Forest structure and heterogeneity increase diversity and alter composition of host–parasitoid networks

Antagonistic host–parasitoid interactions can be quantified using bipartite and metanetworks, which have the potential to reveal how habitat structural elements relate to this important ecosystem function. Here, we analysed the host–parasitoid interactions of cavity‐nesting bees and wasps, as well as their abundance, diversity and species richness with forest structural elements from 127 forest research plots in southwestern Germany. We found that parasitoid abundance, diversity and species richness all increase with host abundance, a potential mediator between parasitoids and forest structure. Both parasitoid abundance and diversity increased with stand structural complexity, possibly mediated by the abundance of hosts. In addition, parasitoid abundance increased with increasing standing deadwood and herb cover. The bipartite networks of host–parasitoid interactions showed higher connectance with increasing standing deadwood, herb cover and host abundance. Analyses of interactions within the host–parasitoid metanetwork revealed that increasing host abundance and decreasing canopy cover diversify the suites of interactions present at the plot level. These results demonstrate that forest structural elements can improve the stability and resilience of host–parasitoid networks by promoting parasitoids and diversifying interactions in ecological networks.


INTRODUCTION
Parasitism can be quantified and visualised as a network of directed bipartite interactions (Dormann et al., 2009;Thierry et al., 2019) between parasitoids and their hosts.Bipartite networks can quantify and visualise parasitism at varying spatial and temporal scales (Mora et al., 2020;Torné-Noguera et al., 2020) and along environmental gradients (Fisogni et al., 2022).In so doing, it is possible to test the importance of habitat components on parasitism using indices quantified from individual networks.Connectance, for example, is calculated by dividing the linkage density by the number of species present with a high value indicative of a generally well-connected network, which has been associated with higher robustness (Estrada, 2007).Parasitism can be further visualised across whole host-parasitoid communities using metanetworks (Librán-Embid et al., 2021), which quantify the co-occurrence of interactions.Thus, metanetworks have the potential to evaluate the prevalence of interactions at large spatial scales and across multiple sites and habitats (Grainger & Gilbert, 2016), where previous approaches have examined spatially or temporally isolated interaction networks (Hagen et al., 2012).Additionally, modelling of metanetwork interaction indices accounts for the spatial distributions of species (Emer et al., 2018;Li et al., 2020), revealing interactions that link distinct networks and thus support ecological functions at large spatial scales.
Parasitism is a grouping of life history strategies characterised by species living in close association with their hosts to utilise them or their resources for survival (Gullan & Cranston, 2014).A subset of parasitic species are parasitoids and kleptoparasites (parasitoids hereafter), which, though having slightly different strategies, typically result in the death of hosts to complete their lifecycles (Sedivy et al., 2013), making them functionally similar.Parasitism is an important ecosystem function, regulating populations of hosts (Lynch et al., 1998), which may be pests (Menalled et al., 2003;Mills, 2010), pathogen vectors (Plowright et al., 2017) or invasive species (Chabert et al., 2012;Duan et al., 2013).
The top-down regulation of host populations by species at a high trophic position has important effects on host population dynamics (Hassell, 2001), thus cascading effects on the biotic and abiotic resources that hosts utilise (Tan et al., 2020;Vidal & Murphy, 2017).In addition, parasitoids are particularly sensitive to bottom-up effects (fluctuations in host populations) (Hassell, 2001;Singh, 2021) and microclimate variation (Kankaanpää et al., 2020;Wenda et al., 2022), making parasitoid interaction networks good proxies for evaluating habitat integrity and quality (Anderson et al., 2011;Grass et al., 2018).
In forests, the structural elements or physical structure characteristics of forest habitat (e.g.tree canopies) influencing parasitism interactions have only generally been described (Eckerter et al., 2022;Laliberté & Tylianakis, 2010;Staab et al., 2016).The structural elements of forests, and more broadly other habitats, can be quantified to allow meaningful investigations of their importance for various aspects of biodiversity (Rappa et al., 2022;Storch et al., 2018).Canopy cover, for example, while varying between the sites, is a characteristic of forest habitat and, thus, can be hypothesised to have a strong influence on forest biota (Fornoff et al., 2021;Gustafsson et al., 2019;Oettel & Lapin, 2021).Forb cover, by contrast, or the percentage of flowering plant species, is a quantifiable metric of meadow habitats (Jiang & Hitchmough, 2022).The spatial arrangement of forest structural elements can be strongly influenced by management practices such as retention forestry (Gustafsson et al., 2012).
