Introduction

Insects play critical roles in ecosystem function and the provision of ecosystem services (Noriega et al. 2018). For example, pollination by insects is crucial to approximately 308,000 (87.5%) angiosperm species and is estimated to have an annual global economic value between $235 and $577 billion USD (Ollerton et al. 2011; IPBES 2016). However, insect populations are declining across the globe (Potts et al. 2010; Hallmann et al. 2017; Sánchez-Bayo and Wyckhuys 2019), which has been at least partly attributed to the destruction, degradation and fragmentation of natural habitats resulting from land-use change and vegetation clearing (Kovács-Hostyánszki et al. 2017; Mir et al. 2017). Consequently, a growing number of studies report threats of cascading extinctions associated with insect declines (Potts et al. 2010; Hallmann et al. 2017; Sánchez-Bayo and Wyckhuys 2019), and severe consequences are expected both for the functional integrity of ecosystems and the provision of ecosystem services that humans rely upon if action is not taken to conserve and restore insect populations (Díaz et al. 2006).

Ecological restoration of degraded ecosystems has been promoted to tackle biodiversity and habitat loss at increasingly large scales (Menz et al. 2013; Perring et al. 2015; Miller et al. 2017). However, restoring natural habitats is a complex, long-term enterprise (Miller et al. 2017), and generally, prioritises the establishment of the vegetation community under the assumption that fauna will recolonise unaided (Palmer et al. 2016; Cross et al. 2020; Tudor et al. 2023). This ‘field of dreams’ hypothesis has been subject to scrutiny as it often fails to consider the complex responses of fauna to restoration succession and specific habitat requirements of recolonising species (Palmer et al. 1997; Kollmann et al. 2016). Despite several calls for greater consideration of fauna in restoration ecology, less than 12% of the current scientific literature report on their recovery in restoration landscapes (Tudor et al. 2023). Failure to understand, promote and maintain functional insect communities in such landscapes may compromise restoration trajectories as insects play critical roles in several ecological processes including pollination, predation, seed dispersal and decomposition (Noriega et al. 2018).

Ecological restoration can re-establish diverse vegetation communities (Koch and Hobbs 2007; Ritchie and Krauss 2012), though it is unclear whether these landscapes support similar recovery of insect communities. Shifts in insect community composition are likely the result of local, small-scale factors that differentially affect species depending on their specific habitat requirements and functional ecology (Hortal et al. 2010). Understanding the factors which either enhance or reduce local biodiversity offers the opportunity to optimise habitat suitability for a diverse suite of functionally important taxa, and potentially ‘fast-track’ their reassembly in restored landscapes (Purvis et al. 2020). Directing attention to reinstating functional insect communities, such as pollinators, may also have positive flow-on effects to plant communities by increasing reproductive success and genetic stability (Grass et al. 2018; Kremen et al. 2018), which can promote long-term restoration success (Menz et al. 2011). However, there are key knowledge gaps concerning how the return of insect communities relate to the diversity, composition, and structure of vegetation communities, particularly in floristically and entomologically diverse ecosystems (Majer et al. 2007; Marques et al. 2017).

When restoration is undertaken in response to ongoing disturbance it can result in a mosaic of different successional ages across the landscape, providing model environments to assess the spatiotemporal effects of ecological restoration (Cristescu et al. 2012). We assessed the compositional changes in both vegetation and insect assemblages across a ~ 40-year sequence of post-mining rehabilitation and restoration in the Northern Jarrah Forest in the biodiversity hotspot of southwest of Western Australia (Thackway and Cresswell 1997), using the adjacent unmined Jarrah Forest as a reference system (Fig. 1). The overarching aim of this research was to examine how the species richness, abundance, diversity and composition of the plant and insect communities varied across the sequence of ageing restoration and determine the convergence in similarity between sampled communities and the unmined reference forest. More specifically, it was hypothesised that the insect communities present in older restoration would be more similar to reference communities when compared to insect communities in early-successional restoration(Palmer et al. 1997). Moreover, we expected the functional composition of the insect community to change across restoration successional ages. As such, we also assessed how certain functional guilds (i.e., detritivores, fungivores, herbivores, parasitoids, predators and pollinators) contributed to the relative abundances of the insect assemblages and whether plant community composition or biodiversity indices were significant drivers of variation in insect reassembly in the floristically and entomologically diverse Jarrah Forest (Marques et al. 2017).

