Impact of an invasive tree on arthropod assemblages in woodlots isolated within an intensive agricultural landscape

Landscape simplification and the spread of invasive species are considered beyond the main threats to global biodiversity. It is well recognized that non‐crop habitats bring complexity to farmland and provide refuge for a wide range of organisms, including arthropods. However, knowledge about the effects of invasive trees on arthropods in non‐crop habitats in intensive agricultural landscapes is still weak. Therefore, we examined differences in the arthropod assemblages between woodlots formed by the invasive black locust (Robinia pseudoacacia L.) and by native deciduous tree species in the intensive agricultural landscape.

In many cases, they are left to spontaneous successional processes or are afforested (as a shelter for wild game) by fast-growing and durable tree species (Benayas, Bullock, & Newton, 2008;Lassoie, Buck, & Current, 2009). In general, biodiversity in forest habitats seems to strongly depend on vegetation structure, which is largely conditioned by the dominant tree species Highland, Miller, & Jones, 2013;Kadlec, Štrobl, Hanzelka, Hejda, & Reif, 2018;Tews et al., 2004). The dominant tree species has a strong effect on heterogeneity of habitat structure and canopy-openness, both of which are positively linked to arthropod diversity in large European lowland forests (Kadlec et al., 2018;Sebek et al., 2015). Similar effects can also be expected for the isolated woodlots in agricultural landscapes.
In this study, we compared the arthropod assemblages from woodlot islands dominated by invasive R. pseudoacacia with those formed by native tree species. The effects of R. pseudoacacia were investigated across several arthropod taxa from different trophic levels, including herbivores, carnivores and detritivores. We adopted this multi-trophic and multi-taxonomic approach to better understand the interactions within and between trophic levels (Seibold, Cadotte, Maclvor, Thorn, & Müller, 2018). The following predictions were made: 1. Similar to large forest stands, R. pseudoacacia will create a more open-habitat structure of isolated woodlots in agricultural landscapes than native tree species.
2. The total arthropod species richness and abundance will be higher in woodlots dominated by native tree species and with a more open-habitat structure, but these effects could vary between taxa and trophic levels. We expect stronger effects in herbivorous taxa than in predators or detritivores.
3. Tree invasion and habitat structure will affect the composition of arthropod assemblages in woodlots. Forest specialists will be more dominant in native woodlots, whereas species of (semi) open habitats will be affiliated with R. pseudoacacia woodlots.

| Study area and sampling design
The study was conducted in a lowland agricultural landscape of the Czech Republic, Central Europe ( Figure 1) in 2016. The study area (50.10°-50.46°N, 14.05°-14.83°E, ~1,300 km 2 , 160-330 m a. s. l.) is located in a region with a moderately continental climate with an average annual precipitation of 500-600 mm and an average annual temperature of 8-9°C (Quitt, 1971). The landscape is dominated (>70%) by large, intensively managed arable fields, with scattered grasslands, cultivated lowland forests and human settlements making up most of the remaining area. The heavily fragmented forests are mainly formed by native broadleaved tree species or non-native trees, mostly by the invasive R. pseudoacacia.
The arthropods were sampled using pitfall traps, sweep-netting and light trapping. In each woodlot, a line of five pitfall traps (two nested plastic cups, 94 mm perimeter × 144 mm height, containing 4% formaldehyde; Spence & Niemelä, 1994) spaced every five metres was established and operated continuously from the beginning of April to the beginning of September (152 trap days, emptied at monthly intervals). The lines of traps were located at least 10 metres from the edge of the woodlot to minimize edge effects (Roume, Deconchat, Raison, Balent, & Ouin, 2011). The captured samples were frozen at −22°C. In parallel with trap emptying, all of the vegetation up to a height of 3 m in the 25 × 5 m strip area centred around the line of traps was swept when weather conditions were suitable (sunny, no strong wind) using a 35 cm diameter sweeping net. The captured arthropods were preserved in 95% ethanol. To sample nocturnal arthropods, portable light traps (Brehm & Axmacher, 2006) equipped with two 8 W UV LED strip lights (total luminous flux 400 lm, wavelength F I G U R E 1 Map showing the location of the study plots (15 woodlots dominated by Robinia pseudoacacia, and 15 woodlots formed by native tree species) range 400-420 nm, powered by 7.2 Ah/12 V lead batteries) were used, and collected specimen was euthanized by evaporating chloroform. A single portable trap was placed approximately in the middle of each woodlot and attracted arthropods within a radius of a few tens of metres (Truxa & Fiedler, 2012). To standardize for the weather and moon-phase (Yela & Holyoak, 1997) the light traps were exposed on the same night under suitable weather conditions (no strong wind, no rainfall and increased cloud cover), from dusk until dawn, at the beginning of each month from April to September. The samples from the light traps were frozen at −22°C.
All samples were sorted according to the target taxa, counted and identified to the species level (see Appendix S2). Data from all of the sampling methods and periods were pooled for the particular taxa and woodlots into a final data set. The conservation status of each species was classified according to the national red lists (Hejda, Farkač, & Chobot, 2017;Řezáč, Kůrka, Růžička, & Heneberg, 2015