Retention forestry emphasises maintaining forest habitat structural elements that would have been removed during timber harvest as a result of traditional management practices (e.g., deadwood removal) (Storch et al., 2020).Unique forest structural elements such as deadwood are important components of forest habitats and can have important influences on parasitism networks via the abundance and species richness of their potential hosts (Eckerter et al., 2021;Rappa et al., 2023), feeding resources (Heimpel, 2019) and microclimates (Laliberté & Tylianakis, 2010).Thus, understanding the potential effects of retention forestry on parasitism via the arrangement of forest habitat structural elements will provide valuable insights into the maintenance of this important ecosystem function (Baho et al., 2017;Holling, 1973).
Most insect species with parasite/parasitoid life history strategies are within the Hymenoptera (bees, ants and wasps) (Gullan & Cranston, 2014), and thus closer examination of these insects is necessary to characterise parasitism as an ecosystem function.Cavity-nesting bees and wasps, as well as their associated parasitoids, can be easily and reliably sampled using trap nests (Staab et al., 2018) and are thus ideal study organisms.Cavity-nesting bees and wasps locate hollow cavities in deadwood or soil exposures and provision resources in a series of cells for their offspring.In forests, cavity-nesting bees and wasps are considered secondary saproxylics, nesting in the deadwood exit holes created by primary saproxylic organisms (Westerfelt et al., 2015), and can thus be useful indicators of forest structure (Eckerter et al., 2021, Rappa et al., 2023).During nest building, or following nest completion, parasitoid species exploit the opportunity to lay their eggs on the resources provisioned and on/within the host egg or larvae.The parasitised cells can then be easily identified by the presence of a parasitoid individual/ cocoon or a host cocoon with parasitoid exit holes.
In the present study, we seek to determine if the gradients of the amounts of structural elements prioritised by retention forestry for their importance for biodiversity can influence the stability and resilience of parasitism as an ecosystem function.Here, we follow Rappa et al., 2023, which investigated the importance of forest structural elements for cavity-nesting bees and wasps, which are the hosts of the parasitoids used in the present study.
To assess the influence of forest structural elements on parasitism as an ecosystem function, we tested the following hypotheses: 1) parasitoid abundance, species richness and diversity will increase with structural elements in forest habitats, namely standing/lying deadwood, herb cover and canopy cover as these are the most important resources for the foraging and nesting of their hosts (Rappa et al., 2023), which differ from parasitoids that are primarily limited by host availability.Furthermore, we hypothesise that the environmental variables structuring parasitoid communities will differ from those structuring host communities because of differing resource requirements, such as the predatory nature of some host species.2) The weight and diversity of the bipartite host-parasitoid interaction networks will increase with the increasing forest structural elements that influence the amount of foraging and nesting resources available to hosts (e.g., greater herb cover) but also the diversity of resources following the habitat-heterogeneity hypothesis (e.g., forest strata occupied by vegetation, greater stand structural complexity and understorey species richness) (Cramer & Willig, 2005;MacArthur & MacArthur, 1961).3) The arrangement of host-parasitoid interactions will be determined by forest structural elements that have the potential to create stand-level heterogeneity via more diverse foraging and nesting resources (e.g., canopy cover and stand structural complexity).Heterogeneity of structural elements at the stand level has the potential to influence species richness via increased available niches and, thus, may diversify the suites of interactions that co-occur.We additionally employ a metanetwork approach to identify the most important interactions in promoting the stability of host-parasitoid networks, which we expect to involve host general parasitoids because these species have greater niche breadth.We further expect that parasitoid biodiversity metrics, community composition and host-parasitoid interactions will be significantly influenced by host abundance.The results of these analyses will present what is to our knowledge, the first metanetwork analysis of parasitoid bees and wasps, as well as the first simultaneous use of bipartite and metanetworks.

