2.1 Study Area and Design

We selected eight landscape circles (r = 500 m; Concepción et al. 2012; Grass et al. 2020) in the Great Hungarian Plain, Central Hungary (Figure 1). The landscape is characterised by vast conventionally managed (ca. 100 kg N/ha on average) agricultural fields and alternating semi-natural grasslands and wetlands; arable land amounted to more than 40% of every landscape circle. Due to rotational agriculture, five different crops (alfalfa, barley, corn, sunflower and wheat, Table A1) were around our wildflower fields (WFFs). We pooled the crop biodiversity data irrespective of which crop was there, as it was clearly distinct from the WFF and road verges where agrochemicals and soil tillage were not applied. The road verges in the area either consist of tree hedgerows or grassy margins just a few metres wide. Each landscape circle contains a 0.5 ha-sized (50 × 100 m) sown WFF in the middle. The fields were sown in early 2020 on the edge of former crop fields with a seed mixture of 32 local insect-visited flowering plant species (for a detailed list of sown species, see Appendix 1 in Báldi et al. 2022) that provide food resources over the whole vegetation period and habitat for nesting, resting and mating for many animal taxa (Báldi et al. 2022).

2.2 Biodiversity Sampling

We recorded vascular plant species number and cover on every transect of each habitat by using two 1 × 1 m quadrats (Martin et al. 2021) per transect (at least 10 m apart). The survey took place in June 2022, as the plant diversity of our WFFs can be estimated the most reliably during that month. In each quadrat, we recorded all vascular plant species using Király (2009), the cover of each plant species, bare soil and litter, with a total cover of 100% (i.e., no multiple vegetation layers). Species with very small cover were noted with 0.01% cover. We used the non-crop plant species for the analyses and used species names according to The World Flora Online (http://www.worldfloraonline.org).

Bees (including wild bees and honeybees) and hoverflies were surveyed during transect walks of 50 m in length and an observation width of 1.5 m. During a net sampling time of 7.5 min (handling time of specimens excluded), we aimed to capture all observed (foraging, flying or basking) individuals (except bumblebee queens and honeybees) with a sweep net for later identification. Captured bees and hoverflies from a given transect were pooled in vials containing 70% alcohol. Specimens which were neither successfully caught nor identified to the species level in the field were noted in broad categories such as ‘wild bee’, ‘bumblebee’ or ‘hoverfly’ (22.12% of bees at the early summer sampling occasion, 23.18% of bees at the mid-summer sampling occasion and 47.06% of hoverflies at the early summer sampling occasion). We also recorded information on the date and the location of the transect (i.e., landscape circle ID, transect ID, habitat type), measured temperature [°C] and estimated cloud cover [%] before starting the sampling to verify that the sampling was conducted during favourable weather conditions (min. temperature 18°C, average 26°C, max. 34°C, not more than 5 m/s wind and min. 0%, average 15%, max. 100% cloud cover). We sampled twice a year in 2022: early summer (late May—early June) and mid-summer (early August). We performed the survey at a time when the target taxa are most active: in early summer between 8:30 and 17:00 and in mid-summer between 7:00 and 17:00. In case the temperatures were higher than 34°C in mid-summer, and the observed activity of the bees and hoverflies decreased, we suspended the work for that period (i.e., 13:00 or 14:00). Specimens were identified to the species level in the lab by Zsolt Józan and Zoltán Soltész, using Móczár (1957), Ebmer (1969, 1971), Speight (2020) and Tóth (2011) as reference guides.

Spiders were caught by suction-sampling using a modified leaf blower (Stihl SH 86) where a gauze bag was fixed into its nozzle; the method was adapted from the BioBio project's protocol (Kovács-Hostyánszki et al. 2013). We took five sub-samples per transect at least 20 m apart (exceeding the length of the pollinator transect). Each of the five sub-samples was taken within a 0.357 m internal diameter sample ring pre-placed on the target vegetation (total sampled area per transect: 0.5 m²). The suction nozzle was placed over and pushed into the vegetation and moved within the sample ring for 30 s. The five sub-samples were pooled in a polyethene zip-seal bag and filled with 70% alcohol. Later in the lab, they were stored in a freezer until sorting. We surveyed twice a year: in landscape circles with barley and wheat 3 and 10 weeks after dandelion (Taraxacum officinalis) was in bloom (early May and mid-June), and in landscape circles with corn, sunflower and alfalfa 6 and 20 weeks after dandelion bloom (late May and early September). Adult spiders were identified at the species level, while juvenile spiders (45.56% of all individuals) were identified at the family level. Heimer and Nentwig 1991 and Nentwig et al. (2023) were used as reference guides during identification.

