Pollinator enhancement in agriculture: comparing sown flower strips, hedges and sown hedge herb layers in apple orchards

In intensive agricultural landscapes semi-natural habitats for pollinators are often limited, although willingness to establish pollinator habitat is increasing among farmers. A common pollinator enhancement measure is to provide flower strips, but existent or improved hedgerows might be more effective. In this study, we compare the effectiveness of three pollinator enhancement measures at edges of conventional apple orchards: (i) perennial flower strips, (ii) existent hedgerows, and (iii) existent hedgerows complemented with a sown herb layer. We used orchard edges without any enhancement as control. The study took place over three consecutive years in Southern Germany. Wild bee abundance and species richness were highest in flower strips followed by improved hedges. Hoverflies were also most abundant in flower strips, but not more species rich than at control sites. Wild bee but not hoverfly community composition differed between control and enhancement sites. The overall pollinator community included only few threatened or specialized species. Flower abundance was the main driver for wild bee diversity, whereas hoverflies were largely unaffected by floral resources. Pollinator enhancement had neither an effect on the abundance or species richness within the orchards nor on apple flower visitation. Perennial flower strips seem most effective to enhance wild bees in intensive agricultural landscapes. Additionally, flower-rich hedgerows should be promoted to complement flower strips by extending the flowering period and to increase connectivity of pollinator habitat in agricultural landscapes.


Introduction
In intensively managed agricultural landscapes, semi-natural habitats are often scarce. The resulting lack of food, shelter and nesting sites is one of the main drivers of the decline in insect abundance and species richness (Dicks et al. 2021;Habel et al. 2019;Seibold et al. 2019). With increasing public awareness of the importance of insects in ecosystems, farmers increasingly apply conservation measures, especially for flagship taxa like bees. A prominent measure is to establish flower strips (Wood et al. 2017), which are shown to enhance bee diversity in different agricultural settings (Ganser et al. 2021;Haaland et al. 2011;Lowe et al. 2021), e.g., in blueberry fields , apple orchards (Campbell et al. 2017) and cereal fields (Buhk et al. 2018). Also hoverflies, another important pollinator taxa (Rader et al. 2016), benefit from flower strips (Tschumi et al. 2016). Flower strips can increase diversity of wild bees and hoverflies even in the wider landscape (Buhk et al. 2018;Jönsson et al. 2015) and thus seem to be a suitable conservation measure to enhance pollinator diversity in agricultural landscapes. However, flower strips do not cover the requirements of all potential pollinator species living in agricultural landscapes: many bees, e.g., Andrena (sand bee) species, are active before flower strips start to bloom and thus do not benefit from the additional floral resources (Campbell et al. 2017;Ouvrard et al. 2018). Furthermore, flower strips provide only limited nesting possibilities for aboveground-nesting bees, and limited larval food sources for hoverflies (Cole et al. 2020;Cope et al. 2019). Moreover, annual as well as perennial flower strips are commonly not persistent at one location, which likely limits their influence on local pollinator population growth (Pywell et al. 2011;Schmidt et al. 2020).
Hedges might be more functional to enhance pollinator diversity in agricultural landscapes (Kremen and M'Gonigle 2015;Morandin and Kremen 2013;Pfister et al. 2017;Ponisio et al. 2016). They provide habitats for many plants and animals (Dondina et al. 2016;Hinsley and Bellamy 2000;Staley et al. 2013) and are typically composed of a herb, shrub and possibly a tree layer (Maudsley 2000). The resulting plant diversity makes hedges attractive for bees and hoverflies as they find floral resources over the whole growing season with shrubs and trees blooming in spring and herbs in summer (Garratt et al. 2017;Morandin and Kremen 2013;Schirmel et al. 2018). Furthermore, hedges provide nesting sites, shelter and overwintering habitat for bumblebees, aboveground-nesting solitary bees and hoverflies (Branquart and Hemptinne 2000;Kremen and M'Gonigle 2015;Osborne et al. 2008;Rodríguez-Gasol et al. 2020, but see Lye et al. 2009).
Though of high functional importance, the herb layer of hedges is usually dominated by grasses due to eutrophication or is absent because agricultural area extends to the shrub zone (Maudsley 2000;Staley et al. 2013). A fully developed herb layer is not only itself an attractive pollinator habitat (Garratt et al. 2017;Pfiffner et al. 2018), it is also crucial for the full seasonal floral provision of hedges. We therefore assume that sowing of a flowering edge at hedges with species-poor or missing herb layer, may increase the habitat quality for pollinators.
Pollinator enhancement is of special interest in orchards as fruit production needs insect pollination for optimal yield (Garratt et al. 2014;Grab et al. 2017;Samnegård et al. 2019). Hoverflies additionally act as biological control of aphid populations, a main pest in orchards (Almohamad et al. 2009;Ramsden et al. 2017;Tschumi et al. 2016). To identify the most effective measures for enhancing the diversity of bees and hoverflies, we conducted a study in apple orchards managed according to Integrated Pest Management (IPM, Reganold et al. 2001) in Southern Germany. We investigated orchards bordered by one of 1 3 three enhancement measures, namely flower strips, existent hedges or existent hedges complemented with a sown herb edge, in comparison to unenhanced orchards. We analyzed bee and hoverfly abundance, species richness and community composition of three consecutive years after flower strip and herb edge establishment. We tested the following hypotheses: (1a) The abundance and species richness of bees and hoverflies are higher in flower strips, unimproved hedges and improved hedges than in unenhanced orchard edges and the bee and hoverfly communities differ between the enhancement measures. (1b) The abundance of flowers drives the differences between enhancement types. (2) The abundance and species richness of bees and hoverflies is higher in the understory of orchards bordering enhancement measures compared to the understory of control sites. (3) The visitation rate of apple flowers is higher in orchards bordering enhancement measures compared to control sites.

