The pollination experiments were performed within agricultural areas at nine different study sites in south-west Germany (see Figure 1A and Supplementary Table S1 ). The study sites were part of the BienABest project (www.bienabest.de; see Neumüller et al., 2020 and Neumüller et al., 2021 for the BienABest study design and establishment of study plots). The BienABest project aimed to safeguard the ecosystem service of pollination and to enhance wild bee diversity in agricultural landscapes. Flower strips were established as conservation measures at various sites in Germany, accompanied by extensive wild bee monitoring during the years 2018 to 2022.

The study sites extended over 180 km latitudinally and 175 km longitudinally. The landscape of each site was an agricultural matrix of arable land and/or intensively-used grassland and incorporated bodies of semi-natural grassland. At each study site, 0.3-ha plots were established on various habitats within a 500-m radius. Two plots each of the two different habitat types were used at each study site, resulting in 36 plots (2 x 2 plots at each of the nine study sites) for the pollination experiments and observed bee data. The pollination experiments were performed on study plots with two different habitat types within each study site: near-natural habitats and perennial wild flower strips ( Figure 1C ). Near-natural habitat plots were placed in extensively managed grassland such as calcareous grassland, mostly in nature conservation areas. The semi-natural grasslands were characterized by a high diversity of native flowering plants and were rich in diverse structures (e.g., open soil for nesting or woody vegetation). The habitats were managed by occasional grazing by sheep and met the needs of a diverse spectrum of naturally occurring flora and fauna. The wild flower strips were established on tilled arable land in 2017 by using a seed mixture consisting of ca. 50 plant species (autochthonous seeds, provided by Rieger-Hofmann, Blaufelden, Germany). The seed mixture consisted of annual and perennial plant species that are regionally native or cultivated plant species and included the study plant species used in the pollination experiments. Half of each wild flower strip was mown once a year at peak bloom in June as a measure to increase plant diversity and to induce a second bloom, prolonging the blooming season to promote pollinators. When the experiments of this project were performed in the year 2020, the wild flower strips were well established, with mainly perennial plants flowering. On average, 38 plant species (minimum 26, maximum 42 species) from the seed mixtures were established in each flower strip. The flower cover varied between seasons but was well pronounced when pollination experiments were performed.

The four different bee-pollinated plant species ( Figure 2 ) were common species occurring at all study sites and were part of the established seed mixture. The species differed in floral cues and phenology ( Figure 2 ) and required cross pollination for seed set (see control plants in pollination experiments). The species were chosen to be frequently visited by a broad spectrum of both generalist and specialist wild bees ( Figure 1B ; Kuppler et al., 2023).

Sinapis arvensis (L.) ( Figure 2D ) is an annual Brassicaceae flowering from May onward. For experiments, S. arvensis seeds (Rieger-Hofmann GmbH, Blaufelden, Germany) were sown into 16 x 16 cm flowering pots with 2 to 3 individuals per pot.

Salvia pratensis (L.) ( Figure 2C ) is a perennial Lamiaceae blooming from May to August. It is self-compatible but requires pollinators for pollen transfer (Moughan et al., 2021). Its flowers are specialized for pollination by large bees, e.g., bumblebees, and are rarely visited by insects other than bees (Moughan et al., 2021). This species can produce a maximum of 4 seeds per floret.

Both Asteraceae species, i.e., Centaurea jacea (L.) ( Figure 2A ), and Cichorium intybus (L.) ( Figure 2B ), are perennial and bloom from June/July until the end of season. Bees are the most abundant visitors of Centaurea jacea (Hirsch et al., 2003). All perennial species ( S. pratensis, C. jacea, C. intybus) were cultivated in 16 x 16 cm flower pots with single plants per pot in the plant-rearing area of the Botanical Garden Ulm for at least one year prior to the experiments.

