2.1 Study sites

Wild bees, flowering insect-pollinated plants (“flowering plants” from here onwards) in the vineyard inter-rows, farm management (organic vs. conventional), inter-row vegetation cover (as proxy for vegetation management intensity), and landscape composition were assessed in two viticultural regions in Austria and South Africa. Austria is located in the Palearctic biome and the Holarctic Floral Kingdom, whereas the study sites in South Africa are located in the Fynbos biome of the Cape Floristic Region, a Mediterranean-type ecosystem that supports expansive species radiation and endemism among indigenous plants (Johnson et al., 2006).

Austrian study vineyards were located in two Eastern Austrian viticultural areas (Carnuntum: 48°04′N, 16°47′E and Neusiedlersee–Hügelland: 47°54′N, 16°41′E) characterized by rain-fed vineyards consisting of small parcels (0.4–1.0 ha) with trellis systems on plain or hilly terrain. The study area is characterized by a small-scaled agricultural landscape with vineyards, arable fields, seminatural habitats (SNHs), woods, and villages (Figure 1a) (Kratschmer et al., 2018). The average distance between sites was 13.4 km. The climate is classified as warm temperate (Cfb according to the Köppen–Geiger climate classification (Kottek et al., 2006)). During the two study years (2015 and 2016), average air temperature was 11.5°C and 11.1°C, and annual precipitation was 508 and 636 mm, respectively (ZAMG, 2017). In Austria, 24 selected vineyard inter-rows were either covered with permanent vegetation (no tillage for >5 years) or temporary vegetation (alternating tillage in every second inter-row) in the center of 16 landscape buffers (Figure 1a,c). In eight of these landscape buffers, paired vineyards differing in inter-row vegetation management regimes (n = 16) were studied (Figure 1a). Management information was gathered by means of personal interviews with the winegrowers (Table 1). Conventional vineyard management used herbicides and mechanical weed control only under grapevines, and additional fungicides. Organic winegrowers only used mechanical weed control and copper and sulfur for fungal control. No synthetic insecticides were used in the study vineyards six years prior to the study. The inter-rows were covered with either seeded cover crops or spontaneous vegetation from the existing seed bank or surrounding vegetation (Table 1).

The South African study vineyards were located in the Western Cape Province (33°57’S, 18°46’E) near the town of Stellenbosch, which are characterized by large rain-fed vineyards (4–10 ha size) located on plain or hilly terrain, with natural habitats (i.e., fynbos and renosterveld vegetation) surrounding the vineyards. However, there were also high-density patches of invasive alien tree species (mostly Pinus spp., Eucalyptus spp., and Acacia spp.) and deciduous fruit and olive orchards (Figure 1b). The climate is Mediterranean-type (Csb according to the Köppen–Geiger classification (Kottek et al., 2006)), the mean annual temperature was 17.9°C and 16.6°C, and annual precipitation was 600.2 and 463.9 mm, respectively (Meijers, 2020) for the two years of investigation (2009 and 2010). Five pairs of organic and conventional vineyards, each within 0.13–1 km distance (Figure 1), were surveyed. The average distance between sites was 12.9 km. The guidelines for conventional and organic vineyard management are similar to those described above, but conventional winegrowers in South Africa use low-risk insecticides sparingly as part of the Integrated Production of Wine scheme of SA (Wine and Spirit Board, 2020). The inter-rows of both organic and conventional vineyards were covered with vegetation (Table 1) seeded with cover crops such as Hypochaeris radicata, Raphanus raphanistrum, Erodium moschatum, Bidens pilosa, Avena fatua, and Vicia spp. (Kehinde & Samways, 2012), as well as species emerging from the soil seed bank.