Study region and orchard site selection
The study took place at the Lake Constance region in southeast Baden-Württemberg (Germany, temperate climate) ( Fig. 1; details on study sites see Annex Tab. 1), one of the largest apple-growing regions in Germany (Garming et al. 2018). We selected 18 sites located in private owned IPM-apple orchards. The sites were at least 1 km apart (except one case). This distance is larger than the typical foraging range of most wild bee species (Hofmann et al. 2020), exceeds the typical daily flight distance of hoverflies (Rodríguez-Gasol et al. 2020) and therefore allowed to study separate pollinator communities. Pollinators were shown to respond to the landscape context at 500 m radius (Kleijn and van Langevelde 2006;Meyer et al. 2009). To minimize the effect of other land-use types, all sites were Fig. 1 Study site locations at Lake Constance 1 3 selected to be surrounded by a large proportion of apple orchards (53 ± 13%) and a low amount of forest (6 ± 9%) in 500 m radius.

Enhancement measures
We selected five apple orchard sites for each of the three enhancement measures (see below) and additionally five orchards without any bordering semi-natural habitat as control sites (photos see Annex Fig. 5). One site each of the measures "flower strip" and "hedge" had to be abandoned, reducing the number of sites to 18 sites. The abandonment was caused by land owner decisions to either exit the study due to internal conflicts or to add an unintended flower strip to a hedge without prior consultation.

Flower strips
Flower strips (n = 4) were sown on a 25 × 2.5 m area adjacent to the apple orchards at the beginning of April 2018. A perennial seed mixture containing native wild and cultivated plants of regional provenance was selected (see Annex Tabs. 2, 3).

Unimproved hedges
In this category an existent hedge (n = 4) bordered the orchard. The vegetation density varied from sparse to dense, and the shrub composition was variable, but all hedges, were at least 10 years old. The entire length of the hedges varied from ca. 40 m to 170 m (mean 124 ± 49 m) with none of them being connected to other habitat types. None of the hedges had a noteworthy herb layer. Where herb layers were present, these were narrow and dominated by grasses.

Improved hedges
Along five existent hedges bordering the orchards, wildflower strips were sown at an area of 0.5 × 25 m in late April 2018. We selected a perennial wildflower mixture of regional provenance adapted for edges containing only native plant species (see Annex Tabs. 2, 3). In 2020, only four of the five sites could be used because at one site the orchard was turned into a field and the wildflower strip was destroyed.

