Introduction
Forest fragmentation and climate change, alone or in combination, are among the greatest threats to biodiversity persistence and ecosystem functioning (Thomas et al., Reference Thomas, Cameron, Green, Bakkenes, Beaumont, Collingham, Erasmus, de Siqueira, Grainger, Hannah, Hughes, Huntley, van Jaarsveld, Midgley, Miles, Ortega-Huerta, Townsend Peterson, Phillips and Williams2004; Morris, Reference Morris2010; Traill et al., Reference Traill, Lim, Sodhi and Bradshaw2010). It is well known that tropical forest fragmentation, via habitat loss and isolation (habitat fragmentation per se), and edge effects (Fahrig, Reference Fahrig2003), is likely to cause population collapse, species extirpation and reorganization of native communities towards simplified and more similar communities in taxonomic and functional terms (Santos et al., Reference Santos, Peres, Oliveira, Grillo, Alves-Costa and Tabarelli2008; Wirth et al., Reference Wirth, Meyer, Leal, Tabarelli, Lüttge, Beyschlag and Murata2008). Relatively little is known, however, about fragmentation effects on temperate forest ecosystems (Honnay et al., Reference Honnay, Verheyen and Hermy2002), most probably because of the traditional focus on forest management, although deforestation is often extensive (e.g., 31% in Germany, MUF, 2002) and many centuries old (Frey & Lösch, Reference Frey and Lösch2010). Our understanding how climate change impacts on the reorganization of tropical or temperate native communities is even more rudimentary, despite increasing attention given to plant–animal interactions in this context (Tylianakis et al., Reference Tylianakis, Didham, Bascompte and Wardle2008; Morris, Reference Morris2010). Generally, there is a wide consensus that global warming has direct impacts on the temporal and spatial dynamics of insect herbivores (Netherer & Schopf, Reference Netherer and Schopf2010). Moreover, in Europe, climate change is expected to profoundly affect the composition of their food plant communities (Pompe et al., Reference Pompe, Hanspach, Badeck, Klotz, Bruelheide and Kühn2010) through distributional shifts, altered competition dynamics (Klanderud, Reference Klanderud2005), such as proliferation of thermophiles (Reid, Reference Reid2006) and local extinction (Pompe et al., Reference Pompe, Hanspach, Badeck, Klotz, Bruelheide and Kühn2010). Forests are particularly threatened (predicted cover reduction of 11.7–19.9%, Pompe et al., Reference Pompe, Hanspach, Badeck, Klotz, Bruelheide and Kühn2010), because of their long life cycle (Lindner et al., Reference Lindner, Maroschek, Netherer, Kremer, Barbati, Garcia-Gonzalo, Seidl, Delzon, Corona, Kolström, Lexer and Marchetti2010). Consequently, re-organization of plant communities may cascade to higher trophic levels and disrupt trophic interactions with insect herbivores, e.g. through phenological and distributional mismatch (Schweiger et al., Reference Schweiger, Settele, Kudrna, Klotz and Kühn2008; Traill et al., Reference Traill, Lim, Sodhi and Bradshaw2010; Kaneryd et al., Reference Kaneryd, Borrvall, Berg, Curtsdotter, Eklöf, Hauzy, Jonsson, Münger, Setzer, Säterberg and Ebenman2012).
Trophic webs represent a key component of ecosystems, with substantial importance for ecosystem-level properties such as functioning (Morris, Reference Morris2010; Rzanny & Voigt, Reference Rzanny and Voigt2012), service provision (Memmott et al., Reference Memmott, Waser and Price2004), biodiversity persistence (Morris, Reference Morris2010) and stability (Emmerson et al., Reference Emmerson, Bezemer, Hunter and Jones2005; Rzanny & Voigt, Reference Rzanny and Voigt2012). Amongst them, plant–insect herbivore interaction networks (PIHNs) harbour the bulk of terrestrial biodiversity (Price, Reference Price2002) and play a significant functional role, e.g. for the distribution of energy and biomass to other trophic levels (Haddad et al., Reference Haddad, Crutsinger, Gross, Haarstad and Tilman2011; Rzanny & Voigt, Reference Rzanny and Voigt2012). By integrating over multiple taxa across trophic levels, PIHNs go far beyond what could be addressed in previous studies on plant–insect herbivore interactions. Therefore, and because they are closely linked to the global loss of species and ecosystem functioning (Morris, Reference Morris2010; Rzanny & Voigt, Reference Rzanny and Voigt2012), ecological networks have received increasing awareness as indicators of trophic disruptions (Heleno et al., Reference Heleno, Devoto and Pocock2012; Rzanny & Voigt, Reference Rzanny and Voigt2012) and progressive simplification of ecosystems in the face of anthropogenic disturbance (Tylianakis et al., Reference Tylianakis, Tscharntke and Lewis2007; Laliberté & Tylianakis, Reference Laliberté and Tylianakis2010; Valladares et al., Reference Valladares, Cagnolo and Salvo2012).