Study region and plot selection
The present study was conducted on 134 1-ha plots, established in 2016 by the 'Conservation of Forest Biodiversity' (ConFoBi) project in the southern Black Forest mountain range (Baden-Württemberg, Germany) (Storch et al., 2020).The Black Forest is mixed-deciduous, consisting of mainly planted Norway spruce (Picea abies L.), European beech (Fagus sylvatica L.), silver fir (Abies alba Mill.), sycamore maple (Acer pseudoplatanus L.) and sessile oak (Quercus petraea Matt.).The transition from timber-focused to close-to-nature forest management has focused on enhancing structure through deadwood and habitat tree retention (Storch et al., 2018;Storch et al., 2020), and reflecting the potentially natural beech-dominated vegetation of the area (Gärtner & Reif, 2005;Standovár & Kenderes, 2003).Initial plot selection focused on deadwood amounts and forest cover at the landscape as the two primary gradients.The high number of plots, large spatial extent of the study area and heterogeneity in the Black Forest have resulted in gradients of numerous environmental variables.For more detailed information on the ConFoBi plot selection and the Black Forest as a study system, as well as additional environmental variables measured, see Storch et al. (2020).For a map of the study area, please see Figure S1.

Forest structural elements and environmental variables
Variables characterising the environment were chosen based on their potential influence on feeding and nesting resources of cavity-nesting bees and wasps, following Rappa et al. (2023).Deciduous tree share (proportion of deciduous tree species), elevation and diameter at breast height of standing and lying deadwood pieces above 7 cm in diameter were obtained during full plot-level inventories conducted in 2017 and 2018.Typically, five stages are applied to classify deadwood according to decay: recently dead or raw wood (I), solid deadwood (II), rotten wood (III), mould wood (IV) and duff wood (V) (Hunter, 1990).Cavitynesting bees and wasps, the host species for parasitoids in our study, prefer fresh and/or moderately decomposed deadwood (Bogusch & Horák, 2018;Eckerter et al., 2021;Westerfelt et al., 2015), as nest building requires stable substrates.To account for this, the cumulative diameter of lying and standing deadwood at plot level of only decay stages I-III was used, excluding decay stages IV and V, in which substrates become soft and unsuitable.Herb cover and understorey species richness were measured from six 5 m Â 5 m subplots in 2017 (Helbach et al., 2022).Forest cover (proportion of forested area in 1 km 2 around plot centres) was calculated using the aerial image data by Storch et al. (2020).The remotely sensed indices stand structural complexity index (SSCI) and effective number of layers (ENL) were derived from terrestrial laser scans at northwest and southeast corners, as well as plot centres (Ehbrecht et al., 2017;Frey et al., 2019;Knuff et al., 2020;Rappa et al., 2022).The SSCI is a measure of geometric complexity of vegetation and structures within a forest stand (Ehbrecht et al., 2017;Stiers et al., 2018).The ENL is an index for measuring the vertical heterogeneity of vegetation layering using voxels in 3D space (Ehbrecht et al., 2016;Ehbrecht et al., 2019).Mean values for each index were calculated using three values taken along northwestsoutheast transects to generate one value per plot.Mean canopy cover was measured in ImageJ using overhead hemispherical photos taken at each trap location in early Fall 2020 (Rappa et al., 2023).Summary information of the environmental variables is available in Table 1.Further explanation of the remotely sensed indices can be found in Supporting Information.

Insect collection, identification and categorisation
Insects were collected during the spring/summer of 2020 using trap nests, which present hollow cavities within reed (Phragmites australis Cav.) internodes packed into polyvinyl chloride tubes (Krombein, 1967;Staab et al., 2018).Traps were secured in pairs to $2 m high wooden poles, which were placed between the plot centre points and the northwest (NW) and southeast (SE) corners, totalling four traps per plot, each with cavities facing in the NW and SE directions.Traps were deployed in early March-April and collected mid-late October to sample the full breadth of phenologies.When occupied with nests, internodes can be easily opened, allowing for the quantification of cells with provisioned resources (abundance) and parasitised cells.From the number of provisioned and parasitised cells, the parasitism rate or the proportion of provisioned cells that were parasitised can be calculated.This sampling method allows highly detailed host-parasitoid interaction data to be gathered (Krombein, 1967) as the interactions can be directly inferred.After collection, nests were removed from traps and refrigerated at $4 C for 8-24 weeks to simulate winter diapause.While refrigerated, nests were opened to quantify the abundance of hosts (cells provisioned) and parasitoids (parasitised cells).Nests were then exposed to room temperature to facilitate the hatching of individuals, which were subsequently collected for morphological identification, using current taxonomic literature (e.g., Jacobs, 2007 for Crabronidae).Individuals who could not be identified at species level ($8% of host cells, $12% of parasitised cells) were excluded before analyses.Following identification, species were categorised according to habitat specialisation as forest and non-forest specialists (hosts) (Rappa et al., 2023) and according to host specificity as general and specific (parasitoids).More detailed information regarding trap construction, and insect identification and categorisation can be found in Supporting Information.