Sampling design
Sampling was conducted at least once a month from late March/early April to late August/ early September in three consecutive years (2018,2019,2020). During apple bloom 2018 and 2019 sampling effort was increased to allow the separate analysis of pollinators on apple flowers (three samplings during the approximately two weeks of bloom). In sum, we carried out 33-37 samplings per site. At each site and sampling day, we selected three 1m 3 plots on a predefined transect of 25 m length within the available enhancement measure (hedge, sown hedge herb layer, flower strip) and in the orchards (one transect at the orchard edge and one 20 m inside the orchards). Plots were regarded as three-dimensional to include vertical vegetation like the hedge shrubs or the apple trees (apple flowers were 1 3 sampled as any other flowering plant). The orchard edge observation at the control sites (excluding apple flowers) was used as control in the analysis.
At each plot, we observed bees and hoverflies landing on flowers within 5 min observation time and caught them for later identification in the lab (identification literature see Annex Tab. 4). Individuals with unique characteristics were identified in the field. Bees were categorized after identification in the following subgroups following Westrich (2018): social wild (bumblebees and other social species), solitary (incl. cuckoo bees), aboveground-nesting, oligolectic (specialized at least on plant family level), threatened (at least classified as near threatened on current Red List of German bees by Westrich et al. 2012) and honeybees. Additionally, we noted for each plot at each observation day the plant species in bloom and the corresponding number of individual flowers. For plant species with tiny or composite flowers, the number of flowers was afterwards aggregated to receive floral units based on the average number of flowers per flower head (see Annex Tab. 5 for details).
Each plot was placed on areas with maximal flower cover and diversity at each sampling day individually to depict the maximum pollinator abundance, species richness and community completeness. In homogenous vegetation, this plot selection resulted in a conservative estimate, whereas in heterogeneous vegetation it resulted in an optimistic estimate. If no flowers were detectable despite intensive search, we included plots with zero plants and zero insects in the dataset. We placed the plots at sunny spots if possible. Observations took place on warm and calm days without rain (25.4 ± 4.7 °C, min. 13 °C; wind speed 2.1 ± 2.1 m/s, max. 11 m/s). We sampled each site at varying times of the day to avoid time-of-day effects. At improved hedges, we sampled three plots in the hedge and three in the sown herb layer. To receive the same number of plots at each enhancement measure in the statistical analysis for all enhancement measures, we afterwards once randomly drew three plots from the resulting six plots at improved hedges (excluding plots with zero plants and zero insects if three plots with flowers were available in order to match the selection of flower rich plots as generally conduced in all other enhancement measures).