Experiments were performed in the spring and midsummer of 2020. Immediately before starting to bloom, plants were placed in pollinator-excluding cages of fine mesh to ensure that flowers could only be pollinated during the field experiments. When in full bloom, plants were brought into the study sites to be pollinated by the local bee community. Some plants remained in the flight cages as controls for self-pollination; no seed set occurred in the controls. The two different blooming seasons resulted in two different flower pairings. For the spring experiments, 4 pots of S. pratensis and 6 pots of S. arvensis, and for the summer experiments, 4 individuals of C. jacea and C. intybus were placed in the field per habitat type per study site, respectively. The plants were left on the study plots for 72h when sunny weather was forecast. All plant pots for one plot were placed on two trays (ca. 35 cm x 50 cm x 4 cm) and were located at least 1.5 m from the edges of the plots. The plants were provided with at least 2.5 l water per plant tray to avoid drought stress during the experiments. Subsequently, plants were moved back into the mesh cage where they were checked regularly for seed maturation. Seeds were harvested when flowers or siliques turned dry and were then spread out to dry to avoid the growth of mold. The seed number was then counted and weighed using accurate weighing scales (accuracy of 0.1 mg).

For S. arvensis, 10 siliques per plant were chosen, and their intact seeds (round, with a diameter of at least 1 mm) were extracted manually and weighed to obtained an average weight per seed. For plants with fewer siliques, all siliques were measured. All seeds of the respective plant were then counted and weighed, and the average weight of intact seeds was calculated. As S. pratensis has a constant number of 4 seeds per flower, only the number of intact seeds of ten flowers were counted, and the relevant seeds were weighed. Again, the total number of flowers per plant was counted, and the seeds were retrieved, sorted, and weighed. For the Asteraceae species, intact seeds were broken off the dried flowers and processed as described above.

Wild bee monitoring data for the season of 2020 were received from the BienABest project (for detailed information on bee monitoring see Neumüller et al., 2020; Neumüller et al., 2022 and Verein Deutscher Ingenieure e.V, 2023). Between April and September 2020, five sampling events took place per plot covering the entire season at intervals of 3 to 4 weeks, starting with the flowering of the Taraxacum sect ruderalis dandelion. Each sampling event consisted of two 25 min transect walks per plot, one in the morning and one in the afternoon, during which collectors were able to move freely within the plot. Sampling was performed by wild bee experts to guarantee reliable species determination. All observed bee individuals (males and females) were caught with an entomological net, except for those that could be determined at first sight, and were identified to species level in the field. Individuals that could not identified in the field were taken to the laboratory and identified using a binocular microscope and standard literature. Sampling was conducted only during sunny weather (cloud cover maximum 30%), at temperatures higher than 10°C and under low wind conditions.

This method for the bee survey was undertaken with the aim of recording high numbers of bee species and floral resources for wild bees with standardized monitoring methods according to Verein Deutscher Ingenieure e.V, 2023 and represents a trade-off between the highest standardization and the recording of the complete species inventory in an appropriate period of time. The methodology was additionally verified by Herrera-Mesías et al., 2022.

All analyses were performed using R (version 4.2.2, The R Foundation for Statistical Computing). The nearest weather station for each study site was determined, and the corresponding weather data for each day and site was retrieved from the Deutscher Wetter Dienst (DWD, www.dwd.de; Supplementary Table S1 ).

A total of 10,922 bee individuals of 210 bee species were recorded, which could be grouped into 4,572 individuals of 187 bee species on near-natural grassland and 6,350 individuals of 163 bee species on wild flower strips ( Figure 3 ), resulting in a significant difference of species richness between habitat types (non-overlapping confidence intervals). Of these 210 species, ca. 68% (143 species) overlapped, 47 species only occurred on near-natural grassland, and 23 species occurred only on wild flower strips. The average number of individuals per species was higher on wild flower strips (mean ± SD [lowest value, highest value]: 38.9 ± 117.4 [1, 1119] individuals/species, 40 singletons) than on near-natural grassland (24.4 ± 70.6 [1, 561] individuals/species, 39 singletons). Extremely abundant species (more than 100 individuals in total) made up ca. 61% of bee individuals observed on near-natural grassland (10 species, most abundant Bombus terrestris bumblebees) and ca. 75% of bee individuals observed on wild flower strips (15 species, most abundant B. lapidarius bumblebees). Shannon diversities H’ of each habitat type within the individual study sites indicated an overall high diversity with average indices of H’ = 2.73 (SD: 0.37; [1.69, 3.33]) for near-natural grassland and H’ = 2.79 (SD: 0.41; [1.77, 3.33]) for wild flower strips.

