: 2008-2010 and T2: 2011-2013) and  Table 1 for rehabilitation site descriptions). habitat (reference, grassland, or rehabilitation) as categorical variables. Where the main effects were significant, Fisher's paired-comparison tests were used to determine if habitats differed in species richness. Whether rehabilitation sites differed from reference sites in the two time periods was of particular interest to the study, so paired-comparison tests based on the interaction between time period and habitat were also applied. Scenarios indicative of successful rehabilitation were based on the interaction between time period and habitat, with two alternatives: rehabilitation sites resembled reference sites in species richness in both time periods or there was a significant difference in T1 but not T2. Using this approach, a rapid attainment of a comparable species richness to reference sites would be indicated by scenario 1 or a positive trajectory would be indicated by scenario 2 with a change in the species richness of rehabilitation sites from being significantly lower to that of reference sites in T1 but not in T2. In contrast, a lack of progress in rehabilitation was indicated if metrics remained distinct from the reference ecosystem in both T1 and T2 or if there was a negative trajectory with no difference from the reference sites in T1 but a significant difference in T2. Grasslands were included to provide a negative control representative of the starting state of the rehabilitation sites: it is expected that diversity metrics of invertebrate assemblages of grasslands would remain fixed relative to the remnant woodlands comprising the reference sites as found in other studies (Vasconcelos 1999;Yates et al. 2011;Bowie et al. 2019) and that rehabilitation sites would show a shift away from this state as vegetation structure develops. Species assemblages of rehabilitation and reference sites were compared using presence-absence data pooled within each 3-year time period and by applying the Jaccard similarity index to pairs of sites (Clarke & Warwick 2001). Nonmetric multidimensional scaling ordination was used to examine relationships between samples based on the invertebrate assemblages. The same approach used for univariate data was used to evaluate rehabilitation success based on species assemblages, using a permutation test, time period by habitat, and their interaction (PERMANOVA) (Anderson et al. 2008). Since this test is sensitive to data dispersion and may confound differences among groups with differences in scatter within groups (Anderson 2001), a multivariate homogeneity of dispersion test (PERMDISP, PRIMER-e v7) was also applied.
The order level of classification was used to evaluate the trophic structure of the invertebrate assemblage. Invertebrate taxa with relatively uniform feeding habits or likely to be dominated by one trophic category were placed in one of three groupsdetritivores (mites, both springtail taxa and book-lice); herbivores (bugs, grasshoppers, thrips, butterflies, and moths); and predators (spiders, pseudoscorpions, and wasps) (Naumann 1991). Taxa with a broad range of feeding habits such as ants, beetles, and flies could not be placed into any category and were not included in the analysis . The influence of time period and habitat (and their interaction) on the trophic structure was evaluated using multivariate analysis of variance (MANOVA). Abundance values were log10 + 1 transformed to meet the assumptions of ANOVA (normality and homoscedasticity) (Quinn & Keough 2002). All statistical tests were considered significant at p < 0.05.

Ants
A total of 130 morphospecies of ants from 38 genera were identified over 6 years ( Table 2). The most species-rich genera were three formicine taxa-Camponotus (16), Melophorus (12), and Polyrhachis (7), three myrmicine taxa-Monomorium (9), Pheidole (8), and Meranoplus (6), and the dolichoderine Iridomyrmex (8). The latter was also the most abundant genus in all habitats, accounting for almost two-thirds of the ant catch. Four species of ants were widespread in reference sites (i.e., present in three or more of the five woodland sites) and absent from rehabilitation: Notoncus sp. AG146, Polyrhachis sp. AG64, Melophorus sp. A2312, and Meranoplus sp. A2230, of which the latter two were recorded from one of the pasture sites. Only one species was restricted to rehabilitation sites and absent from reference sites, Paratrechina sp. A2490. This species was also absent from grassland sites. Of the species found in one habitat only, only Notoncus sp. AG146 comprised more than 0.1% of the total catch of ants, the remainder being uncommon. Ant species richness varied significantly by habitat and the interaction between habitat and time period but not by time period (two-way ANOVA, time period F 1,16 = 3.247, p = 0.090, habitat F 2,16 = 4.047, p = 0.038, time period Â habitat F 2,16 = 4.047, p = 0.038). Rehabilitation sites had significantly fewer ant species than reference sites (29 AE 2 compared to 37 AE 2; paired-comparison test, p < 0.05; Table S1), as did grassland sites compared with reference sites (24 AE 2, p < 0.05). However, paired-comparison tests on time period by habitat showed that rehabilitation sites had significantly less ant species than reference sites in T1 but a similar number of species in T2 (Table S2; Fig. 2, p < 0.05), indicating successful rehabilitation based on this metric based on scenario 2, that is a positive trajectory toward reference sites. This partly reflected a fall in ant species richness (i.e., species loss) from T1 to T2 in reference sites (paired-comparison test, p < 0.05). Unlike rehabilitation sites, numbers of ant species in grassland sites were significantly less than in reference sites in both T1 and T2 (paired-comparison test, p < 0.05). With regard to the negative control provided by grassland sites, rehabilitation sites had comparable numbers of ant species in T1 and a trend toward more in T2, although this was not significant (p = 0.053).