In the case of temperate forests, more research on fragmentation effects on PIHNs is needed, as the scarce evidence suggests contrasting patterns to the far better researched tropics. For example, fragmented forests in temperate regions harbour higher plant diversity than continuous forests (Honnay et al., Reference Honnay, Verheyen and Hermy2002). This may lead to greater herbivore diversity (van Halder et al., Reference van Halder, Barbaro and Jactel2011) and more potential feeding links for polyphagous insect herbivores in forest edges, thus leading to increased complexity of trophic interactions and hence food web stability (Haddad et al., Reference Haddad, Crutsinger, Gross, Haarstad and Tilman2011). On the other hand, PIHNs in managed continuous forests can be expected to be composed of generalist insect herbivores with narrow realized niches, as silviculturally important tree species (e.g., beech, Fagus sylvatica) with a low natural set of specialists (Sprick & Floren, Reference Sprick, Floren, Floren and Schmidl2008) are predominant and therefore offer little resource diversity. In relation to climate change, some inferences may be drawn from previous research on single trophic levels and their interactions, such as the intensifications of particular trophic links via higher plant productivity (Emmerson et al., Reference Emmerson, Bezemer, Hunter and Jones2005; Traill et al., Reference Traill, Lim, Sodhi and Bradshaw2010) or temperature-driven increase in the abundance/performance of insect herbivores (reviewed in Bale et al., Reference Bale, Masters, Hodkinson, Awmack, Bezemer, Brown, Butterfield, Buse, Coulson, Farrar, Good, Harrington, Hartley, Jones, Lindroth, Press, Symrnioudis, Watt and Whittaker2002). Differential responses of generalist and specialist insects to climate change may alter competition dynamics towards more uneven and simplified PIHNs. Bottom-up extinction cascades are likely because of proportionally higher extinction probabilities for primary producers, compared with other trophic levels (Kaneryd et al., Reference Kaneryd, Borrvall, Berg, Curtsdotter, Eklöf, Hauzy, Jonsson, Münger, Setzer, Säterberg and Ebenman2012), and increasing variability of overall population densities and temperatures (Emmerson et al., Reference Emmerson, Bezemer, Hunter and Jones2005).
The present study investigates for the first time the conjoined impacts of temperate forest fragmentation and climate change on the structure of PIHNs and relates network topology to network robustness under realistic extinction sequences. For this we adopted an extensive sampling regime of insect herbivores on more than 1300 individual trees across 36 plots in forest edges, small fragments and the interior of continuous control forests in a highly fragmented and managed temperate forest landscape in SW Germany. Following previous evidence indicating a fragmentation-induced rise in species numbers of both plant and insect herbivores (Honnay et al., Reference Honnay, Verheyen and Hermy2002; van Halder et al., Reference Weiner, Werner, Linsenmair and Blüthgen2011), we expected that PIHNs in edges and small fragments have: (i) higher complexity, cohesiveness and trophic redundancy as well as (ii) increased robustness against secondary extinctions in climate change-based extinction scenarios, as opposed to networks from the forest interior. We assume this decline in climate change susceptibility to be caused by a combination of higher thermotolerance across the plant communities (Reid, Reference Reid2006) and higher network complexity, cohesiveness and redundancy, which can be demonstrated by linking network topology to PIHN robustness. Furthermore, this paper aims at discussing management implications arising from the potential anthropogenic changes in PIHN integrity.