The global models across all plant species showed that the standardized seed weight was significantly positively correlated with the bee species richness of the experimental day ( Figure 4B and Table 1 ) and with the Shannon index H’ of the entire season ( Figure 4A and Table 1 ). Furthermore, both models revealed significant differences between the habitat types, with near-natural habitats outperforming wild flower strips in both cases and differences among the plant species ( Table 1 ).

When testing for the dependency of standardized seed weight on the bee abundance, we found no significant effect, although here again, we determined significant differences between the habitat types, with near-natural habitats outperforming wild flower strips, and among the various plant species ( Figure 4C and Table 1 ).

Because of significant differences between plant species in the global models, we performed pairwise comparisons: no significant differences between the two Asteraceae species, C. intybus and C. jacea, were revealed, but these species showed significant differences from S. arvensis and S. pratensis, which again did not exhibit significant differences from each other (see Supplementary Tables S3 and S4 ).

For the Asteraceae specific models, we found a significant positive relationship for greater seed weight with greater Shannon index H’ and bee species richness ( Figure 5 and Table 2 ). Neither of the models showed a significant difference between the habitat types ( Figure 5 and Table 2 ). For the S. pratensis Shannon index H’ model, we observed a significant positive relationship for the Shannon index H’ and a difference between the habitat types with near-natural habitats again outperforming wild flower strips ( Figure 5 and Table 2 ). Similar results were obtained for the S. pratensis species richness model ( Figure 5 and Table 2 ).

Significant differences were also found between habitat types for both S. arvensis models, with plants on wild flower strips performing better than on near-natural habitats ( Figure 5 and Table 2 ). Nevertheless, only the Shannon index H’ was significantly positively correlated to seed weight, whereas species richness was not ( Figure 5 and Table 2 ).

The study locations of our pollination experiments hosted a high bee diversity and abundance. Higher species richness was found on near-natural grassland. Permanent habitats such as near-natural grasslands are normally rich in flowers and also provide diverse nesting sites for wild bees, thus explaining the high bee diversities (Requier and Leonhardt, 2020). In comparison, wild flower strips generally had a higher bee abundance while being inhabited by fewer species. Despite these differences, the Shannon index H’ was not significantly different over the two habitat types and across all study sites. The relatively high Shannon index H’ indicated the high proportions of intermediate and rare species (Peet, 1974). The wild bee monitoring data and previous analyses of the bee diversity at the study locations (Neumüller et al., 2020) confirmed the occurrence of rare species that are often threatened according to the German Red List of wild bees (Westrich et al., 2011). We also found a substantial overlap of bee species between both habitats. These findings indicate that the artificially created flower strips were highly attractive for wild bees and can be considered as a successful conservation measure for improving the availability of floral resources within a region. Many commercially available seed mixtures with high proportions of annual non-native plants often only increase the abundance of a few abundant bee species (Albrecht et al., 2020). In contrast, the floral resources of the examined flower strips were sustainably relatively flower-rich over several years because of an effective mowing regime and the use of a species-rich seed mixture (Neumüller et al., 2021). The perennial flower strips still provided valuable food resources including important pollen hosts of wild bees (Kuppler et al., 2023), and the high flower cover resulted in a high bee abundance. Nevertheless, permanent habitats such as near-natural grassland hosted the highest numbers of bee species. These habitats, which were extensively managed, were highly diverse in flowers and also provided diverse nesting sites for wild bees, factors that explain high bee diversities (Requier and Leonhardt, 2020).

The high numbers of bee species found in near-natural grasslands might explain the significant influence of the habitat type on the seed setting rates. We found a significant influence of the habitat type, with higher seed setting rates in near-natural grassland compared with those on flower strips. Previous studies have shown that structure-rich landscapes have a positive effect on bee diversity (Kennedy et al., 2013; Neumüller et al., 2020) and, consequently, on seed set (Albrecht et al., 2020). Nevertheless, increasing distances to pollinator-friendly habitats have a negative effect on plant reproductive success (Steffan-Dewenter and Tscharntke, 1997; Steffan-Dewenter et al., 2001), and high proportions of arable land in a landscape, often combined with decreased structure richness, can negatively affect the number of produced seeds (Söderman et al., 2018; Herbertsson et al., 2021; Ammann et al., 2024). In contrast to these previous studies, we have not analyzed effects at the landscape scale, as the two habitat types were in close proximity at a study site and therefore were surrounded by the same landscape within the flight distance of most wild bee species (Steffan-Dewenter et al., 2002). In our study, the bee community was determined on 0.3 ha plots established within different habitats. Although the previous studies mentioned above did not directly monitor bee diversity in the surrounding landscape elements, we can draw similar conclusions from our results. We conclude that both the preservation of near-natural habitats as a reservoir of wild plant species diversity and the conservation of reproduction-assuring pollinators are of great importance for seed setting.