2.2 Sampling designs for wild bees and flowering plants

In Austria, wild bees were sampled using 100- to 130-m-long transects (transect width given by inter-row width) in two neighboring inter-rows per vineyard (Figure 2a). Transect length was adapted according to the width of the inter-row, which ranged between 1.5 and 2 m. Sampling was conducted monthly between April and August, resulting in 5 transect walks in every vineyard in both study years. Each sampling was done within 2–3 days between 9 a.m. and 4 p.m. on sunny, less windy days with temperatures above 15°C and dry vegetation. To avoid time of day bias, each vineyard was visited at different times of the day throughout the sampling period. During a sampling period of 15 min per transect, all bees observed were collected with a handheld insect net for later identification in the laboratory. Honey bees and most bumblebee species were identified and counted in the field. The five sampling dates per study year were adjusted to the vine's phenological stages, also complying with wild bee sampling recommendations (Schindler et al., 2013), starting in April (first leaf buds) until September (start of grape maturation). Bees were identified to species level based on identification keys (Schmid-Egger & Scheuchl, 1997; Amiet et al., 1999, 2001, 2004, 2007, 2010; Scheuchl, 2000, 2006; Gokcezade et al., 2010; Dathe et al., 2016) using the nomenclature according to Gusenleitner et al. (2012). Voucher specimens were deposited at the Institute for Integrative Nature Conservation Research (INF) at BOKU (Vienna). Number and cover of flowering plants were also recorded along each sampling transect and identified to species using Fischer et al. (2008). Vegetation cover (%) and plant species richness including non-insect-pollinated plants (Hall et al., 2020) were estimated twice per study year (at the beginning of the vegetation period and 2 months later) in four 1 × 1 m subplots of one inter-row per vineyard.

The South African wild bee data represent a combination of three sampling methods performed in each vineyard in spring to summer (August to December) 2009 and 2010 (Figure 2b). In 2009, 12 yellow (non-UV) pan traps (capacity of 1,000 ml/trap) and two window intercept traps (0.5 × 0.5 m) were left in the field for five consecutive days on two sampling dates. Pan and window traps were arranged in pairs (traps in a pair were 2 m apart) with a distance of 20 m between each site and from the edge of the field. In 2010, sampling was performed by walking the transect, collecting wild bees with an aerial net in a predefined plot (100 × 50 m) per vineyard. Within these plots, three 50 × 2 m transects were placed randomly and sampled six times with a two-week interval between visits, with each transect sampled for 20 min resulting in 1 hr of sampling per plot (Kehinde & Samways, 2012, 2014a, 2014b). Transect walks were done on days without rain, minimal wind, minimal cloud (< 5%), temperature >15°C, and between 9 a.m. and 4 p.m. Bees were identified to species or morphospecies (especially Halictidae) (Michener, 2000). Voucher specimens were deposited at the National Collection of Insects, Pretoria, South Africa. Flowering plant species data were collected in 2010 during the plant–pollinator interaction field survey (Kehinde & Samways, 2014c) and considered to be similar to 2009 as management remained constant (Lososová et al., 2003). Both vegetation cover (%) and total plant species richness were assessed in 2009 along two transects per vineyard plot. Each transect consisted of six 2 × 2 m subplots with 5-m intervals located in the vineyard center (Kehinde & Samways, 2012).