| Statistical analysis
To reduce the complexity of habitat structure and landscape structure data without substantial loss of information and to describe the main gradients of habitat structure and land cover characteristics of the studied woodlots, two principal component analyses were conducted in cAnoco 5.0 (PCA; ter Braak & Šmilauer, 2012): one for the habitat structure and one for the land cover characteristics. We used the scree plot method (Jackson, 1993) to distinguish the principal components explaining most of the variability in the data. Based on this criterion, in both PCAs, the scores from the first two principal components (PC1 and PC2) of habitat structure (henceforth called 'HAB1' and 'HAB2') and land cover characteristics ('LAND1' and 'LAND2') were used as predictors in the following analyses.
To compare the habitat structure between the native and Robinia woodlots, linear models were fitted with the principal components of habitat structure (HAB1 or HAB2) as the respective response variables and the woodlot type (WOODLOT TYPE: native or R. pseudoacacia) as the predictor.
The two native woodlots were excluded from most of the analyses because the majority of the pitfall traps were destroyed by wild animals. Therefore, the data from 13 native and 15 Robinia woodlots were used in analyses, except for the models of Lepidoptera and Neuroptera, as these data were not based on pitfall traps.
As the first step, we examined the differences in the total abundance and total species richness (both summed across all taxa) between the two types of studied woodlots (WOODLOT TYPE: R. pseudoacacia/native) as the only explanatory variable. This approach is often used in studies on the effects of plant invasions (van Hengstum et al., 2014;Litt et al., 2014). Thus, we used generalized linear models (GLMs) with Poisson or negative binomial distributions (to reduce overdispersion) of the errors. In contrast, the simple effect of plot (WOODLOT TYPE in our study) may represent the combined effects of the origin of the dominant tree species (as a measure of food availability for herbivores) and woodlot habitat structure (as a measure of ecological niche diversity). Therefore, in the next GLMs, we examined the direct effects of dominant tree origin (predictor TREE TYPE: R. pseudoacacia/native) and the effects of habitat structure (predictors HAB1 and HAB2). The effects of the surrounding landscape composition (LAND1 and LAND2) and woodlot area (AREA) were also included in these models. Full GLMs with the total abundance or species richness per woodlot (for each taxon and summed for all taxa) as response variables with all the mentioned predictors were performed. Distributions of errors employed in models are mentioned in Table 1. Potential spatial autocorrelation of the residuals was checked by a Mantel test (integrated into r package 'Ade4'; Dray & Siberchicot, 2018), and geographic coordinates were added to these models to account for autocorrelation if needed (according to Carrié, Ekroos, & Smith, 2018).
Furthermore, an information-theoretic approach (r package 'MuMin', Bartoń, 2018;Burnham & Anderson, 2002) was used for model selection and multimodel inference. The candidate models containing all possible predictor combinations were compared TA B L E 1 Model-averaged estimates of the effects of particular predictors on the total number of individuals and the total number of species of the studied arthropod taxa between the native and Robinia woodlots by AICc (Akaike, 1974;Burnham & Anderson, 2002). Models with ΔAICc < 2 were considered superior. These models were used for inference employing model averaging using AIC weights (Bartoń, 2018;Burnham & Anderson, 2002). Univariate analyses were performed in R 3.5.1 (R Core Team, 2018).
Differences in species composition between the woodlot types were analysed by multivariate ordination methods. Based on the gradient lengths (for all models a gradient was at least 1.9 SD units long), canonical correspondence analyses (CCAs) were used (Šmilauer & Lepš, 2014). In the first step, CCA with the species data pooled across all taxa was performed to investigate differences in the total species composition between the woodlot types. The species compositions of particular taxa were compared between the native and Robinia woodlots using separate CCAs. Prior to this, we checked for possible correlations between the effects of TREE TYPE (R. pseudoacacia/native) and habitat structure (HAB1 and HAB2) in the woodlots on the arthropod species compositions (Kadlec et al., 2018). We used the variation partitioning approach