2.3 Statistical Analysis

We used generalised linear mixed models (GLMM) with Poisson distribution to study the effect of the different habitat types on species and individual numbers; in the models landscape and transect IDs were nested random factors. In the case of vegetation cover, we applied beta distribution in the GLMM. We used Tukey's test to evaluate the significance of differences in species and individual/cover numbers for the three habitat groups. We included only the individuals identified to the species level for species number but included all individuals (i.e., observed-only bees, hoverflies, and juvenile and adult spiders) to the individual number. The two sampling rounds were analysed separately for bees, while data analysis of the second sampling round was impossible for hoverflies and spiders due to the low number of individuals.

We used the glmmTMB (‘1.1.8’, Brooks et al. 2017) packages for modelling, the multcomp (‘1.4.25’, Hothorn, Bretz, and Westfall 2008) for post hoc Tukey test to know which habitat types were significantly different from each other, the tidyverse (‘2.0.0’, Wickham et al. 2019) and dplyr (‘1.1.4’, Wickham et al. 2022) packages for data manipulation and ggplot2 (‘3.5.1’, Wickham 2016) and ggpubr (‘0.6.0’, Kassambara 2023) to visualise the number of species, individuals and the vegetation cover.

We performed PERMANOVA using zero-adjusted Bray-Curtis dissimilarities (the number of permutations was 999) to evaluate the effect of habitat types on the assemblages of non-crop vegetation, bees, hoverflies and spiders. We also visualised the assemblage composition of different habitat types for non-crop vegetation, bees, hoverflies and spiders with non-metric multidimensional scaling (NMDS). Since we did not observe individuals on every transect, a dummy species with abundance = 1 was added to all transects to facilitate the inclusion of blank or nearly empty transects (Clarke, Somerfield, and Chapman 2006; Borcard, Gillet, and Legendre 2018) and data points were jittered for better visibility. We also added convex hulls to the figures to visualise the overlap between the three habitat types. To perform the NMDS, we used the following packages: vegan (‘2.6.4’, Oksanen et al. 2022), labdsv (‘2.1.0’, Roberts 2023), TeachingDemos (‘2.13’, Snow 2024) and goeveg (‘0.7.5’, Goral and Schellenberg 2024).

Based on the average species composition in each taxon in each habitat type, we calculated the Bray–Curtis dissimilarity (as a beta diversity index; see Schroeder and Jenkins 2018) for pairwise comparisons of habitat types (road verge—crop; WFF—crop; WFF—road verge) for non-crop vegetation, bees, hoverflies and spiders. Finally, we calculated the significance of difference in dissimilarity using the adonis function of the vegan package (‘2.6.4’, Oksanen et al. 2022).

We used the R statistical environment (version 4.4.0) to perform all the analyses.

3.1 Primary Producers—Plants

The number of non-crop plant species was significantly different in the three habitats according to the GLMM (Tukey tests), with the crop field habitats having the lowest species numbers, followed by road verges and WFFs (all p < 0.001; Figure 2a; Table 1; Table A2). We found similar results for the cover percentages of non-crop plant species: the crop habitats were covered with only a few percent of non-cultivated plants, while in the WFF and the road verge habitats, the cover was significantly higher (all p < 0.001; Figure 2b; Table 1; Table A2). The NMDS analysis showed that the assemblage composition of the three habitats is distinctly different (PERMANOVA: R² = 0.36221, p = 0.001; Figure 2c; Figure A1, Table A3). According to the Bray–Curtis dissimilarity, WFF and crop habitats were the most different in their plant assemblages, while WFF and road verge habitats were more similar (Table 2).

3.2 Pollinators—Wild and Honey Bees and Hoverflies

We found no significant differences in species numbers of bees (Figure 3a,d; Table 1; Table A2). We observed significant differences between the three habitats' individual numbers, with only a few bee individuals inside the crop fields, intermediate levels in the road verge habitat and highest in the WFF (Figure 3b,e; Table 1; Table A2). In the early summer sampling occasion, road verge and WFF habitats were different in their bee assemblage composition according to the NMDS (R² = 0.4468, p-value = 0.001; Figure 3c; Table A3). In the later sampling session, the assemblages of the bees captured within the crop field overlapped with the assemblage of the road verge, and the assemblage of the road verge and WFF habitats partially overlapped with each other, but still differed significantly (R² = 0.52732, p-value = 0.001; Figure 3f; Table A3). In terms of species compositions, the pairwise Bray–Curtis dissimilarities suggest a bigger difference between WFF and crop, and WFF and road-verges, while showing a smaller dissimilarity between crop and road verge in both sampling occasions (Table 2).