Each bee species was described in terms of 7 functional traits (Table 2) that are recognized as important for bee autecology (Michener, 2007). All but two of the traits were obtained from relevant literature (Greenleaf et al., 2007; Scheuchl & Willner, 2016; Westrich, 2018) and expert evaluation (for the South African species, inferences were made based on available literature (Eardley, 2013; Eardley & Urban, 2010; Gess & Gess, 2014; Kuhlmann et al., 2011; Michener, 2007). Lecty and sociality were only used for the Austrian bee data, because ecological information on the South African bees was limited. Further, species of the genera Lasioglossum and Halictus vary greatly in their sociality; thus, inferences based at the genus level for bees identified to morphospecies were not accurate. For lecty or the degree of pollen specialization, we assigned two categories: polylecty and oligolecty. According to Cane and Sipes (2006), polylecty includes broad polylecty, polylecty, and mesolecty, whereas oligolecty joins narrow oligolecty and monolecty. The foraging range of wild bees has been shown to increase with body size, a standard measure for body size is intertegular distance (ITD) or the distance between the tegulae (Greenleaf et al., 2007). We measured the ITD (Cane, 1987; Greenleaf et al., 2007) of 1–5 individuals per bee species from Austria using a Keyence VHX-5000 digital microscope and 9–24 individuals per bee species from South Africa using a Leica Z16 APO stereoscope. Austrian bumblebees were identified in the field; thus, we measured the ITD from five specimens per species selected from the collection at INF, BOKU (Vienna), sampled in eastern Austria. Mouthpart length (i.e., length of proboscis of each bee species was estimated based on bee family and average ITD (Table S1) using the R package BeeIT (Cariveau et al., 2016). As Melittidae bees were not included in this package, the species’ mouthpart length was estimated using the parameter values of the allometric power function as reported by Melin et al. (2019) and revised using the R package pollimetry (Kendall, 2018).

Each flowering plant species was described by 6 functional traits (Table 2) selected according to the relevance for pollination, flower ecology, and morphology (e.g., accessibility of reward, flower morphology, or flower color). They were obtained from TRY plant trait database, version 5, released in March 2019 (www.try-db.org; Kattge et al., 2020; see full list of references in Appendix—TRY References). Additionally, information on flower morphology was extracted from Kugler (1970), and seasonality from Fischer et al. (2008) for the Austrian taxa, and from Manning and Goldblatt (2012) for South Africa. The information derived from the TRY database and other literature was nominally scaled, but due to their great detail, these were not suitable for our statistical analysis; therefore, they were categorized prior to data analysis (Table S2).

2.3 Landscape evaluation

Landscape composition in both countries was evaluated within a 500 m buffer around each vineyard, as this radius covers the flight range of most wild bee species (Zurbuchen et al., 2010; Zurbuchen & Müller, 2012). In Austria, field mapping was performed in July 2015 following the EUNIS habitat-type classification (European Environment Agency, 2016), using the Austrian land utilization map (BMLFUW, 2012). The digitizing of field data and calculation of the proportions of different landscape features (Table 3) was done in ArcGIS (ESRI, 2013). For South Africa, landscape mapping was based on high-resolution (50 cm) orthorectified aerial photographs taken in 2008 and 2010 (NGI, 2020) and digitizing done in QGIS v3.12.1 (QGIS Development Team, 2020). The habitat-type classification used in AT was adapted to account for the South African landscape characteristics (Table 3).

For further analysis, landscape features were aggregated to habitat classes (Table 3): SNHs only included landscape features characteristic for open areas (e.g., natural and seminatural grasslands, natural sclerophyllous vegetation, fynbos, fallow land, grass strips, hedges). In Austria, SNHs include a greater proportion of seminatural elements, whereas in South Africa, SNHs represent more pristine habitats such as the fynbos and renosterveld vegetation. Woody structures (e.g., woodlots, tree rows, alien tree stands, forests) were pooled in a separate class for several reasons: Firstly, most bee species colonize open habitats with many of those subsumed in SNH. Secondly, woody structures may be particularly important for aboveground nesting bee species. Thirdly, woody structures represent different characteristics in the two countries. While in Austria, they consist mostly of native tree species and represent seminatural structures (Burgenländische Landesregierung, 2018), in South Africa, a high proportion of introduced, invasive tree species (pine, eucalyptus, and acacia species) are present in the woody structures, with natural woodland uncommon in the shrub-dominated fynbos vegetation (Melin et al., 2018; Rebelo et al., 2006).

2.4 Data analysis

Numerical variables of repeated measurements per study vineyard were aggregated (species numbers of wild bees and flowering plants) or averaged (mean plant species aggregated across seasons, mean vegetation cover). The only exception was one vineyard in Austria, which was permanently vegetated in 2015, but tilled in early spring 2016, and therefore, the two years were treated as separate observations. Honey bees (Apis mellifera) were excluded from analysis, because managed hives may lead to biased results. Additionally, brood–parasitic wild bee species were excluded from analysis, because they predominantly depend on the occurrence of the host species and no parasitic bee species were reported in the South African vineyards. Except for data standardization with z-scores, all statistical analysis was performed in R (R Core Team, 2019).