Methods
The study landscape is located in the Northern Palatinate highlands, a low, undulating mountain range (250–687 m asl) of Permian origin covering an area of 1556.4 km2 in SW Germany (fig. 1). It is characterized by temperate Central European climate under oceanic influence (mean annual precipitation: 800 mm; mean annual temperature: 9.4°C, 1970–2010, Deutscher Wetterdienst, 2013). Forests are deciduous, broad-leaved woodlands, phytosociologically classified as Carpino–Fagetalia mixed forests with varying transitional degrees of Fagion and Carpinion betuli stands. Extensive deforestation in the Middle Ages occurred mainly in sand and siltstone-dominated valleys, while the agriculturally less valuable igneous hilltops were mostly forested. This has led to a landscape of hyper-fragmented forests, embedded in a matrix of cultivated fields, pastures and meadow orchards. Forest cover of the selected portion (32%) of this landscape (1010 km2, 49°36′N and 7°44′E) is representative for rest of Germany (31%, MUF, 2002). Despite its high fragmentation degree with over 1300 forest fragments ranging from 0.1 to 5616 ha (ca. 80% of them <10 ha) the region still harbours large forest tracts exceeding 1000 ha.
Fig. 1. Maps showing the location of the study landscape, the Northern Palatinate highlands, with respect to Central Europe (a) and SW Germany (b), where it is indicated as white rectangle in the state of Rhineland-Palatinate (Rh.-Pal.). The study landscape (c) shows forest fragments (grey polygons) embedded in a matrix of agricultural land uses (white) and 36 randomly established sampling sites in the centre of small forest fragments (<1000 ha, squares), along forest edges (diamonds) and in forest interiors within continuous control forests (>1000 ha, circles).
Study sites were established in 2008 as 36 permanent plots (20 × 50 m2; 0.1 ha) in three habitat types reflecting fragmentation-induced and continuous forest conditions: (i) small fragments: centre of 12 small forest fragments (ranging between 2.2 and 145.1 ha) entirely surrounded by matrix, reflecting habitat loss, isolation and fragmentation effects per se; (ii) forest edges: peripheral areas within 50 m of the border of large forest tracts (continuous control forests, the three largest forest tracts in the study region, 1049, 3512 and 5616 ha) to address edge effects; (iii) forest interior: core areas of control forests beyond 100 m of the border and without detectable edge influence. Inter-plot distance ranged from 0.3 to 35.4 km with 13.9 ± 7.6 km (mean ± SD). All woody plant species >1.3 m height were identified in these plots.
Insect herbivores were sampled alive in the understory stratum (1.3–2 m) of the plots during two field campaigns (May–August 2010 and August–September 2010) by beating the foliage of each woody plant individual regardless of age class over a funnel-shaped trap (see fig. S1). Sampling intensity totalled 207 h of beating and 1353 woody plant individuals. Separation of transient insects (tourists) from true herbivores was achieved by using no-choice feeding-assays and analysing published host plant records of identified species (see more detail on feeding assays in Supplementary Material text S1.1). Insects were assigned to morphospecies and later identified to species level whenever possible (56%) using standard keys or experts (table S1). Caterpillars were reared to adults, if possible, and then identified. Voucher specimens were deposited at the Natural Museum of Basel (Psylloidea) and at the department of Plant Ecology and Systematics at the University of Kaiserslautern (all other taxa).
For network-related definitions and terminology we followed Dormann et al. (Reference Dormann, Fründ, Blüthgen and Gruber2009). To quantify topological network properties related to network robustness (complexity, cohesiveness and trophic niche redundancy) we used quantitative indices, i.e. Shannon diversity of interactions (H2, Rzanny & Voigt, Reference Rzanny and Voigt2012), nestedness (Bascompte et al., Reference Bascompte, Jordano, Melián and Olesen2003) and the complementary specialization at network level (H2′, Blüthgen et al., Reference Blüthgen, Fründ, Vázquez and Menzel2008) (see more detail on the use of this metrics in Supplementary Material text S1.2). To maximize standardization and comparability of our data, we adopted three approaches proposed by Blüthgen (Reference Blüthgen2010): (i) Sampling of equal areas (12,000 m2 in each of the two sampling campaigns) in each habitat, (ii) rarefaction (Blüthgen et al., Reference Blüthgen, Menzel and Blüthgen2006) of PIHNs to the size of the smallest network (forest interior, m = 217) with 100 rarefaction cycles per habitat (figs S2, S3) and (iii) comparison of PIHNs with null models (Patefield algorithm, Blüthgen et al., Reference Blüthgen, Fründ, Vázquez and Menzel2008, 10 per step in each rarefaction cycle). Habitat-wise comparisons of network metrics were made between the large networks (comprising all interactions minus one randomly deleted interaction in the rarefaction procedure, hereafter ‘full networks’) and between the small networks (rarefied to the size of m = 216 interactions, hereafter ‘rarefied networks’) (Kruskal–Wallis one-way analysis of variance with a Nemenyi–Damico–Wolfe–Dunn post hoc test). Empirical network metrics (n = 100) and their counterparts received from null models (n = 1000) were statistically compared using Mann–Whitney/Wilcoxon tests.