In early summer, we did not capture any hoverflies that could be identified at the species level in the crop habitats and only a few species were captured on the transects in the road verge and WFF habitats. We did not find any significant differences in either species (Figure 4a; Table 1; Table A2). Hoverfly individual numbers at the WFF habitats were significantly higher compared to the crop habitats (Figure 4b; Table 1; Table A2). The NMDS plot showed some overlap between the road verge and the WFF habitats‘hoverfly assemblage composition at the earlier sampling occasion (PERMANOVA: R² = 0.37678, p = 0.003; Figure 4c, Table A3). In terms of species compositions, the road verges and the WFF habitats were similar, as the pairwise Bray–Curtis dissimilarity suggests (Table 2). The number of species and individuals captured and observed during the second sampling round was too low to use in a robust statistical analysis and to have a reliable conclusion.

3.3 Predators—Spiders

According to the Tukey test, the number of spider species was similar in the three habitats (Figure 5a; Tables 1 and 2). However, we found a significantly lower number of individuals (Figure 5b; Table 1; Table A2) at the crop and the WFF habitats compared to the road verges in the early summer sampling. This could also be observed in the spider assemblage composition, where the crop and WFF differed from the road verge (R² = 0.19178, p = 0.002; Figure 5c; Table A3). In terms of species compositions, the pairwise Bray-Curtis dissimilarities suggest a bigger difference between road verge and crop, and WFF and road verge, while showing a smaller dissimilarity between crop and WFF (Table 2). Due to the high number of juvenile individuals that could not be identified at the species level, the data from the second sampling round was omitted.

4.1 Primary Producers—Plants

Unsurprisingly, the crop fields were the most species-poor habitat type in terms of plant species due to the frequent soil ploughing, the subsequent crop rotation and herbicide treatments. Non-crop plant species are regularly eradicated in intensive agriculture as they may host crop pests (Kumar, Bhowmick, and Ray 2021) and are considered weeds potentially interfering with crop growth (Gallandt and Weiner 2015; Korav et al. 2018). Road verges are valuable semi-natural habitats in farmland landscapes due to their connectivity function (Phillips et al. 2020; Dániel-Ferreira et al. 2022) and important abiding habitat in an otherwise impoverished arable surrounding (Phillips et al. 2020). In our study, we found these road verges to mainly consist of a few plant species, such as grasses and weedy, generalist and ruderal species, with few flower resources. That is in strong contrast to the sown WFFs, where we found more specialists (according to the definition of Borhidi (1995) and Horváth et al. (1995)), including sown plant species, but even more species that had emerged from the soil seed bank or dispersed into the fields (Malo and Suárez 1995). Continued management, such as mowing, will be needed for a sustained diverse habitat (Báldi, Batáry, and Kleijn 2013; Kiss et al. 2017). The number of plant species in the WFF habitat was the highest and may therefore support more diverse and complex insect assemblages (Ebeling et al. 2018). The three habitat types had distinct plant assemblages that supported separate pollinator and predator assemblages, a finding which concurs with previous studies (Schaffers et al. 2008; Hussain et al. 2021; Brandl et al. 2022).

Species associated with the crop habitat were mostly weeds (classification after Borhidi (1995)), while the road verges mainly hosted disturbance tolerants. The WFFs were characterised by natural pioneers and stress tolerants for the seeded and naturally appearing species.