Comprehensive information, biodiversity metrics and species composition
Plots with missing environmental variables (seven plots) were omitted before all analyses.Environmental variables were assessed for collinearity using Spearman's coefficient (R package 'ggpubr') (Dormann et al., 2013;Kassambara, 2020).Following this procedure, if a pair of variables share a coefficient greater than 0.70, only one should be retained for analyses.In our data, no pair of environmental variables was found to be collinear (Table S1).
Species data (hosts and parasitoids) were pooled per plot, before calculating abundance, Shannon diversity and species richness, yielding one data point per plot for each metric.Sampling completeness was assessed with species accumulation curves using the 'specaccum' function (R package 'vegan') (Oksanen et al., 2022) for hosts and parasitoids, with jackknife1 estimators of expected total species richness of each (Figure S2).In addition, an accumulation curve of host-parasitoid interactions was calculated, with jackknife1 estimation of the expected total number of bipartite interactions estimated (Figure S3).To analyse the influence of environmental variables on overall parasitoid biodiversity metrics, abundance and species richness were each analysed using negative binomial generalised linear models, diversity was analysed using a linear model and parasitism rate was analysed using a binomial generalised linear mixed model (GLMM), with parasitism represented as successes and failures of brood cells due to parasitoids at plot level.An observation-level random effect was included in the GLMM, analysing parasitism rate to account for overdispersion.All environmental variables listed in Table 1 were included in each model, with the addition of log-transformed (log 10 [x + 1]) host abundance at the plot level as a covariate.Host availability is the most limiting resource for parasitoids (Pitcairn et al., 1990;Vogel et al., 2021), and thus host abundance is potentially highly influential for parasitoid biodiversity metrics as well as bipartite and metanetwork interactions.was similar to two-dimensional ordinations (Table S3), and thus, only the first two dimensions are displayed in Figure S5.All environmental variables listed in Table 1 were fitted post hoc to the scores of the first two ordination axes of each ordination using the 'envfit' function with 1000 permutations.

Bipartite networks
Weighted bipartite networks were calculated at plot level, with network properties quantified with several indices using the 'bipartite' package (Dormann et al., 2009), calculated for each plot-level network.The pooled bipartite network was calculated across all sites to examine the diversity of interactions sampled.While numerous indices are available to characterise bipartite networks (Almende et al., 2021;Dormann et al., 2009), weighted connectance, linkage density, link diversity and specialisation (H2 0 ) were chosen to test the relationship between the host-parasitoid network structure and environmental variables.Models of connectance and linkage density tested whether the environmental variables listed in Table 1 potentially influence the weight of interactions between the parasitoids and host species.Models of link diversity and specialisation (H2 0 ) tested whether environmental variables diversified and compartmentalised networks.Indices were calculated for only networks with more than one host-parasitoid interaction (one parasitised cell or one interaction).Preliminary analyses were conducted using values from all 90 networks where meaningful indices could be calculated (four models), and later excluding networks with fewer than 10 parasitised cells, analysing values from the resulting 68 networks (four models).Each index in both sets of analyses was analysed using linear models, including log-transformed host abundance at plot level as a covariate.To test if host-parasitoid networks and thus interaction indices differed significantly from chance (and are thus not random), we calculated Patefield null models using 1000 random model runs each and compared null indices with observed indices (Blüthgen et al., 2006;Dormann et al., 2009).Additional information about host-parasitoid interactions (Table S7) and bipartite network indices can be found in Supporting Information.