We simulated co-extinction scenarios to explore the effects of climate change on PIHNs in edge/fragmentation-affected vs. continuous forest habitats by consecutively deleting individual plant species from the networks according to sequences obtained from four proxies for susceptibility to climate change. For each proxy and forest habitat the relative proportion of consecutive primary extinctions (plant level) was plotted against the relative proportion of remaining herbivore species (i.e., herbivores that did not lose all trophic links), resulting in extinction curves. Robustness of interaction networks towards primary extinction was measured by calculating the integral under these curves (Burgos et al., Reference Burgos, Ceva, Perazzo, Devoto, Medan, Zimmermann and Delbue2007; Menke et al., Reference Menke, Böhning-Gaese and Schleuning2012). The first two proxies for susceptibility to climate change (and hence extinction sequences) were based on Ellenberg's indicators values for plant requirements for temperature and moisture (Ellenberg & Leuschner, Reference Ellenberg and Leuschner1996; Pompe et al., Reference Pompe, Berger, Bergmann, Badeck, Lübbert, Klotz, Rehse, Söhlke, Sattler, Walther and Kühn2011). Plants classified as indifferent were removed last; extinction order among plant species with identical Ellenberg values was randomized. As a third proxy for extinction sequences we used modified risk classes (R1–R5) for plant species based on predicted net changes in species range size from the GRAS (GRowth Applied Strategy, Pompe et al., Reference Pompe, Hanspach, Badeck, Klotz, Bruelheide and Kühn2010) climate change scenario (Pompe et al., Reference Pompe, Berger, Bergmann, Badeck, Lübbert, Klotz, Rehse, Söhlke, Sattler, Walther and Kühn2011): R1 < 0% (range increase), 0% ≤ R2 < 25%, 25% ≤ R3 < 50%, 50% ≤ R4 < 75% and R5 ≥ 75% range loss. The GRAS scenario is based on an average rise in temperature from 1961–1990 (reference time period) to 2051–2080 (future time period) of 3.8°C (Pompe et al., Reference Pompe, Hanspach, Badeck, Klotz, Bruelheide and Kühn2010). Unclassified plants contributed 7% (fragments), 16% (edges) and 13% (forest interior) of interactions to the networks and were set to die out randomly. A more conservative analysis, in which those plants were left out completely, did not change results qualitatively. A fourth proxy used continuous percentage data of range gain or loss from the GRAS scenario (Pompe et al., Reference Pompe, Berger, Bergmann, Badeck, Lübbert, Klotz, Rehse, Söhlke, Sattler, Walther and Kühn2011). We calculated extinction curves and their respective integrals 100 times for each habitat and susceptibility proxy. Habitat-wise comparisons of the integrals were examined using Kruskal–Wallis one-way analysis of variance with a Nemenyi–Damico–Wolfe–Dunn post hoc test. Integrals for each proxy and habitat were compared with null-models (random extinction sequence, 100 cycles, Kaiser-Bunbury et al., Reference Kaiser-Bunbury, Muff, Memmott, Müller and Caflisch2010) using Mann–Whitney/Wilcoxon tests.
To evaluate the predictive capacity of network topology regarding the robustness of PIHNs against climate change, we related network metrics to integrals of extinction curves (climate change-based and random). For this we split our data: Analysis of network topology and extinction robustness for the three habitats was repeated twice during the vegetation period (early to midsummer, campaign 1 and mid-to-late summer, campaign 2), thereby generating six networks. As these season-specific networks were not independent from each other, we employed linear mixed models with network robustness as response variable, network topology as fixed factor and season as random intercept (package nlme, version 3.1–122, Pinheiro et al., Reference Pinheiro, Bates, DebRoy and Sarkar2017). (Marginal) R 2 values for the fixed effects were calculated following Nakagawa & Schielzeth (Reference Nakagawa and Schielzeth2014).