Functional trait richness and community-weighted means (CWMs) of wild bees and flowering plants were calculated with the function dbFD in the R package FD (Villéger et al., 2008; Laliberté et al., 2015). The calculation of wild bee FRic included all functional traits (Table 2). To estimate the flowering plants’ FRic, presence/absence data were used, because the cover of flowering plant species was not assessed in all studied vineyards. The Cailliez correction method was applied to avoid biased estimations of FRic of wild bees and flowering plants (Laliberté et al., 2015). As only categorical traits were selected to represent flower traits (Table 2), FRic was measured as the number of unique trait combinations per vineyard. In order to include floral traits as predictor variables, the dominant class (i.e., CWM) per floral trait (Table 2) and vineyard was obtained from the dbFD function.

To analyze the effects and possible interaction of environmental filters at different spatial scales on wild bees and flowering plants, the datasets were analyzed together with linear mixed models. To enable the joint statistical analysis of the two countries, and to account for the differences in methodology and sampling effort, numerical variables (y) (i.e., count data of wild bee species, flowering and mean plant species richness, FRic indices, inter-row vegetation cover, proportion of SNHs, and woody structures) were standardized per vineyard (i) and country (j) by calculating z-scores (z_ij = (y_ij – y̅_J) /SD_j). The z-scores do not modify the relationship between response and predictor variables (c.f. Dainese et al., 2019; Garibaldi et al., 2015).

For each response variable (wild bee species richness and FRic, FRic of flowering plants), we fitted a model set of linear mixed-effect models (i.e., generalized linear models of the Gaussian family) using the R packages lme4 (Bates et al., 2015) and nlme (Pinheiro et al., 2020). One model set either included single predictors (summarized in Table 4) or a priori defined combination and the interaction of two noncollinear predictors (e.g., farm type × SNH) according to our research questions. The flower trait CWMs were also included in the models for wild bees. To account for spatial autocorrelation, we included the locality (i.e., the landscape buffer IDs for AT data; localities with plot pairs for SA data) as random structures in the models. Although using the z-score standardization method, the country effect should be ruled out, we included country as fixed effect in models per set, but none of these models turned out to be most parsimonious. As the flowering plant FRic was strongly correlated with the total species richness of flowering plants (Spearman's correlation: p < .01, r = .64), we did not model total flowering plant species richness. Model selection was based on the second-order Akaike information criterion, which corrects for small sample sizes (AICc; Motulsky & Christopoulos, 2003). The cutoff to evaluate the most parsimonious models for a response variable was set at ΔAICc ≤ 2 (R package AICcmodavg; Mazerolle, 2016). Model quality of the highest ranked models was assessed by diagnostic plots (R package DHARMa; Hartig, 2017) and by calculating the conditional and marginal R² (R package MuMIn; Barton, 2016). Additionally, the variance inflation factor (VIF) was calculated for the models with two predictors, to ascertain correct parameter estimation and absence of collinearity between predictors (Zuur et al., 2010). VIFs <3 were considered as absence of collinearity in the models. Graphical visualization of the best model was performed with the effects package (Fox, 2018).

To analyze country-specific associations of different functional bee traits and flower traits, and their dependence to vineyard management intensity and landscape parameters, a fourth-corner analysis was conducted. Using the function traitglm in the R package mvabund (Wang et al., 2012, 2020), we modeled functional bee and flowering plant traits together with vineyard and landscape parameters (not included in flower trait models). Additionally, a LASSO penalty was applied to reduce the environmental–trait relationships to 0 when the correlation was small, which ultimately improves the interpretability of the models’ result. The bee trait models were fitted with negative binomial distribution and the flower trait models with binomial distribution. To study comparable patterns affecting bee and plant traits in the Northern and Southern Hemispheres, each country was analyzed separately with nonstandardized data. Sociality and lecty were excluded from the South African analyses due to missing trait information for many of the morphospecies (mainly from the Halictidae family, which are not fully described in the region, and no species identification keys are available). Flowering seasonality was excluded from the flower trait models (both countries), because some species were only recorded at genus level. For nominal trait variables with two categories (e.g., eusocial and solitary), the results for one category are presented, because the results are the inverse for the two variables.