Fig. 1
figure 1

Locality of the Jarrah Forest Bioregion (shaded in green; as defined by the Interim Biogeographic Regionalisation of Australia) in the southwest corner of Western Australia (a) and location of study area (white square) in the Jarrah Forest (b) with sites sampled across a landscape matrix of post-mining restoration (c). The polygons in panel (c) represent patches of restoration in a broader landscape of unmined Jarrah Forest last disturbed by logging approximately 80 years prior to our study. The green shading gradient in panel (c) indicates restoration age, where the darkest shading indicates the oldest restoration. Sites in restoration plots are represented by black circles, and reference sites in unmined forest are represented by black triangles. The numbers next to each restoration site indicate the year that restoration was completed

Full size image

Methods

Study area

Bauxite mining has occurred in the study region since 1963, with approximately 550–600 hectares of native vegetation cleared, mined and restored annually (Koch 2007). The mining is shallow (< 5 m deep) and takes place in a rolling progression of uniform intensity, leaving a mosaic of restored land amongst adjacent unmined forest following ecological restoration (Koch 2007). Ecological restoration techniques have developed markedly over the nearly 60-year period of operation but standard procedures include landscaping, topsoil and ‘habitat’ return, contour ripping, seeding, planting and fertilisation (Koch 2007). Prior to 1988, post-mining activities were characterised by overstory plantations of Eucalyptus species that were resistant to jarrah dieback, but not indigenous to the Jarrah Forest (Shearer and Tippett 1989). Since 1988, recovery activities have returned only indigenous species with an overstorey of Eucalyptus marginata Sm. and Corymbia calophylla (Lindl.) K.D.Hill & L.A.S.Johnson. Current restoration objectives are to develop post-mining landscapes that are reflective of the surrounding unmined forest that integrate with forest management regimes (Gardner and Bell 2007). However, virtually no pristine old-growth Jarrah Forest remains (McCaw et al. 2011), and therefore the most ecologically intact reference forest is represented by unmined areas that were logged nearly a century ago (Heberle 1997). As rehabilitation and restoration both fall in the restorative continuum of activities (Gann et al. 2019), and recognising that the goal of post-1988 activities was to return functional, resilient, self-sustaining native vegetation communities, all post-mining activities with the aim of vegetation reinstatement are collectively referred to hereafter as restoration.

Seventeen survey sites were laid out at an average of 2.24 ± 0.24 km apart (range 4.66–1.34 km) to ensure spatial independence (Zurbuchen et al. 2010). Fourteen sites were identified across a successional restoration age gradient (Fig. 1; 1981–2018) and three reference sites were selected in adjacent unmined forest, interspersed across the study area. There was an unavoidable north–south gradient in restoration age resulting from the temporal progression of bauxite extraction and subsequent restoration, however, our study area was only 14.3 km long, and was unlikely to impart climatic correlates that influenced the ecology at sampled sites (Hawkins and Felizola Diniz-Filho 2004).

Sample collection and processing

Insect sampling was conducted for a fortnight each month between July and December 2019, using yellow and blue ultraviolet polypropylene vane traps (SpringStar Inc., Woodinville, Washington). Vane traps are very effective in sampling pollinator communities, with evidence to suggest they are also suitable for non-pollinating taxa and capture a representative sample of the flying-insect community (Hall and Reboud 2019). Despite the established biases this trapping methodology has for flying-insects, we were targeting this bias with our sampling efforts as other taxa, such as ground-dwelling invertebrates, have been explored elsewhere (reviewed in Majer et al. 2007). Each survey site consisted of a 50 × 50 m quadrat of four traps (two yellow and two blue) suspended approximately 1.2 m above the ground from wooden stakes. The colour of the trap mounted at each of the four corners at each survey site was randomised at the time of establishment by coin flip. Each trap was filled with a preservative solution comprising ethylene glycol (CAS Number: 107-21-1) and denatonium benzoate (CAS Number: 3734-33-6) diluted 2:1 with water. All traps were cleared and shrouded with black polyethylene covers between trapping periods to ensure no insect capture occurred during these times and that the stimulus of the traps was removed from the local insect community between surveys.