| Vegetation and land cover characteristics of the native and Robinia woodlots
PC1 axis of the habitat characteristics (HAB1, 33.16% of the variation in the habitat structure explained) described the gradient from woodlots with larger trees, a more developed shrub layer and a continuous canopy to more open woodlots with smaller and thinner trees, a more developed taller herb layer and a higher number of dead trees (see Figure S1.2 in Appendix S1 and Figure 2a). The Robinia woodlots had significantly higher scores along the gradient of HAB1 than the native woodlots (t = 4.814, p < .001; Figure 2a). Regarding the surrounding land cover characteristics, PC1 axis (LAND1, 45.86% of the variation in the land cover structure explained) reflected the gradient from landscapes with a larger proportion of arable fields to landscapes with a higher proportion of non-crop habitats, such as coniferous woodlands, urban areas and grasslands ( Figure 2b). PC2 axis of the surrounding land cover characteristics (LAND2, 20.64% of the variation in the land cover structure explained) mainly described the gradient from landscapes with a larger proportion of wetlands, exposed rocks and broadleaved forests to landscapes without these habitats (Figure 2b). The native and Robinia woodlots did not differ along their land cover gradients ( Figure 2b).

| Arthropod abundance and species richness
Altogether, 62,133 individuals (see Table S1.1 in Appendix S1) of 989 arthropod species (742 species in native/767 in Robinia/523 shared by both woodlot types) were recorded (see Appendix S2 and Table   S1.2 in Appendix S1). As indicated by the GLMs with WOODLOT TYPE as the only predictor, the total abundance of arthropods in the Robinia woodlots (mean ± SD = 1,782 ± 479) was lower than that in the native woodlots (2,665 ± 887; z = −3.497, p < .001), while the total species richness did not differ between the Robinia (mean ± SD =217 ± 22) and native woodlots (220 ± 28; z = 0.542, p = .588). Overall, 89 of the species recorded (ca. 10% of all species; 28 in native/38 in Robinia/28 in both) are included in the national red lists.
The more detailed analyses considering habitat characteristics showed a significant negative relationship between the presence of R. pseudoacacia and both the total abundance and the total species richness of arthropods (Table 1). The total species richness also increased along the gradients of HAB1 (towards a more developed taller herb layer, a more open canopy and a higher number of dead trees) and LAND2 (towards a higher proportion of broadleaved forests and water habitats in the vicinity of the woodlots) ( In the woodlots with more developed shrub and canopy layers and larger trees, the abundance of Neuroptera increased (Table 1).
Furthermore, the abundance of Elateridae significantly decreased along the gradient of HAB2 (younger woodlots with a high proportion of clearings in the canopy and lower trees) (Table 1). By contrast, the species richness of Heteroptera increased along the gradient of HAB 2 (Table 1).
The land cover in the surrounding landscape also contributed to the variation in the arthropod communities. Along the gradient of LAND1, the abundances of Elateridae were higher in the woodlots surrounded by a higher proportion of non-crop habitats (