3.1 Common drivers of species and functional richness in vineyards across countries

Overall, regression models revealed a strong relationship between FRic of wild bees and insect-pollinated flowering plants, with these variables improving the model fit considerably. Furthermore, vineyard management and less importantly landscape parameters played an important role in wild bee models. This was not the case for flowering plant FRic, which was only affected by vineyard-scale parameters.

The regression analysis revealed three equally parsimonious models explaining wild bee FRic in vineyards (Table 5), highlighting the strong positive effect of increasing flowering plant FRic interacting with organic versus conventional vineyard management (Figure 3a; Table S3). The strong positive relationship between bee and plant FRic is further underlined by the solely positive effect of flowering plant FRic on wild bee FRic (Figure 3b; Table S3). At the landscape scale, the increasing proportion of woody structures had a weak positive effect on wild bee FRic (Figure 3c; Table S3).

Wild bee species richness, for the four models, were equally parsimonious (Table 5). They revealed similar predictors as for wild bee FRic, with positive effects of flowering plant FRic (Figure 3d; Table S3), as well as an increase in vegetation cover in the inter-rows improving wild bee species richness (Figure 3e; Table S3). At the landscape scale, wild bee species richness was positively affected by increasing proportions of woody structures (Figure 3f; Table S3) and negatively affected, although weakly, by proportion of SNHs (Figure 3g; Table S3).

The flowering plant FRic was equally well explained by three models (Table 5), which did not include any landscape parameter. The results highlight a positive effect of organic vineyard management interacting with increasing wild bee FRic (Figure 3h; Table S3). Wild bee FRic increased flowering plant FRic (Figure 3i; Table S3), whereas increasing inter-row vegetation cover had a weak negative effect on the FRic of flowering plants (Figure 3j; Table S3).

3.2 Unique characteristics of bee and flowering plant functional traits in Austrian and South African vineyards

In both countries, most flowering plant taxa in the vineyard inter-rows had white or yellow flowers with radial symmetry and belonged to the Asteraceae and Fabaceae family. The nectar of most of the flowering plant species was totally hidden in the flower, and flag, ray, disk, and ray–disk blossoms were the predominant flower shapes. In Austria, the inter-row vegetation mainly flowered from early summer to mid-summer, but some vineyards were covered by a high proportion of plant species that potentially flower the whole vegetative period (e.g., Stellaria media, Veronica persica). In South Africa, inter-row flowering was predominantly in spring (see Table S4 for CWMs per vineyard).

In organic vineyards in Austria, there was increased abundance of plant species with bell or funnel flowers (coef = 0.12), but decreased abundance of plants with yellow flowers (coef = −0.08) (Figure 4a). Higher inter-row vegetation cover increased wind-pollinated plant species or plants where flowers solely provided pollen (coef = 0.1). Plants with radial flowers (coef = −0.07), as well as flowers providing only pollen, and wind-pollinated plants (coef = −0.08) were negatively associated with increasing wild bee FRic. Fabaceae with flag blossoms were positively associated with high wild bee FRic in Austrian (coef = 0.06) and South African vineyards (coef = 0.15) (Figure 4b).