Analyses were carried out with the statistical computing software R version 3.0.2 (R Core Team, 2013). All network representations and network analyses were performed with the package ‘bipartite’ version 2.3–1 (Dormann et al., Reference Dormann, Gruber and Fründ2008).
Results
We observed a total of 696 interactions, involving 24 woody plant species (seven families) and 134 insect herbivore taxa (five orders and 28 families). Plant species richness declined almost threefold (Kruskal–Wallis test; χ2 = 15.25, df = 2, P < 0.001) from forest edges (5.8 ± 2.0, mean ± SD) over forest fragments (3.8 ± 2.7) to interior control forests (2.1 ± 1.6). While the tree communities showed high overall dominance of F. sylvatica (44.27%) and Carpinus betulus (29.86%) across all habitats, the species abundances were more evenly distributed in edges and fragments (table S2, fig. S4). Insect herbivore assemblages along forest edges (11.0 ± 3.5 taxa) were ca. 90 and 8% more species rich than in forest interiors (5.8 ± 2.9) and forest fragments (10.2 ± 4.2), respectively (one-way ANOVA; F = 7.68, df = 2, P < 0.01). The proportional species richness of specialized herbivores was four times higher in edge (32% of species), compared with interior habitats (8%, Fisher's exact test, P < 0.001). Within the study region, generalist herbivores were more abundant than specialists, the three most abundant species being Chelidurella guentheri (Forficulidae), Issus coleoptratus (Issidae) and Polydrusus marginatus (Curculionidae) (see more details on insect species in table S1).
Forest fragmentation profoundly affected PIHNs, which can be inferred at different levels of analytical scrutiny, from visual inspection, basic network attributes and topological network metrics associated with robustness. Interaction networks declined in species richness, link abundance (L), and complexity at both trophic levels with declining influence of edge effects (fig. 2), while the highest amount of individual interactions (m) amongst forest habitats was found in forest fragments (m = 318). Interaction evenness of plant species (evenness of lower black bar width, fig. 2) was similar in fragmentation-related habitats (0.78 in fragments and 0.77 in forest edges) and lowest in the forest interior (0.63), with increasing dominance of F. sylvatica and C. betulus (accounting for 53.8, 51.4 and 82.1% of the interactions in fragments, edges and interior forests, respectively). While topological network metrics showed clear positive responses to forest fragmentation (increased complexity, nestedness and redundancy), the adopted standardization procedures underline their ecological validity, as the metrics of full and rarefied PIHNs were similarly affected (table 1) and empirical values always significantly differed from null model values (Mann–Whitney/Wilcoxon tests, all P < 0.001). For the full networks PIHNs declined in complexity, nestedness and niche redundancy from forest edges over fragments to the forest interior (table 1). Network complexity, as measured by Shannon diversity of interactions, significantly decreased by 24% from forest edges to the interior, suggesting a drop in link abundance and evenness. Generally high nestedness across all habitats, with values never exceeding 18 for any network, implies the existence of hub-sections (i.e., cores of many strong links) in PIHNs. Nevertheless, nestedness significantly declined by more than half from forest edges to the forest interior (table 1). Trophic niche complementarity (H2′) was 26% lower in forest edges than in forest interior habitats, indicating that PIHNs along forest edges had the highest overall degree of overlap of realized niches. In a nutshell, forest fragmentation and particularly edge creation significantly increased the complexity, cohesiveness and trophic niche redundancy of interaction networks between trees and insect herbivores.
Fig. 2. Quantitative bipartite graphs of PIHNs for temperate forest edges, forest fragments and interior of continuous control forests in the Northern Palatinate highlands, SW Germany. Upper (lower) bars depict individual insect (plant) species according to their relative interaction strength (bar width). Grey bars represent links between species respective to their relative link weight. L = sum of trophic links; m = sum of interactions within the network. Species abbreviations indicate the three most interacting plant- and insect herbivore species in each network. Plants: C. b. (Carpinus betulus), F. s. (Fagus sylvatica), Q. r. (Quercus robur), and Q. p. (Quercus petraea). Insect herbivores: A. m. (Apterygida media), C. g. (Chelidurella guentheri), C. s.5 (Cicadellidae spec. 5), I. c. (Issus coleoptratus), and P. m. (Polydrusus marginatus).
Table 1. Effects of forest fragmentation on topological network-metrics of full and rarefied PIHNs.