Metanetwork
The metanetwork was constructed using a data frame of unique hostparasitoid interaction co-occurrences.Each unique pair of interactions was used as a node and the frequency of its co-occurrence with another interaction as the edge connecting them (Figure 5).At the centre, or core of a metanetwork are the most central nodes (hostparasitoid interactions) that co-occur with the largest number of other interactions.The metanetwork indices interaction degree (a measure of centrality) and interaction closeness (a measure of distal branching) were calculated for each host-parasitoid interaction throughout the metanetwork using the 'igraph' package (Csardi & Nepusz, 2006).were taken for each plot and then compared to bipartite network indices using Spearman's correlation coefficients (Table S12).Additional information regarding metanetwork construction, detailed description of the metanetwork core and the selection of interaction indices can be found in Supporting Information.
Residuals of four models of biodiversity metrics, eight models of bipartite indices and two models of metanetwork interaction indices were tested for spatial autocorrelation using Moran's I calculations, performed using the 'testSpatialAutocorrelation' function (R package 'DHARMa') (Hartig, 2022), respectively.No model residuals exhibited spatial autocorrelation (Table S11).

Biodiversity metrics and species composition
In total, 2220 parasitised brood cells (from a total of 14,957 provisioned by hosts) from 39 species (Table S2) were collected, representing 85% of the expected total parasitoid species richness (Figure S2).
Only two obligate hyperparasitoid individuals were collected and, thus, were excluded from our data which had no influence on our results.Abundance, diversity and species richness of parasitoids were all positively related to the abundance of hosts at plot level (Figure S4), whereas abundance (z = 2.036, p = 0.042) and diversity (t = 2.284, p = 0.024) were additionally positively related to standing structural complexity (Table S4) (Figures 1a and 2).Parasitoid abundance increased with standing deadwood (z = 3.368, p < 0.001) (Figure 1b).Parasitism rate increased with only increasing standing deadwood (z = 2.998, p < 0.001) (Figure 3).These results changed little following exclusion of the most common parasitoid species (Melittobia acasta) from our data, with parasitoid abundance and diversity no longer increasing with stand structural complexity (Table S13).

Bipartite networks
The pooled host-parasitoid interaction network had a Shannon interaction diversity of 3.453 and specialisation (H2 0 ) of 0.567 (Figure 4).

Parasitoid reliance on hosts
The reliance of parasitoids on their hosts results in close and relatively specialised interactions compared with other interaction types such as pollination, which can be characterised as being more general (Fontaine et al., 2009;Soares et al., 2017).Host abundance was the primary determinant of parasitoid biodiversity metrics, particularly parasitoid abundance but also species composition.These results support the more-individuals hypothesis (Srivastava & Lawton, 1998), where an increase in host abundance yields not only an increase in the abundance of parasitoid species but increases in their richness and diversity as well.This has been observed in several studies Pooled quantitative bipartite host-parasitoid network (H2 0 = 0.567) representing all plots where parasitism was observed.Width of upper bars represents brood cells parasitised by each species.Width of lower bars represents total number of parasitised brood cells for each host species.Arrow width represents the number of interactions (parasitised brood cells) between each parasitoid (above) and host (below) species.Numbers correspond to species listed in Table S7.Species names are listed for the strongest (highest numbers of parasitised brood cells) interactions.
examining parasitoid taxa (e.g., Vogel et al., 2021), as host availability is likely an important component of parasitoids' ecological niche.Our results, therefore, add further support to the significance of bottomup influences in trophic interactions (Mehrparvar et al., 2019).The analyses of host biodiversity metrics in Rappa et al. (2023) show that structural components of forest habitat, namely standing deadwood and SSCI, promote greater abundance and species richness of cavitynesting bees and wasps through increased and more diverse foraging and nesting resources.The potential influences of these relationships were observed in our study of parasitoids at higher trophic levels, with greater abundance, diversity and species richness.Thus, the retention of forest structural elements can potentially promote this important ecosystem function via bottom-up focused conservation and potentially enhance top-down influences by increasing resilience through redundancy (Sanders et al., 2018;Thierry et al., 2022).
The density-dependent regulation of common host species potentially fosters greater ecosystem stability by buffering competitive exclusion (Brown, 2022), which would occur if a highly abundant species consumes foraging and nesting resources.Under these circumstances, other species with overlapping foraging and nesting requirements would experience a population decline or a decreased carrying capacity.In our data, the most abundant species Trypoxylon figulus built $28% of all nests collected.Given its highly general habitat tolerance (Jacobs, 2007) and tendency to nest in a wide range of cavity diameters, this species could, if very abundant, occupy potential nest sites before other species could utilise them.This is unless higher host density increases parasitism rate (Wang et al., 2020), thus creating a more stable population dynamic, buffering the exclusion effect.
The increase in parasitism rate with greater standing deadwood observed in our data is possibly mediated by higher stability of host nesting substrates and is, thus, an effect of resource stability.This is consistent with studies examining the philopatric tendencies of solitary bees and wasps (Murray et al., 2009;Polidori et al., 2006) and their propensity to nest close to other individuals of the same species, and in close proximity to where they themselves hatched.Additional analyses of parasitism rate from our data, including the abundance of host nests as a covariate cannot support this however (Table S14).Interestingly, parasitism rate was influenced by only standing deadwood in our additional analyses, indicating that it may not be the density or abundance of host nests increasing parasitism rate, but rather the stable presence of hosts as resources, facilitated by stable deadwood structures.
Parasitoid and host species' compositions shared only canopy cover as a significant environmental variable.The activity of many host species included in our study is strongly reliant on sun exposure (Eckerter et al., 2022;Fye, 2012;Hilmers et al., 2018), and most cavity-nesting species tend to prefer sun-exposed deadwood substrates (Bogusch & Horák, 2018).Interestingly, no relationships were observed between parasitoid biodiversity metrics and canopy cover.
Following the habitat-heterogeneity hypothesis (Cramer & Willig, 2005;MacArthur, 1972); however, the potential influence of forest structure could possibly be mediated by host abundances.The creation of canopy gaps (e.g., by tree felling) can diversify light conditions and, thus, the communities of photophilic bees and wasps in forests.The resulting increase in host abundance (see also Achury et al., 2023) and thus parasitoid species richness could improve functional resilience of parasitism networks to environmental changes (Evans et al., 2016;Gladstone-Gallagher et al., 2019;Laliberté & Tylianakis, 2010;Standish et al., 2014) such as those resulting in unfavourable microclimates (Bernaschini et al., 2021) via redundancy.