Statistical analysis
We pooled the pollinator observations for abundance and species richness for each site and year and calculated abundance and species richness for each pollinator group. To account for the unequal numbers of samplings per year, we divided the abundance and species richness values by the number of samplings per year resulting in mean values per observation (53 replicates in total). We used Generalized Linear Models ('GLM') with a gamma distribution with log-link to test the relation of each pollinator subgroup's abundance or species richness with enhancement measure and year. As the gamma distribution does not cover zeros, we added + 0.1 to all values to allow model fit. We calculated Moran's I to test for spatial autocorrelation (Dormann et al. 2007, see Annex Tab. 6) and in two cases switched from glm to glmmTMB with a structured variance-covariance matrix (exponential autocorrelation) from the R-package glmmTMB (Brooks et al. 2017) to account for spatial autocorrelation (models for abundance of all wild bees and of solitary bees). To compare the effects between enhancement measures, we calculated pairwise comparisons with the R-package lsmeans (Lenth 2016) with Bonferroni-Holm-adjustment of p-values.
To test for community differences, we used permutational multivariate analyses of variance ('adonis'; Anderson 2001) testing for enhancement measure and year using Morisita-Horn dissimilarities on n = 10.000 permutations and species' abundances per year. For 1 3 the illustration of the species variation, we used non-metric multidimensional scaling (2-dimensional) (metaMDS) based on bray dissimilarities (Minchin 1987). For this, we combined the observations of the three years per site and omitted the site that was sampled in only two years instead of three. Species composition was tested for spatial autocorrelation using a Mantel test ('mantel'; Legendre and Legendre 2012; results see Annex Tab. 6); no autocorrelation was apparent.
To validate direct and indirect effects between predictors, we constructed structural equation models (SEM; 'psem'; Lefcheck 2016), one for each of social wild, solitary bees and hoverflies. The SEMs contained five component GLMs, each with a gamma distribution with log-link: (1) bee abundance in relation to enhancement measure, flower abundance, flower species richness and year, (2) bee species richness in relation to enhancement measure, flower abundance, flower species richness, year and bee abundance, (3) flower abundance in relation to enhancement measure and year, (4) plant species richness in relation to the enhancement measure and year and (5) a correlated error operator (% ~ ~ %) connecting plant species richness and abundance of flowers. Flower abundance (= number of floral units) was log-transformed to account for saturation effects at high flower numbers. Categorical variables (enhancement measure, year) were transformed into dummy variables.
To answer whether potential effects reached into the orchards, we applied GLMs with a gamma distribution with log-link to test the relation of pollinator abundance or species richness (social wild bees, solitary bees, hoverflies) with enhancement measure, year and location in the orchard (edge or inside) and the interaction between enhancement measure and location. For this we pooled the plots at the edge and the inside of the orchards per site and year.
To test the relation of apple flower visits with the enhancement category, we used a data subset consisting only of observations of blooming apple flowers from the plots at the edge or inside the orchards. We applied Generalized Linear Mixed-effects Models of the negative binomial family ('glmer.nb') with a control parameter (optimizer = "nlminbwrap", nAGQ = 0.) and the following predictors: enhancement category, the location in the orchard (edge or inside) plus their interaction, and the year. We applied the model on the following subgroups of pollinators: all wild bees, all bees (incl. honeybees) and all insects. The sites were included as random effect.
We conducted all analyses with R version 3.5.3 (2019-03-11) and inspected the residual distribution of all regression models with the DHARMa R-package (Hartig 2021). The residuals conformed to the expectation in all models.

Results
Overall, we observed 4239 honeybees, 3194 wild bees from 106 species and 1425 hoverflies from 29 species (see Annex Tabs. 7,8,9). The abundance and species richness of both solitary and wild social bees was highest in the flower strips ( Fig. 2; Annex Tab. 10). Especially social wild bee abundance and species richness was increased in flower strips ( Fig. 2; Annex Tab. 11). Oligolectic bees were most abundant in improved hedges, but more species-rich in flower strips (Table 1). Red-listed bees were most abundant and species-rich in flower strips. Honeybees and aboveground-nesting bees showed no preference for one of the enhancement measures (Table 1). Hoverflies were most abundant in the flower strips, but species richness was not different between the enhancement measures and the control ( Fig. 2; Annex Tab. 10).
The SEM analysis revealed that all enhancement measures increased flower abundance and species richness compared to control sites (Fig. 4, Annex Tab. 11). Flower abundance was similar among the enhancement measures, whereas flower species richness was higher in flower strips and improved hedges than in hedges. Flower abundance increased both social and solitary wild bee abundance and indirectly species richness. The presence of flower strips increased social wild bee abundance directly. No direct influences of enhancement measures were identified on solitary bees. Hoverflies were not influenced by floral resources, but were driven by year-to-year variation.
Within orchards, the abundance and species richness of social and solitary bees was not influenced by the presence of any enhancement measure (Table 2). At orchard edges, solitary bee abundance and species richness was higher than inside the orchard; regardless of presence of enhancement measures. Hoverfly species richness was higher in hedges compared to flower strips and improved hedges, but not in comparison to the control ( Table 2).
The number of apple flower visits by wild bees, wild bees and honeybees, or all insects was not influenced by the presence of enhancement measures (Table 3). Honeybees made up 83% of the apple flower visits, wild bees 9% and non-bees 8%.

Enhancement measures
Pollinator enhancement measures increased pollinator abundances and species richness compared to non-enhanced orchards. Among the three measures flower strips were most  and hedges least attractive. Bees, especially social wild bees responded stronger than hoverflies to the enhancements. Therefore, although all enhancement measures contributed to the conservation of pollinators, flower strips seem most promising for overall wild bee conservation. Although our number of replicates per measurement type was limited, the high number of observations per site and year is reassuring. This representativeness of our sampling was supported by the species accumulation curves, which flattened within the available number of sites and enhancement measures (see Annex Fig. 6).