Full networks consist of all interactions of the respective habitat minus one randomly deleted interaction, while rarefied networks were rarefied by random deletion of interactions to the size of the smallest PIHN minus one interaction (forest interior, m = 161) for standardization purposes. H2: Shannon diversity of interactions; Nestedness: ranges from 0 (perfectly nested) to 100 (maximal entropy); H2′: Trophic niche complementarity on network level, ranges from 0 (maximal redundancy) to 1 (maximal complementarity).
Network robustness, as depicted by the area under extinction curves (fig. 3), was positively affected by forest fragmentation, revealed similar patterns across all extinction scenarios (Kruskal–Wallis tests, table S3), and consistently differed from null models (random extinctions, Man–Whitney/Wilcoxon tests, table S3). Forest interior networks experienced a more pronounced collapse (figs 3 and 4), while highest robustness was found either in forest fragments (Ellenberg's temperature) or edges, depending on the underlying proxy reflecting the extinction sequence (fig. 4). To give an example, removing 50% of woody plant species using Ellenberg's temperature values as a proxy resulted in 78 and 70% of herbivore species remaining in forest fragments and edges, respectively, while PIHNs in the forest interior suffered nearly two-thirds of co-extinction (36% remaining insect herbivore species).
Fig. 3. Effects of forest fragmentation on climate change-driven extinction scenarios of PIHNs. Extinction curves depict secondary extinctions of herbivores upon loss of their host plants (primary extinction) for three forest habitats: forest edges (squares, dotted lines), forest fragments (diamonds, continuous lines), and interior of continuous control forests (circles, dashed lines). Sequence of plant extinction followed predicted sensitivity to climate change, based on four proxies (Ellenberg's temperature, Ellenberg's moisture, risk-groups and relative range change). Mean and standard deviation (error-bars) are obtained from 100 iterations of extinction sequences. SDs equalling zero stem from constant co-extinction proportions at particular steps, regardless of the iteration.
Fig. 4. Effects of forest fragmentation on robustness of PIHNs against simulated extinction cascades under climate change. Robustness (integrals of extinction curves in fig. 2, ECI) is depicted for forest fragments, edges and interior of continuous control forests (grey boxes) and corresponding null models (random extinction, white boxes) for four models of extinction sequences (Ellenberg's temperature, Ellenberg's moisture, risk groups and relative range change). Box-plots show the median (line), interquartile range (box) and range (whiskers). Habitat-wise comparisons yielded significant differences for all models (Kruskal–Wallis, results in table S2) indicated by different letters (Nemenyi–Damico–Wolfe–Dunn post-hoc test). Network robustness in climate change-based models significantly differed from null models in most cases, as denoted by asterisks (*P < 0.05; ***P < 0.001; Mann–Whitney/Wilcoxon, results in table S2.
While topological network metrics were entirely unrelated to network robustness against random extinctions (table 2), they revealed several strong relationships and trends with ecologically realistic extinction scenarios. Among the tested metrics, the most important factor was complementary specialization on network level (H2′), which showed significant negative relations with Ellenberg's moisture (marginal R 2 = 0.71), risk groups (marginal R 2 = 0.78) and predicted range change (marginal R 2 = 0.79). In addition, there was a trend between the Shannon diversity of interactions (H2) and Ellenberg's temperature (marginal R 2 = 0.56), whereas network cohesiveness (as measured via nestedness) showed no relation to any extinction scenarios. Thus, the higher the degree of complexity and realized niche redundancy in the network, the more stable it performs against climate change-based extinction scenarios.
Table 2. Relationships between network robustness against random and climate change-driven extinction sequences (response variable) and topological network metrics (fixed effects), modulated by seasonality (random intercept) using linear mixed models (all n = 6).
H2, Shannon diversity of interactions; H2′, Trophic niche complementarity on network level. Est, Estimated slope value of the network metric. R 2, R 2 of the fixed effect; P, P-value of the fixed effect. Bold letters indicate statistically significant values (P < 0.05); statistical trends (0.05 < P < 0.1) are denoted by a cross †.
Discussion
To our knowledge, this is the first study to address the impacts of temperate forest fragmentation on PIHNs, while accounting for their susceptibility to climate change. Our results indicate that fragmentation, particularly edge proliferation, positively affects the complexity and climate change-related robustness of PIHNs, thereby reducing their proneness to simulated climate-driven extinction. These findings provide insight into how network-inherent attributes drive the susceptibility of trophic web members to climate change in fragmented forest landscapes and allow the inference of silvicultural and insect conservation implications.