Interaction bipartite networks
Our hypotheses regarding the influence of forest habitat structural elements on networks were only partially confirmed.Among models including indices from all possible networks, only standing deadwood and herb cover are revealed as potentially supporting more stable parasitism networks by increasing connectance.This is partially supported by our finding that parasitism rate increased with standing deadwood, though parasitism rate and connectance differ.However, networks at plots with more standing deadwood exhibited lower specialisation, indicating that parasitoid host range potentially increases with host abundance, or when host populations are denser (Arneberg et al., 1998;Stanko et al., 2006).It is interesting that greater herb cover increases connectance in our data, considering that nectar foraging by adult parasitoids (Zemenick et al., 2019) is typically overlooked in favour of their more relevant resources (hosts).In addition, it is possible that the abundance of prey items for predatory cavitynesting wasps (e.g., aphids, spiders) increases with greater herb cover, creating a potentially compounding positive influence on host abundance, and therefore parasitoids (Ziesche & Roth, 2008).
The most common interaction in our data involved the highly host general parasitoid species M. acasta, which was observed on $52% of research plots and accounts for 36% of interactions.The importance of generalist species for stabilising networks through redundancy has been well studied (Fornoff et al., 2019).However, networks dominated by one or only a few generalist parasitoids could be overly simple (Dehling, 2018;Poisot et al., 2012) and potentially more vulnerable to disturbances or stochastic changes.For example, a network with only M. acasta would be highly robust, and top-down density dependant population regulation would occur for (theoretically) all species.Removal of half the population of M. acasta in this case may not collapse the network (Nuwagaba et al., 2017;Vizentin-Bugoni et al., 2019).One potential consequence could be, however, greater vulnerability due to decreased redundancy driven by lower richness, meaning a stochastic effect on this single species could collapse the network.Greater resilience provided by the promotion of generalists may be particularly relevant when networks contain highly general species, which have the potential to parasitise also non-native hosts, as is the case with M. acasta.The potential to parasitise nonnative hosts may provide a potential buffer against species invasions (Magal et al., 2008), which without population regulation mechanisms could otherwise collapse ecosystems (Hensel et al., 2021;Morales et al., 2013;Reaser et al., 2007;Walsh et al., 2016).
When only networks with more than 10 parasitised cells are considered, similar effects from only host abundance and canopy cover were observed compared with analyses from networks including fewer than 10 parasitised cells.These analyses together highlight the potential importance of considering small networks to reveal trends.
The decrease in network connectance with increasing host abundance is possibly due to dilution (Civitello et al., 2015;Okuyama, 2021), especially when considering the observed increase in parasitoid abundance with host abundance.The significance of canopy cover for linkage density and link diversity could potentially indicate that promoting the abundance of only hosts may not be a sufficient measure to foster new interactions within networks.This could then mean that consideration of also forest structure is important for actions meant to promote greater network resilience.The diversification of networks with decreasing canopy cover is a somewhat contrary conclusion to Laliberté and Tylianakis (2010), where networks were homogenised by the removal of forest canopy.Furthermore, it has been found that canopy cover re-establishes communities of cavitynesting bees, wasps and parasitoids (Fornoff et al., 2021), albeit these studies were conducted in subtropical forests.It is important to acknowledge that in the context of our study; only forests were sampled, and, thus, we have not sampled the full gradient of canopy cover or examined variables unassociated with forests, which would no doubt provide additional insights.