Trap contents were transported to the Kings Park Science laboratory (Kings Park, Western Australia), where specimens were filtered from the ethylene glycol solution using a 355-micron sieve before being rinsed in water to remove excess ethylene glycol and preserved in 70% ethanol. Analyses were restricted to insect Orders; Diptera, Hymenoptera, Lepidoptera and Coleoptera (Ollerton 2017). Over 70% of Australia’s insect taxa remain undescribed (Austin et al. 2004), and therefore, analyses were conducted upon morphospecies to be representative of species diversity without losing taxonomic resolution. However, formal identifications were reported at family level because genus and species level identifications are difficult to obtain for all specimens due to the incomplete systematics of the insect community in the south-western regions of Western Australia. Specimen family identifications were confirmed with the aid of open-access galleries and databases of insect taxa in the Atlas of Living Australia (ALA, http://www.ala.org.au), Australian National Insect Collection Database (CSIRO, http://anic.ento.csiro.au/database), and Western Australian Museum (http://museum.wa.gov.au/research/research-areas/terrestrial-zoology), along with interactive insect identification keys (CSIRO, http://anic.ento.csiro.au/insectfamilies), and the references provided therein. Families were assigned to one of six functional guilds: Detritivores, fungivores, herbivores, parasitoids, pollinators, or predators (Table A.3). Determination of functional guilds was based on mouth-part morphology of adult specimens and existing literature (CSIRO 1991).

Site characterisation

Vegetation surveys were conducted across August and September 2019, using four replicate 4 × 4 m plots at each site to determine plant species richness (species counts) and abundance (individual counts). All species that were present above ground during the survey period were included. However, some annuals and ephemeral plants may not be encompassed during this survey period. While this sampling effort is smaller than previous studies (e.g. Koch and Ward 1994), and did not fully encompass the temporal variation in flowering phenology across all seasons, the aim was to develop consistent vegetation correlates against which to assess the response of the insect community. Simpson’s diversity of the plant community was calculated from measures of species richness in each plot.

Statistical analysis

All statistical analyses and ordinations were performed in the R statistical environment (version 3.6.2) using RStudio Version 1.2.5001 (R Development Core Team 2022). Summary statistics are presented as means ± standard error for restoration age classes (categorical; year ranges). We were limited to the categorical classifications as restoration approaches changed over time and did not reflect a continuous treatment gradient. Moreover, we were unable to assign a definitive ‘age’ to the unmined sites due to a lack of data on the last date of disturbance. The main effect of restoration age class on plant species richness, abundance and Simpson’s diversity, as well as insect species richness were examined using one-way analysis of variance (ANOVA; Kaufmann and Schering 2014) using the ‘anova’ function in the ‘stats’ package (R Development Core Team 2022). Insect species abundance and Simpson’s diversity across the restoration age classes were examined by non-parametric Kruskal–Wallis test using the ‘kruskal.test’ in the ‘stats’ package (R Development Core Team 2022). Kruskal–Wallis test was selected for these two metrics because the data did not meet the assumption of normality for parametric analysis, tested by Shapiro–Wilk test for normality using ‘shapiro.test’ function in the ‘stats’ package (R Development Core Team 2022). Post hoc pair-wise comparisons were conducted to determine differences between restoration age classes using the TukeyHSD function in the ‘stats’ package following parametric analysis (R Development Core Team 2022), while the ‘dunnTest’ function in the ‘FSA’ package was used following the non-parametric Kruskal–Wallis analysis (Ogle and Ogle 2017).

Generalised linear mixed-effect models (GLMM; McCulloch and Neuhaus 2005) assuming a Poisson error distribution were used to examine the effects of the vegetation abundance, richness and diversity on the total insect abundance and functional guild abundances. Models were fit using the glmer-function in the R statistical package ‘lme4’ (Bates 2007) with the syntax; ‘response variable ~ abundance + richness + diversity + (1|Site/Month)’. Sampling site and month were held as random effects in the model, with month nested within site. Prior to GLMM analysis, Pearson correlation coefficients and variance inflation factors (VIF) were assessed using the packages ‘PerformanceAnalytics’ and ‘Car’ setting thresholds for variable exclusion set at < 0.7 and < 3, respectively (Fox et al. 2012; Peterson et al. 2018). Model reduction using the ‘MuMIn’ package (Bartoń 2014) was used to identify the most parsimonious model by Akaike’s Information Criterion for small sample sizes (AICc; Burnham and Anderson 2002). However, model reductions did not substantially increase model parsimony and full models were maintained for final analysis. The predictions from the GLMMs were generated in the R package ‘ggeffects’ with the ggpredict-function (Lüdecke 2018) to visualise significant fixed effects on functional guilds.