Contrary to numerous published reports that describe disruption and simplification of ecological networks by human disturbance across many ecosystems and trophic levels (Fortuna & Bascompte, Reference Fortuna and Bascompte2006; Tylianakis et al., Reference Tylianakis, Tscharntke and Lewis2007; Weiner et al., Reference Weiner, Werner, Linsenmair and Blüthgen2011) including PIHNs (Valladares et al., Reference Valladares, Cagnolo and Salvo2012), our findings show positive effects of forest fragmentation (notably edge effects) on robustness-related network parameters of PIHNs. This may result from two sometimes overlooked features of temperate forest in human-dominated landscapes. First, the flora of forest edges is relatively more species-rich than the forest interior (Gehlhausen et al., Reference Gehlhausen, Schwartz and Augspurger2000) due to a generally high proportion of light-adapted, thermotolerant species that received century-long anthropogenic facilitation in the cultural landscape (Hermy et al., Reference Hermy, Honnay, Firbank, Grashof-Bokdam and Lawesson1999). Second, greater tree diversity along edge zones and small fragments may also be caused by their partly release from intensive silvicultural regimes, for practical and forest edge management reasons (Coch, Reference Coch1995). As a plausible consequence, this may have enhanced the species richness of insect herbivores, link abundance and average width of realized trophic niches via a wider range of food sources to polyphagous herbivores. This in turn increased the complexity of interactions and niche redundancy (as broader niches were more likely to overlap) – network features that have previously been reported to be affected by different components of forest fragmentation (Menke et al., Reference Menke, Böhning-Gaese and Schleuning2012; Valladares et al., Reference Valladares, Cagnolo and Salvo2012; Albrecht et al., Reference Albrecht, Berens, Blüthgen, Jaroszewicz, Selva and Farwig2013). The relevance of food source variability is further highlighted by the fact that edge communities were more strongly coined by specialized herbivores (monophagous and oligophagous species, e.g. aphids and psyllids) than those of the forest interior, where generalists constituted at least 65% of all interactions. Such prevalence of generalist insects in the interior did not translate into higher PIHN complexity and redundancy because managed forests are typically dominated by a few silviculturally desirable tree species (e.g., the dominance of F. sylvatica and C. betulus reached up to 100% in individual plots; fig. S4). The concomitant restriction of potential feeding links forced generalist herbivores to establish narrow realized niches, resulting in low redundancy on network level. Taken together with the findings from previous studies indicating both beneficial and adverse consequences for the structure of plant–animal networks (Menke et al., Reference Menke, Böhning-Gaese and Schleuning2012; Valladares et al., Reference Valladares, Cagnolo and Salvo2012), our study suggests that the response of ecological networks to forest fragmentation is highly dependent on network type and the particularities of the studied biome.
Our results provide insights into the interaction of two major human impacts on food-webs by showing that PIHNs in forest edges and, to a lesser extent, in small forest fragments were relatively more robust to climate change-based extinction scenarios (figs 3 and 4). This occurred because the above-mentioned higher proportion of light-adapted, thermotolerant species (e.g., common hawthorn, Crataegus monogyna, table S1) that were more likely to die out later in the extinction sequence than shade-tolerant ones because the corresponding links were conserved longer in the network, making fragmented habitats more robust against climate change-induced coextinctions. Such positive effects of land use on the redundancy of functional traits have been previously documented for specific combinations of ecosystem type and disturbance regimes (Laliberté et al., Reference Laliberté, Wells, DeClerck, Metcalfe, Catterall, Queiroz, Aubin, Bonser, Ding, Fraterrigo, McNamara, Morgan, Merlos, Vesk and Mayfield2010). Further, our results corroborate earlier findings that extinction risk can be influenced by network structure (Melian & Bascompte, Reference Melian and Bascompte2002), as network redundancy was lowest in the forest interior and highest in forest edges, indicating higher niche overlap in edge-influenced forests. Therefore, a higher proportion of insect herbivores persisted longer during extinction series due to the delayed loss of all available host plants. On the other hand, the severe co-extinction risk for insects in managed forest interiors has potential relevance for insect conservation. It implies that the intentional promotion of silviculturally important tree species may unintentionally threaten the diversity of insects including their ecosystem services. Network architectures and how they modulate responses to local and global disturbances should therefore be considered for the future enhancement of silvicultural and conservation strategies. While conventional approaches prioritize species richness at single trophic levels or even single charismatic flagship species (Segura et al., Reference Segura, Castaño-Santamaría, Laiolo and Obeso2014), network studies allow us to identify and specifically target critical key parts of food webs (Traill et al., Reference Traill, Lim, Sodhi and Bradshaw2010), for example those insects and plant species that can be relevant for ecosystem robustness and functioning.