Bees
Flower strips provided the most attractive floral resource for wild bees Buhk et al. 2018;Haaland et al. 2011;Jönsson et al. 2015, but see Wood et al. 2017), especially to social wild bees. This can be explained by several factors: (1) The flowering time coincides with highest population densities of social wild bees (Westrich 2018). The flower strips bloomed from early to late summer, the months in which social wild bees are most abundant (Westrich 2018), which may also explain the occurrence of otherwise rare and threatened bumblebee species.    bee abundance and species richness due to the species-area relationship and the chronosequence of solitary bee species . (3) The flower strips included most attractive flowering species (Rosa García and Miñarro 2014;Sutter et al. 2017). Additionally, flower strips included plant families for several oligolectic bee species e.g. Apiaceae, Asteraceae, Boraginaceae, Brassicaceae, Fabaceae and Resedaceae. Hedges were less attractive for wild bees than flower strips. Especially social wild bees occurred in lower number and species richness. This is mostly attributable to the comparatively low flower abundance in the hedges, which mirrors the low year-round-availability and patchiness of floral resources. All hedges showed several months with low availability of floral resources, partly even in the spring months. Furthermore, some hedges were dominated by shrubs, whose flowers are relatively unattractive to wild bees, like Euonymus europaeus or Viburnum sp. (Westrich 2018, see Annex Tab. 12). Other hedges with larger amounts of e.g. Salix sp., Prunus spinosa or Rubus sp. attracted more bees, partly even oligolectic species (mostly on Salix). Though the hedges turned out less attractive as foraging habitat in our analysis, they may still play an important role for local bee populations. Social wild bees like bumblebees depend on early-flowering high-quality forage to build up their populations and hedges furthermore potentially provide nesting and hibernation possibilities (Cole et al. 2020;Hopfenmüller et al. 2014;Morandin and Kremen 2013;Scheper et al. 2015). Several oligolectic bee species depend on Willows, which often grow in hedges.
Improving hedges with a sown herb layer increased bee abundance and species richness by extending the flowering period into the summer months, but the attractiveness of improved hedges varied considerably due to differences in plant establishment success. This was partly due to the seed mixture, which lacked fast-growing annual plants, resulting in low plant cover in the first year. Furthermore, the growth conditions along the hedges were partly not adequate for the sown plant species and the narrow width of the sown herb layers facilitated growth and dominance of grasses and other unwanted plants (Graham et al. 2018). Apple trees in general reach the border of the orchard, which is where hedges typically grow, leaving only limited space for herbs. The establishment of herb layers is only promising if wide enough sunny space is available. Hedges with herb layers are thus not per se less attractive for pollinators than flower strips (Pfiffner et al. 2018), but are difficult to establish in the specific settings of intensive apple orchards.
Bee communities differed between the enhancement measures probably mostly due to differences in flowering periods. The flower strips provided most floral resources in summer, whereas peak flower in the hedges and the control orchards was in spring. Thus the flower strip community included more later-flying species, but missed out early spring flying species. Differences in community composition were expected from other studies, which examined pollinators in different semi-natural habitats (Pfiffner et al. 2018;Sanchez et al. 2019). Aboveground-nesting bees, which potentially rely on the reservoir of nesting sites in hedges (Morandin and Kremen 2013), were to our surprise most abundant in the flower strips (Ganser et al. 2021).