The bee assemblages in both countries were characterized by a high proportion of ground nesting (Austria: 68%, South Africa: 78%), solitary (Austria: 69%, South Africa: 59%), and polylectic wild bee species (Austria: 85%, South Africa: 100%). However, in Austrian vineyards, eusocial wild bees were more abundant (over 60% of the specimens) than in South Africa (11% of the specimens). In South Africa, the sociality of seven species (59 specimens) and the lecty of 2 species (15 specimens) remain unclear. There were 13 oligolectic species (89 specimens) reported from Austrian inter-rows, but none were documented here for South African vineyards. Wild bees ITD ranged between 1.58 and 2.61 mm, and the mouthpart length varied between 2.18 and 4.71 mm (see Table S4 for CWMs per vineyard). In comparison with South Africa, the CWM of the Austrian bee assemblages was characterized by smaller (Austria: 2.06 ± 0.31, South Africa: 2.35 ± 0.22) but longer tongued bee species (CWM of proboscis length, mean ± SD per study vineyard: Austria: 3.23 ± 0.79, South Africa: 2.99 ± 0.41).

The fourth-corner analysis of Austrian vineyards (Figure 5a) showed low correlation between the analyzed wild bee traits and the environmental variables. An increasing frequency of blue flowers in the inter-rows correlated with a lower abundance of solitary bee species (coef = −0.06). Purple flowers decreased abundance of bees collecting pollen on the abdomen (coef = −0.04), but benefitted crop-collecting species (coef = 0.05). Yellow flowers (coef = −0.06) and flowers with radial symmetry (coef = −0.06) were associated with lower abundances of long-tongued wild bee species (thus increased short-tongued species). Vineyard management and the proportion of woody structures were associated with a higher abundance of bee species with different social organizations. Solitary species were likely more abundant in organic (coef = 0.05) than in conventional vineyards, but tended to decrease with increased inter-row vegetation cover (coef = −0.05) and woody structures (coef = −0.15) in the surrounding landscape.

The wild bees in South African vineyards showed different associations with flower traits, management intensity, and landscape composition compared with the Austrian vineyards (Figure 5b). Long-tongued bees were more abundant in vineyards where plant with ray–disk flower heads (e.g., Asteraceae; coef = 0.3), plants with flower associations with totally hidden nectar (coef = 0.16), and yellow flowers (coef = 0.11) were present. In contrast, ITD decreased with the presence of plants with flower associations with totally hidden nectar (coef = −0.46). At the landscape scale, a higher proportion of SNHs (coef = −0.06) and woody structures (coef = −0.03) slightly decreased belowground nesting species. These species showed a positive association with plants with radial flower symmetry (coef = 0.12) and an increased mean plant species richness (coef = 0.06). The abundance of bee species collecting pollen on a scopa on legs increased with a higher mean plant species richness (coef = 0.05) but was negatively correlated with plants with completely hidden nectar (coef = −0.08).

4.1 Common drivers of bee diversity in vineyards across countries

The interacting effect of plant FRic and organic farming on bee FRic suggests that maintaining diverse plant communities in inter-rows may also enhance other biodiversity-friendly practices. Compared with organic vineyards, conventional vineyards had higher bee FRic at low levels of flowering plant FRic. This indicates bees may be able to exploit resources better at low levels of flowering plant FRic, likely due to trait matching. For example, we showed that in Austria, yellow flowers are more frequent in conventional vineyards and are beneficial for short-tongued bees. In South Africa, yellow flowers promoted bees with a tibial scopa and long-tongued bees, which contributes to increased wild bee FRic. However, a direct positive effect on yellow flowers by conventional vineyard management in South Africa was not detected.