Ordination of the ant composition data showed clustering by time period and habitat (Fig. 3). Permutation analysis confirmed this (PERMANOVA: time period F 1,16 = 2.019, p = 0.002, habitat F 2,16 = 2.924, p = 0.001, time period Â habitat F 2,16 = 1.019, p = 0.412). Habitats had similar levels of dispersion (PERMDISP: F 2,19 = 2.217, p = 0.248) showing that differences in habitat were due to location and not differences in dispersion. Pairwise tests confirmed that both rehabilitation and grassland habitats had different ant composition to the reference sites and from each other (Table S3). However, when partitioned into time periods, pairwise tests showed that the ant composition of rehabilitation sites resembled the reference sites in T2 but not T1 (Table S4). In contrast, the ant composition of grassland sites remained distinct from reference sites in both T1 and T2. Thus, the ant composition of rehabilitation sites showed a positive trajectory toward that of reference sites whereas grassland sites did not.  : 2008-2010 and T2: 2011-2013) for ants, beetles, and springtails. Regional diversity levels for each taxon in T1 and T2 were 101 and 93 for ants, 78 and 97 species for beetles and 43 and 35 species for springtails. Results of paired-comparison tests on time period by habitat for each taxon indicated by letters, those with significant differences (p < 0.05) have different letters; all comparisons for the springtail group were non-significant. Invertebrate diversity of post-mining revegetation

Beetles
A total of 133 morphospecies of beetles representing 36 families were captured over the 6 years of sampling (Table 2). The most speciose families were Staphylinidae (20), Scarabaeidae (17), and Carabidae (15). There were four species that were widespread at reference sites (i.e., three or more of the five sites) and absent from rehabilitation sites: two dung-feeding scarabs-Scarabaeidae sp. G836 and Scarabaeidae sp. G742, the detrital feeding Bolboceratidae sp. G1185, and a wood borer, Anobiidae sp. G749. Except for a single specimen of Scarabaeidae sp. G836, these species were also absent from pasture sites. Three species were found only at rehabilitation sites (i.e., two or more of the three) and absent from reference sites: Chrysomelidae sp. G1004, Cleridae sp. G877, and Nitidulidae sp. G643. Of these, only the latter was present at pasture sites. Two species were typical of pastures, being present in all three pasture sites but absent from the other habitats-Anthicidae sp. G720 and Melyridae sp. G803. Beetle species richness did not vary significantly by time period, habitat, or their interaction (two-way ANOVA, time period F 1,16 = 2.846, p = 0.111, habitat F 2,16 = 2.916, p = 0.083, time period Â habitat F 2,16 = 2.400, p = 0.123). Overall, both rehabilitation and grassland sites had only slightly fewer beetle species than reference sites (19 AE 1 compared to 22 AE 1). The number of beetle species found in rehabilitation sites increased from T1 to T2, matching the increase in reference sites. In contrast, grassland sites showed the opposite pattern, declining from T1 to T2, with significantly fewer beetle species than reference sites in T2 but not in T1 (Table S2).
Ordination of the beetle composition data showed clustering by time period and habitat ( Fig. 3; PERMANOVA: time period F 1,16 = 2.398, p = 0.001, habitat F 2,16 = 1.952, p = 0.001, time period Â habitat F 2,16 = 1.271, p = 0.056). Habitats had similar levels of dispersion (PERMDISP: F 2,19 = 0.426, p = 0.753) showing that differences in habitat were due to location and not dispersion. Pairwise tests confirmed that both rehabilitation and grassland habitats had different beetle composition to the reference sites (Table S3). However, when partitioned into time periods, pairwise tests showed that the beetle composition of rehabilitation sites resembled the reference sites in T2 but not T1 (Table S4), as did the beetle composition of grassland sites. Overall, the beetle composition of rehabilitation sites showed a positive trajectory toward that of reference sites whereas this was only weakly expressed in grassland sites.