A Bray–Curtis dissimilarity matrix was used for a subsequent, non-parametric permutational multivariate analysis of variance (PERMANOVA; Anderson 2005) to determine the effect of restoration age and sampling period on insect and plant community assemblage. Post hoc pair-wise comparisons between variables were conducted to determine site-level differences using the pairwise.perm.manova function in the ‘RVAideMemoire’ package (Anderson 2014; Hervé and Hervé 2020). An indicator species analysis was performed using the multipatt-function in the ‘indispecies’ package (Cáceres and Legendre 2009) to identify insect and plant species that are indicators of restoration and unmined sites. Non-metric multidimensional scaling (NMDS; Agarwal et al. 2007) was used to construct an unconstrained ordination of the plant and insect communities. Vectors were superimposed onto NMDS biplots with significance established through 999 permutations using the envfit function in the ‘vegan’ statistical package (Oksanen et al. 2007). A predictive co-correspondence analysis (CoCA; ter Braak and Schaffers 2004) was used to relate the plant community composition to the composition of the insect community and carried out using the coca-function in the ‘cocorrresp’ statistical package (Simpson 2005). Cross-group co-occurrences between the plant community and insect community were determined using the spiec.easi function in the ‘SpiecEasi’ statistical package (Kurtz et al. 2015), by adapting the supplementary code available from Alric et al. (2020) to community data.

Results

Plant community

A total of 5569 individuals from 145 plant species across 39 families were identified from sample sites (Table A.1). Indicator species analysis revealed that younger restoration (1–5-year-old) was characterised by native Fabaceae including Acacia lateriticola and A. drummondii (Fig. 3a; Table A.1), while older restoration sites and unmined forest were dominated by the native overstorey trees, Eucalyptus marginata, Corymbia calophylla (both Myrtaceae) and the midstorey shrub Bossiaea aquifolium Benth (Fabaceae; Fig. 3a). However, 20 species were unique to unmined forest sites, including Xanthorrhoea preissii Endl. and X. gracilis Endl. (Xanthorrhoeaceae) and Pteridium esculentum (G.Forst.) Cockayne (Dennstaedtiaceae). The oldest restoration sites (> 30 years old) were dominated by the introduced non-native overstorey species Eucalyptus resinifera Sm (Fig. 3a).

The abundance of plants in each plot increased from 1–5-year-old restoration (155 ± 24.3 plants/plot) to 5–15-year-old (468 ± 120.7 plants/plot) and 15–20-year-old sites (464 ± 1.5 plants/plot), then decreased markedly along the restoration gradient to 30–45-year-old sites (160 ± 31.5 plants/plot; F1,11 = 6.26, P < 0.001; Fig. 2a). However, plant species richness and Simpson’s diversity increased along the same age gradient, being highest in 30–45-year-old sites (richness: 36 ± 1.5 species/plot; diversity: 0.9 ± 0.01 species/plot) where it was comparable with both species’ richness (44.7 ± 3.2 species/plot; F1,11 = 1.97, P = 0.162) and Simpson’s diversity (0.9 ± 0.01 species/plot; F1,11 = 2.66, P = 0.08) of unmined forest (Figs. 2b, c). There was also a significant effect of restoration age on plant species composition (Pseudo-F1,15 = 9.73, P = 0.001), with post-hoc pairwise comparisons indicating that community composition in restoration sites remained significantly different from unmined forest even after four decades of development (Fig. 3a).