Functional richness and species richness are orthogonal to each other, FRic either increases or remains the same with increasing species richness (Mason et al., 2005; Schleuter et al., 2010). In this study, FRic increased with higher species richness, which explains the similar results of wild bee species richness and FRic of wild bees. The positive effect of higher vegetation cover on bee species richness was also reported for wild bees in other crop systems (Nicholson et al., 2017; Shuler et al., 2005) and vineyards (Kratschmer et al., 2019), as well as for other beneficial organisms (Buchholz et al., 2017; Fiera et al., 2020; Hall et al., 2020; Winter et al., 2018). Benefit is derived from undisturbed soil conditions for eusocial species (Williams et al., 2010), which are predominantly ground nesting species in our study. Additionally, eusocial bee reproduction often depends on a single fertile female for the whole colony, which makes them more vulnerable to soil disturbance (Kratschmer et al., 2018). The positive effect of undisturbed soil conditions on wild bee FRic was not detected in models that combined both countries. Possibly because the 23 ground nesting and eusocial species belong to the Halictidae family (Halictus spp. and Lasioglossum spp.) and their contribution to FRic is low, as most are very small, short-tongued species that collect pollen with a tibial scopa.

In the current study, we did not see evidence that higher proportions of SNHs compensate for the negative effects of intensive vineyard management. However, higher proportions of woody structures at the landscape scale increased wild bee FRic in the study vineyards in both countries, but this has to be interpreted with care, because the characteristics of the woody structures vary across the study regions. In Austria, large woody structures represent seminatural structures of species-rich thermophilic oak and oak-hornbeam forests partly belonging to the European Natura 2000 protected area network (Burgenländische Landesregierung, 2018). Although only one third of the Austrian bee species found in vineyards were eusocial, this group contributed to functional richness, with the species varying in nesting type, body size, mouthpart length, and pollen collection with different structures (e.g., Bombus spp vs. Lasioglossum spp.). In the Cape Floristic Region, natural habitats (SNHs) consist of low-growing shrubs with few trees. The woody structures recorded here consisted mainly of non-native invasive species, which probably do not provide habitat for many of the native bee species, as plant species richness and ground-dwelling arthropod diversity decline under stands of alien trees (Magoba & Samways, 2012; Richardson et al., 1989; Schoeman & Samways, 2011). However, dead-wood nesting species, such as Xylocopa rufitarsis, Lithurgus spiniferus, or Allodape tridentipes, may find suitable nesting sites in these wooded areas. Additionally, certain flowering alien trees, such as Acacia saligna and Eucalyptus cladocalyx, can provide floral resources to local pollinators (Gibson et al., 2013; De Lange et al., 2013).

The models revealed a weak negative effect of SNHs on wild bee species richness. However, this result should be interpreted with caution (Kratschmer et al., 2018), as either a pull effect due to good habitat quality or, in contrast, a generally poor habitat quality of the SNHs could be responsible for this result. As Kehinde and Samways (2014a) found higher bee and flowering plant diversity in natural fynbos sites compared with vineyards, the pull-effect explanation is probably more likely in South Africa. It also explains why only 28 bee species were documented in vineyards in a country known as a bee diversity hot spot. We cannot underpin the pull-effect explanation directly for Austria, but the importance of SNHs over vineyards as habitat for wild bees was recently reported (Pascher et al., 2020). Another reason for the low species diversity could lie in the different sampling methods and frequencies in the two countries. Although the combination of sampling methods used in South Africa (trapping, transect sampling) should provide a representative sample of the bee community (Prendergast et al., 2020; Schindler et al., 2013; Vrdoljak & Samways, 2012), the very short activity period of many bee species in combination with the short sampling period may have missed many bees in the South African vineyards. Also, different sampling methodology used in the two countries may have influenced the traits recorded, especially those selected from the colored pan traps (McCravy et al., 2019).

4.2 Common drivers of flowering plant diversity in vineyards between countries

Functional richness of flowering plants increased with wild bee FRic, organic management, and lower vegetation cover. The positive effect of higher bee species richness on insect-pollinated plant species richness was also shown by Papanikolaou et al. (2017) across 24 European study sites. Furthermore, they also did not detect any effect of landscape composition on functional richness of insect-pollinated plants. It seems contradictory that less intensive management associated with higher vegetation cover results in lower insect-pollinated plant FRIc. Most likely the effect is related to the higher proportion of grasses dominating vineyard inter-rows with permanent vegetation cover, as reported from vineyards across Europe (Hall et al., 2020). In accordance with the intermediate disturbance hypothesis (Connell, 1978), infrequent and alternating tillage in vineyard inter-rows increases species richness. Just like wild flower strips, which tend to be dominated by grasses without disturbance (Schmid-Egger & Witt, 2014), vineyard inter-rows should also be occasionally tilled to increase cover by ruderal plants (Hall et al., 2020) and less competitive species (Gago et al., 2007), which contribute to the resource provision by a diverse flowering plant community.