Springtails
Of the 53 morphospecies of springtails, elongate-bodied springtails were the most speciose with four species of Poduromorpha and 22 species of Entomobryomorpha, the remaining 17 species being globular springtails (Symphypleona). Eleven species of springtail were found only at reference sites, seven at grassland sites and two at rehabilitation sites, all uncommon. Two species were relatively widespread (i.e., occurred in three or more of the reference sites) but absent from rehabilitation. These were two species of poduromorph elongate springtails in the family Onchiuridae sp. 696 and 900, the latter also found at one pasture site. In contrast, all species that were widespread in rehabilitation (i.e., two or more of the three sites) were present in at least one reference site. Only one species was relatively widespread in pasture sites and absent from other habitats-Sminthuridae sp. 1290.
Springtail species richness of rehabilitation sites matched that of reference sites (17 AE 1 compared to 18 AE 1) as did grassland sites (18 AE 1) (two-way ANOVA, time period F 1,16 = 0.287, p = 0.599, habitat F 2,16 = 0.797, p = 0.468, time period Â habitat F 2,16 = 0.073, p = 0.930). Springtail species richness changed little from T1 to T2 in all three habitats, all comparisons being non-significant. Considering pairwise tests within the time period, species richness of both grassland and rehabilitation sites resembled that in reference sites in both T1 and T2.
Ordination of the springtail composition data showed clustering by time period but not for habitat or their interaction (PERMANOVA: time period F 1,16 = 3.869, p = 0.001, habitat F 2,16 = 1.549, p = 0.063, time period Â habitat F 2,16 = 0.889, p = 0.606). Habitats had similar levels of dispersion (PERMDISP: F 2,19 = 0.726, p = 0.561) confirming that dispersion did not influence these results. All three habitats in T1 were in close proximity in ordination space indicating a similar springtail composition at this time, and all three shifted in the same direction in T2. Pairwise tests showed that the springtail composition of rehabilitation sites resembled that of reference sites and grassland sites, irrespective of the time period (Table S4), demonstrating that the springtail composition of rehabilitation sites was comparable with other habitat types at the outset of the study and consistent with scenario 1, that is rapid attainment of comparable species richness to reference sites.

Trophic Structure
To examine the effects of the time period and habitat on trophic abundance, a two-way MANOVA was performed on logtransformed data. The abundance of invertebrates declined from T1 to T2 (MANOVA: time period F 3,14 = 10.952, p = 0.001, habitat F 6,28 = 2.070, p = 0.089, time period Â habitat F 6,28 = 1.652, p = 0.170), mostly due to a fall in detritivore numbers (Fig. 4). However, irrespective of the time period, all three habitats had a comparable trophic structure with detritivores dominant in abundance, herbivores intermediate and predators least abundant. Of the three habitats, rehabilitation sites changed the least in abundance from T1 to T2. Overall, this indicates that epigaeic invertebrate assemblages of rehabilitation sites rapidly attained a similar trophic structure to that present in reference sites.

Discussion
This study demonstrates that invertebrate diversity metrics provide useful criteria for evaluating whether rehabilitated landforms are developing a natural ecosystem. Springtail assemblages of rehabilitation sites were found to show congruence in relatively early-stage rehabilitation vegetation succession, while ant assemblages of rehabilitation sites had more congruence with the fauna of remnant woodland habitat in the latter half of the study, indicating a positive trajectory in invertebrate species composition toward reference sites. Overall, these findings suggest that rehabilitation sites are developing diverse invertebrate assemblages comparable to local remnant woodlands.