Fig. 2
figure 2

The effect of restoration age class (years) on plant () and insect () a) species abundance; b) species richness and; c) Simpson’s diversity in restored Jarrah Forest of different ages in southwestern Australia. Reference values from unmined forest are indicated by Mature Reference Forest (MRF). Data are presented as mean ± standard error with differences in lettering in the corresponding colours indicating significant differences in means between restoration age classes as defined by either Tukey’s post-hoc test or Dunn test

Full size image
Fig. 3
figure 3

Unconstrained non-metric multidimensional scaling (NMDS) showing the relative differences in a) plant species composition and indicator species (ɑ < 0.002), and b) insect community composition and indicator insect families (ɑ < 0.05) across a sequence of post-mining restoration in the Northern Jarrah Forest. Points represent restoration age class centroids and 95% confidence intervals; 1–5 year (), 5–15 (), 15–20 (), 20–30 (), 30–45 () and unmined mature remnant forest (MRF; )

Full size image

Insect community

Trapping yielded 15,427 insect specimens representing 169 morphospecies from 86 families (Table A.2). The insect community was dominated by Chironomidae (non-biting midges), which accounted for 46.8% of insect specimens across all sites. Halictidae and Apidae accounted for 18.4% and 15.3% of the remaining 8211 specimens, respectively. Indicator species analysis revealed that the youngest sites (1–5 years old) were characterised by Hymenoptera, namely Apidae and Halictidae, and Coleoptera such as Buprestidae and Coccinellidae (Fig. 3b; Table A1.2). Restoration sites > 5 years of age were characterised by coleopteran families including Curculionidae, Cleridae and Chrysomelidae, the dipteran families Chloropidae and Mycetophilidae and Hymenoptera including Braconidae (Fig. 3b; Table A.2).

Insect abundance was significantly higher in 1–5-year-old (296 ± 43 individuals) and 5–15-year-old restoration (205 ± 41.5 individuals) than in older restoration (KW x2 5,96 = 40.8, P < 0.001; Fig. 2a). Insect species richness was higher in 1–5-year-old (21.6 ± 2.36 species) and 5–15-year-old restoration (20.9 ± 1.76 species) than older restoration, however these differences were marginally non-significant (F5,96 = 2.04, P = 0.07; Fig. 2b). There were no significant effects of restoration age on insect Simpson’s diversity (KW x2 5,96 = 2.06, P = 0.84; Fig. 2c). Neither plant abundance, species richness nor Simpson’s diversity influenced the total abundance of insects (plant abundance: z = − 1.31, P = 0.192; plant species richness: z = − 0.06, P = 0.949; plant species diversity: z = − 0.18, P = 0.860), leaving the univariate metrics of characterising the insect community unexplained by those of the plant community.

There was a significant difference in insect community composition across the restoration age classes (Pseudo-F5,96 = 3.18, P < 0.001). However, this difference was driven by the unique composition present in the 1–5-year-old restoration and insect community compositions > 8 years old became statistically indistinguishable from those of the unmined forest (Fig. 3b; Table A.4). Neither plant abundance (R2 = 0.037, P = 0.169), Simpson’s diversity (R2 = 0.009, P = 0.602) or plant species richness (R2 = 0.003, P = 0.873) explained the compositional variation observed across the restoration landscape.

A predictive co-correspondence analysis based on the first two axes (CoCA 1: P = 0.001; CoCA 2: P = 0.075; Fig. SI.1) demonstrated that the plant community explained 5% of the insect community inertia (3.4% and 1.6% for axis 1 and 2, respectively). A cross-group analysis of co-occurrence highlighted 149 associations between the plant and insect community, 23 of which were negative (Fig. A.2). However, of the 196 morphospecies sampled only 82 (42%) were found to have at least one instance of co-variance with the plant community. The remaining 114 (58%) morphospecies did not present any association with any particular plant species (Fig. A.2).