Hoverflies
Flower strips attracted the highest abundance of hoverflies (Haenke et al. 2009;Sutherland et al. 2001), followed by improved hedges, but species richness did not differ between enhancement measures and control. Adult hoverflies require nectar and pollen as food source (Almohamad et al. 2009) and diverse floral resources show positive influences on hoverfly diversity (Meyer et al. 2009;Sutherland et al. 2001). However, in our study only year, not flower abundance and species richness, explained the abundance and species richness of hoverflies. The occurrence of several easily accessible flowers from Apiaceae and Asteraceae in the flower strips and sown hedge layers, which hoverflies prefer, may play a role (Branquart and Hemptinne 2000). Yet, environmental factors like the presence of their larval food sources seem to be more important for hoverfly abundance than floral resources (Alignier et al. 2014;Schirmel et al. 2018). With more than 85% of the hoverflies in our study being aphidophagous, which is common in landscapes of high levels agricultural intensification (Inclán et al. 2016;Pfister et al. 2017), aphid population variability might be the driver of our observations. Hoverflies thus do not necessarily profit from the same measures as bees (Cole et al. 2020;Jauker et al. 2009). Several studies have shown that hedges are attractive habitat for hoverflies-even more attractive than herbaceous vegetation -, not only as foraging but also as shelter and overwintering habitat (Haenke et al. 2014;Pfister et al. 2017;Schirmel et al. 2018). It may be that our sampling method did not adequately reflect the role of hedges as shelter and overwintering habitat as hoverflies are highly mobile insects (Almohamad et al. 2009).

Orchard understory flower visitation
Within orchards the bee communities were not influenced by the enhancement measures at the orchard edges. This is not surprising because, in most cases, apart from early spring the orchards provided only scarce floral resources. Thus, bees had no reason to fly into the orchards . Mowing reduces the flower abundance in orchards which may explain why the bees rarely use the orchard interiors but stay at the edges. The hoverfly communities visiting understory flowers were also not influenced by enhancement measures. Floral resources inside the orchards are probably of little importance for hoverflies as they are no central foragers but place their eggs directly next to or in the larval food source (Almohamad et al. 2009). They can store food reserves for egg production (van Rijn et al. 2006). Thus unlike bees they do not constantly visit flowers but spend much of their time searching for oviposition sites (Rodríguez-Gasol et al. 2020). Our results suggest that flower strips and hedges are ineffective tools for increasing pest regulation by hoverflies in apple orchards, at least on the local scale. Some studies found spillover of hoverflies from semi-natural habitats to agricultural fields (Haenke et al. 2009;Sutherland et al. 2001), others not (Inclán et al. 2016;Morandin and Kremen 2013). However, the abundance of adult hoverflies gives only limited information on the actual level of pest regulation.

Apple flower visitation
The share of wild pollinators observed on apple flowers was low, whereas honeybees dominated the apple flower visits. Osmia bicornis was the most common apple-visiting wild bee, probably because it is managed for this purpose in the region. Other studies found much higher proportions of wild bees among apple flower visitors (Hutchinson et al. 2021). Reasons for this difference may be higher proportions of semi-natural habitat in the surrounding landscape in other studies or the intensity of honeybee-keeping in our study region (Bartholomée et al. 2020;Klein et al. 2012;Mallinger and Gratton 2015;Marini et al. 2012). Neither flower strips nor hedges increased wild pollinator visits of apple flowers. For the flower strips this is probably due to a phenological mismatch between the flight period of the main apple visiting bee species, like early-flying Andrena and Osmia, and the bloom of the flower strip (Campbell et al. 2017). Hedge shrub bloom would phenologically better fit the flight period of apple flower visiting bee species and, indeed, hedges were found to increase wild pollinator numbers e.g. in adjacent strawberry and tomato fields (Castle et al. 2019;Morandin and Kremen 2013). However, strawberries and tomatoes bloom later than apples, when e.g. social bees are naturally more abundant (Westrich 2018). In a current review, hedges were found to have no significant effect on pollination services ).

Conclusion
In summary, perennial flower strips are a valuable conservation strategy to enhance bee diversity in intensive agriculture areas. Additionally, high-quality hedges should be promoted as they provide early flower resources and pollen sources for several oligolectic bee species. To increase the positive effect on specialized and threatened bee and hoverfly species and crop pollinators a long-term maintained network of flower strips complemented with other enhancement measures may be necessary (Boetzl et al. 2021;Buhk et al. 2018;Sutter et al. 2018).

3
Funding Open Access funding enabled and organized by Projekt DEAL. The research project was funded by Bayer Crop Science. Bayer Crop Science was involved in the development of the study design. Apart from that, Bayer Crop Science was not involved in the research process.