Ant species richness of rehabilitation sites showed congruence with reference sites in the latter half of the study (T2), partly from declines in diversity levels of reference sites. This decline in ant species richness of reference sites coincided with the peak of the La Niña influence from late 2010 to 2012, with widespread flooding and twice the annual rainfall before sampling. It is possible that the low ant diversity in reference sites in T2 may have been influenced by changes in ant activity associated with saturated soils or even loss of species due to the direct effects of flooding on ground-nesting species (Ballinger et al. 2007;Mertl et al. 2009;Castro et al. 2012). In contrast, rehabilitation sites showed no indication of a loss of ant species in T2, possibly due to lower susceptibility to flooding of the relatively elevated landforms created by rehabilitation practices. It is also possible that rehabilitation sites have developed an ant fauna that is a subset of that found in reference sites and representative of disturbance-tolerant species as found in other studies of reconstructed landscapes (Campbell & Crist 2017). However, other factors may also be involved since grassland sites, which resembled reference sites in land slope and were likely to be equally vulnerable as reference sites to impacts of flooding, had only a slight fall in ant species richness in T2. As expected, the ant composition of grassland sites remained distinct from reference sites in T1 and T2, demonstrating a persistent difference in ant composition. Habitats displaying different patterns of temporal change have also been observed in other studies of ant assemblages (Deblauwe & Dekoninck 2007;Barrow & Parr 2008;Castro et al. 2012). These differences were attributed to variable environmental conditions between habitats and the influence of this on food resources, nest sites, and predators. In particular, moisture availability and vegetation composition and/or structure were hypothesized to affect the distribution of food resources and suitable nest sites (Levings 1983;Castro et al. 2012).
While a relatively rapid rate of attainment of epigaeic springtail assemblages has been observed in other rehabilitation studies (Bowie et al. 2019;Fernandes et al. 2019), this is not always so (Dunger et al. 2004). The reasons for this rapid development in this study are not understood but springtails have been found to be influenced by both plant cover and litter (Majer et al. 2007;Korboulewsky et al. 2016;Errington et al. 2018). Rehabilitation sites had developed more than 60% groundcover (i.e., low vegetation, leaf, and woody litter cover) compared to over 90% in remnant woodlands. It is possible that the rapid development of substantial amounts of plant and litter cover in rehabilitation sites may have contributed to the development of favorable microhabitat conditions for springtail colonization. However, other factors, such as dominance by introduced grasses, leading to the homogenization of faunal assemblages, may also be involved (Samways et al. 1996;Majer et al. 2021).
Beetle assemblages of rehabilitation sites were more problematical, showing a trend toward rehabilitation sites in T2, although this trend was matched by grassland sites. While beetle assemblages have been found to be useful in demonstrating the trajectory of rehabilitation (Davis et al. 2003;Bowie et al. 2019), the matching trend shown by the assemblage in grassland sites suggests that other factors besides succession of vegetation within rehabilitation sites were in play. Beetles tend to have relatively higher dispersal capacity than ants and springtails, leading to greater turnover of species and differences in metacommunity dynamics to the other species groups (Tonkin et al. 2016).
Characterization of the trophic structure in reference sites provided a benchmark against which the rehabilitation sites could be evaluated. Classic trophic pyramid theory predicts that detritivores are most abundant followed by herbivores and then predators (Krebs 2009). Demonstrating that a functional trophic structure was present, all elements of the trophic pyramid (detritivores, herbivores, and predators) were present in similar abundance to that found in the reference sites.
Like remnant woodlands, rehabilitation sites were dominated by several species of dolichoderine ants in the genus Iridomyrmex. Members of this genus are typical of open habitats, such as woodlands, throughout Australia (Andersen 2000;Hoffmann & Andersen 2003). One species, a myrmicine, Paratrechina sp. A2490 was found to typify rehabilitation sites, being present only in rehabilitation sites. Species of Paratrechina are recognized as being "weedy" ants that are characteristic of disturbed habitats throughout northern Australia .
This study has demonstrated that it is necessary to sample more than one taxonomic group when evaluating the progress of rehabilitated sites. While springtail assemblages of rehabilitation sites rapidly attained congruence with those of remnant woodlands, ant and beetle assemblages took longer. The utility of using order level assemblages to evaluate the restoration of trophic capacity was also demonstrated. As in other studies (Majer et al. 2007;Gollan et al. 2011), the ant composition of rehabilitation sites showed a trajectory toward reference sites. The sensitivity of invertebrate assemblages to climatic disturbance events indicates that to fully evaluate rehabilitation success, further research is required over longer time periods (Cristescu et al. 2012;Majer et al. 2013), ideally incorporating several climatic cycles. Yates ML, Gibb H, Andrew NR (2011) Habitat characteristics may override climatic influences on ant assemblage composition: a study using a 300-km climatic gradient. Australian Journal of Zoology 59:332-338. https://doi. org/10.1071/ZO11096

Supporting Information
The following information may be found in the online version of this article: Table S1. Results of Fisher's paired-comparison tests on species richness for each taxon based on habitat (df = 16). Table S2. Results of Fisher's paired-comparison tests on species richness for each taxon based on habitat within two time periods, T1 and T2.