In our pollination experiments, we found increases in seed set with greater wild bee diversity (species richness and Shannon index H’) across habitats and study sites. Although the bee species richness was calculated for the experimental day and specific plot, the Shannon index H’ represents the bee diversity of the entire season. We thus compared two bee parameters and included two different time-scale approaches. We were able to show that both parameters significantly influenced the seed weight, as greater species richness and a greater Shannon index H’ resulted in greater seed set, and both seemed to be a suitable measure for seed set prediction. Both parameters allowed conclusions to be made from monitoring data without the performance of time-consuming pollination experiments and direct observations of bee visits. The studies reviewed by Garibaldi et al., 2013 and the meta-analysis by Woodcock et al., 2019 have especially particularly highlighted the important role of wild bees in pollination and have shown that the visits of diverse wild bee species, in particular, explain the pollination success. For example, distinct bee assemblages increased the seed set of Helianthus anuus by up to 45% (Mallinger et al., 2019).

Wild bee diversity highly depends on the structure of the habitat, but the effects of the habitat type on pollination success and its dependency on wild bees have previously often been measured only indirectly as effects of surrounding landscape elements. For example, the seed set of Raphanus sativus is increased by flower strips in the surroundings (Albrecht et al., 2007) and, for the seed set of Sinapis arvensis, by near-natural habitats. Although these previous studies have not directly monitored the wild bee community of the habitats, as carried out in our study, the results support our findings: habitats that naturally host or are established to promote high bee diversities increase the seed set in wild plant species. Concurrently, pollinator diversity, in addition to bees and thus resulting niche complementarity, has been shown to impact fruit set positively (Albrecht et al., 2012; Magrach et al., 2021), but this was not a focus of our study.

As the chosen plant species benefit from cross-pollination by bees and are highly attractive to generalists and specialist bees, we assume that the higher seed weights are attributable to cross-pollination by diverse bee species. Seed weight is particularly linked to the initial growth of the plantlet (Lamont and Groom, 2013). In this phase, the fitness of the plantlet in its surroundings is crucial for its survival. Survival is assured, for example, by its genotypic characteristics such as higher heterozygosity presumably arising from cross-pollination, adaptability resulting from cross-pollination, and its fast growth as a competitive advantage (e.g., competition for light by out-growing other plants). We conclude that the high bee diversities of our habitat types provides the necessary pollination services to maintain the next generations of study plant species, which are important host plants of diverse bee species.

In addition to bee diversity, a greater bee visitation rate has been reported to result in a higher seed set (Herbertsson et al., 2021). As a high bee abundance leads to more potential flower visits, we expected to observe this effect in our experiments. However, this was not confirmed by our study. The bee abundance at the study sites was mostly driven by a few widespread bee species that occurred in high numbers, such as the bumblebees Bombus lapidarius and B. terrestris. An effective seed set thus seems not to be strongly affected by abundant bee species. More specialized visitors that are often part of species-rich wild bee communities can, in contrast, greatly improve pollination (Mallinger et al., 2019). Similarly, we found an increase in pollination via higher bee diversity, which can largely explain the seed sets in the studied plant species. The foraging behavior and functional traits of intermediate or rare bee species are more important than high abundancies of common bee species. An alternative explanation for the non-significant effect of bee abundance might be that a saturation in pollination occurs (Morris et al., 2010). This means that at least the minimum bee abundance needed for pollination was assured in both our habitats. On the other hand, over-pollination attributable to an excess of bee visits did not seem to be a factor in our experiments, as this would have resulted in a decreased seed set (Sáez et al., 2014; Garratt et al., 2021).

The pollination experiments revealed that the bee diversity differently impacted the seed set of the four studied wild plant species. In all represented taxa, the seed set was positively correlated with greater Shannon indices H’ and, in two models, also to bee species richness, although the effects depended, in some of the plant species, on the habitat type.