| Arthropod species composition
The Robinia woodlots were more heterogeneous in their overall arthropod species composition and differed from the native woodlots (pseudo-F = 2.8, p < .001; see Figure S1.3 in Appendix S1). For most of the taxa, significant marginal effects of habitat structure on the species compositions were found (see Table S1.3 in Appendix S1).
For the majority of these taxa, conditional effects of the dominant tree species and habitat structure were still significant (except for Diplopoda), but the percentage of explained variance was rather low compared to that of the marginal effects (see Table S1.3 in Appendix S1). Moreover, the species compositions of particular taxa (except for Neuroptera, Orthoptera and Silphidae) were significantly different between the native and Robinia woodlots (Table 2) after controlling for spatial (PCo scores from the PCNMs) and environmental variables (LAND1, LAND2 and AREA). Forest specialists occurred primarily in the native woodlots and were more frequent in the woodlots characterized by a more developed canopy and shrub layer (decreasing HAB1; Figure 3). In contrast, open-habitat species were more frequent in the Robinia woodlots, with the exception of Carabidae (Figure 3), and preferred woodlots with more developed taller herb layers, more open canopies and a higher number of dead trees (increasing HAB1). These trends were also evident in the majority of the threatened species with an affinity for a given habitat structure type. The majority of the predominantly herbivorous taxa (Curculionoidea, Heteroptera and Lepidoptera), which are typical of scattered greenery, were more abundant in the native woodlots ( Figure 3). Note: The effects of habitat structure (HAB1 and HAB2) were included in case of their significance for particular taxa. All of the CCAs were controlled for the environmental (LAND1, LAND2 and AREA) and spatial effects (PCo scores from PCNMs). The significant effects (p < .05) are highlighted in bold. Robinia and native woodlots differed significantly in their habitat structure, which may be another key driver for the observed differences in the arthropod communities. A higher total arthropod species richness was found in the woodlots with a more open canopy, a higher coverage of taller herbs and a higher number of F I G U R E 3 Canonical correspondence diagrams showing the species distribution of the studied taxa between the native and Robinia woodlots. The effects of habitat structure (HAB1 and HAB2) were included in case of their significance for particular taxa. Only the best fitting species (˃5%) in the ordination models are displayed. The symbols refer to forest specialists: green stars; species bounded to scattered greenery: blue diamonds; open-habitat species: yellow down triangles; and habitat generalists: black crosses. Red-listed species are displayed by red-coloured symbols. See Table 2 for model details dead trees (represented by increasing HAB1). Such habitat structure (more common in the Robinia woodlots) offers a more irradiated and warmer understorey due to higher amounts of solar radiation penetrating through the canopy layer (Cierjacks et al., 2013;Vítková et al., 2017;Xu et al., 2009). Contrary to the recent findings from larger R. pseudoacacia forests (Buchholz et al., 2015;Kadlec et al., 2018), the shrub layer was better developed in the native woodlots. The majority of the Robinia woodlots in this study could be included in phytocoenological units dominated by wellcompetitive grasses (e.g. false oat-grass, Arrhenatherum elatius (L.)