Interaction metanetwork
The metanetwork allowed for the visualisation of parasitism over the entirety of our study area, and revealed which interactions are cooccurring more frequently and potentially bridging distal groups of interactions.For example, the interaction between the spider-hunting wasp T. figulus and the gregarious parasitoid M. acasta was the most frequent in terms of parasitised cells in the trap nests, yet did not co-occur with other interactions as frequently as the interaction between T. figulus and the kleptoparasitoid Trichrysis cyanea.This indicates that the latter interaction is potentially more important for connecting distal networks and is a more important feature of parasitism networks in forests.
It was contrary to our expectations that more specific parasitoids comprise more of the metanetwork core than generalists, given that generalists are more frequently encountered due to greater niche breadth (Kassen, 2002;Robinson & Strauss, 2018).Several studies have highlighted the importance of rare or unique parasitoids in stabilising metanetworks (Santos et al., 2020), but the importance of more specific parasitoids in our analyses of metanetwork interactions warrants further study.Moreover, more abundant hosts may support more specialist parasitoids due to the stability of hosts as a resource and sensitivity of specialist parasitoids to fluctuations in host populations (Cagnolo et al., 2009;Holzchuh et al., 2010).
Canopy cover partially confirmed our hypothesis that increasing forest structural heterogeneity would diversify interactions in the metanetwork, manifesting as the distal branches of the metanetwork where unique species are present.The leftmost distal branch, for example, is comprised of interactions, which occurred on very few plots, for example, the leaf-cutter bee Megachile versicolor parasitised by the cuckoo bee Coelioxys inermis.Interestingly, the host species in these interactions are often forest specialists (Rappa et al., 2023), and thus characteristic among forest biota.This result is contrary to our expectation that removal of canopy cover would foster unique interactions with non-forest species (Fricke & Svenning, 2020), following transformation of forest habitat.These findings indicate that discrimination by host habitat specialisation is necessary to assess the value of forest structures for parasitism networks.
Future research could reveal the impact of habitat structural elements on specialist and non-specialist-based interactions, by sampling research plots across habitats and analysing the interaction values from the resulting metanetwork.Furthermore, analyses of interaction indices extracted from the metanetwork can reveal habitat structural elements with important roles in maintaining the complexity of interactions and thus ecological communities (Mougi & Kondoh, 2012;Xing & Fayle, 2021).The metanetwork approach could be taken yet another step further and include multiple interaction types, and interactions involving multiple actors to more precisely assess the importance of habitat structural elements in structuring multi-trophic interaction networks.
For example, using the co-occurrences of both host-parasitoid and plant-pollinator interactions sampled from forests and grasslands, hypothetically revealing the most important habitats for interaction types and more importantly, the interactions which connect them spatially.

