Figure 1: Schematic sketch of ant observation sites in each study location. Each of the three transects contained six sampling plots at increasing distance to the adjacent old grassland (semi-natural habitat remnant; grey area on top). Hatched squares (OG ) = first sampling plot of each transect in old grassland (reference plots); black squares (NG ) = sampling plots within newly established grassland; grey squares (CN ) = sampling plots in adjacent cereal field near to NG; white squares (CF ) = sampling plots in control cereal field far from NG.
The new grasslands had been established in August 2016 in five winter cereal fields directly adjacent to selected areas of old grassland. In order to mimic the native plant community of the old grasslands, the new grasslands were sown with a variety of seeds from 54 different plant species native to the region (30% grasses, 55% herbaceous plants and 15% legumes). The new grasslands were mowed once every year in late summer and the old grasslands in late June. The use of tillage in the cereal field transects was avoided by the farmers during the sampling period between April and June, but otherwise field management (such as the use of pesticides) continued here.

2.2 Data recording/sampling

Recording of ant activity and species diversity was done by hand collecting of worker ants with fine tweezers. Hand collection has been discussed as the most efficient method for sampling ants (Gotelli et al. 2011). It generates results comparable to those from pitfall traps (Andersen, 1991; Sanders, Barton, & Gordon, 2001), and is not biased in favour of behaviourally dominant species that monopolize food-resources (Andersen, 1997), which may occur when using bait traps. Over the sampling period a total of three consecutive survey runs on each of the 90 sampling plots was performed, with 14 to 21 days between each run. For statistical evaluation, the results of all three runs were aggregated for each transect/habitat. On each sampling plot two 1x1 m sized quadrants around the center were searched for foraging worker ants for four minutes each per run. Worker ants active around nests were also sampled and the total aboveground nest activity (in ants per four minutes) estimated. Prior to the hand sampling also the vegetation cover (0-100 % of soil covered) was estimated in a radius of 2 m around the plot center. All collected individuals were preserved in 70 % ethanol and later identified to species level according to Seifert (2018) using a stereo-microscope at 10-fold magnification.
Measuring of biocontrol potential was done with sticky-card experiments, using adult Drosophila melanogaster (Meigen) flies as baits (Lys, 1995). Over the sampling period a total of four consecutive survey runs on each sampling plot was performed. For statistical evaluation, the results of all four runs were aggregated for each sampling plot/habitat. For each sticky-card, thirty flies were glued to the upper side of a 6x8 cm cardboard, which had a plastic underlay (to protect the card from soil moisture), and fixed to the ground with a long nail. Flies were glued to the cardboard with well-diluted fish-glue enabling ground-dwelling predatory arthropods to remove the prey which guarantees successful predation (Lys, 1995). Each cardboard was covered by an enclosure with an appropriate mesh size (1x1 cm) preventing the access of rodents and birds, thus enabling effectively to measure predation on flies by arthropods (Hulme, 1996). Two cardboards were placed on each sampling plot per survey and exposed to predatory arthropods for two and a half to 3 hours. Afterwards, predation rates (number of destroyed/killed flies) and the estimated vegetation cover of the sampling plots (0-100 % of surface covered) were recorded directly in the field.

2.3 Ant traits

Life history traits of all ant species encountered were taken from Seifert (2017) and (2018) and Arnan et al. (2017). All trait data and a detailed description of trait categories are provided in the appendix: see Tables S1 and S2. The subsequent statistical analysis determined the overall functional trait space covered by the ant communities and examined in detail a selection of traits which are closely linked to biocontrol services.