| D ISCUSS I ON
J. Presl et C. Presl) in their understories, which could effectively suppress shrub seedlings (Campagnaro, Nascimbene, et al., 2018;Vítková & Kolbek, 2010;Vítková et al., 2017). This effect could be enhanced by heat and water stress in the understorey caused by R. pseudoacacia (Xu et al., 2009). Arthropods linked to such insolated understorey vegetation in the Robinia woodlots could partly compensate for loss of forest canopy herbivores due to plant invasion (Kulfan, 2012;Litt et al., 2014;Liu & Stiling, 2006) by filling new available niches (e.g. insolated herbs, rotten wood; Highland et al., 2013;Tews et al., 2004). Nevertheless, the abovementioned direct negative relationship between the presence of R. pseudoacacia and herbivorous Lepidoptera and Curculionoidea was stronger than the effect of changes in vegetation structure.
This was probably caused by a higher contribution of herbivorous canopy specialists within the moth assemblages (Kadlec et al., 2018) and high abundances of Curculionoidea exploiting broadleaved trees in the native woodlots (Koch, 1992), but not able to feed on exotic Robinia (Kulfan, 2012;Litt et al., 2014;Liu & Stiling, 2006). By contrast, higher abundance of Elateridae and abundance and species richness of Heteroptera is probably linked to the habitat structure as many open-habitat specialists were present mainly in the Robinia woodlots. Similarly, Buchholz et al. (2015) found an increased abundance of Heteroptera within stands with a more open canopy. Simultaneously, Elateridae had higher abundances in the woodlots with older and larger trees (represented by negative HAB2 scores), probably due to the higher occurrence of xylophagous species in such conditions (Irmler, Heller, & Warning, 1996).
No direct relationship between the presence of R. pseudoacacia and the abundance and species richness of carnivorous taxa was found, except for Carabidae. This is in accordance with the weak impact of woody invaders on carnivorous arthropods that have been found elsewhere (Buchholz et al., 2015;Litt et al., 2014;Van der Colff et al., 2015). The lower abundance of carabids in the Robinia woodlots is surprising, as it contradicts the earlier findings from R. pseudoacacia forests (Buchholz et al., 2015). We suppose that carabids benefit from the more favourable microclimate in the native woodlots, in which relatively greater humidity may support more ample food resources, such as springtails, earthworms and gastropods. Similar to our results, Knapp and Řezáč (2015)  Matusch)).
The abundance and species richness of detritivorous Diplopoda did not differ between the two woodlot types. A high amount of nitrogen in the R. pseudoacacia litter (Tateno et al., 2007) and more decaying vegetation in its understorey (Vítková et al., 2017) could compensate for the lack of leaf litter from the native tree species.
Detritivorous arthropods are often even positively influenced by plant invasions (Harris et al., 2004;Litt et al., 2014), but it has not been shown in the case of Robinia woodlots or in large R. pseudoacacia forests (Buchholz et al., 2015).
The surrounding land cover composition was also significantly linked to the woodlot arthropod assemblages. Increasing proportions of broadleaved forest (decreasing rates of woodlot isolation; Baz & Garcia-Boyero, 1995;Torma et al., 2014) and wetlands (represented by LAND2) in the vicinity of the woodlots were positively related to the total arthropod species richness. In contrast to generally positive species-area relationship (Mac Arthur & Wilson, 1967) as well as the previous examinations of species-area relationships for woodlot arthropods (Baz & Garcia-Boyero, 1995;Knapp & Řezáč, 2015), the woodlot area was not linked to the species richness and abundance of almost any of the investigated arthropod taxa in our study. This could be caused by a limited variation in sizes of our woodlots (0.11-1.31 ha). Nevertheless, within a limited range of areas, the effects of vegetation cover and habitat structure may outweigh the importance of area (see also Knapp & Řezáč, 2015;Torma et al., 2014).
Similar to the species richness and the abundance of the target groups, the differences in habitat structure between the native and Robinia woodlots were also reflected in the species compo-

| CON CLUS IONS
The results of this multi-taxonomic study covering various trophic levels highlight the importance of habitat structure in assessments of the impacts of tree invasion on native arthropod communities.
Despite their small size, the forest islands isolated within the intensively managed agricultural landscapes of Central Europe host diverse arthropod assemblages and are enriched by a considerable number of threatened species. Although R. pseudoacacia is considered one of the most harmful invasive trees for native ecosystems (Campagnaro, Brundu, et al., 2018;Vítková et al., 2017), our study on the arthropod assemblages in woodlots does not fully support this view. Due to their more open-habitat structure, the Robinia woodlots support open-habitat arthropod species, including endangered specialists. These specialists could also use woodlots as short-term ref- uges or shelters during agricultural disturbances in the arable fields.
Nevertheless, the majority of forest specialists, including canopy herbivores, were negatively influenced by R. pseudoacacia and were more common in the native woodlots. Thus, presence of Robinia and native woodlots scattered across intensively managed arable fields deliver substantial support for arthropod biodiversity and provide refuges for arthropods with different ecological and trophic requirements. Moreover, forest management of small woodlots supporting biodiversity is limited due to their isolation by arable land. The habitat structure of the native woodlots could turn to shaded dense stands in later successional stages, whereas similarly old Robinia woodlots form spontaneously more open stands (Vítková et al., 2017) inhabited by different arthropod assemblages. Moreover, due to the limited spreading of R. pseudoacacia (Cierjacks et al., 2013;Vítková et al., 2017) from isolated woodlots, its negative impact on more valuable native habitats in the surroundings is minimized. For these reasons, we conclude there is no need for eradicating R. pseudoacacia from existing woodlots within agricultural landscapes, as has been recommended for valuable native habitats (Campagnaro, Brundu, et al., 2018;Cierjacks et al., 2013;Vítková et al., 2017).

ACK N OWLED G EM ENTS
We thank Barbora Tojflová, Ondřej Štrobl, Jiří and Lenka Skalová and Michaela Černá for help with fieldwork. We also thank Barbora

DATA AVA I L A B I L I T Y S TAT E M E N T
The data are provided in the Supporting Information.