The negative effect of conventional management on vascular plant species richness due to the use of herbicides in vineyards was shown for Italian vineyards (Nascimbene et al., 2012) and in an experimental field trial in California (Sanguankeo & León, 2011). Without the use of herbicides in conventional vineyards, Bruggisser et al. (2010) and Kehinde and Samways (2014a) did not find any significant difference between management type on plant species richness. It is surprising that in this study, where winegrowers did not use any herbicides in conventional vineyard inter-rows, organic viticulture still increased FRIc of flowering plants.

Despite the fact that Plantago lanceolata was the only flowering plant species documented in inter-rows in both countries, several taxa occur in both countries. This larger overlap is due to the high proportion of alien species (43% of all flowering plants) found in South African vineyards: Echium plantagineum, Medicago polymorpha, Helminthotheca echioides, Raphanus raphanistrum, Trifolium angustifolium, and Vicia benghalensis. Raphanus raphanistrum and V. benghalensis were probably introduced through seed mixtures.

4.3 Unique characteristics of bee and flowering plant functional traits in Austrian and South African vineyards

Apart from the previously discussed effects of flower traits on wild bee functional traits, the detailed trait analysis provided additional and country-specific patterns, which can be explained by trait matching. However, due to the low correlation between traits and environmental variables (especially in Austria) these results should be interpreted with care. It is important to keep in mind that only plant occurrence data, but not flower abundance or cover, were available for this analysis. Thus, clearer trait matching patterns may be derived from plant–pollinator interaction data; however, those data were not available. Further, it should be noted that bees flower color perception is shifted toward the UV light spectrum. In this study, flower colors, as perceived by humans, were used, due to missing information about flower UV reflectance for >25% of the plant species considered in our study. As pointed out by Burr and Barthlott (1993) and Burr et al. (1995), there exists a relationship between these different color perceptions, which supports the selection of flower colors as perceived by humans used for this analysis.

In Austria, yellow flowers and radial flower symmetry, which are predominantly plants belonging to the Asteraceae family, benefitted short-tongued bee taxa. The nectar of Asteraceae flowers is not easily available, but hidden in each flower, which excludes other, less reliable pollinators such as flies from collecting it. Short-tongued bees (e.g., Andrena, Lasioglossum, or Halictus species, Figure S2) are matched with these flower traits and able to collect nectar and pollen from the flowers efficiently (Mani & Saravanan, 1999). In contrast to the Austrian results, long-tongued bee species benefitted from Asteraceae flowers in South African vineyards. Although no flower abundance data are available, we cannot rule out that yellow Asteraceae predominantly flowered in the South African inter-rows and that long-tongued bee species (e.g., Xylocopa rufitarsis, Tetraloniella junodi) collect mainly pollen from these flowers (Mani & Saravanan, 1999). However, it is puzzling that although Fabaceae species flowered in most of the South African inter-rows, tongue length was not associated with the specific flower morphology or nectar accessibility, especially as these plants would seem to be ideal for long-tongued bee species. We propose the following explanations: Firstly, the reported bee species with longer mouthparts (> 4 mm) may not prefer or be adapted to collecting floral resources from the introduced Fabaceae plants. Secondly, the 2010 dataset also includes plant–pollinator interactions (Kehinde & Samways, 2014c), which reveals that Fabaceae species were most frequently visited by Apis mellifera. This could indicate either a niche shift of other bee species to other plant species due to competition with the honey bees, or that honey bees are more adaptable to using introduced plant species compared with the other wild bee species (Mallinger et al., 2017).