CONCLUSIONS
Parasitoid biodiversity in forests is influenced most strongly but not exclusively by the abundance of their hosts.Stand heterogeneity (measured here by canopy cover, and stand structural complexity) may have strong influences on parasitoid abundance, species richness and diversity directly but also via increasing host abundance.Forest structural elements such as deadwood have the potential to enhance ecosystem functions, such as parasitism in the case of our study.This research demonstrates the potential insightfulness of concurrent analyses of bipartite and metanetworks for evaluating interactions and ecosystem functions with important considerations for conservation.Furthermore, the derivation of metanetworks from bipartite interactions necessitates their use in tandem to thoroughly answer research questions.
To test the influence of environmental variables on species composition, NMDS (Non-metric MultiDimensional Scaling) was performed for hosts and parasitoids using the 'metaMDS' function (R package 'vegan') with 1000 random starting draws each.Ordinations were made using 'Bray-Curtis' dissimilarities on three axes to reduce stress while ensuring ordination, and fitting of environmental variables could still be reliably interpreted.Procrustes errors of the first two axes of host and parasitoid ordinations were compared separately to ordinations with two axes using the 'protest' function.The representation of the first two axes in three-dimensional ordinations T A B L E 1 Environmental variables and summary statistics characterising the 127 plots used in analyses, following exclusion of plots deficient of remotely sensed indices (seven plots).
While many indices are available for interactions within the metanetwork (R package 'igraph')(Csardi & Nepusz, 2006), interaction degree and closeness were chosen to measure the centrality and distal branching of interactions(Librán-Embid et al., 2021).A high value of interaction degree indicates a high number of uniquely co-occurring interactions, and thus a more central node forming a greater part of the metanetwork core.Closeness measures how many steps are required to reach another node from a given node.While also used to infer interaction centrality, a high value of closeness does not indicate a greater connection to other interactions but fewer steps necessary to reach all other interactions.Importantly, a low value of closeness indicates an interaction that is far from the metanetwork core.The indices were applied to each interaction in the plot-level dataset, resulting in groups of values of interaction degree and interaction closeness for each plot.To model the similar and repeated interaction degree (one model) and closeness (one model) values, where the number of interactions varied between the plots, GLMMs were calculated including the log-transformed (log[x + 1]) number of interactions as an offset and plot as a random term, as well as log-transformed host abundance at plot level as a covariate.Modelling interaction degree tests the influence of environmental variables on the co-occurrence of interactions, and thus the potential to connect plot-level networks with shared function (e.g., standing deadwood promoting frequently occurring interactions).Modelling interaction closeness determines which environmental variables have the potential to diversify suites of interactions, appearing as unique.To compare the indices calculated for interactions in the metanetwork with indices calculated from bipartite networks, mean values for interaction degree and closeness

1
Abundance of parasitoids of cavity-nesting bees and wasps and significant fixed effects: (a) stand structural complexity index (SSCI) and (b) standing deadwood cumulative diameter (cm).Trend lines from negative binomial generalised linear models are depicted for abundance in both figures, with 95% confidence intervals coloured in grey.F I G U R E 2 Diversity (Shannon) of parasitoids of cavity-nesting bees and wasps and stand structural complexity index (SSCI).Trend line for the linear model of diversity is depicted, with 95% confidence intervals coloured in grey.F I G U R E 3 Parasitism rate of cavity-nesting bees and wasps and standing deadwood cumulative diameter (cm).Trend line for the binomial generalised linear model (GLM) is depicted with 95% confidence intervals coloured in grey.p = 0.016) and link diversity (t = À2.093,p = 0.039) both decreased with increasing canopy cover.

F
I G U R E 5 Host-parasitoid interaction metanetwork representing all 115 plots where parasitism was observed.Nodes in the metanetwork represent host-parasitoid interactions from local bipartite networks while edges represent co-occurrences with other interactions at plot-level.For each node (interaction) parasitoid species are depicted above dashed lines while hosts are depicted below.Edge width corresponds to the number of local networks where both interactions co-occurred.Point size corresponds to degree, or number of co-occurring interactions, with the most common 'core' interactions represented with images.Point shape corresponds to the habitat specialisation of hosts while point colour corresponds to specificity of host utilisation by parasitoids.The host-parasitoid interactions depicted are (from top to bottom): Trypoxylon figulus-Trichrysis cyanea, T. figulus-Melittobia acasta, Deuteragenia subintermedia-M.acasta, Anicistrocerus trifasciatus-C.solida, T. figulus-Nematopodius debilis, Symmorphus gracilis-C.corusca, Passaloecus insignis-Omalus aeneus, P. insignis-O.puncticollis and T. clavicerum-N.debilis.Additional descriptions of the interactions pictured, and the metanetwork core can be found in Supporting Information.
None of the variables were excluded on the basis of assumed collinearity (ρ > 0.70) following pairwise analyses.A summary of Spearman's correlations among environmental variables can be found in TableS1.