Insect functional guilds

Analysis of functional guilds (excluding Chironomidae due to their disproportionate contribution to total abundances) revealed that most insects sampled were pollinators and detritivores which accounted for 41% and 28% of the remaining assemblage, respectively. All functional guilds excluding fungivores varied in abundance across restoration age classes (Fig. 4; Table A.5). Pollinators and predators were more abundant in 1–5-year-old restoration (Pollinators: 129 ± 32 individuals, F5,96 = 11.9, P < 0.001; Predators: 13.1 ± 3.4 individuals, F5,96 = 4.4, P = 0.001). Herbivorous insects had higher mean abundance in the mature restoration (18 ± 5.1 individuals, F5,96 = 5.8, P < 0.001), while detritivores had peak abundance in the early-successional restoration (34.8 ± 8.2) and unmined forests (33.3 ± 8.4) with significantly lower abundance in the mature restoration (8 ± 1.4 individuals; F5,96 = 3.14, P = 0.012). Parasitoids had peak mean abundance in the mature restoration (6.2 ± 1.9 individuals) and lower mean abundance in the unmined forest (1.0 ± 0.3 individuals, F5,96 = 5.04, P < 0.001). Neither vegetation species richness, abundance nor diversity could explain the trends observed for detritivores, predators, fungivores or herbivores (Table A.5). However, vegetation abundance significantly explained the decreases observed for pollinators (Fig. 4a; z = − 2.11, P = 0.035) and parasitoids (Fig. 4b; z = − 2.28, P = 0.023) across the sequence of restoration (Fig. 5).

Fig. 4
figure 4

The proportional contributions of different functional guilds: Pollinators (yellow; solid), detritivores (light green; horizontal stripes), fungivores (green–blue; vertical stripes), herbivores (teal; wide diagonal stripes), parasitoid (purple; narrow diagonal stripes), and predators (deep purple; dotted) to insect community assemblages in post-mining restoration (summarised by age range since establishment) and unmined mature reference forests (MRF). Chironomidae (non-biting midges; Diptera) were excluded from this analysis due to their over-representation in sampling

Full size image
Fig. 5
figure 5

Abundance of a) pollinators, b) parasitoids across a gradient of increasing vegetation abundance (indivudal counts/16 m2) in restored and unmined Jarrah Forest of south-western Australia. Both pollinators and parasitoids were modeled to decline as forest habitats increased in vegetation abundance. Shaded regions around marginal effect regression lines represent 95% confidence intervals

Full size image

Discussion

The recovery of degraded habitats through various means of ecological restoration and rehabilitation is a slow process, often taking many decades to ‘successfully’ recover (Suding and Gross 2006). The insect communities at our restored sites were similar in composition to those of adjacent unmined forests following nearly a decade post-restoration, despite the lack of similar convergence in the composition of restored plant communities. This convergence, however, reflected a concomitant decline in insect abundance between early- and late-successional restoration. Similar successional changes have been documented previously (Hemerik and Brussaard 2002), and are broadly consistent with the intermediate disturbance hypothesis (Connell 1978). However, plant abundance, richness and diversity failed to explain the patterns of compositional reassembly for the broader insect community. Functional guild analysis revealed a shift from communities dominated by pollinators in the early-successional restoration, to those dominated by herbivores in the late-successional restoration and detritivores in the unmined habitats.

Compositional shifts and functional drifts

Among the different insect functional guilds that we recorded at our sites, pollinators were the main drivers of compositional variation in response to ecological restoration and were significantly less abundant in late-successional restoration and unmined forests. Pollinator abundance declined in response to increasing plant abundance across the restoration landscape and we speculate that associated increases in vegetation density contributed to this trend with increased density of the vegetation having led to marked changes in forest structure at our sites. Parasitoids also responded significantly to plant abundance; however, we speculate this to be due to an indirect effect driven by host abundances. There is accumulating evidence that changing vegetation structure is a significant driver of insect abundance and diversity (McCabe et al. 2019; Odanaka et al. 2020). This is often attributed to changes in habitat structure leading to modified abiotic conditions which ultimately drive the microclimatic suitability of a site for inhabiting insects (Tomlinson et al. 2017). However, there is also an intimate relationship between flowering plants and the pollinators that are heavily reliant on the floral community for both nutritional and nesting requirements (Ghazoul 2006; Winfree et al. 2011). The early-successional restoration supported an abundant native legume understory of Acacia lateriticola and A. drummondii which provided copious foraging resources for female bees to provision nests when in flower (Bernhardt 1987). However, many larger understorey legumes reach maximum heights following 5 years of growth (Daws and Koch 2015), and many of these legumes begin to senesce, and are outcompeted by other species (Grant et al. 2007; Daws and Koch 2015). While some sources recommend that this legume understory be reduced during rehabilitation or restoration programs because they suppress a more diverse understory (Koch and Davies 1993; Grant and Loneragan 1999), we suggest that their dominance in the early-successional sites may explain the rich pollinator abundance in early-successional restoration through increased foraging resource availability.