2.4 Statistical Analyses

All statistical analyses were conducted in the statistical programming environment R (Version 3.6.2, R Core Team 2019). Cumulated ant species richness (ant species in transects pooled per habitat type) was compared across habitat types according to a Monte-Carlo randomization test procedure (Manly, 2006) using the “rich”-package (Rossi, 2011). To investigate how species replacement (turnover) and species loss (nestedness) account for the variation in species composition (beta diversity), the total dissimilarity expressed as Sørensen index (βSOR) across the four habitats, as well as its respective turnover (βSIM) and nestedness (βSNE) components, were calculated using the package “betapart” (Baselga & Orme, 2012).
In order to study the influence of habitat type on ant species composition of the transects, a constrained ordination analysis was performed. A dummy species with an abundance of one in all samples was added to the presence/absence data, to deal with low numbers of species per transect (Clarke, Somerfield, & Chapman, 2006). Based on this dataset a Sørensen dissimilarity matrix was created using the package “vegan” (Oksanen et al., 2018). Subsequently, a canonical analysis of principal coordinates with two axes on the Sørensen dissimilarity matrix was performed, with habitat type as a constraint variable. Differences between the habitat types were tested for significance with a PERMANOVA using “adonis” function, where the habitat type served as fixed factor and the study region (Elsbach, Ollern) as random factor.
A principal component analysis of the species-trait data was performed using the package “FactoMineR” (Lê, Josse, & Husson, 2008) and the first two principal coordinates of each species plotted in a two-dimensional diagram. In order to display the functional trait space covered by ants in the different study habitats, a convex hull (polygon) was drawn around the respective species communities. Differences between the habitat types were tested for significance with a PERMANOVA based on an Euclidean distance matrix of the species trait data.
Community weighted mean (CWM) values of selected ant species traits were calculated using the “FD”-package (Laliberté & Legendre, 2010; Laliberté, Legendre, & Shipley, 2014). The calculated CWM values refer to the average of species trait values at each sampling transect weighed by the relative species abundance (Lavorel et al., 2008; Ricotta & Moretti, 2011). As the observed abundance of foraging workers showed high fluctuations caused by e.g. life cycle stage of ant colonies (Seifert, 2018), the analysis was based on a pseudo-abundance matrix, which refers to the presence of the respective species on the number of runs (0-3) on each plot pooled per transect. Cereal field habitats were excluded from the analysis, as a reliable calculation of CWM requires at least three species, which was not given for the majority of transects of these habitats. The analysis focussed on three distinct traits:proportion of animal-based resources in ant diet (Zoopha ; food resources acquired via predation or scavenging, see Table S1),recruitment behaviour of workers (FS ; foraging strategy) and colony size (CS ; number of individuals). In order to increase the suitability of the CWM values for linear models and to attain normal distribution, logit transformation was applied for the CWM values of the traits Zoopha and FS and log transformation for the trait CS . Using the CWM values of the selected traits as response variable, the habitat type (OG, NG) as predictor variable and study region as random factor three generalized linear mixed models (GLMM) were created using the package “lmerTest” (Kuznetsova, Brockhoff, Christensen, & Jensen, 2019), which computes p-values via the Satterthwaite approximation. To access p-values for the comparison among fixed factor levels, Tukey’s post-hoc tests were conducted using the package “multcomp” (Hothorn, Bretz, & Westfall, 2008). The same approach was applied for the subsequent models.
To investigate the effect of habitat type and mean vegetation cover (0-100 % of sampling plot surface covered) on predation intensity on sticky cards, a predation rate (0-1) was calculated based on the number of eaten flies per sampling plot summed across all four survey runs (n per 240 flies in total; 30 flies x 2 cards per plot x 4 runs). Subsequently, a GLMM was created, with the predation rate per sampling plot as response variable, habitat type and logit transformed mean vegetation cover of the sampling plot as fixed factors and study region as random factor. Furthermore, the same GLMM approach was used to test whether logit transformed mean vegetation cover showed significant differences between habitat types. Fixed effect structures and GLMMs were compared using the packages “multcomp” and “MuMIn” (Bartoń, 2019).
To study the effect of habitat type on aboveground ant activity, the number of observed worker ants per sampling plot was summed across all three survey runs and transformed with Tukey’s Ladder of Powers, in order to attain normally distributed values, using the package “rcompanion” (Magnificio 2019). Subsequently, a GLMM was created with Tukey transformed number of observed workers per sampling plot as response variable, habitat type and logit transformed mean vegetation cover of the sampling plot as predictor variables and study region as random factor. Further, the correlation of predation rate on sticky cards and Tukey transformed aboveground ant activity was tested with a GLMM with study region as random factor, again. Marginal and conditional R² values (R²m/R²c) of the GLMM were calculated using the package “MuMIn”.