4.2 Pollinators—Wild and Honey Bees and Hoverflies

We observed rather few bees and hoverflies inside the crop field, although some species of the latter lay eggs close to aphid colonies (Almohamad et al. 2007; Almohamad, Verheggen, and Haubruge 2009; Miličić et al. 2021; Vujanović et al. 2023) and are considered an important pest control (and pollinator) agent (Almohamad et al. 2007; Rodríguez-Gasol et al. 2020). Since most crops in our study area were wind-pollinated, it might not surprise to observe only a few pollinators, although honey bees have been shown to forage on maize pollen, among others (Danner, Härtel, and Steffan-Dewenter 2014). In the road verges, bees were significantly less abundant than in the WFFs, which can be explained by the wide array of pollinator-friendly plant species in the latter and the low number of herbs in the verges (Albrecht et al. 2020; Threadgill et al. 2020; Wen et al. 2022). A further reason might be the higher percentage of bare soil in the WFFs, which ground-nesting bees may profit from for nesting (Gardein et al. 2022). We observed a similar pattern for hoverfly species numbers, although individual numbers did not significantly differ between road verges and WFFs. This observation is supposedly due to a preference for areas with higher plant density (Dániel-Ferreira et al. 2022) and shelter from predators or—for larvae—the provision of additional food resources in road verges (Sutherland, Sullivan, and Poppy 2001). Therefore, we conclude that the pollinator groups do not benefit equally from newly established wildflower fields, but for hoverflies, permanent semi-natural grasslands such as road verges may be more important. Pairwise Bray–Curtis dissimilarities were the lowest among crops and road verges, indicating higher similarity in terms of their bee assemblages (Table 2).

4.3 Predators—Spiders

Although spiders are considered important pest control agents in farmland (Nyffeler and Sunderland 2003; Michalko et al. 2019), we found only a few spiders in the crop fields (e.g., Agyneta rurestris, Trichoncoides piscator, Xysticus kochi). This may be due to regular ploughing that recurrently erases the spider populations (and their prey) in the crop fields (Schneider, Krauss, and Steffan-Dewenter 2013; Plath et al. 2021), meaning that the road verges could become refugia in the absence of more natural habitat. The varied and dense vegetation structure provided by the road verge habitat can be an important benefit for spiders (Balzan, Bocci, and Moonen 2014; Plath et al. 2021; Mei et al. 2021). Surprisingly, the WFFs hosted only a few spiders, supposedly because the time since the establishment of the field was insufficient for them to occupy the new habitat (Maas et al. 2021; Hussain et al. 2021) or because the vegetation is not yet dense enough to provide optimal habitat (McDonald 2007). WFFs have the potential to provide additional resources to predators, such as undisturbed overwintering and reproductional sites and increased prey resources (Hoffmann et al. 2021; Mei et al. 2021; Plath et al. 2021), but some of these are also provided by road verges. Furthermore, sown WFFs enhance pest control in the adjacent or surrounding agriculturally managed fields, as not only spiders but other pest control agent groups are also promoted (Blaauw and Isaacs 2012; Albrecht et al. 2020; Bischoff et al. 2022; but see Török et al. 2021).

We found generalist grassland spider species in the road verge habitats (e.g., Aulonia albimana; Talavera aequipes) and disturbance tolerant agrobiont species (i.e., that reach high dominance in agroecosystems; Samu and Szinetár 2002) inside the crop fields (e.g., Agyneta rurestris). The latter use road verges as overwintering sites when the crop fields lay bare. Our WFFs may be similar to the crop field in that they are relatively newly established and do not have a complex vertical structure yet. We expect that spiders will increasingly use the WFFs habitats within a few years because the WFFs will become more similar to the road verges—currently, their assemblage composition is already moderately similar (Table 2)—as grass species will colonise the WFFs and more vertical structures will develop.

4.4 The Broader Context

Our results suggest no habitat fits all, and a diversity of habitat types is key if different functional groups are to be supported (Balzan, Bocci, and Moonen 2014; Gayer et al. 2021; Hussain et al. 2022). Furthermore, our results show similar patterns for the studied insect groups as in other agricultural systems in western and northern Europe. In conclusion and beyond our results, there is massive evidence of the beneficial role of non-crop habitats in supporting a wide range of essential ecosystem service providers and biodiversity (e.g., Haaland, Naisbit, and Bersier 2011; Balzan, Bocci, and Moonen 2014; Dolezal, Esch, and MacDougall 2022). Our study showed, however, that different non-crop habitats support different functional groups; thus the ecology of the different targeted functional groups should be taken into account when planning effective conservation and restoration (Holl and Aide 2010). Beyond diversity protection per se, the maintenance of ecosystem functioning also requires a diversity of habitats on the landscape. Such knowledge will be valuable input for the implementation of the EU's Nature Restoration Law by member states when agricultural landscapes and pollinators are targeted, and more broadly to the transformation of agriculture (Báldi et al. 2023). Note, however, that due to large yearly variations in arthropod abundance, and the diversity of farm systems across Europe, these results need to be validated locally before applying to the design of local or regional multifunctional landscapes. Thus, we encourage further research to help fill the knowledge gaps, especially in East Central European landscapes.