The mid-successional restoration fostered insect communities with significantly lower pollinator abundance and supported a functional composition akin to that of the unmined forest. However, the late-successional restoration transitioned into an herbivore-dominated community. Neither plant abundance, species richness or Simpson’s diversity had a significant influence on herbivore abundance, but increased leaf biomass that is common among mature forests, along with the establishment of host plants, may explain the population increases (Juan-Baeza et al. 2015). Leaf herbivores such as weevils (Curculionidae) and leaf beetles (Chrysomelidae) were the primary contributors to the increased herbivore abundance in late-successional restoration (> 30 years old) and several genera in these families are common pests of the Jarrah Forest. Interestingly, stands in this age class were characterised by a novel overstory of Eucalyptus resinifera, a fast-growing species indigenous to the eastern states of Australia incorporated into rehabilitation practices for its resistance to dieback (Phytopthora cinnamomi), which may also contribute to the increased herbivore abundance (Koch and Hobbs 2007). The notable difference in the seeding of non-native overstory species highlights the importance of initial seeding selections in driving differences across higher trophic levels and insect communities that may diverge away from a ‘desired’ trajectory.

The unmined reference forest hosted insect assemblages which were dominated by detritivores, such as crane flies (Tipulidae), which form an important component of the saproxylic forest fauna and contribute to decomposition (Byriel et al. 2020). While no significant effects were observed in response to vegetation abundance, we expect this trend to be partly attributed to increased accumulation of leaf litter and organic matter in the understorey (Byriel et al. 2020). It is equally likely that increases in detritovore abundances across the restoration gradient are also the result of changes in the physical habitat structure that contribute to cooler, shaded microclimates and increased moisture, which facilitates decomposition (Stone et al. 2020).

Interestingly, no significant drivers were identified for predators or fungivores. These functional guilds may respond to emergent vegetation across developing ecological restoration in different ways to other invertebrate groups and the mechanisms underpinning their reassembly are complex (Brown 1991). These mechanisms may relate to a combination of biotic and abiotic factors limiting insect dispersal, fitness and survival that may be meditated by species physiological thresholds (Tomlinson et al. 2017), nutritional and nesting requirements (Vaudo et al. 2015), trophic interactions (Loch et al. 2020), and phenological factors (Havens and Vitt 2016). For example, predator prey dynamics suggest that predators are unlikely to recolonise restoration sites in the absence of a large and diverse community of prey (Johansson et al. 2007), while fungivores are likely to return when cool, damp understories have developed that are most likely to support fungal communities (Thoresen et al. 2021). As such, landscape scale restoration requires the reconstruction of habitat with increased structural complexity and heterogeneity to maximise habitat suitable for a diverse suite of taxa and their specific requirements. Homogenous landscapes are likely to result in significant energetic, reproductive and survival costs to taxa with more complex habitat requirements and this may ultimately impart considerable and cascading impacts on the ecology of restored systems (Tomlinson et al. 2018).

Ecological drivers of insect community reassembly

Vegetation composition might be expected to have the strongest influence on faunal communities by having both direct and indirect effects via structure and by extension, microclimate (Humphrey et al. 1999; Geber and Moeller 2006; Ghazoul 2006), as has been reported previously (Schaffers et al. 2008). However, the poor statistical association between the reassembly of the plant community and that of the insect community of this study is inconsistent with previous reports examining the factors driving insect community reassembly (Moir et al. 2005; Winfree et al. 2011). Only 42% of morphospecies samples presented covariance with the plant community and this may be explained by the generalist nature of many plant–insect interactions, reducing the capacity for certain plants to predict the presence of specific flower-visitors (Grundel et al. 2010). Communities with higher species diversity also often reflect reduced mutual dependencies and increased functional redundancy (Kaiser-Bunbury et al. 2017), further weakening the predictive capacity of plant species composition specific to insect taxa. Stronger relationships between the insect and plant community may be established with a phenological comparison of floral compositions that accounts for the abundance and diversity of foraging resources, rather than taxonomic composition of the broader floristic plant community. As the sampled insects were predominantly nectarivorous pollinators, they are more likely to respond positively to seasonal patterns in the availability of pollen and nectar (Blaauw and Isaacs 2014), and potentially the diversity of foraging sources available in a landscape (Fuccillo Battle et al. 2021). Moreover, insect populations vary greatly from year-to-year, especially in the face of significant interannual shifts in local and regional climate (Dakos et al. 2009). Consequently, the survey efforts here represent only one assessment of the local insect community and the spatiotemporal trends in insect community structure and composition could be further disentangled through concerted long-term monitoring to capture extraneous abiotic drivers of change that were beyond the scope of this study.