Another aspect of high relevance for conservation practices is the identification of hub species (Olesen et al., Reference Olesen, Bascompte, Dupont and Jordano2007), like European beech (F. sylvatica) in our PIHNs. This suggests that F. sylvatica should receive priority attention with regard to susceptibility against climate change, as all extinction models were largely dependent on its high extinction scores, as well as its high dominance and association with many exclusive herbivores (e.g., 35% of all herbivore species in the forest interior). In addition, such high scores are justified by a projected distributional decline of beech at its southern border following climatic warming (Kramer et al., Reference Kramer, Degen, Buschbom, Hickler, Thuiller, Sykes and de Winter2010). This is consistent with strong climate change-driven shifts in plant (Pompe et al., Reference Pompe, Hanspach, Badeck, Klotz, Bruelheide and Kühn2010) and insect species (Netherer & Schopf, Reference Netherer and Schopf2010) distributions predicted for Europe. Different, but nonetheless far-reaching conservation implications arise from the apparent impression that fragmented, edge-dominated forest habitats seemed pre-adapted to climate change and (provocatively suggested) should therefore be promoted as reservoirs for insect diversity and PIHN integrity. Instead, as edge richness is largely due to the release from sylvicultural management, conservation measures should aim at strengthening PIHNs in the continuous forest interior at the local and the landscape scale (e.g., via close-to-nature practices, such as single-tree harvesting, site-specific tree species matching, or mimicking natural gap dynamics as suggested by Brunet et al., Reference Brunet, Fritz and Richnau2010). While management issues were beyond the scope of this study, future research efforts should more closely examine whether such diversity-oriented silviculture translates into higher PIHN complexity and ecosystem resilience via increasing functional redundancy. This is, in fact, a plausible scenario, as we saw higher levels of redundancy in more diversified edge habitats, although these habitats were more strongly coined by specialized herbivore species, such as aphids and psyllids.
In synthesis, this paper reinforces the notion that trophic interactions are highly sensitive to human impacts and sheds some light on the interrelations between two key disturbance agents in forest ecosystems. Using biologically realistic models of plant species extinction risk, our study provided strong evidence that fragmentation of temperate forests positively affects the structural integrity/complexity of plant-herbivore food webs under climate change, while forest interior networks rapidly decline following simulated loss of their plant components. These findings contrast with previous studies in the fragmentation context, e.g. fragment size-related simplifications of PIHNs (Valladares et al., Reference Valladares, Cagnolo and Salvo2012) and thereby point to the potential importance of floristic and management-related peculiarities of European timber forests. The documented impoverishment of PIHNs in managed forest interiors and the suggested loss of insect diversity from climate-induced co-extinction highlight the need for further research efforts focusing on the promotion of PIHN complexity and implications for ecosystem functioning by adequate silvicultural and conservation approaches.
Supplementary Material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007485317000062
Acknowledgements
We are particularly grateful to the following entomological experts for their generous help in species identification: Peter Sprick, Thomas Thieme, Sabine Walter, Jürgen Deckert, Daniel Burckhardt, Erwin Rennwald and Heidrun Melzer. We further thank field assistants Carina Brenner, Tobias Küpper, Elsbeth Bähner, Eve Caputula, Sarah Herzog, Anna Schmitz, Katrin Gerricke, Dorina Strieth, Katharina Meier, Ireen Lutter and Philippe Golfiere for their valuable contributions. Moreover, we greatly acknowledge Sven Pompe for supplying data and for his guidance on the use of projected plant distributional shifts. Finally, we are grateful for the effort and constructive comments on the manuscript of two anonymous reviewers, as well as Michelle Gehringer. This study was supported by the Rhineland-Palatinate Ministry for Environment, Agriculture, Nutrition, Viticulture and Forestry.