Restoration and landscape context in community re-assembly

Ecological restoration is bound by the concept of the ‘reference’ or ‘target’ ecosystem, such that restoration aims to recapitulate these ecosystems and ‘success’ is often determined based on the degree of similarity between the restored ecosystem and the reference (Miller et al. 2017). Our data suggest that insect communities may reassemble to a state representative of unmined Jarrah Forest without intervention following the re-establishment of native vegetation. However, it is important to note that virtually no pristine old-growth Jarrah Forest remains in the Southwest Australian Floristic Region to act as reference communities and the surrounding unmined reference forests assessed in this study are themselves extensively managed landscapes. These sites are recovering from timber harvesting from the late nineteenth century through until the mid-twentieth century (Heberle 1997), and have been subjected to prescribed burning and forest management regimes (Bradshaw et al. 2018). These management strategies have generated forests that are more dense and shaded than natural, undisturbed forests (Mills 1989). Additionally, Abbott et al. (1989) estimate that Jarrah has a life cycle ranging between 500 and 1000 years, and the unmined forest in the study region is unlikely to have reached the state of a mature ‘climax’ Jarrah Forest community or be reflective of a natural, undisturbed forest. Undisturbed Jarrah Forests were not particularly dense and shaded (Heberle 1997), and likely permitted a more diverse flowering understory to support pollinator foraging, so many pollinating taxa, such as bees, were plausibly more abundant in ‘climax’ Jarrah Forests.

The insect pollinators in the Northern Jarrah Forest seem to have responded to post-mining restoration in accordance with the ‘Field of Dreams’ hypothesis (Palmer et al. 1997), which is in stark contrast to other examples of ecological restoration in the southwest of Western Australia, where autonomous post-disturbance reassembly has not occurred (Frick et al. 2014). The landscape context at our study site may have been uniquely beneficial in facilitating this return, in that the disturbance represented a suite of small sites among a matrix of unmined forest. Consequently, each area to be restored is relatively small, and surrounded by an adjacent source population of insects that can readily recolonise developing vegetation (Fig. 1), potentially lending itself to the rapid convergence of the restored insect communities with the surrounding forest. However, this outcome may not be achieved in more degraded or highly fragmented landscapes (Menz et al. 2013), or in restoration programs that tackle clearing and fragmentation at scales exceeding thousands of hectares, such as agricultural restoration. Isolated broad-scale restoration may experience slower rates of reassembly or be recolonised by non-target taxa, leading to longstanding compositional differences (Cusser and Goodell 2013). This highlights the imperative of allocating resources towards maintaining landscapes with high degrees of connectivity and heterogeneity to facilitate biodiverse, resilient, functional, and sustainable ecological communities, whether that be through active conservation, ecological restoration, or other environmental management initiatives. While landscape connectivity is expected to have substantial impacts on the insect community ecology and delivery of pollination services (Kremen et al. 2007), supporting empirical data are rare (Mitchell et al. 2013). Therefore, concerted efforts to understand the mechanisms by which connectivity influences the structure and function of insect communities, and at what scale, are also necessary to better inform restoration and landscape management.

Conclusions

Ecological restoration may consider trade-offs between pollinator enhancement, designated land-use objectives, and other forest values. This could be achieved through maximising landscape heterogeneity and habitats with increased structural complexity across later successional stages along with increasing phenological diversity in seed mixes and floral abundances to increase nutritional resource availability. Enhancing the insect biodiversity value of restored landscapes requires a greater understanding of the factors driving insect habitat suitability and selection. This cannot be achieved by descriptive science alone and requires a mechanistic approach to develop causal understanding of landscape and land use change. Therefore, future research should seek to understand the reproductive, physiological, and energetic requirements of strategically selected taxa, for example, ecosystem service providers or keystone species (Tomlinson et al. 2022), and explore the role of certain management techniques that may work to optimise the biodiversity value of restored habitats.